Protist community composition during early phytoplankton blooms in the naturally iron-fertilized Kerguelen area (Southern Ocean)

Microbial eukaryotic community composition was examined by 18S rRNA gene tag pyrosequencing, during the early phase of spring phytoplankton blooms induced by natural iron fertilization, o ﬀ Kerguelen A total of 999 operational taxonomical units (OTUs), a ﬃ liated to 30 known high-level 5 taxonomic groups, were retrieved from 16 samples collected in the upper 300 m water column. The alveolata group was the most abundant in terms of sequence number and diversity (696 OTUs). The majority of alveolata sequences were a ﬃ liated to Dinophyceae and to two major groups of marine alveolates (MALV-I and MALV-II). In the upper 180 m, only 13 % of the OTUs were shared between of the fertilized stations and 10 the reference site characterized by high nutrient low chlorophyll (HNLC) waters. Fungi and Cercozoa were present in iron-fertilized waters, but almost absent in the HNLC samples, while Haptophyta and Chlorophyta characterized the HNLC sample. Finally, the 300 m depth samples of all stations were di ﬀ erentiated by the presence of MALV-II and Radiolaria. Multivariate analysis, examining the level of similarity between di ﬀ erent 15 samples, showed that protistan assemblages di ﬀ ered signiﬁcantly between the HNLC and iron-fertilized stations, but also between the diverse iron-fertilized blooms.

82F (5'-GAAACTGCGAATGGCTC-3', López-Garcia et al., 2003) and Euk-516r (5'-ACCAGACTTGCCCTCC-3', Amann et al., 1990). These primers have been designed to amplify the variable V2 and V3 eukaryote 18S rRNA gene regions. A 10 bp tag sequence specific to each sample, a 4 bp TCAG key, and a 26 bp adapter for the GS FLX technology, were added to the primers. Polymerase chain reactions were carried 10 out according to standard conditions for Platinum Tag High-Fidelity DNA polymerase (Invitrogen) with 10 ng of environmental DNA as a template. After the denaturation step at 94 • C for 2 min, 30 cycles of amplification were performed with a GeneAmp PCR System Apparatus (Applied Biosystems) as follows: 15 s at 94 • C, 30 s at 50 • C, 1 min at 72 • C, and 7 min at 72 • C. Tag pyrosequencing was carried out by the com-15 pany GenoScreen (Lille, France). The library was prepared following the procedures described by Roche (Basel, Switzerland) and used in a 1/4 plate run on a 454 GS FLX Titanium sequencer. Pyrosequences were submitted on GenBank-SRA under the accession number SRP041236. 20 The sequences were processed using the MOTHUR 1.28.0 software (Schloss, 2009) following the standard operating procedure (http://www.mothur.org/wiki/Schloss_SOP) (Schloss et al., 2011). First, flowgrams were extracted and demultiplexed according to their tag. The resulting sixteen flowgrams were denoised using the MOTHUR 1.28.0 implementation of PyroNoise (Quince, 2009). Primer sequences, TAG, and key frag-maining sequences were clustered into operational taxonomical units (OTUs) at 97 % similarity threshold. Single singletons (unique amplicons after 97 % clustering that occurred exclusively in only one sample) were removed from downstream analyses, as these are most likely erroneous sequencing products (Reeder and Knight, 2009;Kunin et al., 2010;Behnke et al., 2010). This dataset showed a representative overview of 10 the diversity as indicated by the rarefaction curves reaching a plateau in most cases ( Fig. S1; Supplement). All OTUs were given a putative taxonomic affiliations based on BLAST (Altschul et al., 1990) identification of the closest cultured or uncultured relatives against the PR2 (Guillou et al., 2013) and the GenBank databases. The OTUs identified as metazoan, were removed from downstream analysis. However, the meta-15 zoan OTUs displayed high and heterogeneous number of sequences between samples, making subsampling of the remaining OTUs unsuitable as it resulted in a drastic loss of diversity. For this reason, the data are presented based on the relative abundance of OTUs in each sample. 20 Rarefaction curves and alpha diversity estimators within particular samples (richness estimator S Chao1 ; the heterogeneity of the diversity; Simpson and Berger-Parker indices) were calculated with the PAST 2.17c software (Hammer et al., 2001). The S Chao1 approach uses the numbers of singletons and doubletons to estimate the number of expected species. According to S Chao1 ,"missing" species information is mostly concen-25 trated on those of low frequency counts. The Simpson index measures the "evenness" of the community and ranges from 0 (one taxon dominates the community) to 1 (all taxa are represented equally). Berger-Parker indicates the relative abundance of the 11184

Study site
The hydrographic conditions during KEOPS2 are reported in detail in Blain et al. (2014, this volume). The "historical" A3 station situated ∼ 500 m on the Kerguelen plateau (Blain et al., 2007(Blain et al., , 2008 was characterized by a deep mixed layer (ML) (153 ± 15 m) ( Table 1 Chl a, respectively, constrained to shallow ML (38 ± 7 m and 61 ± 11 m, respectively; Table 1). The highest temperature was recorded in the ML of the F-L station (4.2 • C, Fig. 2), indicating the influence of sub-Antarctic waters. The reference site (station R-2) in HNLC waters had low concentrations of chl a (0.25 ± 0.08 µg L −1 ), and a temperature of 2.1 • C (Fig. 2) in the ML (105 ± 15 m). The macronutrient concentrations in all and the lowest at the E-4W station (387 OTUs), while A3-2 and the HNLC R-2 stations had similar number of OTUs (550 and 496, respectively). The Simpson index, was relatively high, ranging from 0.76 (F-L station in the ML) to 0.99 (HNLC, R-2 station at 300 m). The Berger-Parker, indicating the relative abundance of the dominant OTU was generally low, except at the F-L station, where it reached its' highest value (0.48; Table 2).

High-level taxonomic groups
The 999 OTUs were affiliated into 30 higher taxonomic groups distributed in all the samples (Table 3) and shown as pie charts for each of the four stations ( Fig. 3). At all stations, Alveolata was the most diverse group (696 OTUs, mainly composed of 15 MALV-II, Dinophyceae, MALV-I, and Ciliophora). The iron-fertilized stations accounted for the highest percentages of Alveolata while the lowest percentage was observed at the HNLC station R-2 ( Fig. 4). Stramenopiles were represented by 133 OTUs belonging to 10 higher taxonomic groups (Table 3). The most representative Stramenopile groups, in terms of OTUs number, were MAST, followed by Bacillariophyceae, and 20 Labyrinthulomycetes (Table 3). The relative abundance of sequences of Stramenopiles ranged between 8 and 29 % in the mixed layer samples (Fig. 4). Radiolaria (belonging to Rhizaria) were present at all stations and were more abundant in the 300 m depth samples. Their relative abundance was particularly pronounced at station F-L, where they represented 55 % of all sequences (Fig. 4). The fertilized stations were characterized by lower relative abundances of Haptophyta and Chlorophyta compared with the HNLC R-2 station (Fig. 4) Table 3). They were found almost exclusively at the fertilized stations, when only three OTUs were detected at the HNLC R-2 station (Fig. 3). Regarding lineages distribution according to depth, the proportions of phototrophic protists (e.g. Bacillariophyceae and Haptophyta) generally decreased below the ML. The relative contribution of MALV-I and MALV-II increased with depth, at all stations 5 except at station F-L.

Most abundant OTUs
The most abundant 207 OTUs, representing > 1 % of the sequences for each higher taxonomic group, accounted for 95 % of the total sequences.
The heterotrophic Gyrodinium spp. was the dominant Dinophyceae genus in all sam-10 ples, while the small autotrophic Gymnodinium spp., also present in all samples, displayed higher relative abundance in the HNLC R-2 samples (Table 4). Among Ciliophora, the genus Strombidium was the most abundant, while different OTUs belonging to Tintinnid species (Choreotrichia) were detected at all stations. The 17 most representative MAST-related OTUs were distributed in eight clades, with a MAST-9 sp.
prevailing at the surface F-L station (Table 4). At the fertilized stations, Bacillariophyceae-related OTUs were dominated by small sized species such as Planktoniella, Thalassiosira, andMinidiscus spp., while Pseudonitzschia was relatively abundant at the HNLC R-2 station (Table 4). Regarding the rest of the Stramenopiles, the photosynthetic picoalgae of the genus Bolidomonas prevailed 20 at all stations. The non-photosythetic Labyrithulomycetes were more often found at the iron-fertilized stations, with the parasitic genus Oblongichytrium sp. being relatively more abundant at the E-4W and A3-2 stations ( were present at the iron-fertilized stations, but almost absent at the HNLC station R-2 (Table 4).

Similarity of protistan assemblages
Altogether, the stations shared 197 OTUs, with 40 OTUs specific to the fertilized stations (Fig. 5). The F-L station contained the highest number of exclusive OTUs (Fig. 5).

5
The Bray-Curtis similarity analysis of 999 OTUs indicated four major clusters (Fig. 6a). The SIMPROF significance test indicated significant differences (P < 0.05) between these four groups and showed significant differences within the groups (i) to (iv) (Fig. 6a). The two-dimensional space nMDS visual representation, based on Bray-Curtis similarity analysis highlighted two major clusters ("shallow" and "deep" samples). 10 An overall low similarity (> 15 %) was observed within each group (Fig. 6b). At a higher level of similarity (40-50 %), the clusters broke roughly into individual stations: HNLC (cluster i); A3-2 (cluster ii); and E-4W (cluster iii); while the F-L 20 m and 65 m samples clustered with E-4W and the HNLC stations, respectively (Fig. 6b). Within the "deep" assemblage (cluster iv), the similarity between samples was low, except for samples R A total of 207 OTUs were classified as "abundant" (each representing ≥ 1 % of sequences in their higher taxonomic group) (Table 4); the most frequent OTUs belonged to Alveolata, followed by Stramenopiles, then Hacrobia (Table 3).

Phytoplankton
Although the tag pyrosequencing of the 18S rRNA gene has become a routine method in marine microbial diversity studies, it is itself subjected to several limitations, including, PCR-related biases, chimera formation, and primer non-universality ( reported recently in an extensive study at the San Pedro Ocean Time-Series station (SPOT; Lie et al., 2013). They can be related to extraction efficiency from thick walled diatoms (Medinger et al., 2010) and/or amplification biases favouring species with high 18S rRNA gene copy number, such as ciliates and dinoflagellates (Potvin and Lovejoy, 2009). It is also worth noting that 28 out of the 52 taxa identified by microscopy 5 (Armand et al., 2008) were not referenced in the GenBank. Finally, regarding the 27 diatom taxa that were "identified" only by pyrosequencing-based on sequence similarity with the closest existing cultured relatives in GenBank, they mainly belonged to the genera previously observed in this area (Armand et al., 2008). The accuracy of BLAST-derived taxonomy, especially at low-level taxa, depends on, sequence length, 10 variability of the 18S region, database coverage for the specific taxonomic group, and correct identification of the reference sequence (Bik et al., 2012). Sequences belonging to the nano-and pico-phytoplanktonic groups of Bolidophyceae, Pelagophyceae, Chrysophyceae, and Cryptophyta were found at relatively low abundances in all samples. Moreover, Haptophyta were dominated by an OTU af-15 filiated as Phaeocystis antarctica (100 % sequence identity). This phylotype has been previously reported as dominant in the south of the polar front (Wolf et al., 2014), in the Ross Sea waters (DiTullio et al., 2000), and in the naturally iron-fertilized bloom around the Crozet plateau (Poulton et al., 2007). 20 Although, Dinophyceae might be over-represented in the sequence data, possibly due to its high 18S gene copy number (e.g. Prokopowich et al., 2003;Zhu et al., 2005), tag pyrosequencing has allowed the highlighting of its extensive diversity (161 OTUs) in the Southern ocean; previously missed by conventional microscopy and/or pigment analysis (see also Wolf et al., 2014). For example, based on microscopy, Gyrodinium 25 is the most abundant dinoflagellate analyzed; however, no reliable distinction has been made between G. spirale and G. rubrum with morphological observations (Saito et al., 2005;Georges et al., unpublisehd KEOPS data Ciliophora, which are ecologically important grazers of small sized phytoplankton, accounted for a relative high number of OTUs (60 OTUs). As with previous microscopic observations in the Kerguelen area (Christaki et al., 2008), the most representative ciliate sequences in this study belonged to Strombidiidae. The relatively large sized Strombidium spp. (≥ 50 µm) can be plastidic (mixotrophic) and, along with Tontonia 5 spp. and Laboea spp. -also present in sequences -, were found to contribute to 40-60 % of the aloricate ciliate biomass during the late bloom on the Kerguelen plateau (KEOPS1, Christaki et al., 2008). Finally, the most relatively abundant sequences of tintinnid taxa -which are also important nanophytoplankton consumers -belonged to the large Cymatocyclis calyciformis (Christaki et al., 2008).

Microzooplankton: dinoflagellates, ciliates, and radiolaria
Radiolaria were another well-represented microzooplankton group (35 OTUs). These can act as particle feeders, by trapping their prey on the peripheral network of rhizopodia, or capture diatoms. They are also hosts of dinoflagellate symbionts and parasites, and may be important reservoirs of MALV taxa (e.g. Bråte et al., 2012). In this study, the relative increase of MALV with depth was consistent with a parallel increase of 15 Radiolaria. This observation is also supported by the hypothesis that MALV taxa are able to parasitize "deeper" planktonic organisms such as Spumellarida (Guillou et al., 2008), which were the most common group and were always well represented in the deeper water samples in this study (Fig. 4,Table 4). Radiolaria and MALV taxa characterizing deeper protistan assemblages have also been reported in the North Atlantic 20 (Countway et al., 2007(Countway et al., , 2010Not et al., 2007) and deep Antarctic polar front samples (López-Garcia et al., 2001).

Symbionts, parasites, and decomposers
This assemblage included the taxonomic groups of MALV-I, MALV-II, Labyrinthulomycetes, Pirsonia, Oomyeta, Apicomplexa, Perkinsea, Fungi, and Cercozoa. Many  -Alió, 2008). Their considerable abundance and diversity suggests interactions with various hosts, and therefore, it has been proposed that the whole MALV assemblage is composed of marine parasites (Skovgaard et al., 2005;Massana and Pedrós-Alió, 2008). Fungi and Cercozoa accounted for 28 and 17 OTUs, respectively. In a recent suc-5 cession study in the English Channel, it was observed that these groups mostly cooccurred with Bacillariophyceae (Christaki et al., 2014). Fungi are possibly related to the polysaccharide degradation of the freshly produced organic material by primary producers (Kimura and Naganuma 2001;Raghukumar, 2004). It is known for diatoms that polysaccharides are their main exudates (Myklestad, 1995 and references therein), 10 and these sugars could promote the growth of Fungi. Many Cercozoa are parasites of marine organisms, including large heavily silicified diatoms (e.g. Tillman et al., 1999;Schnepf and Kühn, 2000), which could explain why Fungi and Cercozoa were detected in the bloom stations and were poorly represented (2-3 OTUs, Table 4) at the HNLC R-2 station. Labyrinthulomycetes were also better represented in terms of numbers of 15 sequences in the bloom stations (Table 4)

Small heterotrophic protists
Among the small heterotrophic protists found in the samples, there were a variety of MAST (46 OTUs), Choanoflagellida (10 OTUs), and Telonemia (12 OTUs). MAST taxa are widely distributed in the world's oceans, and have been identified as free-living bacterivorous heterotrophic flagellates through a combination of FISH and other measure-

Variability of protistan assemblages relative to iron-fertilization
In general, the stability of OTUs richness and diversity indices between the HNLC R-2 and iron-fertilized stations indicated that the environment maintained an overall 5 diversity across stations and depths (Table 2). These observations are in agreement with previous molecular studies based on protistan diversity (e.g. Countway et al., 2007;Monchy et al., 2012). However, community structure analysis showed clear differences inside and outside the blooms (Fig. 6a). 10 Based on trophic organization, HNLC areas seem conceptually similar to oligotrophic regions dominated by small producers and an active microbial food web (e.g. Hall and Safi, 2001;Oliver et al., 2004;Christaki et al., 2008Christaki et al., , 2014bObernosterer et al., 2008). The characteristic contributors of the HNLC cluster (i) were Haptophyta, Chlorophyta, and MAST, which included mainly nanoplanktonic organisms. During KEOPS2, the rel-15 ative importance of small-sized cells at the HNLC station is in accordance with the flow cytometry data (4.8 ± 1.9 × 10 3 m L −1 nano-picophytoplankton cells in comparison to 1.8 ± 1.3 10 3 m L −1 at the bloom stations; KEOPS2 data). The factors influencing phytoplankton community composition (e.g. diatoms vs.Phaeocystis sp.) in the Southern Ocean are a complex interplay between bottom up (iron-silicate-light availability; 20 controlling growth) and top down effects (grazing; controlling mortality) (Cullen, 1991;Arrigo et al., 1999;Smetacek et al., 2004;Schoemann et al., 2005

Iron-fertilized sites
The mechanisms that fertilize the surface water in the region around Kerguelen are complex, which results in a patchwork of blooms with diverse biological and biogeo- to 170 m above the plateau at station A3. In accordance with these hydrographic characteristics, multivariate analysis of sequences showed that the ML sample of the F-L (20 m) was found in the same cluster as the E-4W samples, while the 65 m F-L sample was grouped with the HNLC samples. The OTUs putatively affiliated to heterotrophic dinoflagellate taxa (Table 5) were the major contributors of clusters (ii) and (iii) (Fig. 6a,   20 b). Dinoflagellate increase during iron-fertilized blooms, in particular, Gyrodinium spp. has been observed with microscopic counts during the iron addition experiments, and has been attributed to the increase of their diatoms prey (Hall and Safi, 2001;Saito et al., 2005;Henjes et al., 2007). Concluding, the tag pyrosequening approach in this study has provided an overview    Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Table 5. Results of SIMPER (similarity percentages) following the Bray-Curtis cluster analysis (Fig. 6a). Forty-one OTUs contributing for at least 1 % of the similarity of each cluster are listed in this table. In parenthesis, the mean of Bray-Curtis similarity is given for each cluster. Dino-Group-I-Clade-1 sp. 1 Total 4.7 6.9 9.2 9 Otu1211 MALV-2 Dino-Group-II-Clade-10 sp.