Nitrate Removal and Dynamics of Microbial Community of A Hydrogen-Based Membrane Biofilm Reactor at Diverse Nitrate Loadings and Distances from Hydrogen Supply End

The back-diffusion of inactive gases severely inhibits the hydrogen (H2) delivery rate of the close-end operated hydrogen-based membrane biofilm reactor (H2-based MBfR). Nevertheless, less is known about the response of microbial communities in H2-based MBfR to the impact of the gases’ back-diffusion. In this research, the denitrification performance and microbial dynamics were studied in a H2-based MBfR operated at close-end mode with a fixed H2 pressure of 0.04 MPa and fed with nitrate (NO3−) containing influent. Results of single-factor and microsensor measurement experiments indicate that the H2 availability was the decisive factor that limits NO3− removal at the influent NO3− concentration of 30 mg N/L. High-throughput sequencing results revealed that (1) the increase of NO3− loading from 10 to 20–30 mg N/L resulted in the shift of dominant functional bacteria from Dechloromonas to Hydrogenophaga in the biofilm; (2) excessive NO3− loading led to the declined relative abundance of Hydrogenophaga and basic metabolic pathways as well as counts of most denitrifying enzyme genes; and (3) in most cases, the decreased quantity of N metabolism-related functional bacteria and genes with increasing distance from the H2 supply end corroborates that the microbial community structure in H2-based MBfR was significantly impacted by the gases’ back-diffusion.


Introduction
Nitrate (NO 3 − ) contamination of surface and groundwater has become a significant challenge due to the uncontrolled discharge of wastewater and the intensive use of fertilizers [1,2]. Given the organic-deficient nature of surface and groundwater (in which, the content of dissolved organic matter is commonly below 18 mg/L) [3,4], the autotrophic denitrification technique is therefore a more reasonable option than its heterotrophic counterpart, despite the higher biodegradation kinetics rate of the latter. As an emerging autotrophic denitrification approach, H 2 -based membrane biofilm reactor (H 2 -based MBfR), has gained widespread popularity in recent years for the purification of NO 3 − -contaminated surface and groundwater, mainly attributed to the unique advantages, namely, allowing efficient and cost-effective NO 3 − elimination with minimal bio-sludge yield and without allowing efficient and cost-effective NO3 − elimination with minimal bio-sludge yield and without need for external carbon source [5][6][7]. In H2-based MBfR, H2 gas, the exogenous electron donor, driven by the H2-concentration gradient across the walls of the hollow-fiber membranes (HFMs), diffuses passively from the intramembrane to the HFMs-attached biofilm. In the biofilm, H2 is oxidized by the denitrifying bacteria (DNB) to reduce the NO3 − that diffuses from the bulk liquid [8].
Given the abundant presence of inorganic carbon (mainly bicarbonate) in most surface and groundwater [9][10][11], the denitrification performance of the H2-based MBfR system is commonly largely dependent on the availability of H2 and NO3 − within the counter-diffusional biofilm. The gas-based MBfR is usually operated with either open-end or close-end HFMs. In the case of open-end operation, due to the markedly higher gas velocity of advective transport in the intramembrane than the diffusive transfer across the walls of HFMs, the intramembrane gas was uniformly distributed with an elevated concentration level, which enables the high microbial activity in the biofilm along the HFMs [12]. Nonetheless, in addition to the massive loss of gas, the open-end operation is inapplicable to H2-based MBfR as it creates an explosive atmosphere. Alternatively, the H2-based MBfR is extensively equipped with close-end HFMs, on account of electron donor (i.e., H2) saving and operational safety [13][14][15]. Unfortunately, close-end HFMs always suffer from the backdiffusion of inactive gases such as nitrogenous and water vapor gases, especially the N2 from bulk liquid and hydrogenotrophic denitrification process, which may severely reduce the overall HFM efficiency in terms of H2 delivery [8,12,16]. According to a previous model-predicted result, the gas transfer rate of HFMs operated at the open-end mode was obviously greater than that of the closeend operation, leading to an approximately 116% increase in the contaminant removal flux of the system [12].
As displayed in Figure 1, in principle, as a result of the continuous H2 consumption by the biofilm along the HFMs, accompanied by the back-diffusion of inactive gases (mainly N2) from the biofilm and bulk liquid into the intramembrane as well as the flow of gases toward the distal end of HFMs, the partial pressure of H2 and N2 gradually increases and decreases with the increase in distance from the H2 supply end, respectively; the diffusion of N2 from the intramembrane into the biofilm is due to its concentration at the end of the HFMs [5]. Until now, to the best of our knowledge, only one single research exists that covers the effects of gas back-diffusion on the gas profiles of biofilm, the gas transfer rate of membrane, and the pollutant removal efficiency of the gas-based membrane biofilm reactor system [12]. No study has been published to give insights into the impacts of gas backdiffusion on the dynamics of a microbial community in H2-based MBfR. Research is needed to address this knowledge gap. The objectives of this study are as follows: (1) to evaluate how the denitrification performance of H2-based MBfR responds to influent NO3 − concentration variation, and (2) to reveal the evolution of the microbial community structure in the biofilm with changing NO3 − loadings and distances from the H2 supply end.

Reactor Configuration
A schematic of the lab-scale H 2 -based MBfR used in this study is shown in Figure 2. The reactor contained 20 polyvinyl chloride made HFMs (effective length = 450 mm, inner diameter = 1.0 mm, outer diameter = 1.5 mm, pore size = 0.01 µm, membrane surface area = 0.042 m 2 ) assembled in a vertical plexiglass cylindrical shell (inner diameter = 45 mm, height = 500 mm, effective volume = 0.6 L). An ultrapure H 2 tank was connected to the lower end of the HFM module for pressurized H 2 supplementation. The upper end of the module was sealed using waterproof epoxy glue. A gas regulator was linked to the H 2 tank for H 2 supplying pressure adjustment. Synthetic medium (see Section 2.2 for details) was pumped from the bottom of the reactor via a peristaltic pump (BT101L-DG-1, Lead Fluid, Baoding, China). To guarantee the complete mixing of bulk liquid, a recirculation pump (BT101L-YZ15/25, Lead Fluid, Baoding, China) was operated at a high flowrate of 100 mL/min. Effluent was collected from the outlet at the top of the reactor.

Reactor Configuration
A schematic of the lab-scale H2-based MBfR used in this study is shown in Figure 2. The reactor contained 20 polyvinyl chloride made HFMs (effective length = 450 mm, inner diameter = 1.0 mm, outer diameter = 1.5 mm, pore size = 0.01 μm, membrane surface area = 0.042 m 2 ) assembled in a vertical plexiglass cylindrical shell (inner diameter = 45 mm, height = 500 mm, effective volume = 0.6 L). An ultrapure H2 tank was connected to the lower end of the HFM module for pressurized H2 supplementation. The upper end of the module was sealed using waterproof epoxy glue. A gas regulator was linked to the H2 tank for H2 supplying pressure adjustment. Synthetic medium (see Section 2.2 for details) was pumped from the bottom of the reactor via a peristaltic pump (BT101L-DG-1, Lead Fluid, Baoding, China). To guarantee the complete mixing of bulk liquid, a recirculation pump (BT101L-YZ15/25, Lead Fluid, Baoding, China) was operated at a high flowrate of 100 mL/min. Effluent was collected from the outlet at the top of the reactor.

Experimental Operation
The inoculated biomass was obtained from a long-term operated denitrifying H2-based MBfR in our lab [19]. A H2 pressure of 0.04 MPa was used throughout the whole experiment period, which is a quintessential empirical parameter extensively implemented in preceding H2-based MBfRs [9,13,20]. The start-up processes of the reactor were as follows: a relatively low influent NO3 −

Synthetic Influent
The synthetic influent was prepared using tap water amended with NaNO 3 , NaHCO 3 , and mineral trace elements. NO 3 − was added according to the demand, and 252 mg/L NaHCO 3 was added as the sole carbon source to maintain the autotrophic bacteria growth [13]. The composition of mineral trace elements was identical to our previous study (in µg/L) [

Experimental Operation
The inoculated biomass was obtained from a long-term operated denitrifying H 2 -based MBfR in our lab [19]. A H 2 pressure of 0.04 MPa was used throughout the whole experiment period, which is a quintessential empirical parameter extensively implemented in preceding H 2 -based MBfRs [9,13,20]. The start-up processes of the reactor were as follows: a relatively low influent NO 3 − concentration of 10 mg N/L was initially used to facilitate the start-up of the reactor, and the influent flowrate was set at 1 mL/min, resulting in a hydraulic retention time (HRT) of 10 h; once the complete removal of NO 3 − was achieved, the influent flowrate was increased to 2 mL/min, corresponding to an HRT of 5 h. Following experiments, in which the influent NO 3 − concentration was sequentially maintained at 10, 20, and 30 mg N/L in phases I, II and III, respectively, were carried out to investigate the effects of NO 3 − availability on the denitrification performance of the system and the dynamics of microbial communities in biofilm. In each phase, 40 days operation was performed to enable the system performance and microbial community structure to reach stabilization. Specifically, a NO 3 − influent concentration of 10-30 mg N/L was adopted since this range is close to the typical NO 3 − concentration in real contaminated groundwater [21][22][23], and is the representative concentration range extensively investigated in H 2 -based MBfRs for NO 3 − removal [14,20,[24][25][26]. concentrations inside the biofilm were determined by a microsensor measuring unit, and the detailed analytical procedure can be found in our previous research [27]. NO 3 − removal flux of the system (J in g/(m 2 ·d)) was calculated by Equation (1) [15,28].

Analytical Methods of Aquatic Samples
where Q is the influent flow rate (m 3 /d); A is the membrane surface area (m 2 ); and S inf and S eff are the influent and effluent NO 3 − concentrations (g/m 3 ), respectively.

Biofilm Sampling and Analysis
In order to evaluate the microbial community structure variation as a function of changing influent NO 3 − concentration, the biofilm sample was obtained by stripping off the entire biofilm from one of the 20 HFMs at the end of phases I, II, and III, named bio-sample N1, N2, and N3, respectively. Additionally, to figure out the microbial community structure at diverse locations of the biofilm fed with a specific NO 3 − loading, as shown in Figure 2, we sampled the biomass on a single HFM from the sampling ports 1, 2, and 3 (with a distance of 5, 20, and 35 cm from the H 2 supply end, respectively) at the end of phase III, which was named bio-sample D, M, and U, respectively. It should be noted that after the biofilm sampling at the end of phases I and II, new biofilms could be reconstructed on the single HFM within 14-21 days; meanwhile, stable NO 3 − removal was achieved at the remaining 19-26 days. All the bio-samples were stored at −80 • C until analyzed. The collected bio-samples were delivered to Novogene Co., Ltd. (Suzhou, China) for highthroughput pyrosequencing analysis to investigate the structure and dynamics of the microbial community. Extraction of genomic DNA was conducted using the cetyltrimethylammonium bromide (CTAB)/sodium dodecyl sulfate (SDS) method [29], and bacterial 16S rRNA genes of V4-V5 regions were amplified using primers 515F (5 -GTGCCAGCMGCCGCGG-3 ) and 907R (5 -CCGTCAATTCMTTTRAGTTT-3 ) [30]. After purification with the Qiagen Gel Extraction Kit (Qiagen, Germany), the amplicon library was generated using the TruSeq DNA PCR-Free Kit (Illumina, San Diego, CA, USA). Paired-end reads were merged by using FLASH (V1.2.7), and high quality tags were screened on the basis of QIIME (V1.9.1) [31], then they were assigned into operational taxonomic units with a similarity threshold of 97% by Uparse V7.0.1001 [32]. The Silva Database [33] was employed on the basis of the Mothur algorithm for taxonomic annotation, and sequence alignment was processed by the MUSCLE (V 3.8.31) [34]. PICRUSt, a classical and powerful platform for predicting functional genes [35,36], was applied based on the whole qualified sequencing results using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database [37], and more details are provided in Langille et al. [38]. In this research, we especially focused on the KEEG orthology of nitrogen metabolism related pathways.

Denitrification Performance
After 60 days running of the H 2 -based MBfR at conditions where HRT and NO 3 − loading equaled 10 h and 10 mg N/L, respectively, biofilms with a relatively uniform thickness of nearly 600 µm were naturally formed on the surface of the HFMs, and complete NO 3 − removal was achieved, implying the successful start-up of the system. Following the stabilization of the system after the HRT was shifted from 10 to 5 h, experiments were conducted to evaluate the effects of electron acceptor availability on the denitrification performance. As delineated in Figure 3, the effluent NO 3 − concentration in phases I, II, and III was 0.12, 0.77, and 7.86 mg N/L, respectively, and NO 2 − was only detected in the effluent in phase III with a concentration of 3.2 mg N/L. According to the existing reference [14], excessively high NO 3 − loading could lead to NO 2 − accumulation in the effluent of H 2 -based MBfR.
Noticeably, the average NO 3 − removal flux (calculated via Equation (1) sequencing results using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database [37], and more details are provided in Langille et al. [38]. In this research, we especially focused on the KEEG orthology of nitrogen metabolism related pathways.

Denitrification Performance
After 60 days running of the H2-based MBfR at conditions where HRT and NO3 − loading equaled 10 h and 10 mg N/L, respectively, biofilms with a relatively uniform thickness of nearly 600 μm were naturally formed on the surface of the HFMs, and complete NO3 − removal was achieved, implying the successful start-up of the system. Following the stabilization of the system after the HRT was shifted from 10 to 5 h, experiments were conducted to evaluate the effects of electron acceptor availability on the denitrification performance. As delineated in Figure 3, the effluent NO3 − concentration in phases I, II, and III was 0.12, 0.77, and 7.86 mg N/L, respectively, and NO2 − was only detected in the effluent in phase III with a concentration of 3.2 mg N/L. According to the existing reference [14], excessively high NO3 − loading could lead to NO2 − accumulation in the effluent of H2based MBfR. Noticeably, the average NO3 − removal flux (calculated via Equation (1)) of the system was markedly enhanced from 0.67 g/(m 2 ·d) to 1.31 g/(m 2 ·d), as the influent NO3 − concentration was increased from 10 mg N/L in phase I to 20 mg N/L in phase II, but the further increase of NO3 − loading to 30 mg N/L in phase III merely resulted in a slightly higher NO3 − removal efficiency (1.50 g/(m 2 ·d)). In combination with the NO3 − removal and NO2 − accumulation results, it can be surmised that the relatively higher NO3 − removal flux at the influent NO3 − concentration of 30 mg N/L was because the quantity of the electron donor (H2) was insufficient to completely reduce NO2 − to N2. The microsensor measurement results, as displayed in Figure 4, indicate that NO3 − was always abundant at varying depths of the biofilm when 30 mg N/L NO3 − was added to the influent, while an influent NO3 − concentration of 20 mg N/L resulted in the formation of an inefficient denitrifying zone in the biofilm interior, where the DNB activity was inhibited because of the lower local NO3 − concentration than the half-maximum-rate concentrations of NO3 − for DNB (K DNB NO3 , 0.2 mg N/L) [39]. The above findings led us to reasonably extrapolate that the availability of electron donor (i.e., H2) was the dominant limiting factor for NO3 − removal in phase III. As illustrated earlier, the H2 availability for biofilm utilization is directly subject to the back-diffusion of N2 from the biofilm interior and bulk liquid. However, no single study exists that adequately covers the influence of the gases' back-diffusion on the dynamics of microbial communities in the counter-diffusional biofilm of a H2-based MBfR. Therefore, we concentrated in this study on the clarification of the microbial community structure characteristics of the counter-diffusional biofilm colonized on the close-end  [39]. The above findings led us to reasonably extrapolate that the availability of electron donor (i.e., H 2 ) was the dominant limiting factor for NO 3 − removal in phase III. As illustrated earlier, the H 2 availability for biofilm utilization is directly subject to the back-diffusion of N 2 from the biofilm interior and bulk liquid. However, no single study exists that adequately covers the influence of the gases' back-diffusion on the dynamics of microbial communities in the counter-diffusional biofilm of a H 2 -based MBfR.
Therefore, we concentrated in this study on the clarification of the microbial community structure characteristics of the counter-diffusional biofilm colonized on the close-end HFMs, which is conductive to furthering our comprehending of biofilm-related contaminant removal behavior.
Water 2020, 12, x FOR PEER REVIEW 6 of 13 HFMs, which is conductive to furthering our comprehending of biofilm-related contaminant removal behavior.

Microbial Community Analysis
To delve into the dynamics of microbial communities, the relatively high abundance of bacteria (top 8) at the genus level were investigated at different phases (bio-samples N1, N2, and N3) as well as distances from the H2 supply end (bio-samples D, M, and U). The qualified sequence reads of biosamples N1, N2, N3, D, M, and U were 59,290, 61,121, 65,304, 61,093, 62,610, and 47,935, respectively. As shown in Figure 5, in phase I, the most abundant genera pertaining to autotrophic denitrification process were found to be Hydrogenophaga and Dechloromonas with a relative abundance of 14.3% and 44.8%, respectively. Hydrogenophaga, known as an unculturable H2-oxidizing bacteria [40], played a dominant role in leading to NO3 − removal in a number of H2-based MBfRs [41][42][43]. The genus of Dechloromonas was deemed to be capable of eliminating NO3 − and ClO4 − using dissimilatory nitrate reductases and/or specialized perchlorate reductases [42,[44][45][46]. The further increase in influent NO3 − concentration from 10 mg N/L in phase I to 20 mg N/L in phase II gave rise to the overwhelming percentage (relative abundance equals 49%) of Hydrogenophaga as well as the strikingly decreased population (<1.0%) of Dechloromonas in the biofilm. This is in accordance with the results of preceding studies that Hydrogenophaga outperformed Dechloromonas in terms of H2 utilization and NO3 − reduction in the case of an elevated NO3 − loading [24,42,47]. Intriguingly, the relative abundance of Hydrogenophaga dropped to 17.2% at the influent NO3 − concentration of 30 mg N/L in phase III. The possible explanations are the following: (1) the shortage of electron donors, which is supported by the observation that a mass of NO3 − and NO2 − was detected in the effluent, as exhibited in Figure 3; (2) the NO2 − accumulation, given the susceptibility of Hydrogenophaga to toxicity of NO2 − [1,13]; and (3) the enrichment of unidentified bacterial strains, which were affiliated to the order of Sphingobacteriales, and might be involved in the conversion process of NO3 − to NO [48].

Microbial Community Analysis
To delve into the dynamics of microbial communities, the relatively high abundance of bacteria (top 8) at the genus level were investigated at different phases (bio-samples N1, N2, and N3) as well as distances from the H 2 supply end (bio-samples D, M, and U). The qualified sequence reads of bio-samples N1, N2, N3, D, M, and U were 59,290, 61,121, 65,304, 61,093, 62,610, and 47,935, respectively. As shown in Figure 5, in phase I, the most abundant genera pertaining to autotrophic denitrification process were found to be Hydrogenophaga and Dechloromonas with a relative abundance of 14.3% and 44.8%, respectively. Hydrogenophaga, known as an unculturable H 2 -oxidizing bacteria [40], played a dominant role in leading to NO 3 − removal in a number of H 2 -based MBfRs [41][42][43]. The genus of Dechloromonas was deemed to be capable of eliminating NO 3 − and ClO 4 − using dissimilatory nitrate reductases and/or specialized perchlorate reductases [42,[44][45][46]. The further increase in influent NO 3 − concentration from 10 mg N/L in phase I to 20 mg N/L in phase II gave rise to the overwhelming percentage (relative abundance equals 49%) of Hydrogenophaga as well as the strikingly decreased population (<1.0%) of Dechloromonas in the biofilm. This is in accordance with the results of preceding studies that Hydrogenophaga outperformed Dechloromonas in terms of H 2 utilization and NO 3 − reduction in the case of an elevated NO 3 − loading [24,42,47]. Intriguingly, the relative abundance of Hydrogenophaga dropped to 17.2% at the influent NO 3 − concentration of 30 mg N/L in phase III. The possible explanations are the following: (1) the shortage of electron donors, which is supported by the observation that a mass of NO 3 − and NO 2 − was detected in the effluent, as exhibited in Figure 3; (2) the NO 2 − accumulation, given the susceptibility of Hydrogenophaga to toxicity of NO 2 − [1,13]; and (3) the enrichment of unidentified bacterial strains, which were affiliated to the order of Sphingobacteriales, and might be involved in the conversion process of NO 3 − to NO [48].
Concerning the microbial communities at diverse locations of the biofilm fed with influent containing 30 mg N/L NO 3 − , Hydrogenophaga, was always the primary genus regardless of the distance from the H 2 supply end, where its relative abundance was dramatically increased from 13.1% at the distance of 5 cm to 21.7% at the distance of 35 cm. This is unsurprising in light of the fact that attributed to the impact of back-diffusion, the quantity of H 2 that could penetrate the HFMs and be available for the utilization of DNB proliferated at different positions of biofilm was negatively correlated with the distance from the H 2 supply end. As a frequently discovered autotrophic DNB in H 2 -based MBfRs [19,47], Methyloversatilis was found to be the second most abundant functional bacteria for NO 3 − removal, possessing a relatively stable relative abundance ranging from 4.0-5.2% at varying locations of the biofilm. The unidentified_Sphingobacteriales, a possible NO 3 − consumer as mentioned earlier, with its content variation tendency with changing positions was similar to that of Hydrogenophaga, but merely occupied a quite lower population (0.3-1.4%). The unidentified_Nitrospiraceae, the well-known autotrophic nitrite-oxidizing bacteria (NOB) [43], preferred to enrich at the upper side instead of the lower side of the biofilm, presumably due to the fact that the low H 2 pressure at the upper side of the intramembrane facilitated the entrance into the biofilm and the subsequent consumption of trace of dissolved oxygen arisen from the influent by the NOB. Concerning the microbial communities at diverse locations of the biofilm fed with influent containing 30 mg N/L NO3 − , Hydrogenophaga, was always the primary genus regardless of the distance from the H2 supply end, where its relative abundance was dramatically increased from 13.1% at the distance of 5 cm to 21.7% at the distance of 35 cm. This is unsurprising in light of the fact that attributed to the impact of back-diffusion, the quantity of H2 that could penetrate the HFMs and be available for the utilization of DNB proliferated at different positions of biofilm was negatively correlated with the distance from the H2 supply end. As a frequently discovered autotrophic DNB in H2-based MBfRs [19,47], Methyloversatilis was found to be the second most abundant functional bacteria for NO3 − removal, possessing a relatively stable relative abundance ranging from 4.0-5.2% at varying locations of the biofilm. The unidentified_Sphingobacteriales, a possible NO3 − consumer as mentioned earlier, with its content variation tendency with changing positions was similar to that of Hydrogenophaga, but merely occupied a quite lower population (0.3-1.4%). The unidentified_Nitrospiraceae, the well-known autotrophic nitrite-oxidizing bacteria (NOB) [43], preferred to enrich at the upper side instead of the lower side of the biofilm, presumably due to the fact that the low H2 pressure at the upper side of the intramembrane facilitated the entrance into the biofilm and the subsequent consumption of trace of dissolved oxygen arisen from the influent by the NOB.
Throughout the whole experiment, other DNBs such as Azonexus [42,49] were detected in the bio-samples but with an exceedingly low relative abundance. Since SO4 2− was included in the influent, some genera involved in sulfur reduction (Desulfovibrio, Sediminibacterium) [50,51] and sulfide oxidation (Flavobacterium and Sulfuritalea) [18,52] were also discovered, thus a sulfur-relating microcirculation could occur in the counter-diffusional biofilm.

Predictive Functional Genes
PICRUSt was applied to predict the functional genes in the bio-samples on the basis of the highthroughput sequencing results of the 16S rRNA gene, and the odds ratios of the predictive functional genes are summarized in Figure 6. The variation tendencies in predicted relative abundance of a few basic metabolic pathways including xenobiotics biodegradation and metabolism, membrane transport, and energy metabolism are in full agreement with those of the detected dominant DNB (i.e., Hydrogenophaga) at conditions of varying NO3 − loadings and distances from the H2 supply end. Xenobiotics biodegradation and metabolism is closely related to the resistance of microorganisms to the toxicity of the exogenous contaminant [53]. The predicted highest abundance (5.87%) of the genes relating to xenobiotics biodegradation and metabolism appeared in phase II, probably due to the largest population of Hydrogenophaga and poor accumulation of noxious intermediates at this stage. Throughout the whole experiment, other DNBs such as Azonexus [42,49] were detected in the bio-samples but with an exceedingly low relative abundance. Since SO 4 2− was included in the influent, some genera involved in sulfur reduction (Desulfovibrio, Sediminibacterium) [50,51] and sulfide oxidation (Flavobacterium and Sulfuritalea) [18,52] were also discovered, thus a sulfur-relating microcirculation could occur in the counter-diffusional biofilm.

Predictive Functional Genes
PICRUSt was applied to predict the functional genes in the bio-samples on the basis of the high-throughput sequencing results of the 16S rRNA gene, and the odds ratios of the predictive functional genes are summarized in Figure 6. The variation tendencies in predicted relative abundance of a few basic metabolic pathways including xenobiotics biodegradation and metabolism, membrane transport, and energy metabolism are in full agreement with those of the detected dominant DNB (i.e., Hydrogenophaga) at conditions of varying NO 3 − loadings and distances from the H 2 supply end. Xenobiotics biodegradation and metabolism is closely related to the resistance of microorganisms to the toxicity of the exogenous contaminant [53]. The predicted highest abundance (5.87%) of the genes relating to xenobiotics biodegradation and metabolism appeared in phase II, probably due to the largest population of Hydrogenophaga and poor accumulation of noxious intermediates at this stage. With a similar trend, membrane transport, known to play a fundamental role in the substantial and ATP transportation process [54], occupied a conspicuously greater relative abundance of 17.69% in phase II than those (9.11-14.47%) in phases I and III. The evolution of denitrification behavior of functional bacteria is correlated to the changes of the nitrogen metabolism-related genes, belonging to energy metabolism-related genes. The nitrogen metabolism-related genes were most abundant with a proportion of 0.95% in phase II. It is noteworthy that the gene abundance involved in nitrogen metabolism was increased from 0.83% in the upside to 0.88% in the downside of the biofilm; this mirrored the variation tendency in the abundance of the genes involved in membrane transport with changing distance from the H 2 supply end. The foregoing function prediction results can support that the microbial metabolism at different locations of the biofilm in the H 2 -based MBfR was significantly affected by the back-diffusion of inactive gases. metabolism was increased from 0.83% in the upside to 0.88% in the downside of the biofilm; this mirrored the variation tendency in the abundance of the genes involved in membrane transport with changing distance from the H2 supply end. The foregoing function prediction results can support that the microbial metabolism at different locations of the biofilm in the H2-based MBfR was significantly affected by the back-diffusion of inactive gases. To figure out the nitrogen reductase involved in the nitrogen metabolism pathway, copy numbers of functional enzymes relating to the autotrophic denitrification process including nitrate, nitrite, nitric oxide, and nitrous oxide reductases were examined, as exhibited in Figure 7. The results regarding the counts of predictive functional genes encoding for nitrate reductase consisting of alpha subunit, beta subunit, gamma subunit, cytochrome, and electron transfer subunit are shown in Table  1. As to the influent NO3 − concentration series, the largest counts of nitrate reductase genes of 57,554 and nitric oxide reductase genes of 19,475 were found in phase II, probably due to the greatest population of Hydrogenophaga genera in the biofilm, as shown in Figure 5. It is worth noting that despite the proportion of Hydrogenophaga and counts of nitrate enzyme genes in phase II were obviously greater than those in phase III ( Figures 5 and 7), however, the NO3 − removal flux in phase II was slightly lower than that in phase III (Figure 3). This can be ascribed to the decreased availability of NO3 − for DNB in the biofilm interior in phase II (Figure 4), which gave rise to the declined overall activity of these functional bacteria. Although the nitrite reductase genes were more enriched in phase III rather than phase II, a considerable amount of NO2 − was accumulated in the effluent in phase III (Figure 3). According to a preceding reference [27], the biofilm depth (from the bulk liquid side) that the electron donor (H2) could reach was negatively correlated with NO3 − loading; thus the activity of the nitrite reductase in DNB that grew in the vicinity of the HFM side might suffer from To figure out the nitrogen reductase involved in the nitrogen metabolism pathway, copy numbers of functional enzymes relating to the autotrophic denitrification process including nitrate, nitrite, nitric oxide, and nitrous oxide reductases were examined, as exhibited in Figure 7. The results regarding the counts of predictive functional genes encoding for nitrate reductase consisting of alpha subunit, beta subunit, gamma subunit, cytochrome, and electron transfer subunit are shown in Table 1. As to the influent NO 3 − concentration series, the largest counts of nitrate reductase genes of 57,554 and nitric oxide reductase genes of 19,475 were found in phase II, probably due to the greatest population of Hydrogenophaga genera in the biofilm, as shown in Figure 5. It is worth noting that despite the proportion of Hydrogenophaga and counts of nitrate enzyme genes in phase II were obviously greater than those in phase III ( Figures 5 and 7), however, the NO 3 − removal flux in phase II was slightly lower than that in phase III ( Figure 3). This can be ascribed to the decreased availability of NO 3 − for DNB in the biofilm interior in phase II (Figure 4), which gave rise to the declined overall activity of these functional bacteria. Although the nitrite reductase genes were more enriched in phase III rather than phase II, a considerable amount of NO 2 − was accumulated in the effluent in phase III (Figure 3).
According to a preceding reference [27], the biofilm depth (from the bulk liquid side) that the electron donor (H 2 ) could reach was negatively correlated with NO 3 − loading; thus the activity of the nitrite reductase in DNB that grew in the vicinity of the HFM side might suffer from the shortage of H 2 . Regarding the nitrous oxide reductase genes, the counts in phase I were found to be strikingly greater than those in phases II and III. The limited expression of nitrous oxide reductase in phases II and III is presumably associated with the insufficient supply of electron donors in the case of high electron acceptor (NO 3 − ) loading.  As the distance from the H2 supply end was increased, the counts of nitrate, nitrite, and nitrous oxide reductase genes gradually decreased, with this variation trend identical to those in relative abundances of Hydrogenophaga and predicted genes involved in N metabolism. For instance, the counts of nitrate, nitrite, and nitrous oxide reductase genes in the biomass at the upside of HFMs was decreased by 1428, 1378, and 626, respectively, in comparison to those at the downside. This result offers evidence that the expression of most denitrifying enzymes of DNB in the biofilm was significantly hampered by the back-diffusion of inactive gases, especially those that grew at the locations far from the H2 supply end. In particular, in most cases of diverse NO3 − loading and distance from the H2 supply end, the counts of genes encoding for nitrous oxide reductase are the fewest among denitrifying enzymes, likely as a consequence of the prioritized consumption of electron donor by other N metabolism-related enzymes in the case of electron donor deficiency. The relatively less expression of nitrous oxide reductase implies the possible accumulation of denitrifying intermediates (i.e., N2O). A previous mechanism study results suggested that biofilm systems could As the distance from the H 2 supply end was increased, the counts of nitrate, nitrite, and nitrous oxide reductase genes gradually decreased, with this variation trend identical to those in relative abundances of Hydrogenophaga and predicted genes involved in N metabolism. For instance, the counts of nitrate, nitrite, and nitrous oxide reductase genes in the biomass at the upside of HFMs was decreased by 1428, 1378, and 626, respectively, in comparison to those at the downside. This result offers evidence that the expression of most denitrifying enzymes of DNB in the biofilm was significantly hampered by the back-diffusion of inactive gases, especially those that grew at the locations far from the H 2 supply end. In particular, in most cases of diverse NO 3 − loading and distance from the H 2 supply end, the counts of genes encoding for nitrous oxide reductase are the fewest among denitrifying enzymes, likely as a consequence of the prioritized consumption of electron donor by other N metabolism-related enzymes in the case of electron donor deficiency. The relatively less expression of nitrous oxide reductase implies the possible accumulation of denitrifying intermediates (i.e., N 2 O). A previous mechanism study results suggested that biofilm systems could result in obviously greater amount of N 2 O emissions than suspended-growth systems, attributed to their nature in terms of microbial stratification and substrate gradients [55].

Conclusions
In this study, the NO 3 − removal performance and characteristics of microbial community structure were investigated in a close-end operated H 2 -based MBfR. Based on the analysis of the concentration variations of NO 3 − and/or its intermediate product (i.e., NO 2 − ) in the effluent and biofilm as a function as changing NO 3 − loading, the H 2 availability for biofilm utilization was found to be the main limiting factor for NO 3 − removal at an influent NO 3 − concentration of 30 mg N/L. Microbial community analysis results suggest that at NO 3 − loadings of 20-30 mg N/L, Hydrogenophaga was always recognized as the dominant functional bacteria in the collected bio-samples, regardless of the distance from the H 2 supply end; an influent NO 3 − concentration of 20 mg N/L was found to facilitate the enrichment of Hydrogenophaga; the relative abundance of Hydrogenophaga was negatively correlated with the distance from the H 2 supply end. Functional genes analysis results corroborate that the variation trends of relative abundance of basic metabolic pathways and counts of functional enzyme genes with varying NO 3 − loading and distance from the H 2 supply end are, in most cases, in good agreement with the population evolution of Hydrogenophaga; due to the impact of the gases' back-diffusion, a majority of functional genes pertaining to the microbial metabolism as well as the denitrification process gradually decreased from the downside to upside of HFMs.