Hydrilla verticillata–Sulfur-Based Heterotrophic and Autotrophic Denitrification Process for Nitrate-Rich Agricultural Runoff Treatment

Hydrilla verticillata–sulfur-based heterotrophic and autotrophic denitrification (HSHAD) process was developed in free water surface constructed wetland mesocosms for the treatment of nitrate-rich agricultural runoff with low chemical oxygen demand/total nitrogen (C/N) ratio, whose feasibility and mechanism were extensively studied and compared with those of H. verticillata heterotrophic denitrification (HHD) mesocosms through a 273-day operation. The results showed that the heterotrophic and autotrophic denitrification can be combined successfully in HSHAD mesocosms, and achieve satisfactory nitrate removal performance. The average NO3−-N removal efficiency and denitrification rate of HSHAD were 94.4% and 1.3 g NO3−-N m−3·d−1 in steady phase II (7–118 d). Most nitrate was reduced by heterotrophic denitrification with sufficient organic carbon in phase I (0–6 d) and II, i.e., the C/N ratio exceeded 4.0, and no significant difference of nitrate removal capacity was observed between HSHAD and HHD mesocosms. During phase III (119–273 d), sulfur autotrophic denitrification gradually dominated the HSHAD process with the C/N ratio less than 4.0, and HSHAD mesocosms obtained higher NO3−-N removal efficiency and denitrification rate (79.1% and 1.1 g NO3−-N m−3·d−1) than HHD mesocosms (65.3% and 1.0 g NO3−-N m−3·d−1). As a whole, HSHAD mesocosms removed 58.8 mg NO3−-N more than HHD mesocosms. pH fluctuated between 6.9–9.0 without any pH buffer. In general, HSHAD mesocosms were more stable and efficient than HHD mesocosms for NO3−-N removal from agricultural runoff during long-term operation. The denitrificans containing narG (1.67 × 108 ± 1.28 × 107 copies g−1 mixture-soil−1), nirS (8.25 × 107 ± 8.95 × 106 copies g−1 mixture-soil−1), and nosZ (1.56 × 106 ± 1.60 × 105 copies g−1 mixture-soil−1) of litter bags and bottoms in HSHAD were higher than those in HHD, which indicated that the combined heterotrophic and autotrophic denitrification can increase the abundance of denitrificans containing narG, nirS, and nosZ, thus leading to better denitrification performance.


Introduction
Agricultural runoff always has high nitrate and low chemical oxygen demand (COD)/total nitrogen (TN) (C/N) ratio. Increased nitrate availability in agricultural runoff may lead to serious eutrophication and deterioration of public health in many cases [1][2][3]. Among the available approaches, the free water surface constructed wetland has recently been identified as an effective and ecologically sustainable technology for the treatment of agricultural runoff [4,5].
Biological denitrification, occurring either heterotrophically or autotrophically, accounts for 60-90% of the total nitrate reduction [6][7][8]. Among popular electron donors for autotrophic denitrification, sulfur has competitive characteristics of cost-effectiveness, nontoxicity, water-insolubility, and wide availability [9,10]. In the heterotrophic denitrification process, some chemical organics such as methanol and glucose are widely used as additional carbon sources for the treatment of nitrate-contaminated wastewater with a low C/N ratio [11]. However, such methods are costly due to the continuous organics dosage [12]. What is more, the residual chemical organics added need to be further treated. To substitute the chemical organics and reduce costs, various plant biomass has been proved to be feasible for the improvement of heterotrophic denitrification efficiency [13][14][15][16].
Previous studies have obtained favorable denitrification performance by applying combined heterotrophic and autotrophic denitrification (HAD) process in the treatment of drinking water [17][18][19], hydroponic wastewater [20], groundwater [21,22], simulated wastewater [23][24][25][26][27][28][29][30][31], etc. A woodchip-sulfur combined mixotrophic process was recently developed to eliminate nitrate from groundwater [31], whose denitrification performance was better than that of the sulfur-based autotrophic denitrification systems. However, it was only operated in serum bottles for less than two weeks with limited experimental data. Another study discussed the effects of coexistent ions including Na + , Cl − , HCO 3 − , Ca 2+ , Cu 2+ , Fe 2+ , Fe 3+ , HCO 3 − , and SO 4 2− on nitrate removal by combined sawdust-based heterotrophic and Fe-based autotrophic denitrification processes [23]. Further investigation is needed to prove that HAD process works better than individual heterotrophic denitrification for the improvement of nitrate removal capacity during long-term operation. Dissolved organic matter (DOM), which is one of the important components in organic compounds of plant biomass, is a key influencing factor for heterotrophic denitrification performance [32]. Recent studies have demonstrated that parallel factor analysis (PARAFAC) could decompose full excitation-emission matrix (EEM) fluorescence into different independent fluorescent groups [33], and it is a valuable tool for the characterization and quantification of DOM fluorescence changes.
Based on the studies mentioned above, the Hydrilla verticillata (H. verticillata)-sulfur-based heterotrophic and autotrophic denitrification (HSHAD) process was put forward. H. verticillata is a fresh-water aquatic plant used as the carbon source. This paper aims to verify the feasibility and mechanism of the HSHAD process for the treatment of high nitrate and low C/N ratio agricultural runoff in free water surface constructed wetland mesocosms during long-term operation. Characteristics of H. verticillata decomposition and denitrification performance in the HSHAD mesocosms were also extensively studied in long-term operation and compared with those in H. verticillata-based heterotrophic denitrification (HHD) mesocosms. DOM changes were also analyzed to elaborate the HSHAD mechanism.

Source of Material and Pretreatment
Lake Erhai is a typical plateau lake in Dali city, Yunnan Province, Southwest China. H. verticillata and soil samples were collected from the water near the shore and the paddy-field along the littoral zone of Lake Erhai (25 • 36 -25 • 58 N, 100 • 06 -100 • 18 E), respectively, in August 2014. H. verticillata samples were cleaned, cut into 1 to 2 cm pieces and oven-dried at 40 • C to constant weight. Soil samples were air-dried to constant weight, milled and screened through 100-mesh sieve. The granular sulfur (1-2 cm, >99.9%) and pebbles (SiO 2 , 1-2 cm) were bought from local companies.

Mesocosm Set-Up and Operation
Six 5 L laboratory-scale, free water surface constructed wetland mesocosms were made using plastic buckets (0.17 m bottom inner diameter, 0.23 m upper inner diameter, 0.21 m height) and fed with 2.0 L synthetic agricultural runoff. Pretreated soil from Lake Erhai shore and paddy-field were mixed homogeneously at a 1:1 mass ratio, and then added into the mesocosms. The litter bag technique could carry out the decomposition and denitrification experiments [34][35][36]. A litter bag (0.16 m length, 0.12 m width) loaded with different media (Table 1) was placed on the bottom soil (Table 1) surface in each mesocosm, and the litter bag was made of nylon with approximately 1.0 mm mesh. Two HSHAD mesocosms (packed with H. verticillata and sulfur in the litter bag), two HHD mesocosms (packed with only H. verticillata in the litter bag), and two control mesocosms (packed with only soil and gravels in the litter bag) ( Table 1) were used. The schematic diagram of the mesocosm was published in a previous study [26]. Based upon one-year monitored data of the agricultural runoff in the Lake Erhai area, the ingredients of the simulated agricultural runoff were 0.5 mg L −1 NH 4 Since the nitrate concentration in the control mesocosms was too high (126.1 mg L −1 ), the water column of all mesocosms was replaced by the simulated agricultural runoff on the 204th day.

Sampling and Analytical Procedure
To avoid agitation of the bottom soil, water was sampled from the top using a 50 mL syringe. Meanwhile, the same volume of deionized water was injected into the mesocosms for compensation. At the beginning and end of the experiments, cellulose, hemicellulose, and lignin contents of H. verticillata were quantified according to Van Soest's method [7]. Organic carbon was analyzed by the potassium dichromate oxidation-external heating method (LY/T 1237-1999) [37]. TN was determined by automatic azotometer (UK152 Distillation and Titration Unit, Velp Co., Milano, Italy). TP was determined by inductively coupled plasma-atomic emission spectrometry (IRIS Intrepid II XSP, Thermo, Waltham, MA, USA). Before analysis, H. verticillata pieces were cleaned, oven-dried at 40 • C to constant weight, and then milled and screened through 60-mesh sieve. NH 4 + -N and TP were determined by standard methods [38]. DOM in water samples was characterized by three-dimensional EEM fluorescence spectrophotometer (Hitachi F-7000 Fluorescence Spectrophotometer, Japan) after glass microfiber filtration (Whatman GF/C glass microfiber filters). A xenon excitation source was used for the spectrometer. Both the excitation (Ex) and emission (Em) were 200-500 nm with 5 nm bandwidth, and the scanning speed was 12,000 nm/min. Milli-Q water was utilized as blank control. Fluorescence spectroscopy coupled with parallel factor analysis (PARAFAC) was applied in DOM fluorescence characterization.

DNA Extraction and Q-PCR
Mixed soil from the bucket bottom, and mixture (gravel/H. verticillata pieces/mixed soil) from litter bags were collected on the 273th day for denitrificans genes analysis using quantitative polymerase chain reaction (q-PCR) method.
Before deoxyribonucleic acid (DNA) extraction, the mixture from litter bags was shaken at 200 r min -1 for 1 h in sterile glass bottles in order to detach the biofilm into a liquid phase. Total genomic DNA was extracted using the E.Z.N.A ® Soil DNA Kit (OMEGA bio-tek, Doraville, GA, USA), and was subject to q-PCR assays targeting the narG, nirS, and nosZ genes.

Volumetric denitri f ication rate g NO
, the NO 3 − -N concentration difference between any two sampling time V (L), the effective mesocosm volume ∆t (d), the sampling time interval The statistical analysis was carried out using SPSS software (version 23.0, Chicago, IL, USA). The paired-sample t-test was used to evaluate the differences between the two mesocosms.

Physical-Chemical Component Variation of H. verticillata
Cellulose, hemicellulose, and lignin are the main components of H. verticillata (Table 2) [42]. Cellulose and hemicellulose are macromolecules from different sugars, whereas lignin is an aromatic polymer synthesized from phenylpropanoid precursors. After 273 days' decomposition, cellulose and hemicellulose of H. verticillata (0.17 and 0.23 g g −1 biomass −1 ) decreased to 0.09 and 0.04 g g −1 biomass -1 in HSHAD mesocosms and 0.15 and 0.03 g g −1 biomass −1 in HHD mesocosms, which indicated that sulfur does not have adverse effect on biomass decomposition. Lignin increased from 0.05 g g −1 biomass −1 to 0.22 g g −1 biomass −1 (HSHAD) and 0.15 g g −1 biomass −1 (HHD), respectively, at the same time ( Table 2). Plant biomass decomposition converts complex organic molecules to simple organic constituents as a result of following processes: (1) physical leaching and fragmentation, (2) extracellular enzyme hydrolysis, (3) aerobic and anaerobic catabolic activities of heterotrophic microorganisms [43]. The high microbial cellulose and hemicellulose utilization (Table 2) indicated that they were likely the main carbon source for H. verticillata heterotrophic denitrification in the HSHAD and HHD mesocosms. Cellulose in crystalline form is less susceptible to enzymatic degradation than that in nonorganized form [44]. Crystalline cellulose accounts for a large portion of the total cellulose, which might be responsible for the large portion of the cellulose remaining in H. verticillata (Table 2). Hemicellulose is a complex carbohydrate polymer and more hydrolysable than cellulose [45,46], which explains the reason why most of the hemicellulose was decomposed at the end of the experiment. Lignin is the recalcitrant fraction of the plant biomass, whose microbial utilization requires initial oxygenation [43]. Therefore, the increase of lignin resulted from the nonenzymatic lignin formation and its low degree of microbial utilization [45][46][47][48]. Pérez et al. [46] confirmed that the degradation of lignin was difficult due to its structural complexity, insolubility, and high molecular weight. Moreover, 50% N and 100% P were released from H. verticillata in both the HSHAD and HHD mesocosms during the 273 days of operation.

Principles of the HSHAD Process
Assuming the main component of the cellulose and hemicellulose in H. verticillata is C 6 H 10 O 5 , the stoichiometric equations of H. verticillate heterotrophic and sulfur autotrophic denitrification in the HSHAD process can be listed as Equations (4) and (5), respectively: Two aspects should be noted for the HSHAD process. The first aspect is its ability to keep pH balanced without a pH buffer. According to Equations (4) and (5) The second aspect relates to the enhancement of denitrification performance. Theoretically, the synergistic improvement of the denitrification performance can be achieved if both the sulfur autotrophic denitrificans and H. verticillata heterotrophic denitrificans grow well in the HSHAD system.

Carbon Availability
As depicted in Figure 1a, COD in the control is low and relatively stable. Ingersoll and Baker [49] suggested that the optimum C/N ratio for plant biomass heterotrophic denitrification was 4.0-5.0. HSHAD and HHD were operated in three phases based on the C/N ratio and its fluctuation. Phase I was the short-term physical leaching period (0-6 d, C/N ratio > 4.0), during which the COD rose quickly to the maximum values (i.e., 204.7 and 352.6 mg L −1 for HSHAD and HHD, respectively) in 1 day (Figure 1a). The rapid COD increase resulted from the fast eluviation of water-soluble substances (i.e., organic acid, protein, and minerals) into water. The dynamic equilibrium between organic carbon release and consumption can be attained on the first day. Then COD gradually decreased to 56.9 and 74.8 mg L −1 in the HSHAD and HHD mesocosms on the sixth day. Phase II (7-118 d, C/N ratio > 4.0) was the long-term steady biological decay period, during which COD was released continuously at lower and steady rate from H. verticillata pieces to support the growth of the heterotrophic denitrificans [47,50]. In phase I and II, C/N ratios in HSHAD and HHD were both higher than 4.0, which meant sufficient organic carbon for heterotrophic denitrification. Phase III (119-273 d, C/N ratio < 4.0) was the COD reduction period, i.e., COD decreased gradually, which suggested that the carbon release speed was lower than its consumption speed. The average C/N ratios in HSHAD and HHD mesocosms were 2.4 and 1.4 in phase III, which meant insufficient organic carbon for heterotrophic denitrification.
The COD variation curve of HSHAD was similar to that of HHD (p > 0.05), indicating that there was no significant difference of organic carbon dynamics between HSHAD and HHD mesocosms.

DOM Analysis
Four main fluorescence peaks of EEM fluorescence spectra were identified for water samples from different phases (Table 3). EEM spectral regions were related with different organic functional groups in DOM [51]. Peak C1 was located at 250-280/325-370 nm excitation/emission wavelengths (Ex/Em), which was considered as soluble microbial byproduct-like substances [51,52]. Peak C2, which was identified by 225-230/320-340 nm Ex/Em, was associated with the aromatic amino acid tryptophan, i.e., protein-like substances [51,53,54]. Peak C3 (260-360/410-460 nm Ex/Em) and Peak C4 (230-250/460-480 nm Ex/Em) were described as humic acid-like and fulvic acid-like substances, respectively [51,52,55,56].    Table 3 illustrate the compositional proportion of each peak intensity in different phases (119-203 d of Phase III), and obvious shifts of peak intensity were observed. From phase I to phase II, rapid decrease of the fluorescence intensity ratio of Peak (C1 + C2) to Peak (C3 + C4) was observed. It can be seen from Figure 2 that massive water-soluble substances (e.g., tryptophan, Peak C1) were released initially, and the fulvic acid-like (Peak C3) and humic acid-like (Peak C4) substances were produced throughout the long-term biological decay process. Fluorescence intensity of Peak (C3 + C4) were highest in phase II for all mesocosms. In the compost, the typical mature components were characterized as humic and fulvic acid-like substances (Peak C3 and C4), while the typical immature components contained tryptophan-like substances and soluble microbial byproduct-like substances (Peak C1 and C2) [33]. The decomposition process can be characterized by the degradation of original tyrosine-like and tryptophan-like material (Peak C1 and C2) and the increase of humic and fulvic-like material (Peak C3 and C4) [33,57]. Therefore, it can be concluded from the peak alternation ( Figure 2) that the soil and H. verticillata/sulfur (or H. verticillata) mixture in the experimental mesocosms were gradually matured in Phase II, which corresponded to the steady COD concentration and nitrate removal ( Figure 1).
Almost no peak was found after 204 days of operation when the water medium of the mesocosms was displaced by the simulated agriculture runoff on the 204th day in phase III (data are not shown), which corresponded to the low COD concentration (around 10.0 mg/L) and denitrification rate (Figure 1), and it further confirmed that organic compound release from H. verticillata attained the decline period.
Almost no peak was found after 204 days of operation when the water medium of the mesocosms was displaced by the simulated agriculture runoff on the 204th day in phase III (data are not shown), which corresponded to the low COD concentration (around 10.0 mg/L) and denitrification rate (Figure 1), and it further confirmed that organic compound release from H. verticillata attained the decline period.  A similar trend of denitrification rate was observed for HSHAD and HHD mesocosms, but the average denitrification rate of HSHAD was higher than that of HHD in phase II and III. The denitrification rate of HSHAD gradually increased from the 222th day, but that of HHD still decreased. On the 273th day, the denitrification rate of HSHAD mesocosms (0.7 g NO 3 − -N m −3 ·d −1 ) was 1.3 times higher than that of HHD mesocosms (0.3 g NO 3 − -N m −3 ·d −1 ).
For both HSHAD and HHD mesocosms, the same maximum NO 3 − -N removal value (100.0%) was achieved on the second day. Then, the NO 3 − -N removal efficiency was maintained above 90.0% until the 118th day. No statistical differences were observed in NO 3 − -N removal efficiency between HSHAD and HHD mesocosms during phase I and II (p > 0.05) (Figure 1a), which might be due to the sufficient organic carbon released from plant biomass decomposition [17]. After 118 days of operation, organic carbon inadequacy became an obvious limiting factor for NO 3 − -N reduction. NO 3 − -N removal efficiency of HSHAD and HHD both decreased slowly. What is more, NO 3 − -N removal efficiency of HHD decreased faster than that of HSHAD, which demonstrated that the HSHAD process was more effective and efficient than the individual H. verticillata heterotrophic dentirification process when the carbon source was insufficient. So the denitrification rate of HSHAD was higher than that of HHD in phase II and III, and the NO 3 − -N removal efficiency of HSHAD was also higher than that of HHD after 118 days of operation. On the 273th day, NO 3 − -N removal efficiency of HSHAD (69.6%) was still 41.0% higher than that of HHD (28.6%), which might be caused by the combination of sulfur autotrophic denitrification. Total NO 3 − -N reduction by HHD was calculated using the stoichiometric equation of the heterotrophic denitrification process (Equation (4)). Theoretically, 912.8 mg NO 3 − -N can be reduced by HHD if cellulose and hemicellulose reduced ( Table 2) in H. verticillata pieces were completely utilized by heterotrophic denitrification. However, only 557.8 mg NO 3 − -N was eliminated within 273 days of operation, i.e., 61% of cellulose and hemicellulose reduction was used by heterotrophic denitrification, and the other 39% might be used by anaerobic and aerobic respiration rather than by heterotrophic denitrification, which was consistent with the 39% COD consumption for anaerobic and aerobic respiration reported in wetlands [47]. Total NO 3 − -N reduction achieved by HSHAD mesocosms (616.6 mg) was 58.8 mg greater than that by HHD mesocosms (557.8 mg), which was attributed to sulfur autotrophic denitrification. In general, the HSHAD process was split into two major stages in this study. One stage occurred in the period 0-118 d (phase I and II), during which heterotrophic denitrification dominated at higher C/N ratio (above 4.0); the other stage spanned 119-273 d (phase III), during which sulfur autotrophic dentirification dominated at lower C/N ratio (less than 4.0), which was different from the simultaneous oxidation of sulfur (or thiosulfate) and organic matter by inoculating denitrificans or activated sludge into heterotrophic and autotrophic denitrification systems [20,[58][59][60]. However, this phenomenon agreed well with the findings of Liu et al. [17], i.e., nitrate was primarily reduced by heterotrophic denitrification when the carbon source was adequate, and the residual was subsequently reduced by autotrophic sulfur denitrification. Also, the balanced growth of autotrophic and heterotrophic denitrificans was difficult to achieve without microbial inoculation probably because of the very different growth rates for both bacterial groups [61]. To a great extent, the advantage of HSHAD was its higher nitrate removal capacity and more stable denitrification performance compared with HHD during the later operation period (119-273 d).
To days of operation, i.e., 61% of cellulose and hemicellulose reduction was used by heterotrophic denitrification, and the other 39% might be used by anaerobic and aerobic respiration rather than by heterotrophic denitrification, which was consistent with the 39% COD consumption for anaerobic and aerobic respiration reported in wetlands [47]. Total NO3 − -N reduction achieved by HSHAD mesocosms (616.6 mg) was 58.8 mg greater than that by HHD mesocosms (557.8 mg), which was attributed to sulfur autotrophic denitrification. In general, the HSHAD process was split into two major stages in this study. One stage occurred in the period 0-118 d (phase I and II), during which heterotrophic denitrification dominated at higher C/N ratio (above 4.0); the other stage spanned 119-273 d (phase III), during which sulfur autotrophic dentirification dominated at lower C/N ratio (less than 4.0), which was different from the simultaneous oxidation of sulfur (or thiosulfate) and organic matter by inoculating denitrificans or activated sludge into heterotrophic and autotrophic denitrification systems [20,[58][59][60]. However, this phenomenon agreed well with the findings of Liu et al. [17], i.e., nitrate was primarily reduced by heterotrophic denitrification when the carbon source was adequate, and the residual was subsequently reduced by autotrophic sulfur denitrification. Also, the balanced growth of autotrophic and heterotrophic denitrificans was difficult to achieve without microbial inoculation probably because of the very different growth rates for both bacterial groups [61]. To a great extent, the advantage of HSHAD was its higher nitrate removal capacity and more stable denitrification performance compared with HHD during the later operation period (119-273 d).
To further demonstrate the critical function of sulfur autotrophic denitrification during the later carbon-limited period (119-273 d), SO4 2− concentrations for HSHAD, HHD, and control mesocosms were analyzed (Figure 3). No notable differences were observed in SO4 2− concentrations between the HSHAD and HHD mesocosms until the 93th day. A total of 680.2 mg more SO4 2− was produced in HSHAD mesocosms in comparison to HHD during the whole operation. This occurred because SO4 2− concentrations increased after 46 days of operation (Figures 3) with higher proportion of NO3 − -N reduced by sulfur autotrophic denitrification in the HSHAD mesocosms. In fact, corresponding to the extra 58.8 mg NO3 − -N reduction in HSHAD mesocosms, 443.6 mg SO4 2− should be produced, based on Equation (5), which was lower than the measured value of 680.2 In fact, corresponding to the extra 58.8 mg NO 3 − -N reduction in HSHAD mesocosms, 443.6 mg SO 4 2− should be produced, based on Equation (5), which was lower than the measured value of 680.2 mg. This phenomenon suggests that SO 4 2− generation did not completely obey the stoichiometric relationship due to sulfite and thiosulfate formation by some intermediate reaction processes [31,62].
Concentrations of NH 4 + -N, NO 2 − -N and TP during different phases were also analyzed. Average NO 2 --N concentrations were 1.7 (2.0), 2.9 (2.8) and 1.9 (2.0) mg L -1 for HSHAD (HHD) mesocosms in phase I, II, and III. By comparison with the data for HSHAD, HHD, and control, fast nutrient release from H. verticillata at the beginning (phase I) resulted in a sharp increase in NH 4 + -N and TP concentrations for both HSHAD and HHD mesocosms, which was similar with former research [48,[63][64][65]. However, at phase II and III, NH 4 + -N and TP concentrations averaged 1.1 (1.2) and 1.0 (0.9) mg/L for HSHAD (HHD) mesocosms, which demonstrated that the initial NH 4 + -N and TP accumulation can be successfully removed after phase I. Presumably, the effluent NH 4 + -N, NO 2 − -N, and TP can be less likely accumulated if the experiment was operated in a continuous feeding mode.

pH Change
The pH in both HSHAD and HHD mesocosms presented a similar trend and fluctuated between 6.9 and 9.0 throughout the entire operation (Figure 1b). The pH of HSHAD mesocosms sharply decreased from 8.3 to a minimum value of 7.0 on the second day, and slowly increased to 8.8 on the 93th day, and then declined to 7.6 on the 204th day. The sharp pH decrease within two days may have been caused by aerobic respiration. In phase III, the pH of the HSHAD mesocosms was slightly lower than that of HHD, which resulted from more NO 3 --N removal by sulfur autotrophic denitrification. pH had a closed relation with denitrificans growth, and therefore influenced denitrification performance [66]. The optimum pH for most strains of heterotrophic denitrificans is 7.0-8.0 [67,68], while that for sulfur autotrophic denitrificans is 7.7-8.6 [68]. After 46 days of operation, pH in both of HSHAD and HHD exceeded the optimum range for heterotrophic denitrificans growth (Figure 1b), which may have led to the decline of denitrification performance. After 9 days of operation, the pH in HSHAD was within the optimum range for autotrophic denitrificans. Sulfur autotrophic denitrificans could grow well under such pH conditions in HSHAD, which resulted in the higher denitrification capability of HSHAD than HHD in phase II and phase III.

Mesocosm Denitrifying Genes
The total abundance of denitrifying genes (narG, nirS, and nosZ) in the mesocosms were calculated using their corresponding genes copies to the weight of the DNA samples, and the average abundance of denitrifying genes with standard errors in bottom soil and in mixture from litter bags for HSHAD, HHD, and control mesocosms on the 273th day are illustrated and characterized in Figure 4a-c.
The abundance of nosZ genes was lower than that of narG and nirS genes, which was similar to previous studies [69,70]. The nosZ genes catalyzed the reduction of N 2 O to N 2 and enhanced the growth of bacteria containing the nosZ gene, thus reduced the emission of N 2 O. However, it has been reported that approximately one-third of genome denitrifying bacteria have N 2 O reductase [71].

Comparison of HSHAD with Other HAD Processes
The nitrate removal performance of the HSHAD process was compared to that of other HAD processes, based on literature survey, and the packed fillers and the highest removal capacity of each process are listed in Table 4. In a pilot-scale, horizontal-flow constructed wetland, only 67.0% NO 3 − -N removal efficiency was attained because of inadequate influent organic carbon [20,60]; however, the HSHAD mesocosms could attain the high NO 3 − -N removal efficiency (100%). In bioreactors, most HAD processes could achieve above 89.0% NO 3 − -N removal efficiency with the highest denitrification rate of 5050.0 g NO 3 − -N/m 3 ·d (Table 4) Different denitrification performance probably resulted from the abundance of denitrifying bacteria, reactor configuration, type of carbon source, denitrification microenvironment, operation conditions, etc. Therefore, the optimum parameters (e.g., plant biomass species, mass ratio of plant biomass/sulfur, soil types, dosing positions of plant biomass, sulfur, etc.) should be further investigated to obtain more desirable nitrate removal performance by HSHAD in free water surface constructed wetlands.  [76] a No data available. b Estimated or calculated based on the data of the references.

Conclusions
To effectively treat nitrate-rich agricultural runoff with low C/N ratio, the HSHAD process in free water surface constructed wetland mesocosms was put forward, and its performance was extensively evaluated and compared with that of HHD in mesocosms through a 273-day operation.

1.
Heterotrophic and autotrophic denitrification can be combined in the HSHAD process, i.e., the former process mainly dominated NO 3 − -N reduction during the 0-118 days of operation with 4.0 or higher C/N ratio, while the latter process dominated during 119-273 days of the operation.

2.
The average NO 3 − -N removal efficiency and denitrification rate of HSHAD mesocosms were 94.4% and 1.3 g NO 3 − -N m −3 ·d −1 in steady phase II (7-118 d). The HSHAD process was much more efficient and stable than the HHD process in the long-term operation. At the end of the experiment, the NO 3 − -N removal efficiency of HSHAD mesocosms (69.6%) was 41.0% higher than that of HHD mesocosms (28.6%). The rapid increase of NH 4 + -N, NO 2 − -N, and TP concentration in the beginning did not affect the HSHAD denitrification performance, and a pH buffer was not necessary for its moderate fluctuation throughout the operation. 3.