Marked spatiotemporal variations in small phytoplankton structure in contrasted waters of the Southern Ocean (Kerguelen area)

In the Southern Ocean, diatom blooms have attracted a lot of attention, while other small nonsilicified phytoplankton groups have been less studied. Here, small phytoplankton (< 20 μm, including small diatoms and nonsilicified small phytoplankton) are focused on in two contrasting areas: the productive Kerguelen plateau and its surrounding low productivity waters. To assess the diversity and spatial structuration of phytoplankton, discrete plankton samples (0–300 m layer) of two size fractions (< 20 and 20–100 μm) were analyzed with 18S rDNA amplicon sequencing in late summer. Phytoplankton seasonal succession was described using flow cytometry, pigments, and environmental data, from two previous cruises (during the onset and decline of the diatom bloom). In the mixed layer, small nonsilicified phytoplankton represented less than 10% of chlorophyll a (Chl a) during the onset and late diatom bloom on the plateau, but they increased on and off the plateau after the bloom (53–70% of Chl a). Phaeocystis antarctica was relatively abundant at all stations after the bloom, but other small phytoplanktonic groups featured marked differences on and off the plateau. Higher NH4+ concentrations on the plateau appeared to stimulate the presence of Micromonas, while Pelagophytes were enhanced off the plateau. A diverse assemblage of small diatoms was also promoted off the plateau, where silicate concentration was still high. Interestingly, P. antarctica represented up to 25% of all reads at 300 m depth off the plateau in the larger size fraction suggesting a significant contribution to carbon export through aggregation in low productive waters.

Small phytoplankton (< 20 μm) play key roles in the global carbon cycle and marine food webs (Marañón et al. 2001;Richardson and Jackson 2007;Uitz et al. 2010). Typically, small phytoplankton have an advantage over larger cells in nutrient uptake due to their higher surface to volume ratio (Chisholm 1992;Agawin et al. 2000;Irwin et al. 2006). The Southern Ocean (SO) is the earth's largest high nutrient low chlorophyll (HNLC) area where phytoplankton production is limited by iron despite high concentrations of macronutrients, such as nitrate, silicate, or phosphate (Martin 1990;de Baar et al. 1995). Phytoplankton in these low productive HNLC waters is typically composed of small cells (< 20 μm), including small nonsilicified phytoplankton and diatoms (Moore and Abbott 2000;Kopczynska et al. 2001;Uitz et al. 2009Uitz et al. , 2010Lasbleiz et al. 2014). Here, annual primary production of small phytoplankton collectively exceeds the large phytoplankton (herein defined as large-size cells, > 20 μm) by a factor of 2.7 (2.5 Gt C yr −1 and 0.9 Gt C yr −1 , respectively; Uitz et al. 2010). Despite their significant functional role, their diversity and spatiotemporal dynamics have remained overlooked in the SO (e.g., Weber and El-Sayed 1987;Wright et al. 2009;García-Muñoz et al. 2013), when large diatom blooms and the classical Antarctic food chain (diatoms-krill-whales) have received much more attention.
In contrast to HNLC low productive waters, discrete areas of the SO naturally, fertilized in iron and highly productive, are characterized by large diatoms bloom from spring to summer Korb et al. 2008;Pollard et al. 2009). This is the case south-east of the Kerguelen archipelago where large diatom blooms appear at this time and result in a chlorophyll-rich area of 45,000 km 2 , constrained by the bathymetry of the plateau ; Mongin et al. 2008). Upward fluxes of deep iron and major nutrients from the shallow continental shelf of the Kerguelen plateau (< 500 m) sustain these blooms Mongin et al. 2008;Park et al. 2008). The blooms over the plateau can be dominated either by small diatoms forming long chains (chain length > 100 μm, e.g., Pseudo-nitzschia, Chaetoceros) or large diatoms (Eucampia, Corethron). In contrast, small phytoplankton (mostly Phaeocystis and small diatoms) dominates in surrounding HNLC waters and production remains low throughout the year Kopczy nska et al. 1998;Armand et al. 2008;Uitz et al. 2009;Lasbleiz et al. 2016).
Community composition, temporal succession, and contribution to carbon export of diatoms during their blooming period around Kerguelen are well documented Rembauville et al. 2015Rembauville et al. , 2017Rembauville et al. , 2018Lasbleiz et al. 2016). Previous cruises KEOPS 1 and 2 (KErguelen Ocean and Plateau compared Study project) investigated phytoplankton composition and C-cycling during the onset and decline of the bloom. The carbon export efficiency during the onset and decline of the bloom was lower on the plateau than in the HNLC area (e.g., Christaki et al. 2014;Laurenceau-Cornec et al. 2015;Planchon et al. 2015) although the amount of carbon exported at 200 m was twofold higher on the plateau Planchon et al. 2015). At a broader scale, this inverse relationship between production and export seems to be a feature of the SO (Maiti et al. 2013;Le Moigne et al. 2016). This pattern could be explained by differences in phytoplankton communities (dominance of either small or large cells) and resulting trophic structures, grazing intensity, and/or microbial mineralization (Maiti et al. 2013;Laurenceau-Cornec et al. 2015;Le Moigne et al. 2016).
As chain-forming and large-diatom blooms occur during relatively short periods, it has been hypothesized that smallsize cells are an essential element of Antarctic food webs, especially during winter and periods preceding blooms (Detmer and Bathmann 1997). Satellite data and sampling has indicated that phytoplankton communities on the plateau switch to an ecosystem dominated by small phytoplankton before (Rembauville et al. 2017), and also likely, after the bloom (Penna et al. 2018). As such, the plateau of Kerguelen could also include a persistent and functionally important small phytoplankton community, on which diatom blooms superimpose themselves (Smetacek et al. 1990).
Despite their ecological importance, little is known about the diversity, spatiotemporal dynamics, and environmental drivers of small phytoplankton in the SO. Phaeocystis antarctica has been identified as a putative key player in the absence of diatom blooms (Froneman et al. 2004;Hashihama et al. 2008;Iida and Odate 2014;Schulz et al. 2018). Early molecular studies specifically targeting Antarctic small plankton used technologies such as clone libraries or molecular fingerprinting techniques (Díez et al. 2001;Díez et al. 2004;Gast et al. 2004;Piquet et al. 2011). These technologies were unable to depict the full microbial diversity that is dominated by low-abundant taxa (Pedròs-Alio 2007). High-throughput sequencing of size fractionated samples allowed further description of the diversity of small microbial eukaryotes Clarke and Deagle 2018). However, summer to autumn transition (March) and winter have not been investigated thus far, although these are the periods when the ecosystem most likely relies on primary production by small phytoplankton. Moreover, phytoplankton seasonal dynamics around the Kerguelen Islands has not been fully described yet.
The MOBYDICK (Marine Ecosystem Biodiversity and Dynamics of Carbon) cruise took place in late February and March 2018, a month after the diatom bloom ended. MOBYDICK's two main objectives were: to trace carbon from its initial fixation at the surface to its final export toward the ocean interior, and to explore how diversity influenced the carbon cycle in contrasted marine ecosystems on and off the iron enriched waters of the Kerguelen plateau. The present study focuses on the composition, seasonal variability, and ecology of small phytoplankton on and off the plateau. Molecular diversity of microbial eukaryotes after the bloom (MOBYDICK) is described through 18S rDNA amplicon Illumina Mi-Seq sequencing of small and largecells size fractions (< 20 and 20-100 μm). Pigment and flow cytometry data from two previous cruises at the early and late stage of the bloom (KEOPS2 and KEOPS 1, respectively) were also used to complete the seasonal succession of phytoplankton on and off the plateau of Kerguelen. Environmental factors likely governing the balance between small phytoplankton vs. large diatoms throughout the season and potential ecological implications for carbon export are also identified.

Sampling strategy overview
Seawater was collected onboard the R/V Marion Dufresne around the plateau of Kerguelen Islands during the MOBYDICK cruise after the diatom bloom (26 February 2018-18 March 2018. We combined samples from the MOBYDICK cruise with data collected during past cruises undertaken at different seasons to describe the seasonal succession. To do so, we restricted the pigment and cytometry data to a similar depth horizon (upper 50 m) and to stations presenting either the same location on the plateau and reference station in the HNLC area off the plateau. KEOPS2 cruise took place during austral spring (15 October 2011-20 November 2011, while KEOPS1 occurred during summer (19 January 2005-13 February 2005 (Table 1). Plateau Sta. A3 of KEOPS1 and 2 corresponds to MOBYDICK Sta. M2 (Fig. 1). This station experiences natural iron enrichment from the plateau resulting in diatom blooms during spring and early summer Lasbleiz et al. 2016). Off-plateau stations M1, M3, and M4 (MOBYDICK) were expected to display HNLC characteristics. Sta. R-2 (KEOPS2) and Sta. C11 (KEOPS1) were the HNLC reference stations (Quéguiner et al. 2007;Cavagna et al. 2015), which will be considered here for the seasonal succession ( Fig. 1; Table 1).

Environmental parameters
For KEOPS 1 and 2, environmental parameters were described in Mosseri et al. (2008) and Cavagna et al. (2015). For MOBYDICK, temperature, dissolved oxygen, and salinity measurements were collected at all CTD casts using a SeaBird 911-plus. Ammonium was measured by fluorometry (Holmes et al. 1999).
For other dissolved inorganic nutrients (PO 3− 4 , Si(OH) 4 , NO − 2 , NO − 3 ), samples were collected in 125 mL polyethylene bottles. The samples were then filtered through 0.45 μm acetate cellulose filters and preserved with 100 μL HgCl 2 (4 g L −1 ) in 20 mL. The samples were then stored in the dark at room temperature until analysis in the laboratory with a segmented flow analyzer (Skalar) equipped with colorimetric detection using methods described in Aminot and Kérouel (2007).

Sample collection for DNA extraction and molecular biology
During the MOBYDICK cruise, four different sites were visited. M2, located on the plateau, was sampled three times during the cruise with about a 10-d interval. Off the plateau, M3 and M4 were sampled twice with a 2-weeks interval, and M1 only once (Table 2). Seawater samples were collected with 12-liter Niskin bottles at all visited stations at four depths (15, 60, 125, and 300 m). The first two depths (15 and 60 m) were chosen to correspond approximately to the surface and bottom of the mixed layer (ML), respectively. Water at 300 m was characteristic of deep nutrient-rich winter water (Park et al. 2008). Water at 125 m corresponded to the "transition layer," defined as the water mass at the interface between the deep stratified ocean interior and the highly turbulent surface ML (Johnston and Rudnick 2009).
Sampled seawater was first prefiltered using a 100 μm nylon mesh (Millipore, U.S.A.) in order to remove Metazoa and large particles. Ten liters of water from each depth were then filtered successively through 20 and 0.2 μm pore-size membranes (Millipore) for microbial eukaryote taxonomic composition analyses of large (100-20 μm) and small sizefractions (< 20 μm). This resulted in 64 samples (eight visits at the different stations, four depths, two size fractions). Filters were directly frozen and stored at −80 C for 18S rDNA  amplicon Illumina Mi-Seq sequencing. Each filter was cut into small pieces before DNA extraction with PowerSoil DNA Isolation Kit (QIAGEN, Germany) following standard manufacturer's protocol. DNA was eluted with 100 μL AE buffer.

Polymerase chain reaction and Illumina sequencing
To describe protist diversity, the V4 hypervariable region of the 18S rRNA gene was amplified using the primers EK-565F (5 0 -GCAGTTAAAAAGCTCGTAGT) and UNonMet (5 0 -TTTAA GTTTCAGCCTTGCG) biased against Metazoa (Bower et al. 2004). Different molecular identifiers of 10 bp were added to both forward and reverse primers to tag amplicons and allow to differentiate them after sequencing. Polymerase chain reaction (PCR) mixtures comprised 1-2 μL of DNA, 12.5 μL of DreamTaq Green PCR Master Mix (Thermo Fisher Scientific, U.S.A.), and 1 μL of each primer (10 μmol L −1 ) in a total volume of 25 μL. PCR conditions included an initial step of denaturation at 94 C for 3 min, 25 cycles of denaturation at 94 C for 30 s, annealing at 58 C for 45 s, and extension at 72 C for 90 s, and a final step of extension at 72 C for 10 min. PCR products from at least five different PCR reactions of each sample were then pooled together and purified using the QIAquick PCR purification kit (QIAGEN, Germany), according to the manufacturer's instructions. DNA concentrations were calculated on a Qubit fluorometer with the dsDNA High Sensitivity Assay Kit (Life Technologies Corp., U.S.A.) and identical concentrations of each sample's amplicons were pooled.

Demultiplexing, quality filtering, and taxonomic affiliation
Paired-end sequences were imported and demultiplexed in Qiime (Caporaso et al. 2010) based on each sample's 10 bp molecular identifier with the functions extract_barcodes.py (Qiime1) and demux emp-paired (Qiime 2-2018.8). Demultiplexed sequences without primers and barcodes were further processed in R-software (R Core Team 2018) using the DADA2 package v.1.10.1 (Callahan et al. 2016) to define amplicon sequence variants (ASVs). Sequences were quality filtered (maximum expected error = 6), trimmed (forward and reverse reads at 270 nt) prior to inferring ASVs and removing chimeric sequences with the DADA2 algorithm. Taxonomy was assigned for each ASV to the best taxonomic level using the RDP naive Bayesian classifier implemented in DADA2 (Wang et al. 2007) in combination with PR 2 database v.4.11.0 (Guillou et al. 2013), a curated database implementing EukRef (del Campo et al. 2018). Taxonomy assigned this way was cross-checked with Sintax algorithm in USEARCH v11 (Edgar 2013). When taxonomy assignment differed between the two algorithms, taxonomy was compared to NCBI blast and/or limited to the highest taxonomic level where both algorithms agreed. All ASVs sequences were aligned in Geneious Prime v.2019 using MAFFT v7.388 plugin (Katoh and Standley 2013) parameterized for FFT-NS-2 analysis to Table 2. Description of MOBYDICK stations. The depth of the mixed layer (ZML) is based on a difference in sigma of 0.02 compared to surface value. The mean ZML is calculated based on all CTD casts performed during the occupation of the stations. Temperature, nutrients, and Chl a are mean values of all measures sampled within the ML (2-3 depths depending on the station). The depth of the euphotic layer (Ze) corresponds to the depth where light intensity was at least 1% of incident light at surface during the CTD cast of the sampling for metabarcoding.

Station Date
Depth (m) deal with the large number of sequences. A phylogenetic tree was then built based on this alignment with FastTree 2.1.5 plugin using default parameters (Price et al. 2009). All data for this study are accessible in the MOBYDICK database (http:// www.obs-vlfr.fr/proof/php/mobydick/mobydick.php#SA). Raw sequencing files in fastq format, as well as ASVs, taxonomy and metadata tables are available at http://www.obs-vlfr.fr/ proof/ftpfree/mobydick/db/DATA/PAR_2251/.

Statistical analysis
Alpha diversity analyses were conducted in R-software version 3.5.1 (R Core Team 2018) on ASVs with phyloseq v.1.16 (McMurdie and Holmes 2013) and microbiomeSeq (Ssekagiri et al. 2018) after rarefying read depths based on sample with the lowest read count. Observed richness and Pielou's evenness (i.e., J = H 0 /ln [S] where H 0 is Shannon Weiner diversity index and S is the total number of species in a sample indexes) were compared between size fractions and locations with ANOVA, followed by pairwise tests using Tukey multiple comparisons (Tukey's honestly significant difference test) with 95% confidence interval.
To illustrate community composition differences between samples, phyloseq was used to plot PCoA based on weighted UniFrac distances, that considers the phylogenetic distance between two sample's composition and ASVs relative abundances (Lozupone et al. 2011). Cluster analysis was conducted based on the wUniFrac distance matrix using the hclust function in R-software and Ward classification method (Murtagh and Legendre 2014). Homogeneity of dispersion (variance) within group was calculated with function betadisperser. ANOVA-like test (permutest) was used to determine if the variances differed by groups.
Heatmaps of the 30 most abundant genera were generated using "Ampvis2" R-package (Andersen et al. 2018) after pooling ASVs belonging to the same genus. To test for statistical differences in community composition according to a variable of interest (size fraction, location of the station on or off the plateau), permutational analysis of variance (PERMANOVA) implemented in the function adonis from the "vegan" Rpackage (Oksanen et al. 2017) was run on weighted UniFrac distances with 1000 permutations. To identify genera differing in abundance between plateau and off-plateau surface samples (10 and 60 m), un-normalized table of ASVs, pooled by genus when possible, representing over 1% of the reads, were transformed using DESeq2 (v1.22.2) (Love et al. 2014). The same package was used to perform the differential analysis. All genera that were significantly different in abundance between plateau and off-plateau surface samples (p value, adjusted for multiple comparisons using the Benjamini-Hochberg false discovery rate procedure, less than 0.05) were visualized by plotting Log2 fold changes in abundance on and off the plateau in ggplot2 v 2.2.1 (Wickham 2011(Wickham , 2016. To explore the influence of abiotic factors on community composition, a canonical correspondence analysis (CCA) was done with the "vegan" R-package on ASVs relative abundances, pooled by genus when this taxonomic level was available. The same analysis was also run only considering ASVs accounting for over 1% of the reads. Significant environmental variables were selected using function ordistep with 1000 permutations, which performs automatic stepwise model building for constrained ordination methods in both directions (forward and backward) based on pseudo-Akaike Information Criterion. No autocorrelation for selected environmental variables was detected using the vif.cca function. Significance of the final model and the different axis was checked using ANOVA (1000 permutations) and the CCA was plotted using ggplot2.

Pigment measurements and CHEMTAX analysis
Pigment extractions for KEOPS 1 and 2 cruises are described in Uitz et al. (2009) and Lasbleiz et al. (2014). In brief, from 1 to 2.2 L, based on particle concentrations, were filtered onto Whatman GF/F filters. For the MOBYDICK cruise, 2.32 L of water was sampled and filtered onto Whatman GF/F filters at each station and depth. Filters were then flash-frozen in liquid nitrogen and stored at −80 C. For all cruises, pigment determination was done using high-performance liquid chromatography (HPLC), following the method of Ras et al. (2008), adapted from Van Heukelem and Thomas (2001). To perform CHEMTAX analysis, HPLC data from all sampling points in the first 50 m were considered. CHEMTAX v1.95  was used to assess the contribution of seven different taxonomic groups to total chlorophyll a (Chl a): chlorophytes, prasinophytes, cyanobacteria (most likely representative of Synechococcus, as Prochlorococcus is absent at this latitude), cryptophytes, diatoms, dinoflagellates (with peridinin), and haptophytes (Phaeocystis like). This software aims at determining the contribution of different photosynthetic groups to total Chl a biomass based on each group's characteristic accessory pigment to Chl a ratio. Pigment composition characteristic of the different groups was assessed using 11 pigments: chlorophyll c3 (Chl c3), perdinin (Peri), fucoxanthin (Fuco), prasinoxanthin (Pras), 19 0 -Hexanoyloxyfucoxanthin (Hexfuco), zeaxanthin (Zea), alloxanthin (Allo), lutein (Lut), and chlorophyll b (Chl b). An initial matrix of pigment to Chl a ratios for these seven taxonomic groups was build based on previous chemo-taxonomic studies of phytoplankton in the SO (Wright et al. 2010;van Leeuwe et al. 2014) (Supporting Information Table S1).
Samples were first clustered based on their pigment : Chl a ratios to form homogeneous bins using R-software, Hierarchical clustering on principle components (HCPC), implemented in the package FactoMineR (Lê et al. 2008). Then, pigment: Chl a ratios were adjusted for each bin using 60 randomized ratio matrix varying by up to AE 35% of the initial ratio matrix to avoid any bias linked to the ratios chosen from the literature. Taxa composition was then inferred for each sample of every cluster based on the average ratios from the six best runs, associated with the lowest root mean square of the residuals and hence the lowest amount of unexplained pigment. It was then finally ensured that ratios of each characteristic pigment : Chl a obtained in the final ratio matrix for each taxonomic group were still in the range accepted in literature sources used in .

Abundance of autotrophic pico-and nanoplankton
Abundance of pico-and nanoautotrophs was determined by flow cytometry. For all cruises, 4.5 mL of seawater were fixed with glutaraldehyde (1% final concentration). Fixed samples were stored for 30 min at 4 C, flash frozen in liquid nitrogen, and stored at −80 C until analysis (Marie et al. 1999). Counts were performed with BD flow cytometers (FACSCalibur for KEOPS 1, and FACSCanto for KEOPS2 and MOBYDICK cruises) equipped with blue (argon 488 nm) and red (633 nm) lasers. Phytoplankton cells were analyzed according to their natural fluorescence (Chl a and phycoerythrin). Approximately 0.75 mL of sample were analyzed at high speed (150 μL min −1 ) for 5 min as described in Christaki et al. (2014).

Brief description of the physicochemical parameters (MOBYDICK)
During MOBYDICK, phosphate and nitrogen were abundant in the ML at all stations (Table 2). On the plateau (M2), silicic acid concentrations were low (< 2 μmol L −1 ), especially during the first two visits, but doubled at the last visit after a storm (10 March 2018). M2 was also characterized by the highest ammonium concentrations (0.75-1.1 μmol L −1 ). Offplateau, Sta. M1, M3, and M4, presented different physicochemical characteristics for the ML in relation with their location North or South of the polar front, in subantarctic or Antarctic waters, respectively (Fig. 1c). Silicate concentrations were lower in subantarctic waters at M3 (minimum of 2.3 μmol L −1 ) than in Antarctic waters at M4 and M1 (from 4.36 to 8.38 μmol L −1 ). M1 and M4 were located in dynamic water masses near the polar front and were weakly stratified. Consequently, the depth of the ML was shallowest at M1 (27 m) and varied considerably between the two visits at M4 (49 and 87 m). At M2 and M3, ML depth varied between 60 and 80 m and also deepened after the storm ( Table 2). Depth of the euphotic layer (Ze) was always shallower at M2 (45 m) than off the plateau (56-75 m). Lowest temperatures were observed at M4 (4.4 C), south of the plateau, and highest temperatures in subantarctic waters at M3 (5.6 C).

Diversity and community composition of microbial eukaryotes after the diatom bloom
After DADA2 quality filtering, 8328 ASVs were retrieved from a total of 3,366,966 reads in 64 samples. ASVs not assigned to Eukaryotes or assigned to Metazoa were removed leaving a total of 8164 ASVs and 3,356,832 reads. After removing singletons, 5831 ASVs and 3,355,898 reads were kept for further analysis. One sample with less than 5000 reads was excluded from the analysis (M2-3, 60 m, < 20 μm).
The number of observed ASVs per sample ranged from 116 (M2-1, 125 m, < 20 μm) to 648 (M4-2, 125 m, < 20 μm). The mean observed richness across all stations and depths was significantly higher in the small than in the large size fraction. Richness was not different on and off the plateau for each size fraction. Within the small size fraction, richness differed according to depth and was overall higher at 125 and 300 m than at 10 m (Fig. 2). Highest richness was observed off the plateau at M4 (125 m) with over 600 different ASVs sequenced at each visit and lowest richness on the plateau at M2 (125 m) with only 116 ASVs (Fig. 2). No differences in evenness between the two size fractions could be identified, although it was overall higher off the plateau. Within the small size fraction, evenness was significantly higher at 125 m than at 10 m (Supporting Information Fig. S1). Although all taxonomic orders were found in the two size fractions, small and large size fraction community composition differed significantly at all depths (PERMANOVA, p < 0.001).
Surface samples of the < 20 μm size fraction were dominated either by parasitic Syndiniales (55-78% at M2-1, M2-2, M3-1) or Haptophytes (Phaeocystis) (10-46%; Fig. 3). Small diatoms (in the < 20 μm fraction) were more abundant at Sta. M1 and M4 off-plateau, where they accounted for 21% and up to 30% of the reads, respectively, but represented less than 10% of the reads at M2 and M3. Other phytoplankton, such as Fig. 3. Relative abundance of high-level taxonomic groups at each station and depth for both size fractions. Bacillariophyta have been distinguished from other Ochrophyta and Syndiniales from core dinoflagellates Dinophyceae (Dinoflagellata) to highlight these two taxonomic groups. "Other groups" include rare taxonomic groups representing together less than 0.1% of the reads. Numbers close to each station refer to the order of visits when stations were visited several times (see Table 2).

Ochrophyta (Pelagophyceae)
and Chlorophyta (Mamiellophyceae and Prasinococcales), were also present in the small size fraction, but accounted for a low proportion of reads at each station (less than 10%). Picograzers such as Telonemia and Stramenopiles (MAST and MOCH; Fig. 3) were also consistently observed at all surface stations in reduced proportions (less than 10% of the reads). An increase of Dinophyceae and Radiolaria was observed at 125 and 300 m depth. Syndiniales, Stramenopiles, and Haptophytes were also present ( Fig. 3). At 300 m, the small size fraction was largely dominated by Radiolaria (31-83%), and Dinophyceae (7-36%).
Surface samples (10 and 60 m) of the large size fraction were dominated by Bacillariophyta (up to 80% at M2-3) or Ciliophora (up to 65% at M3-3, 60 m; Fig. 3). Diatoms were still present (14-91%) deeper at 125 m and contributed up to 91% of the reads at Sta. M2-1 with a marked contribution of Odontella and Eucampia (Supporting Information Fig. S2). The relative abundance of Dinophyceae and Radiolaria increased at this depth as in the small size fraction. Finally, the large size fraction consisted mainly of an assemblage of dinoflagellates, Radiolaria, Syndiniales, and Bacillariophyta at 300 m depth. Small phytoplanktonic Haptophyta (Phaeocystis) represented up to 25% of the reads at 300 m at M3-1 and M4-1 (Fig. 3, Supporting Information Fig. S2) when they were almost absent in this size-fraction in the upper layers.
When looking only at autotrophic taxa of the small size fraction in surface samples (10 and 60 m), phytoplankton communities were different on and off the plateau (PERMANOVA; p = 0.04). Phaeocystis was present in all surface samples, but a few common autotrophic taxa were enhanced in these distinct regions (DESeq2 analysis; Fig. 4a). Pelagomonas and unidentified Pelagophyceae were more abundant off the plateau in contrast with the prasinophyte Micromonas (Mamiellophyceae) that was more abundant on the plateau. Diatoms belonging to the genera Fragilariopsis, Pseudo-nitzschia, Dactyliosolen, Thalassiosira, and Chaetoceros were enriched off the plateau (Fig. 4a). Fragilariopsis, Pseudo-nitzschia, Dactyliosolen, Thalassiosira, and Chaetoceros were also enriched off the plateau in the large size fraction, whereas reads of the diatoms Corethron and Guinardia were more abundant on the plateau (Fig. 4b).

Relation between environmental variables and community composition in late summer
CA analyses of surface euphotic layer samples and main environmental variables emphasized a clear distinction between on and off-plateau communities (Fig. 5). Phaeocystis, which was abundant in surface waters of all stations, appeared in the middle of the CCA, as it was not related to any particular environmental parameter. Among the environmental variables tested, ammonium, phosphate, temperature, and silicic acid were significant predictors of community composition within the small size fraction when considering all ASVs and/or only the most common ASVs (automatic stepwise model building analysis, p < 0.01). None of these variables showed significant autocorrelation and they explained 45.8% of the variance of the entire community composition within the small size fraction (CCA1 = 15.1%; CCA2 = 12.4%; CCA3 = 10.3%; CCA4 = 8.0%). When plotting on the CCA small phytoplankton taxa that were differentially abundant on and off the plateau (cf. Fig. 4a), plateau station M2 was characterized by higher ammonium concentrations along with higher abundances of the prasinophyte Micromonas (Mamiellaceae family). Diatoms such as Fragilariopsis, Thalassiosira, and Pseudo-nitzschia were associated off the plateau Fig. 4. Phytoplankton genera presenting differential read abundances between surface samples collected on and off plateau after Deseq2 analysis (p < 0.05) for small (a) and large size-fractions (b). Only most abundant genera (> 0.1% of total reads and base mean after Deseq2 normalization superior to 100 reads) are represented. A negative log2 fold change indicates genera that are relatively more abundant off the plateau, while a positive log2 fold change indicates genera relatively more abundant on the plateau. with M1 and M4 as well as the highest concentrations of silicic acid and phosphate (Fig. 5).

Seasonal changes in phytoplankton assemblages
CHEMTAX analyses highlighted that phytoplankton communities were contrasted on and off the plateau at the onset and the decline of the Kerguelen bloom (KEOPS 1 and 2, respectively). In late summer (MOBYDICK), small phytoplanktonprymnesiophytes particularly-was important over the entire area (Fig. 6). On the plateau, contribution of the different phytoplankton groups varied markedly over seasons (Fig. 6a,c). Diatoms dominated phytoplankton blooms in spring and summer (KEOPS 2 and 1), while small flagellates represented higher proportions of the phytoplankton community in late summer (MOBYDICK, Fig. 6c). During spring and summer blooms (KEOPS 2 and 1), Chl a was sixfold higher on the plateau than off-plateau reference stations (1.21 AE 0.26 μg L −1 and 0.19 AE 0.05 μg L −1 , respectively) and diatom pigments represented over 90% of Chl a (Fig. 6a,c). Concomitantly, only prymnesiophytes remained at low background levels within small phytoplankton taxa. In late summer (MOBYDICK), Chl a on the plateau dropped to values similar to the ones found off the plateau (0.3 AE 0.02 μg L −1 at M2-1 and M2-2; Fig. 6a,b). These low Chl a concentrations were mainly due to a strong (22-fold) decrease of diatoms' contribution to Chl a (from up to 1.8 μg L −1 in November to 0.08 μg L −1 end of February). As a consequence of the diatom decline, small nonsilicified phytoplankton's contribution strongly increased on the plateau after the bloom, with prymnesiophytes and prasinophytes accounting for 53-58% of Chl a (Fig. 6a,c). The contribution of prasinophytes notably increased from 5.7% to 16.6% between the first and last visit at M2 during late summer (MOBYDICK). They also contributed to 8.9% to Chl a biomass at the very early stage of the bloom (KEOPS2, first visit). Total abundances of small phytoplanktonic cells after the bloom (3200-8300 cells mL −1 ) were also much higher after the bloom than during the bloom (700-2000 cells L −1 ).
Off the plateau, Chl a was low during the three cruises (0.20 AE 0.05 μg L −1 ; Fig. 6b). Prymnesiophytes, followed by diatoms, were the most represented groups in pigment data, with average relative contribution to Chl a of 51.6% and 39.3%, respectively. Pigments characteristic of small nonsilicified phytoplankton dominated at all off-plateau reference stations (Fig. 6b), with the exception of Sta. C11-1 . Black marks are representative of the community composition of each sample, labeled by station and depth. The different genera (or ASVs when this taxonomic level was not available) are represented by dots colored by division. Only the genera/ASVs representing more than 1% of total reads were plotted, to not overload the chart. Only phytoplankton genera considered as differentially abundant between stations on and off plateau according to DESeq2 were labeled. Phaeocystis position in the middle of the CCA plot indicated its presence at all stations.
(KEOPS1 HNLC reference station) where diatom pigments accounted for 52.1% of Chl a. Prymnesiophytes were the most important group and represented up to 70% of Chl a (M3-1; Fig. 6d). Other groups were minor contributors to Chl a, with the exception of prasinophytes at R2 (20.7%). As a consequence of these contrasted phytoplankton communities on and off the plateau, Chl a was significantly correlated with the abundance of small phytoplankton only off the plateau (R 2 = 0.89; p < 0.0001; Supporting Information Fig. S3).

Discussion
The MOBYDICK cruise provided first insights into the phytoplankton community around Kerguelen after the summer bloom. Data from this study highlight the dominance of small-sized phytoplankton off the plateau (HNLC) throughout the year and document for the first time their importance in phytoplankton communities in surface waters on the iron-fertilized plateau. The intense diatom blooms on the Kerguelen plateau are thus followed by an increase of small phytoplankton in terms of absolute and relative contribution to Chl a (Fig. 6a,b), displaying a seasonal community succession observed in other regions of the SO (e.g., Moline et al. 1996;Fiala et al. 1998;Wright et al. 2010). This study has further highlighted that small phytoplankton communities differed on and off the plateau in relation with contrasting nutrients concentrations (ammonium and silicic acid) after the bloom (Fig. 7).

Diversity of pico-and nanophytoplankton and their response to environmental variables
Both CHEMTAX and sequencing analyses suggested that Prymnesiophytes, dominated by Phaeocystis antarctica, were the most abundant phytoplankton group in late summer. This phytoplanktonic group represented up to 53% and 70% of the Chl a on and off the plateau, respectively (Fig. 6c,d). Contrasted environmental conditions on and off the plateau did not seem to affect their distribution (Fig. 5). The ability of P. antarctica to thrive in HNLC waters has been explained by its large phenotypical plasticity, being able to reduce its cell volume and adapt its photo-physiology under Fe-limitation (Alderkamp et al. 2012;Koch et al. 2019). Iron replete conditions are known to trigger Phaeocystis' colony formation Bender et al. 2018). Colony formation and size increase could in turn serve as defense mechanisms against small grazers such as copepods and ciliates (Schoemann et al. 2005;Tang et al. 2008). As a consequence, previous studies conducted in the SO observed that Phaeocystis population was mostly composed of single cells before and after the phytoplankton bloom. In spring, optimal light and iron concentrations lead to the formation of colonies (Smith et al. 2003). At the onset of the spring bloom (KEOPS2, November), when surface dissolved iron concentration was at its highest Bowie et al. 2015), Phaeocystis was observed on and off the plateau in small colonies (Georges et al. 2014). This feature probably made them more resistant to grazing by protists than free cells like prasinophytes, as shown by the strong decrease of prasinophytes between the first two visits on the plateau during KEOPS2 (Fig. 6a,c). In late summer (MOBYDICK), Phaeocystis was only observed as small flagellated single cells (approximately 3 μm, electron microscopy observations, Supporting Information Fig. S4) on and off the plateau, in agreement with a potential iron limitation after the phytoplankton bloom (Smith et al. 2003).
While sequencing data indicated that P. antarctica was relatively abundant at all stations after the bloom, the structure of the other small phytoplanktonic groups featured marked differences on and off the plateau. Free Phaeocystis cells, Micromonas, and Pelagomonas are mostly grazed by ciliates, whose prey generally falls within the nanophytoplankton size class (Hansen et al. 1994). During MOBYDICK, ciliate communities showed similar structures and abundances on and off the plateau (Christaki, pers comm), suggesting that the observed spatial variability of small phytoplankton was most likely related to the availability of specific nutrients. High NH + 4 concentrations on the plateau (0.75-1.1 μmol L −1 ) coincided during MOBYDICK with an increase in relative abundance of Micromonas (Figs. 5,7). Increased ammonium concentrations on the plateau have previously been observed during the bloom decay (KEOPS1),  (Fig. 4a). Off the plateau, silicate concentrations were not limiting and promoted a diverse assemblage of diatoms (Fragilariopsis, Thalassiosira, Pseudo-nitzschia, Chaetoceros) in both small and large-size fractions (Fig. 4a). Small and large heterotrophic dinoflagellates were well represented everywhere and ciliates were particularly abundant in some samples of the large size fraction off the plateau (Fig. 3). Below the ML (125 m), the relative abundance of heterotrophic dinoflagellates and Radiolaria increased. At 300 m, the small size fraction was largely dominated at all stations by Radiolaria, probably as swarmers stages (Fig. 3). Large, heavily silicified diatoms (Eucampia and Odontella on the plateau; Thalassiosira and Fragilariopsis off the plateau) characterized phytoplankton community (Supporting Information Fig. S2). Chaetoceros was also present, mostly in the small size fraction on the plateau and at M1. Phaeocystis, probably forming aggregates, was detected in large proportions only at 300 m off the plateau (Fig. 3). and were attributed to high heterotrophic activity . Ammonium is less energetically costly to assimilate than nitrate and is as such the preferred dissolved nitrogen source for most microalgae (McCarthy 1981;Raven et al. 1992), especially for prasinophytes (Litchman et al. 2007). They are thus able to grow more efficiently than other phytoplanktonic groups when ammonium concentrations are high. Slightly warmer temperatures on the plateau than off the plateau might also represent an advantage for prasinophytes. In the SO, other CHEMTAX studies have reported prasinophytes mostly north of 51 S , particularly in the Sub Tropical Front, where they could represent up to 10% of Chl a biomass (Iida and Odate 2014). In our study area, prasinophytes were mostly found close to the polar front during the onset (R-2, KEOPS2, Georges et al. 2014) and after the bloom (Sta. M2 and M3, MOBYDICK, Fig. 6d). When the Arctic Micromonas thrive in a cold and ice-covered ocean (Lovejoy et al. 2007), its close phylogenetic counterpart inhabiting the SO seems to be favored by warmer waters, suggesting differentiation into two distinct ecotypes.
Pelagophytes were enriched in Antarctic waters off the plateau (Figs. 4a, 7). During the LOHAFEX artificial iron experiment, pelagophytes were common in iron limited waters, but their relative abundance in sequencing datasets decreased after iron fertilization, whereas Micromonas became more abundant (Thiele et al. 2014). Experimental studies also demonstrated that Pelagomonas had higher affinities for dissolved inorganic iron than another prasinophyte Prasinomonas (Timmermans et al. 2005). If the same is true for other prasinophytes, Pelagomonas could thus be more competitive when iron is limiting, whereas Micromonas could be characterized as an opportunistic species, able to rapidly grow on ammonium.

Balance between small nonsilicified phytoplankton and diatoms (small and large)
At the bloom onset, when silicate and iron were not limiting, diatoms were able to build up biomass more efficiently than small nonsilicified phytoplankton whose abundance decreased ( Fig. 6a; Supporting Information Table S2). Diatom grazers have relatively long generation times, leaving time for diatoms to proliferate until predation pressure control their population (Smetacek et al. 2004). Small nonsilicified phytoplankton are more rapidly controlled by predators, composed of fast growing microzooplankton such as ciliates and dinoflagellates (Coale et al. 1996;Landry et al. 2002Landry et al. , 2011Smetacek et al. 2004). During the decline and after the bloom on the plateau, the scarcity of silicate (18.7 μmol L −1 in October in KEOPS2 to less than 2 μmol L −1 during MOBYDICK; Supporting Information Table S2) and iron likely explains the collapse of diatom populations .
During MOBYDICK, diatom communities differed between on and off-plateau stations. According to sequencing data of the < 20 μm fraction, a diverse assemblage of small diatoms, such as Fragilariopsis, Pseudo-nitzschia, Thalassiosira, and Chaetoceros, was enriched in Antarctic waters off the plateau (M4 and M1), probably related to higher silicate concentrations (Figs. 5, 7). The same diatom genera were also enriched in the large size fraction (Fig. 4b) and microscopic observations confirmed that their size range overlapped the two size fractions. In contrast, a few large diatoms were enriched on the plateau (Fig. 4b), where the lowest concentrations in Si(OH) 4 were observed (< 2 μmol L −1 during the first two visits). Silicate requirements as well as Si uptake kinetics can vary from one species to the other. However, Hoffmann et al. (2008) suggested that silicate depletion may limit small diatoms sooner than large diatoms since the amount of silica needed relative to the volume of the cell is higher for small diatoms. In contrast, small diatoms were likely advantaged off the plateau (Sta. M1 and M4), where silicic acid concentration was higher. When silicate concentrations remain sufficient to build the frustules, cell size reduction is a morphological adaptation of diatoms to low iron concentrations, as an increased surface to volume ratio could facilitate iron uptake (Leynaert et al. 2004;Marchetti and Cassar 2009). In addition to their small size, most diatom genera that were enriched off the plateau in our study possess particular physiological adaptations to cope with low iron concentrations. The pennate diatoms Pseudo-nitzschia and Fragilariopsis use the iron-concentrating protein Ferritin, which allows them to store more iron than centric diatoms ). The centric diatom Thalassiosira is adapted to low iron concentrations by possessing a modified photosynthetic apparatus, with electrontransport proteins modified to use copper instead of iron (Peers and Price 2006). It is also supposed to have an efficient iron vacuolar storage mechanism (Nuester et al. 2012). Sta. M3 was located in subantarctic waters and differed from the two other Antarctic water influenced off-plateau stations (M1, M4). M3 presented low silicate concentrations and lowest Chl a (Table 2). M3 showed the lowest proportion of diatom pigments (25% of Chl a) and number of reads in both small and large size fractions suggesting silicate and iron colimitation which disadvantaged growth of all diatoms.

Microbial eukaryotes below the ML and potential implications
Surface phytoplankton communities dominated by small cells, as observed during MOBYDICK cruise, are expected to sustain an active microbial loop. This implies little direct export of small phytoplankton production to the deep ocean, most of it being rapidly remineralized (Michaels and Silver 1988;Martin et al. 2013). Molecular analysis showed that the relative contribution of Phaeocystis strongly decreased below the ML in the small size fraction. The relative contribution of diatoms to the total amount of reads remained unchanged at 125 m and decreased at 300 m in this size fraction (Fig. 3). In the large size fraction, diatoms still accounted for over 25% of total number of reads in some off-plateau deep samples (M3-2 and M4-1, 300 m). Thalassiosira and Fragilariopsis, which were recorded in both small and large size fractions at the surface in off-plateau waters, were found only in the large size fraction at 300 m (Supporting Information Figs. S2, S5). Lightly silicified Corethron dominated at the surface on the plateau during MOBYDICK (Supporting Information Fig. S2) while Chaetoceros, Eucampia, and Odontella were the main diatoms detected at 125 m depth (M2-3; Supporting Information Figs. S2, S5b). This suggests that small diatoms are not as efficiently exported as their larger counterparts. It also supports the hypothesis that the higher carbon export efficiency recorded off the plateau in summer (up to 58% of net primary production during KEOPS1) could be due to direct export of large fast sinking heavy silicified diatoms, assumed to be grazing-resistant (Laurenceau-Cornec et al. 2015).
Interestingly, Phaeocystis was found in relatively high proportions off the plateau (M3 and M4) at 300 m in the large size-fraction (Fig. 3). Although vertical carbon flux attenuation is higher for Phaeocystis than for diatoms (Reigstad and Wassmann 2007), Phaeocystis blooms have been shown to greatly contribute to carbon export in the Ross Sea (DiTullio et al. 2000). In our study, only single cells of Phaeocystis were observed in surface samples (Supporting Information Fig. S5). Considering the essential role of iron in colony formation, we consider unlikely that the presence of a relatively abundant number of reads at 300 m could result from direct sinking of colonies forms in previous weeks. This suggests that single cells of Phaeocystis could be exported directly through aggregation and/or indirectly through grazing (Reigstad and Wassmann 2007;Richardson and Jackson 2007;Richardson 2018). Grazers, such as dinoflagellates and Radiolaria, were the most common organisms in both size-fractions in the deeper samples (Figs. 3, 7). These predators are able to feed on a wide variety of size classes, from small cells to chainforming diatoms (Swanberg and Caron 1991;Sherr and Sherr 2007). Previous studies have reported their high abundance at the bottom of the ML where they may feed on diatoms and other sinking cells and particles (Christaki et al. 2008(Christaki et al. , 2015Gomi et al. 2010). The small fraction of deeper samples was largely dominated on and off-plateau by Radiolaria swarmers belonging to Chaunacanthidae family (cluster 4; Supporting Information Fig. S5b; Fig. 7). This group can form large cysts that can sink from the surface to deep waters where they release thousands of swarmers (< 5 μm) after the bloom. They may then use surface-produced organic carbon as a nutritional source while ascending through the water column (Martin et al. 2010;Decelle et al. 2013).
Concluding, global-change models project more stratified waters over the plateau of Kerguelen, increasing silicate and iron limitations (Freeman et al. 2018). Iron and silicate depleted waters are known to favor small phytoplankton communities over large ones-as observed during MOBYDICKand eventually reduce carbon export due to promotion of microbially dominated food webs and greater remineralization (Bowie et al. 2011;Ebersbach et al. 2011;Martin et al. 2013).
Our study has shown contrasting small phytoplankton community structures relative to available nutrients on and off the Kerguelen plateau, and also underlined the capacity of Phaeocystis to thrive all over the studied region. In addition, this study has also highlighted that the possible contribution of Phaeocystis to carbon export in low productivity conditions deserves to be further studied.