Content uploaded by Jun Sun
Author content
All content in this area was uploaded by Jun Sun on Jul 23, 2021
Content may be subject to copyright.
Content uploaded by Chao Wu
Author content
All content in this area was uploaded by Chao Wu on May 13, 2021
Content may be subject to copyright.
journal of environmental sciences 111 (2022) 324–339
Available online at www.sciencedirect.com
w w w . e l s e v i e r . c o m / l o c a t e / j e s
Dynamics of bacterial communities during a
seasonal hypoxia at the Bohai Sea: Coupling and
response between abundant and rare populations
Chao Wu
1 , 2
, Jinjun Kan
3
, Dhiraj Dhondiram Narale
1 , 2
, Kun Liu
4
,
Jun Sun
1 , 2 , 5 , ∗
1
Tianjin Key Laboratory of Marine Resources and Chemistry, Tianjin University of Science and Technology, Tianjin
300457, China
2
Research Centre for Indian Ocean Ecosystem, Tianjin University of Science and Technology, Tianjin 300457, China
3
Department of Microbiology, Stroud Water Research Center, Avondale, PA 19311, USA
4
Laboratory of Marine Biology and Ecology, Third Institute of Oceanography, Ministry of Natural Resources, Xiamen
361005, China
5
College of Marine Science and Technology, China University of Geosciences (Wuhan), Wu ha n 430074, China
Article history:
Received 14 September 2020
Revised 14 April 2021
Accepted 15 April 2021
Keywords:
Bacterial community
Hypoxia
Stratication
The Bohai Sea
Nitrogen cycles
High-throughput sequencing
Marine bacterial community plays a vital role in the formation of the hypoxia zone in coastal
oceans. Yet , their dynamics in the seasonal hypoxia zone of the Bohai Sea (BHS) are barely
studied. Here, the 16S rRNA gene-based high-throughput sequencing was used to explore
the dynamics of their diversity, structure, and function as well as driving factors during the
gradual deoxygenation process in the BHS. Our results evinced that the bacterial community
was dominated by Proteobacteria, followed by Bacteroidetes, Firmicutes, Actinobacteria, and
Cyanobacteria, etc. The abundant subcommunity dominated in the number of sequences
(49%) while the rare subcommunity dominated in the number of species (99.61%). Although
abundant subcommunity accounted for most sequences, rare subcommunity possessed
higher diversity, richness and their population dramatically changed (higher turnover) dur-
ing the hypoxia transition. Further, co-occurrence network analysis proved the vital role
of rare subcommunity in the process of community assembly. Additionally, beta diversity
partition revealed that both subcommunities possessed a higher turnover component than
nestedness and/or richness component, implying species replacement could explain a con-
siderable percentage of community variation. This variation might be governed by both en-
vironmental selection and stochastic processes, and further, it inuenced the nitrogen cycle
(PICRUSt-based prediction) of the hypoxia zone. Overall, this study provides insight into the
spatial-temporal heterogeneity of bacterial and their vital role in biogeochemical cycles in
the hypoxia zone of the BHS. These ndings will extend our horizons about the stabiliza-
tion mechanism, feedback regulation, and interactive model inside the bacterial community
under oxygen-depleted ecosystems.
© 2021 The Research Center for Eco-Environmental Sciences, Chinese Academy of
Sciences. Published by Elsevier B.V.
∗Corresponding author.
E-mail: phytoplankton@163.com (J. Sun).
https://doi.org/10.1016/j.jes.2021.04.013
1001-0742/© 2021 The Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences. Published by Elsevier B.V.
journal of environmental sciences 111 (2022) 324–339 325
Introduction
Coastal regions are inuenced by various adverse environ-
mental issues as a result of complicated natural processes
as well as anthropogenic pollution. Among them, hypoxia
is recognized as one of the major stressors which adversely
inuence the ecological balance ( Ravaglioli et al., 2019 ). Hy-
poxia exerts enormous hazardous inuences on the aquatic
ecosystems, it not only threatens aquatic biota but also gener-
ates various types of greenhouse gases and toxic compounds.
The term hypoxia is typically dened as the concentration
of dissolved oxygen (DO) in water lower than 94 μmol O
2
/L
( Rabalais et al., 2010 ). In general, hypoxia is the resultant of
a complex interaction between multiple factors such as up-
welling, eutrophication, stratication, food web dynamics, etc
( Helly and Levin, 2004; Levin et al., 2009 ). Furthermore, it varies
with the different hydrodynamic regimes, seasonally chang-
ing environmental factors, and basin topography ( Lu et al.,
2018 ; Rabalais et al., 2014 ). For most coastal oceans, eutrophi-
cation and stratication are recognized as the most crucial in-
ducers of hypoxia. Concurrently, with deoxygenation, the high
organic matter aerobic respiration stimulates the accumula-
tion and dissolution of carbon dioxide (CO
2
), which results in
the decline of pH and ocean acidication ( Melzner et al., 2013 ;
Zhai et al., 2019 ). More seriously, globally, the hypoxia zones
are predicted to inate continually in both size and frequency
( Fennel and Tes ta, 2019; Naqvi et al., 2009 ) .
Recent studies revealed that hypoxia can have profound
and lasting inuences on the marine microbial community,
which eventually changing the marine ecosystem function-
ing and global biogeochemical cycling ( Carolan et al., 2015;
Orsi et al., 2012; Stevens and Ulloa, 2008 ). The low DO con-
centration results in loss of xed nitrogen from the ocean
through suppressing aerobic nitrication and enhancing den-
itrication ( Bulow et al., 2010 ). The global oxygen minimum
zones (OMZs) are estimated to remove up to 30% of oceanic
xed nitrogen though they only represent less than 1% of
global seawater volume ( Thamdrup et al., 2012 ). Furthermore,
many greenhouse gases (e.g., nitrous oxide and methane)
which are usually generated as intermediate products of the
deoxygenation process, can aggravate global climate change
( Naqvi et al., 2009 ). Notably, microbial-driven cryptic sulfur cy-
cle, i.e., sulfurs reduction and sulde oxidation facilities en-
ergy ux and elemental cycling in OMZs as evidenced else-
where ( Caneld et al., 2010 ). This analogous sulfur cycle is
closely associated with the nitrogen cycle, which could largely
inuence the global biogeochemical cycle ( Caneld et al.,
2010 ).
The Bohai Sea (BHS) is a shallow and semi-enclosed
marginal sea located towards the Northeast part of the Chi-
nese mainland. Over the last few decades, the BHS is suf-
fered from the severe anthropogenic eutrophication resultant
of massive terrestrial input of anthropogenic pollutants into
the ocean, which further stimulated harmful algae blooms in-
cidences in this region ( Xu et al., 2017 ). Through the intensive
efforts of the government and public in recent years, the trend
of environmental pollution aggravation in the BHS has now
been controlled. However, water quality is not signicantly
improved yet ( Wan g et al., 2018 ). In particular, the increas-
ing reports of summertime hypoxia in the BHS indicated that
this coastal ocean is distressed with serious environmental
and ecological degradation ( Wang et al., 2015 ). The summer-
time hypoxia in the BHS has been extensively attributed to the
combined effect of seasonal stratication and organic matter
remineralization ( We i et al., 2019 ; Zhai et al., 2012 ; Zhao et al.,
2017 ). Furthermore, the bottom seawater pH and aragonite
saturation state in the BHS could be drastically reduced dur-
ing summertime, which also threatens the survival of numer-
ous benthonic organisms ( Zhai et al., 2019 ). To date, however,
no detailed study has been conducted to investigate the dy-
namics of the bacterial community in the hypoxia zones of
the BHS.
Here, we present the rst detailed investigation on bacte-
rial community in the seasonally stratied and hypoxic inner
bay region of the BHS from May to August 2017. The 16S rRNA
gene-based high-throughput sequencing was used to investi-
gate the inuence of deoxygenation and ocean acidication
over the temporal distribution of the bacterial community.
Overall, this study aims to, (1) evaluate the bacterial commu-
nity structure and diversity; (2) distinguish the abundant and
rare subcommunity, and uncover their interaction patterns;
(3) recognize the keystone species via co-occurrence network
analysis; (4) explore the dynamic of abundant as well as rare
subcommunity and their functions along the temporal gradi-
ent; and (5) furthermore, to assess the control of regional envi-
ronmental cues on shaping the bacterial community structure
in the seasonally stratied hypoxic BHS region. In brief, the
ndings of this study will widen our understanding of the mi-
crobial community dynamics and their functional variations
in the changing aquatic ecosystems.
1. Materials and methods
1.1. Study location, sample collection, and analysis
Wat er samples were collected from six stations (A1-A6) lo-
cated outwards the Qinhuangdao shelf region, the Bohai Sea
during four consecutive monthly cruises from May to August
2017 ( Fig. 1 ). At all stations, vertical samples were collected for
the assessment of environmental parameters from two to ve
depths using a Niskin water sampler equipped with RBR620
multiple sensors. In August, we observed obvious deoxygena-
tion in the bottom layers at sites A3 to A5. Finally, a total of 46
DNA samples were collected at the three stations (sampling
depths were roughly at 1, 5, 10, 15, and 25 m) from May to Au-
gust 2017 for high-throughput sequencing.
Tem perat ure , salinity, and turbidity were recorded on-
board. The DO and pH were recorded at desired depths with
pre-calibrated sensors prior to each cruise using a standard
protocol ( Luo et al., 2018 ). Collected seawater samples were
transferred into separated 10 L pre-rinsed PE buckets (with
10% HCl and Milli-Q water), and stored in an on-deck incu-
bator cooled by running seawater till further subsampling.
For estimation of Chl a , 500 mL seawater was subsampled
and vacuum-ltered ( < 100 mmHg) through 25 mm GF/F lters
(Waterman, Florham Park, NJ, USA). These lters were pack-
aged in aluminum foil bags and freeze stored ( −20 °C) until fur-
ther analysis. In addition, to analyze the nutrients, the ltrate
326 journal of environmental sciences 111 (2022) 324–339
Fig. 1 –The map showing the sampling stations in the Bohai Sea (BHS). In summary, samples from 6 sites (A1-A6) were
used for physicochemical analysis, and 3 sites (A3-A5) were used for molecular analysis.
of respective depths was transferred into separate 100 mL pre-
rinsed (10% HCl) PE bottles and frozen in the same refrigerator.
The detailed protocols of the Chl a and nutrients estimation
are given in Wu et al. (2019) .
For estimation of particulate organic nitrogen (PON) and
particulate organic carbon (POC), 300 mL seawater samples
ltered onto pre-combusted (450 °C for 2 hr) 25 mm GF/F l-
ters. The lters were stored as similar to Chl a samples un-
til further evaluation. Prior to analysis, the lters were fumed
with concentrated HCl for 3 hr to remove inorganic carbon.
These fumed lters were then analyzed with a Costech Ele-
mental Combustion System (Costech Analytical Technologies
Inc., Valencia, CA) using standard procedures ( Solórzano and
Sharp, 1980 ).
1.2. DNA sample collection, extraction, amplication, and
Illumina sequencing
For DNA estimation, 500-800 mL seawater from different
depths were ltered individually through 0.22 - μm GTTP lters
(47 mm diameter; Millipore, Eschborn, Germany) under low
vacuum pressure ( < 100 mmHg). These lters were transferred
in 2 mL microtubes, ash-frozen in liquid nitrogen (-196 °C)
until analysis. In the lab, genomic DNA was extracted using
DNeasy PowerSoil Kit (Qiagen, Hilden, Germany) according to
the manufacturer’s protocol. The quality and quantity of ex-
tracted DNA were checked by 1.8% Agarose gel (BioWest, Cas-
tropol, Spain) electrophoresis and an ND-2000 Nanodrop spec-
trometer (Thermal Scientic, Wilmington, DE, USA). Then, the
extracted DNA was diluted to a nal concentration of 1 ng/ μL
and stored at a refrigerator (-20 °C) until further amplication.
To characterize the total microbial composition, V3-V4
variable regions of the 16S rRNA genes were amplied with
the pair-wise common primer 343F (5’-TACGGRAGGCAGCAG-
3’) and 798R (5’-AGGGTATCTAATCCT-3’). The V3-V4 fragments
were amplied from the diluted DNA through polymerase
chain reactions (PCRs) by using a Bio-Rad thermocycler (Bio-
Rad, Redmond, WA, USA). Detailed protocol of PCRs and high
throughput sequencing is given in Appendix A Te xt S1 . All
libraries were constructed and sequenced at Shanghai OE
Biotech CO., Ltd. (Shanghai, China) via paired-end chemistry
on an Illumina Miseq PE300 platform (Illumina, San Diego,
CA, USA). The raw sequencing data obtained from the present
study have been submitted to NCBI Sequence Read Archive
(SRA) with accession no. SUB7174894. The raw sequencing
data obtained from the Illumina platform were stored in
FAS TQ format which contains the information of raw se-
quence reads and corresponding sequencing quality. Accord-
ing to the corresponding barcodes, the raw sequencing data
were separated by samples which permitting up to one mis-
match ( Zhang et al., 2017 ). The following bioinformatics anal-
ysis was achieved using the open-source software pipeline
QIMME v1.8.0 ( Caporaso et al., 2010 ), and detailed protocol
is given in Appendix A Text S2 . Finally, to homogenize se-
quences across samples a random resampling was conducted.
1.3. Statistical analysis
1.3.1. Denition of abundant and rare taxa
In this study, to estimate the relative abundance of abun-
dant and rare subcommunities, the thresholds were dened
as 1% and 0.01% of the total sequence reads ( Wu et al., 2016 ;
Jiao et al., 2017 ). Further to subdivide bacterial communities,
the six categories were dened with reference to recent publi-
cations ( Xue et al., 2018 ; Chen et al., 2019 ): (1) always abundant
taxa (AAT): relative abundance great than 1% in all samples,
(2) always rare taxa (ART): relative abundance less than 0.01%
in all samples, (3) moderate taxa (MT): relative abundance be-
tween 0.01% and 1% in all samples, (4) conditionally abundant
taxa (CAT): relative abundance great than 0.01% in all samples
and 1% in some samples, (5) conditionally rare taxa (CRT): rel-
ative abundance less than 0.01% in some samples but never
journal of environmental sciences 111 (2022) 324–339 327
great than 1% in any sample, (6) conditionally rare and abun-
dant taxa (CRAT): relative abundance ranging from 0.01% to
1%.
1.3.2. Alpha diversity analysis
To compare the variation in the bacterial community, alpha di-
versity indices including Chao1 richness estimator, Ace rich-
ness estimator, Shannon diversity indices, and Simpson di-
versity index were calculated based on OTU table using R
v3.6.1 software (R Foundation for Statistical Computing, Vi-
enna, Austria). Further to calculate the intersections of OTUs
between different months, the Venn diagrams were applied.
Rarefaction curves were calculated by PAST 3 software and vi-
sualized with origin v8.5 software. Furthermore, our data t-
ted to Preston log-normal model using the functions “preston-
t ”and “prestondistr ”in “vegan” of R packages as described in
Logares et al. (2014) .
1.3.3. Beta diversity analysis
Beta diversity (BD) analysis was used to illustrate the varia-
tion of community composition and assembly mechanisms.
The Bray-Curtis dissimilarity between each pair of samples
was calculated in the “vegan” package of R v3.6.1 software.
Then, the monthly difference of Bray-Curtis dissimilarity be-
tween samples was compared and visualized with a boxplot
in Origin v8.5 software. The BD values of different categories
were further partitioned into two components, BD turnover
(species replacement) and BD nestedness (abundance loss and
gain), through BSA and POD method as given in previous
studies ( Baselga and Orme, 2012; Cardoso et al., 2015 ). BSA
and POD methods were performed in R v3.6.1 software us-
ing the “betapart” and “BAT” package, respectively. Non-metric
multidimensional scaling (NMDS) analysis based on the Bray-
Curtis similarity matrix was also used to reveal the spatial
and temporal distribution patterns of bacterial communities
in PRIMER v6.0 software ( Clarke and Gorley, 2006 ).
1.3.4. Co-occurrence network analysis
A co-occurrence network was applied to reveal the interac-
tion between bacterial communities. Firstly, all possible pair-
wise Spearman’s rank correlations between bacterial com-
munities were calculated by a pre-constructed correlation
matrix ( Ju et al., 2014 ). Notably, bacterial communities ap-
peared in at least 20% of the total samples with more than
20 sequences were retained during computation to reduce
the network interface. The correlation between two OTUs
was considered robust and used for later network construc-
tion if Spearman’s correlation coefcient ( R ) was > 0.7 or < -
0.7, and the P -value was < 0.01. Network topological param-
eters including betweenness centrality, closeness centrality,
neighborhood connectivity distribution, path length, shared
neighbors’ distribution and stress centrality distribution were
also calculated in this process. The network visualization and
modular analysis were conducted in Gephi v0.9.2 software.
Keystone species were recognized as the nodes with a high
degree ( > 100) and low betweenness centrality ( < 5000) in the
networks ( Ma et al., 2016 ). In addition, 1000 Erdös–Réyni ran-
dom networks were generated which had the same num-
ber of nodes and edges with the real networks ( Erdös and
Rényi, 1960 )). Each node in the random networks has the same
opportunity to assign to any node as the real networks. The
topological properties of random networks were also calcu-
lated and compared to the ones form the real networks. All the
above statistical analyses were calculated by “vegan”, “igraph”
and “Hmisc” packages in R v3.6.1 software.
1.3.5. Neutral community model for bacterial communities
A neutral community model (NCM) was used to predict the
relationship between OTU’s occurrence frequency and their
relative abundance among the meta-community. According
to Sloan’s NCM, abundant sub-community in the larger mi-
crobial populations tends to disperse stochastically and more
widespread, while rare sub-community is controlled by eco-
logical drift and always occurred randomly ( Sloan et al., 2006 ;
2007 ). Compared with the unied neutral theory of biodiver-
sity, Sloan’s NCM does not contain the process of speciation.
The Sloan’s NCM was tted by non-linear least-squares in the
present study. The efciency of the tting was explained by R
2
which represents the overall degree of the tting. Nm is also an
important parameter to explain the relationship between bac-
terial frequency and their relative abundance, where N is the
size of metacommunity and m is the immigration rate. In ad-
dition, the 95% condence intervals around the tting model
were also calculated using the Wilson score interval. All statis-
tical analysis of Sloan’s NCM was conducted by “Hmisc”, “min-
pack.lm” and “stats4” package in R v3.6.1 software following
the protocol shown in Chen et al. (2019) .
1.3.6. Volcano plot
OTUs with a signicant difference were identied by calcu-
lating the log
2
fold change (log
2
(FC)) of microbial abundance
between months ( Edwards et al., 2015 ). Meanwhile, a paired t -
test was also performed to assess the signicant differences
in this process. These OTUs were signicantly different if the
absolute value of log
2
(FC) > 3 and P -value < 0.01. All statisti-
cal analysis and visualization were conducted by the “DESeq2”
and “ggrepel” package in R v3.6.1 software ( Love et al., 2014 ).
Notably, only OTUs with the absolute value of log
2
(FC) > 8 were
marked in the volcano Plot.
1.3.7. Functional prediction based on PICRUSt
The functions of bacterial communities were predicted us-
ing PICRUSt genome prediction software v0.9.2 ( Langille et al.,
2013 ). The OTU table was rst standardized to remove the in-
uence of 16S rRNA gene copies number variations among
bacteria. Then, the standardized OTUs were mapped with the
Greengenes database v13.5 to get the COG (Clusters of Orthol-
ogous Groups of proteins) functions ( Natale et al., 2000 ). The
predicted proteins related to nitrogen cycles were discovered
in our study and visualized by heatmap using R v3.6.1 soft-
ware.
1.3.8. Correlation analysis
Tw o - w a y ANOVA was used in the present study to examine
the spatial and temporal effects on Alpha diversity by IBM
SPSS Statistics 25. Mantel test was also applied in our study
to test the relationships between environmental parameters
and bacterial communities by the “vegan” package in R v3.6.1
software. The signicances in the Mantel test are tested based
328 journal of environmental sciences 111 (2022) 324–339
Fig. 2 –Depth proles of temperature ( °C, top), dissolved oxygen (DO, μmol O
2
/L, middle), and pH (bottom) from May to
August. The dashed lines showed distinct stratication in the study area, and the hypoxia zone formed in August.
on 999 permutations, and the environmental parameters with
P values less than 0.05 were recognized as signicant param-
eters.
2. Results
2.1. Hydrography and environmental parameters
The spatial and temporal variation in the hydrographic char-
acteristics along the study region showed in Fig. 2 and Ap-
pendix A Fig. S1 . In the study region, a continuous increase
in seawater temperature was observed from May to August.
In addition, distinct stratication was observed below 10 m at
St. A3 to A5 throughout the sampling period, where the ver-
tical temperature differences ranged between 3 to 8 °C. How-
ever, stratication intensied further during August. In con-
trast to temperature, DO and pH presented a reverse temporal
trend with gradually decreased intensity from May to August.
Ultimately, the hypoxia zone developed below 10 m at St. A3
to A5 during August where the DO concentration dropped to
less than 94 μmol O
2
/L. The nutrient (nitrate, nitrite, and sil-
icate) concentrations were also continuously increased over
time, and accumulated in the hypoxia zones ( < 10 m) in Au-
gust ( Appendix A Fig. S1 ). However, Chl a concentration in the
hypoxia zones presented an inverse trend with a gradual in-
crease from May to July and then rapidly declined in August
( Appendix A Fig. S1 ).
2.2. Sequencing analysis and diversity estimates
In this study, the valid tags varied from 22,192 to 48,657 per
sample after quality control. To minimum the sample differ-
ence, resampling was conducted in our study. Finally, the valid
tags in each sample were 15,991 after resampling. Overall,
735,586 sequences and 3043 OTUs at 97% 16S rRNA gene simi-
larity were obtained for further analysis. The total rarefaction
curve was nearly plateau, but the separate rarefaction curves
in each month were not plateau ( Appendix A Fig. S2 ). This
imperfection might lead to a slight distortion of the computa-
tional solution (e.g. Bray-Curtis dissimilarity). It is the effect of
sequencing depth, and, not been resolved till date ( Chen et al.,
2019 ; Mo et al., 2018 ; Xue et al., 2018 ; Zhang et al., 2014 ). This ef-
fect was mainly reected in losing some extremely rare OTUs.
Even though we still detected a considerable amount of rare
community, and the alpha and beta diversity all presented
great differences. The maximum likelihood and the Quasi-
poisson methods were used to t sequencing data to reect
the sequencing depth. The tting results revealed that our
sequencing covered almost 88.2% of the total OTUs in the
present study. Furthermore, the good coverage in this study
was greater than 98% (data was not shown here), suggesting
that the sequencing efforts were rational, though a small per-
centage of species were still not detected. From the alpha di-
versity index, samples in July and August were estimated to
have higher diversity than samples from other months ( Ap-
pendix A Fig. S3 ).
According to the denition, CRAT was the most domi-
nant group in the present study, followed by AAT, CAT, CRT,
ART, and MT ( Appendix A Table S1 ). In general, a total of 84
OTUs (2.76%) were identied as CRAT, and they contributed
to 31.14% of the total sequences. Though only 1 OTU (OTU1)
was recognized as AAT, it was the most dominant species ac-
counting for 29.54% of the total sequences. For CAT, 17 OTUs
(0.56%) were classied and represented 19.46% of the total se-
quences. In summary, only 18 OTUs were classied as abun-
dant taxa (AAT + CAT), however, they contributed more than
49% to the total sequences. In contrast, RT (ART + CRT) con-
tributed to the vast majority of OTUs (96.61%), but they only
represented 19.56% of the total sequences. MT was also de-
journal of environmental sciences 111 (2022) 324–339 329
Table 1 –Two-way analysis of variance (ANOVA) revealing the effects of time and depth on the alpha-diversity of microbial
communities
∗.
All AT (AAT + CAT) CRAT RT (ART + CRT)
F P F P F P F P
Time
Richness 14.442 0 / / 94.192 0 14.295 0
Chao1 9.31 0 / / 50.139 0 10.494 0
ACE 7.417 0.001 / / 43.135 0 8.535 0.001
Shannon 24.787 0 32.205 0 48.231 0 17.006 0
Simpson 15.843 0 23.848 0 14.737 0 18.607 0
Pielou 24.501 0 32.205 0 26.123 0 5.706 0.004
Depth
Richness 0.954 0.526 / / 1.395 0.256 0.875 0.59
Chao1 0.643 0.788 / / 5.879 0 0.53 0.875
ACE 0.711 0.73 / / 3.468 0.009 0.621 0.805
Shannon 1.538 0.2 1.194 0.36 1.809 0.125 1.578 0.187
Simpson 2.607 0.033 1.203 0.354 1.164 0.378 1.907 0.106
Pielou 1.634 0.17 1.194 0.36 2.396 0.047 1.451 0.233
Time ×Depth
Richness 0.55 0.842 / / 0.713 0.712 0.873 0.581
Chao1 0.305 0.975 / / 2.489 0.044 0.657 0.758
ACE 0.474 0.895 / / 1.082 0.428 0.744 0.686
Shannon 1.538 0.206 1.887 0.116 2.333 0.057 0.672 0.746
Simpson 2.141 0.077 1.514 0.214 1.317 0.295 0.69 0.731
Pielou 1.894 0.115 1.887 0.116 2.657 0.034 0.44 0.916
All: the whole prokaryotic communities; AT: abundant taxa; AAT: always abundant taxa; CAT: conditionally abundant taxa; RT: rare taxa; ART:
always rare taxa; CRT: conditionally rare taxa; CRAT: conditionally rare and abundant taxa; MT: moderate taxa.
∗Bold font means the signicance at P < 0.05 level; “/” means two-way analysis of variance was not calculated for the lack of alpha diversity, MT
was not included in the two-way ANOVA for only a single OTU.
tected in our study, but only contain 1 OTU and 1436 se-
quences (0.02% of the total sequences).
2.3. Spatio-temporal dynamics of the bacterial
community
Tw o - w a y ANOVA revealed that time (months) have signicant
effects on all OTUs, AT, RT, and CRAT, whereas depth and their
interactions have no signicant effects on the bacterial com-
munity ( Ta ble 1 ). Further, we also found that the bacterial
community showed a striking similarity with CRAT and RT but
distinct from AT from the two-way ANOVA analysis. Spatio-
temporal distribution patterns of the bacterial communities
in different categories were presented by non-metric multidi-
mensional scaling (NMDS) analysis ( Fig. 3 a ). In the four NMDS
gures, samples collected in the same months were almost
clustered together. This emphasizes that time has signicant
effects on the microbial community as revealed in two-way
ANOVA ( Ta b le 1 ). The Venn diagram further illustrated a to-
tal of 2559 OTUs (84.1%) varied in the four months, and the
changed species belonged to CRAT, ART, and CRT ( Fig. 3 b ).
In addition, the time-lag regression analysis of Bray-Curtis
dissimilarity, Jaccard dissimilarity, and Sørensen dissimilar-
ity presented positive slopes. Especially the slopes of RT and
CRAT steeper than those of All and AT indicated that the
bacterial communities were undergone a directional change
( Figs. 3 c and 4 ). The results of beta diversity partitioning indi-
cated that species replacement (turnover) was the major fac-
tor determining the beta diversity and species change over
time ( Fig. 4 ). However, nestedness and/or richness component
only account for a minor contribution ( Fig. 4 ).
2.4. Species composition and diversity changes
The 3043 OTUs belonged to 26 phyla, wherein Proteobac-
teria and Bacteroidetes were the most dominant groups in
the study area, followed by Firmicutes, Actinobacteria, and
Cyanobacteria, etc. ( Appendix A Fig. S4 and Table S2 ). Nearly
half of the OTUs (49.49%) were identied as Proteobacte-
ria, where 539 OTUs belonged to Alphaproteobacteria, 76
to Betaproteobacteria, 616 to Gammaproteobacteria, 194 to
Deltaproteobacteria, 14 to Epsilonproteobacteria, and the rest
67 OTUs were unidentied Proteobacteria. Within the iden-
tied bacterial taxa, OTU1 ( Candidatus Pelagibacter , Alphapro-
teobacteria) were dominated in all samples and the only
species classied in the group of AAT. In addition, 12 and 42
OTUs in Proteobacteria were clustered as CAT and CRAT, in-
dicated Proteobacteria as the major contributor within abun-
dant taxa. Bacteroidetes was the second dominant phylum,
which accounts for 31.25% of the total OTUs and 22.44% of
the total sequences. Overall, 4 and 29 OTUs in Bacteroidetes
were classied in CAT and CRAT, respectively, whereas the rest
OTUs belonged to rare taxa. Other 24 phyla only contributed
to 6.33% of the total sequences, and the vast majority of these
OTUs were clustered into rare taxa. In the context of temporal
variation, the relative abundance of Flavobacteriia and Sphin-
330 journal of environmental sciences 111 (2022) 324–339
Fig. 3 –Microbial communities structuring over the four months in the Bohai Sea. (a) Non-metric multidimensional scaling
(NMDS) analysis of different microbial communities. Note that the gradation of color in the map represent the change of
sampling depths; (b) Ve nn diagram showing numbers of unique and shared OTUs between different months for different
communities; (c) Time-lag regression analysis of Bray-Curtis dissimilarity of different microbial communities.
gobacteriia decreased, while Gammaproteobacteria, Deltapro-
teobacteria, Cytophagia and Actinobacteria (at the phylum
level) increased from May to August ( Appendix A Fig. S4 ). The
volcano plots showed that the number of differential OTUs
(|log
2
(FC)| > 3) were higher in all samples ( Fig. 5 a ) than that in
the hypoxia zones samples ( Fig. 5 b ). In addition, we found that
the OTUs with the biggest changes (|log
2
(FC)| > 8, OTU names
were showed in boxes in Fig. 5 ) were nearly the same in the
two groups. Further statistical analysis presented that the dif-
ferential OTUs were mainly from the phyla of Proteobacteria
and Bacteroidetes, and 98.8% signicantly enriched and de-
pleted OTUs belonged to the groups of CRAT and CRT ( Ap-
pendix A Tab l e S2 ). This result tted with the time-lag regres-
sion analysis of Bray-Curtis dissimilarity that CRAT and RT all
showed higher slopes than other groups ( Fig. 3 c ).
2.5. Network analysis of the co-occurrence patterns
Co-occurrence networks were constructed and presented to
expound the interaction between bacterial sub-communities
( Fig. 6 ). In general, a total of 519 nodes and 6767 edges were
included in the network analysis, with a much higher number
of strong positive correlations than negative correlations. CRT
was the most dominant group in the networks which account-
ing for 81.31% of the nodes ( Fig. 6 a ), followed by CRAT (15.22%),
CAT (3.28%), AAT (0.19%), and MT (0.19%). However, ART was
not detected in the networks. The positive and negative corre-
lations between CRT and CRAT within network analysis were
greater than other pairwise correlations, implied these two
groups might play a signicant role in the ecosystem.
In addition, identical-sized Erdös–Réyni random networks
were generated in the present study to compare with the
real networks ( Appendix A Fig. S5) . These results highlighted
that the modularity, clustering coefcient (CC), average path
length, network diameter of the real co-occurrence networks
was greater than the corresponding random networks ( Ap-
pendix A Fig. S6 and Tabl e 2 ). The high ratio of CC/CC
r
(11.66,
wherein CC
r was the clustering coefcient of correspond-
ing random networks) proved that the real networks con-
tained “small world” properties, where nodes were more con-
nected in the real networks than in the corresponding ran-
dom networks. Further, the degree of the real networks and
the identical-sized random networks were tted by different
methods. This revealed that the real network structure fol-
lowed exponential distribution (non-random), while the iden-
tical sized random networks followed Gaussian distribution
journal of environmental sciences 111 (2022) 324–339 331
Fig. 4 – Time-lag regression analysis of total and compositional beta diversity based on BAS method and POD method. (a)
Time-lag regression of Jaccard dissimilarity ( βjac
) and partitioned turnover component ( βjtu
) and nestedness-resultant
component ( βjne
) by BAS method; (b) Time-lag regression of Sørensen dissimilarity ( βsor
) and partitioned turnover
component ( βsim
) and nestedness-resultant component ( βsne
) by BAS method; (c) Time-lag regression of Jaccard
dissimilarity ( βcc
) and partitioned turnover component ( β-3
) and richness component ( βrich
) by POD method.
Table 2 – Topological properties of the real co-occurring networks and random networks of bacterial communities.
MD CC APL ND AD GD
The real co-occurrence networks 0.678 0.583 3.292 8 26.077 0.050
The corresponding random networks
∗0.136 0.050 2.205 3 26.077 0.050
MD: Modularity; CC: Clustering coefcient; APL: Averag e path length; ND: Network diameter; AD: Average degree; GD: Graph density.
∗Random network was generated based on 1000 Erdös–Réyni model networks which shared the same nodes and edges with the real networks.
( Appendix A Fig. S5 ). The neutral community model (NCM)
was used to t the relationship between the occurrence fre-
quencies. The mean relative abundances of OTUs explained
community variance with 57.2%, 26.4%, and 52.1% for all,
abundant and rare OTUs, respectively (Appendix A Fig. S7 ).
The low explanation of NCM indicated that the bacterial com-
munities might not be solely driven by stochastic processes,
especially for abundant species.
2.6. Modular structure of the co-occurrence network and
keystone species
The value of modularity in a real network was 0.678 ( > 0.4),
which implied the network had a good modular structure
( Newman, 2006 ). Overall, the real network was divided into
four major modules ( > 10%), accounting for more than 90%
of the total nodes ( Fig. 6 b ). The OTUs in the same mod-
332 journal of environmental sciences 111 (2022) 324–339
Fig. 5 –Volcano plot showing differential OTUs between two consecutive months. (a) all samples; (b) samples in hypoxia
zones ( < 10 m). Each point represented an individual OTU, and the position along the x -axis represents the abundance fold
change. The dashed line showed the threshold of signicant differential OTUs (|log
2
(FC)| > 3). Blue dots, red dots, and grey
dots represented signicant enriched OTUs (up), signicant depleted OTUs (down), and no difference OTUs, respectively.
Padj: adjusted P -value.
ule were almost clustered together, while the OTUs at phy-
lum/class level appeared disorganized ( Fig. 6 c ). Terna ry plot
analysis indicated the OTUs in different modules appeared
in different months ( Appendix A Fig. S8 ). For instance, the
OTUs in module I was more abundant in July, while the OTUs
in module II tended to appear in August. However, in the
ternary plot at the phylum/class level, they were disorganized
but followed a pattern. For example, Alphaproteobacteria
and Gammaproteobacteria were widely distributed through-
out the study period, whereas Deltaproteobacteria was more
abundant in August and Flavobacteria during July and Au-
gust. During the analysis, we set the OTUs with a degree > 100
and betweenness centrality values < 5000 as keystone species
( Xue et al., 2018 ). Based on thresholds, a total of 16 OTUs were
recognized as keystone species, including Alphaproteobacte-
ria (5 OTUs), Gammaproteobacteria (5 OTUs), Flavobacteriia
(4 OTUs), Deltaproteobacteria (1 OTUs), and unclassied (1
OTUs). All 16 OTUs have belonged to the group of CRAT (7
OTUs) and CRT (9 OTUs). The majority of these OTUs were
grouped into other modules.
2.7. Inuence of environmental characteristics on
bacterial community distribution
The Mantel tests illustrated the inuence of different environ-
mental factors on shaping the bacterial community structure
in the study region ( Tabl e 3 ). Time, temperature, salinity, pH,
DO, DIC, nutrients like nitrate nitrogen, and silicate presented
signicant positive relationships with the groups of all bacte-
rial communities (All), CRAT, and RT. Moreover, Chl a and POC
were signicantly correlated with the variation of all bacte-
rial communities ( P < 0.01). In contrast to other environmen-
tal factors, only time, ammonia, and Chl a was signicantly
correlated with AT ( P < 0.01). Though we detected many envi-
ronmental factors that might inuence the bacterial commu-
nity, the explained variations reected in redundancy analysis
(RDA) were not high ( Appendix A Fig. S9 ). The explained varia-
tions were 51.5%, 62.5%, 45.1% and 28.8% for All, AT, CRAT and
RT, respectively. In short, the explained variation of AT was
higher than the other three categories. It was greatly different
from the result generated from NCM (Appendix A Fig. S7 ).
journal of environmental sciences 111 (2022) 324–339 333
Fig. 6 –Network analyses showing the co-occurrence pattern of abundant and rare species in the Bohai Sea. Each node
represented an individual OTU, and the edges showing the correlation between two points. A connection stands for a very
strong (Spearman’s R > 0.7 or R < −0.7) and signicant ( P -value < 0.01) correlation. (a) Co-occurring network colored by
different categories. The size of the node represented the number of connections (i.e. Degree). The value outside and inside
the brackets represented positive and negative correlations, respectively. (b) Co-occurring network colored by different
modules. The size of the node represented the value of betweenness centrality. (c) Co-occurring network colored by class
level. The size of the node represented the relative abundance of different species.
2.8. Dynamic of nitrogen cycles along the time gradient
The putative functions of the microbial communities were
predicted by PICRUSt, and the proteins related to nitrogen pro-
cesses had been selected for further analysis ( Fig. 7 ) . Accord-
ing to the COG orthologs, we found that change in an envi-
ronment signicantly affects the abundance of predicted pro-
teins associated with nitrogen pathways, including ammo-
nia assimilation, urea hydrolysis, nitrogen xation, nitrica-
tion, ammonium oxidation (anammox), dissimilatory nitrate
reduction to ammonium (DNRA) and denitrication. We also
provided a preliminary result of the nitrogen cycle change in
the hypoxia zone of the BHS ( Fig. 8 ). In summary, the pre-
dicted functional genes related to nitrogen xation, nitrica-
tion, anammox, ammonia assimilation, and urea hydrolysis
decreased, whereas DNRA and nitrate reduction (rst step of
denitrication) were enhanced during the hypoxia transition.
Meanwhile, later steps of denitrication (stepwise reduction
of nitrite to NO, N
2
O, and N
2
) showed a different pattern i.e.
increased rst and reached a maximum in July, and further
decreased dramatically in August.
3. Discussion
3.1. Spatio-temporal patterns of abundant and rare
microbial communities
Marine bacterial communities play a fundamental role in bio-
geochemical cycles, including carbon, nitrogen, and sulfur cy-
cles, etc. ( Mo et al., 2018 ; Zhang et al., 2014 ). Thus, under-
standing the composition change of bacterial communities
334 journal of environmental sciences 111 (2022) 324–339
Table 3 –Results of Mantel tests between microbial community similarity matrices (Bray-Curtis distance) with spatio-
temporal factors and environmental factors (Euclidean distance) across four sampling months in the Bohai Sea
∗.
All AT (AAT + CAT) CRAT RT (ART + CRT)
R P- value R P -value R P -value R P -value
Time 0.545 0.001 0.189 0.005 0.8 0.001 0.799 0.001
Depth -0.044 0.861 -0.035 0.794 -0.013 0.612 -0.015 0.63
Tem pe ra ture 0.154 0.009 0.026 0.292 0.334 0.001 0.327 0.001
Turbidity 0.062 0.112 0.054 0.142 -0.006 0.486 0.017 0.346
Salinity 0.225 0.001 0.04 0.201 0.413 0.001 0.413 0.001
pH 0.194 0.002 0.048 0.168 0.14 0.001 0.173 0.001
Dissolved oxygen 0.341 0.001 0.09 0.098 0.374 0.001 0.378 0.001
Phosphate phosphorus 0.228 0.004 0.046 0.23 0.168 0.008 0.206 0.001
Nitrite nitrogen 0.038 0.291 -0.068 0.797 0.002 0.458 0.08 0.107
Nitrate nitrogen 0.307 0.001 0.058 0.209 0.331 0.001 0.377 0.001
Ammonia 0.294 0.001 0.153 0.009 0.443 0.001 0.382 0.001
Silicate 0.39 0.001 0.148 0.026 0.4 0.001 0.406 0.001
Chlorophyll-a 0.195 0.01 0.233 0.002 0.048 0.2 0.081 0.068
Dissolved inorganic carbon 0.231 0.005 0.028 0.321 0.483 0.001 0.397 0.001
Particulate organic nitrogen -0.032 0.6887 -0.044 0.721 0.101 0.07 0.082 0.079
Particulate organic carbon 0.186 0.008 0.064 0.196 0.175 0.017 0.155 0.005
∗The signicances are tested based on 999 permutations, and boldface indicates the value is signicant at P < 0.05.
Fig. 7 –Heat map showing the relative abundance of different enzymes related to nitrogen cycles and their corresponding
genes based on the COG database. Notably, the relative abundance of the predicted proteins was log-transformed in the
heatmap.
and their relationships with environmental parameters in the
hypoxia zones of the BHS can extend our horizons about the
stabilization mechanism and feedback regulation of ecosys-
tems ( Caneld et al., 2010 ; Helly et al., 2004). In this study, the
bacterial community from the same month almost clustered
together, and the Bray-Curtis dissimilarity between samples
signicantly increased over time ( Fig. 3 ). Moreover, the Mantel
tests and two-way ANOVA also proved that the bacterial com-
munity changed over time rather than their sampling depths
( Tab l es 1 and 3 ). This emphasizes that temporal differences
of bacterial composition overwhelmed the spatial differences,
and might be inuenced by the changing environmental fac-
tors ( Chen et al., 2019 ; Xue et al., 2018 ). This is also the case
with eukaryotic plankton communities, which were reported
signicantly altered over time rather than depth following a
reservoir cyanobacterial bloom ( Xue et al., 2018 ). Our results
showed that the similarity between samples signicantly de-
creased with the increase in time-lag ( Fig. 3 c ), demonstrating
that bacterial communities were not resilient and had reached
an alternative state ( Xue et al., 2018 ). This change signicantly
alters the bacterial composition of RT and CRAT groups, while
has a minor impact on AT groups. Additionally, our results also
journal of environmental sciences 111 (2022) 324–339 335
Fig. 8 –Schematic representation of nitrogen processes occurring in the hypoxia zones of the Bohai Sea. Every colored arrow
represents a nitrogen transformation: denitrication (red), nitrate deduction (sapphirine), ammonia (orange), nitrogen
xation (gray), DNRA (dark green), nitrication (modena), remineralization (mauve), assimilation (black). Letters in the
brackets indicate downregulation (D), upregulation (U), upregulation rst, and downregulation later (UD) or unknown (?).
OrgN: organic nitrogen.
presented that different taxa did not respond equally to envi-
ronmental changes. Bacterial communities from RT and CRAT
were more capable to respond to environmental changes than
that of AT. This illustrated that the bacterial community from
AT is more resilient and robust to all possible changes under
adverse conditions ( Mo et al., 2018 ; Xue et al., 2018 ).
Signicantly, the two-way ANOVA analysis showed that
the time and depth on alpha-diversity presented strikingly
consistency between all bacterial communities and CRAT/RT
( Tab l e 1 ) . The possible reason could be that alpha-diversity
is concerned about the number of species in local homo-
geneous habitats, and the member from RT and CRAT ac-
counted for the overwhelming majority of total species num-
bers ( Mo et al., 2018 ; Xue et al., 2018 ). Recently, various studies
concerned about the extremely low abundance but high diver-
sity of the bacterial community in various ecosystems. This
fraction of the microbial community is often called a “rare
biosphere” ( Jiao et al., 2017 ). This study detected enormous
amounts of rare microbial subcommunity possessed higher
richness and diversity as also reported in other coastal oceans
( Mo et al., 2018; Natale et al., 2000 ). Although the rare bio-
sphere only accounts for a minor fraction of the whole mi-
crobial community, they might mediate multiple metabolic
capabilities and ecosystem functions ( Coveley et al., 2015 ).
Campbell et al. (2011) suggested that a signicant portion
of the rare community is active, and they might also con-
tribute signicantly to the biogeochemical process. Moreover,
the co-occurrence networks showed that most of the key-
stone species belonged to a rare community (9/16 OTUs) ( Ap-
pendix A Tab le S4 ). Thus, the rare community might play a
signicant role in mediating the elemental cycles in the BHS
( Coveley et al., 2015 ; Mo et al., 2018 ). In addition to the rare
biosphere, CRAT may also a vital group correlated with the
biogeochemical process. The most immediate evidence was
that the rest keystone species in the co-occurrence networks
were all belonged to CRAT (7/16 OTUs) ( Appendix A Table S4 ),
demonstrating the fundamental role of CRAT groups in the
construction of bacterial communities. The time-lag regres-
sion slope of CRAT ( R
2
= 0.898, P < 0.01) even greater than that
of the RT ( R
2
= 0.840, P < 0.01), illustrating the members of CRAT
could undergo a more drastic change. We found that many
species in CRAT turned from RT into AT or from AT into RT
over time. For instance, OTU1556 (Rhodobacterales), the sec-
ond dominant species in this region, emerged with the de-
crease of oxygen and was only detected in July and August.
Presumably, the members of CRAT were more sensitive to the
change of environmental parameters.
3.2. Dynamics of microbial communities and their
function changes
Previous studies have documented that DO scarcity and as-
sociated ocean acidication may lead to profound and last-
ing consequences on bacterial community structure and their
biogeochemical functions ( Bulow et al., 2010; Carolan et al.,
2015; Orsi et al., 2012; Stevens and Ulloa, 2008 ). In the present
study, Proteobacteria and Bacteriodetes were the most dom-
inant groups in all water samples and their dominance re-
sembled with earlier studies from a variety of coastal hypoxia
zones ( Campbell et al., 2011 ; Mo et al., 2018 ). Alphaproteobac-
teria had the highest abundance among Proteobacteria groups
and contributed the largest fractions of relative abundance in
all samples ( Appendix A Fig. S4 ). The phylotypes afliated to
Alphaproteobacteria detected in our study including SAR11
clade, Rhodobacterales, Rickettsiales, and Sphigomonadales.
SAR11 clade was reported to adapt to various habitats, and
not only thrive in oxygen-rich surface waters but also oxygen-
depleted water ( Morris et al., 2002 ; Rappe et al., 2002 ). A re-
cent study showed that SAR11 nar G genes, a gene that encodes
proteins catalyzing the rst step of denitrication, constituted
about 40% of the OMZ nar G transcripts ( Tsementzi et al., 2016 ).
The Alphaproteobactrial Rhodobacterales group was reported
336 journal of environmental sciences 111 (2022) 324–339
to correlate with fundamental metabolic processes such as
sulfur oxidation, nitrogen xation, and autotrophic carbon
xation ( Jain et al., 2014 ; Liu et al., 2015 ). The prevalence
of Rhodobacterales (OTU1556) reected the near anoxia and
highly polluted condition in the BHS. Liu et al. (2015) also re-
ported that members of Rhodobacterales increased when oxy-
gen was depleted in the Pearl Estuary. We observed a contin-
uous increase of the relative abundance of Gammaproteobac-
teria over time in the present study, and Gammaproteobac-
teria even dominated in some samples below 10 m in Au-
gust ( Appendix A Fig. S4 ). The dynamic of Gammaproteobac-
teria over time were mainly induced by the growth and de-
cline of Oceanospirillales. Members of Oceanospirillales are
known to be symbionts with various marine invertebrates
( Liu et al., 2015 ; Pajares et al., 2020 ). In addition, some mem-
bers of Oceanospirillales were reported to contain genes for
carbon xation (RuBisCO genes), nitrogen xation, and sul-
fur oxidation, revealing the importance of this order in ele-
mental cycling ( Pajares et al., 2020 ; Pujari et al., 2019 ). Among
Bacteroidetes, Flavobacteriia and Sphingobacteriia were two
groups dominant in the study region, and all tended to de-
crease over time, as previously observed in the Arabian Sea
OMZs. Pinhassi et al. (2004) reported that Bacteroidetes were
abundant during and following algal blooms in aquatic habi-
tats. Presumably, the relative abundance of Bacteroidetes de-
creased with the exhaustion of organic matter over time.
The nitrogen cycle has been thought to be more sensitive
than other elemental cycles at low levels of oxygen ( Bulow
et al., 2010; Fennel and Tes ta, 2019 ). The main nitrogen path-
ways including nitrogen xation, nitrication, denitrication,
anammox, and DNRA were predicted by PICRUSt in our study
( Fig. 7 ) . Results showed that the associated proteins of these
pathways varied dramatically during the time series. Earlier
it was believed that the hypoxia zones are losing xed in-
organic nitrogen via microbial-mediated denitrication and
anammox, but the efciency of nitrogen removal by the two
processes is now controversial ( Bulow et al., 2010; Fennel and
Tes ta, 2019; Kong et al., 2013 ). Until the discovery of anammox,
denitrication was thought to be the most important nitro-
gen removal pathway in the oxygen-depleted region. Further
studies evinced that anammox is also an important nitrogen
removal pathway in some of the OMZs ( Bulow et al., 2010 ).
Nitrate reduction was the most active step in the progress of
denitrication in this study, and these observations matched
well with the nitrite accumulation measured by chemical as-
say. Nitrite accumulation or secondary nitrite maxima also re-
ported from the OMZs of the Arabian Sea, Peruvian region,
and Eastern Tropical South Pacic regions ( Claudia et al., 2016 ;
Stewart et al., 2012 ). In the BHS, the later steps of denitrica-
tion (stepwise reduction of nitrite to NO, N
2
O, and N
2
) were
rst increased and reached a maximum in July, then dropped
rapidly in August. In this tendency, the expression of nir BD,
nor B, and nos Z genes were following the dynamic of Chl a .
Thus, it can be hypothesized that the organic carbon source
was not enough to support denitrication in August since
denitrication needs organic carbon as an electron acceptor
( Kong et al., 2013 ). In contrast to denitrication, anammox was
continuously declined during the study period ( hzs ABC gene).
Theoretically, a decline in DO concentration could be bene-
cial for the survival and expression of anammox bacteria, as
anammox is an anaerobic process ( Lam et al., 2007 ). In the
low oxygen waters, the excess nitrite mainly affects the dis-
tribution of anammox bacteria, although it’s a substrate for
the anammox process ( Kong et al., 2013 ). The oxygen-nitrite
co-limit the distribution of anammox bacteria was also re-
ported in many other oceans, such as the Arabian Sea and
Black Sea ( Kuypers et al., 2005 ; Bulow et al., 2010 ). Thus, the
increased nitrite concentration could be the reason for the in-
hibition of the anammox bacteria community in the present
study. However, there is no consensus yet about which one is
the main nitrogen removes pathway in the global ocean OMZs
ecosystems ( Kuypers et al., 2005 ; Lam et al., 2007 ). It is sug-
gested that the ratio of anammox to denitrication in OMZs
depends on the stoichiometry of organic matter supply, and
is likely transformed with season and location ( Bulow et al.,
2010 ; Claudia et al., 2016 ).
3.3. Factors regulating the microbial communities in the
BHS
Many studies discussed the control of diverse environmen-
tal factors on modulating the microbial community in var-
ious habitats ( Chen et al., 2019 ; Jiao et al., 2017 ; Xue et al.,
2018 ). However, the microbial community structure in na-
ture is not shaped by environmental factors solely. Recent ad-
vanced studies revealed that the microbial communities are
co-regulated by both stochastic processes (NCM) and deter-
mined parameters (Niche model) ( Chen et al., 2019 ; Mo et al.,
2018 ; Xue et al., 2018 ). In other words, the NCM is challeng-
ing the traditional thinking “everything is everywhere, the en-
vironment selects” ( Bass-Becking, 1934 ). This study revealed
the multiple environmental cues, such as temperature, DO,
pH, etc. inuence the dynamic of the bacterial community
( Tab l e 3 ). Furthermore, the Mantel tests showed the relation-
ship of environmental parameters with CRAT and RT were
nearly consistent with all bacterial communities, while it var-
ied with AT. Such as, pH and DO only affected the distribu-
tion of CRAT and RT, but not signicantly inuenced AT. Even
the identical parameters and their correlation with different
categories (All, AT, CRAT, and RT) were also varied. For in-
stance, the correlation of time with CRAT ( R = 0.8, P < 0.001) and
RT ( R = 0.799, P < 0.001) was greater than that of AT ( R = 0.189,
P < 0.005). This might be caused by the wide distribution of
the most dominant species ( Candidatus Pelagibacter ), and their
stronger adaption to various environments (e.g., nutrient de-
pletion). Our results proved that abundant community was
ubiquitous and more likely inuenced by dominant parame-
ters. Similar observations also noted from different (soil, reser-
voir, and lake) ecosystems ( Chen et al., 2019 ; Xue et al., 2018 ;
Jiao et al., 2017 ). Xue et al. (2018) reported that abundant eu-
karyotic plankton subcommunities had higher niche breadth
than the rare ones. Presumably, in the present study, the
abundant sub-community is also inuenced by higher niche
breadth. However, the rare microbial communities were more
likely inuenced by stochastic processes ( Mo et al., 2018 ). Fur-
ther, the NCM results conrmed this hypothesis, where the
NCM of rare sub-community ( R
2
= 0.521) had a higher tting
degree than the abundant ones ( R
2
= 0.264) ( Appendix A Fig.
S7 ). From the combined results of NCM and RDA, it can be ob-
served that the neutral processes explained more than half of
journal of environmental sciences 111 (2022) 324–339 337
variations of rare taxa, but local environmental explained less
than 30% variations of rare sub-community ( Appendix A Figs.
S7 and S8 ). This scenario was different for abundant taxa, the
neutral processes explained less variations whereas local en-
vironmental factors explained most of the variations. Hereby,
we speculate that though the bacterial communities were co-
regulated by stochastic processes and environmental ltering,
the inuence degree of the two processes to abundant and
rare subcommunities was totally different ( Chen et al., 2019 ;
Mo et al., 2018 ).
In the Mantel tests, the R -value ( R = 0.341, P < 0.01) between
DO and all bacterial communities was higher than other envi-
ronmental factors except silicate. This proves DO plays a vital
role in shaping the bacterial community. However, the oxygen
tolerance capacity varies between different species with aer-
obic and anaerobic respiration. On a deeper level, some en-
zymes involved in the cells are sensitive to oxygen. For in-
stance, nitrogenase, an enzyme responsible for nitrogen x-
ation, is easy to get inactivated by oxygen, and thus nitrogen
xation is a strict anaerobic reaction. In brief, oxygen is not the
sole parameter to regulate bacterial community, but it must
be a vital parameter. Likewise DO, many other environmental
parameters including nutrients (nitrate, ammonia, and phos-
phate) also showed signicant correlations with the bacterial
community. This might be because nutrients play a vital role
in the growth and development of the bacterial community
( Mo et al., 2018 ). Silicate was also a major limiting factor in
shaping the biogeography of bacterial community in the study
region ( Ta b le 3 ) as observed previously ( Zhang et al., 2014 ).
That probably because silicate is an essential nutrient for the
growth of various diatoms, and the ourishing of diatoms re-
leases plentiful DOC to support the growth of bacterial com-
munities ( Töpper et al., 2010 ). Diatom is reported to dominant
in the ocean ecosystems when silicate and other mineral nu-
trients are adequate. Turbidity was also measured in our study
but showed no correlation with all subcommunities. This re-
sult was typically different from a previous study conducted
in the junction of the BHS and the yellow sea ( Yu et al., 2018 ).
As we know, high turbidity can result in a decrease in light
penetration, and affect the growth of photosynthetic bacteria
(i.e. Synechococcus ). Cyanobacteria only accounted for a minor
fraction in our study, but it was the dominant bacteria in the
previous study ( Yu et al., 2018 ). That is why we get a different
result about the relationship between turbidity and bacterial
community.
4. Conclusions
This study provided the evidence of composition, distribution,
and variation of the bacterial community in the hypoxia zone
of the BHS. Through a continuous sampling effort from May
to August 2017, an intensication of hypoxia was detected be-
low 10 m depths towards the north of the dual-core struc-
ture in this region. Our 16S rRNA gene-based high-throughput
sequencing detected a large amount of bacterial community
including various rare species, which were often overlooked
previously. Though the bacterial communities co-regulated by
stochastic processes and environmental ltering, the inu-
ence of the two processes on abundant and rare subcommu-
nities could be different. Rare subcommunities could be more
capable to respond to environmental changes, whereas the
abundant bacterial communities were more resilient and ro-
bust to all possible changes under adverse conditions. More-
over, this study demonstrated that rare subcommunity had
stronger temporal heterogeneity than that of abundant sub-
community, and might play a vital role in regulating the ni-
trogen cycles in the BHS. Here, DNRA and nitrate reduction
(rst step of denitrication) enhanced, whereas nitrogen xa-
tion, nitrication, anammox, ammonia assimilation, and urea
hydrolysis all decreased. Further, the later steps of denitri-
cation (stepwise reduction of nitrite to NO, N
2
O, and N
2
) were
increased rst and decreased later. This result foreboded that
nitrite was accumulated in the hypoxia zone, and matched
well with the chemical assay. Even though, our results only
supplied a rough and preliminary estimation. Many functional
groups were not included in our study for the limitation of
PCR amplication. For example, ammonia-oxidizing archaea
(AOA) and ammonia-oxidizing bacteria (AOB) all participate in
the process of anammox oxidize. However, the 16S rRNA gene-
based high-throughput sequencing only detected the group of
AOB, and AOA was lost in the library. Thus, further investi-
gations on niche specialization and eco-physiological charac-
terization of various functional groups are needed to, (1) bet-
ter understand the interactions between different functional
groups, and (2) evaluate how these interactions inuence the
biogeochemical cycles in changing aquatic ecosystems.
Acknowledgments
This work was supported by the National Key Research and
Development Project of China (No. 2019YFC1407803), the Na-
tional Natural Science Foundation of China (No. 41876134),
and the Changjiang Scholar Program of Chinese Ministry of
Education (No. T2014253) to Jun Sun, and the Endowment
Fund from Stroud Water Research Center to Jinjun Kan.
Supplementary materials
Supplementary material associated with this article can be
found, in the online version, at doi:10.1016/j.jes.2021.04.013 .
Baselga, A., Orme, C.D.L., 2012. Betapart: An R package for the
study of beta diversity. Methods Ecol. Evol. 3, 808–812 .
Bass-Becking, L.G.M., 1934. Geobiologie of Inleiding tot de
Milieukunde. WP Van Stockum & Zoon, The Hague, the
Netherlands .
Bulow, S.E., Rich, J.J., Naik, H.S., Pratihary, A.K., Ward, B.B., 2010.
Denitrication exceeds anammox as a nitrogen loss pathway
in the Arabian Sea oxygen minimum zone. Deep-Sea Res. Pt. I
57 (3), 384–393 .
Campbell, B.J., Yu, L.Y., Heidelberg, J.F., Kirchman, D.L., 2011.
Activity of abundant and rare bacteria in a coastal ocean. P.
Natl. Acad. Sci. USA 108 (31), 12776–12781 .
Caneld, D.E., Stewart, F.J., Thamdrup, B., Brabandere, L.D.,
Dalsgaard, T., Delong, E.F., et al., 2010. A cryptic sulfur cycle in
338 journal of environmental sciences 111 (2022) 324–339
oxygen-minimum-zone waters off the Chilean Coast. Science
330 (6009), 1375–1378 .
Caporaso, J.G., Kuczynski, J., Stombaugh, J., Bittinger, K.,
Bushman, F.D., Bushman, F. D., e t al ., 2010. QIIME allows
analysis of high-throughput community sequencing data.
Nat. methods 7 (5), 335–336 .
Carolan, M.T., Smith, J.M., Beman, J.M., 2015. Transcriptomic
evidence for microbial sulfur cycling in the eastern tropical
North Pacic oxygen minimum zone. Front. Microbiol. 6, 334 .
Cardoso, P. , R i g a l , F. , C a r v a l h o, J. C., 2015. BAT – Biodiversity
Assessment Tools, an R package for the measurement and
estimation of alpha and beta taxon, phylogenetic and
functional diversity. Methods Ecol. Evol. 6, 232–236 .
Chen, W.D . , R e n , K.X., Isabwe, A., Chen, H.H., Liu, M., Yang, J., 2019.
Stochastic processes shape microeukaryotic community
assembly in a sub-tropical river across wet and dry seasons.
Microbiome 7, 138 .
Clarke, K.R., Gorley, R.N., 2006. PRIMER v6: User manual/tutorial.
PRIMER-E, Plymouth, UK .
Claudia, L., Speth, D.R., Kox, M.A.R., Villanueva, L., Jetten, M.S.M.,
2016. Metagenomic analysis of nitrogen and methane cycling
in the Arabian Sea oxygen minimum zone. Peerj 4 (4), e1924 .
Coveley, S., Elshahed, M.S., Youssef, N.H., 2015. Response of the
rare biosphere to environmental stressors in a highly diverse
ecosystem (Zodletone Spring, OK, USA). PeerJ 3 (8), e1182 .
Edwards, J., Johnson, C., Santos-Medellín, C., Lurie, E.,
Podishetty, N.K., Bhatnagar, S., et al., 2015. Structure, variation,
and assembly of the root-associated microbiomes of rice. P.
Natl. Acad. Sci. USA 112 (8), 911–920 .
Erdös, P., Rényi, A., 1960. On the evolution of random graphs. T.
Am. Math. Soc. 5, 17–61 .
Fennel, K., Testa, J.M., 2019. Biogeochemical controls on coastal
hypoxia. Annu. Rev. Mar. Sci. 11 (1) 4.1-4.26 .
Helly, J.J., Levin, L.A., 2004. Global distribution of naturally
occurring marine hypoxia on continental margins. Deep Sea
Res. Part I: Oceanogr. Res. Papers 51 (9), 1159–1168 .
Jain, A., Bandekar, M., Gomes, J., Shenoy, D., Meena, R.M., Naik, H.,
Khandeparkar, R., Ramaiah, N., 2014. Temp or ally invariable
bacterial community structure in the Arabian Sea oxygen
minimum zone. Aquat. Microb. Ecol. 73 (1), 51–67 .
Jiao, S., Chen, W.M., Wei, G.H., 2017. Biogeography and ecological
diversity patterns of rare and abundant bacteria in
oil-contaminated soils. Mol. Ecol. 26 (19), 5305–5317 .
Ju, F., X i a, Y., Guo, F., Wang, Z.P., Zhang, T., 2014. Taxo n omi c
relatedness shapes bacterial assembly in activated sludge of
globally distributed wastewater treatment plants. Environ.
Microbiol. 6 (8), 2421–2432 .
Kong, L.L., Jing, H.M., Kataoka, T., B uchw a l d , C., Liu, H.B., 2013.
Diversity and spatial distribution of hydrazine oxidoreductase
( hzo ) gene in the oxygen minimum zone off Costa Rica. PLoS
One 8 (10), e78275 .
Kuypers, M.M.M., Lavuk, G., Woebken, D., Schm id, M., Fuchs, B.M.,
Amann, R., et al., 2005. Massive nitrogen loss from the
Benguela upwelling system through anaerobic ammonium
oxidation. P. Natl. Acad. Sci. USA 102 (18), 6478–6483 .
Lam, P., J e n s e n , M.M., Lavik, G., McGinnis, D.F., Müller, B.,
Schubert, C.J., et al., 2007. Linking crenarchaeal and bacterial
nitrication to anammox in the Black Sea. P. Natl. Acad. Sci.
USA 104 (17), 7104–7109 .
Langille, M.G.I., Zaneveld, J., C aporaso, J.G., McDonald, D.,
Knights, D., Reyes , J.A., et al., 2013. Predictive functional
proling of microbial communities using 16S rRNA marker
gene sequences. Nat. Biotechnol. 31 (9), 814–821 .
Levin, L.A., Ekau, W., Gooday, A.J., Jorissen, F., Middelburg, J.J.,
Naqvi, S.W.A., et al., 2009. Effects of natural and
human-induced hypoxia on coastal benthos. Biogeosciences 6
(10), 2063–2098 .
Liu, J.W., Fu, B.B., Yang, H.M., Zhao, M.X., He, B.Y., Zhang, X.H.,
2015. Phylogenetic shifts of bacterioplankton community
composition along the Pearl Estuary: the potential impact of
hypoxia and nutrients. Front. Microbiol. 6 (64), 64 .
Logares, R., Stéphane, A., Bass, D., Bittner, L., Boutte, C.,
Christen, R., et al., 2014. Patterns of Rare and Abundant
Marine Microbial Eukaryotes. Curr. Biol. 24 (8), 813–821 .
Love, M.I., Huber, W. , A n d e r s , S., 2014. Moderated estimation of
fold change and dispersion for RNA-seq data with DESeq2.
Genome Biol 15 (12), 550 .
Lu, Z.M., Gan, J.P., Dai, M.H., Liu, H.B., Zhao, X.Z., 2018. Joint effects
of extrinsic biophysical uxes and intrinsic hydrodynamics
on the formation of hypoxia west off the pearl river estuary. J.
Geophys. Res.- Oceans 123 (9), 6241–6259 .
Luo, X.F., Wei, H., Fan, R.F., Liu, Z., Zhao, L., Lu, Y. Y. , 2018. Hypoxia
disaster in waters adjacent to the Changjiang estuary. Cont.
Shelf. Res. 152, 1–13 .
Ma, B., Wang, H.Z., Dsouza, M., Lou, J., He, Y. , D a i , Z.M., et al., 2016.
Geographic patterns of co-occurrence network topological
features for soil microbiota at continental scale in eastern
China. ISME J 10, 1891–1901 .
Melzner, F. , T h o m s e n , J., Koeve, W., Oschlies, A., Gutowska, M.A.,
Bange, H.W., et al., 2013. Future ocean acidication will be
amplied by hypoxia in coastal habitats. Mar. Biol. 160 (8),
1875–1888 .
Mo, Y. Y. , Z h a ng , W.J., Yang, J., Lin, Y. S ., Yu , Z., Lin, S.J., 2018.
Biogeographic patterns of abundant and rare
bacterioplankton in three subtropical bays resulting from
selective and neutral processes. ISME J 12, 2198–2210 .
Morris, R.M., Rappe, M.S., Connon, S.A., Vergin, K.L., Siebold, W. A .,
Carlson, C.A., et al., 2002. SAR11 clade dominates ocean
surface bacterioplankton communities. Nature 420, 806–810 .
Naqvi, S.W.A., Bange, H.W., Farías, L., Monteiro, P.M.S.,
Scranton, M.I., Zhang, J., 2009. Coastal hypoxia/anoxia as a
source of CH
4
and N
2
O. Biogeosciences 6 (5), 9455–9523 .
Natale, D.A., Shankavaram, U.T., Galperin, M.Y., Wolf, Y. I .,
Aravind, L., Koonin, E.V., 2000. Genome annotation using
clusters of orthologous groups of proteins (COGs) –towards
understanding the rst genome of a Crenarchaeon. Genome
Biol 1 (5) research0009.1-0009.19 .
Newman, M.E.J., 2006. Modularity and community structure in
networks. P. Natl. Acad. Sci. USA 103 (23), 8577–8582 .
Orsi, W., S o n g , Y.C., Hallam, S., Edgcomb, V. , 2012. Effect of oxygen
minimum zone formation on communities of marine protists.
ISME J 6 (8), 1586–1601 .
Pajares, S., Varona-Cordero, F., Hernández-Becerril, D.U., 2020.
Spatial distribution patterns of bacterioplankton in the
oxygen minimum zone of the tropical Mexican Pacic. Microb.
Ecol. 80, 519–536 .
Pinhassi, J., Sala, M.M., Havskum, H., Peters, F. , G u a d a v o l , O.,
Malits, A., Marrase, C., 2004. Changes in bacterioplankton
composition under different phytoplankton regimens. Appl.
Environ. Microb. 70 (11), 6753–6766 .
Pujari, L., Wu, C., Kan, J., Li, N., Wang, X., Zhang, G., et al., 2019.
Diversity and spatial distribution of chromophytic
phytoplankton in the Bay of Bengal revealed by RuBisCO
genes ( rbc L). Front. Microbiol. 10, 1501 .
Rabalais, N.N., Díaz, R.J., Levin, L.A., Turner, R.E., Gilbert, D.,
Zhang, J., 2010. Dynamics and distribution of natural and
human caused hypoxia. Biogeosciences 7 (2), 585–619 .
Rabalais, N.N., Cai, W.J., Carstensen, J.D., Fry, B., Hu, X.P.,
Quiñones-Rivera, Z., et al., 2014. Eutrophication-driven
deoxygenation in the coastal ocean. Oceanography 27 (1),
172–183 .
Rappe, M.S., Connon, S.A., Vergin, K.L., Giovannoni, S.J., 2002.
Cultivation of the ubiquitous SAR11 marine bacterioplankton
clade. Nature 418, 630–633 .
journal of environmental sciences 111 (2022) 324–339 339
Ravaglioli, C., Bulleri, F., Saskia, R., Sophie, J.M., Helen, S.F.,
Stephen, W. , e t a l., 2019. Ocean acidication and hypoxia alter
organic carbon uxes in marine soft sediments. Glob. Change
Biol. 25 (12), 4165–4178 .
Sloan, W.T., Lunn, M., Woodcock, S., Head, I.M., Nee, S., Curtis, T.P.,
2006. Quantifying the roles of immigration and chance in
shaping prokaryote community structure. Environ. Microbiol.
8 (4), 732–740 .
Sloan, W.T., Woodcock, S., Lunn, M., Head, I.M., Curtis, T. P., 2007.
Modeling taxa-abundance distributions in microbial
communities using environmental sequence data. Microb.
Ecol. 53 (3), 443–455 .
Solórzano, L., Sharp, J.H., 1980. Determination of total dissolved
phosphorus and particulate phosphorus in natural waters.
Limnol. Oceanogr. 25 (4), 754–758 .
Stevens, H., Ulloa, O., 2008. Bacterial diversity in the oxygen
minimum zone of the eastern tropical South Pacic. Environ.
Microbiol. 10 (5), 1244–1259 .
Stewart, C.F.J., Ulloa, O., Delong, E.F., 2012. Microbial
metatranscriptomics in a permanent marine oxygen
minimum zone. Environ. Microbiol. 2012 14 (1), 23–40 .
Thamdrup, B., Dalsgaard, T., R evs b e c h , N.P., 2012. Widespread
functional anoxia in the oxygen minimum zone of the
Eastern South Pacic. Deep-Sea Res. Pt. I 65, 36–45 .
Töpper, B., Larsen, A., Thingstad, T.F., Thyrhaug, R., Sandaa, R.A.,
2010. Bacterial community composition in an Arctic
phytoplankton mesocosm bloom: the impact of silicate and
glucose. Polar Biol 33, 1557–1565 .
Tsementzi, D., Wu, J.Y., Deutsch, S., Nath, S., Rodriguez-R, L.M.,
Burns, A.S., et al., 2016. SAR11 bacteria linked to ocean anoxia
and nitrogen loss. Nature 536, 179–183 .
Wan g, C., Lv, Y., Li, Y., 2018. Riverine input of organic carbon and
nitrogen in water-sediment system from the Yellow River
estuary reach to the coastal zone of Bohai Sea. China. Cont.
Shelf Res. 157, 1–9 .
Wan g, G.G., Peng, J.L., Yang, D.D., Zha ng, D.H., L i, X.G., 2015.
Current levels, composition proles, source identication and
potentially ecological risks of polychlorinated biphenyls
(PCBs) and polybrominated diphenyl ethers (PBDEs) in the
surface sediments from Bohai Sea. Mar. Pollut. Bull. 101 (2),
834–844 .
Wei , Q.W., Wang, B.D., Yao, Q.Z., Xue, L., Sun, J.C., Xin, M., et al.,
2019. Spatiotemporal variations in the summer hypoxia in the
Bohai Sea (China) and controlling mechanisms. Mar. Pollut.
Bull. 138, 125–134 .
Wu, C., Kan, J.J., Liu, H.J., Pujari, L., Guo, C.C., Wang, X.Z., et al.,
2019. Heterotrophic bacteria dominate the diazotrophic
community in the Eastern Indian Ocean (EIO) during
pre-southwest monsoon. Microb. Ecol. 78, 804–819 .
Wu, W.X., Logares, R., Huang, B.Q., Hsieh, C.H., 2016. Abundant
and rare picoeukaryotic sub-communities present contrasting
patterns in the epipelagic waters of marginal seas in the
northwestern Pacic Ocean. Environ. Microbiol. 19 (1), 287–300 .
Xue, Y.Y., Chen, H.H., Yang, R.J., Liu, M., Huang, B.Q., Yang, J., 2018.
Distinct patterns and processes of abundant and rare
eukaryotic plankton communities following a reservoir
cyanobacterial bloom. ISME J 12, 2263–2277 .
Xu, X., Yu, Z.M., Cheng, F. J., He, L.Y., Cao, X.H., Song, X.X., 2017.
Molecular diversity and ecological characteristics of the
eukaryotic phytoplankton community in the coastal waters of
the Bohai Sea. China. Harmful Algae 61, 13–22 .
Yu, S.X., Pang, Y.L., Wang, Y. C . , L i , J.L., Qin, S., 2018. Distribution of
bacterial communities along the spatial and environmental
gradients from Bohai Sea to northern Yellow Sea. PeerJ 6 (1),
e4272 .
Zhai, W.D . , Z h a o , H.D., Zheng, N., Xu, Y. , 2012. Coastal acidication
in summer bottom oxygen-depleted waters in north western
northern Bohai Sea from June to August in 2011. Chinese Sci.
Bull. 57 (9), 1062–1068 .
Zhai, W.D . , Z h a o , H.D., Su, J.L., Liu, P.F., L i , Y.W., Zheng, N., 2019.
Emergence of summertime hypoxia and concurrent carbonate
mineral suppression in the central Bohai Sea. China. J.
Geophys. Res.- Biogeo. 124, 2768–2785 .
Zhang, Y.Y. , Ya ng , Q.S., Ling, J., Van, Nostrand J.D., Shi, Z.,
Zhou, J.Z., et al., 2017. Diversity and structure of diazotrophic
communities in mangrove rhizosphere, revealed by
high-throughput sequencing. Front. Microbiol. 8, 2032 .
Zhang, Y., Z h a o , Z.H., Dai, M.H., Jiao, N.Z., Herndl, G.J., 2014.
Drivers shaping the diversity and biogeography of total and
active bacterial communities in the South China Sea. Mol.
Ecol. 23 (9), 2260–2274 .
Zhao, H.D., Kao, S.J., Zhai, W.D., Zang, K.P., Zheng, N., Xu, X.M.,
et al., 2017. Effects of stratication, organic matter
remineralization and bathymetry on summertime oxygen
distribution in the Bohai Sea. China. Cont. Shelf Res. 134,
15–25 .
