Enhanced nitrogen and phosphorus removal by natural pyrite–based constructed wetland with intermittent aeration

Four subsurface flow constructed wetlands (SFCWs) filled with different substrates including ceramsite, ceramsite+pyrite, ceramsite+ferrous sulfide, and ceramsite+pyrite+ferrous sulfide (labeled as SFCW-S1, SFCW-S2, SFCW-S3, and SFCW-S4) were constructed, and the removal of nitrogen and phosphorus by these SFCWs coupled with intermittent aeration in the front section was discussed. The key findings from different substrate analyses, including nitrification and denitrification rate, enzyme activity, microbial community structure, and the X-ray diffraction, revealed the nitrogen and phosphorus removal mechanism. The results showed that the nitrogen and phosphorus removal efficiency for SFCW-S1 always remained the lowest, and the phosphorus removal efficiency for SFCW-S4 was recorded as the highest one. However, after controlling the dissolved oxygen by intermittent aeration in the front section of SFCWs, the nitrogen and phosphorus removal efficiencies of SFCWs-S2 and S4 became higher than those of SFCW-S1, and SFCW-S3. It was noticed that the pollutants were removed mainly in the front section of the SFCWs. Both precipitation and adsorption on the substrate were the main mechanisms for phosphorus removal. A minute difference of nitrification rate and ammonia monooxygenase activity was observed in the SFCWs’ aeration zone. The denitrification rates, nitrate reductase, nitrite reductase, and electron transport system activity for SFCW-S2 and SFCW-S4 were higher than those detected for SFCW-S1 and SFCW-S3 in the non-aerated zone. Proteobacteria was the largest phyla found in the SFCWs. Moreover, Thiobacillus occupied a large proportion found in SFCW-S2, and SFCW-S4, and it played a crucial role in pyrite-driven autotrophic denitrification.


Introduction
Urban inland rivers in China are mostly seasonal rivers, the water quality of these rivers is often affected by the sewage treatment plants' effluent intrusion and rainstorm runoff mixing. As a result, these seasonal rivers gradually form black and smelly water bodies, called low pollution water, which is a great threat to ecological environment sustainability and human water safety concerns. In recent years, the purification of low pollution water polluted by the sewage treatment plants effluent intrusion and storm runoff mixing has attracted extensive attention from researchers. Constructed wetlands (CWs) are widely utilized for advanced wastewater treatment purposes due to their relatively low cost, low energy demand feasibility, ease of operation, and beautification effect (Abou-Kandil et al. 2021;Baldovi et al. 2021;Rizzo et al. 2020). Although CWs have many advantages, natural reoxygenation and oxygen provided by plant roots can not provide sufficient dissolved oxygen (DO) for CWs; therefore, the total nitrogen (TN) removal efficiency is not high when ammonia nitrogen (NH 4 + -N) is the main form of TN. Aeration in CWs can accelerate the transfer and diffusion of DO in water, enhance the activity of biofilm, and improve the degradation capacity of organic matter and nitrogen removal efficiency. On the other hand, aeration in CWs can significantly reduce the occupying area, slow down the blockage of wetland substrate, reduce the maintenance frequency, and extend the service life of CWs (Lai et al. 2020;Stefanakis 2020). Strong correlation (Pearson's coeffi-cient=0.96) between the areal oxygen consumption rate and organic loading rate was observed in the literature, and the systems with high organic load will "consume" more of the readily available oxygen than the low loaded systems (Rous et al. 2019).
Heterotrophic denitrification is the main denitrification mechanism in traditional CWs, and carbon source is considered the main limiting factor for the denitrification process Si et al. 2018;Tan et al. 2020;Wang et al. 2021). Theoretically, the estimated C/N ratio for heterotrophic denitrification is 3.74 (Anjali and Sabumon 2015;Chiu and Chung 2003). In the case of application of the CWs for low pollution water purification, the C/N ratio for most low pollution water is lower than the value reported above, which makes it challenging to meet the carbon source requirements of heterotrophic denitrification process. Thus, this results in limited biological denitrification processes, as well as poor denitrification efficiency. To boost the denitrification process, carbon sources can be added to improve denitrification efficiency. Generally, it is suggested that the higher carbon content availability in the influent is conducive to converting nitrate or nitrite into nitrogen, which produces less N 2 O. However, excessive organic carbon availability in the influent will lead to the increase of unused carbon sources in the denitrification process, which become a cause of secondary pollution spreading, reactor blockage, and high effluent turbidity.
Moreover, in the case of autotrophic denitrification, denitrifying bacteria reduce NO 3 − -N to N 2 by using H 2 , sulfide, sulfur, and reductive iron as electron donors, and NO 3 − as electron acceptor. The compounds including CO 2 , HCO 3 − , and CO 3 2− are used as inorganic carbon sources. The sludge yield in the autotrophic denitrification process remains comparatively low, and no additional organic carbon source is needed to accomplish the task, which can effectively remove NO 3 − -N from low C/N sewage. Previous studies have shown that elemental sulfur's autotrophic denitrification process will lead to a drop in solution pH and produce a large amount of sulfate Sahinkaya et al. 2011;Wang et al. 2020). Therefore, pyrite is considered a good choice for autotrophic denitrification, low cost, stable pH, and less sulfate formation. The production of sulfate in the process of FeS autotrophic denitrification also remains much lower, when compared with elemental sulfur autotrophic denitrification.
In the present study, subsurface flow constructed wetlands (SFCWs) with different substrates ceramsite (SFCW-S1), ceramsite+pyrite (SFCW-S2), ceramsite+ferrous sulfide (SFCW-S3), and ceramsite+pyrite+ferrous sulfide (SFCW-S4) were used for the purification of low pollution water, i.e., seasonal river water contaminated by the sewage treatment plant effluent and storm runoff. The main research objectives for this study were as follows: (1) comparing the purification efficiency of CWs filled with different substrates with and without intermittent aeration; (2) exploring the nitrogen and phosphorus removal effects of CWs filled with different substrates, as well as the nitrogen removal effects of ferrous sulfide and pyrite as electron donors; (3) analyzing the migration and transformation of nitrogen elements, microbial activity, and community structure in CWs, as well as the nitrogen and phosphorus removal mechanism.

Experimental apparatus and methods
Total four SFCWs labeled as SFCW-S1, SFCW-S2, SFCW-S3, and SFCW-S4 were constructed having same dimensions (length×width×height=80 cm×30 cm×60 cm). The experimental apparatus diagram has been presented in Fig. 1. Moreover, the filling structure of substrates has been given in Table 1. The grain size of the substrate is 15-25 mm for gravel, 0.5-1 mm for pyrite, 20-80 mm for ferrous sulfide, and 2-6 mm for ceramsite. The chemical composition of substrates used in the four SFCWs is shown in Table S1; the data suggested that the substrates in SFCW-S1 were mainly composed of quartz and hercynite, the substrates in SFCW-S2 were mainly composed of clinochlore and quartz, the substrates in SFCW-S3 were mainly composed of ferrous sulfide and magnesioferrite aluminian, and the substrates in SFCW-S4 were mainly composed of quartz and ferrous sulfide. The porosity for SFCW-S1, SFCW-S2, SFCW-S3, and SFCW-S4 was 36%, 36%, 31%, and 29%, respectively. Three cannas plants were planted at equal intervals in total of four SFCWs. A certain amount of compounds CH 3 COONa, NH 4 Cl, KNO 3 , and KH 2 PO 4 was dissolved in the tap water to prepare the synthetic wastewater. The influent water quality of the four SFCWs has been given in Table S2. The water temperature of the water samples during the experiment was 15.6°C-28.2°C (Fig. S1).
The operation mode of continuous inflow and continuous outflow was adopted during the operation of the four SFCWs. The experiment was carried out from September 17, 2020, to January 8, 2021. From September 17, 2020, to October 16, 2020, it was a non-aeration stage, and the air pump was not turned on. The aeration time of the four SFCWs was from October 17, 2020, to January 8, 2021. The aeration pipes were located in the first chamber of the four SFCWs. The aeration mode was as follows: continuous aeration for 2.5 h, nonaeration for 0.5 h, then continuous aeration for 2.5 h, nonaeration for 0.5 h, the total aeration time was 20 h per day, 8 cycles per day. The aeration rate was 1.5 L/min. The inflow of each SFCW was 25.5 L/d; the hydraulic retention time was 2 d.
Water sampling points (P1, P2, P3, P4, P5) and substrate sampling points (A, B, A1, A2, B1, B2) are shown in Fig. 1. Sampling frequency of the water sampling points P1 and P5 for water quality determination was 1-2 times a week. On the 55th day of the experiment, samples were taken at P2, P3, and P4 along the length of the SFCWs, and the water quality indexes were determined. Water samples at P2, P3, and P4 were taken from the sampling pipes embedded in each chamber of the SFCWs by pipette (Fig. 1). Water sampling points P2, P3, and P4 were 15 cm, 42 cm, and 65 cm away from the influent along the length of the SFCWs.
Substrate sampling was conducted on the 74th, 78th, and 88th days of the experiment at points A, B, A1, A2, B1, and B2 ( Fig. 1). Substrate samples were taken from the aeration  Fig. 1 Experimental apparatus diagram of subsurface flow constructed wetlands zone (sampling point A) and non-aeration zone (sampling point B) in SFCW-S1, SFCW-S2, and SFCW-S3. In SFCW-S4, two samples were taken from the aeration zone (sampling points A1 and A2), which were mixed as one sample A; two samples were taken from the non-aeration zone (sampling points B1 and B2), which were mixed as one sample B. For the eight substrate samples, 10 g substrate was added to a 250 mL beaker for each substrate sample, then 75 mL ultrapure water was added to the beaker, the beaker was placed in the ultrasonic instrument for 2 h of ultrasonic treatment. The sample after ultrasonic treatment was stirred evenly to obtain the biofilm mixture. The obtained biofilm mixture was used for the analysis of nitrification rate, denitrification rate, microbial enzyme activity, electronic transmission system activity (ETSA), and microbial community structure. The biofilm mixture used for microbial community structure analysis was prepared from the substrate samples obtained on the 74th day. The biofilm mixture used for nitrification rate and denitrification rate analysis was prepared from the substrate samples obtained on the 78th day. The substrate and biofilm mixture used for microbial enzyme activity and ETSA analysis was prepared from the substrate samples obtained on the 88th day.

Analysis methods
The conventional water quality index was analyzed by adopting standard methods (Table S3). X-ray diffraction (XRD) characterization of substrates was performed by the X-ray diffractometer (X'Pert Pro MPD, PANalytical, Netherlands).

Denitrification rate determination
A mixed solution containing sodium nitrate, potassium dihydrogen phosphate, and glucose was prepared; the initial concentrations of sodium nitrate, potassium dihydrogen phosphate, and glucose were 46 mg/L, 50 mg/L, and 1000 mg/L respectively. The obtained biofilm mixture (50 mL) was added into a conical flask containing 300 mL of the mixed solution containing sodium nitrate, potassium dihydrogen phosphate, and glucose solution, and then nitrification inhibitor (1-allyl-2-thiourea) (1 mg/L) was added into the conical flask. Each conical flask was rinsed with nitrogen for 10 min, then sealed and placed on the oscillator. The temperature, oscillation frequency, and oscillation time of the oscillator were set at 25°C, 120 rpm, and 12 h. The NO 3 − -N concentration before and after the reaction was measured. Biofilm denitrification rate (DNR) of the SFCWs substrates was calculated by the decrease in NO 3 − -N concentration over time and divided by the initial biomass (Eq. (1)) (Dong and Reddy 2012;Zhang et al. 2017): where A is the amount of removed nitrate during the reaction (mg), T is the reaction time (h), and B is the amount of biomass used during the reaction (g).

Nitrification rate determination
The biofilm nitrification rate of the CWs substrates has represented by ammonium oxidation rate (AOR). A mixed solution (300 mL) was prepared with ammonium chloride and potassium dihydrogen phosphate. The initial concentrations of ammonium chloride and potassium dihydrogen phosphate were 44 mg/L and 50 mg/L, respectively. The mixed solution (300 mL) and the obtained biofilm mixture (50 mL) were added into a 500-mL conical flask. Aeration was conducted in the conical flask to keep the DO concentration at 5.5 mg/L. Samples were taken at different times and the concentration of NH 4 + -N was determined after filtration (pore size of the membrane 0.45 μm). AOR was calculated by the decrease in NH 4 + -N concentration over time and divided by the initial biomass (Eq. (2)) (Zhang et al. 2017): where C is the amount of NH 4 + -N removed during the reaction (mg), T is the reaction time (h), and B is the amount of biomass used during the reaction (g).
As shown in Eqs.
(1) and (2), B is the amount of biomass used during the experiment for biofilm denitrification rate and nitrification rate determination. Ten grams of substrate was added to a 250-mL beaker, then 75 mL ultrapure water was added to the beaker, the beaker was placed in the ultrasonic instrument for 2 h of ultrasonic treatment. The sample after ultrasonic treatment was stirred evenly to obtain the biofilm mixture. The obtained biofilm mixture with a certain volume was put into a crucible and dried in an oven at 105°C. After cooling, it was weighed and recorded as W1. Then, it was put into a muffle furnace and calcined at 605°C. After cooling, it was weighed and recorded as W2. The difference between W1 and W2 was counted as the amount of biomass B.
The determination methods of ammonia monooxygenase (AMO), nitrite reductase (NIR), and nitrate reductase (NAR) activities have been described in the previous study , which were shown as follows:

Determination of AMO activity
The wetland substrate sample (10 g) was added into a 150-mL conical flask and mixed with 20 mL solution (NaCl 8.0 g/L, KCl 0.2 g/L, Na 2 HPO 4 0.2 g/L, NaH 2 PO 4 0.2 g/L, (NH 4 ) 2 SO 4 1 mM, KClO 3 10 mM). The conical flask was placed in a water bath oscillator, the water temperature was set at 37°C, the oscillation frequency was 150 rpm, and the oscillation time was 4 h. Then, 10 mL KCl solution (2 mol/L) was added to the conical flask, which was oscillated for 5 min at an oscillation frequency of 200 rpm. After centrifugation at 4000 rpm for 5 min, the NO 2 − -N concentration of the supernatant was measured by spectrophotometry. The activity of AMO was expressed as μg NO 2 − -N/(kg gravel·h).

Determination of NIR and NAR activities
The obtained biofilm mixture (40 mL) was centrifuged at 4000 rpm for 10 min. Then, it was washed three times with phosphate buffer solution (PBS) (30 mL, 100 mM) and resuspended in PBS (30 mL, 100 mM). The mixture was treated with an ultrasonic cell breaker (4°C, 200 W, 5 min), then centrifuged for 20 min (4°C, 8000 rpm). Supernatant after centrifugation (1 mL), electron donor mixture (3 mL, 0.01 M PBS (pH 7.4), 0.01 M methyl viologen and 0.005 M sodium hydrosulfite), and electron acceptor (1 mL, 1 mM NaNO 3 or NaNO 2 ) were added into a 10-mL anaerobic culture bottle. Incubation was carried out at 30°C for 30 min, and then the produced or remained NO 2 − -N was determined. The activities of NAR and NIR were expressed as μg NO 2 − -N/(kg gravel·h).

ETSA determination
The ETSA of the bacteria was determined by reducing 2-(piodophenyl)-3-(p-nitrophenyl)-5-phenyl tetrazolium chloride (INT, a kind of exogenous electron acceptor) to formazan (INF). The obtained biofilm mixture (30 mL) was centrifuged at 4000 rpm for 10 min, rinsed twice with PBS (30 mL, 50 mM), and then resuspended in PBS (30 mL, 50 mM). Then, INT (200 μL, 0.5%) and reduced nicotinamide adenine dinucleotide (NADH) (0.2 mg) were added to the above mixture (1 mL). The mixture was then incubated in the dark at 30°C for 30 min; then, 100 μL of formaldehyde was added to terminate the reaction. The mixture was then centrifuged for 3 min (10,000 g), 500 μL of methanol (96%) was added to extract INF, the mixture was centrifuged for 10 min (10,000 rpm), and finally, the supernatant after centrifugation was measured at 490 nm by a spectrophotometer. ETSA was calculated according to the following formula (Eq. (3)) : where ABS 490 is the sample absorbance at 490 nm, 15.9 is the specific absorptivity of INT-formazan, V 0 and V 1 are the initial volumes of bacteria and the total volume of methanol (mL), t is the incubation time (min), 32/2 is the constant for the transformation of μmol INT-formazan to μg O 2 , and m is the protein concentration per milliliter of bacteria (mg protein/mL bacteria).

Microbial community structure determination
Microbial community structure was determined by Sangon Biotech (Shanghai) Co., Ltd., China. The main processes of metagenome sequencing were sample DNA extraction, library construction, and sequencing. Briefly, an appropriate amount of PBS solution (30 mL) was used to clean the obtained biofilm mixture. After cleaning, the mixture was centrifuged at 12,000 rpm for 2 min. DNA was extracted from the centrifuged precipitate with a kit (E.Z.N.A™ Mag-Bind Soil DNA Kit, OMEGA, USA). Then, the samples were inspected to get the electrophoretic test map and quality inspection results. Two rounds of PCR amplification were performed. In the first round of amplification, qubit3.0 DNA detection kit was used to quantify genomic DNA accurately to determine the amount of DNA to be added in PCR reaction. In the second round of amplification, Illumina bridge PCR compatible primers were introduced. The selected amplification area was 16S V3-V4, the amplified fragment was 464bp, and the sequencing primer was 341F (CCTACGGGNGGCWGCAG) /805R (GACTACHVGGGT ATCTAATCC) (Xu et al. 2016). Quality control of the sequencing library was then conducted; the size of the library was detected by 2% agarose gel electrophoresis. In order to get uniform long cluster effect and high-quality sequencing data, the sequencing library was detected by a fluorescence quantitative analyzer (Qubit® 3.0, Invitrogen, USA). Finally, the sequencing data were analyzed by bioinformatics.

Effects of SFCWs' substrates on the removal of pollutants
The overall performance of SFCWs' series for the removal of COD Cr was monitored. The removal of COD Cr by SFCWs having different substrates has been presented in Fig. S2.
As-obtained results from one-way ANOVA showed that there was no significant difference found for COD Cr removal efficiency for all SFCWs before and after aeration (p>0.05). This endorsed that the substrates and aeration have little effect on the COD Cr removal by SFCWs' series.
Furthermore, as depicted in Fig. 2a, the effluent concentration of total nitrogen (TN) fluctuated between 1.44 and 10.18 mg/L. During the initial stage in SFCWs, the adsorption and ion exchange of substrates were responsible for the removal of TN and NH 4 + -N (Buelna et al. 2008;Yuan et al. 2020), which supported that the initial removal efficiency of TN and NH 4 + -N was relatively higher (Fig. 2b and Fig. 2d). As the absorption of NH 4 + -N by the substrate tended to be saturated, the abundance of ammonia-oxidizing bacteria in the system was not enough to oxidize NH 4 + -N, and the removal efficiency of NH 4 + -N began to decline. As both COD Cr and microorganism of SFCWs competed for DO, so DO level in the wetland got decrease; then, the conversion rate of NH 4 + -N was slowed down and the removal efficiency of TN was reduced Sun et al. 2004). For 30-65 days (aeration stage), the influent concentration of TN was ranged from 17.56 to 22.86 mg/L, the influent load increased from range 0.32-0.39 g/day to 0.49-0.60 g/day. Both SFCW-S2 and S4 exhibited a relatively higher TN removal efficiency (35.74-71.79%), while SFCW-S1 and S3 have shown comparatively low TN removal efficiency. In addition, SFCW-S1 reflected a consistently low TN removal efficiency even before and after aeration.
Before aeration (1-30 days), the DO level in the wetland was 0-0.7 mg/L (Fig. S3), and the NH 4 + -N average removal efficiencies for SFCW-S1, S2, S3, and S4 were 71.58%, 77.58%, 58.74%, and 48.36%, respectively, which suggested the moderate nitrification capacity at low DO level (Anjali and Sabumon 2015;Bernat et al. 2011). Under anaerobic and anoxic conditions, NH 4 + -N could be removed by the following processes: (1) as a kind of nitrogen source, NH 4 + -N was assimilated by anaerobic bacteria; (2) adsorbed by the substrate and reused by microbes. In the early stage of operation, the substrate of SFCWs was unsaturated, adsorption played an important role in the removal of NH 4 + -N. With the extension of the operation time, the substrate of SFCWs gradually saturated, and the biofilm attached to the substrate gradually matured, and the NH 4 + -N could be reused by the microorganisms in the biofilm (Kong et al. 2016).
After 30-65 days, aeration was done to reduce the accumulated NH 4 + -N in the wetland. After aeration, DO levels in the wetland increased dramatically (Fig. S3), which was favorable for the growth and metabolism of aerobic organisms (Yuan et al. 2020). The NH 4 + -N concentration in effluent of the four SFCWs decreased significantly after aeration, which met the Class III standard of the Environmental Quality Standard for Surface Water (EQSSW) (GB 3838-2002, China), and the removal efficiency of NH 4 + -N was basically above 90%. In addition to aeration, the reason for the high NH 4 + -N removal efficiency during the later stage might be the good growth of plants, which released a large amount of oxygen as an electron acceptor to promote the oxidation of NH 4 + -N (Yuan et al. 2020). Furthermore, there was no production of NO 3 − -N detected in SFCW-S3 and S4 during the period 1-30 days before aeration as reflected in Fig. 2e, and the denitrification rate was the highest. While the concentration of NO 3 − -N in S1 effluent was 3.30-5.28 mg/L, and the denitrification rate was relatively the lowest.
After the aeration of the wetland (30-65 days), the increase in DO level led to the conversion of NH 4 + -N to NO 3 − -N, and the concentration of NO 3 − -N in the effluent of the four SFCWs got increased. The NO 3 − -N could be removed by denitrification, partial denitrification, and anammox in a constructed wetland. Excessive oxygen content in water inhibited the metabolic process of heterotrophic microorganisms and increased the concentration of NO 3 − -N in water. Compared with SFCW-S2, S3, and S4, the concentration increase of NO 3 − -N in the SFCW-S1 (ceramsite) during the aeration stage was the most significant, which indicated that the denitrification of SFCW-S1 was obviously inhibited. In addition, because of the low carbon-nitrogen ratio in the influent, the NO 3 − -N concentration in wetland SFCW-S1 was higher due to lack of denitrification carbon source, while the NO 3 − -N concentration in wetland SFCW-S2, S3, and S4 was lower, indicating that pyrite and ferrous sulfide in SFCW-S2, S3, and S4 played an essential role in the removal of NO 3 − -N. Wu et al. (2016) used sludge-ceramsite as the main substrate of intermittent-aerated SFCW to treat sewage; the accumulation of NO 3 − -N in the later stage of the experiment showed that complete denitrification could not be realized due to the lack of carbon source, which was the same as the low denitrification efficiency of SFCW-S1 in this study. The pyrite and ferrous sulfide in SFCW-S2, S3, and S4 could promote the removal of NO 3 − -N by autotrophic denitrification (Eqs. (4) and (5)).
For the 55-65 days (aeration stage), the average influent concentration of NO 3 − -N in wetland was 3.63 mg/L, and the effluent NO 3 − -N concentration of four SFCWs was 5.74-10.43 mg/L. The effluent NO 3 − -N concentration of SFCW-S2 and S4 was lower than those of SFCW-S1 and S3. The substrates of SFCW-S2 and S4 contained pyrite, which was an electron donor material for autotrophic denitrification. Si et al. (2021) found that pyrite improved NO 3 − -N removal in CWs and autotrophic denitrification with pyrite as electron donor was an important way to remove NO 3 − -N and TN from polluted water. In addition, Ge et al. (2020) also found that adding pyrite to the CWs significantly improved the denitrification efficiency; the pyrite not only promoted the autotrophic denitrification process, but also stimulated the growth of heterotrophic denitrifying bacteria.
It was noticed that the effluent NO 2 − -N concentration in the four SFCWs increased first and then gradually decreased as -N (f), TP (g, h) by SFCWs filled with different substrates (HRT= 2d; IC refers to influent concentration; ID refers to influent loading; EC refers to effluent concentration; SFCW-S1, SFCW-S2, SFCW-S3, and SFCW-S4 refer to SFCWs with different substrates: ceramsite (S1), ceramsite+pyrite (S2), ceramsite+ferrous sulfide (S3) and ceramsite+pyrite+ferrous sulfide (S4); "Class III" and "Class V" refer to the Class III and Class V standard of the Environmental Quality Standard for Surface Water (GB 3838-2002, China)) demonstrated in Fig. 2f. By comparing the overall four SFCWs, it was found that the effluent NO 2 − -N concentration in SFCW-S3 and S4 was the highest among others. The accumulation of NO 2 − -N can be attributed to incomplete nitrification and partial denitrification. It has been stated by many recent studies that incomplete nitrification in CWs could lead to the increase of NO 2 − -N concentration in the effluent . The effluent concentration of NO 2 − -N in the four SFCWs was significantly different before and after aeration (p<0.01). In view of the characteristics of the substrate in the four SFCWs, the rapid accumulation and removal of NO 2 − -N might be attributed to the rapid growth of microorganisms and the acceleration of ETSA (Wu et al. 2018). In addition, the average concentration of total phosphorous (TP) in the influent was 2 mg/L as depicted in Fig. 2g. For the period, 1-30 days (non-aeration stage), the TP concentration in SFCW-S1 and S3 effluent was low at the beginning; then, TP concentration gradually increased with an increase in operating time, and the removal efficiency of TP gradually decreased (Fig. 2h). The adsorption of phosphorus by the substrate and the absorption by plant roots in CWs were important ways for TP removal (Baldovi et al. 2021). Substrates in SFCW-S1 and S3 contained ceramsite, which had a good adsorption capacity for phosphorus in water (Dong et al. 2021). The plant roots in CWs could absorb soluble inorganic salts as components of cells and tissues; substrate adsorption and plant roots absorption both played important roles for TP removal in the early stage of CWs (Higgins et al. 2016;Hussein and Scholz 2018).
The decreased TP removal efficiency with increasing time for SFCW-S1 and S3 (1-30 days, non-aeration stage) might also be due to the gradual thickening of biofilm on the substrate, which hindered the effective contact and mass transfer between pollutants and substrate, resulting in the decrease of TP removal efficiency (Ye et al. 2011). In addition, the increase of biomass would quickly consume DO in water, which made phosphorus accumulating bacteria release phosphorus under anaerobic conditions (Xu et al. 2021;Zhang et al. 2021). Because the substrate of SFCW-S3 contained ferrous sulfide, the Fe 2+ dissociated from ferrous sulfide substrate would inevitably be oxidized to Fe 3+ although the SFCWs were not aerated. Fe 3+ could react with phosphorus to form stable complex compounds; however, Fe 3+ was easily reduced to Fe 2+ under anaerobic conditions, which made the complexed phosphorus be released, and then the phosphorus removal efficiency reduced (Yang et al. 2017).
The TP removal efficiency in SFCW-S3 drastically increased as soon as aeration mode started (30-65 days); the main reason was that aeration was beneficial to the transformation of substrate released Fe 2+ to Fe 3+ , Fe 3+ hydrolyzed rapidly in water to form a series of hydroxyl complexes, the generated hydroxyl complex adsorbed aqueous phosphorus, and then iron phosphorus complex was formed resulting in the removal of aqueous phosphorus (Kong et al. 2016). The reaction of phosphorus and iron to form precipitation was also accelerated by the aeration. Furthermore, the above reaction rate was low with non-aeration due to the large particle size of ferrous sulfide substrate (20-80 mm) in this study. Aeration promoted the renewal of the reaction interface; therefore, the TP removal efficiency was significantly improved. The removal of TP mainly depended on the chemical precipitation and complexation of the substrate-generated complexes; aeration could enhance the TP removal efficiency of SFCWs.
The TP removal efficiency for the SFCW-S4 was the highest, and the effluent TP concentration (except for the 5th day) could meet the Class III standard of the EQSSW (GB 3838-2002, China). The average concentration of TP in the effluent of SFCW-S2 was 0.27 mg/L, and the average removal efficiency of TP was 86.57%. Except for the 22nd day, the effluent TP concentration could meet the Class V standard of the EQSSW (GB 3838-2002, China). The main reason for the high TP removal efficiency by SFCW-S2 and S4 was that the chemical composition of the substrate contained Fe, Ca, Al, Mg, etc. (Table S1). Aqueous phosphorus could react with metal ions such as Fe, Ca, Al, and Mg to generate phosphate precipitation or hydroxyl metal phosphate precipitation, thus realizing the removal of phosphorus. In addition to the phosphorus removal by precipitation, there was also a mechanism of phosphorus removal by adsorption. Under neutral conditions, the Fe 3+ produced by the substrate iron release hydrolyzed rapidly in water to form a series of hydroxyl complexes, the generated hydroxyl complex adsorbed aqueous phosphorus and then iron phosphorus complex was formed resulting in the removal of aqueous phosphorus (Torrentó et al. 2010). Wang et al. (2019) clarified that Fe 2+ , Fe 3+ , and Fe(OH) 3 as metabolic intermediates of the substrate could be used to remove TP from wastewater by precipitation or adsorption.
According to Fig. S3, the DO concentration of the effluent ranged from 0 to 0.7 mg/L (1-30 days, without aeration), and the effluent was anoxic and anaerobic in most conditions. Because the substrate of SFCW-S2 and S4 contained pyrite and ferrous sulfide respectively, the Fe 2+ dissociated from pyrite and ferrous sulfide substrate would inevitably be oxidized to Fe 3+ although the SFCWs were not aerated. Therefore, phosphorus removal by precipitation and the hydroxyl complexes adsorption was less affected at this stage (1-30 days, without aeration); high TP removal efficiency by SFCW-S2 and S4 was still observed.
For the 31-65 days (aeration), the TP concentration in the effluent of SFCW-S1 (ceramsite) gradually decreased and the TP removal efficiency gradually increased. The increase of aqueous DO enhanced the phosphorus uptake of phosphorus accumulating bacteria under aerobic conditions. Furthermore, the ability of plants to absorb and utilize phosphate gradually increased with the gradual growth of plants, which was conducive to the removal of TP. The effluent TP concentrations of SFCW-S2, S3, and S4 were stable during the aeration time, with average concentrations of 0.12, 0.35, and 0.07 mg/L, respectively. TP concentration in effluent of SFCW-S2 and S4 (except for the 42nd day) could meet the Class V standard of the EQSSW (GB 3838-2002, China).
The XRD peaks data presented in Fig. S4 reflected that the substrate of the four SFCWs primarily consisted of Si, Al, Fe, S, Ca, and Mg-bearing mineral phases, which indicated that the substrate had the potential for phosphate removal via precipitation and adsorption. The high TP removal efficiency of SFCWs was mainly due to the fact that the chemical precipitation formed by the reaction of metals with phosphorus, and the adsorption by the metal oxides and hydroxides in the SFCWs' substrate (Torrentó et al. 2010). The reactions were shown as follows (Eqs. (6)-(16)) (Ge et al. 2019;Omwene and Kobya 2018).
Fe 2þ þ 0: Both FeS 2 and FeS were considered as potential highquality autotrophic denitrification electron donors, which could reduce nitrate-nitrogen to nitrogen; iron was oxidized to ferric hydroxide, and sulfur was oxidized to sulfate (Eqs. (4) and (5)). Under the aerobic conditions, FeS 2 was also oxidized to ferrous iron, ferric iron, and sulfate (Eqs. (6) and (7)). Ferrous iron and ferric iron reacted with phosphorous to form iron phosphate and ferrous phosphate which was difficult to be dissolved in water (Eqs. (8) and (9)). Calcium oxide contained in the substrate reacted with water to generate calcium hydroxide, calcium ions were dissociated from the generated calcium hydroxide, and aqueous phosphorous reacted with calcium hydroxide and calcium ions to generate insoluble phosphate compounds (Eqs. (10)- (14)). Similarly, both aluminum and magnesium contained in the substrate could also react with aqueous phosphorous to form insoluble aluminum phosphate compounds and magnesium phosphate compounds (Eqs. (15)-(16)).

Water quality changes along the length of the SFCWs
On the 55th day of the experiment, samples were taken at P2, P3, and P4 along the length of the SFCWs (Fig. 1). It was noticed that the effluent TN concentration gradually decreased along the length of the SFCWs, and correspondingly, the TN removal efficiency gradually increased as shown in Fig. 3a and Fig. 3b. The highest TN removal efficiencies were noticed for SFCW-S2 and S4, and the TN concentration for SFCW-S3 wetland was higher, when compared with other SFCWs. The TN removal potential for both SFCW-S2 and S4 was much better than that for SFCW-S3 and control wetland SFCW-S1.
It was detected that NH 4 + -N concentration in the sample was completely removed, when water reaches 15 cm away from the inlet point; the removal efficiencies found for SFCW-S1, S2, S3, and S4 were 84.47%, 96.51%, 96.34%, and 97.66%, respectively ( Fig. 3c and Fig. 3d). Because this sampling place was very near to the wetland aeration port, the DO level was higher, which was supportive for the removal of NH 4 + -N by nitrosation reaction. The average effluent NH 4 + -N concentrations for SFCW-S1, S2, S3, and S4 were 1.18, 0.22, 0.27, and 0.26 mg/L, respectively. There was no significant difference for NH 4 + -N removal efficiency among four SFCWs (p>0.05).
Furthermore, NO 3 − -N and NO 2 − -N concentrations firstly increased and then decreased along the length of the SFCWs as presented in Fig. 3e and Fig. 3f. In the first zone of the wetland (15 cm away from the water inlet), which was closest to the aeration port, the higher DO content was found (Fig.  S5), which was beneficial for the growth of nitrite and nitrifying bacteria, and existed NH 4 + -N easily oxidized to NO 2 − -N and NO 3 − -N. However, with an increase of the water flow path, DO content gradually decreased (Fig. S5), which was conducive to accelerate the conversion of NO 2 − -N and NO 3 − -N to N 2 .
At the outlet point, 80 cm away from the water inlet, the concentration of NO 3 − N accumulated in SFCW-S1 and S3 was higher than those determined for SFCW-S2 and S4. Because the particle size of pyrite was much smaller than that of ferrous sulfide and ceramsite in the present study, the substrate with small particle size could get good contact with microorganisms in wetlands and provide attached biological sites for the growth of microorganisms, thus accelerating the removal of NO 3 − -N by autotrophic denitrifying bacteria (Bosch et al. 2011;Juncher Jørgensen et al. 2009;Torrentó et al. 2010). The reason for the low denitrification efficiency of SFCW-S3 might also be that the particle size of ferrous sulfide was relatively large (Ye et al. 2011), and the low solubility of ferrous sulfide decreased its utilization speed during autotrophic denitrifying process; furthermore, the coverage of biofilm on the ferrous sulfide surface hindered the contact between microorganisms and sulfur-containing substances Miot et al. 2011).
The average ratio of COD Cr /NO 3 − along the wetland was 1.82, 1.65, 2.00, and 2.41, respectively, which belonged to the low C/N ratio wastewater (Anjali and Sabumon 2015;Xu et al. 2018). No substrate containing sulfur or iron was -N (f), TP (g, h) along the length of the SFCWs introduced to SFCW-S1, the carbon sources were insufficient for heterotrophic denitrification; therefore, NO 3 − -N could not be effectively removed in SFCW-S1. As shown in Fig. 3f, the cumulative concentration of NO 2 − -N in SFCW-S3 was significantly higher than that in SFCW-S1, S2, and S4 (p<0.01). Torrentó et al. (2010) also observed the phenomenon of NO 2 − -N accumulation in the experiment of groundwater denitrification; the results of isotope tracer analysis showed that the main source of NO 2 − -N was NO 3 − -N; high concentration of NO 3 − -N would inhibit the reduction of NO 2 − -N and lead to NO 2 − -N accumulation. As shown in Fig. 3e, the NO 3 − -N concentration in SFCW-S3 was also the highest among the four SFCWs, which might be the main reason for the high accumulation concentration of NO 2 − -N along the length in SFCW-S3.
The TP concentration gradually decreased with the increasing flow path of the SFCWs, and the removal efficiency gradually increased ( Fig. 3g and Fig. 3h). The TP removal efficiency for SFCW-S4 was above 90%, while the TP removal efficiency of SFCW-S1 was relatively low (<75%). The maximal removal efficiencies for SFCW-S1, S2, S3, and S4 were 72.64%, 96.40%, 81.19%, and 97.88%, respectively. In the whole sampling site of the SFCWs (except 15 cm away from the water inlet), the removal efficiency of TP was in the following order SFCW-S4>SFCW-S2>SFCW-S3>SFCW-S1. There were significant differences found in TP removal efficiency among the four SFCWs (p<0.05), which indicated that the chemical reaction occurred between phosphorus and iron in the substrate; the adsorption of phosphorus by iron oxides and hydroxides in the substrate play an important role in the removal of aqueous phosphorus (Eqs.(4)-(16)). As shown in Fig. 3g, TP concentration decreased with increasing water flow path. At this stage (aeration stage), the SFCWs were always in the aerobic conditions (DO=1-11 mg/L, Fig. S5); the increase of DO in water strengthened the phosphorus absorption by the phosphorus accumulating bacteria under aerobic condition, which subsequently improved the phosphorus removal performance. It should be noted that the removal of TP mainly occurred in the front section of the SFCWs along the water flow path, while the latter half section mainly ensured the stability of effluent water quality, especially for the SFCW-S4 (Fig. 3g); similar results were also observed by other researchers (Wu et al. 2015). Ge et al. (2019) constructed two SFCWs based on natural pyrite and limestone; it was found that pyrite had no negative impact on the growth of reed; the removal efficiencies of COD, NH 4 + -N, TN, and TP were improved by the addition of natural pyrite; the mechanism of long-term simultaneous removal of TN and TP in pyrite constructed wetland was anaerobic and aerobic oxidations of pyrite; the phosphorus remaining in wetland was mainly in the form of (Fe+Al)bound phosphorus (Ge et al. 2019).

Analysis for nitrification and denitrification rate
As shown in Fig. 4a, the nitrification rate of SFCW-S1 was the lowest for the aeration zone. The nitrification rate of SFCW-S2, S3, and S4 was significantly different from that of SFCW-S1 (p<0.01), and the nitrification rates of SFCW-S2, S3, and S4 treatments were 1.37, 1.43, and 1.23 times higher than that of SFCW-S1, respectively. This indicated that the nitrification rates of SFCW-S2, S3, and S4 containing inorganic electron donor (pyrite or ferrous sulfide) were significantly higher than that of SFCW-S1 without inorganic electron donor. The denitrification rate results for the non-aerated zone of SFCWs exhibited that the denitrification rate of SFCW-S2 and S4 was significantly higher than that obtained for SFCW-S1 (p<0.05; Fig. 4b). However, there was no significant difference found between SFCW-S1 and S3 (p>0.05). The denitrification rates for SFCW-S2 and S4 were 2.31 and 2.22 times greater than calculated for SFCW-S1, respectively, suggesting that SFCW-S2 and S4 had higher denitrification efficiency than SFCW-S1.
Analysis for microbial enzyme activity and electron transport system activity AMO is one of the key enzymes found in heterotrophic nitrification process, which assists in NH 4 + -N conversion into NO 3 − -N . As shown in Fig. 4c, AMO activity in the aeration zone of SFCW-S2, S3, and S4 were 1.17, 1.21, and 1.26 times higher than the control group SFCW-S1, respectively. There was a significant difference between the AMO activity of SFCW-S2, S3, and S4, and the control group SFCW-S1 AMO activity, respectively (p<0.05), which was similar to the nitrification rate result (Fig. 4a).
Denitrifying enzymes are responsible for the bio-reduction of NO 3 − -N, NO 2 − -N, NO, and N 2 O in the denitrification process (Wu et al. 2018;Zhao et al. 2020). NAR activity is critical for the denitrification process, reducing NO 3 − -N to NO 2 − -N. Figure 4d shows that the NAR activity for SFCW-S2 and S4 was 1.52 and 2.08 times of that was calculated for SFCW-S1, and 1.41 and 1.93 times of that was calculated for SFCW-S3, respectively, indicating that the decrease in NAR activity for SFCW-S1 and S3 led to the accumulation of NO 3 − -N (Wu et al. 2018). The NIR activity is also considered a key factor for denitrification. Figure 4e shows that there was no significant difference found in NIR activity among SFCW-S2, S3, S4, and S1 (p>0.05). However, the NIR activity for SFCW-S2 and S4 was relatively higher than that obtained for SFCW-S1 and S3. For SFCW-S2 and S4, denitrification efficiency could be promoted by improving the activities of NAR and NIR, and then catalyzing the reduction of NO 3 − -N and NO 2 − -N, which was considered to be the reason for the increased TN removal efficiency and the decreased NO 3 − -N accumulation ( Fig. 2b  and Fig. 2e).
The efficiency for the nitrification and denitrification process mainly depends on the efficient transport of electrons that can be assessed using ETSA (Wan et al. 2016;Wu et al. 2018). AMO, NAR, and NIR obtained electrons from the electron transport system, and the activities of these enzymes are directly related to nitrogen removal efficiency. The AMO, NAR, and NIR activity obtained for SFCW-S1 was comparatively low (Fig. 4c, Fig. 4d, and Fig. 4e), which was an important reason for the low ETSA (Fig. 4f).

Analysis for microbial community structure
The structure and composition of microbial community in SFCWs at the phylum level has been shown in Fig. 5a. The dominant bacterial community has divided into 13 phyla, Proteobacteria accounts for the highest relative abundance at each sampling point, ranging from 44.60 to 87.16%. Proteobacteria is usually the largest phyla found in constructed wetland systems (Guan et al. 2015;Ibekwe et al. 2016).
Proteobacteria contain bacteria responsible for nitrification and denitrification activities and various metabolic bacteria, which play a vital role in the removal of organic matter, nitrogen, and phosphorus.
Cyanobacteria are often detected in high nutrient wastewater as an indicator bacterium (Adyasari et al. 2019), which was enriched in SFCW-S1 and S4. Cyanobacteria with high relative abundance detected in SFCW-S1 indicated that the TN removal performance for SFCW-S1 was lower than that observed for the other SFCWs. Relative higher abundance of Cyanobacteria noticed for SFCW-S4 might be assigned to the accumulation of phosphorus in SFCW-S4 substrate and biofilm. Moreover, Planctomycetes contain bacteria related to Anammox (Bae et al. 2010). Planctomycetes not only has the function of Anammox, but also may play denitrification function (Du et al. 2020;Ishimoto et al. 2020;Kumar et al. 2020;Park et al. 2020;Vipindas et al. 2020). The relative abundance of Planctomycetes in ceramsite substrate SFCW-S1 was higher than that in SFCW-S2, S3, and S4, it might be the   Fig. 4 Nitrification rate in aeration zone (a), denitrification rate in non-aeration zone (b), AMO activity in aeration zone (c), NAR (d), NIR activities (e), and ETSA (f) in non-aeration zone of the SFCWs (the asterisk indicates significant difference between the experimental group (S2, S3, and S4) and the control group (S1), one-way ANOVA, *p< 0.05,**p< 0.01) reason that ceramsite substrate SFCW-S1, which did not contain pyrite and ferrous sulfide substrates, was easier to cultivate the bacteria Planctomycetes. While autotrophic denitrification was the main reaction for TN removal in SFCWs (S2, S3, and S4) containing pyrite and ferrous sulfide substrates, which was not conducive to the growth of Planctomycetes. It was found that the proportion of Bacteroidetes has increased with the outbreak of cyanobacterial blooms and gradually became the dominant bacteria found in eutrophic water (Kolmonen et al. 2004;Rashidan and Bird 2001;Wu et al. 2007). Bacteroidetes were also related to denitrification, no matter in an aerated zone or non-aerated zone, the relative abundance of Bacteroidetes in SFCW-S1 was much higher than that monitored for SFCW-S2, S3, and S4, respectively, indicating that the addition of pyrite and ferrous sulfide substrates to the SFCWs inhibited the growth of Bacteroidetes. Figure 5b shows the microbial community structure composition of the overall four SFCWs at the genus level. Acinetobacter commonly exists around the roots of CWs, and most of them are aerobic bacteria, which can metabolize by using ammonia, nitrogen, and glucose. This is consistent with the fact that the proportion of Acinetobacter in the aeration zones of the overall four SFCWs was higher than that in the non-aerated zones. Rhodobacter is a kind of phototrophic bacteria related to Fe (II) oxidation and nitrogen removal (Chakraborty and Picardal 2013). Higher abundance of Rhodobacter detected in SFCWs containing iron compared with control SFCW-S1 suggested that the addition of pyrite or ferrous sulfide in SFCW-S2, S3, and S4 changed microbial community structure and then improved denitrification efficiency.
Thiobacillus has the potential to use the sulfur present in pyrite for autotrophic denitrification, and the proportions of Thiobacillus found in SFCW-S1, S2, S3, and S4 were 0.26%, 13.00%, 0.20%, and 1.90%, respectively. This trend indicates that Thiobacillus played an important role for the autotrophic denitrification of SFCW-S2 (containing pyrite substrate) and S4 (containing pyrite and ferrous sulfide substrate).
Aerobic denitrifying microorganisms such as Paracoccus, Pseudomonas, Rhizobium, Novosphingobium, and Sphingomonas were detected in aerated zones of the overall four SFCWs. Bacterial denitrification is not strictly an anaerobic process, denitrification may also exist under aerobic conditions. Aerobic denitrifying microorganisms have potential for nitrification and denitrification process simultaneously. Geobacters are iron-reducing bacteria with organic matter as an electron donor (Ge et al. 2019;Leang et al. 2003), which contribute to the removal of COD Cr . In both aerated and nonaerated zones, the proportion of geobacters in SFCW-S2, S3, and S4 was higher than that found in SFCW-S1, indicating that iron-reducing process was more dominant for SFCW-S2, S3, and S4 than for SFCW-S1.
It has been reported that Paracoccus is more suitable to live in the water environment having higher TN content Zhang et al. 2020). The proportion of Paracoccus detected for SFCW-S1, S2, S3, and S4 was 0.31%, 0.05%, 0.98%, and 0.16%, respectively (Fig. 5b). For SFCW-S1 and S3, the proportion was relatively higher, while that of SFCW-S2 and S4 was relatively low, indicating that the TN content for SFCW-S1 and S3 was relatively higher, which was inconsistent with the results presented in Fig. 2a.

Canonical correspondence analysis
Canonical correspondence analysis (CCA) is a nonlinear multivariate direct gradient analysis method, which is commonly used to study the correlation between microbial community structure and water environmental factors (Ter Braak and Prentice 1988). Figure 6 demonstrates the relationship among the microbial community at the phylum level and seven environmental factors. The correlation coefficient between the first and second axes of environmental factors was 0, which indicates that the analysis results are reliable. The principal components (AX1 and AX2) can explain 79.11% of the bacterial structure, with AX1 explaining up to 63.01% and AX2 up to 16.1% of the total variation. Among the environmental factors analyzed, water temperature T and NH 4 + -N were positively correlated with the first ordering axis (AX1), while TP, COD Cr , TN, NO 2 − -N, and NO 3 − -N were negatively correlated with the AX1.
Furthermore, temperature T, NH 4 + -N, and TP were positively correlated with the second sorting axis (AX2), while COD Cr , TN, NO 2 − -N, and NO 3 − -N were negatively correlated the AX2. Proteobacteria was positively correlated with TP, COD Cr , TN, NO 2 − -N, and NO 3 − -N, and negatively correlated with NH 4 + -N and water temperature T. In addition, Bacteroidetes, Planctomycetes, Acidobacteria, and Actinobacteria were positively correlated with water temperature T and NH 4 + -N and negatively correlated with COD Cr , TN, NO 2 − -N, and NO 3 − -N. According to the correlation analysis of the AX1 and AX2, it can be noticed that NO 2 − -N, TP, and water temperature T can significantly affect the microbial community at the phylum level.

Conclusion
The concentration of NH 4 + -N in effluent drops rapidly after intermittent aeration in the front section of the SFCWs, and the removal efficiency of NH 4 + -N gradually reaches over 90%. The removal efficiency for SFCW-S2 and S4 exhibited better removal performance of TN and TP than wetland SFCW-S1 and S3. The average effluent TP concentration for wetland SFCW-S4 could meet the Class III standard of EQSSW (GB 3838-2002, China), and the phosphorus removal efficiency was the highest. Nitrification rate and AMO activity in aeration zone of overall four SFCWs were not significantly different. The denitrification rates, NAR, NIR, and ETSA of SFCW-S2 and S4 were higher than those of SFCW-S1 and S3 in the non-aerated zone. Proteobacteria accounts for the highest relative abundance found at each sampling point, ranging from 44.60 to 87.16%. Heterotrophic denitrification was the main process in wetland SFCW-S1, Thiobacillus plays an important role in pyrite-driven autotrophic denitrification process of wetland SFCW-S2 and S4. NO 2 − -N, TP, and water