Sampling campaigns were performed in 2014 as described in , , , and , during five different months, to capture the various stages of the Arctic primary production: January (6–15 January, winter), March (5–10 March, pre-bloom), May (15 May–2 June, spring bloom), August (7–18 August, post-bloom), and November (3–10 November, early winter). The cruises were undertaken west and north of the Svalbard archipelago, where warmer Atlantic water in the West Spitsbergen Current meets colder water from the Arctic Ocean (Fig. ). Bottom depth varied from 327 m (November station N03) to c. 3000 m (March station M05). The area and locations for each sampling campaign were as similar as possible, but constrained by the sea ice cover, from 79 to 82.6∘ N. During each cruise, transects of three to six stations were sampled at three or four depths: in the epipelagic zone at 1 m and at the deep chlorophyll maximum (usually between 15–25 m) and in the mesopelagic zone at one or two depths, as a rule 500 and 1000 m, or as deep as the bathymetry of the station permitted.

DNA was extracted with the DNeasy Plant mini kit (Qiagen), according to the protocol from the manufacturer, except for the following step: to disrupt the thick cell walls of certain protist groups, the frozen samples in cryovials were incubated at 95 ∘C for 15 min and then shaken in a bead-beater 2× 45–60 s. Subsequently, 4 µL RNase was added, and the lysate was incubated on a heating block at 65 ∘C for 15–20 min, with vortexing in between. Purity and quantity of the extracted DNA was assessed with NanoDrop.

The V4 region of the 18S rRNA gene was amplified with the primer pair 18S TAReuk454FWD1 (5'-CCAGCASCYGCGGTAATTCC-3') and V4 18S Next.Rev (5'-ACTTTCGTTCTTGATYRATGA-3') , which yields a fragment of 410–420 base pairs. This primer pair is an improvement over the widely used primers developed by , which have been used in more than 60 studies. The Piredda primers aim to reduce the biases against Haptophyta seen in the Stoeck primers . The samples were prepared for Illumina sequencing with a so-called dual-index approach e.g.,, where a 12 bp internal barcode was added to both the forward and reverse amplification primers for the initial amplification. In order to pool several samples into one library preparation, 19 unique barcodes for each direction were used. The internal barcodes were designed to give a balanced distribution of the four bases, following the recommendations of . Polymerase chain reactions (PCRs) consisted of 12.5 µL KAPA HiFi HotStart ReadyMix 2× (KAPA Biosystems, Wilmington, MA, USA), 5 µL of each primer (1 µM) and 10 ng DNA template and PCR-grade water to a final volume of 25 µL. The PCR was run on an Eppendorf thermocycler (Mastercycler, ep gradient S, Eppendorf), with an initial denaturation step at 95 ∘C for 3 min, followed by 25 cycles of denaturation at 98 ∘C for 20 s, annealing at 65 ∘C for 60 s, and elongation at 72 ∘C for 1.5 min, with a final elongation step at 72 ∘C for 5 min. The reactions were performed in triplicate for each sample and pooled prior to purification and quantification. The length of the PCR products was assessed by gel electrophoresis. In all samples, there was a strong band at about 470 bp and no other bands (data not shown). The PCR products were purified with AMPure XP beads (Beckman Coulter, Brea, USA) using the standard protocol with elution buffer EB (Qiagen), quantified with a Qubit dsDNA High-Sensitivity kit (Thermo Fisher, Waltham, MA, USA) and pooled in equal concentrations to create nine pools with ca. 19 samples in each. The pools were sent to library preparation at the Norwegian Sequencing Centre (Oslo, Norway) and GATC GmbH (Constance, Germany) with the KAPA library amplification kit (Kapa Biosystems). Further quality control of the amplicons was carried out with Bioanalyzer at the sequencing centers prior to Illumina sequencing. Due to worldwide supply problems with the Illumina MiSeq chemistry in 2015, the sequencing was done with a modified HiSeq protocol on two HiSeq runs at the GATC in October 2015. This modified protocol yielded 250 bp paired-end reads. The HiSeq sequencing runs were spiked with 20 % PhiX (viral DNA added to ensure homogeneity of bases during sequencing). To assess variation between DNA extracts and annealing temperature, we separately sequenced replicate DNA extractions and replicate PCR runs with 60 ∘C annealing temperature for a few samples (indicated in Table ). After initial analysis of the HiSeq data, samples with low numbers of reads were re-amplified with 30 cycles with the original DNA as template to increase the concentration of the PCR product and re-sequenced with Illumina MiSeq at the Norwegian Sequencing Centre. The MiSeq protocol yielded 300 bp paired-end reads. In total, we sequenced 199 samples separately. The taxonomic composition of sequencing replicates of the same sample was inspected visually, and in all cases the different sequencing protocols were found to give similar taxonomic compositions.

PhiX sequences were removed, and the raw reads were sorted according to the Illumina index by the Illumina software at the sequencing provider. For the HiSeq datasets, the samples within each Illumina library were demultiplexed with cutadapt v2.10 with Python 3.6.11 , requiring 0 errors in the internal barcodes. The amplification primers were removed with cutadapt v2.8 with Python 3.7.6, with setting --trim-n (trim N's on ends of reads). The reads were denoised and merged with dada2, v1.16 . For the HiSeq reads the settings were truncLen = c(240,200), minLen = c(240, 200), truncQ = 2, maxEE = c(10, 10), and max_number_asvs = 0. Chimeras were detected with isBimeraDenovo with default settings and removed with removeBimeraDenovo, with “method_chimera” = “pooled”. For the MiSeq reads truncLen and minLen were set to c(270, 240), and the other settings were the same as for HiSeq. The reads were subsequently classified with assignTaxonomy, the dada2 implementation of the naive Bayesian classifier method , against the Protist Ribosomal Reference Database, version 4.12.0 (https://github.com/pr2database/pr2database/releases/tag/v4.12.0, last access: 25 January 2021; ). ASVs with less than 90 % bootstrap value at class level and/or which comprised fewer than 10 reads in total were removed. As this study is focusing on the protists, all reads assigned to Metazoa and Viridiplantae (Embryophyceae) were excluded from the processed ASV tables (Table ).

Preparation of the ASV tables was done in R v. 3.6.0 . To be able to compare ASV richness between samples (i.e., the number of ASVs in each sample), the DNA samples were subsampled to equal read number. Prior to subsampling, data from fastq files that map onto the same size-fractionated DNA sample were merged by taking the sum of the read number for each ASV. The merging was done to increase the number of reads in the samples that initially had a low read number after HiSeq sequencing. The DNA samples were subsampled according to size fraction as follows: 0.45–3 µm: 40 000 reads; 3–180 µm: 88 000 reads; 3–10 µm: 40 000 reads; 10–50 µm: 40 000 reads; and 50–200 µm: 8 000 reads. It should be noted that the number of reads in the 3–180 µm fraction samples was in some cases lower than the subsample size. Subsampling to equal read number was performed 100 times, and the average read number per ASV was used, rounded to 0 decimals. Subsampling was done with the function rrarefy() from the “vegan” package , v. 2.5-7. The low number of protist reads in the 50–200 µm fraction was due to a high proportion of Metazoan reads in this fraction. To calculate the ASV richness of each sample, the subsampled ASV table was transformed to presence–absence. An overview of the available versions of the ASV table is given in Table . To assess whether the sequencing depth was sufficient, we plotted rarefaction curves for each DNA sample and calculated the slope at the endpoint of the curve with the functions “rarecurve” and “rareslope” respectively. Figures were made with the R package “ggplot2” . In addition to the figures and tables presented in the paper, interactive versions of figures, tables, and supplementary material are available as a Shiny app (available at https://micropolar-protists.metapr2.org/, last access: 20 September 2021). Interactive figures and tables were made with the packages “DT” and “plotly” , respectively.

In total we obtained 44 water samples from Niskin bottles and 8 net hauls, which were fractionated into 140 and 15 size-fractionated samples, respectively (the DNA isolation of the 10–50 µm fraction from the net haul taken at station P04 in May failed). The samples from Niskin and net hauls are referred to as DNA samples in the following, and they are denoted month_station_depth_minfract_maxfract or month_station_net_minfract_maxfract, respectively. On some DNA samples, we performed replications of DNA extraction, PCR with variable annealing temperature, and/or replicate sequencing. Thus, one or more fastq-file pairs can map onto the same DNA sample. The fastq files were deposited individually to the European Nucleotide Archive (ENA, project accession number PRJEB40133) and are referred to as a “sequencing event” (“seq_event”) in Table ). In total the dataset consists of 199 sequencing events, some of which were merged, to form in total 155 DNA samples. Description of metadata available for each fastq-file pair can be found in Table . Table  describes all environmental parameters obtained from each water acquisition event (i.e., from the Niskin samples). These are referred to as “env_sample” and labeled month_station_depth.

After quality filtering, dada2 processing, removal of chimeras, and non-target taxonomic groups, the dataset comprised 6536 protist ASVs, corresponding to 32 164445 reads. After subsampling to equal number of reads per sample within each size fraction, the dataset was reduced to 6430 ASVs and 5 729 358 reads. Number of ASVs per division or class within each size fraction, after subsampling, is shown in Table . In total, we recovered 3339, 2720, 2799, 1153, and 3172 ASVs in the 0.45–3, 3–10, 10–50, 50–200, and 3–180 µm size fractions, respectively. Note that the numbers are not directly comparable, as the fractions were not obtained from the same number of samples (e.g., 3–180 µm were only sampled in January and March). Syndiniales and Dinophyceae had the highest number of assigned ASVs, with 2166 and 1723, respectively. Ciliophora, Bacillariophyta, Radiolaria, and Chlorophyta had between 400 and 200 assigned ASVs each (Table ).

Slopes of rarefaction curves at the endpoint, after subsampling, ranged from 0 to 0.014 (Fig. ), which means that for every 1000 extra reads sequenced, we could expect to find between 0 and 14 new ASVs . There was no correlation between the number of ASVs detected in a sample and the slope of the rarefaction curve (r2=-0.13, p=0.11), which means that the DNA samples with a low number of ASVs were not necessarily undersampled.

The proportional taxonomic composition of the metabarcoding reads, at division or class level, is shown in Fig. . The taxonomic composition of the ASV richness in each sample is shown in Fig. . The metabarcoding data reveal variation in taxonomic composition by both season and depth, in all size fractions. In the following, the fractions are defined as follows: 0.45–3 µm is picoplankton, 3–180 µm is nano-micro, 3–10 µm is small nanoplankton, 10–50 µm is large nanoplankton, and 50–200 µm is microplankton. All the major protist groups varied from less than 1 % of the reads to up to 99 % for the most abundant (e.g., Syndiniales in the picoplankton fraction and diatoms in the microplankton fractions; Table ).

In January (winter) and March (pre-bloom), heterotrophic or parasitic groups (e.g., certain dinoflagellates, Syndiniales, and Picozoa) were dominating at all depths. In the picoplankton size fraction, the parasitic dinoflagellate group Syndiniales had the highest relative abundance these months, with up to 99 % of the reads, followed by the heterotrophic group Picozoa, with up to 35 % of the reads, and Pseudofungi with up to 12 % (previously categorized as marine stramenopiles, MAST). Syndiniales also had the highest ASV richness in all samples. In the nano-micro fraction, Dinophyceae had generally higher relative abundance, with 20 %–55 % of the reads in most samples. Syndiniales and Picozoa had up to 82 % and 40 %, respectively. Syndiniales also had the highest ASV richness in this fraction, followed by Dinophyceae. Other heterotrophic groups notably present in this fraction were Pseudofungi and Radiolaria, with 2 %–20 % of the reads each, and Ciliophora and Choanoflagellida with up to 6 % of the reads. ASVs assigned to phototrophic groups (e. g. diatoms, haptophytes, and chlorophytes) were detected in these months but constituted less than 3 % of the reads in all samples.

The May samples were characterized by higher proportions of phototrophs in all size fractions. In the pico- and small nanoplankton fractions, there was a pronounced difference between the epipelagic and mesopelagic samples this month. In the picoplankton fraction, Chlorophyta (mainly represented by the genera Micromonas and Bathycoccus) had high relative abundance in the epipelagic samples, with 17 %–43 % of the reads. In the small nanoplankton fraction, Haptophyta (mainly represented by the genus Phaeocystis) and Dinophyceae were the most abundant groups in the euphotic samples with 25 %–47 % and 14 %–39 % of the reads, respectively. The mesopelagic samples in the pico- and small nanoplankton fractions were characterized by a high abundance of Syndiniales, with 47 %–85 % of the reads. In these fractions, ASV richness was generally higher in the mesopelagic than in the epipelagic samples. Syndiniales generally had the highest number of ASVs, despite having lower relative abundance. In the large nanoplankton and microplankton fractions, diatoms were dominating in both the epi- and mesopelagic samples, with up to 99 % of the reads. Dinoflagellates (Dinophyceae) was the second most abundant group in the large nanoplankton fraction, with up to 50 % of the reads. In the microplankton fraction, Phaeocystis was also abundant in certain samples, with up to ca. 30 % of the reads. In the net haul samples from May, the diatoms were dominating with up to 97 % of the reads. Dinophyceae had 10 %–11 %, and Haptophyta constituted 11 % in the microplankton fraction from station P01. These fractions generally had lower ASV richness than the pico- and nanoplankton, and there was no clear difference in ASV richness by depth. The groups with the highest ASV richness in these samples were Dinophyceae, Bacillariophyta, and Syndiniales.

In August, in the picoplankton fraction of the epipelagic samples, Dinophyceae had the highest relative abundance, with 13 %–64 %. Haptophyta had ca. 4 %–14 % and Chlorophyta ca. 7 %–25 % in these samples. In the mesopelagic samples in this fraction, Syndiniales also dominated in August, with up to 68 % of the reads. Radiolaria accounted for 10 %–13 % of the reads in these samples, whereas Picozoa had 3 %–15 %. Picozoa relative abundance also reached up to 12 % in the epipelagic samples. In the nanoplankton size fraction, Dinophyceae was dominating, with 27 %–79 % of the reads. Similar to in May, Syndiniales generally had the highest ASV richness in the picoplankton fraction. In the nanoplankton fraction, Syndiniales and Dinophyceae had similar ASV richness. In the large nanoplankton fraction, Dinophyceae dominated with 31 %–88 % of the reads. Diatoms constituted up to 41 % of the reads, and there was no clear difference in proportion of this group between the epi- and mesopelagic samples. In the microplankton fraction, the diatoms dominated, with 30 %–73 % of the reads. Dinophyceae was the second most abundant group in this fraction, with 3 %–46 % of the reads. In the net haul samples, Dinophyceae and diatoms were the most abundant in the large nanoplankton and microplankton fraction, with 64 %–77 % and 12 %–20 % of the reads, respectively. In the microplankton fraction Radiolaria were also abundant, representing 7 %–30 % of the reads. ASV richness was slightly higher than in May in the large nanoplankton and microplankton fraction. Syndiniales and Dinophyceae also had the highest richness in these fractions, followed by diatoms and ciliates.

In November, the proportion of reads assigned to phototrophs was less than 3 % in most samples in the pico- and nanoplankton. In the large nanoplankton and microplankton fraction, diatom reads constituted 1 %–33 %. In the picoplankton fraction, Syndiniales and Picozoa were the most abundant, with 40 %–75 % and up to 25 % of the reads, respectively. Radiolaria represented 44 % of the reads in the sample N04_1000. Dinophyceae was the most abundant group in the fractions between 3 and 50 µm, with 28 %–76 % of the reads. In the microplankton fraction, Radiolaria, Dinophyceae, and Syndiniales were the most abundant with up to 77 %, 43 %, and 42 % of the reads, respectively. In the net hauls, Ciliophora was also abundant, with up to ca. 30 % of the reads in each size fraction. ASV richness was generally higher this month than in May and August, especially in the small and large nanoplankton fractions. Syndiniales and Dinophyceae also had the highest richness this month.