Whole Community Metatranscriptomes and Lipidomes Reveal Diverse Responses Among Antarctic Phytoplankton to Changing Ice Conditions

The transition from winter to spring represents a major shift in the basal energy source for the Antarctic marine ecosystem from lipids and other sources of stored energy to sunlight. Because sea ice imposes a strong control on the transmission of sunlight into the water column during the polar spring, we hypothesized that the timing of the sea ice retreat influences the timing of the transition from stored energy to photosynthesis. To test the influence of sea ice on water column microbial energy utilization we took advantage of unique sea ice conditions in Arthur Harbor, an embayment near Palmer Station on the western Antarctic Peninsula, during the 2015 spring–summer seasonal transition. Over a 5-week period we sampled water from below land-fast sea ice, in the marginal ice zone at nearby Palmer Station B, and conducted an ice removal experiment with incubations of water collected below the land-fast ice. Whole-community metatranscriptomes were paired with lipidomics to better understand how lipid production and utilization was influenced by light conditions. We identified several different phytoplankton taxa that responded similarly to light by the number of genes up-regulated, and in the transcriptional complexity of this response. We applied a principal components analysis to these data to reduce their dimensionality, revealing that each of these taxa exhibited a strikingly different pattern of gene up-regulation. By correlating the changes in lipid concentration to the first principal component of log fold-change for each taxa we could make predictions about which taxa were associated with different changes in the community lipidome. We found that genes coding for the catabolism of triacylglycerol storage lipids were expressed early on in phytoplankton associated with a Fragilariopsis kerguelensis reference transcriptome. Phytoplankton associated with a Corethron pennatum reference transcriptome occupied an adjacent niche, responding favorably to higher light conditions than F. kerguelensis. Other diatom and dinoflagellate taxa had distinct transcriptional profiles and correlations to lipids, suggesting diverse ecological strategies during the polar winter–spring transition.


INTRODUCTION
The western Antarctic Peninsula is a region characterized by a highly dynamic sea ice regime. Seasonal and higher frequency shifts in sea ice cover are superimposed on large, inter-annual variations in ice thickness and extent. In the coastal zone this variability coincides with the zone of highest biological productivity, setting the stage for strong physical control of biological processes that enable the region's rich marine ecosystem (Bowman et al., 2016). During the winterspring seasonal transition, sea ice rapidly advects in and out of the shallow coastal zone. This sea ice impacts phytoplankton communities by limiting light penetration into the surface ocean and stabilizing the surface mixed layer. Though sea ice cover leads to a shallow mixed layer, a scenario conducive to a strong spring bloom, it also decreases the light available to drive photosynthesis. In regions of highly variable sea ice cover phytoplankton must respond to rapidly changing light conditions independent of the circadian rhythms available to coordinate metabolic responses in lower latitude oceans.
Phytoplankton utilize a number of strategies to survive periods of prolonged darkness. While some utilize cysts or resting spores, and are thus metabolically inactive, others rely on lipid or carbohydrate energy stores to survive periods of prolonged darkness. Recent analysis of the translational response of Fragilariopsis cylindrus, a model sea ice diatom, to prolonged darkness suggests regulation of multiple systems to sustain cell function and enable a fast response to light (Kennedy et al., 2019). Specifically, those authors identified an increase in proteins diagnostic of catabolic processes including glycolysis, the Entner-Doudoroff pathway, and the tricarboxylic acid cycle. Significantly, however, proteins that are necessary for photosynthesis and related anabolic functions were not necessarily less abundant during prolonged darkness. This suggests an adaptation to the extreme seasonal cycle of light and dark based on a fast response to light. As soon as the photon flux is sufficient to drive the photosynthetic machinery F. cylindrus is able to undergo primary production (PP); there is no need to expend additional resources retooling the cell's molecular machinery.
Lipids are a key component for the survival of phytoplankton in the dark, but these compounds and their derivatives play important roles beyond energy storage . While triacylglycerols (TAGs) are primarily involved in energy storage, phospholipids, including phosphatidylcholine (PC), phosphatidylethanolamine (PE), and phosphatidylglycerol (PG), are key components of cell membranes. Free fatty acids (FFA) are rare in healthy cells, and, thus, are indicative of cell degradation. Oxidized FFA (i.e., oxylipins) can be deleterious to bacterial growth (Edwards et al., 2015). The glycolipids sulfoquinovosyldiacylglycerol (SQDG), digalactosyldiacylglycerol (DGDG), along with PG, are the primary structural lipids associated with the chloroplast.
The community lipidome can change on scales much shorter than seasonal. Recent work in the North Pacific subtropical gyre revealed a diel cycle in TAGs, with these compounds being synthesized during the day and catabolized at night (Becker et al., 2018). Given the complexity of light during the polar winter-spring seasonal transition, with day length and sea ice cover interacting in unpredictable ways, we sought to identify major contributors to the lipidome and patterns in lipid synthesis and catabolism through a coupled metatranscriptomic and lipidomic study. We made use of unusual sea ice conditions during the 2015 Austral spring to investigate the response of Antarctic phytoplankton to changing light conditions. Samples were collected from below sea ice, the marginal ice zone, and microcosms to identify different physiological responses to variable conditions associated with the winter-spring transition. Although metatranscriptomes are easily disaggregated into their component transcriptomes by mapping to references, it is much more difficult to assign lipids to specific phytoplankton (except for lipids that are uniquely produced by a given taxonomic group). To overcome this limitation we used principal components analysis (PCA) as a dimension reduction technique, correlating changes in lipid concentration to principal components describing up-or downregulated genes for each reference phytoplankton transcriptome. This approach allowed us to identify key phytoplankton taxa responding to our experimental conditions, and different lipid utilization strategies employed by these phytoplankton during the Antarctic winter-spring seasonal transition.

Ethics Statement
Written informed consent was obtained from the individuals for the publication of any potentially identifiable images or data included in this article.

Experimental Design and Sampling
Persistent land-fast sea ice in Arthur Harbor allowed us to establish an ice station at 64.77253 • S and 64.05411 • W, on November 4, 2015 ( Figure 1A). On November 10, 2015 we initiated a microcosm experiment to simulate the effects of ice removal and concomitant exposure to sunlight. Using a peristaltic pump and acid-washed tubing we filled 12 × 10 L acid-washed carboys with seawater from the ice station ( Figure 1B). The carboys were transported to an outdoor recirculating incubator ( Figure 1C) and shaded with screening to approximate light conditions at 10 m with no sea ice cover (approximately 30% irradiance as determined with a LI-COR Spear sensor for November 21, 2014 at Station B) (Supplementary Figure 1). As a control, three carboys were covered with additional layers of screening to maintain light levels below an estimated 1% irradiance, which was similar to under the sea ice. Three carboys each were sampled on November 11 (24 h; day 1), November 13 (72 h; day 3), and November 18 (192 h; day 8) (Figure 1E). On November 18 we also sampled the three dark controls.
PAR profiles were collected from the Ice Station on November 4 (6 days prior to T 0 ) and November 15 (day 5), and from Station B on November 17 (day 7). Profiles were collected with a LI-COR 193 against a LI-COR 193 surface reference. Care was taken to position the casts on the unshadowed side of the Zodiac (for Station B) and ice sampling team (Ice Station).
Concurrent with the ice removal experiment, seawater samples were collected from the Ice Station on November 4, November 15, November 25, and November 30 ( Figure 1E), at which time the ice was deemed too unstable for travel. At each sampling opportunity an ice core was removed with a Sipre-type corer. Acid-washed peristaltic pump tubing secured to a bamboo pole was inserted into the ice to a depth of 1 m below the ice surface. For the metatranscriptomes, a fieldportable peristaltic pump was used to filter water onto sterile 47 mm 0.2 µm Durapore filters (PALL). Immediately after filtering, the filter was inserted into a bead-beating tube preloaded with RNeasy PowerWater kit (Qiagen) buffer #1 and 10 µL 2-betamercaptoethanol (Sigma). The tube was inverted to fully coat the filter surface, and the filters were placed on ice and protected from sunlight until extraction within 4 h of sampling.
For lipidomics, the peristaltic pump was used to fill acidwashed polycarbonate bottles. Care was taken to shade the bottles during collection and transport. Within 4 h the samples were filtered on 47 mm diameter, 0.2 µm pore size, hydrophilic PVDF membranes (Millipore) under gentle vacuum. The membranes were flash frozen in liquid nitrogen and stored in a −80 • C freezer.
Samples were collected from the Palmer Long Term Ecological Research project (PAL-LTER) Station B, located at approximately 64.78205 • S and 64.07934 • W, on November 22, November 26, December 2, December 7, and December 15 (Figures 1A,D). Samples were collected by peristaltic pump from a Zodiac, at 5 m depth, following the same methods employed for sampling at the ice station.
RNA extraction followed the manufacturers recommended protocol, including DNase treatment. RNA was quantified using the Qubit HS RNA kit (Invitrogen). Parallel samples were collected for DNA and extracted using the DNeasy PowerWater kit (Qiagen), quantification was carried out using the Qubit HS DNA kit (Invitrogen). Measurements for chlorophyll a, bacterial production (BP), PP, and fluorescence induction and relaxation (FIRe) were carried out by the Palmer Long Term Ecological Research Program using the standard methods available on the Palmer LTER DataZoo (Palmer LTER Science Team, 2021).

Sequencing and Sequence Processing
Sequencing was carried out via PE sequencing on the Illumina HiSeq platform at either the Joint Genome Institute (12 samples) or the UC San Diego IMG Genomics core facility (64 samples). Demultiplexed reads were merged and quality controlled with the Paired-End reAd mergeR (PEAR) (Stamatakis et al., 2014), using a p-value of 0.01, minimum PHRED score of 33, and eliminating merged reads <100 bp in length.
Merged, quality controlled reads were mapped to reference transcriptomes of the Marine Microbial Eukaryotic Transcriptome Sequencing Project (MMETSP) (Keeling et al., 2014) and reference genomes from GenBank with an extension of the paprica bioinformatics pipeline (i.e., paprica-mt) (Bowman and Ducklow, 2015a,b). In brief, reference bacterial, archaeal, and viral genomes were downloaded from GenBank, and reference eukaryotic transcriptomes were downloaded from MMETSP, on February 1, 2017. All unique genes coding for a product with an Enzyme Commission (EC) number were identified and concatenated in a database, along with the taxonomic information of the originating genome or transcriptome, and the name of the gene product. BWA (Langmead et al., 2009) was used to search all query reads against this database. The BWA output in SAM format was parsed to create a tab-delimited file of query reads, best mapped reference sequence, and taxonomy and gene product of the reference sequence.

Data Analysis
For each sample and reference transcriptome/genome, the unique EC numbers were tallied to produce tables of transcript abundance. This approach allowed an independent comparison of each reference transcriptome/genome between replicate libraries for select samples. To make this comparison we employed DESeq2 (Love et al., 2014); differentially expressed transcripts are described as those with a p-value < 0.05 in a Wald test. p-Values reflect Holm-Bonferroni correction for multiplecomparisons as is the default in DESeq2. This experiment was designed such that ice station samples were the initial condition, followed by samples from Station B, then the microcosm incubations. Where comparisons were made within each of these groups, the initial condition was that which occurred first in the time series.
To identify which reference transcriptomes responded similarly to changing environmental conditions we applied PCA to the differential expression data for all points of comparison. PCA was applied to each reference independently. The step was important for dimension reduction; the PCA effectively collapsed one dimension of the data allowing us to compare all reference transcriptomes for all points of comparison in only two dimensions. Transcriptomes with similar magnitudes in the first principal component (PC1) according to a simprof analysis based on Euclidean distance (in deference to positive and negative values) and hierarchical clustering were considered to have a similar response to changing environmental conditions. To explore specific gene expression patterns within each group we then considered the contribution of each enzyme (cos 2 ) to PC1.
To relate the lipidomes to the transcriptomes we constructed a correlation matrix based on Pearson's correlation coefficient (r). For each transcriptome, PC1 was correlated with the change in lipid concentration between each sample. The resulting matrix was organized according to hierarchical clustering based on Euclidean distances (in deference to the positive and negative values).
All statistical analyses were carried out in R v3.6 (R Core Team, 2014) using R Studio and R Studio Server. The simprof analysis was carried out using the "simprof " function in package clustsig (Whitaker and Christman, 2014). For the PCA analysis, the cos 2 values were extracted using the "get_pca_var" function from the factoextra package (Kassambara and Mundt, 2017).

Ecophysiology
Chlorophyll a values responded quickly to the microcosm treatment (Figure 2A). Initial (T 0 ) samples for the microcosms and ice station samples were 0.23 mg m −3 on November 10, 2015. At the time of the final sampling on November 25, 2015 (day 15), the ice station had risen to 0.63 mg m −3 . By contrast, after only 7 days the microcosms had increased to 1.20 mg m −3 (mean of three replicates, SD = 0.50 mg m −3 ). At Station B the values increased from 0.52 mg m −3 on day 17 to 2.08 mg m −3 on day 34. The PP values largely followed the pattern set by chlorophyll a, increasing from initial values of 1.71 mg C m −3 d −1 at the ice station and mesocosm T 0 samples to 41.75 mg C m −3 d −1 in the microcosms on day 8 (mean of three replicates, SD = 26.93 mg C m −3 d −1 ) ( Figure 2B). One exception was the Ice Station day 20 value, which jumped to 24.05 mg C m −3 from 6.03 mg C m −3 d −1 on day 15. Bacterial production values followed PP values with no indication of a lag ( Figure 2C). These values ranged from 3.34 pmol leucine l −1 h −1 on day 6 at Station B to 19.95 pmol leucine l −1 h −1 on day 20 at the Ice Station. For all ecophysiology parameters, only single values are available for most time-points from the Ice Station and all time-points from Station B.
Fluorescence induction and relaxation parameters measured included F v /F m (a key indicator of photosynthetic efficiency), ρ PSII (the cross-sectional absorption area of photosystem II), and ρ (the reaction center connectivity). F v /F m remained relatively stable at the Ice Station over the course of the experiment ( Figure 2D). The high light associated with the microcosms induced rapid perturbations in this parameter, with F v /F m falling from 0.46 on day 0 to 0.41 on day 3, before recovering to 0.50 on day 8. Overall the Station B samples were lower than either the Ice Station or microcosm values, and increased over the course of the experiment from 0.26 on day 6 to 0.41 on day 16. ρ PSII declined over the course of the microcosms, from 440 to 302 ( Figure 2E). ρ PSII was consistently higher at the Ice Station than at Station B. For the microcosms, ρ followed an inverse relationship to F v /F m , falling from a day 0 value of 0.26 to 0.09 on day 1, before recovering to 0.25 on day 3 (Figure 2F). At the start of the experiment ρ was higher at the Ice Station than at Station B, these values converged near 0.08 by day 16.
On November 4 (6 days prior to T 0 ) the 1% irradiance depth at the Ice Station was estimated to be 1.25 m (Supplementary Figure 1). On day 5 it was not possible to get a clean value for the uppermost portion of the water column and we could not estimate the depth of the 1% light level. The shape of the profile, however, suggests a similar value to that estimated for November 4. The lowest irradiance values observed for Station B on day 7 was 17.0%. The irradiance at 5 m depth (the sampling depth for our Station B metatranscriptomes) was 51.7%.

Metatranscriptome Results
We produced 76 merged sequence libraries with a minimum quality score of 33, ranging in size from 3,492,999 to 104,508,538 (mean = 4.94 × 10 7 , SD = 2.49 × 10 7 ). The fraction of reads mapped to a reference transcriptome/genome for each sample ranged from 13.06 to 81.87% (mean = 48.48%, SD = 13.48%). All sequence data is available from the NCBI SRA at BioProject PRJNA494856, and a flat file of differentially expressed genes is provided as Supplementary Table 1. Heatmaps for each enzyme in the paprica-mt database that had significant log 2 foldchange differences for one or more reference transcriptomes are provided as a Supplementary Material.
We identified 93 reference genomes/transcriptomes with significantly differentially expressed genes. Of these, 37 had differentially expressed genes for all points of comparison. These were primarily eukaryotic reference transcriptomes from the MMETSP, but included one prokaryotic reference genome; GCF_000012305.1 (Psychrobacter arcticus 273-4). The prokaryotic genome was not considered further in our analysis. The remaining reference transcriptomes included 23 members of Bacillariophyta (diatoms), 10 Dinophyta (dinoflagellates), 1 Ascomycota (fungi), 1 Ochrophyta (a chrysophyte or golden algae), and 1 Sarcomastigophora ( Table 1). Each reference transcriptome could represent a range of more or less closely related phytoplankton present in the system. Here, when we refer to a reference transcriptome by name we are referring to the collection of phytoplankton with gene transcripts that mapped to that reference transcriptome. This approach does not imply a close evolutionary relationship between the reference and environmental phytoplankton. For brevity the strain name of each reference transcriptome is included only at first use.
Photosystem I (Supplementary Figure 3) and II (Supplementary Figure 4) expression patterns were dominated by dinoflagellates. While most phytoplankton associated with the reference transcriptomes up-regulated Photosystem I over the course of the microcosm experiment, and in comparisons between earlier and later experimental time-points, MMETSP1367 (Symbiodinium sp. C1) and MMETSP0503 (Heterocapsa rotundata SCCAP-K-0483) strongly downregulated the genes coding for Photosystem I at the start of the microcosms. A similar pattern was observed for Photosystem II, though MMETSP0796 (Pyrodinium bahamense RCC2770) and MMETSP0228 (Protoceratium reticulatum CCCM535 = CCMP1889) had a strong transcriptional response, and generally up-regulated Photosystem II in response to conditions associated with higher light levels. It should be noted that here we have retained the MMETSP nomenclature for the Symbiodinium reference transcriptomes, however, Symbiodinium C1 is now a member of the genus Cladocopium (LaJeunesse et al., 2018).
Dinoflagellates also dominated the expression pattern of genes coding for RUBISCO (Supplementary Figure 5). MMETSP1036 (Azadinium spinosum 3D9), MMETSP0258 (Amphidinium carterae CCMP1314), and Symbiodinium sp. C1 strongly upregulated these genes in response to conditions of higher light. A large number of diatoms also significantly up-or downregulated the genes coding for RUBISCO over the course of the experiment, albeit at lower levels than the dinoflagellates.
The total number of differentially expressed genes for a given reference and comparison ranged between 1 and 328 (Figure 3). We consider this value as the response complexity of the reference genome/transcriptome for the point of comparison. Hierarchical clustering of response complexity for the eukaryotic reference transcriptomes identified 19 significant clusters according to a simprof analysis. Three reference transcriptomes stood out as having high response complexity; F. kerguelensis, C. pennatum, and T. rotula, with the latter two forming a statistically significant cluster. These reference transcriptomes also had the strongest DNA polymerase signal across the microcosm portion of the experiment. The greatest response complexity was observed for F. kerguelensis between Ice Station day 5 (November 15) and Station B day 7 (November 17), and between Ice Station day 5 and Station B day 27 (December 7).
In contrast to response complexity we defined response strength as the summed fold-change for all differentially expressed genes for a given reference and comparison (Figure 4). Hierarchical clustering of response strength identified 23 significant clusters according to a simprof analysis, with F. kerguelensis and C. pennatum responding most strongly to changing conditions. To identify which references responded similarly to changing conditions we considered the magnitude of each sample comparison in PC1. principal component 1 accounted for between 18.0 and 50.3% of variance across the points of comparison for all reference transcriptomes ( Table 1; mean = 29.0%, SD = 9.2%). Using the magnitude of each sample comparison in PC1, we identified 23 distinct clusters of reference transcriptomes when organized by response to environmental gradients (normalized magnitude of points of comparison in PC1) (Supplementary Figure 6). Although they did not form a statistically significant cluster, the 10 dinoflagellate references formed a coherent group. These references showed a strong positive expression response for the Ice Station day 20 samples in comparison to the microcosms, and a strong negative expression response for the Station B day 7 samples when compared to the day 8 microcosms (November 18). It is important to recognize that the magnitude in PC1 for a given sample comparison only considers the degree of response according to log 2 foldchange, it does not imply that similar genes are being up-or down-regulated. The three reference transcriptomes with the largest response strength (F. kerguelensis and C. pennatum) and complexity (F. kerguelensis, C. pennatum, and T. rotula) clustered separately from one another in the PC1 magnitude analysis, though C. pennatum clustered with MMETSP0139 (Pseudonitzschia australis 1024910AB), and F. kerguelensis clustered with MMETSP0105 (Acanthoeca sp. 10tr).
To take a closer look at the three reference transcriptomes that exhibited the highest response complexity (T. rotula, F. kerguelensis, and C. pennatum) we considered the transcripts for each reference with the highest cos 2 values for PC1 (Figure 5). These transcripts were presumed to contribute strongest to the transcriptional response described by PC1. Each of these FIGURE 3 | Transcriptional response complexity. Transcriptional response complexity was determined as the number of genes in each reference transcriptomes determined to be differentially expressed between two points of comparison. The dendrogram at the top of the plot indicates similarity according to the number of unique genes differentially expressed between points of comparison but does not consider the identify of those genes. Significant clusters are indicated by alternating light and dark green color.

Lipid Distribution and Correlation With Metatranscriptomes
Triacylglycerols were the most abundant lipid class, followed by FFA (Figure 6 and Supplementary Figures 7, 8). TAGs rapidly increased in the microcosms, from 1847.3 pM at T 0 (November 10) to 10,252.6 pM at the end (day 8). Over the same period, TAGs also increased at the Ice Station, reaching 11,132.3 pM on day 15. TAG concentrations at Station B reached 11,327.9 pM that same day, and increased by more than fourfold to 53,391.4 pM on day 34. At T 0 the dominant TAGs were 48:2 and 48:1. For the microcosms, by day 8 the dominant TAG was 52:9, while at the end of the sampling period the dominant TAGs at Station B were 50:6, 52:6, and 52:7 (Supplementary Figure 7).
Free fatty acids were the next most abundant lipid class (Figure 6) and exhibited a more variable distribution (Supplementary Figure 8). FFAs increased only slightly from 4409.3 pMol at the start of the microcosm experiment to 4901.2 pMol at the end of the experiment (day 8). Over the same period FFA increased at the Ice Station to 5697.5 pMol on day 5 (November 15), but then decreased to 3802.2 on day 15. FFA initially declined at Station B, from 7509.9 pMol on day 6 to 6233.1 pMol on day 15, however, FFA increased dramatically at Station B after day 15, ending at 36,632.8 pMol on day 34. Initially the dominant FFA was 18:0, however, 20:5 was the most abundant FFA in the microcosms by day 8, and this FFA was also dominant at Station B after day 22 (Supplementary Figure 8).
To look for correlations between changes in lipid concentration ( lipid) and gene expression, we evaluated Pearson's r for lipid and PC1 for all reference transcriptomes ( Figure 7A). Mean variance in lipid explained by PC1 ranged from 10.41% for C. pennatum and 0.53% for T. rotula GSO102. C. pennatum and Aulacoseira subarctica yielded the strongest correlations between PC1 and lipid (Figures 7C-F), ranging from r = 0.57 for MGDG 24:0 ( Figure 7C) to r = −0.71 for DGCC 37:0 ( Figure 7D). Clustering the reference transcriptomes by FIGURE 4 | Transcriptional response strength. Transcriptional response strength was determined as the summed, normalized log 2 fold-change for all significantly differentially expressed genes in each reference transcriptome between two points of comparison. The dendrogram at the top of the plot indicates similarity according to the total number of transcripts associated with differentially expressed genes between points of comparison but does not consider the identify of those genes. Significant clusters are indicated by alternating light and dark green color. r yielded two groups. One group, exemplified by P. australis 1024910AB, F. kerguelensis, and C. pennatum, was characterized by largely negative correlations with PC1, except for FFA and select DGCC lipids ( Figure 7A). The other group was exemplified by MMETSP1362 (Leptocylindrus danicus CCMP1856), MMETSP1057 (Cyclotella meneghiniana CCMP338), and Aulacoseira subarctica CCAP1002-5. These had largely positive correlations between lipid and PC1, and weak or negative correlations between FFA and PC1.

DISCUSSION
Dynamic light conditions are associated with the winter to spring seasonal transition in the coastal Antarctic. Superimposed on the steady increase in irradiance associated with solar declination and day length are variable-term variations in irradiance caused by sea ice cover. We observed that the development of the spring phytoplankton bloom was significantly impeded by sea ice cover at our ice station, with only a slight increase in chlorophyll a over the observational period. Nonetheless, the concentration of TAGs increased at the ice station from day 0 to day 15, consistent with the downregulation of the gene coding for triacylglycerol lipase (EC 3.1.1.3) in F. kerguelensis and increasing PP. Combined with the up-regulation of genes coding for DNA polymerase, this suggests that F. kerguelensis was acclimated to the extreme low light conditions present and actively undergoing PP at the Ice Station throughout the experimental period.
This early transcriptional activity suggests a diverse lipid utilization strategy by the phytoplankton associated with the F. kerguelensis transcriptome. F. kerguelensis was by far the most transcriptionally active reference for the gene coding triacylglycerol lipase. Although the gene was generally downregulated from earlier to later season samples, and from the Ice Station to Station B, it was up-regulated in the microcosms on days 1, 3, and 8 in comparison with day 0. One possible explanation is that F. kerguenlensis had become photoinhibited within 24 h of exposure to high light conditions, and reverted to lipid catabolism for energy. There is strong evidence for photoinhibition in the microcosms by day 1. Although F v /F m increased slightly before dropping on day 3, ρ, a measure of reaction center connectivity, dropped sharply.
Other members of the phytoplankton community responded quickly to the simulated removal of ice and release from light limitation. Within 192 h, chlorophyll a concentration had more than doubled, TAG lipid concentrations had reached parity with those observed at Station B the day prior (despite the up-regulation of triacylglycerol lipase by F. kerguelensis), and the transcriptional response was dominated by a mixed diatom community. Diatoms represented by the reference FIGURE 5 | Major transcriptional response associated with the first principal component (PC1) for Fragilariopsis kerguelensis, Thalassiosira rotula, and Corethron pennatum. The cos 2 metric is a measure of the contribution of each variable to a PC. The higher the cos 2 value, the greater the contribution of that variable. transcriptomes T. rotula, F. kerguelensis, and C. pennatum were observed to have the strongest and most complex transcriptional response to changing environmental conditions. The specific response, however, and underlying transcriptional patterns varied widely across these references. As seen by the magnitude in PC1 for these points of comparison (Figure 7B), C. pennatum responded strongest to the experimental conditions in the microcosms, with high values for the transition from Ice Station day −6 and day 0 samples to the microcosm day 8 and Station B day 16 samples. We consider the phytoplankton associated with the C. pennatum reference transcriptome as high light specialists capable of rapidly responding to a stable light regime. Based on the % variance in lipid concentration explained by PC1 (Figure 7B), C. pennatum was also the dominant driver of the composition of the lipidome over the course of the experiment. Overall, our analysis suggests that the lipidome and transcriptome were only weakly connected. This may reflect a more complex and transient transcriptional regulation of the lipidome that is not easily detected by correlation, or alternate modes of regulation.
In contrast to C. pennatum, F. kerguelensis down-regulated the genes coding for DNA polymerase between the microcosm day 1 and day 8 samples, but up-regulated them between the Ice Station day 0 and Station B samples. Similarly, F. kerguelensis had a neutral or down-regulated response to these conditions in genes coding for Photosystems I and II. This is consistent with the idea of F. kerguelensis-associated phytoplankton as lowlight specialists, capable of early-season growth below sea ice and within the marginal ice zone, but inhibited in the higher light regime represented by the microcosm experiments.
The phytoplankton associated with the T. rotula reference transcriptome may represent an intermediate case. Like F. kerguelensis, T. rotula responded strongly in comparisons between Ice Station day 0 and Station B days 27 and 34. Unlike F. kerguelensis, T. rotula showed an early positive growth response to the microcosm conditions through up-regulation of the genes coding for DNA polymerase. This regulation was not sustained over the course of the microcosm experiments, however, and could represent an effort to repair damage brought on by the rapid increase in light, or a quick transition from a preferential to inhibiting photon dose.
Transcripts associated with several dinoflagellates were observed over the course of the experiment. The dinoflagellate references exhibited similar transcriptional response strength and complexity, and low mean % variance explained by PC1 (Figure 7B), suggesting a weak connection to the lipidome. Although very few studies have evaluated the ecological role of dinoflagellates along the WAP, they are thought to form blooms (Schofield et al., 2017) and to play a significant role in the regional carbon budget (Lin et al., 2017). Dinoflagellates are well known for their unusual modes of transcriptional regulation, which can lead to a reduced transcriptional response to changing environmental conditions. Because our analysis considered only the log 2 fold-change of significantly differentially expressed genes, the role of dinoflagellates may be underrepresented in this study. Despite this potential issue, dinoflagellate associated transcripts for PSI, PSII, and RUBISCO dominated transcriptional responses to increased light. This suggests that they are not poised to take immediate advantage of available light, as the diatoms may be, perhaps because they are able to supplement their nutritional needs during low light conditions through mixotrophic grazing.
Our findings come with two important caveats. One is that the true composition of the community does not necessarily reflect the mapping of reads to our reference transcriptomes, however, many of the references mapped are expected members of the WAP phytoplankton assemblage. F. kerguelensis, C. pennatum, and T. rotula in particular have been observed in the region or in coastal Antarctica. T. rotula is highly cosmopolitan (Whittaker et al., 2012), while C. pennatum and F. kerguelensis have previously been observed in WAP phytoplankton blooms (Lange et al., 2007;Świło et al., 2016). It is important to note that the model sea ice diatom F. cylindrus, a known low-light specialist with broad distribution in the Antarctic, is not present in the MMETSP database. Thus, read recruitment to F. kerguelensis could represent the presence of F. cylindrus. Other species not present in our database may also have been present in our samples. To work around this we focused our analysis on the specific response of those taxa we were able to observe.
The second caveat is that transcriptional response to the day 8 dark control in our microcosm experiments is very similar to that of the day 8 treatment. The amount of screening needed to achieve a 1% irradiance level was determined experimentally with the PAR meter, however, it was not possible to verify this light level after the carboys had been screened. Thus we suspect light leakage resulted in a higher light level than anticipated. Although we do not know what range of light was represented by our treatment and control, the expected wide difference in irradiance yet similar response points to the broad niche defined by "normal" light levels.
Here we used a novel analytical approach to couple two highdimensional datasets. Although our findings are inferential, and further experiments will be required to confirm the changes in gene expression and lipid composition that we observed for different reference transcriptomes, we have identified phytoplankton responding to different light levels brought on by variable sea ice conditions. Putative low light and high light specialists had distinct transcriptional signatures, and impacts on the community lipidome. As sea ice cover along the western Antarctic Peninsula continues to change on annual and decadal scales, we expect these changes to impact the composition of the phytoplankton community, and the concentration and composition of the lipidome. Because of the ecological role played by lipids, which function as key sources of energy, as signaling molecules, and even toxins, we expect these changes to have wider ecosystem ramifications during the critical winter-spring seasonal transition.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: NCBI BioProject, PRJNA494856.

ETHICS STATEMENT
Written informed consent was obtained from the individuals for the publication of any potentially identifiable images or data included in this article.

AUTHOR CONTRIBUTIONS
JB and JC designed the study with input from HD, BV, and NC. JB, JC, and NC carried out the sampling with assistance from CH. DL, HF, and BV contributed the lipidomics analysis. JB carried out the metatranscriptomics analysis and drafted the initial version of the manuscript. RG and CH contributed discussions of protist physiology and enzyme functions. All authors edited the manuscript.

FUNDING
JB was supported by NSF-OPP 1641019, NSF-OPP 1846837, and the Simons Foundation Early Career Marine Microbial Investigator program. BV, DL and JC were supported by NSF (OPP-1543328 and OCE-1756254). CH was supported by NSF OCE-1355720. The Palmer LTER project is support by NSF-OPP 1440435. A small-scale Community Sequencing Project (CSP) award from the DOE Joint Genome Institute supported part of the sequencing effort.

ACKNOWLEDGMENTS
We would like to thank Ashley Goncalves and Chelsea Farischon for assistance with sampling at Palmer Station, and the Palmer Station logistics personnel for their dedication during the 2015-2016 season. Satellite imagery in support of this project (see Figure 1A) was kindly provided by the Polar Geospatial Center.