Introduction
The eastern tropical North Atlantic (ETNA) region is influenced by an
eastern boundary upwelling system (EBUS) off northwest Africa, which along
with nutrient supply via Saharan dust deposition, fuels one of the most
productive ocean regions in the world. A moderate oxygen minimum zone (OMZ)
is associated with this EBUS, with lowest oxygen (O2) concentrations
just below 40 µmol kg-1 present at intermediate depths
(Chavez and Messié, 2009; Jickells et al., 2005; Karstensen et al.,
2008).
O2 records over several years from the Cape Verde Ocean Observatory
(CVOO) mooring (located at 17∘35′ N, 24∘15′ W, Fig. 1)
confirmed the well-ventilated character of the ETNA. However, the
observation of distinct events of very low-O2 concentrations (< 1 µmol kg-1)
at depths around 40 to 100 m over periods of more
than 1 month challenged our understanding of the biogeochemistry in that
area (Karstensen et al., 2015a). The meridional current
structure observed during these low-O2 events revealed the passage of
anticyclonic modewater eddies (ACME) crossing the CVOO mooring
(Karstensen et al., 2015a). The ocean is filled with eddies
(Chelton et al., 2011) but only a few of them have the
dynamical and biogeochemical boundary conditions that support formation of a
low-O2 core. Anomalous low salinity within the ETNA low-O2
eddies suggested the water mass originated from the EBUS off Mauritania,
which was confirmed by analyzing sea-level anomaly data. In combination with
other data from the upwelling region, Karstensen et al. (2015a) showed that
O2 concentrations decreased over a period of a few months during
westward propagation of the eddies into the open north Atlantic Ocean.
Respiration in these eddies was estimated to be about three to five times
higher than typical subtropical gyre values (Karstensen et al., 2008).
(a) MODIS-Aqua 4 km monthly mean chl a distribution in
the ETNA (mg m-3) in November 2013. Markedly increased chl a
concentrations are associated with the low-oxygen ACME, located between 21c
and 22∘ W and 17.5 and 19∘ N. Analyses and visualizations
were produced with the Giovanni online data system, developed and maintained
by the NASA GES DISC. Eddy location indicated by sea level anomaly (SLA)
during the time of the two surveys: (b) first eddy observation; +
denotes the eddy_1 station, (c) second eddy observation + denotes
the eddy_2 station, an additional station was sampled at the eddy rim for C
uptake measurements, indicated by the blue +. White triangle marks the
sampling station for the potential source water of the eddy. The dashed
circles indicate the location of the eddy during the R/V Islandia survey, the
black circle indicates the eddy location during the R/V Meteor survey, and the
dashed black line indicates the direction of eddy propagation. Sampling
stations are shown with white triangles.
Mesoscale eddies are increasingly recognized as biogeochemical hot-spots of
basin-wide relevance for the world's oceans (Altabet et al., 2012; Baird
et al., 2011; Chelton et al., 2011; McGillicuddy et al., 2007; Oschlies and
Garcon, 1998; Stramma et al., 2013). Upward nutrient supply to the euphotic
zone through mesoscale eddy dynamics enables intense primary productivity
(Lévy et al., 2001, 2012; McGillicuddy et al., 2007).
Classically, primary producers in the ETNA open waters area are dominated by
a range of diatom clades, flagellates and cyanobacteria (Franz
et al., 2012), but so far no specific information on the primary producers
in productive ETNA eddies has been reported. As a result of enhanced primary
production in the surface, increased organic matter export flux below the
euphotic zone is expected, which in turn supports increased respiration at
intermediate depths. Indeed, particle maxima a few meters above the O2
minimum have been reported based on autonomous observations of
O2-depleted eddies in the ETNA (Karstensen et al.,
2015a), indicating enhanced organic matter export and providing environments of
enhanced remineralization (Ganesh et al., 2014). Observations
from a low-O2 eddy from the ETNA revealed a remarkable impact on all
productivity-related processes in that particular system
(Fischer et al., 2015). Estimated productivity was
three-fold higher in the surface layer compared to surrounding waters along
with a multiple times increase in mass flux in bathypelagic during the eddy
passage. Furthermore, Fiedler et al. (2015)
determined export flux derived from carbon remineralization rates within the
eddy and found a 3–4-fold enhanced export flux compared to background
conditions in the open-ocean ETNA.
O2-depleted conditions are supposed to act as a critical switch for the
marine microbial community, both with regard to functionality and diversity.
O2 begins to limit oxidative pathways and reductive pathways are
induced (Stewart et al., 2011; Ulloa et al., 2012; Wright et al., 2012). A
loss in microbial diversity related to vertical O2 gradients has
previously been described for the Pacific Ocean (Beman and
Carolan, 2013; Bryant et al., 2012), but to date no comparable data are
available from the ETNA. O2-loss related microbial community shifts and
modified functionality are supposed to favor heterotrophic communities
dominated by Flavobacteria, α- and γ-Proteobacteria, which
efficiently recycle organic matter (Buchan et al., 2014).
Furthermore, marine nitrogen (N) and carbon (C) cycling are significantly
altered under low O2 conditions (Vaquer-Sunyer and Duarte,
2008; Wright et al., 2012). Substantial N loss (Altabet et al.,
2012) along with enhanced nitrous oxide production
(Arévalo-Martínez et al., 2015) has been described
in low-O2 eddies in the OMZ off Peru in the eastern tropical South
Pacific.
Classically, the N cycle in the open ETNA is assumed to be dominated by
nitrification. An N loss signal is not present due to comparably high
background O2 concentrations (≥ 40 µmol kg-1, (Löscher et al., 2012;
Ryabenko et al., 2012)). However, any drop in O2 concentration in the water column, as
potentially induced by the low-O2 eddies, could potentially activate
anammox and/or denitrification. During recent decades, the ETNA OMZ has been
expanding both in terms of vertical extent and intensity and is predicted to
expand further in the future (Stramma et al., 2008) with
unknown consequences for the ecology and biogeochemistry of that system.
Thus, it is critical to understand the biogeochemical response to changing
O2 concentrations in that region.
In this study, we investigated differences in microbial community structure
in an O2 depleted eddy, surrounding ETNA open waters, and upwelled
waters on the Mauritanian shelf. This was achieved using a combined
high-throughput 16S rDNA amplicon sequencing/qPCR approach along with carbon
uptake rate measurements and hydrochemical observations. This study aimed to
understand the microbial community response to O2 depleted conditions
with regard to primary production and remineralization in these
poorly-described anomalies, to improve understanding of the sensitivity of
the ETNA biogeochemistry to future ocean deoxygenation.
Material and methods
Data collection
Remotely sensed sea level anomalies (SLA), in combination with temperature
and salinity data measured by Argo floats (an overview is presented by
Schütte et al., 2015) were used for general eddy identification and
tracking in this area. After identification of a low-O2 eddy candidate
that was propagating towards CVOO, a pre-survey was started using autonomous
gliders (see Karstensen et al., 2015b). Once the glider data had confirmed
the low O2 concentration in the candidate eddy, a ship-based survey was
started. First, we performed a survey with the Cape Verdean R/V Islandia on
6 March 2014 (samples from this survey are further referred to as eddy_1),
followed by a second survey with the German R/V Meteor (cruise M105; 19 March
2014; samples from this survey are further referred to as eddy_2). Moreover,
the background signal (i.e. waters outside the eddy) was measured, in order
to compare the eddy with the typical open-ocean ETNA environment. For this
purpose, we used metagenomic samples from the CVOO time series monitoring
site (collected on 19 March 2014 during cruise
M105). Samples from the Mauritanian shelf collected during R/V
Meteor
Cruise M107 (station 675, 18.22∘ N/16.56∘ W, collected on
24 June 2014) represent data from the eddy
formation area. Station 675 was chosen according to its location within the
area that Schütte et al. (2015) identified as the region of eddy
formation and further because of the observed low O2 concentrations of
33.9 µmol kg-1 at 115 m depth (which corresponds to a
potential density of σT= 26.4 kg m-3, thus similar to the
core density of minimal O2 concentrations in the eddy).
In addition to metagenomic sampling, carbon uptake measurements were
performed during the R/V Meteor M105 survey at two stations: no. 186
(profile 10, 19.3∘ N, 24.77∘ W) and no. 190 (profile 15,
18.67∘ N, 24.87∘ W, see Fig. 1c, blue crosses).
Water sampling and hydrographic parameters
Discrete samples for salinity, dissolved O2 and nutrients on all
surveys were taken from a CTD rosette equipped with Niskin-bottles. The CTD
data were calibrated against salinity samples and CTD oxygen probe data (SBE
43 Clark electrode sensor) were calibrated against O2 concentrations,
determined following the Winkler method using 50 or 100 mL samples. Salinity
and nutrient concentrations were determined as described in Grasshoff et al. (1999).
The CTD on R/V Meteor was equipped with double sensors for
conductivity, temperature, and oxygen. Calibration followed standard
procedures (GO-SHIP Manual; Hood et al., 2010).
Oxygen respiration
In order to estimate the net O2 consumption as a potential driver for
microbiological community shifts a simple calculation was performed as
follows:
ΔO2=O2(S)-O2(E),
where O2(S) denotes the lowest O2 concentration detected on the
shelf (36.69 ± 6.91 µmol kg-1 at σT= 26.3 ± 0.15 kg m-3,
cruise M107, average of shelf stations
between 18.10∘ N/16.59∘ W and 18.25∘ N/16.45∘ W).
This region was chosen as it was identified
(Schütte et al., 2015) to be the
area where the eddy most likely originated. O2(E) denotes the lowest
O2 concentration measured in the eddy core at the same potential
density (4.8 µmol kg-1 at σT= 26.35 kg m-3
during M105).
The daily O2 loss rate (ΔO2d) was calculated as follows,
assuming a lifetime of 180 days of the eddy (Schütte et al., 2015):
ΔO2d=ΔO2/180.
Chlorophyll a measurements
Sea water samples (0.5–1 L) for chlorophyll a (Chl a) analyses were filtered
(200 mbar) on GF/F filters (25 mm, 0.7 µm; Whatman, Maidstone, UK).
Filters were transferred to a plastic vial and 1 mL of MilliQ water was
added. Filters were immediately frozen at -20 ∘C and stored for at
least 24 h. Afterwards, 9 mL acetone (100 %) was added to the vials and
the fluorescence was measured with a Turner Trilogy fluorometer (Sunnyvale,
CA, USA). Calibration took place using a Chl a standard dilution series
(Anacystis nidulans, Walter CMP, Kiel, Germany). Chl a concentrations were determined as
described by Parsons et al. (1984).
Molecular methods
Seawater samples were taken from the Niskin-Bottles at selected CTD casts.
For nucleic acid purification 2 L seawater was rapidly filtered (exact
filtration volumes and times were recorded continuously) through 0.2 µm
polyethersulfone membrane filters (Millipore, Billerica, MA, USA). The
filters were immediately frozen and stored at - 80∘C until
further analysis. Nucleic acids were purified using the Qiagen DNA/RNA
AllPrep Kit (Qiagen, Hilden, Germany) with modifications as previously
described (Löscher et al., 2012).
Extracts of DNA and RNA were quantified using a spectrophotometer (Thermo
Fisher Scientific, Waltham, MA, USA). To remove DNA from RNA extracts, a
DNase I treatment (Invitrogen, Carlsbad, CA) was performed; purity of RNA
was checked by PCR amplification before random reverse transcription with
the Quanti Tect® Reverse Transcription Kit (Qiagen, Hilden,
Germany). HNLC, HLII and other Prochlorococcus ecotypes were qPCR-amplified using primers
and PCR conditions as previously described (Ahlgren et al., 2006).
Reactions were performed in technical duplicates in a final volume of 12.5 µL
using 0.25 µL of each primer (10 pmol µL-1), 3.25 µL
nuclease-free water and 6.25 µL SYBR qPCR Supermix W/ROX (Life
Technologies, Carlsbad, CA, USA) on a ViiA7 qPCR machine (Life Technologies,
Carlsbad, CA, USA) according to established protocols
(Ahlgren et al., 2006; West et al., 2011). TaqMan-based qPCRs
were performed for picophytoplankton (Prochlorococcus/Synechococcus) and bacteria as previously
described (Suzuki et al., 2001) in a final volume of 12.5 µL
with primer/probe concentrations as shown elsewhere (Table 1,
West et al., 2011), but with the addition of 0.5 µL BSA
(20 mg mL-1) and 6.25 µL TaqMan Mix (Life Technologies, Carlsbad,
CA, USA). Dilution series of plasmids containing the target gene were used
as standards as described (Lam et al., 2007; Löscher et al., 2012).
Nitrogen cycle key functional genes amoA, nirS, hzo and
nifH were amplified and quantified from
DNA and cDNA following established protocols (Lam et al., 2007; Langlois
et al., 2008; Löscher et al., 2012, 2014). Detection
limits of qPCR assays were determined from no-template controls, which were
run in duplicate for each primer (and probe) set, and were undetectable
after 45 cycles, thus setting the theoretical detection limit of our assay
mixtures to one gene copy. However, detection limits additionally depend on
the amount of filtered seawater per sample, elution volume after extraction,
and the amount of sample loaded to the qPCR assay. Based on a filtration
volume of 2 L seawater, a detection limit of 20 copies L-1 has been
determined. qPCR efficiencies were calculated using the formula E = 10-1/slope-1,
and were between 95.3 and 96.8 %.
PCR amplification of bacterial and archaeal 16S rDNA for Illumina
MiSeq amplicon sequencing
For the analysis of the bacterial community, hypervariable regions V1 and V2
of the 16S rDNA was amplified from genomic DNA using the primer set 27
forward (Frank et al., 2007) and 338 reverse
(Fierer et al., 2008). Beside the target-specific region the
primer sequence contained a linker sequence, an 8-base barcode and the
Illumina-specific region P5 (forward primer) or P7 (reverse primer),
respectively, as recently described (Kozich et al. 2013). The PCR reaction
mixture consisted of 13.6 µL DEPC H2O (Roth, Karlsruhe,
Germany), 0.4 µL of 10 mM dNTPs (Thermo Fisher Scientific),
4 µL 5× HF-buffer (Thermo Fisher Scientific, Waltham, MA, USA), 0.8 µL
primers (5 µM, Eurofins, Ebersberg, Germany), 0.2 µL Phusion
high-fidelity polymerase (2 U µL-1, Thermo Fisher Scientific,
Waltham, MA, USA) and 1 µL genomic DNA with a concentration between
10 and 100 ng µL-1. Negative controls consisted of the reaction
mixture as described above without the addition of DNA. PCR reaction
conditions started with an initial denaturation step for 5 min at
95 ∘C followed by 30 cycles of 15 s denaturation at 95 ∘C,
30 s primer annealing at 52 ∘C and 30 s elongation at
72 ∘C and a final elongation at 72 ∘C for 5 min.
For analysis of the archaeal community, hypervariable regions V5-V7 of
the 16S rDNA were amplified from genomic DNA using the primer set 787
forward and 1059 reverse (Yu et al., 2005) with 8-base barcode and
Illumina specific adapters. Reaction mixture, PCR protocol and purification
were identical to the amplification of bacterial community DNA
amplification, the only difference was the annealing temperature
(58 ∘C). Amplification was checked for correct size and band
intensity on a 2.5 % agarose gel. Amplicons were purified using the
MinElute Gel Extraction Kit (Qiagen, Hildesheim, Germany) and quantified on a
spectrophotometer (Nanodrop 1000, Thermo Fisher Scientific, Waltham, MA,
USA). Pooled purified amplicons were prepared and sequenced according to the
manufacturer's protocol on a MiSeq Instrument using the MiSeq reagent Kit V3
chemistry (Illumina, San Diego, CA, USA). Sequences were submitted to NCBI
Sequence Read Archive under accession number PRJNA288724.
Sequence analysis of 16S rDNA gene amplification
Sequence processing was performed using mothur software version 1.32.1
(Kozich et al., 2013; Schloss et al., 2009). 4 054 723 bacterial sequence
read pairs could be concatenated to contiguous sequences (contigs) using the
command make.contig. Contigs containing ambiguous bases, homopolymers longer than
eight bases or contigs longer than 552 bases were deleted from the data set.
Redundant sequences were clustered using the command unique.seqs, which led to 645 444
unique sequences. Sequences were consecutively aligned with align.seqs against a
modified version of the SILVA database release 102 (Pruesse et al.,
2007) containing only the hypervariable regions V1 and V2. The alignment was
optimized by removing sequences not aligning in the correct region with
screen.seqs, and by the removal of gap-only columns using filter.seqs. The optimized alignment
contained 636 701 sequences of lengths between 255 and 412 bases. Rare
sequences with up to three positional differences compared to larger sequence
clusters were merged with the latter by the pre.cluster command. Chimeric sequences
were removed with the implemented software UCHIME (Edgar et al.,
2011) using the command chimera.uchime, followed by remove.seqs.
Taxonomic classification of the remaining sequences was done using the Wang
approach based on a modified version of the Greengenes database
(DeSantis et al., 2006) with a bootstrap threshold of
80 %. Sequences of archaea, chloroplasts and mitochondria were removed
with remove.lineage. Operational taxonomic units (OTUs) were formed by average neighbor
clustering using the cluster.split command, parallelizing the cluster procedure by
splitting the data set at the taxonomic order level. A sample-by-OTU table
was generated with make.shared at the 97 % sequence similarity level. The resulting
table contained 15 509 OTUs. OTUs were classified taxonomically using the
modified Greengenes database mentioned above and the command classify.otu.
Archaeal sequences showed lower quality in the reverse read, which lead to
multiple ambiguous bases in the contigs formed. For this reason only the
forward read starting from base 36 was used for analysis. Sequence analysis
was performed as described above for bacterial 16S sequences, except that
the alignment (align.seqs) was accomplished using the SILVA archaeal reference release
102 (Pruesse et al., 2007) fitted for hypervariable regions
V5-V7. Classification (classify.seqs and classify.otu) was conducted using the RDP database file
release 10 (Cole et al., 2014; Wang et
al., 2007). Results and additional information on the archaeal community
structure are listed in the Supplement.
An overview of the sequencing output is given in Table S1 in the Supplement.
Statistics
Low-abundance OTUs were removed to reduce noise and computation time.
Statistical downstream analysis was performed in R v3.1.3 (R Core Team,
2015) with custom scripts (available from the authors on request). As OTUs
of very low abundance only increase computation time without contributing
useful information, they were removed from the data set as follows: after
transformation of counts in the sample-by-OTU table to relative abundances
(based on the total number of reads per sample), OTUs were ordered by
decreasing mean percentage across samples. The set of ordered OTUs for which
the cumulative mean percentage amounted to 99 % was retained in the
filtered OTU table.
Distribution of OTUs across samples was modeled by a set of environmental
variables (Table S2) with minimal interdependence. The variance in OTU
composition (i.e., the extent of change in OTU abundance across samples)
explained by the measured environmental variables was explored by redundancy
analysis (RDA) with Hellinger-transformed OTU counts (Langfeldt et al.,
2014; Stratil et al., 2013, 2014) using the R package vegan
(Oksanen et al., 2013). In order to minimize collinearity
of explanatory variables in the RDA model, a subset of the recorded
environmental variables was chosen according to their variance inflation
factor (VIF), employing vegan's functions rda and vif.cca. Starting with an RDA model
that contained all explanatory variables, the variable with the highest VIF
was iteratively determined and removed from the model until all remaining
explanatory variables had a VIF < 2.5.
OTU distribution was subject to “Realm” depending on O2 concentration. Model selection started with a full RDA model containing
all main effects and possible interactions based on the set of explanatory
variables with minimal collinearity. This model was simplified by backward
selection with function ordistep. The final RDA model exhibited a significant
interaction effect “Realm:O2” (see results section). For plotting and
indicator analysis (see below), the continuous variable “O2” was
converted into a factor with two levels “high O2” (> 90 µmol L-1)
and “low O2” (≤ 90 µmol L-1);
the threshold of 90 µmol L-1 was chosen for two reasons: (1) to
obtain sample groups of fairly equal size between stations, which include
low O2 parts of the water column at all sampling stations in order to
enable a comparison between the ETNA OMZ (outside the eddy) and the eddy
OMZ. (2) 90 µmol L-1 has previously described the highest
concentration of O2 at which denitrification has been detected to be
active (Gao et al., 2010). The presence of nirS transcripts
(see Sect. 3.4) indicated a potential importance for denitrifiers in the
eddy, therefore the theoretical upper limit of 90 µmol L-1 was
chosen.
We determined OTUs typical for a given combination of levels of factors
“Realm” and “O2”. OTUs significantly correlated with any axis in the
final RDA model were determined using the function envfit with 105
permutations, followed by Benjamini-Hochberg correction (false discovery
rate, FDR) (Benjamini and Hochberg, 1995). In order to reduce the
number of tests in this procedure, OTUs were pre-filtered according to their
vector lengths calculated from corresponding RDA scores (scaling 1) by
profile likelihood selection (Zhu and Ghodsi, 2006).
OTUs significant at an FDR of 5 % were further subject to indicator
analysis with function multipatt of the R package indicspecies v1.7.4 (De
Cáceres and Legendre, 2009) with 105 permutations. Indicator OTUs
– in analogy to indicator species sensu De Cáceres and Legendre (2009)
– are OTUs that prevail in a certain sample group (here: a level of factor
“Realm” within a chosen O2 level) while being found only irregularly
and at low abundance in other sample groups. In order to remove the effects
of the covariate “Depth” in indicator analysis, Hellinger-transformed
counts of significant OTUs were first subjected to a linear regression with
“Depth”; residuals of this regression were then transformed to positive
values by subtraction of their minimum and used as input for indicator
analysis.
3-D visualizations of the RDA model were produced in kinemage format
(Richardson and Richardson, 1992) using the R package R2Kinemage
developed by S.C.N., and displayed in KiNG v2.21 (Chen et
al., 2009).
Diversity within samples was related to environmental variables by advanced
linear regression. For alpha diversity analysis, effective OTU richness
(Shannon numbers equivalent, 1D, Jost, 2006, 2007) was calculated
from the filtered OTU table. 1D was fitted to the set of explanatory
variables with minimal collinearity in a generalized least squares (GLS)
model using function gls of the R package nlme v3.1-120 (Pinheiro et
al., 2015). The variable “NO2” was square root-transformed to decrease the
potential leverage effect of its two highest values (0.25
and 0.28 µmol L-1, respectively) on 1D. Apart from
main effect terms, the interaction term “Realm:O2” was included into
the GLS model for comparability with beta diversity analysis (see results
section). The variance structure of the GLS model was chosen to account for
both different variances per level of “Realm” and an overall decreasing
variance by “Depth”. The resulting model was validated following the
recommendations of Zuur et al. (2009). While only the
“Realm” effect was significant, the other terms were kept in the model to
maintain a valid residual distribution. For visualization of the (partial)
effect of only factor “Realm” on 1D, partial response residuals were
extracted from the full GLS model re-fitted without the “Realm” main
effect. These partial response residuals were then modelled by the “Realm”
main effect alone, using the same variance structure as for the full GLS
model.
Carbon fixation rate measurements
Seawater incubations were performed in triplicate at two stations, one
inside the eddy (station 10, M105 cruise) and one in ETNA open waters
(station 15, M105 cruise, both stations indicated in Fig. 1c). Seawater was
sampled from a CTD system and directly filled into 2.8 L polycarbonate
bottles (Nalgene, Thermo Fisher Scientific, Waltham, MA, USA). For carbon
fixation measurements, NaH13CO3 (Cambridge Isotope Laboratories,
MA, USA) was dissolved in sterile deionized water (> 18.2 MΩ cm-1,
MilliQ, Merck-Millipore, Darmstadt, Germany;
5 g/294 mL). A volume of 1 mL (2.8 L bottles) was added to the incubations
with a syringe (∼ 4.4 at % final). After amendment,
bottles were stored on deck in a seawater-cooled Plexiglas incubator covered
with light foils (blue-lagoon, Lee filters, Andover, Hampshire, UK) that
mimic light intensities at corresponding sampling depths (5/10/30/70 m).
Samples from below the euphotic zone were stored at 12 ∘C in the
dark. The depth of the euphotic zone was estimated from photosynthetically
active radiation (PAR) sensor measurements from CTD profiles as the depth
where PAR is < 1 % of the surface value. This corresponded to 60 m
water depth during this survey. After 24 h of incubation, 1.5–2.8 L of
seawater was filtered onto pre-combusted (450 ∘C, 5 h) 25 mm
diameter GF/F filters (Whatman, Maidstone, UK) under gentle vacuum (-200 mbar).
Filtrations were stopped after 1 h since high particle load of
surface water led to a clogging of the filters. Filters were oven dried
(50 ∘C) for 24 h and stored over desiccant until analysis.
Environmental samples of 2.8 L untreated seawater were filtered and prepared
in the same way to serve as blank values. For isotope analysis, GF/F filters
were acidified over fuming HCl overnight in a desiccator. Filters were then
oven-dried for 2 h at 50 ∘C and pelletized in tin cups. Samples
were analyzed for particulate organic carbon and nitrogen (POC and PON) and
isotopic composition using a CHN analyzer coupled to an isotope ratio mass
spectrometer.
Results and discussion
Hydrography of low-O2 eddy reveals similarities to shelf
waters
As the detailed properties of the investigated eddy are described in
Schütte et al. (2015) only the main
characteristics are mentioned here.
The surveyed low-O2 eddy belongs to the group of the anticyclonic
modewater eddies (ACME) (Karstensen et al. 2015a). It has been reported that
ACME promote intense primary production in surface and mixed layer waters
(Mahadevan, 2014) fueled by nutrient supply to the euphotic zone. The
surveyed eddy had a diameter of about 100 km and was characterized by highly
elevated mixed-layer chlorophyll a (chl a) concentrations, a positive SLA
signature (Fig. 1) and a low O2/low-salinity core (Fig. 2). The
O2-depleted core, with concentrations of less than
5 µmol kg-1, was centered rather deep for an ACME at ∼ 100 m
depth. Concentrations of less than 30 µmol kg-1 were observed in
the eddy water column between 70 to 150 m depth (Figs. 2, 3a), which is
significantly below average O2 concentrations in that region. O2
concentrations in the core decreased over the survey period (March 2014),
(see Fiedler et al. (2015), for a detailed
description of O2 properties). During the metagenomic sampling of the
background signal (“no eddy”) on the shelf (Meteor M107 cruise station
675, 18.22∘ N/16.56∘ W, Fig. 1), O2 concentrations
of 33.9 µmol kg-1 were observed at 115 m depth, which
corresponds to the potential density layer of the low O2 core in the
eddy. The open-ocean background minimum O2 concentrations of about 70 µmol kg-1
were detected at ∼ 250 m depth at CVOO
(Fig. 1). This can be considered average O2 concentrations for the
open ETNA (Karstensen et al., 2008).
Temperature (left panel), salinity (middle panel) and O2
concentration (right panel) measured during a section of R/V Meteor Cruise
M105 across the studied eddy. Minimum O2 was
4.8 µmol kg-1 at ∼ 100 m water depth on that section;
however, even lower O2was detected with a glider
(1.2 µmol kg-1). Isopycnals are indicated by white lines.
(a) O2 and (b) nitrate and (c) nitrite
concentrations measured at the open-ocean station CVOO (black circles), in
the first observation (eddy_1, open circles), second observation (eddy_2,
black triangles) and on the Mauritanian shelf (open triangles). (d)
Nitrate vs. phosphate concentration at the four sampling stations. The color
code denotes the O2 concentration and the black line indicates the
Redfield ratio of N : P = 16 : 1.
In the low-O2 eddy core, we observed nitrate and phosphate
concentrations around twice as high as background concentrations at CVOO at
the same depth (Fig. 3). However, N : P ratios below the mixed layer were
close to Redfield stoichiometry (16.15 ± 0.63, Fig. 3) and thus
comparable to surrounding waters. Nitrate concentrations in the O2-min
core (∼ 100 m depth) were similar to concentrations on the
Mauritanian shelf at 100 m depth (Fig. 3) and most likely generated by very
efficient local remineralization of nitrate from the sinking material
(Karstensen et al. 2015b).
Loss of phylogenetic diversity in low-O2 eddy waters
A critical issue regarding climate change induced pressures on ocean
ecosystems is to understand the effects of ocean acidification and
deoxygenation on microbial communities as major drivers of the ocean's
biogeochemistry (Riebesell and Gattuso, 2015). Thus, we investigated
phylogenetic diversity of the microbial community with a 16S rDNA amplicon
sequencing approach of bacteria and archaea inside and outside the eddy.
Although the bacterial community was dominated by Proteobacteria in all
samples, there were distinct differences between the community structures
inside compared to outside the eddy (Fig. 4). Increased abundances of the
uncultivated SUP05 clade (up to 20 % of proteobacterial sequences) have
been recovered from eddy samples compared to surrounding waters (Supplement Fig. S1,
Table S3). This clade is known to occur frequently in O2 depleted
environments (Swan et al., 2011). Phyla such as
Bacteroidetes, Actinobacteria and Firmicutes were only present in the eddy
and increased in relative abundance over time. Those phyla were also
detected in potential source waters on the shelf (Supplement Fig. S2). Interestingly,
the family of Pelagibacteraceae, which belong to the ubiquitous SAR11 clade
(Giovannoni et al., 1990), were strongly decreased in the eddy
(to ∼ 1 % of all reads), compared to CVOO samples
(∼ 65 % of all reads). SAR11 was previously described as
being sensitive to decreasing O2 concentrations (Forth
et al., 2014), which may explain the absence of this classically highly
abundant group from the eddy. In addition to the dissimilarity in bacterial
diversity, we also detected a substantial difference in archaeal community
composition between eddy stations and CVOO (Fig. S3). This was most obvious
in samples from the eddy_2 station, where Methanomicrobia
dominated the archaeal community in the O2-depleted parts of the water
column but was absent in CVOO samples. The presence of methanogens in the
low-O2 eddy core samples may indicate potential for methanogenesis.
Although the eddy has not been shown to become fully anoxic, methanogenesis
tolerates O2 concentrations at low ranges (Angel et al.,
2011).
Distribution of bacterial phyla along vertical profiles of
(a) CVOO, (b) first observation (eddy_1) and (c)
second observation (eddy_2) is shown along with the O2 gradient (black
line). Data sets result from 16S rDNA amplicon sequencing (an overview on
archaeal sequence distribution is given in the Supplement).
Redundancy analysis (RDA) confirmed that the distribution of bacterial OTUs
strongly differed between the two eddy stations and CVOO samples (Fig. 6a;
RDA model: F6,24= 4.48, p < 0.001). Changes in OTU composition
mirrored the depth gradient (RDA “Depth”: F1,24= 2.08, p≈0.03;
Fig. 5) and were thus strongly correlated to chemical (PO43-,
NO3-, SiO2) and physical (T, S) properties (Fig. S4). The RDA
model indicates a noticeable interaction effect of habitat (“Realm”) and
O2 concentration (RDA “Realm:O2”: F2,24= 2.03,
p≈ 0.02), meaning that the “Realm” effect on bacterial community
structure depends on the O2 level and vice versa. An overview of the
parameters included in the RDA model is given in Table S2. O2 and
nutrient availability can thus be considered the major determining variables
for the composition of the microbial community.
Redundancy analysis (RDA) of OTU distribution in samples from the
first eddy observation (eddy_1), from the second eddy observation (eddy_2)
and from CVOO based on 16S rDNA sequences. (a) First and second
axis, (b) third and fourth axis of the RDA model, illustrating the
interaction effect of factor “Realm” and O2 concentration. For
plotting, the continuous variable “O2” was converted into a factor
with two levels “high O2” (> 90 µM) and “low
O2” (≤ 90 µM).
Alpha diversity analysis of eddy sampling stations (first
observation (eddy_1), second observation (eddy_2)) and CVOO expressed as
Shannon numbers equivalent (1D). A strong and significant decrease in
diversity is observed in the eddy. Partial response residuals (black symbols)
were extracted from full GLS model re-fitted without the “Realm” main
effect. Predicted values for partial residuals modelled by the “Realm” main
effect alone (and thus adjusted for differences in O2 concentration) are
shown as blue symbols. Error bars represent 95 % confidence interval for
fitted values.
Our results further show a significant decrease in bacterial alpha diversity
in the eddy relative to CVOO (Fig. 6). The community in eddy_2
samples was also markedly less diverse compared to those of the other
realms (Fig. 6; generalized least squares (GLS) model: F7,23= 5.37,
p= 0.001; GLS “Realm”: F2,23= 16.26, p < 0.0001). This may be
attributed to an aging effect of the eddy, and corresponds to progressive
O2 loss and consecutive changes in the eddy biogeochemistry. We
calculated an overall O2 loss of 0.18 µmol kg-1 d-1 at 100 m
depth by respiration, when comparing the eddy core water to the
potential origin waters on the shelf, assuming a lifetime of 180 days for
the eddy (average O2 concentrations on the shelf from Meteor M107 were
36.69 ± 6.91 µmol kg-1 compared to observed minimum
O2 concentrations of 4.8 µmol kg-1 in the eddy core).
These results are comparable to previous estimates on low O2-eddies in
that region (Karstensen et al., 2015a). Likewise, Fiedler et al. (2015)
also observed a significant increase in pCO2 and
dissolved inorganic carbon compared to coastal waters, indicating enhanced
remineralization and respiration. Although our data set does not allow
differentiation between high-pCO2 and low-O2 effects on the
microbial community, it supports the view of a general loss in diversity.
This may be attributed to a direct or indirect response to factors related
to deoxygenation and increasing pCO2, such as the impact on nutrient
stoichiometry, as previously suggested (Bryant et al., 2012).
Hence, climate change-related ocean deoxygenation and consequent shifts in
nutrient stoichiometry may mean an overall loss of microbial diversity, with
potential for substantial loss in the spectrum of metabolic functions in the
future ocean.
Specific Prochlorococcus clade contributes to primary production in the
eddy
The detected ACME was characterized by shoaling of the mixed layer depth in
the center of the eddy. This coincided with a pronounced surface chl a maximum
as observed by ocean color based and remotely sensed chl a estimates (Figs. 1a,
7), which was slightly deeper (∼ 50–70 m water depth)
outside the eddy. In accordance with increased chl a concentrations, enhanced
carbon uptake was observed via direct rate measurements of
H13CO3- uptake which was potentially fueled by increased
nutrient availability from intermediate depths. We found a 3-fold increase
in depth-integrated carbon uptake rate in the chl a maximum of the eddy (178.3 ± 30.8 m mol C m-2 d-1)
compared to surrounding waters (59.4 ± 1.2 mmol C m-2 d-1).
Chlorophyll a (chl a, µg L-1) distribution as
determined from discrete measurements and carbon uptake rates (a)
inside the eddy (eddy_2, second observation) and (b) at the eddy
rim (location denoted in Fig. 1). Error bars indicate the standard deviation
of three replicate samples for C uptake.
While the upper chl a maximum in the eddy may likely be ascribed to eukaryotic
primary producers such as diatoms and flagellates that are widely
distributed and abundant in that region (Franz et al., 2012),
confirmed by increased abundances of plastids in surface samples of our
amplicon data set (Table S3). A secondary chl a maximum dominated by
cyanobacteria was detected in the eddy at about 100 m water depth,
coinciding with the O2 minimum.
The quantitative analysis of cyanobacterial primary producers by 16S
rDNA-qPCR further revealed dominance of a specific clade of Prochlorococcus
in the secondary chl a maximum (Fig. S5 depicts phylogenetic relations of detected
Prochlorococcus clades). This ecotype has so far not been identified in the ETNA and is
only known from high nutrient low-chlorophyll (HNLC) regions of the eastern
tropical Pacific Ocean (West et al., 2011). Its described
adaptation to high nutrient conditions such as those present in this
O2-depleted ACME points towards a selective advantage for this clade.
Gene abundance of this ecotype – for convenience further referred to as
HNLC-PCC (results of an ecotypespecific16S rDNA based qPCR) – showed a
strong correlation with chlorophyll (R2= 0.95, n= 22) below the
euphotic zone within the eddy. This correlation was not present outside the
eddy, where HNLC-PCC abundance was approximately one third compared to the
second eddy observation (Fig. 8). The Prochlorococcus community in surrounding waters was,
however, dominated by another high-light ecotype of Prochlorococcus (further referred to as
HL-PCC (West et al., 2011)). Contrary to HNLC-PCC, HL-PCC was not
detected inside the eddy. The difference between the CVOO,
eddy_1 and eddy_2 observations points towards
a community shift of Prochlorococcus related clades depending on specific characteristics of
the eddy (O2, nutrient availability) with the potential to alter
primary productivity in that region. Under increasing pCO2 levels, Prochlorococcus is
predicted to substantially increase in abundance (Flombaum et al.,
2013). Elevated pCO2 levels in the eddy core water may therefore – apart
from favorable elevated nutrient concentrations – explain the additional
selective advantage of specific Prochlorococcus clades, in this case of HNLC-PCC. This may
be critical as Prochlorococcus is one of the most abundant photosynthetic organisms in the
ocean and contributes to ∼ 40 % of dissolved organic carbon
supporting bacterial production (Bertillson et al., 2005).
Vertical distribution of Prochlorococcus and
Synechococcus ecotypes quantified by qPCR. While the HNLC-PCC
(a) dominates the eddy water mass and increases from the first
observation (eddy_1) to the second observation (eddy_2) it is nearly absent
outside the eddy (CVOO). HLII-PCC (b) occurs in highest abundances
outside the eddy, while being close to the detection limit inside the eddy.
(c) shows the distribution of pico-phytoplankton as detected with a
general primer-probe system (Suzuki et al., 2001).
Besides a direct impact of O2, nutrients and pCO2, increased
abundances of Prochlorococcus in the eddy may be explained from an interaction effect in the
microbial community present in the eddy. Prochlorococcus is supposed to play a major role in
sustaining heterotrophs with organic carbon compounds such as glycine and
serine, thus favoring their growth (Biller et al.,
2015; Carini et al., 2013). Conversely, Prochlorococcus benefits from the presence of
heterotrophs as they diminish the concentration of reactive oxygen species
in their immediate surroundings, which is not feasible for Prochlorococcus due to the lack
of catalase and peroxidase genes (Berube et al., 2014; Morris
et al., 2008). The close proximity of increased abundances of the HNLC-PCC
maximum to the O2 minimum in the eddy may thus point towards a
beneficial relationship between the HNLC-PCC and the heterotroph-dominated,
eddy core water microbial community.
Increased primary productivity promotes a specific heterotrophic
microbial community in underlying waters
We analyzed species indicative for the eddy and CVOO for either high-O2
conditions (> 90 µmol kg-1) or low-O2 conditions
(≤ 90 µmol kg-1). Indicator OTUs for high O2 in the
eddy were mostly associated with different clades of Proteobacteria, whereas
Pelagibacteraceae dominated at CVOO in accordance with several studies
describing those organisms as ubiquitous in open-ocean oxic waters
(Morris et al., 2002; Rappé et al.,
2002; Poretsky et al., 2009; DeLong, 2009; Brown et al., 2014).
High-O2 samples of all three sampling stations were dominated – as most
parts of the ocean – by indicator OTUs belonging to the Proteobacteria. The
Prochlorococcus clade HNLC-PCC targeted by qPCR could be recovered in the 16S rDNA amplicon
sequences, as well.
For low-O2 conditions, indicator species present in the eddy were
mostly affiliated to the Cytophaga-Flavobacteria-Bacteroides (CFB) group
(Glöckner et al., 1999) (Table S4). Members of Bacteroidetes
and Proteobacteria (Gramella, Leeuwenhoekiella marinoflava, unclassified Comamonadaceae species) were found to
be indicative for the low-O2 realm. Gramella-like organisms are usually a
quantitatively important fraction of the heterotrophic marine
bacterioplankton, often attached to marine snow but also found free-living
in nutrient-rich microenvironments (Buchan et al., 2014).
Frequently associated with extensive phytoplankton blooms
(Buchan et al., 2014), their ability to degrade high molecular
weight compounds in both the dissolved and particulate fraction of the
marine organic matter pool points towards a specific role in respiration
processes and the marine C cycle (as described for “Gramella forsetii” KT0803,
Bauer et al. (2006). Karstensen et al. (2015a) described a
particle maximum associated to the low-O2 core of those eddies which
likely harbors this specific heterotrophic community. Further, in the core
of the ACME presented here, the integrated abundance (upper 600 m) of large
aggregates was five times higher than in surrounding waters
(Hauss et al., 2015).
Enhanced productivity and consecutive respiration and O2 decrease may
enable N loss processes to occur in the open ETNA, which have previously not
been described for the ETNA waters (Löscher et al., 2012, 2015; Ryabenko et al., 2012). qPCR results of key gene distribution
(amoA for nitrification as sum of bacterial and archaeal nitrifiers, nirS as key
gene for denitrification) in that area show a decrease of amoA in the eddy,
while nirS shows higher abundances inside the eddy with ∼ 3000 copies L-1
at depth of the O2 minimum (compared to ∼ 100 copies L-1 outside the eddy). Besides a direct sensitivity of
nitrifiers to anoxic conditions, the decrease in amoA gene abundance (determined
by qPCR) towards the O2 minimum in the eddy may result from an effect
of elevated pCO2 (see Fiedler et al., 2015)
and the corresponding drop in pH on ammonia due to a shift in the
ammonia/ammonium equilibrium. The latter has previously been described to
alter the efficiency of nitrification (Beman et al., 2011). Further,
nirS transcripts as quantified by qPCR were detected in abundances up to 3600 transcripts L-1
in the eddy O2 minimum, while no transcripts were
detected outside the eddy (Fig. 9).
Gene and transcript abundance vs. O2 concentrations of samples
from the eddy observations (eddy_1 and eddy_2) and CVOO. (a) shows
the key gene for denitrification, nirS, coding for the nitrite
reductase, (b) shows archaeal amoA as key functional gene
of ammonia oxidation, coding for the ammonia monooxygenase. Gene abundances
are denoted in red, transcript abundances are indicated by black circles.
The presence and expression of nirS supports the view that potential for N loss
is also present in the usually oxic open ETNA. This is in line with another
study on nitrous oxide (N2O) production from the same eddy
(Grundle et al., 2015), where the authors observed
massively increased N2O concentrations in the oxygen deficient eddy
core waters in connection with denitrification. Observations from e.g. the
eastern tropical Pacific Ocean demonstrated previously that mesoscale eddies
are drifting hotspots of N loss (Altabet et al., 2012). This
might be explained by feedback mechanisms between eutrophication, enhanced
primary productivity and consecutive enhanced export production, which may
promote denitrification in those systems as suggested by Kalvelage et al. (2013).
Our results strongly suggest that N loss is
possible in eddy systems of that region, thus altering one major
biogeochemical cycle with unknown consequences for the ETNA biogeochemistry.
In case of the described eddy, we neither detected key genes for anammox
(hzo, Schmid et al., 2008) nor significant abundances of the key genes for
dinitrogen fixation. The latter has been investigated by screening for the
functional key gene, nifH, which has been tested for classical diazotrophs as
Trichodesmium, UCYN-A, UCYN-B, UCYN-C, gamma proteobacterial diazotrophs and DDAs; all of
which were not quantifiable by qPCR. This may be explained by the high
availability of inorganic N sources, as well as the prevalence of N : P close
to the Redfield ratio of 16 : 1 as mentioned above.
Although N2 fixation does not appear to play a role in the low-oxic
core waters or adjacent surface waters of the eddy, it may occur as a result
of increasing N loss and resulting excess P as previously discussed for
other O2 depleted marine habitats (Deutsch et al., 2007; Fernandez et
al., 2011; Löscher et al., 2014; Ulloa et al., 2012).
Conclusions
We investigated the microbial community structure and gene expression in a
severely O2-depleted anticyclonic modewater eddy in the open waters of
the ETNA OMZ region. This was then compared to the eddy observations to
background signals from the ETNA open-ocean CVOO time series site and the
Mauritanian upwelling region, where the eddy was likely formed.
A significant difference between microbial communities outside and inside
the eddy along with an overall loss in bacterial diversity in the
low-O2 core of the eddy was observed. Similarity was found between the
microbial community in the eddy core and on the shelf. This unique microbial
community may shape the specific character of this O2-depleted eddy
progressively over time.
We observed enhanced primary production in the eddy, presumably due to an
increased nutrient supply related to the eddy dynamics (Karstensen et al.
2015b). We found a specific HNLC ecotype of Prochlorococcus, which may play a role in
mediating inorganic C to certain organic C sources for the associated
heterotrophic community present in the eddy. Importantly, we found the first
indication for N loss processes in the ETNA region. Low-O2 eddies in
that region thus represent an isolated ecosystem in the open ocean, forced
by strongly elevated biological productivity, which travels with the eddy.
This leads to consequent enhanced respiration and further deoxygenation in
its core waters.
At one stage the low-O2 eddies will lose coherence and the extreme
signatures will be released into and mixed with the surrounding waters
(Karstensen et al., 2015a). The ACME formation frequency for the ETNA
(12–22∘ N and 15–26∘ W) has been
estimated to be about 2 to 3 yr-1 (Schütte et al. 2015),
hence no large-scale impact of the eddies are
expected. However, an unexpected shift in elemental ratios or other
anomalies, normally expected for regions with much lower minimal oxygen
levels than the ETNA, may be detected and explained by the dispersal of
low-O2 eddies. Another factor to consider is the impact of
deoxygenation of the ETNA (Stramma et al., 2008) as it may
result in even lower O2 conditions to be created in the low-O2
eddies. With regard to the distinct character of the low-O2 eddies and
the critical shift in microbial diversity and biogeochemistry that occur
over relatively short times, this study contributes to understand and
evaluate the far-reaching effects of future and past ocean deoxygenation.