Introduction
Precipitation changes caused by global climate change are predicted to be
increasingly severe over the coming century (IPCC, 2007; Seager et al.,
2007). Future projected precipitation patterns vary spatially and temporally,
and the complexity and unpredictability of precipitation changes have
exceeded other global changes such as elevated CO2 and temperature
(Beier et al., 2012). In addition to the frequency and intensity of
precipitation events, seasonal precipitation changes are of increasing
severity in some regions of the world (Easterling et al., 2000). For example,
an analysis of 60 years of precipitation data showed remarkable seasonal
precipitation redistribution in subtropical China, with more frequent
droughts in the dry season and extreme rainfall events in the wet season (Zhou et
al., 2011). In contrast to changes in total annual precipitation,
redistribution of seasonal precipitation may be more important in controlling
ecosystem function in subtropical forests, due to strong contrasts between dry
and wet seasons (Wang et al., 2009). Recent meta-analyses on precipitation
manipulation experiments pointed out the lack of data in the warm and humid
monsoon zones (Wu et al., 2011; Liu et al., 2016), and that more than
60 % of all manipulative field experiments only focused on changes in
annual precipitation amounts (Beier et al., 2012). The consequences of
seasonal precipitation redistribution at ecosystem levels are still under
investigation. Field experiments simulating seasonal precipitation changes in
subtropical regions are urgently needed for better understanding of the
ecosystem responses.
Changes in precipitation can strongly affect soil nitrogen (N) cycling and
balance, thus exerting a feedback on climate (Davidson et al., 2008; Wieder
et al., 2011). For instance, annual N2O emission was decreased by a
rainfall exclusion experiment in the moist tropical forest, but recovered
within the first year after rainfall exclusion was stopped (Davidson et
al., 2008). In grasslands, the net N mineralization rate declined sharply in
response to increased rainfall, but increased during drought (Jamieson et
al., 1998). Contrasting responses of N transformation have also been obtained
in temperate forests (Emmett et al., 2004; Chen et al., 2011; Fuchslueger et
al., 2014). However, limited information exists about the responses of the N
cycle to seasonal precipitation changes in subtropical forests, which serve
as important sources of N2O emission and inorganic N leaching (Fang et
al., 2009; Isobe et al., 2012). Seasonal precipitation changes may affect N
transformations by disturbing the seasonal dynamics of microbial activities,
soil moisture, temperature, plant nutrient uptake, and carbon (C) and N
availabilities (Reichmann et al., 2013). Although the direct effects of soil
physicochemical properties and microbial communities on N transformations are
well documented, the dominant factors in determining N transformations under
precipitation changes are still debatable (Petersen et al., 2012; Auyeung et
al., 2015).
Ammonium oxidation, the central and rate-limiting step in N cycling, is
driven by ammonia-oxidizing archaea (AOA) and bacteria
(AOB), which are marked by the amoA functional gene (van
der Heijden et al., 2008). The release and consumption of N2O by
denitrification are mainly driven by nitrite-reducing bacteria carrying the
nirK and nirS genes and nitrous oxide-reducing bacteria
carrying the nosZ gene (Schimel and Bennett, 2004; Levy-Booth et
al., 2014). Thus, changes in these functional microorganisms can shed light
on the underlying mechanisms of N transformation responses. The abundance,
composition and activity of these microbial functional groups largely depend
on soil moisture, temperature, O2 diffusion, and C and N availabilities
– all of these factors are strongly influenced by precipitation (Bell et
al., 2014). For instance, previous research has shown that reduced
precipitation decreases soil moisture and increases aeration and O2
diffusion, which stimulates the activity of nitrifiers
(AOA/AOB) and nitrification (Stark and Firestone, 1995;
Zhalnina et al., 2012). In contrast, reduced precipitation could constrain
the activity of denitrifiers, and consequently reduce the N2O and/or
N2 emissions (Stark and Firestone, 1995; Zhalnina et al., 2012). Both
denitrifiers and nitrifiers might be suppressed by decreased moisture and
available C during drought (Bárta et al., 2010; Zhalnina et al., 2012).
In addition, increased precipitation may raise the
NH4+ : NO3- ratio, as NO3- is easily leached
(Reichmann et al., 2013). High NH4+ : NO3- ratios can
consequently alter the predominant microbial groups (Nautiyal and Dion,
2008). The potential for mixotrophic growth and starvation tolerance of
nitrifying communities (Levy-Booth et al., 2014) suggests a broader
ecological niche occupied by the nitrifying groups. Therefore, the nitrifying
and denitrifying microorganisms may respond differently to seasonal
precipitation changes, leading to non-synchronous changes in nitrification
and denitrification, and consequently different changes in soil NO3-,
NH4+ contents and N2O emission. However, the extent to which
microorganisms control N transformations remains unclear because soil
physicochemical properties can also affect N pools through erosion, leaching,
plant uptake and physiological changes in microorganisms, regardless of
microbial composition or abundance (Cregger et al., 2014; Auyeung et al.,
2015). As a result, the effects of soil physicochemical properties and
microbial communities on N transformation rates are difficult to
differentiate, which makes it difficult to uncover the underlying drivers.
In order to investigate responses of N transformations to seasonal
precipitation changes and the main controlling factors, a precipitation
manipulation experiment was conducted in a subtropical forest in southern
China, where the precipitation is predicted to increase in wet seasons and
decrease in dry seasons (Zhou et al., 2011). We simulated this seasonal
precipitation pattern for 2 years. Changes in soil physicochemical
properties, net N transformation rates, and nitrifying (bacterial and
archaeal amoA) and denitrifying (nirK, nirS and
nosZ) gene abundance were analyzed and integrated in a hypothetical
path model which assumed that the precipitation-induced changes in soil
physicochemical properties and microbial abundance could alter N
transformation rates (Fig. 1). The path coefficients and model fitness were
analyzed by a structure equation model (SEM; Petersen et al., 2012;
Delgado-Baquerizo et al., 2014). We hypothesized the following: (1) decreasing
precipitation in the dry season will reduce N transformation rates by
decreasing SWC, C and N availabilities, as well as microbial abundance, but
(2) precipitation addition during the wet season will have little impact on N
transformation due to the originally sufficient SWC and substrate supply;
(3) the responses of N transformation rates to the precipitation change will
be associated with changes in functional gene abundance, because N
transformation processes are primarily catalyzed by specific enzymes coded by
functional genes; and (4) microbial abundance is directly influenced by soil
physicochemical properties, but denitrifiers will be more strongly affected
than nitrifiers, because the nitrifiers have the potential for mixotrophic
growth and are tolerant of low N and C substrate availabilities.
A conceptual model illustrating the effects of physicochemical
properties and functional microorganisms on N transformation rates. Soil
water content (SWC), ammonium (NH4+), nitrate (NO3-) and
extractable organic carbon (EOC) concentrations were included in the group of
soil physicochemical property. Microbial biomass carbon (MBC) and nitrifying
(amoA) and denitrifying (nirK, nirS and
nosZ) gene abundance were included in the microbial attributes
group. The solid lines with arrows indicate the direction of the effect.
Materials and methods
Site description
The study site is located at the Heshan National Field Research Station of
Forest Ecosystem, Chinese Academy of Sciences (22∘41′ N, 112∘54′ E), Heshan City, Guangdong province, southern China. This
area has a pronounced wet season (April to September) receiving 80 % of
the annual rainfall, and a dry season (October to March) with only 20 %
of the annual rainfall (Wang et al., 2009). The soil is typical laterite (or
Oxisols based on the USDA soil taxonomy), developed from sandstone, and is
easily leached. This study was conducted in a 35-year old evergreen
broadleaved mixed species forest dominated by Schima superba and Michelia macclurei. The vegetation inventory was
conducted in the study forest by recording species name, diameter at breast
height (DBH), tree height and density prior to the experiment. Overall, the
forest consists of about 30 woody species, with average tree height of 8 m,
average DBH of 9.5 cm, stem density of
1430 trees ha-1 and basal area of 11.6 m2 ha-1.
Experimental design
A replicated manipulative experiment of precipitation reduction in the dry season
and precipitation addition in the wet season was employed for 2 years from
October 2012 to September 2014. Eight 12 m × 12 m experimental
plots were randomly assigned to four replicates of each of the two treatment
types: the seasonal precipitation change manipulation (hereafter
“precip-change”) and the trenched control (hereafter “control”). Distance between
the adjacent plots was at least 2 m. Prior to the experiment, the stand
characteristics between the precip-change and control plots were compared,
and no significant differences were detected. Generally, the four
precip-change plots have average tree height and DBH of 10.2 ± 5.0 m
and 10.7 ± 6.3 cm, respectively, with average crown width of
46 ± 11 m2 and total number of 64 tree individuals. The average
tree height, DBH, crown width and total tree number in the four control plots
are 7.7 ± 3.5 m, 9.5 ± 5.2 cm, 49 ± 13 m2 and 68,
respectively. Around the perimeter of each of the eight plots, a 60–80 cm deep
trench was excavated and 1 m height PVC segregation board was imbedded to
reduce the potential for lateral movement of soil water from the surrounding
areas into the plots. The precipitation reduction and addition was realized
by throughfall exclusion and water addition facilities, respectively.
Throughfall exclusion and water addition facilities were established in the 4
precip-change plots, but not in the control. The facilities included
supporting structures, rainout shelters and water addition subsystems
(Fig. S1 in the Supplement). Within each of the four precip-change plots, 16
galvanized steel pipes (2.5–3 m length × 10 cm diameter) were
vertically fixed in concrete bases which were imbedded in soil for 60 cm
depth, and were welded together with 8 horizontal stainless steel frames
(12 m length) at the top. Rainout sheets were fixed in two stainless steel
frames and hung on the supporting system with steel hook rivets. There were
about 8–12 rainout sheets (with a width of 50–100 cm) within each
precip-change plot, depending on the density of tree stems. The rainout sheets
were made from polyethylene plastic with > 90 % light transmission and
installed at approximately 1.5 m height above the soil surface. The total
area of all the rainout sheets was 67 % of the plot area (i.e.,
144 m2). The sheets were opened to exclude throughfall during the dry
season (1 October to 31 March) but folded without throughfall exclusion
during the wet season (1 April to 30 September). Therefore, we reduced about
67 % of the full incoming throughfall in the dry season. The intercepted
rainfall was routed into an iron gutter placed at the lower slope of the
plots, and then drained outside the plot with PVC pipes.
The water added into precip-change plots in the wet season was pumped from a
pond (about 800 m away from the experimental plots) and transported with PVC
pipes to the rubber sacs fixed on the supporting system, and then sprinkled
out via 25 sprinklers distributed evenly in each plot. The pH was similar in
the throughfall (6.42) and pond water (6.19) but the nutrient (e.g., nitrogen
and organic carbon) contents were higher in throughfall than in the pond water
(Zhao et al., 2017), which ensures that we did not enrich nutrients while
adding water. The amount of water added into a precip-change plot during the
wet season was calculated as a product of the above-canopy dry-season
rainfall, the throughfall ratio, and the throughfall exclusion ratio (i.e.,
0.67). The above-canopy rainfall was obtained from a standard meteorological
station (Davis, Vaisala, Finland) about 80 m away from the experimental
site. The throughfall ratio was 0.86, obtained from 8 rain gauges (TB4MM,
Techno Solutions, Beijing, China) installed about 80 cm above soil surface
in the eight plots. As a result, the intensity of the dry season rainfall events
was reduced and the frequency of large rainfall events in wet season was
increased, while the annual total quantity of the throughfall was not
changed. More specifically, the throughfall excluded was 220 mm in the 2013
dry season (1 October 2012 to 31 March 2013) and the same amount of water was
added back into each precip-change plot with four large events
(55 mm day-1) in June through September 2013 (i.e., one event in each
month) to mimic the projected occurrence of more large rainfall events in the wet
season in the region (Zhou et al., 2011). The throughfall exclusion was
170 mm in the 2014 dry season (1 October 2013 to 31 March 2014) and the same
amount of water was added back into each precip-change plot with three large events
(57 mm day-1) in June through August 2014 (Fig. 2a).
Soil sampling and analyses
Soil samples were collected at the beginning and end of January, March, May,
August and October from May 2012 to September 2014 for physicochemical
properties, and from January 2013 to September 2014 for microbial functional
genes analyses. Soil samples were collected from 0 to 10 cm depth with an
auger (Φ35 mm), sieved through a 2 mm mesh to remove litter and
stones. One composite soil sample, consisting of six subsamples randomly
collected within each plot, was used for the physicochemical (stored at
4 ∘C) and microbial (stored at -20 ∘C) analyses. All
samples were analyzed within 2 weeks.
Seasonal dynamics of precipitation and soil physicochemical
properties in control and precipitation change (precip-change) plots over the
course of experiment. Points and bars with standard error (n=4) show mean
values at each sampling time and in dry (DS) and wet (WS) seasons. Grey
shades indicate the periods of precipitation reduction. The significance
levels are presented as *p<0.05.
Soil physicochemical properties were measured using the methods as described
by Liu et al. (1996). Briefly, soil water content (SWC) was obtained by
drying fresh soils in an oven at 105 ∘C for 24 h. Total nitrogen
(TN) and total phosphorus (TP) were determined using the H2SO4
digestion–indophenol blue colorimetry and H2SO4
digestion–Mo–Sb colorimetry methods, respectively. NH4+ and
NO3- contents were determined from the 2 M KCl extraction liquid by
using the indophenol blue colorimetry and copperized cadmium reduction
methods, respectively.
Soil extractable organic carbon (EOC) and microbial biomass carbon (MBC) were
measured immediately after the soil sampling using the fumigation extraction
method described in Vance et al. (1987). In detail, a pair of
fresh soil subsamples (10 g) was placed into two glass breakers. One was
fumigated in a vacuum dryer with alcohol-free chloroform and NaOH solution
for 24 h in the dark, and the other one was placed in the dark for 24 h without
fumigation. The two subsamples were extracted with 0.5 M K2SO4
after fumigation, and the EOC concentration was determined using a total
organic C analysis instrument (TOC-VCSH, Shimadzu, Japan). The difference in
EOC concentration between the fumigated and un-fumigated was multiplied by
0.45 to calculate MBC content.
Soil total DNA was extracted from 0.3 g fresh soil using the HiPure Soil DNA
Mini Kit (Magen, Guangzhou, China), quantified with a NanoDrop 2000
spectrophotometer (Thermo Fisher Scientific Inc., USA) and stored at
-20 ∘C for further analyses. The abundance of bacterial and
archaeal ammonia monooxygenase gene (amoA), nitrite reductase genes
(nirK and nirS) and nitrous oxide reductase gene
(nosZ) were quantified by using absolute real-time polymerase chain reaction (PCR) on an ABI
7500 thermocycler system with primers and thermal profiles presented in the
Supplement (Table S1). The real-time PCR reactions were performed on 96-well
plates (Axygen, USA), with 20 µL volume in each well including
12.5 µL SYBR Premix Ex Taq (TaKaRa Biotechnology, Japan),
1 µL of each primer (10 mmol L-1), 2 µL of DNA
template (10 ng), 1 µL dimethyl sulfoxide and 4.5 µL
RNase free Ultra-Pure water. The standards were constructed using the method
described in Henry et al. (2006) and Isobe et al. (2012). Briefly, the target
functional gene PCR products were obtained with the same primers used in
real-time PCR and the extracted soil DNA as template. The PCR products were
cloned using the pMD20-T vector (TaKaRa, Dalian Division), and then
transformed into Escherichia coli JM109 strains. The recombinant
Escherichia coli JM109 strains carrying the target functional gene
recombinant plasmids were inoculated into LB broth with ampicillin and
incubated at 37 ∘C overnight. The plasmid DNA was then extracted
using the HiPure Plasmid Mini Kit (Magen, Guangzhou, China) and quantified on
a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific Inc., USA). The
DNA copy numbers of the extracted plasmid DNA carrying the target functional
gene was calculated from the plasmid DNA size, concentration and average
base pair molecular weight, which could stand for the copy numbers of the
standard functional gene. Finally, the standard curve was generated from a
tenfold serial dilution (103–108 copies per µL) of the
plasmid DNA.
Measurement of N transformation rates
Net N mineralization and nitrification rates were measured through the in
situ soil incubation using the resin-core method (Reichmann et al.,
2013). Six paired soil cores (0–10 cm) were randomly sampled within each
plot at the beginning of January, March, May, August and October from
May 2012 to September 2014. One core of each pair was sieved through a 2 mm
sieve after removing litter and stones, and stored at 4 ∘C for the
initial pre-incubation measurements of SWC, NO3- and NH4+.
The other core was incubated for 1 month in a PVC pipe that was open on
both sides and was oriented vertically with an ion exchange resin bag placed
at the bottom to collect inorganic N leached from the core. Soil cores and
resin bags in the PVC pipes were collected after the 1 month incubation,
and the soil was sieved and stored at 4 ∘C for the final
post-incubation measurements of SWC, NO3- and NH4+. The net N
mineralization rate was calculated as the final NO3- and NH4+
content minus the initial NO3- and NH4+ content, and the net
nitrification rate was calculated as the final NO3- content minus the
initial NO3- content (Reichmann et al., 2013). Concentrations of
NO3- and NH4+ extracted from the resin were considered as the
leaching rates of NO3- and NH4+ per month.
Soil nitrous oxide (N2O) efflux was measured twice per month, from
October 2012 to September 2014, using static chamber and gas chromatography
techniques. The static chambers were made from white PVC materials and
consisted of a removable cover box (26 cm diameter × 35 cm height)
and a base (33 cm diameter × 11 cm height). The bottom of the base
was inserted into soil at a depth of 5 cm. Two months before gas sampling, four
static chambers were deployed randomly at each plot to minimize effects of
installation disturbance. The N2O samples were collected between 09:00
and 11:00 LT. The static chamber was closed for 30 min, and
gas samples (80 mL) were taken using 100 mL plastic syringes at the initial
closed time as well as every 10 min thereafter during the closed period. At
the same time, values of atmospheric pressures and air temperatures inside
static chambers were measured 3 times. N2O concentrations were
analyzed in the laboratory with a gas chromatograph (Agilent 7890A, Agilent
Technologies, USA) equipped with an electron capture detector set at
300 ∘C and a stainless porapak-Q column set at 70 ∘C within
24 h after gas sampling. The N2O flux was calculated by changes
in the N2O concentrations inside static chamber during periods of gas
sampling, with the equation as follows:
F=ρ×VA×PP0×T0T×dCdt,
where F stands for the flux of N2O (mg m-2 h-1), ρ
stands for the density of N2O under standard condition (g L-1),
V stands for the effective volume of chamber (m3), A stands for the
area of soil covered by chamber (m2), P and T stand for the
atmospheric pressures (Pa) and absolute air temperature inside chamber (K)
when gas sampling, P0 and T0 stand for the atmospheric pressures
(Pa) and the absolute temperature (K) under standard condition, and
dC/dt stands for changes of N2O concentrations in the
chamber during gas sampling.
Statistical analyses
Two-way repeated-measure analysis of variance (ANOVA) with sampling time as
the repeated factor was used to examine the effects of precip-change and
sampling time on all measured parameters. Pillai's trace from multivariate
testing was used for within-subject testing when the assumption of multisample
sphericity was not met. Independent-sample t tests were used to detect the
difference of each variable between precip-change and control at each
sampling time. All the parameters were explored for normality
(Kolmogorov–Smirnov test) and homogeneity of variances (Levène test)
prior to the analyses, and log-transformed if necessary. All statistical
analyses described above were performed using SPSS v.16.0 (SPSS Inc.,
Chicago, IL, USA).
Structural equation modeling (SEM) is often used to detect complex
relationships between one or more dependent or independent variables by using
a series of statistical methods. The complex relationships among the target
variables are expressed as paths in a hypothetical model, and finally tested
by a series of statistical methods, such as univariate and multivariate
regressions, ANOVA and factor analysis (Bagozzi and Yi, 2012). In this study,
we used SEM to test the hypothetical causal relationships among soil
physicochemical properties, microbial abundance and N transformation rates in
the conceptual model (Fig. 1), and the SEM was performed with AMOS 21.0 (SPSS
Inc., Chicago, IL, USA). How the effects of soil physicochemical properties
and microbial abundance determine the responses of N transformation rates
were evaluated. In order to explicitly illustrate the pathways of soil
physicochemical properties and microbial abundance involved in each N
transformation process, three individual models were constructed
corresponding to the conceptual model to explain the responses of (a) net
nitrification, (b) net N mineralization and (c) N2O emission rates. The
hypothetical relationships among variables in the models are constructed
based on the results of correlation analyses (Fig. S2). We used three models
since it would be easier to discover the controlling factors than using one
complex model that implicates all the measured processes (Delgado-Baquerizo
et al., 2014). In these models, the precip-change treatments are categorical
exogenous variables with two levels: 0 representing control and 1
representing seasonal precipitation changes (Delgado-Baquerizo et al., 2014).
Abundance of both nirK and nirS genes were shown to be
correlated with nitrification or N mineralization rates (Levy-Booth et al.,
2014). Therefore, nirK and nirS abundance were added as one
(nirK + nirS) endogenous factor in the model. Net
nitrification rate was included in model (b) as an endogenous factor because
it may influence N2O emission through altering the production of
NO3- as the substrate for N2O production. Prior to the SEM
analyses, normal distribution of all the involved variables were examined,
and gene abundances were log-transformed. Goodness of model fits was
evaluated by chi-square test (p > 0.05), comparative fit index
(CFI > 0.95), and root mean square errors of approximation
(RMSEA < 0.05) (Hu and Bentler, 1998; Schermelleh-Engel et al., 2003).
Pathways without significant effects were not shown (p > 0.05) in the
final models.
Nitrogen transformation rates measured in control and precipitation
change (precip-change) plots over the course of experiment. Points and bars
with standard error (n=4) show mean values at each sampling time and in
dry (DS) and wet (WS) seasons. Grey shades indicate the periods of
precipitation reduction. The significance levels are presented as *p<0.05.
Results
Responses of soil physicochemical properties, N transformation
rates and microbial abundance to precipitation changes
Before the precipitation manipulation from May to September in 2012, average
net N transformation (i.e., N nitrification, mineralization and leaching)
rates, N (NO3-, NH4+, TN) and organic C (MBC, EOC, TOC)
contents as well as soil temperature were similar among all plots (Table S2).
In the two dry seasons with precipitation reduction, SWC decreased by
16 % in 2013 and by 21 % in 2014 (p < 0.01; Table S3 and
Fig. 2d). Similarly, NO3- concentration decreased by 35 and 24 %
in 2013 and 2014, respectively (p < 0.01; Table S3 and Fig. 2j).
Opposite patterns were observed for NH4+ concentration, which
increased with the precipitation reduction (Fig. 2l). In the wet seasons with
precipitation addition, SWC, NO3- concentration, EOC and MBC remained
lower in the precip-change plots than in the control plots in both years
(Table S3 and Fig. 2d, f, h and j). After the experiment, soil pH in the
precip-change plots was 3.82 ± 0.02 in dry seasons and
3.78 ± 0.07 in wet seasons. In the control plots, it was
4.06 ± 0.05 in dry and 3.86 ± 0.1 in wet seasons. It has no
significant changes when compared with the pH values before the experiment, with
4.01 ± 0.04 and 4.05 ± 0.08 in dry and wet seasons of the
precip-change plots, and 4.23 ± 0.01 and 4.11 ± 0.07 in dry and
wet seasons of the control plots.
Copy numbers of archaeal amoA, nirK, nirS
and nosZ gene per gram of dry soil measured in control and
precipitation change (precip-change) plots over the course of experiment.
Points and bars with standard error (n=4) show mean values at each
sampling time and in dry (DS) and wet (WS) seasons. Grey shades indicate the
periods of precipitation reduction. The significance levels are presented as
*p<0.05.
Precipitation reduction strongly decreased the average dry-season net
nitrification rate by 13 % in 2013 and by 20 % in 2014, and decreased
net N mineralization rate by 16 % in 2013 and by 18 % in 2014
(p < 0.1; Table S4 and Fig. 3b and d). The NO3- leaching also
declined with precipitation reduction, especially in 2014, with a marked
decrease by 22 % (p < 0.001; Table S4 and Fig. 3e and f).
Contrastingly, the rates of three N transformation processes increased by
50 % with precipitation addition in the 2013 wet season, whereas they changed
little in the 2014 wet season (Fig. 3b, d and f). Throughout the 2 years,
moderate decreases were detected in N2O emission either during
dry-season precipitation reduction (35 %) or during wet-season
precipitation addition (15 %) (Table S4 and Fig. 3j).
No amplification of the bacterial amoA gene was detected in soil, either
from the precip-change plots or from the control plots, which was mainly
because soil AOB community abundance in the studied forest was under
the detection limitation caused by low soil pH (4.08 ± 0.05) (Isobe et
al., 2012). The average seasonal archaeal amoA gene was
6.5 × 106 ± 1.9 × 106 copies g-1 dry
soil, and varied significantly according seasonal precipitation changes. With
precipitation reduction, the archaeal amoA gene abundance changed
little in the 2013 dry season but decreased by 70 % in the 2014 dry
season (Fig. 4a and b). The abundance of three denitrifying genes
(nirK, nirS and nosZ) increased with precipitation
reduction by 30–80 % in the 2013 dry season (p < 0.05; Table S5
and Fig. 4d, f and h). In both seasons of 2014, neither dry-season
precipitation reduction nor wet-season precipitation addition had significant
impacts on the abundance of the three denitrifying genes (Table S5 and
Fig. 4c, d, e, f, g and h).
Paths determining N transformation rates and functional microbial
abundance
Although the annual precipitation amount was kept constant, the
redistribution of seasonal precipitation imposed an overall negative impact
on SWC and NO3- concentration (Fig. 5). SWC affected net
nitrification and N mineralization through a direct negative path and
N2O emission through a direct positive path (Fig. 5). Net N
mineralization, nitrification and N2O emission rates were also affected
by the functional-gene abundance and MBC paths. Since the bacterial
amoA gene was not detected, we only use the archaeal amoA
abundance as the dominant nitrifying microbial abundance in the SEM analyses.
Specifically, the archaeal amoA gene abundance and MBC had direct
positive impacts on net N mineralization and nitrification rates, whereas the
nosZ gene abundance had a direct negative impact on N2O
emission (Fig. 5). As a result, 21 and 22 % of the net N mineralization
and nitrification variability are explained, respectively (see the r2 in
Fig. 5a and b). Among the direct influential factors, archaeal amoA
abundance showed the strongest correlations, with both net N mineralization
or with net nitrification rates. Soil N2O emission was mostly affected
by positive effects of net nitrification rate and SWC, followed by negative
effects of nosZ abundance and MBC, and as much as 42 % of the
total variation could be explained (see the r2 in Fig. 5c).
Path diagrams demonstrating the effects of soil physicochemical
properties and functional-gene abundance on net nitrification, N
mineralization and N2O efflux rates in response to precipitation change
(precip-change) over 2 years. Numbers adjacent to arrows are path
coefficients, which indicate the relationships between the two variables on
both sides of the arrows. Solid and dashed lines represent positive and
negative paths, respectively. The r2 above or below each response
variable in the model denotes the proportion of variance which could be
explained. The size of the lines indicates significant levels of path
coefficients.
Precip-change-induced changes in SWC had no direct impacts on functional-gene abundance. Instead, the functional-gene abundance was indirectly
affected by the precip-change-induced alterations in NO3-,
NH4+ concentrations and EOC (Fig. 5). Specifically, NO3- and
NH4+ had direct positive effects on archaeal amoA abundance,
whereas EOC had a direct negative effect on nirK + nirS
abundance. Both NH4+ and EOC concentration had direct positive
impacts on the nosZ abundance (Fig. 5c). Changes in MBC were
directly positively influenced by SWC and EOC.
Discussion
Drivers of N transformation processes
Consistent with our hypotheses, seasonal precipitation redistribution induced
significant changes in net N mineralization and nitrification rates by
altering SWC, MBC and archaeal amoA gene abundance. N2O
emission was decreased by both precipitation enhancement (wet season) and
precipitation reduction (dry season), which indicated that soil N loss by
N2O emission in subtropical forests would be alleviated by the predicted
seasonal precipitation changes. In contrast, increased NO3- leaching
during precipitation addition in the wet seasons led to significant losses
from the soil NO3- pool. During the 2 year experiment, SWC was
always lower in precip-change plots than in control plots, despite the
precipitation addition in the wet seasons (Fig. 2c and d). One reason for this is the
higher transpiration loss resulting from relatively bigger trees in the
precip-change plots (tree height: 10.2 ± 5.0 m, DBH:
10.7 ± 6.3 cm) than that in the control plots (tree height:
7.7 ± 3.5 m, DBH: 9.5 ± 5.2 cm). There were no significant
differences in these stand characteristics, but the bigger trees in
precip-change plots might have had greater transpiration rates and therefore
caused more soil water loss in the summer wet season (Gao et al., 2017).
Another reason might be the large amount of precipitation added (55 mm per
event). Large precipitation events may result in flood-irrigation that can
break the soil pores or reduce pore number, leading to soil structural
decline (Barber et al., 2001). These changes in soil structure may affect
soil water content, as soil water retention capacity is related to pore size
and pore distribution (Loll and Moldrup, 2000).
Initially, we hypothesized that decreased precipitation in the dry season
would suppress N transformation, and precipitation addition during the wet
season would have little impact on N transformation processes because the
soils are water-saturated and substrate sufficient. In agreement with the
first hypothesis, net nitrification and N mineralization rates decreased
sharply with the reduction of throughfall in the dry season (Fig. 3a, b, c
and d). However, contrary to the second hypothesis, nitrification and N
mineralization rates increased markedly with precipitation supplementation in
the wet seasons (Fig. 3a, b, c and d). These results can be explained by the
interactions between microbial abundance, soil moisture and substrate
availability (Figs. 5a, b and S3). Specifically, soil EOC of the dry season
was less in the precip-change plots than in the control plots (Fig. 2e and
f), which is probably attributable to reduced C input due to lower root production and
exudation after drying (Kuzyakov and Domanski, 2000; Borken and Matzner,
2009). The reduced supply of soil C substrate (i.e., EOC) could have
restricted the growth of soil microorganisms (e.g., MBC and AOA),
resulting in decreased net nitrification and mineralization rates (Fig. 5a
and b). Although increased NH4+ concentrations with reduced
precipitation could provide more N substrate for nitrifiers, the negative
effects of decreased SWC and EOC may have outweighed the positive effects of
increased NH4+. Instead, the accumulated NH4+ after dry
season precipitation reduction might have had a positive legacy effect on
soil microbial activity in the wet season, leading to increased N
transformations. In addition, SWC differences are also known to directly
affect N transformations by stimulating physiological changes in microbial
activity, regardless of microbial abundance and composition (Auyeung et al.,
2015). The increased N transformation rates (Fig. 3b, d) in response to
decreased SWC, MBC (Fig. 2d, h) and archaeal amoA gene abundance
(Fig. 4a) with precipitation addition might be such a case (also see
Fig. S2). A 10 % lower SWC in the precip-change plots in natural humid
wet season might create better redox conditions for microbial nitrification,
as excessive soil moisture could reduce soil oxygen concentration. According
to Borken and Matzner (2009), the increases of soil microbial activity by
rewetting usually occurred due to an increased pulse in organic substrate
availability as well as reconstituting mineralization of SOM. Substantial
decreases in MBC and archaeal amoA gene abundance in our study
indicated that some microorganisms may die from starvation or competition
caused by limited substrate concentrations, and consequently release MBC and
microbial biomass nitrogen (MBN). These available substrates released by dead
microorganisms could be reused by the surviving microorganisms, which could
support the increased energy demand of accelerated microbial processes
(Borken and Matzner, 2009).
We also hypothesized that N transformation processes are associated with
functional microbial abundance. As expected, net N mineralization and
nitrification rates showed stronger relationships with archaeal amoA
abundance than with MBC or other soil properties (Fig. 5a and b). However,
MBC and denitrifying gene abundance had similar effects on N2O emission.
Our results also showed that only nosZ gene abundance exerted a
pronounced effect on N2O emission (Fig. 5c), probably by reducing
N2O consumption (Henderson et al., 2010; Levy-Booth et al., 2014). No
significant correlation between N2O emission and
nirK + nirS gene abundance was detected, in contrast to
previous studies (Levy-Booth et al., 2014; Gao et al., 2016). The N2O-emission-related denitrification can also be performed by nitrifiers and
fungi in soils with high aeration and limited substrate availability
(Levy-Booth et al., 2014). The experimental seasonal precipitation strongly
decreased SWC and EOC content (Fig. 1), leading to higher aeration while
lowering substrate availability. These changes in soil physicochemical
properties could enhance the role of nitrifier and fungi denitrification in
controlling N2O emission. In addition, SWC and nitrification rate also
directly affected N2O emission by altering substrate availability and
consequently microbial activity, despite high microbial abundance (Fig. 5c).
Although functional microbial abundance showed the most significant
correlations with N transformation rates and could explain more than 20 %
of their variation, a large proportion of the variation remained unexplained
(Fig. 5). This unexplained variation is mainly attributed to the changes in
other functional microbial genes involved in the nitrogen cycle, such as
narG and napA responsible for NO3- reduction, and
nifH responsible for N fixation (Widmer et al., 1999; Tavares et
al., 2006). Moreover, gene abundance based on DNA may not fully reflect gene
expression.
Determinants of nitrifying and denitrifying gene abundance
The responses of both nitrifying and denitrifying genes were mainly related
to the changes in substrate concentrations. SEM analysis showed that both
amoA and nosZ gene abundance was positively affected by EOC
and NH4+ concentration, suggesting substrate constraints for these
two functional microbial groups. This disagreed with previous studies that
reported that the AOA community had greater potential for
mixotrophic growth and better low-substrate tolerance than its counterpart
AOB (Erguder et al., 2009; Shen et al., 2012). However, these
previous results were mainly due to greater competitiveness of AOA
than AOB, as these studies mainly focused on the comparison of
effects of substrate availability on AOA and AOB
communities. Both nosZ and amoA gene abundance increased
with EOC and NH4+ concentration (Fig. 5), which indicated that the
AOA community could be constrained by C and N substrates when
competing with other microbes that have different functions. Otherwise, the
existing AOA species that have the potential for mixotrophic growth
and starvation tolerance would not dominate in the subtropical
forest studied, as the soil is rich in SOM (Zhou et al., 2006; Chen et al., 2015).
Therefore, the AOA community in the studied soil could be strongly
influenced by changes in soil C and N availability.
The abundance of nirK and nirS genes was positively
controlled by soil NH4+ concentration and negatively controlled by
EOC content (Fig. 5). This confirmed that higher NH4+ content could
favor more abundant microorganisms containing nirK or nirS
genes (Yi et al., 2015), because higher NH4+ concentration could
supply sufficient NO3- as the direct substrate or provide optimum pH
values for growth of the denitrifying microorganisms. The negative effect of
EOC on nirK and nirS gene abundance was inconsistent with
previous reports that denitrifiers are primarily heterotrophic (Bárta et
al., 2010). One reason is that high EOC concentrations can constrain the
growth of microorganisms carrying nirK and nirS genes
through effects on other factors, such as pH and C : N ratio (Henderson et
al., 2010; Levy-Booth et al., 2014). Generally, the abundance of both
nitrifying and denitrifying genes changed with precipitation redistribution,
and the direction and magnitude of the changes depended mainly on soil N and
C substrate availabilities.