Seasonal samples were collected onboard the Canadian Coast Guard Research Icebreaker CCGS Amundsen every 2–4 weeks from November 2007 to July 2008 from Amundsen Gulf and adjacent bays (Figure 1A). We targeted the polar mixed layer (PML) near the surface (5–20 m), and the top of a halocline (22–80 m), which separates the PML from Pacific Summer Water (PSW). In late spring the Beaufort Sea SCM typically forms at this halocline (Bergeron and Tremblay, 2014). During winter and early spring, the halocline sample depth was identified by salinity and temperature profiles with salinities of 31–32, and from May to July confirmed by the deeper chlorophyll fluorescence peak.

Samples used to investigate interannual variability of ciliate and dinoflagellate communities in the SCM were collected during multiple missions to the Amundsen Gulf and Beaufort Sea between 2003 and 2010, aboard the CCGS Amundsen (2003-2006, 2009-2010), Louis S. St. Laurent (2007), and Sir Wilfred Laurier (2008). Details of these missions have been reported elsewhere (Barber et al., 2010, 2012; Comeau et al., 2011; Terrado et al., 2011).

Except for 2010, all sample were preserved using the same methods, and following extraction, all were amplified and sequenced using the same protocols. Briefly, water samples for DNA (ca. 6 L) and RNA (ca. 4 L) were collected into cleaned acid rinsed carboys from 12-L (CCGS Amundsen) or 10-L (CCGS Louis St Laurent, Sir Wilfred Laurier) Niskin-type bottles mounted on a Rosette system equipped with a conductivity, temperature and depth (CTD; Sea-Bird Electronics Inc., California), chlorophyll fluorescence (Seapoint Sensors Inc.), and relative nitrate from ultraviolet spectrophotometer (ISUS, Satlantic) profilers. Target depths were identified by fluorescence, salinity, temperature, and relative nitrate from the downcast and water collected on the upcast. All samples were filtered and preserved within 2 h of collection. Following pre-filtration through a 50-μm mesh to remove macrozooplankton, samples were sequentially filtered using a peristaltic pump (Cole-Parmer, USA) onto a 3-μm pore-size 47-mm polycarbonate filter (PC, AMD Manufacturing), and 0.2 μm pore-size Sterivex™ unit (Millipore) for the DNA and a 47-mm 0.2 μm pore-size PC filter for the RNA. All PC filters were placed in 2-mL cryovials. Prior to 2010, DNA was preserved by adding 1.8 mL of lysis buffer to the Sterivex units (0.2–3 μm fraction) and cryovials (3–50 μm fraction), and RNA was preserved in RLT buffer (Qiagen) as in Terrado et al. (2011) and Thaler and Lovejoy (2015). Samples were either frozen immediately at −80°C or placed in liquid nitrogen after adding buffer. Samples collected in 2010 were preserved using RNALater (Ambio). After 30–60 min in buffer, the filters were flash frozen in liquid nitrogen and then kept at −80°C until processing in the laboratory. We note that preliminary tests on paired samples from both marine and freshwater environments showed no significant differences in DNA and RNA recovery or community compositions between the protocols (Lovejoy pers. comm.). For samples collected before 2010, DNA was extracted from the filters as in Terrado et al. (2011) using a salt based method (Aljanabi and Martinez, 1997) while RNA was extracted using the RNAEasy micro Kit (Qiagen). The sample collected in 2010 was extracted using the AllPrep DNA/RNA Mini Kit (Qiagen) with DNA and RNA from the same filters. Conversion of RNA to cDNA was carried out using the High Capacity Reverse Transcription Kit (Applied Biosystems, Foster City, CA, USA). The V4 region of 18S rRNA in both DNA and cDNA samples was amplified using eukaryotic specific forward primers E572F with the Roche A adapter and reverse primer E1009R as described by Comeau et al. (2011). For the seasonal samples, large and small fractions were pooled after normalization based on relative chlorophyll concentrations (Supplementary Table 1). Only the 0.2–3 μm fraction was used in the interannual study. Separate sequencing runs were carried out for the seasonal and interannual studies. For each run, equimolar concentrations of amplicons from each sample were pooled for multiplex sequencing using the GS FLX Titanium Roche 454 platform (Roche/454 Life Sciences, Branford, CT, USA) at IBIS/Université Laval Plateforme d'Analyses Génomiques (Quebec, QC, Canada). All reads have been deposited in the Short Read Archive of NCBI under accession codes: PRJNA283142 (IPY-CFL) and SRA029114 (Amundsen Gulf Time Series, Comeau et al., 2011).

The seasonal data reads were quality filtered using the Quantitative Insights into Microbial Ecology program (QIIME; Caporaso et al., 2010b). Short sequences and primers were identified in mothur and removed (Schloss et al., 2009). Data were then de-noised following Reeder and Knight (2010). Chimeric sequences were identified in UCHIME (Edgar et al., 2011) and removed. Interannual data was processed using mothur (Schloss et al., 2009) as described in Comeau et al. (2011). The two data sets were then processed for OTU picking at 98% similarity utilizing USEARCH (Edgar, 2010) against the Silva Reference Database v.102 (Pruesse et al., 2007). All OTU representative reads were aligned in PyNast (Caporaso et al., 2010a), manually curated in BioEdit v.7.2.5 (Hall, 1999) and used to construct a phylogenetic tree (FastTree; Price et al., 2010). Assignment of taxonomic identity was performed in mothur with a 0.8 confidence threshold against the Northern Reference Database v.1.0 (Lovejoy et al., 2016), which follows the Silva taxonomy but includes high-quality longer environmental sequences from the Arctic and North Atlantic (Terrado et al., 2009, 2011; Charvet et al., 2012; Monier et al., 2013; Dasilva et al., 2014). However, using this database, around 30–45% of the reads were still not classified beyond “Other Dinoflagellates” or “Other Ciliates.”

To further increase taxonomic resolution, dinoflagellate sequences from Saldarriaga et al. (2001, 2004), Logares et al. (2007), and Potvin et al. (2013), and ciliate sequences from Dunthorn et al. (2014), along with Evolutionary Placement Algorithm, Randomized-Axelerated Maximum Likelihood- (EPA RAxML, Stamatakis, 2014) abundant Arctic-derived HTS reads were added to the v.1.0 of Northern Reference Database (v.1.1, Lovejoy et al., 2016). Taxonomic classifications of the dinoflagellate sequences used as references were verified in the AlgaeBase.org (Guiry and Guiry, 2015) and by literature searches, while ciliate taxonomic classification followed Lynn (2012). Using the v.1.1 of the database, taxonomy assignment was re-performed to generate the final OTU matrix (OTU reads per sample) with improved results (i.e., “Other Dinoflagellates” and “Other Ciliates” went down to 5–10%). OTUs matching metazoa, fungi, higher plants and reads that could not be classified to any taxonomic level below Eukaryota were removed and not analyzed further. This cleaned data set was considered as the total microbial eukaryotic community. To facilitate diversity comparisons, the OTU matrix was then rarefied, based on the sample with the fewest reads in the separate data sets, resulting in 4953 reads per sample for the seasonal study, and 7041 reads per sample for the interannual comparison.

To investigate the OTUs that contributed to the changes in July 2008 for the interannual data (see Section Results), the remaining abundant “Other Ciliate” OTUs were searched for in NCBI and their most similar nearly full length 18S rRNA gene sequences were used to construct reference trees as in Thaler and Lovejoy (2015). Then, the abundant but unclassified ciliate OTUs were then mapped back onto the reference trees using EPA RAxML v.8 (Stamatakis, 2014). We then inferred their probable taxonomic identity based on the nearest or most similar annotated reference sequences.

Statistical tests were carried out in the R environment v 3.0 (R Development Core Team, 2008). Spearman's Rank Correlation (rho) in the Vegan package (Dixon and Dixon, 2003) was used to test correlations between taxa and environmental parameters, and results were plotted in igraph (Csárdi and Nepusz, 2006). Only correlations with Spearman's rho ≥ 0.3 and significant at p < 0.001 were retained (Barberán et al., 2012). Species-variable relationships were then visualized by network analysis as target-source plots in Cytoscape 3.0 (Shannon et al., 2003; see Supplementary Methods 1). Analysis of variance (ANOVA) was applied to determine significant differences between samples and linear regression tests were used to infer relationships, both analysis were carried out using PAST v3.0 (Hammer et al., 2001).

Alpha diversity (Chao1 index) of the ciliate and dinoflagellate communities in each seasonal sample was estimated as implemented in QIIME (Chao and Shen, 2003; Caporaso et al., 2010b). Since the samples used for the interannual part of the study were collected in different months (Jul to Nov) over the 8 years, we tested for variability associated with the month of collection and found no significant differences in species richness of OTUs in ciliate or dinoflagellate abundance between summer and autumn (Supplementary Figure 1). Phylogenetic unweighted UniFrac dissimilarities (beta diversity) among the taxa were computed separately for seasonal and interannual datasets by the jackknife method and generalized UniFrac procedure (Lozupone and Knight, 2005; Chen et al., 2012). All dinoflagellate and ciliate OTUs in the seasonal study were included in the beta diversity measure. To avoid potential rarefication artifacts (Ramette, 2007) in the interannual data, rare OTUs that were defined as < 1% of the total ciliate or dinoflagellate reads, were not included in the beta diversity and remaining analysis. The pairwise dissimilarity matrices were used to generate dendrograms using UPGMA (Price et al., 2010). The weighted contributions of seasons, depths (surface or SCM-halocline) and templates (RNA or DNA) to community structuring were computed using ADONIS (Fierer et al., 2012). The checkerboard score (C-score) was used to test for non-random co-occurrence patterns under the null hypothesis parameter using the oecosimu function in the R library bipartite with 1000 simulations (Dormann et al., 2008).

For the seasonal study (Figure 1A), surface PML waters remained cold (−1.7 to 2°C) from mid-January to mid-May and increased toward summer reaching 8°C in July, when the region was ice-free. Phosphate concentration changed little over time with concentrations from 0.6 to 1.6 mg m⁻³. Over the 9 months, nitrate concentrations were significantly greater in the halocline compared to the surface (t-test, p < 0.001), with greatest concentrations from February to April (Table 1). Nitrate concentrations started to decrease beginning in early April in the surface and in May at the halocline. The chlorophyll a (Chl a) concentrations, as estimated from in situ fluorescence, were negligible until 9 April and were greater in the surface waters compared to halocline through 19 May. Concentrations were below 1 mg m⁻³ in both the surface and halocline until July when a strong SCM was apparent at the halocline (Table 1).

The physical characteristics of the time series stations (Figure 1B) have been reported elsewhere (Comeau et al., 2011). Briefly, over the 8 years, nutrients in the SCM ranged from 0.09 to 8.22 mmol m⁻³ for NO3−, 0.46–1.29 mmol m⁻³ for PO₄ and the highest Chl a (total) was detected on September 2005 reaching 2.14 mg m⁻³ (Supplementary Table 2). After binning the data from before and after 2007, total summer-fall Chl a values went from 0.43 mg m⁻³ to 0.31 mg m⁻³ in the SCM.

The seasonal rDNA and rRNA OTU communities separated into distinct surface and halocline clusters, with clear separation between the rDNA and rRNA communities within the depth defined clusters (Figure 2). Within the template clusters samples separated by season as Autumn-Winter (Nov-April) and Spring-Summer (May-July) categories, except for one rRNA surface sample collected on June 14, which clustered with the winter rRNA samples (Supplementary Figure 2). The similarity was mainly driven by the presence of several OTUs that occurred in the rRNA surface community on Jan 3 and June 14 but were absent in the other summer samples (Supplementary Table 3). We note that, because of ship time constraints, the June 14 samples were collected from Darnley Bay, which is an extension of Amundsen Gulf. For all samples, the ADONIS test also showed individual contributions to clustering of around 14% (adj. R² = 0.148 and 0.146, p < 0.001) grouping by depth category and sample type, while season contributed the least to the variance observed (adj. R² = 0.073, p < 0.001). The C-score test further supported the non-random distribution of the microzooplankton communities (p < 0.001).

Since DNA can be retrieved from both living and senescent cells, and free DNA can persist in the environment (Torti et al., 2015), the RNA-sourced samples were used to investigate potential influence by local conditions. Based on rRNA reads, ciliates, and dinoflagellates were found in all samples (Figure 3, Supplementary Tables 4 and 5), with greater proportions during summer (ca. 40%) and lowest in spring (ca. 5%) when larger photosynthetic taxa (mostly diatoms, Joli et al., accepted) dominated the planktonic community. The combined mean seasonal alpha diversity (Chao 1 index) of both ciliates and dinoflagellates based on rDNA, decreased from 125 in winter to 83 after spring (Supplementary Figure 3). In terms of total relative abundance, ciliate proportions were similar in both depths during winter (Figures 3A,C) while dinoflagellates were better represented in the surface (Figures 3B,D). Ciliate read relative abundance increased in the surface starting early June (Figure 3A), while dinoflagellates increased in the halocline (Figure 3D).

Significant associations (Spearman's rho > 0.3, p < 0.001) among taxa and environmental variables were detected in the network analysis (Figure 4). The associations were consistent with the seasonal succession and depth (water mass) seen in the UniFrac clustering (Figure 2). For example, surface-associated ciliates: Strombidium, Pseudotontonia, Askenasia, and dinoflagellates Gymnodinium-like, Adenoides-like, Gyrodinium c.f. gutrula, and Scripsiella (Figure 4, taxa 1–7) were associated with higher DO, which was a characteristic of the surface PML. In addition, abundances of Amphidoma, “Arctic Clade 1,” Blastodinium, Monodinium, and Novostrombidium (Figure 4, taxa 8–13) were higher when Chl a concentrations were greater, which was associated with longer day lengths and warmer water, suggesting characteristic surface summer taxa. Tintinnidium, Parauronema, Oligohymenophorea, Gyrodinium helveticum, and Amphidinium-like taxa (Figure 4, taxa 15–18) were more associated with the deeper, more saline and nutrient-richer SCM or halocline layer. Although Laboea (Figure 4, taxon 14) was also strongly correlated with the SCM/halocline conditions, it was most abundant when the days were longer. In contrast, Woloszynskia, Gymnodinium sp., and “Other Gyrodinium” (Figure 4, taxa 19, 21, 22) were more abundant in the halocline during winter when ice cover was more extensive. These species, correlated with specific environmental variables, provide evidence for succession and seasonality within the two water masses. Binned OTUs of poorly defined groups that were not classified beyond their respective ranks, including “Other” Dinoflagellates, Ciliophora, Spirotrichea, Litostomatea, and Strombidium (Figure 4, taxa 23–30), did not correlate with any of the environmental variables tested.

Ciliate communities were consistently dominated by “Other Spirotrichea” with annual mean abundance of 9.4 ± 2% relative to the total microbial eukaryotic reads. “Other ciliates” (1.7 ± 1%), Strombidium sp., Pelagostrombidium sp., “Other Oligohymenophorea,” “Other Litostomatea,” Askenasia sp., and Monodinium sp. abundances ranged from 0.01 to 1.5% of total microbial eukaryote relative abundance (Supplementary Figure 4A). The most abundant dinoflagellate OTUs were from Gymnodiniales, Blastodinium-like, “Arctic clade 1,” Gyrodinium, Adenoides-like, and unclassified Dinoflagellates, each accounting for 1–6% of the total microbial eukaryotic reads (Supplementary Figure 4B).

Here, using our refined taxonomic database we found that microzooplankton community composition also changed, especially in summer 2008. Specifically, based on abundant OTUs (>1%), the July 2008 sample showed the greatest dissimilarity from other communities (Figure 5A). This dissimilarity was driven by the July 2008 increase in both abundance and OTU richness of Strombidium, Laboea, Monodinium, and “Other Ciliates” (Figure 5B), with a significant correlation with decreasing salinity (r² = 0.82, p < 0.01). After 2008, the relative abundances of these groups returned to previous levels by 2010. Their relative abundances were also inversely correlated (multiple regression) with SiO₃ (r² = 0.29, p < 0.001), salinity (r² = 0.22, p = 0.1), and nitrate (r² = 0.18, p = 0.2). Phylogenetic placement of the “Other Ciliate” reads further revealed that these were in the Class Spirotrichea, and included sub-classes Oligotrichia and Choreotrichia. Most of these Spirotrichea clustered with environmental sequences, which we refer to as Environmental Clades 1, 2, 3, 4, 5, and 6 (Supplementary Figure 5). Aggregated read counts of Oligotrichia, and Environmental Clades 1, 4, 5, and 6 showed significant increase after 2007 (Krustal Wallis, p < 0.01), while Choreotrichia, and Environmental Clades 2 and 3 decreased.

Eukaryotic microbial communities from rDNA have been used to infer water mass history (Hamilton et al., 2008; Monier et al., 2013; Zhang et al., 2014), whereas taxa from rRNA are thought to more closely indicate active representatives of the community (Campbell and Kirchman, 2013; Hunt et al., 2013). Here, we found that for both DNA and RNA sourced samples, the surface communities always clustered together separate from the SCM communities. Earlier microscopy (Lovejoy et al., 1993; Okolodkov and Dodgeb, 1996; Jiang et al., 2013) and clone library studies using longer 18S rRNA sequences (Bachy et al., 2011; Lovejoy and Potvin, 2011) have also highlighted the differences between SCM and surface communities over the Arctic. Within the two water masses, communities separated by template suggests a pool of historic DNA, which could include dormant or less active stages, e.g., cysts (Verni and Rosati, 2011; Bravo and Figueroa, 2014), advected non-active species (Lovejoy and Potvin, 2011), preserved free DNA, or non-living material in marine snow (Nielsen et al., 2007; Boere et al., 2011). All of these sources can be transported by currents and persist over long distances (Heiskanen, 1993; Brocks and Banfield, 2009).

Although RNA-derived reads cannot be directly used to estimate biomass (Medinger et al., 2010; Blazewicz et al., 2013), the patterns of relative read abundance were consistent with specific environmental drivers operating on particular species or clades, with communities separating by season. One exception was the June 14 rRNA surface sample from Darnley Bay, which based on OTU composition clustered with the January 3 rRNA sample. The placement of the June sample in an otherwise winter cluster was consistent with deeper unmodified Pacific Winter Water appearing in Darnley Bay, which was evident in the physico-chemical sample clustering (Supplementary Figure 2). Upwelling and on-shelf transport of deeper Pacific Water from Amundsen Gulf has been reported previously (Garneau et al., 2006; Paulic et al., 2012). Unfortunately, CTD transects were not taken along the shelf break in June and the source of this water remains speculative. The biological clustering was driven by 15 OTUs belonging to “Arctic Clade 1,” Gyrodinium, Gymnodinium, “Other Dinoflagellates,” Askenasia, Tintinndinium, “Other Spirotrichea,” and “Other Ciliates.”

While the dependence on phytoplankton prey can generate annual bimodal or unimodal microzooplankton bloom patterns in some regions, with temperature having an influence (Levinsen and Nielsen, 2002; Godhantaraman and Uye, 2003), there is little understanding of species composition over annual cycles. Here we found that ciliate and dinoflagellate taxa showed distinct seasonal patterns that could be partially explained by salinity, temperature, and nutrient changes from winter to summer (Figure 4). Specifically, the proportion of ciliate reads were generally higher under ice free conditions, when the surface waters were less saline and at the end of the spring bloom when nitrate concentrations were drawn down by photosynthetic activity (Forest et al., 2010; Tremblay et al., 2011). Although many ciliates have a wide range of salinity tolerance (Montagnes, 1996), this has not been investigated for Arctic species. In contrast dinoflagellate proportions were higher in the surface in winter, but more abundant in the SCM toward summer. Dinoflagellates can produce the osmolyte dimethylsulfoniopropionate (DMSP), which allow cells to regulate internal homeostasis (see review by Stefels, 2000), which may provide an advantage relative to ciliates under conditions of slightly higher salinity.

During late spring and toward summer with increasing light availability, the putative plastid-containing ciliates and dinoflagellates increased proportionally in the rRNA OTUs, suggesting that photosynthesis may have provided additional energy for photosynthetic or mixotrophic taxa. In Amundsen Gulf, in summer, the higher sun angle and longer day-length mean that more light reaches the nutrient-rich halocline leading to the establishment of the SCM. We found that ciliate and dinoflagellate reads made up on average ca. 60% of the total microbial eukaryotic reads in early summer in the SCM. In contrast, a microscopy study in 2005-2006 reported that photosynthetic and non-photosynthetic dinoflagellates constitute only around 3–6% of the total mean phytoplankton cell counts in the SCM (Martin et al., 2010). The discrepancy is likely related to cell breakage during collection, misidentification with smaller dinoflagellates grouped with other flagellates, and loss during storage. Ciliates, can also be fragile and tend to be either ignored or under sampled using standard Lugol's techniques (Lovejoy et al., 1993). Ciliate and dinoflagellate dominance in 18S rRNA gene libraries has also been attributed to their comparatively larger genomes and multiple copies of the rRNA gene (Godhe et al., 2008). Irrespective of the actual cell abundance, our analysis clearly revealed otherwise undetected changes in species over seasons (Supplementary Tables 4 and 5). Our seasonal data ended in July, but the interannual data in the SCM showed that both alveolate groups remain through summer to late autumn.

Season and depth were also the major factors that grouped ciliates and dinoflagellates using the network analyses. The seasonal changes in microbial composition also suggested shifts in dominant function within the two groups (Figure 4). For example, winter communities were mainly dominated by heterotrophic taxa. The surface winter communities would have access to small planktonic species such as Micromonas (Lovejoy et al., 2007), haptophytes, and smaller heterotrophic flagellates (Terrado et al., 2011). The relative scarcity of food resources during winter likely selected for particular groups as reported for Kongsfjorden, Svalbard (Seuthe et al., 2011), and Disko Bay (Levinsen et al., 2000; Levinsen and Nielsen, 2002). Dinoflagellates can also graze on smaller ciliates using pseudopodia or feeding veils (Jacobson and Anderson, 1986, 1996). Such trophic interactions, with dinoflagellates most likely preying on ciliates, could contribute to differences in relative abundances in different depths and times (Hansen et al., 1999; Møller et al., 2006; Seuthe et al., 2011). Additional trophic interactions were indicated in the summer when primary production was higher and zooplankton were more abundant in surface waters (Seuthe et al., 2007; Forest et al., 2011), with the appearance of the putative copepod parasite Blastodinium.

After spring and toward summer, the correlation of some ciliates and dinoflagellates with Chl a might also be associated with predation on phytoplankton. Ciliates and dinoflagellates consume from 30% (15.6 g C m⁻²) of gross primary production in Amundsen Gulf to as much as 56% in some Arctic fjords (Seuthe et al., 2007; Forest et al., 2011). The availability of light in spring and summer could favor mixotrophy as well. For example, the obligate mixotrophic Laboea and most litostomatids, and the potentially plastidic Scripsiella-like, “Other Gyrodinium,” “Other Gymnodiniales,” and Adenoides further increased in abundance in the SCM over this period when both light and prey would be available.

The summer-autumn ciliate and dinoflagellate assemblages tended to be similar in the different years, with a consistent predictable re-establishment of communities most years, despite samples being collected from across a large geographical distance. The SCM in the Beaufort Sea forms near the upper surface of the Pacific Winter Water, which is transported as a distinct entity over long distances (Carmack and Macdonald, 2002). Previous studies consistently show that protist communities in the SCM of Beaufort Sea and Amundsen Gulf change little over geographical space and for a given year are strongly associated with their water mass of origin (e.g., Comeau et al., 2011; Lovejoy and Potvin, 2011; Monier et al., 2015). The strong association with water mass was evident in the two samples collected in July 2008 that were analyzed separately because of differences in sample preparation protocols between the seasonal study and the interannual study. The internannual study was only from the small fraction, while the seasonal study included amplicons from large and small fractions. Despite this, the dinoflagellate and ciliate OTUs from Amundsen Gulf collected 8 July, were similar to those collected further offshore in the Beaufort Sea on 21 July, consistent with both 2008 samples from the same water mass and similar conditions.

Lovejoy et al. (2002) showed in microcosm experiments in the Arctic that ciliates and dinoflagellates dominated when nutrients were depleted following blooms of large phytoplankton. Ciliates are able to exploit oligotrophic conditions, by grazing on smaller phytoplankton and bacteria either as strict heterotrophs or by way of mixotrophy. By the end of July 2008, the Beaufort Sea SCM was nitrate limited and had the lowest mean phosphate concentrations measured from 2003 to 2010. Overall primary productivity was low (Martin et al., 2013) with small cells (<3 μm) dominating the Chl a biomass (Comeau et al., 2011). The mean salinity at the SCM was lower in 2008 compared to the preceding years (2003-2007), which was due to the input of additional freshwater from multi-year ice melt in 2007. This event was part of the ongoing trend in decreasing summer ice extent that has been recorded since the onset of the satellite era. The continuing low ice volume over winter was associated with the 2008 ice break-up 2 weeks before the historic average. This precocious ice breakup stimulated an earlier pelagic spring bloom and earlier depletion of nutrients in the euphotic zone (Forest et al., 2011). These July 2008 conditions were reflected in the microzooplankton assemblage, characterized by the greater prevalence of Strombidinium, Laboea, Monodinium sp., other unclassified ciliates and relative decrease in most dinoflagellates, which was also noted in the seasonal data set. The relative abundance of ciliate taxa in the above community gradually returned to earlier levels along with the partial recovery of summer ice extent. At the OTU level, Spirotrichea showed different trends, before and after 2007. Spirotrichea ciliates are obligate mixotrophs, feeding on smaller prey and presumably able to use retained chlorophyte or prymnesiophyte-derived plastids for photosynthesis (Stoecker et al., 1988, 2009; McManus and Katz, 2009). Similar observations were reported by Jiang et al. (2013) in Western Arctic Ocean where a few ciliate groups became dominant during the 2012 record ice melt, supporting the notion that microzooplankton are highly sensitive to changing physico-chemical regimes, including those associated with low-ice or freshening events.

The mean vertical positions of the SCM and the nitracline in the Beaufort Sea and Amundsen Gulf deepened from 2003 to 2011 (Bergeron and Tremblay, 2014). The increasing distance between the depth at which light is still sufficient for photosynthesis and the nitracline will continue to accelerate under a changing climate regime (Yamamoto-Kawai et al., 2008; McLaughlin and Carmack, 2010; Krishfield et al., 2014), which could drive both the surface and SCM layers to become more nutrient-limited. Such an effect has already been reported for the Canada Basin with an increase in smaller phytoplankton and lower nitrate levels in the upper 200 m (Li et al., 2009). Under these new conditions ciliates and dinoflagellates that are better able exploit smaller prey would be selected for. Our results support recent mesocosm and modeling experiments that suggest that under future conditions, the microzooplankton and microbial loop will become more prominent as efficient intermediates between bacteria-picoplankton and the classical food webs (Skjoldborg et al., 2003; O'Connor et al., 2009; Montagnes et al., 2010; Lewandowska et al., 2014; D'Alelio et al., 2016).

In summary, this study shows that in Amundsen Gulf, ciliates and dinoflagellates exhibit complex temporal dynamics and are influenced by biological and physico-chemical regulators. Ciliate reads were more abundant in the surface when salinity was lower toward summer while dinoflagellates dominated in the SCM where nutrients and light were available for potential mixotrophic or photosynthetic activity. Despite this strong seasonality, the microzooplankton assemblages were also highly similar every summer from 2003 to 2010 except in 2008, following the 2007 summer ice minimum. Mixotrophic taxa increased drastically when nutrient concentrations were low and small prey were presumably available. Low nutrient-small cell conditions are predicted to occur more frequently in the Arctic, and this may lead to a change in dominant microzooplankton species that link to the classical food webs.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.