Biogeography of Heterotrophic Flagellate Populations Indicates the Presence of Generalist and Specialist Taxa in the Arctic Ocean

Abstract

Heterotrophic marine flagellates (HF) are ubiquitous in the world's oceans and represented in nearly all branches of the domain Eukaryota. However, the factors determining distributions of major taxonomic groups are poorly known. The Arctic Ocean is a good model environment for examining the distribution of functionally similar but phylogenetically diverse HF because the physical oceanography and annual ice cycles result in distinct environments that could select for microbial communities or favor specific taxa. We reanalyzed new and previously published high-throughput sequencing data from multiple studies in the Arctic Ocean to identify broad patterns in the distribution of individual taxa. HF accounted for fewer than 2% to over one-half of the reads from the water column and for up to 60% of reads from ice, which was dominated by Cryothecomonas. In the water column, many HF phylotypes belonging to Telonemia and Picozoa, uncultured marine stramenopiles (MAST), and choanoflagellates were geographically widely distributed. However, for two groups in particular, Telonemia and Cryothecomonas, some species level taxa showed more restricted distributions. For example, several phylotypes of Telonemia favored open waters with lower nutrients such as the Canada Basin and offshore of the Mackenzie Shelf. In summary, we found that while some Arctic HF were successful over a range of conditions, others could be specialists that occur under particular conditions. We conclude that tracking species level diversity in HF not only is feasible but also provides a potential tool for understanding the responses of marine microbial ecosystems to rapidly changing ice regimes.

INTRODUCTION

Heterotrophic flagellates (HF) have a central role in marine food webs, particularly in the Arctic, where they can control phytoplankton biomass (1). Modeling studies indicate that they likely consume the majority of bacterial biomass in this region (2). In other marine environments, they have been found to be more important than viral lysis (3) or ciliate grazers (4) for controlling bacterial concentrations. HF range from pico- to nanosized cells (0.8 to 20 μm) and are phylogenetically diverse, with representatives in most branches of the Eukaryota tree. Because taxon-specific differences in feeding behavior, population dynamics, and life histories may be important over broad ecosystem scales (5), the identification and quantification of the diverse groups can add substantially to our understanding of marine ecosystem function. However, HF taxa often have few diagnostic morphological features that enable easy identification (6) and microscopy-based studies rarely differentiate between HF taxonomic groups. Molecular approaches, such as cloning and amplicon sequencing, have provided additional information on the deep phylogenetic diversity of HF, especially in the Arctic (7). More recently, high-throughput multiplex tag sequencing of hypervariable regions of the 18S rRNA gene has become commonplace (8, 9). This is considered a semiquantitative method for retrieving thousands of taxonomically informative sequences from environmental samples. However, the level of taxonomic placement, for example, from phylum level to ecotype, is completely dependent on the depth and accuracy of the reference databases. For this reason, alignment and placement within phylogenetic trees remain the best method for accurately portraying diversity within groups of closely related organisms (10). While amplicon sequencing shares some of the inherent biases of other PCR-based methods (11, 12), it is useful for highlighting differences and similarities among samples, provided similar protocols are followed. In addition, diversity can be compared between environments, and the high output enables the detection of rarer species. Sufficient coverage will depend on the taxonomic richness of an environment, and even when using high-throughput sequencing, it may be difficult to achieve in extreme cases (13).

Current changes in the Arctic Ocean, such as the loss of multiyear ice, may be impacting marine photosynthetic protist communities. For example, Li et al. (14) documented a shift in size structure toward smaller cells. However, the consequences for HF taxa are not known, in part because few studies have characterized the diversity and biogeography of different HF (15, 16). Although there have been a number of studies using V4 18S rRNA gene amplicon pyrosequencing in the Arctic (9, 17–19), none were focused on pan-Arctic comparisons. All of these studies have detected an effect of environmental gradients on one or more HF taxa, but the extent to which HF with similar putative ecological roles (e.g., bacterivory) might vary in relative abundance over space and time is not yet understood, and a comprehensive analysis is lacking.

Our aim was to examine smaller HF taxa (nano- and picoplankton, 0.2 to 20 μm), whose sporadic and low occurrence in individual studies has eluded detailed analysis, and to increase the environmental coverage by additional high-throughput sequencing from samples collected mostly from offshore regions of the Eastern and Western Arctic. We chose to focus on HF taxa of smaller cell size that are usually considered bacterivores and detritivores. These included the cercozoan order Cryomonadida, the phylogenetically diverse marine stramenopiles (MAST), the Telonemia, the Picozoa (formerly known as picobiliphytes [20]), and the Choanoflagellida. All of these HF taxa have been detected in 18S rRNA gene clone libraries from the Canadian Arctic (7, 21, 22). Katablepharids were excluded from the analysis, although they have been detected in enrichment studies in the Arctic (23), because they were rare in our samples. Although chrysophytes are detected in the Arctic (24), this group includes both photosynthetic and heterotrophic taxa that cannot readily be distinguished at this time based solely on the V4 region 18S rRNA gene sequence that we used to identify taxa, and it was therefore also omitted. Our large data set of 117 samples, collected from different regions, depths, and ice conditions, enabled us to examine the distribution of the five major groups at multiple taxonomic scales. Our first step was to identify the taxonomic diversity within the five groups. In general, MAST phylogeny is well studied, named clades form robust clusters based on sequence similarity (16, 25, 26), and reads can be identified by comparison with a reference database. However, clades from the other four HF groups are less well defined. Cryomonadida, Telonemia, and Picozoa have very few cultured representatives, and choanoflagellate phylogeny has recently been extensively revised (27), requiring an update of tools and reference databases used for taxonomic identification from sequence data. Because the uncertainty of the phylogenetic placement of short reads makes simple identification by sequence similarity ineffective, our approach for the four taxa with problematic phylogenies was to map reads onto phylogenetic trees using an environmental placement algorithm (EPA). Reads placed at the same node of a phylogenetic tree were considered to belong to the same phylotype. We then determined where these phylotypes occurred across Arctic Seas.

MATERIALS AND METHODS

Experimental design.

Existing data were drawn from four different studies that used pyrosequencing of the V4 variable region of the 18S rRNA gene to describe eukaryotic communities. All four studies were carried out over the Beaufort Sea, and the areas included the Canada Basin, Amundsen Gulf, and across the Mackenzie Trough (Table 1; Fig. 1; see also Table S1 in the supplemental material). The studies sampled surface waters or the subsurface chlorophyll maximum (SCM) as defined by Martin et al. (28) and sometimes both depths. The Mackenzie Trough study, which was in conjunction with the French International Polar Year project “Malina,” included additional depths above and below the SCM. One of the projects focused on sea ice sampled in the Amundsen Gulf region in spring 2008 (17). Sample collection and pyrosequencing are described in associated publications (9, 17, 19, 24, 29) and were similar for all data sets. To increase pan-Arctic coverage, additional samples that had been collected during various missions between 2005 and 2011 were also sequenced and analyzed using similar methods; these samples were from the Chukchi Sea, the Canadian Arctic Archipelago (Archipelago), Baffin Bay, Hudson Bay, and the Laptev Sea as detailed below. For most samples, 2,000 to 12,000 reads were analyzed per sample (see Table S1 in the supplemental material). Accession numbers for the NCBI Sequence Read Archive (SRA) for all data sets are given in Table 1. Metadata for stations from all studies, including sampling depths, is provided in Table S1 in the supplemental material.

TABLE 1.

Summary of studies analyzed for heterotrophic nanoflagellates

Region	Data set name (reference)	Collection dates (yr/mo)	No. of sites	Size (μm) fraction(s) selected for sequencing	SRAa accession no. for raw read data
Beaufort Sea	Amundsen Gulf (9)	2003–2010/July–November	11	<3	SRA029114
Beaufort Sea	Sea Ice (17)	2008/March–May	5	Pooled (<3 and >3)	SRA054160
Mackenzie Trough	Malina (19)	2009/July	6	<3	SRA063446
Beaufort Sea	Canada Basin	2007/August	4	>3	SRA099217
Chukchi Sea	ICESCAPE	2010/June–July	12	<3 μm and >3	SRP029300
Archipelago, Hudson Bay, Baffin Bay, Laptev Sea	Arctic Ocean Survey	2005–2011/July–October	20	Pooled (<3 and >3)	SRP040734

^aSRA, NCBI Sequence Read Archive.

FIG 1.

Stations sampled by the multiple studies in this paper: Beaufort Sea, Baffin and Hudson Bays, Chukchi Sea, and the Laptev and Siberian Seas. The map was created in Ocean Data View version 4 (R. Schlitzer, Ocean Data View, http://odv.awi.de, 2014).

Additional samples.

Chukchi Sea samples were collected onboard the U.S. Coast Guard ship (USCGS) Healy mostly from the Chukchi Shelf (Fig. 1) from 13 June to 22 July 2010 as part of the ICESCAPE project (http://www.espo.nasa.gov/icescape/). Seawater was collected into rinsed carboys from 30-liter Niskin-type bottles mounted onto a SeaBird carousel rosette equipped with a SBE9plus CTD. Genomic DNA was collected by sequentially filtering 5 liters of seawater through a 3-μm-pore-size, 47-mm-diameter polycarbonate (PC) filter (Millipore) and a 0.22-μm-pore-size Sterivex cartridge (Fisher Scientific) to obtain two size fractions, 0.22 to 3 μm (small) and >3 μm (large), to enrich for and concentrate organisms in the targeted size classes (30). Samples were preserved in a lysis buffer (50 mM Tris HCl [pH 8.3], 40 mM EDTA [pH 8.0], and 0.75 M sucrose) and frozen at −80°C.

The Archipelago, Baffin Bay, and Hudson Bay samples were collected in conjunction with ArcticNet missions (http://www.arcticnet.ulaval.ca/) from 2005 to 2011. Samples were collected aboard the Canadian Coast Guard Ship (CCGS) Amundsen as described above but with the following modifications. Seawater was collected using 12-liter Niskin-type bottles mounted on a rosette system (9, 19), and 6 liters was sequentially filtered as described above. In 2010 and 2011, RNAlater (Qiagen, Germantown, MD) was used as a preservative instead of the lysis buffer. Laptev Sea samples were collected onboard the Russian research vessel (RV) Viktor Buynitsky from 17 to 30 September 2007 as part of the 2007 Nansen and Amundsen Basins Observational System (NABOS) mission (http://nabos.iarc.uaf.edu/). Water was collected directly from 5-liter Niskin-type bottles and was filtered sequentially as above, preserved in the lysis buffer, and frozen at −80°C. These ArcticNet and NABOS samples were sequenced as part of the Arctic Ocean Survey (AOS) project (31).

DNA was extracted from filters and Sterivex cartridges using an AllPrep DNA/RNA Minikit (Qiagen). Primers and conditions for amplification and generating amplicons and reads followed the protocols described by Comeau et al. (9, 17). For the Chukchi Sea, the two size fractions were sequenced separately except for the <3-μm CHA1B (Kotzebue Sound) sample, which was lost (Table 2). Because of limited funds, the large- and small-size fractions for the AOS project were pooled after the first amplification step of the amplicon tag sequencing protocol (9) following the ratio of small and large cells as determined by microscopy counts.

TABLE 2.

Pyrosequencing raw data, filtering, and OTU statistics for data from the ICESCAPE 2010 study^a

Processing step	Value for indicated size fraction
	>3 μm (12 samples)	0.22–3 μm (11 samples)
Prefiltering
Total no. of reads	138,658	137,727
Mean length (bp)	343	308
Postfiltering
No. (%) of reads retained after quality control	64,139 (46)	49,451 (36)
Mean length (bp)	435	431
No. (%) after metazoans, fungi, and land plants removed	54,319 (85)	47,594 (96)
No. (%) after chimeras and poorly aligned reads removed	53,600 (99)	47,200 (99)
No. (%) after singletons removed	50,645 (94)	44,643 (95)
OTU analysis (97% level)b
No. of OTUs for all samples (nonredundant)	1,132	986
No. of OTUs for all samples (cumulative)	3,111	3,228
Mean no. of OTUs per sample	283	293

^aPercentages are given relative to reads available from the preceding step.

^bReads from station CN14 >3 µm are excluded from the OTU analysis because after metazoan reads (mostly a single genus of copepod) were removed, there were insufficient remaining reads for the analysis.

Sequence processing.

The software Qiime (32) was used for all sequence processing and downstream analyses. Reads were denoised using the default FLX settings in Qiime. Reads of <200 bp or >1,000 bp were discarded. Chimeras were removed using the Uchime algorithm (33) by checking both de novo and against the Silva database. Only nonchimeric sequences detected by both methods were retained. Reads were clustered into operational taxonomic units (OTUs) at ≥98% similarity as described in reference 9. Representatives from each OTU were aligned using MUSCLE (34) and assigned taxonomy using Mothur (35) with a minimum confidence score of 0.8 against our own curated 18S rRNA gene database, which includes Arctic Ocean 18S rRNA gene sequences from clone libraries (9).

Comparative phylogenetic community diversity was analyzed using unweighted UniFrac distance metrics (36), which clusters samples based on phylogenetic lineages. UniFrac distances were calculated using reads belonging to all five major HF groups and to MAST reads alone. Dendrograms were constructed using Randomized Accelerated Maximum Likelihood (RAxML v.7.2.7) software (37, 38) with a general time-reversible (GTR) model of nucleotide substitution using four discrete rate categories to approximate a gamma distribution. Distance matrices were used to build an unweighted pair group method using average linkages (UPGMA) tree. The robustness of UPGMA clustering was determined using jackknife analyses with the default parameters of 10 iterations of 140 reads per sample, which represented 70% of the total number of sequences per sample. A similarity percentage (SIMPER) test using OTU abundance, as implemented in PAST v3.01 (39), was used to test which taxa contributed most to the clustering by UniFrac.

Placement of reads on phylogenetic trees.

For phylogenetic analysis, 18S rRNA gene reference sequences were selected based on previously published phylogenies: Cryomonadida (15, 40), Telonemia (41), Picozoa (20), and Choanoflagellida (27, 42). Because there are relatively few Picozoa-related full-length 18S rRNA gene sequences in GenBank, we constructed two trees: one with only full-length sequences and one that omitted the V1 and V3 variable regions but included more environmental sequences. Many of the choanoflagellate sequences contained large insertions that were omitted from the analysis, as they are not taxonomically informative.

Alignments of short-read OTUs were checked against the reference sequences to look for potential artifacts, particularly errors in homopolymer length, which is poorly resolved using pyrosequencing (43). Short-read and reference alignments were then assembled into a single alignment by manually adding gaps in the sequence-editing program SeaView (44). Aligned reference sequences were then extracted and used to reconstruct a reference tree using RAxML as described above. Short reads were mapped onto the reference tree using the EPA of RAxML, which sequentially places each short query sequence (read) at each edge of a reference tree previously constructed with longer sequences and calculates the likelihood of the resulting tree (45). The resulting tree was visualized using NJ Plot (46). Only read placements with likelihood weight of >0.5 were retained. A combination of EPA and ordination with nonmetric multidimensional scaling (NMDS) run in R used the nodes identified in EPA as taxa, with an aim to identify phylotypes that could be indicators of regional differences or sampling techniques. Phylotypes were defined as reads that were placed at the same node.

Nucleotide sequence accession numbers.

Reads from the Chukchi Sea-ICESCAPE project were deposited in the NCBI SRA with accession number SRP029300, while reads from the multiple sites of the Arctic Ocean Survey project (31) are together under accession number SRP040734 (Table 1).

RESULTS

In total, approximately 1,020,000 reads passed the quality control steps. HF groups comprised 1.6 to 54% of eukaryote sequences in the water column samples and up to 60% in the ice samples (see Fig. S1 in the supplemental material). All of the HF phylum level groups were found in all of the samples.

Phylogenies and evolutionary placement of reads.

The reference trees of Cryomonadida, Telonemia, Picozoa, and the Choanoflagellida indicated the high phylogenetic diversity of the four groups, with well-supported branching orders, and usually recovered previously published clusters. Specifically, the Cryothecomonas clade (15) was recovered with high bootstrap support in the Cryomonadida phylogeny (Fig. 2). Within the Telonemia, we recovered the groups TEL1 and TEL2 described by Bråte et al. (41) with very high bootstrap support, along with their freshwater clades 1d and 2e (Fig. 3). Two Picozoa phylogenies were constructed, one from an alignment of 16 sequences and 1,633 characters and a second one from an alignment of 55 sequences and 1,168 characters. There was good agreement between the topologies, so the tree with a shorter residue range and more sequences was used for EPA. Some of the same clades as those described by Seenivasan et al. (20) were recovered, albeit with a different branching order (Fig. 4). In the choanoflagellate phylogeny, we recovered the Acanthoecida and Crespedida clades found by Nitsche et al. (27), although the latter had low bootstrap support (Fig. 5). Within the Acanthoecida, we recovered Stephanoecidae as paraphyletic and containing the Acanthoecidae clade. As a group, the choanoflagellates had only moderate bootstrap support, and overall the RAxML tree should be interpreted cautiously, as the 18S rRNA gene is poor at resolving deeper branches among the choanoflagellates (27).

FIG 2.

Phylogenetic mapping of Cryomonadida reads from all studies. The rooted Cryomonadida reference phylogenetic tree was constructed using maximum likelihood from an alignment of 92 sequences and 1,689 characters. Some non-Cryomonadida reference sequences have been omitted for clarity. Closed circles indicate nodes with bootstrap support of >50 (of 100). Reads are mapped onto the nodes marked with blue circles using the RAxML evolutionary placement algorithm; only placements with likelihood weight of >0.5 are shown. Pie charts show the proportion of reads from each study at a given node. The left scale bar indicates the number of substitutions per position. Protaspis longipes (*) was formerly identified as Cryothecomonas longipes (74), and Rhogostoma minus (**) was formerly identified as Lecythium sp. (75). Outgroups (not shown) are two radiolarian sequences and an acantharean (Spongaster tetras [AB101542], Lithomelissa setosa [HQ651801], and Acanthometra fusca [KC172856]).

FIG 3.

Phylogenetic mapping of Telonemia reads from all studies. The rooted Telonemia reference phylogenetic tree was constructed using maximum likelihood from an alignment of 52 sequences and 1,942 characters. Closed circles indicate nodes with bootstrap support of >50 (of 100). Reads are mapped onto the nodes marked with blue circles using the RAxML evolutionary placement algorithm; only placements with likelihood weight of >0.5 are shown. Pie charts show the proportion of reads from each study at a given node. The left scale bar indicates the number of substitutions per position. Labeled clades correspond to the system of Bråte et al. (41). Outgroups (not shown) are a haptophyte and a katablepharid (Prymnesium parvum [AJ246269] and Katablepharis japonica [AB231617]).

FIG 4.

Phylogenetic mapping of Picozoa reads from all studies. The rooted Picozoa reference phylogenetic tree was constructed using maximum likelihood from an alignment of 55 sequences and 1,168 characters. Closed circles indicate nodes with bootstrap support of >50 (of 100). Reads are mapped onto the nodes marked with blue circles using the RAxML evolutionary placement algorithm; only placements with likelihood weight of >0.5 are shown. Pie charts show the proportion of reads from each study at a given node. The left scale bar indicates the number of substitutions per position. Clades are labeled according to the system of Seenivasan et al. (20). Outgroups (not shown) are a haptophyte, a katablepharid, and a Telonemia (Prymnesium parvum [AJ246269], Katablepharis japonica [AB231617], and Telonema antarcticum [AJ564773]).

FIG 5.

Phylogenetic mapping of choanoflagellate reads from all studies. The rooted choanoflagellate reference phylogenetic tree was constructed using maximum likelihood from an alignment of 49 sequences and 1,963 characters. Closed circles indicate nodes with bootstrap support of >50 (of 100). Reads are mapped onto the nodes marked with blue circles using RAxML evolutionary placement algorithm; only placements with likelihood weight of >0.5 are shown. Pie charts show the proportion of reads from each study at a given node. The left scale bar indicates the number of substitutions per position. Outgroups (not shown) are two metazoan sequences, Mnemiopsis leidyi (AF293700) and Beroe ovata (AF293694), a sponge (AY348876), two ichthyosporeans (Y16260 and AF232303), and Corallochytrium (L42528).

Reads were mapped onto the reference RAxML trees to identify key taxa in different data sets. In total, 3,165 Cryomonadida reads, of an original 23,870, could be placed (Fig. 2). One phylotype (node a), found in sea ice and the Laptev Sea, was associated with the putative diatom parasite Cryothecomonas aestivalis. The node c phylotype within the Cryothecomonas clade was broadly distributed in all regions, though not found in sea ice, while the node d phylotype was found mostly in the Chukchi Sea. The most abundant phylotype (e; 1,394 reads) branched at the base of the Cryothecomonas clade and was composed almost exclusively of sea ice reads. One phylotype comprising 596 reads, mostly from sea ice, branched outside the Cryomonadida (node k), likely with the genus Ebria.

For Telonemia, 9,288 of 14,874 reads could be placed on the reference tree (Fig. 3). The largest number of reads (4,174) came from the Mackenzie Trough (Malina data set); the lowest number (79) came from the Laptev Sea. Phylotypes nodes b, g, h, and i were broadly distributed in most or all of the regions studied. Three Telonemia phylotypes (nodes a, c, and e) were found mostly in the Canada Basin, with a restricted to surface water samples. Nodes c and e were found in higher abundances in the RNA than the DNA sample from station PP-6.5, although this difference was not observed for other sites where both RNA and DNA were sampled (data not shown). Over 90% of the reads from the Mackenzie Trough were placed with environmental sequence Nor26.Telo.6, in clade 2d (node f). In contrast, reads from the Canada Basin were phylogenetically more broadly distributed through the marine clades in TEL2.

For the Picozoa, 12,902 of 37,904 reads could be placed on the reference tree (Fig. 4). The largest number of reads came from the Amundsen Gulf Time Series (3,925) and the Mackenzie Trough (2,685). Very few came from studies that sampled the >3-μm size fraction exclusively, consistent with the <2-μm size range reported for Picozoa (20). Reads were mapped to just four nodes, all of which were broadly distributed.

For the choanoflagellates, 8,690 of 17,216 reads could be placed on the reference tree (Fig. 5). The largest number of reads (2,402) came from the Hudson Bay; the lowest number (122) came from the >3-μm size fraction from the Chukchi Sea. The majority of reads (7,742 total) were placed with the tectiform acanthoecid Didymoeca costata (also referred to as Diplotheca costata by Doweld [47]; nodes f, g, and h). A few nudiform acanthoecid reads (64; node a) and Crespedida reads (420; nodes j and k) were detected. One hundred sixty reads were placed at the base of the choanoflagellate tree (node l) and could not be definitely identified.

The combination of EPA and ordination with NMDS enabled identification of phylotypes that appeared to be indicators of regional differences or sampling techniques. While broadly distributed taxa were found in all groups, more region-specific taxa were found for organisms that are usually >3 μm, such as Cryomonadida and Telonemia, than for smaller organisms such as MAST and Picozoa. For example, while Cryomonadida reads were found in all three >3-μm data sets, the phylotypes found in each one were different, with very little overlap (Canada Basin, node f; sea ice, nodes a, b, e, h, i, and k; Chukchi Sea, nodes c, d, g, and j) (Fig. 2 and 6).

FIG 6.

Nonmetric multidimensional scaling (NMDS) of heterotrophic flagellate reads in the Arctic, excluding sea ice, using taxa identified by EPA. Letters correspond to nodes from Fig. 2 to 5, while numbers refer to MAST clades.

HF community composition and beta diversity analysis.

Unweighted UniFrac distance measures revealed some clustering by region, albeit with low jackknife support (see Fig. S1 in the supplemental material). Both sea ice samples and the >3-μm size fraction of the Chukchi Seawater column formed separate clusters with a higher proportion of Cryomonadida sequences. The August and October 2009 Amundsen Gulf time series and SCM samples from the Mackenzie Trough (Malina data set) mostly clustered together. Within the Canada Basin surface water cluster, samples based on either RNA or DNA from the same site tended to cluster together, with the exception of those from station PP-6.5. Telonemia and choanoflagellates were more dominant in the Canada Basin. A SIMPER test identified the same four OTUs as the largest contributors to dissimilarity between clusters (defined at 85% similarity) for weighted and unweighted UniFrac, accounting for 17% of total dissimilarity. These were an acanthoecid choanoflagellate, a MAST-4, a MAST-7, and a member of the Telonemia. All were either not recovered or very rare in samples from the >3-μm fraction in the Chukchi Sea, Canada Basin, and sea ice. MASTs were always a greater proportion of reads in data sets from the <3-μm fraction and were dominated by clades MAST-1 and -7 (see Fig. S2 in the supplemental material). As with the analysis of all HF, clusters had very low jackknife support.

DISCUSSION

Here we examined high-throughput sequence data from nine regions across the Arctic, focusing on heterotrophic flagellates, a functional guild that consumes bacteria and picoeukaryotes and a key component of microbial food webs (48, 49). Although all of the major HF groups were found in all of the samples examined in this study, the taxonomic resolution provided by the EPA analysis revealed differences in distribution between clades for some of the HF groups. All Picozoa and choanoflagellate clades, as well as most MAST clades, were widely distributed. In contrast, some Cryomonadida and Telonemia phylotypes had more-restricted distributions.

Data interpretation.

Inferring taxon abundance from the proportion of reads and interpreting this information in an ecological context are subject to a number of caveats, including limitations inherent to amplicon sequencing, biases introduced by the collection and analysis of samples from different studies, and incomplete taxon sampling in reference trees and databases.

The greater sampling depth achievable by high-throughput sequencing enabled the detection of taxa that had not been recorded using fluorescent in situ hybridization (FISH) or 18S rRNA gene clone libraries. For example, a phylotype related to Cryothecomonas aestivalis, a species that parasitizes diatoms (50), was detected by high-throughput sequencing of sea ice and samples from the Laptev Sea (Fig. 2). While cryothecomonad-like cells have been observed to be associated with diatoms during the spring bloom in northern Baffin Bay (C. Lovejoy, unpublished data), Thaler and Lovejoy (15) did not find any Arctic C. aestivalis-related sequences in 18S rRNA clone libraries or in public databases. Cryothecomonas was originally described as a free-living predator in sea ice (51); however, to our knowledge, no similar FISH studies using a Cryothecomonas-specific probe have been carried out in ice environments.

While high-throughput sequencing provided a more complete picture of taxonomic diversity than did earlier studies, some anomalies were evident. For example, the MAST-1A clade had the highest proportion of reads in nearly all samples, in contrast with the cell count data of Thaler and Lovejoy (16) and Lin et al. (52), who found MAST-1A at much lower relative concentrations than MAST-1B or -1C in the Arctic and the North Pacific, respectively. MAST-1A also tends to be retrieved more frequently in Arctic Ocean clone libraries (7, 22). This could be due to PCR bias toward certain taxa (11), possibly because of preferential primer binding or RNA structure (12). Alternatively, FISH studies suggest that MAST-1A cells tend to be larger than MAST-1B cells (16) and could also have multiple 18S rRNA gene copies. In this context, as with other eukaryotic microbes, variable copy numbers of 18S rRNA genes in different taxa make inference of taxon abundance relative and should not be treated as equivalent to absolute estimates (53).

The opportunistic nature of the collection and analysis of the different samples from studies with slightly different methods could introduce biases that mask true biogeographical signal. Of particular concern is the interannual variability resulting from the changes that have taken place in the Arctic over the last 10 years (9, 14). The effect of this type of variability could not be analyzed directly since, with the exception of the Amundsen Gulf Time Series, no region was sampled more than once.

Finally, EPA was able to place only 15 to 60% of reads. This loss of information resulted from our rather stringent cutoff likelihood weight of >0.5 in EPA. Likelihood weights are calculated for the entire tree after adding the inserted query sequence; it is therefore logical that lower likelihood weights were found for taxa such as Cryomonadida and Picozoa, for which large numbers of reads mapped to poorly supported regions of the reference tree (for example, Cryomonadida phylotype e or Picozoa phylotype c), whereas higher likelihood weights were found for Telonemia, for which most reads were mapped to well-supported nodes. This highlights the need for improved reference trees for better identification.

Because of the caveats discussed above, we make an effort to discuss only those trends that are supported by very marked differences in read abundance or by a combination of more than one line of evidence (EPA, clustering, NMDS) and are therefore presumably robust.

Regional and water mass signals.

Over most of the Arctic, different water masses result in strongly stratified water columns (54), which display very sharp vertical gradients in resources (28). The protist community composition in the Arctic is closely associated with water masses rather than geographic location (19, 55, 56). Samples from the Amundsen Gulf Time Series were all from the interface of the Polar Mixed Layer and water originating from the Pacific; the SCM forms at this point because of tradeoffs between sufficient light and nutrients (28). Vertical structuring of the HF community by water mass along the Mackenzie Shelf has already been discussed by Monier et al. (19), who found that samples from the same water mass were more similar to each other than samples from different water masses at the same station. Our new analysis, comparing a much larger data set, found much less evident separation detected by UniFrac analysis of the HF communities (see Fig. S2 in the supplemental material). Factors that could mask the water mass signal by contributing to similarities and differences between samples include interannual variability as well as the hydrography of the Arctic Ocean itself, where water masses are modified over distances (57). There was little difference between surface and SCM samples from Hudson Bay and the Archipelago, which may reflect a shallower, well-mixed water column in the Archipelago, and the complex hydrography of the Hudson Bay Region. The samples from Baffin Bay were collected with an aim to document the maximum diversity and included both Arctic- and Atlantic-influenced water masses (31). The complex interleaving of water masses in Baffin Bay (58) could also mask the effects of water mass or depth, because sampling did not occur at a sufficiently fine vertical scale to capture the structure of water column. Interestingly, many of the <3-μm size fraction Chukchi Sea samples tended to cluster with Amundsen Gulf SCM samples, consistent with the Pacific Origin water being nearer the surface in the Chukchi Region (57).

Indicator taxa.

A combination of EPA and ordination with NMDS enabled identification of phylotypes that appear to be indicators of regional differences or sampling techniques. More region-specific taxa were found for larger organisms such as Cryomonadida and Telonemia than for smaller organisms such as MAST, Picozoa, and choanoflagellates, although broadly distributed taxa were found in all groups.

Cryomonadida reads dominated sea ice samples from the Beaufort Sea and were also abundant in water column samples from the Chukchi Sea (see Fig. S1 in the supplemental material). However, there was little overlap of phylotypes between the two data sets (Fig. 2 and 6), and the most common phylotypes (nodes e and k [Fig. 2]) found in the Amundsen Gulf Sea Ice were rare in the Chukchi Sea. This suggests that some Cryomonadida may be more adapted to open water rather than ice, such as species associated with nodes c and f (Fig. 2).

The Chukchi Sea samples were collected from sites within or near the marginal ice zone, and the associated ice melt may have released sea ice organisms, including Cryomonadida, into the water column. However, in July 2010 the region was characterized by widening and merging polynyas (U.S. National Ice Centre [http://www.natice.noaa.gov), as is typical for this time of year (59), that would be more consistent with the surface waters having been ice free for much longer. Interestingly, no sea ice reads were placed with reference sequence MC5-1, a cultured bacterivorous strain isolated from sea ice in Antarctica (60). Instead, nearly all reads from this data set were found in the node e phylotype (node k, also with a high abundance of sea ice reads, is probably not a member of the Cryomonadida but of the Ebriida). This phylotype must have a very broad halotolerance, since it includes an environmental sequence (KRL01E39) from a brackish lake, while the salinity of brine channels within sea ice can be many times that of seawater (61, 62).

Telonemia clades were most abundant in the offshore Mackenzie Trough, associated with the node f phylotype, and the Canada Basin, with the greatest relative proportions in surface waters (see Fig. S1 in the supplemental material), associated with phylotypes a, c, and e. In contrast, Telonemia in the Canada Basin were both more abundant and phylogenetically more diverse (Fig. 3). In the Canada Basin, the anticyclonic Beaufort Gyre causes the vertical downward stacking of water masses, with different densities at the center creating a bowl-like structure (57). The downwelling effect depletes the surface waters of nutrients and deepens the nitracline and the SCM to approximately 60 m or deeper (63). Low nutrient concentrations found in surface waters support low concentrations of phytoplankton, as exhibited by low concentrations of chlorophyll a (chl a) offshore of the Mackenzie shelf at the time of sampling (64). While chl a concentrations were not measured in all samples, the average surface concentration in the Canada Basin was one-half that of the southern Beaufort Sea and more than 2 orders of magnitude lower than in the Chukchi Sea (see Table S1 in the supplemental material). Such distributions suggest that Telonemia have an advantage in low-productivity systems and are able to maintain high populations of one or a few phylotypes, for example, node f in the Mackenzie Trough (Malina data set) and nodes c and e in the Canada Basin (Fig. 3); however, this trend cannot be statistically evaluated with our present data. Higher relative abundances of the Telonemia phylotypes c and e, along with lower abundance of MAST-1A (UniFrac Analysis [see Fig. S2 in the supplemental material]), were found in the RNA than in the DNA template samples from site PP-6.5, which was adjacent to the marginal ice zone of Canada Basin (C. Lovejoy and K. Scarcella, unpublished data). No such difference was observed for sites CB-15 and CB-17.5, located under >90% ice cover, implying that ribosomal activity detected from 18S rRNA for these taxa may be associated with community changes in a mobile ice environment; unfortunately, no RNA sample was available from the other marginal ice site, PP-2. The node l phylotype, containing over one-half the reads from sea ice, was associated with an environmental sequence from a sediment sample and may represent a taxon adapted to interstitial environments.

In the Chukchi Sea samples, where size fractions were analyzed separately, the smaller size fraction was enriched in Picozoa and MAST (about 20-fold enrichment in the Chukchi Sea data), consistent with cell diameters given for HF taxa in the literature (16, 20, 25, 65). For all samples, Picozoa and MAST were geographically less distinct in the Arctic than Cryomonadida and Telonemia, which are generally larger and in the nanoplankton size range (2 to 20 μm) (15, 66). On the contrary, our results highlighted the ubiquity and abundance of the clade MAST-7 (see Fig. S2 in the supplemental material), which comprised 25% of all MAST reads in our data and was enriched in the <3-μm data sets. While this clade has been identified in 18S rRNA gene clone libraries (67), to our knowledge there have been no studies on MAST-7 using fluorescent in situ hybridization (FISH), which could give a more reliable picture of its abundance in the Arctic. Similarly, EPA analysis showed that Picozoa in the Arctic contained only four broadly distributed phylotypes (Fig. 4), indicating that this group may be dominated by a few taxa. The single cultured representative of Picozoa is thought to consume colloidal material and perhaps small viruses (20). If so, this group might be less constrained by food availability than other HF that may have a preference for specific bacteria or phytoplankton prey, thus enabling the Arctic picozoans to grow across different regions and depths. However, it may also indicate that better taxon coverage is needed in the reference tree or that V4 rRNA gene sequences are highly conserved in this group.

While many eukaryotic environmental gene surveys, including in the Arctic, have focused on the <3-μm size fraction (9, 68), the results reported above suggest that the large-size fraction of HF can be indicators of regional differences. In NMDS, samples from both size fractions in the Chukchi Sea overlapped (Fig. 6), but the >3-μm size fraction occupied a larger ordination space, suggesting that small organisms were being retrieved in the larger-size fraction but not vice versa. Size fractionation by filtration is imperfect, as flexible cells may be able to pass through pores smaller than their longest dimension, while pores that are clogged will retain cells smaller than the pore diameter. In addition, free DNA from larger cells can be retained on the 0.2-μm filter (69). In practice, size fractionation enriches for a particular size class (30), and this needs to be taken into account.

Most choanoflagellate phylotypes were found in all of the studied regions. There is a rich literature on the distribution of choanoflagellate taxa using microscopic methods; however, comparison with molecular results is complicated by historic misidentifications and a lack of cultured representatives. Most studies in polar water columns and sea ice have focused on Acanthoecida (70–73), reflecting both the comparative wealth of diagnostic features in this loricate group and the dominance of acanthoecids in marine planktonic environments, which is corroborated by our EPA results. Nearly all choanoflagellate reads that could be mapped onto the tree belonged to the order Acanthoecida, whose members are considered to be mainly planktonic. Ambiguities in the reference tree made it impossible to place reads at a species level; however, in most regions the majority were placed in the genus Didymoeca. del Campo and Massana (42) identified a Didymoeca clade containing sequences from the Arctic water column, the Mediterranean, and a deep anoxic basin, indicating a possibly cosmopolitan distribution for this phylotype.

Conclusions.

We reanalyzed new and publicly available V4 18S rRNA gene short-read data to detect biogeographic signal for heterotrophic flagellates in the Arctic Ocean. Placement on phylogenetic trees allowed reads to be identified at a high level of taxonomic resolution, although improved reference trees will be needed to fully characterize communities. While many phylotypes were widespread, others were specific to regions and environments. Phylotypes with restricted distributions were more likely to be found among taxa of larger cell size, such as Cryomonadida and Telonemia. These taxa may make good targets for future research, functioning as indicators of the dramatic changes in ice regimes that are already taking place across the Arctic.