The first principle of the MBfR is that it delivers H₂ without forming bubbles. Several types of hollow-fiber membranes can be used in the MBfR; the most common are dense, single-layer polypropylene fibers and composite fibers made of a dense internal layer of polyethylene sandwiched by external layers of macroporous polyurethane. In either case, the key characteristic is that they can be pressurized with H₂ gas without forming bubbles. This prevents off-gassing and allows H₂ delivery to be on-demand and 100% efficient. In addition, the use of these dense (non-porous) polymeric gas-permeable membranes in MBfRs provides long-term stable gas transmission without fouling problems.

A second principle is that the membranes serve as an active substratum for a microbial biofilm community that is composed of microorganisms able to biologically reduce oxyanions in the water (Rittmann, 2018). Figure 1 shows schematically the principles by which the MBfR efficiently and safely delivers H₂ by diffusion through the membrane wall to a biofilm that accumulates on the membrane's outside surface. The microbial communities in the MBfR biofilms are naturally occurring consortia. Changes in operating conditions (explained later) cause the communities to respond in terms of total accumulation and ecological composition.

Biofilms on H₂-supplying membranes are called counter-diffusional biofilms, because the electron donor and acceptor penetrate the biofilm from opposite sides. This is illustrated in Figure 2. Counter-diffusional biofilms have special characteristics compared to conventional, co-diffusional biofilms, where the donor and acceptor diffuse from the same side of the biofilm. When H₂ and the oxyanion are supplied from the bulk (co-diffusional biofilm), the most metabolically active region of the biofilm is the exterior, where H₂ and the electron acceptor substrate are at their highest concentrations. In contrast, in a counter-diffusional biofilm the most active zone may be anywhere within the biofilm, depending on the donor and acceptor concentrations in the biofilm. Counter diffusion of donor and acceptor leads to unique behaviors: development of unique microbial community structures, greater sensitivity to biofilm accumulation, and reduced susceptibility to liquid diffusion layer (LDL) resistance (Nerenberg, 2016). Each is explained below.

The MBfR technology began in the laboratory, where tests were conducted with a variety of membrane materials, reactor configurations, and water characteristics (Lee and Rittmann, 2002). Most of the MBfR laboratory experiments, as exemplified in Figure 3, were performed with an unstructured, loose bundle of gas-permeable hollow fibers inside an outer tube; the water flow was along the axial length of the fiber bundle. While this configuration has been effective for assessing process feasibility and microbiology, it was not easily scaled up to larger devices required for municipal and industrial applications.

Scale-up and commercial development of the H₂-based MBfR began in 2005 by Applied Process Technology, Inc., now APTwater, LLC. Numerous pilot and demonstration tests (summarized in Table 1) were conducted to develop methods of fiber handling and to improve biofilm exposure, fluid flow distribution, and biomass management. APTwater had to define large-scale manufacturing techniques and create component supply chains. To reduce market confusion with membrane filtration processes, APTwater adopted the name ARoNite™ for the autotrophic reduction of nitrate, and similar ARo names are used for other oxyanions (e.g., ARoPerc for perchlorate reduction).

Commercial-scale MBfR modules are manufactured by APTwater by weaving gas-permeable hollow fibers into a fabric sheet, several of which are wound onto a perforated core tube along with alternating layers of a spacer mesh which provide fabric sheet separation and water flow distribution. A depiction of the module assembly is shown in Figure 4. Each end of the fabric and spacer assembly is potted in resin and machined to open the lumen ends of each hollow fiber. A cap is placed on each end to provide H₂ to the lumen of the fibers. Water to be treated is fed into the perforated core tube and travels spirally outward across the surface of the hollow-fiber fabric. Treated water exiting the module can be partially recycled to the module, with the remaining water forwarded downstream.

Approval of the ARoNite™ system by the State of California for municipal water treatment for NO3- was received in 2013. The first full-scale system was constructed in mid-2018 at a municipal well-site in La Crescenta, California, with startup and commissioning completed in August 2018. After validation testing, it is expected to be in full operation by early 2019. Figure 5 shows a photograph of the latest commercial system.

To compare the MBfR with other existing technologies, a triple-bottom-line analysis was conducted to evaluate life-cycle costs and non-monetary benefits for three nitrate-removal approaches in the context of a 10 million gallons/day groundwater treatment facility for the City of Glendale, Arizona (Meyer et al., 2010). The three approaches were: (1) MBfR + Ozone + Biologically Active Carbon Filtration, (2) Static Upflow Bed with Plastic Media + Ozone + Biologically Active Carbon Filtration (a conventional biological nitrate removal approach), and (3) Fixed Bed Ion Exchange (a conventional physicochemical approach). Compared to the second approach, the first approach provided more environmental benefits due to better water utilization, more drinking water quality benefits due to better contaminant removal, and more residuals benefits due to less production of solids/biomass. However, the second approach represented a slightly lower total cost and offered more societal benefits due to less maintenance complexity. The third approach had the highest score for permitting, since it has already been widely applied, but the poorest scores for societal and environmental impacts due to the production of large quantity of residues. The cost of the third approach was between the first two approaches.

A biofilm's capacity to reduce oxyanions depends on the biofilm accumulation and activity. Accumulation is the total mass of biofilm per unit surface area, and it is the product of the biofilm's thickness and biomass density. As discussed previously, a biofilm that is too thin has inadequate biomass to catalyze the reaction, but a biofilm that is too thick increases the mass transfer resistance into the biofilm. Biofilm accumulation can be managed in part via a loop that recirculates part of the treated water back into the MBfR, because the recirculation rate affects biofilm detachment through shear stress and turbulence (Rittman, 1982; Cunningham et al., 1991). Biofilm accumulation also is managed by periodic (e.g., daily) high-shear events, in which excess biomass is detached (usually by a gas sparge to create a two-phase gas-water high velocity stream) and then removed from the reactor (Evans et al., 2013).

Biofilm activity is affected by the rate of detachment. Minimal detachment allows non-active biomass to accumulate in the biofilm, and this slows the biofilm's metabolic activity. Regular detachment keeps the biofilm more metabolically active.

Biofilm activity is also affected by the pH, which usually increases in the MBfR used for oxyanion reduction due to proton consumption. Table 2 summarizes the biological reactions for H₂ and key oxyanions. The molar proton consumption per mole of oxyanion varies from 0.11 to 2, depending on the oxyanion. Too much pH increase not only can inhibit the microorganisms' activity, but also can lead to precipitation of minerals, such as calcium carbonate, calcium hydrogen phosphate, calcium dihydrogen phosphate, hydroxyapatite, and β-tricalcium phosphate (Lee and Rittmann, 2003). The minerals in the biofilm increases mass transfer resistance, which makes the biofilm less active. The pH increase and precipitation can be quantified by a model developed based on mass and charge balance (Tang et al., 2011). The pH can be controlled by two methods. The first method is adding acid (e.g., HCl) into the influent at a concentration that balances the proton consumption, and the second method is sparging carbon dioxide into the reactor at a set point using a pH-control loop (Evans et al., 2013). The model in Tang et al. (2011) predicts the amount of acid needed in the first method and the pH set point in the second method. The second method is implemented at the commercial scale using pre-mixed gas containing CO₂.

When H₂ gas is supplied via hollow-fiber membranes, other dissolved gases in the bulk liquid or gases formed within the biofilm can diffuse into the membrane, diluting the H₂ gas. When operating an MBfR with membranes sealed on one end, these gases concentrate at the distal end of the membrane, possibly decreasing their effectiveness for H₂ transfer and leading to thinner biofilms (Ahmed et al., 2004; Perez-Calleja et al., 2017).

Operating with open-ended membranes eliminates this problem, but leads to loss of H₂, which wastes H₂ and also creates a potential combustion hazard. A better solution is to periodically vent the membrane. By opening the sealed end for a few seconds every few minutes, the H₂ level can be re-established within the membrane with minimal waste (Perez-Calleja et al., 2017).

A biofilm in the H₂-based MBfR used for removing an oxyanion usually consists of four biomass types: (1) H₂-oxidizing bacteria, (2) SMP (soluble microbial products)-oxidizing bacteria, (3) extracellular polymeric substances (EPS), and (4) inert biomass. As illustrated in Figure 6, the four biomass types interact with each other through four dissolved chemical species: H₂, the oxyanion, utilization-associated products (UAP), and biomass-associated products (BAP) (Laspidou and Rittmann, 2004; Tang et al., 2012a; Liu et al., 2016). In brief, the H₂-oxidizing bacteria use H₂ to reduce the oxyanion to a harmless or immobilized form, and they produce EPS and UAP as byproducts. The EPS help the microorganisms to stick together and to the membrane surface area. EPS accumulation is balanced by its hydrolysis to BAP. The UAP and BAP together are called SMP, which stimulate the growth of the SMP-oxidizing bacteria that also reduce the oxyanion. Both active biomass species can be converted to inert biomass through endogenous respiration.

The co-existence of several oxyanions is common and adds more interactions among the biomass species and the dissolved chemical species. The added interactions depend on the co-existing oxyanions. When the oxyanions (e.g., nitrate and sulfate) are reduced by different bacteria (Moura et al., 1997), the interactions can be described by including one more kind of H₂-oxidizing bacteria that use H₂ in parallel (e.g., H₂-oxidizing denitrifying bacteria and H₂-oxidizing sulfate-reducing bacteria) (Tang et al., 2013). The two kinds H₂-oxidizing bacteria compete for H₂ and space in the biofilm. Previous experimental and modeling studies have shown that often a subordinate electron acceptor (e.g., sulfate) is not reduced unless the more dominant electron acceptor (e.g., NO3-) surface loading (defined below) is low (Tang et al., 2013).

The co-existence of nitrate and perchlorate represents a more complicated interactive scenario, since most H₂-oxidizing perchlorate-reducing bacteria also can use nitrate as an electron acceptor (Kengen et al., 1999; Giblin and Frankenberger, 2001; Chaudhuri et al., 2002; Okeke et al., 2002); this leads to competitive inhibition between nitrate and perchlorate on the one hand, but promotion of perchlorate reduction by nitrate on the other hand (Tang et al., 2012a). The interactions are quantified in some case studies in the next section.

To close this section, Table 3 identifies needs for moving the MBfR technology forward for the range of applications in water-quality engineering.

The performance of an MBfR can be established using four criteria: effluent concentration, percentage removal, surface loading, and removal flux. Each is defined here and used to characterize a set of case studies that follow. An oxyanion's effluent concentration (denoted C_eff) determines whether or not the treatment goal has been attained. C_eff has concentration units, or M/L³. C_eff can be compared to an MCL or discharge standard. The percentage removal (%Rem) compares C_eff to the influent concentration, C_inf:

The surface loading (SL) is

in which SL has units of M/L²-T, Q is the influent flow rate (L³/T), and A is the biofilm surface area (L²). The surface loading is used to define the size of the MBfR in terms of its biofilm surface area, or A = QC_inf /SL. As MBfR modules have a set area, computing A from the SL makes it possible to determine the number of modules, which leads to estimate of the capital costs and areal footprint. Finally, the removal flux (J, also in units of M/L²-T) is

The removal flux is the design parameter for achieving the desired C_eff for a given oxyanion and for mediating competition among different electron acceptors (Ziv-El and Rittmann, 2009; Zhao et al., 2013a,b; Ontiveros-Valencia et al., 2014a). Ideally, a study provides all of the performance criteria, and this makes it possible to assess mechanisms, regulatory compliance, and economics.

The performance of the H₂-based MBfR for denitrification has been summarized previously by Di Capua et al. (2015) and Karanasios et al. (2010). It is common to report MBfR denitrification performance in terms of removal fluxes. However, it is important to note that removal fluxes can vary significantly for the same reactor, depending on the H₂-supply pressure, the type of membrane, whether the membranes are operated as open ended or closed ended, the bulk nitrate concentration, the biofilm thickness, the type of membrane material, the mixing regime, and the water temperature.

Reported denitrification fluxes ranged from <0.3 g N m⁻² d⁻¹ to over 14 g N m⁻² d⁻¹, but were commonly between 0.5 and 2 g N m⁻² d⁻¹. Higher fluxes tended to occur with high influent loadings, higher H₂ supply pressures, and batch operation, which allows higher bulk nitrate concentrations for much of the treatment cycle. Percent removals from 90 to 99 percent are common. In practice, the nitrate flux and loading are usually controlled by the effluent concentration of nitrite (a nitrate-reduction daughter product), rather than the nitrate concentration, since the maximum contaminant level (MCL) set by USEPA is 10 times lower for nitrite (1 mg N/L) than for nitrate (10 mg N/L) (USEPA, 2012). For example, in a pilot-scale study conducted by Tang et al. (2010), the effluent nitrite concentration reached the MCL of 1 mg N/L at a nitrate surface loading rate of 6 g N/m⁻² d⁻¹, when the effluent nitrate concentration was below its MCL of 10 mg N/L.

The simultaneous removal of nitrate and perchlorate was studied experimentally and via modeling (Zhao et al., 2011; Tang et al., 2012a,b). Nitrate affects perchlorate removal in two opposing ways. On the one hand, nitrate promotes perchlorate removal by stimulating the perchlorate-reducing bacteria, since most perchlorate-reducing bacteria can use nitrate as the primary electron acceptor. On the other hand, nitrate can inhibit perchlorate by competing for common resources, such as H₂ and space in the biofilm. A multispecies biofilm model based on these mechanisms was developed in Tang et al. (2012a,b) and validated by experimental results in Zhao et al. (2011). The model predicted that nitrate promotes perchlorate reduction when the nitrate loading is <0.1 g N/m²-day, because the promotion mechanism dominates; however, it inhibits perchlorate reduction when the nitrate loading is higher than 0.6 g N/m²-day, because the inhibition mechanism dominates. Nitrate's effect on perchlorate reduction is negligible when the nitrate loading rate is between 0.1 and 0.6 g N/m²-day. Perchlorate can be removed to a few ppb level when the nitrate loading is below 0.6 g N/m²-day and the perchlorate loading is below 0.01 g ClO4-/m²-day (Tang et al., 2012b). In practice, the influent nitrate concentration usually is a few orders of magnitude higher than the influent perchlorate concentration (Kimbrough and Parekh, 2007). In this situation, a two-stage MBfR can be used (Zhao et al., 2013b; Ontiveros-Valencia et al., 2014b). The first stage can remove nitrate at a very high loading rate up to 6 g N/m²-day (Tang et al., 2010), and the second stage can remove both nitrate and perchlorate simultaneously at a nitrate loading rate lower than 0.6 g N/m²-day.

Flue-gas desulfurization (FGD) and coal mining are two of the major sources of selenium in the environment. FGD is used to remove sulfur dioxide from exhaust gases of power plants, and FGD wastewaters have high concentrations of total dissolved solids (TDS), nitrate, nitrite, sulfate, and selenate. Despite the challenging conditions that FGD brines present, they have been successfully treated in the MBfR (Van Ginkel et al., 2010, 2011a,b). In those studies, high TDS concentrations (e.g., 15–33 g/L) did not interfere with the reduction of nitrate and selenate by the biofilm microbial community. Furthermore, sulfate reduction did not occur in the above-mentioned studies. Microbial competition for electron donor (H₂ gas) and the high TDS levels seemed to be related with the lack of sulfate reduction when denitrification and selenate reduction were the dominant and subordinate microbial processes, respectively (Van Ginkel et al., 2010, 2011a,b). Nitrate at 500 mg N/L and selenate at 36 mg/L were reduced up to 98 and 99%, respectively, in synthetic FGD brines. Experimentation with real FGD brines (nitrate/nitrite concentration of 56 mg N/L and 10 mg/L of selenate) also achieved the same reduction rates observed with synthetic FGD. Thus, the MBfR achieved a total-Se concentration in the ppb range (Van Ginkel et al., 2011b). It appears that the co-reductions of nitrate/nitrite and selenate increased biofilm accumulation and improved the removal of both oxyanions.

Because denitrification and selenate reduction can raise the pH, it was necessary to control the pH to avoid precipitation of calcium and magnesium ions from the brines. A pH of 6.8 in the bulk liquid was sufficient to avoid an excessively high pH inside of the biofilm, and, consequently, the reduction rates for nitrate and selenate were not affected (Van Ginkel et al., 2011a).

A synthetic mining wastewater, contained up to 45 mg N/L nitrate and selenate of 0.6 mg/L, was treated in a two-stage MBfR system. The first stage, called lead MBfR, reduced the concentration of the dominant oxyanion (Ontiveros-Valencia et al., 2014a), nitrate in this case. The second stage, called the lag MBfR, finished the treatment by taking the concentration of the sub-ordinate oxyanion, selenate for this study, to a low concentration. The lead MBfR was responsible of average nitrate reduction rate of 91–94%, and the lag MBfR achieved selenate reduction rates up to 87% (Mehta, 2016). Depending on the flow rate at which the lead and lag MBfRs were operated, the range of selenate surface loadings was 0.004–0.04 g Se/m² day, while the nitrate loading range was 0.1–1 g N m⁻² d⁻¹.

Not only is the MBfR capable of removing selenate from FGD brines and mining wastewater, but it also produces and retains elemental selenium (Se⁰) in the biofilm matrix. Transmission electron microscopy coupled with energy dispersive x-ray spectroscopy (TEM-EDX) analysis of the solids from the MBfR biofilm documented the presence of elemental selenium (Ontiveros-Valencia et al., 2016). Thus, the MBfR holds promise to recover a valuable resource.

Heavy metal contamination of the environment is a widely recognized concern. Oxidized metals in the form of oxide anions are very soluble, which leads to contaminant transport after discharge into aqueous streams and threatens ecological function and human health. For instance, hexavalent chromium [Cr(VI)], present as soluble chromate ion (CrO42-), causes irritation of the skin, eyes, and mucous membranes (Leonard and Lauwerys, 1980).

Some of these toxic metal oxyanions can be reductively transformed by microorganisms into less toxic and/or insoluble forms. In the example of chromium, bacteria utilizing H₂ as the electron donor, reduce the hexavalent chromate anion to the trivalent chromite (Cr³⁺) cation:

In neutral or basic conditions (pH > 6), Cr³⁺ spontaneously precipitates with OH⁻:

The Cr(OH)₃ precipitate tends to associate with bacterial cell surfaces and extracellular polymeric substances (EPS), particularly in biofilm matrices, and it is consequently separated and immobilized from water.

A number of MBfR studies reported successful bio-reduction of Cr(VI) as the sole electron acceptor(Long et al., 2017; Zhong et al., 2017) or along with primary electron acceptors including nitrate, sulfate, and/or selenate (Chung et al., 2006c, 2007c; Lv et al., 2017). The Cr-only tests support that the biofilms were able to grow by respiring Cr(VI), and Cr(VI) reduction was not only a detoxification response against oxidative stress (Sedláček and Kučera, 2010). The multi-acceptor tests reveal that other electron acceptors affected Cr(VI)-reducing capacity in distinct ways. On one hand, substantial nitrate and/or sulfate reduction caused considerable retardation of Cr(VI) removal (Lv et al., 2017). For example, after 10 mg N/L of nitrate was added, the Cr(VI)-reducing flux significantly dropped from 0.12 to 0.08 g Cr(VI)/ m² day, or from 86 to 57% removal of the initial 1 mg Cr(VI)/L (Chung et al., 2006a).

DF is an employee of APTwater, which is commercializing the MBfR. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.