Samples were collected from 9 stations in the Amazon River plume-influenced region of the WTNA (between 0 and 13°N and 44–57°W) as part of the Amazon INfluence on the Atlantic: CarbOn export from Nitrogen fixation by DiAtom Symbioses (ANACONDAS) project. Data presented here are from a cruise aboard the R/V Knorr May 22–June 24, 2010, during the period of peak plume discharge (Figure 1) and a large scale DDA bloom of Hemiaulus-Richelia (Goes et al., 2014). Underway monitoring of sea surface salinity (SSS) and photosynthetic pigments as well as satellite monitoring of plume indicators such as chromophoric dissolved organic matter (CDOM) were the measurements used to determine station location. Station categorizations were determined by SSS at time of the respective tow for each sample. We sampled from 1 plume station (23) with SSS < 30; 4 mesohaline stations (2, 3, 19, and 21) with SSS between 30 and 35; and 4 oceanic stations (5, 6, 20, and 27) with SSS > 35.

Mesozooplankton were collected with a 1-m Multiple Opening and Closing Net and Environmental Sensing System (MOCNESS; Wiebe et al., 1976) fitted with ten 202 μm mesh nets. Daytime tows were performed between 1000 and 1400 h while nighttime tows were collected between 2200 and 0200 h (local time). Four depth intervals within the top 150 m were sampled with the MOCNESS, although molecular analysis was performed on animals collected from the two shallowest depth intervals (0–25 and 25–50 m) since shipboard observations and cell abundances detected the highest number of microscopically identified diazotrophs (e.g., Trichodesmium, DDAs) in the upper 50 m. Once the nets were onboard, zooplankton were immediately anesthetized with carbonated water to prevent gut evacuation (Gannon and Gannon, 1975). Samples were subsequently split into either ¼ or ½ of the total sample volume using a Folsom plankton splitter, then size fractionated using nested sieves into the following size fractions: 0.2–0.5, 0.5–1.0, 1.0–2.0, 2.0–5.0, and >5.0 mm. Each size fraction was then concentrated onto a pre-weighed, 0.2 mm Nitex mesh filter, rinsed with Milli-Q to remove salt, and frozen at -80°C until processed in the laboratory.

To compare grazing at stations with different diazotroph assemblages, mesozooplankton selection for molecular analysis was based on phytoplankton distribution from Goes et al. (2014) utilizing pigment analysis, physical and chemical properties described in Loick-Wilde et al. (2016), and phytoplankton cell counts presented here. Cells for microscopy counts were collected directly from the CTD at distinct depths (5–6) in the upper 150 m and approximately at midday (local noon). Stations 2, 23, and 27 were sampled once per day for 2 days. The entire contents of the Niskin bottles were gravity filtered onto a 47 mm diameter Poretics (Millipore) membrane filter with a pore size of 10 μm. The filter was removed and mounted onto an oversize microscope slide (75 mm × 50 mm × 1 mm) and examined under 400× magnification using a Zeiss Axioskop Epifluorescence microscope (Zeiss, Berlin, Germany). Colonial and single trichomes of Trichodesmium, as well as the symbiotic Richelia, were identified by phycoerythrin and chlorophyll a (Chl a) excitation with green (510–560 nm) and blue (450–490 nm) excitation wavelengths, respectively. The diatoms associated with Richelia were identified as Hemiaulus or Rhizosolenia spp. based on cell ultrastructure. In many instances the entire filter was scanned and symbiotic diatom cells enumerated as number of Richelia heterocysts L^-1 for each respective diatom and for Trichodesmium as colonies L^-1 (col L^-1) or single trichomes L^-1. In some instances when symbiotic cells were at high densities, several smaller grid areas (62.5 μm²) of the filter were scanned, and at least 1000 cells were enumerated and corrections to cell abundance were made by the area scanned (Supplementary Table S1).

Zooplankton from the 0.5–1.0 and 1.0–2.0 mm size fractions were selected after visual inspection under a stereomicroscope. Harpacticoid copepods (Macrosetella gracilis and Miracia spp.) and the decapod shrimp Lucifer faxoni were identified to species or genus level, while calanoid copepods, fish larvae, and decapod larvae from the family Thalassinidae were not (Supplementary Table S2). These targets were chosen based on mesozooplankton community composition of each station, onboard microscopic counts from the cruise, pigment concentrations (Goes et al., 2014), and stable isotope analysis (Loick-Wilde et al., 2015) for the same cruise. Calanoid copepods were selected, as they were present across all salinities and with only a few exceptions were the most abundant taxa in all samples, while the harpacticoid copepods were selected because of known associations with Trichodesmium. The L. faxoni, fish, and decapod larvae were included given their periodic high abundance. The taxa analyzed were mostly calanoid copepods (and proportionately representative of the zooplankton community composition; Conroy and Steinberg, in preparation), with the relative taxonomic representation as follows: calanoid copepods (69.4%), harpacticoid copepods (19.4%), decapod larvae including crab megalopae and Thalassinidae larvae (4.2%), fish larvae (4.2%), and L. faxoni (2.8%) (Supplementary Table S2).

Animals were sorted and placed in autoclaved artificial seawater and inspected for exterior contamination with phytodetritus in appendages and mouthparts. Animals were picked clean of any obvious large particles using a needle and forceps, and following the procedure of Boling et al. (2012) subsequently rinsed five times with autoclaved artificial seawater. Animals were inspected again for contaminating phytoplankton and cyanobacteria using blue (450–490 nm) and green (510–560 nm) excitation filters on an epifluorescent compound microscope at 200–450× magnifications. This procedure ensured animals chosen for molecular analysis were phytoplankton-free on their exterior. Between 25 and 50 animals were pooled per DNA extraction (Supplementary Table S1); the number of individuals varied depending on size and availability of target and the results from preliminary PCR assays which determined the lowest number of pooled individuals needed for consistent amplification (see below). Samples were extracted using a modification to the Qiagen DNeasy^® Blood and Tissue Kit Animal Tissue (Spin-Column) protocol. Briefly, a 12-h lysis step was performed and all recommended reagent volumes were halved during the extraction. The final elution volume was 35 μl in the provided AE buffer.

We performed qPCR assays for three of the major diazotrophs in the WTNA, two DDAs (het-1, Richelia associated with Rhizosolenia diatoms and het-2, Richelia associated with Hemiaulus diatoms), and Trichodesmium spp. using the previously described oligonucleotides (Church et al., 2005a,b; Foster et al., 2007) and a modified TaqMAn assay (see below). A total of 72 samples were analyzed for all three targets (het-1, het-2, Trichodesmium) with the exception of two calanoid copepod samples collected from St. 2 that were not run with the het-1 assay, and six samples from St. 3 were not run with the Trichodesmium due to low template (Supplementary Table S2). In preliminary attempts (data not shown) to optimize the extraction and detection by qPCR, we identified a minimum number of individuals for replicable amplification. From those results we used a minimum of 25 pooled individuals per extraction but, unless limited by abundance of the taxa in a sample, we pooled 50 individuals per extraction (Supplementary Table S2).

For all TaqMan PCR, the 12 μL reactions contained 6.25 μL TaqMan 2X Master Mix (Applied Biosystems), 0.5 μL forward and reverse 0.5 μM primers, 0.25 μL fluorogenic 0.5 μM probe, 2.5 μL of nuclease free water, and 2 μL of DNA template. All reactions were run in triplicate, and for the no-template controls, 2 mL of 5-kD filtered nuclease free water was added to each reaction. All qPCR assays were performed on an ABI 7500 Fast machine (Applied Biosystems) with the following thermal cycling conditions: 50°C for 2 min, 95°C for 10 min, and 45 cycles of 95°C for 15 s, followed by 60°C for 1 min. Gene copy abundances were calculated from the mean cycle threshold (C_t) value of three replicates and the standard curve for the appropriate primer and probe set. For each primer and probe set, triplicate standard curves were made from 10-fold dilution series ranging from 10⁸ to 1 gene copies per reaction. The standard curves were made from linearized plasmids of the target nifH. Regression analyses of the results (number of cycles = C_t) of the duplicate standard curves were analyzed in Excel. In some samples only one of the three replicates produced an amplification signal; these were noted as detectable, but not quantifiable (dnq). For samples where two or three of the replicates amplified the values were averaged and reported as nifH-gene copies per organism. We note that while we report gene copies per organism, as is convention with qPCR, we do not scale our numbers to an estimation of feeding rate. Instead we consider amplification of our targets as confirmation of direct grazing on either DDAs or Trichodesmium (see detailed explanation in Discussion).

Sequencing of 16S rRNA genes in zooplankton (containing their gut contents) followed a modified protocol (Arfken et al., 2015) using the Ion Torrent PGM sequencer (Life Technologies). DNA concentrations from the extractions were measured on a NanoDrop 2000 (Thermo Scientific) and PCR was performed on samples normalized to 5 ng μl^-1 per reaction, except for fish larvae, which were normalized to 20 ng μl^-1 (see Supplementary Table S3 for samples included in sequencing). The V4 hypervariable region of 16S rRNA genes was targeted with 515F and 805R primers for PCR reactions using GoTaq Green Master Mix (Promega) (Caporaso et al., 2011). The 805R primers have barcodes with fusion sequences while the 515F contains fusion sequences only. PCR reactions were performed as follows: an initial denaturing step for 3 min at 95°C, then 30 cycles of 30 s at 95°C, 1 min at 55°C, 1 min at 72°C, and finally at 72°C for 5 min. PCR products for each sample were then combined and sequencing was conducted. Barcoded samples were sequenced using the Ion Torrent 400 base pair (bp) sequencing protocol with samples pooled onto a 316 chip. We note that for the 16S rRNA analysis we consider our results representative of the zooplankton “microbiome,” similar to other studies (Scavotto et al., 2015; Shoemaker and Moisander, 2015), however, we focus our results on the cyanobacteria and are confident that the cleaning methods adapted from Boling et al. (2012) were adequate enough and our results represent the cyanobacteria consumed by the various mesozooplankton.

Sequencing output was downloaded from the Torrent Server using Torrent Suite v3.0 to obtain the FastQ file. A total of 33 libraries were created using the barcoded sequences. Denoising was performed with Acacia (Bragg et al., 2012) and the open-source bioinformatics program Mothur (Schloss et al., 2009) was used to trim and align the sequences. Chimera sequences were removed using UCHIME and remaining sequences for each library were classified using the Greengenes database¹ at confidence of greater than or equal to 0.80. Relative abundance of each taxa was calculated by dividing the number of classified sequences by total number of sequences in each library. All cyanobacteria sequences identified through Greengenes database were subsampled and subsequently reclassified through the SILVAngs pipeline² with the SILVA 128 reference library (Quast et al., 2013). Additionally, principal coordinate analysis (PCoA) was performed on square-root transformed relative abundance data in PRIMER 7 to identify patterns in microbial diversity between libraries. The raw sequence reads have been submitted to National Center for Biotechnology Information³ and are available under Bioproject number PRJNA387277.

At all stations, except Station 6, the Hemiaulus-Richelia DDA was more abundant than the Rhizosolenia-Richelia DDA. As such, below we only report the former, although all cell counts are available in Supplementary Table S1 for comparison.

The diazotroph assemblage within the low salinity plume (SSS < 30) at station 23 was dominated by Hemiaulus-Richelia DDAs (39.4–7.29 × 10³ heterocysts L^-1; Supplementary Table S1) primarily in the upper 75 m (Figure 2). At several stations in the mesohaline plume (30 > SSS > 35; stations 2, 3, 19, and 21), cell counts for stations 2 and 19 showed the DDA assemblage described by Goes et al. (2014) was dominated by Hemiaulus-Richelia, and the highest cell density of the cruise (7.66 × 10⁵ heterocysts L^-1; Supplementary Table S1) occurred at 20 m at Station 2 (Figure 2). At station 19 in the upper 100 m, Hemiaulus-Richelia and Trichodesmium were similar in abundance, ranging from 34.0–2.39 × 10³ heterocysts L^-1 and 415–1.59 × 10³ trichomes L^-1, respectively (Supplementary Table S1). In comparison, stations 3 and 21 had low DDA abundances. At station 3, Hemiaulus-Richelia abundance (8.4 heterocysts L^-1) was low in surface waters (∼2 m), free-living Richelia cells in the surface 20 m were the highest observed on the cruise (34.9–338 heterocysts L^-1; Supplementary Table S2). Goes et al. (2014) characterized the phytoplankton community of Station 21 as dominated by Trichodesmium and Synechococcus spp. based on pigments, and microscopy confirmed Trichodesmium as the most dominant cyanobacterial diazotroph in the phytoplankton assemblage, with 4.6–141 trichomes L^-1 in the surface 50 m (Figure 2).

Oceanic stations (SSS > 35; stations 5, 6, 20, and 27) were dominated by Trichodesmium and Synechococcus spp. Trichodesmium was present at all four oceanic stations but was not always the most abundant diazotroph in the phytoplankton assemblage (Supplementary Table S1) as previously described in Goes et al. (2014) based on pigments. At Stations 6 and 27 Hemiaulus-Richelia were more abundant than Trichodesmium, particularly at Station 27 with a 1–2 order of magnitude difference (5.9–4.66 × 10⁴ heterocysts L^-1 vs. 25.6–1.25 × 10³ trichomes L^-1 in the surface 150 m, respectively; Supplementary Table S1).

DDAs– Both the het-1 (Richelia associated with Rhizosolenia diatoms) and het-2 (Richelia associated with Hemiaulus diatoms), targets were successfully amplified from the gut content extractions (Table 1 and Supplementary Table S2). Het-2 was the most common target amplified across all extractions, with a total of 25 samples amplified–24 from calanoid copepods and one from Macrosetella gracilis. Of those, 11 (10 calanoids and 1 M. gracilis), were detectable not quantifiable (dnq). The remaining samples detected het-2 with a range of 1.6–16.8 nifH-copies/organism. All stations except for station 3, the furthest inshore mesohaline station, had samples that amplified the het-2 target.

In comparison to het-2, the other DDA, het-1, was detected only in calanoid copepods collected at night at Station 19 (Table 1 and Supplementary Table S2). Each of the two size fractions at this station had very low detection with the smaller calanoid copepods dnq and larger calanoid copepods 0.10 nifH copies/organism. No pattern was observed between night (n = 12) and day (n = 13) samples for het-2, whereas the only het-1 detection occurred at night (Supplementary Table S2).

Trichodesmium– The nifH of Trichodesmium spp. was also successfully detected in the gut contents of the mesozooplankton samples (Table 1 and Supplementary Table S2). Twenty samples out of 72 (14%) showed amplification: 14 in calanoid copepods (7 dnq), 4 in M. gracilis (3 dnq), and 2 in crab megalopae (1 dnq) (Figure 2). All oceanic stations (St. 5, 6, 20, and 2; Figure 1), as well as one mesohaline station (St. 19), included in this analysis detected Trichodesmium. Values ranged from 1.1 to 4.0 Trichodesmium nifH copies/organism with the highest value found in crab megalopae. No pattern was observed between day (n = 11) and night (n = 9) samples for detecting Trichodesmium.

Sample selection for 16S rRNA NGS analysis was guided by our qPCR assays and resulted in the 33 samples selected for analysis. Not all samples with amplification from the qPCR were included due to template limitation, but a representative number of samples provided over 1.7 million sequences for analysis (see Supplementary Table S3 for NGS samples).

PCoA of 16S rRNA sequences showed a tight clustering of two groups (Figure 3). Calanoid copepods (blue squares in Figure 3) and all harpacticoid copepods (green triangles in Figure 3) clustered by taxa. Other variables analyzed (salinity, size fraction, depth interval, time of day) resulted in no significant grouping.

Proteobacteria were the most abundant phyla represented in our samples, with 16 of the 33 samples having over 50% of sequences associated with proteobacteria (Figure 4). Cyanobacteria represented between 0 and 45% of bacterial composition across all samples and represented 11.3% of total sequences (Supplementary Table S3). In the non-calanoid copepod samples (n = 14), the decapod shrimp L. faxoni, had cyanobacterial sequences associated with its gut which represented >1% of total sequences. Harpactacoid copepods M. gracilis and Miracia spp. had cyanobacterial sequences present but were all <1% of total sequences, and the fish larvae sample analyzed was devoid of cyanobacterial sequences (Figure 4 and Supplementary Table S3).

When composition of only cyanobacterial sequences (n = 197,298) was analyzed with the GreenGenes database, the most abundant class was Synechococcophycideae (Figure 5). When investigated to genus level, the most abundant within this class was Synechococcus (n = 83,745, representing 42.4% of all cyanobacterial sequences). Prochlorococcus was the next largest contributor at the genus level (n = 22,734 representing 11.4% of all cyanobacterial sequences) although 21.2% of cyanobacterial sequences in Synechococcophycideae were unidentifiable to genus classification. Additionally, in a several samples chloroplast sequences composed up to 100% of the cyanobacterial sequences (Figure 5).

Reclassification of cyanobacterial sequences based on SILVA 128 database agreed that Synechococcus and Prochlorococcus were still the most abundant representatives (56 and 26%, respectively). However, at station 5 and station 21 UCYN-A and UCYN-B sequences were found within the smallest size fraction calanoid copepod samples, representing 1–2% of sequences.

DDAs–Stable isotope studies in the subtropical and tropical Atlantic Ocean (Montoya et al., 2002; Landrum et al., 2011; Loick-Wilde et al., 2015), including the region of our study, established that new N attributed to N₂-fixation from diazotrophs is incorporated in the planktonic food web, but the pathways were not elucidated (i.e., through the microbial loop, exudation, or grazing). The qPCR results of the gut content show N_D can enter the food web via consumption of DDAs, as both DDA targets (het-1 and het-2) were detected in the zooplankton gut contents. This is the first evidence for consumption of DDAs. These results are consistent with DDA distributions observed in our microscopy, where het-2 was the most dominant of the two DDAs present during our study, as well as in a prior studies of the Amazon-influenced WTNA (Carpenter et al., 1999; Foster et al., 2007; Luo et al., 2012). Consumption of DDAs is also consistent with the decreased δ¹⁵N content of zooplankton reported by Loick-Wilde et al. (2016) in the mesohaline plume, particularly at Stations 2 and 19 (their Figure 9). At St. 19, the only station where het-1 was detected in the guts of mesozooplankton, microscope counts observed both Hemiaulus and Rhizosolenia symbioses were present. However, Hemiaulus-Richelia symbioses (het-2) were 1–2 orders of magnitude more abundant and found deeper in the water column than the Rhizosolenia-Richelia symbiosis (het-1).

The pattern of DDA distribution in the WTNA largely follows the nutrient and surface salinity gradients outlined in Subramaniam et al. (2008), with DDAs occurring in the mesohaline region of the plume (Goes et al., 2014; Loick-Wilde et al., 2015). However, the diazotrophic cyanobacterial counts and molecular results reported here indicate there is not a strict boundary between the salinity regimes for DDA presence, particularly along the transition from mesohaline to oceanic waters. All four of the oceanic stations included in this study had low abundances of het-2 symbioses along with amplification of this target in zooplankton gut contents. These stations were characterized as an oceanic assemblage with abundant Trichodesmium and Synechococcus by Goes et al. (2014) (note Station 6 was not included in their study). Therefore, het-2 consumption at all four oceanic stations sampled is unexpected and can be explained by plume meanders or development of frontal zones between mesohaline and oceanic waters. These physical features are common in river plumes and represent areas of zooplankton aggregation and increased grazing (Woodson et al., 2005; True et al., 2015; Woodson and Litvin, 2015) and could promote episodic, opportunistic feeding in oceanic mesozooplankton.

Our results, while novel, are not unexpected. Conroy et al. (2016) characterized the mesozooplankton food web from grazing rate estimations at low salinity plume and mesohaline waters with SSS < 33 as an export style food web. Export food webs are characterized by shorter diatom and mesozooplankton food chains compared to retention food webs dominated by microzooplankton and the microbial loop (Michaels and Silver, 1988; Legendre and Michaud, 1998; Pomeroy et al., 2007). Both DDAs investigated form chains, often suggested as a grazing deterrent, yet copepods actively graze diatoms of similar size (Bergkvist et al., 2012). Therefore in the plume and mesohaline stations DDA consumption supports a shorter export food web; DDA consumption we observed at oceanic stations is likely more occasional, but when it occurs could increase export.

Trichodesmium were consumed by zooplankton at all four oceanic stations, as well as at two mesohaline stations. Calanoid copepods were again the primary consumers, although the highest nifH gene copies organism^-1 were amplified from crab megalopae. Despite the higher abundance of M. gracilis at oceanic stations relative to plume stations (Conroy et al., in preparation) and its known association with, and grazing on, Trichodesmium (O’Neil and Roman, 1994; O’Neil et al., 1996; O’Neil, 1998), this species only showed one amplification for Trichodesmium that was not dnq (Figure 2 and Supplementary Table S2). This is not to say M. gracilis does not graze on Trichodesmium, which was reported to be a pathway of atmospheric nitrogen incorporation into zooplankton in the eastern tropical Atlantic (Sandel et al., 2015), but that this result is an artifact of our methodology, discussed in detail below.

Prior to our study there was little evidence of calanoid copepod grazing on Trichodesmium, and a general consensus emerged that it did not occur or was severely limited (Capone et al., 1997; Carpenter and Capone, 2008). The detection of Trichodesmium in the guts of calanoid copepods in the WTNA is novel and builds on other limited evidence for calanoid grazing on Trichodesmium globally (Hawser et al., 1992; Guo and Tester, 1994). Calanoid copepods were typically the most abundant organism across all stations in the three smallest size fractions (0.2–0.5, 0.5–1.0, and 1.0–2.0 mm) (Conroy et al., in preparation), as also found in zooplankton community composition studies in the subtropical North Atlantic (Steinberg et al., 2008; Eden et al., 2009). Thus, grazing by copepods on Trichodesmium in the oligotrophic tropical and subtropical ocean where Trichodesmium is abundant (Capone et al., 1997, 2005) has potential for a significant input of N_D into the food web. Furthermore, these results support the observed low δ¹⁵N of mesozooplankton at oceanic stations (Loick-Wilde et al., 2015), was due, at least in part, to direct grazing on Trichodesmium. Similar evidence was recently reported in the Eastern tropical North Atlantic where δ¹⁵N of zooplankton was correlated with Trichodesmium spp. colony abundance (Sandel et al., 2015).

Other work from stable isotope analysis suggests that decreases in δ¹⁵N in Trichodesmium-dominated waters is due to N_D exudation and incorporation into the food web via the microbial loop (Montoya et al., 2002; Mulholland, 2007) rather than direct grazing, predominately due their potentially allelopathic toxins (Hawser et al., 1992). In the eastern tropical North Atlantic Ocean, Sandel et al. (2015) observed decreased δ¹⁵N in carnivorous, omnivorous, and Trichodesmium-grazing copepods. Aside from the known Trichodesmium grazers included in that study (M. gracilis and M. efferata) they did not suggest direct consumption of Trichodesmium by other copepod species. However, Hawser et al. (1992) showed that toxicity was species dependent and that not all Trichodesmium spp. are toxic to zooplankton. Furthermore, Guo and Tester (1994) investigated the effect of Trichodesmium sp. on the calanoid copepod Acartia tonsa after a meander from the Gulf Stream transported Trichodesmium inshore toward Albemarle Sound, North Carolina. They found no toxic effects of Trichodesmium on A. tonsa when fed healthy cells, but observed toxic effects, including distended guts and mortality, when fed aging or senescing cells or when treated with a filtered cell homogenate (Guo and Tester, 1994). Given the ephemeral exposure A. tonsa has to Trichodesmium as a predominantly coastal copepod, it is reasonable that calanoid copepods exposed to Trichodesmium during development in the open ocean would have a similar ability to consume Trichodesmium. Furthermore, a study from the tropical North Atlantic between the Cape Verde Islands and Barbados reports that δ¹⁵N values for zooplankton and Trichodesmium are similar, so that direct consumption is the likeliest explanation (McClelland et al., 2003, see their Figure 6). Zooplankton grazing on Trichodesmium is further supported by zooplankton-cyanobacteria interactions from freshwater habitats. Kâ et al. (2011) performed feeding experiments with calanoid copepods, cladocerans, and rotifers and found grazing could be a significant factor in controlling filamentous cyanobacteria. While we are unable to scale our numbers to estimate the grazing impact on Trichodesmium (see below), our results support that Trichodesmium is directly consumed by mesozooplankton.

Pairing the highly specific qPCR approach with a 16S rRNA NGS approach provided insight into the broader zooplankton-cyanobacterial dynamics in the WTNA. Our 16S rRNA sequences show abundant phyla that varied between samples, with proteobacteria and cyanobacteria consistently a large percentage of all sample sequences. Similar to findings of Shoemaker and Moisander (2015) from the subtropical North Atlantic, our results indicate a distinct partitioning of microbes based on taxonomic groups rather than environmental factors (Figure 3). While an understanding of the complete microbiome of zooplankton is important (Scavotto et al., 2015; Shoemaker and Moisander, 2015), particularly in the oligotrophic ocean where microenvironments (e.g., zooplankton) support unique bacterial assemblages (Azam and Malfatti, 2007), we limit the scope of our discussion to the cyanobacteria given their potential role in primary production, N₂ fixation, and carbon export.

Predominance of Synechococcus and to a lesser extent Prochlorococcus sequences within the most abundant cyanobacteria class (Figure 5) is similar to results from the same cruise investigating diversity of free-living and particle-associated cells. Metatranscriptomic studies conducted by Satinsky et al. (2014) showed that the prokaryotic assemblage was dominated by the Synechococcaceae family of cyanobacteria (Satinsky et al., 2014, See Supplementary Figure S1 of that study). Those results are limited to one station, which we did not sample; however, it does provide insight into why our sequences were dominated by Synechococcus and Prochlorococcus. Synechococcus and Prochlorococcus are globally abundant (Flombaum et al., 2013) but generally considered too small to be directly consumed by most mesozooplankton. On the other hand, consumption of the latter are often considered indirectly through feeding on marine snow aggregates or fecal pellets (Wilson and Steinberg, 2010), or by ingesting microzooplankton which previously consumed small cells. Satinsky et al. (2014) showed a higher percentage of Synechococcaceae cyanobacteria in “particle-associated” sequences compared to “free-living,” while HPLC pigment data showed abundant Synechococcus throughout the plume and WTNA (Goes et al., 2014). We conclude that predominance of Synechococcus and Prochlorococcus sequences in the crustacean zooplankton are likely from consumption of cyanobacteria-containing aggregates and/or consumption of microzooplankton, which had consumed picocyanobacteria previously.

The presence of UCYN-A and UCYN-B sequences in two different calanoid copepod samples indicates a pathway of N_D directly into the food web (Zehr, 2011; Thompson et al., 2012, 2014). Scavotto et al. (2015) is the only other study to show consumption of UCYN-A by zooplankton with their results showing the calanoid copepod Acartia spp. regularly had UCYN-A in full gut contents. To our knowledge there are no studies reporting UCYN-B consumption by zooplankton. UCYN-A is comparable in size to Synechococcus and Prochlorococcus, but UCYN-A is suggested to be an obligate, or at least mutualistic, symbiont with a larger haptophyte (Thompson et al., 2012; Farnelid et al., 2016) and UCYN-B can be found in free-living, colonial and symbioses with diatoms. Therefore, UCYN-A and UCYN-B may be directly targeted by zooplankton or consumed as components of larger sinking aggregates in the water column.

We did not analyze water column UCYN-A or UCYN-B abundance; however, results from prior studies in the tropical Atlantic suggest both are widely distributed, particularly in the eastern Atlantic and outer most regions of the plume-influenced WTNA (Foster et al., 2007; Goebel et al., 2010; Moisander et al., 2010). Globally, UCYN-A and UCYN-B are observed throughout the tropics and thus our results suggest a broad potential pathway for N_D from unicellular diazotrophs into planktonic food webs, particularly in areas such as the south Pacific where high concentrations for both are observed (Moisander et al., 2010; Farnelid et al., 2016).

Lastly, we offer explanations for the chloroplast sequences identified in the cyanobacterial sequences (Figure 5). Our NGS protocol required a PCR amplification step and it is possible that these chloroplasts could represent a bias introduced during amplification (Acinas et al., 2005). More likely is that the sequences represent phytoplankton, likely cyanobacteria, that were ingested but are not included in the most recent GreenGenes reference taxonomy (McDonald et al., 2012).

While molecular methods to quantify zooplankton grazing rates are becoming more common (Nejstgaard et al., 2008; Troedsson et al., 2009; Cleary et al., 2012, 2016; Shoemaker and Moisander, 2015), we hesitate to extend our qPCR results beyond a qualitative assessment of grazing. The nifH Taqman probes are highly specific, thus other non-targeted organisms consumed, which our NGS analysis confirms, would not be detected by the qPCR assays but could interfere with the amplification (Kanagawa, 2003; Nejstgaard et al., 2008). On the other hand, given the evidence of polyploidy in heterocystous cyanobacteria (Griese et al., 2011; Sukenik et al., 2012) and recently observed in Trichodesmium (Sargent et al., 2016) there is potential to overestimate the nifH copy number by qPCR. However, target DNA degradation and inefficient DNA extraction in the gut is a significant factor in underestimating zooplankton grazing, particularly in copepods, with molecular methods (Simonelli et al., 2009; Troedsson et al., 2009). Conroy et al. (2016) observed elevated grazing in both the low salinity plume stations and in the intermediate mesohaline stations utilizing the gut pigment method, yet our qPCR assays all yielded low gene copies per organism regardless of station category. Similarly, Nejstgaard et al. (2008) found a pattern of underrepresentation of gut contents when comparing qPCR estimates to gut pigment estimates. Regardless of the low gene copies, we are confident our results are indicative of grazing on DDAs and Trichodesmium given the high specificity of our qPCR assays and our sanitation techniques. In order to account for low gene copies due to gut degradation, we suggest future work include controlled grazing experiments using a culture of the targeted diazotroph for analyses so that a differential length amplification qPCR (dla-qPCR) method, similar to that utilized in Troedsson et al. (2009), could account for DNA degradation, and be used to estimate grazing rate on diazotrophs. Furthermore, the molecular methods utilized in this study could be paired with fluorescent in situ hybridization as well as analysis of gut contents through traditional electron microscopy. Lastly, there is the potential for contamination from environmental DNA (eDNA) that at present is not accounted for in this or other studies. While likely a low percentage of the genetic material analyzed, methodologies such as the one used here, that will likely have low amplification, need to consider a method for determining contamination from eDNA.