Comparison of fouling between aerobic and anaerobic MBR treating municipal wastewater

The key driver for anaerobic membrane bioreactors (AnMBR) for municipal wastewater treatment is enabling the transition to energy neutral wastewater treatment. However, municipal wastewater delivers a comparatively constrained methane yield, which means energy conservation must be prioritised to achieve the proposed energy neutral ambition. A critical focus on membrane fouling is therefore warranted, as membrane operation represents the primary energy demand in MBRs. This review seeks to quantify the characteristics of the prevailing AnMBR biological suspension and to ascertain whether knowledge transfer exists between fouling characteristics in aerobic and anaerobic MBRs for municipal applications. Analysis of literature data revealed that the level of extractable extracellular polymeric substrate is slightly higher in aerobic MBRs than in anaerobic MBRs. However, AnMBR comprises considerably higher soluble microbial product concentrations, which have been widely reported to increase fouling propensity in aerobic systems. More distinct is the difference in the colloidal and fine solids fraction (between 1 and 10–15 μm), which is likely to dominate fouling in anaerobic systems and limit knowledge transfer from aerobic MBRs. Literature data on energy production was compared to that employed for membrane operation, and evidences that despite the challenging character of the particle matrix, energy neutral operation is achievable for AnMBR applied to municipal wastewater treatment.


INTRODUCTION
Anaerobic membrane bioreactors (AnMBRs) hold significant potential to reduce the overall energy demand of municipal (domestic) wastewater treatment through obviating the need for aeration, increasing energy recovery from methane production and producing lower secondary sludge (Martin- . Several reviews have been published on AnMBR over the past ten years. Liao et al. (2006) focused on AnMBR technology for high strength industrial wastewater treatment. More recent reviews have since updated the position of AnMBR for industrial treatment (Skouteris et al. 2012;Lin et al. 2013), which include a particularly insightful narrative on the thirty years of AnMBR commercial development for industrial application (Lin et al. 2013).
Development of AnMBR for municipal wastewater treatment has been comparatively limited due to the past perception that municipal wastewater could not be treated anaerobically since the low organic concentration was insufficient to support microbial growth (Lester et al. 2013). Integrating membrane technology into anaerobic systems helps respond to this challenge by decoupling hydraulic retention time (HRT) from solids retention time (SRT), thereby diminishing washout. The capability of AnMBR to achieve chemical oxygen demand (COD) and suspended solids compliance to international discharge standards has been experimentally demonstrated on municipal wastewater (Martin Garcia et al. 2013). The impact of reactor configuration and operating conditions on organic biodegradation, coupled with a description of potential methods for integrating AnMBR technology into the flowsheet for municipal wastewater, is presented in recent reviews (Smith et al. 2012;Ozgun et al. 2013).
While the capability of AnMBRs to treat municipal wastewater has been demonstrated, the principal driver for their implementation is in enabling the transition toward energy neutral wastewater treatment. McAdam et al. (2011) demonstrated this conceptually for anaerobic treatment of temperate settled wastewater, evidencing around 0.28 kWh m À3 of additional energy production (including recovery of the dissolved methane fraction) (Cookney et al. 2016). However, through comparison of the typical energy demand of membrane operation in commercial aerobic MBR technology (0.19-0.70 kWh m À3 ; Judd 2011; Krzeminski et al. 2012), it can also be evidenced that energy conservation must be prioritised to achieve the proposed energy neutral ambition. Consequently, membrane fouling can be considered one of the critical barriers in achieving commercially viable AnMBR for municipal wastewater treatment, as this governs the energy demand (Judd 2011).
The mechanisms underpinning membrane fouling in aerobic MBR for municipal wastewater have been studied extensively and summarised (Le-Clech et al. 2006;Meng et al. 2009). The factors that influence membrane fouling in both aerobic and anaerobic MBR can be generally defined by the membrane characteristics, biomass properties, reactor and membrane operational conditions. Very few studies have directly compared aerobic and anaerobic MBR experimentally, for either municipal wastewater or blackwater (Baek & Pagilla 2006;van Voorthuizen et al. 2008;Martin-Garcia et al. 2011). These studies acknowledge an important distinction in fouling behaviour between aerobic and anaerobic systems, however, an in-depth characterisation and comparison of foulants has not yet been completed. What is clear from the literature is that size distribution and organics concentration in the AnMBR mixed liquor determines the characteristics and mechanisms of fouling. Providing greater resolution on the characterisation can thus inform the most effective fouling control strategies, as compared with the more extensively studied aerobic MBRs (AeMBRs). This review therefore proposes to complement existing knowledge through characterisation of biomass properties within AnMBRs for municipal wastewater treatment, evaluating their impact on membrane fouling and comparing to those in AeMBR, so as to ascertain the main factors that determine differences in fouling behaviour and characteristics between these two systems. Since membrane fouling strongly influences energy demand, published data on both energy production from AnMBR for municipal wastewater treatment and membrane specific energy utilisation was further analysed to ascertain the position of the existing literature relative to the aspiration of 'energy-neutral' wastewater treatment.

BIOMASS CHARACTERISTICS
Differences in characteristics between aerobic and anaerobic sludge are most readily attributed to the different mechanisms involved in the biological process. Aerobic biological suspensions mainly comprise microorganisms, decay products and influent solids forming microbial aggregates which are held together by high molecular weight polymers secreted by bacteria. This allows them to exist at high population densities (Laspidou & Rittmann 2002) in virtue of their high biomass yield and growth rates. It has been reported that the structure, morphology and surface properties of such suspensions can be altered by changes in physiological state of the biomass induced by changes in bioreactor operational parameters such as SRT and food to microorganism (F:M) ratio (Liao et al. 2001).
On the other hand, anaerobic degradation of wastewater with dissolved, colloidal and particulate organic matter involves several sequential steps such as hydrolysis, acidogenesis, acetogenesis and methanogenesis (van Lier et al. 2008). Hydrolysis is thought to be an extracellular reaction where solids are converted to simple monomers by extracellular enzymes secreted by hydrolytic and fermentative bacteria (Sanders et al. 2000;Vavilin et al. 2008). As a result, and due to the low hydrolysis rates and biomass yield of anaerobic bacteria in high rate anaerobic reactors, the reactor solids inventory is considered to be mainly constituted by influent particles (Soto et al. 1993) that are of reduced particle size (Elmitwalli et al. 2001) and density (Lant & Hartley 2007). Therefore, as opposed to aerobic systems, sludge properties are probably more dependent on influent characteristics than on bioreactor operational parameters.

Mixed liquor suspended solids
Although sludge flocs are not considered the main contributors to fouling under low flux operation, mixed liquor suspended solid (MLSS) concentration has been shown to negatively affect membrane fouling in AnMBRs (Table 1). For instance, Jeison & van Lier (2006) found that changing biomass concentration had a greater impact on the formation of a cake layer than varying gas sparging intensity. Le Clech et al. (2003) also reported that the effect of MLSS was higher than gas sparging in an aerobic MBR. However, in their study, critical flux increased at higher biomass concentrations of 12 g MLSS L À1 as compared to the lowest MLSS concentration of 4 g L À1 . Robles et al. (2013a) also indicated that higher MLSS concentration led to lower membrane permeability, which supports the observations of Jeison & van Lier (2006). Martin Garcia et al. (2013) compared a flocculated AnMBR configured as a continuous stirred tank reactor (CSTR) and a granulated AnMBR configured as an upflow anaerobic sludge blanket (UASB) reactor for settled municipal wastewater treatment, which presented MLSS concentrations for membrane filtration of 7.7 and 0.1-0.6 g L À1 , respectively. Higher fouling rates and lower critical fluxes have been determined in CSTR configured AnMBR with higher MLSS concentrations.
The contradicting influence that MLSS concentration has on membrane hydraulic performance between aerobic and anaerobic MBRs can be attributed to differences in the relationship between biomass and dissolved/colloidal compounds in the mixed liquor. In aerobic MBRs, it has been widely reported that higher levels of soluble microbial products are found at lower MLSS when short sludge retention times are applied (Massé et al. 2006;Liang et al. 2007) while in anaerobic systems, soluble microbial products (SMP) tend to accumulate together with biomass (Harada et al. 1994) or at high sludge ages.

Particle size distribution
According to Lant & Hartley (2007), anaerobic sludge particle sizes are one order of magnitude lower than aerobic floc (activated sludge floc), even though the range of particle sizes covers three orders of magnitude as opposed to only one for aerobic biomass. Data collated from the literature identified similar median particle size ranges of 5.2-220 and 0.8-138 μm in aerobic and anaerobic MBRs respectively (Table 2), however, a significant difference is the presence of a population of fine colloidal matter in AnMBR which has been shown to negatively affect membrane performance. This was confirmed by Martin-  who found a higher colloidal content in a flocculated AnMBR (CSTR) and granulated AnMBR (UASB), when compared directly with AeMBR. The authors attributed the higher degree of dispersive growth to higher fouling propensity.
Particle size analysis in AnMBRs (Table 2) indicates that externally configured (side-stream) membrane systems yield considerably lower particle sizes when compared to immersed systems using gas sparging (Imasaka et al. 1989;Bailey et al. 1994;Jeison & van Lier 2006)  which can be attributed to the shear imposed by the pumping demanded to maintain cross-flow velocity (Imasaka et al. 1989;Jeison et al. 2009). Analogous effects have been demonstrated following conversion of conventional anaerobic reactors to AnMBR (Ho & Sung 2009;Ozgun et al. 2015a) and it is suggested that such particle reduction can also decrease biomass activity (Brockmann & Seyfried 1997). Apart from hydrodynamic conditions, operating temperature has been shown to influence biomass characteristics. Ozgun et al. (2015b) observed a reduction in median particle size from 80-137 to 25-88 μm when the temperature was reduced from 25 to 15°C during the treatment of synthetic municipal wastewater. This was corroborated by Robles et al. (2013a), following the transition from mesophilic to psychrophilic conditions. The authors proposed that the smaller particle sizes led to lower cake layer porosities, and higher cake layer tortuosity, which resulted in considerable cake layer resistance. Transients in particle size distributions have been noted in both anaerobic and aerobic MBR systems following changes to SRT. Huang et al. (2011Huang et al. ( , 2013 reported a decrease in median particle size for AnMBR when SRT increased from 30 to 60 d, which is similar to the observations of Martin- . Similar effects of particle size reduction in aerobic MBR in response to an increase in SRT have been observed (Huang et al. 2001;Martin-Garcia et al. 2011 (2011) indicated higher protein and polysaccharides accumulated at long SRT, which resulted in increased fouling. These biopolymers have been fractionated according to whether they are found in the sludge supernatant as SMP or bound to the sludge flocs and are thus extracted from the cell walls (eEPS). Although the term SMP implies that these substances are of bacterial origin, they may also be the result of recalcitrant or partially transformed influent organics. This is particularly the case for anaerobic systems at lower temperatures where lower biodegradation rates apply, which increase SMP concentration above those of aerobic biomass . However, independent of their origin, analysis of eEPS and SMP has contributed to further characterisation both of the solid and the colloidal/soluble fractions of biological suspensions respectively.

Bound/extractable EPS
The surface properties of the sludge are primarily determined by eEPS. The hydrophobicity and surface charge have been correlated with the total EPS concentration and the ratio of proteins to carbohydrates in both conventional activated sludge and MBR systems (Liao et al. 2001). For instance, a higher ratio of proteins to carbohydrates in activated sludge has been reported to enhance bioflocculation through the reduction of surface charge and increase in hydrophobicity (Liao et al. 2001). On the other hand, high proportions of carbohydrates in the eEPS are associated with a more dispersed sludge structure due to the greater repulsion between sludge particles and interaction with the aqueous phase resulting from the higher negative surface charge and reduced hydrophobicity (Liao et al. 2001). Literature data regarding surface properties of anaerobic sludge do not correlate as consistently as those from aerobic systems, and no conclusions can be drawn regarding comparative concentrations of eEPS and fractions thereof, to allow comparison between aerobic and anaerobic sludges with respect to surface charge and hydrophobicity. Morgan et al. (1990) reported that aerobic sludge was more negatively charged, contained higher levels of total eEPS and lower ratios of proteins to carbohydrates than anaerobic sludge. Similarly, analysis of eEPS literature data from MBR operated with municipal wastewater (Crude, Settled and Synthetic) demonstrates a considerably higher protein to carbohydrate ratio in AnMBR when compared to AeMBR (Table 3). In terms of specific eEPS concentration, the distinction between aerobic and anaerobic systems is less clear. Comparison between anaerobic and aerobic MBRs operated with complete retention of solids and fed with settled sewage (Baek & Pagilla 2006) showed levels of EPS to continuously decrease. In a further study on AnMBR, Fawehinmi et al. (2004)    n Use MLVSS/MLSS ratio ¼ 0.63 to convert the unit from mg g SS À1 to mg g VSS À1.
o Values in the bracket were calculated directly from the average value.
values above 200 mg g VSS À1 in a synthetic wastewater. Analysis of the arising cake layer revealed that the specific eEPS deposited on the membrane surface was twice that found in the granular sludge (Chu et al. 2005) with higher eEPS protein to carbohydrate ratio on the membrane, whilst Lin et al. (2009) reported a lower eEPS protein to carbohydrate ratio in the fouling layer when compared to the bulk material. Regardless of the major fouling component, the discrepancies between biomass and cake layer eEPS composition found in these studies suggest that soluble or colloidal compounds are also responsible for the increase in membrane resistance in AnMBRs.

Soluble-colloidal EPS: SMP
It has been reported that soluble organic matter in the effluent from the biological treatment processes is predominantly SMP, which comprises the soluble cellular components released during cell lysis, lost during synthesis, or otherwise secreted for some purpose (Laspidou & Rittmann 2002). Soluble microbial products are classified according to their origin as products associated with biomass growth and are produced at a rate proportional to substrate utilisation and non-growth associated products related to cell lysis. In conventional (i.e. non-membrane based) systems, the concentration of SMP normalised to influent COD is higher in aerobic than anaerobic processes (Barker & Stuckey 1999). This can be explained by the lower biomass uptake and decay rates of anaerobic microorganisms compared to aerobic biomass. The SMP concentration within aerobic and anaerobic MBR is higher than for conventional processes Massé et al. 2006). Ozgun et al. (2015a) directly compared the SMP of conventional UASB with UASB coupled with membrane and reported that SMP increased over three times after membrane addition, from about 37 to 120 mg L À1 . This is due to an increased retention of the high molecular weight organic fraction by the membrane (Massé et al. 2006) as well as an increase in net SMP productivity through endogenous decay and cell lysis which are enhanced by long SRT and high organic loading rates (Harada et al. 1994). Anaerobic MBRs have considerably higher SMP concentrations than are present in aerobic MBR (Table 4). Normalising literature SMP COD data to influent COD (w/w), suggests arising SMP COD ranges from 10 to 48% and from 9 to 59% for aerobic and anaerobic MBRs, respectively. This has been confirmed through comparative experimental study of anaerobic and aerobic MBR systems for municipal wastewater treatment, which has evidenced SMP concentrations up to five times higher in anaerobic MBR (Martin- . The higher colloidal content within AnMBR, compared with that of AeMBR, may reflect the higher levels of free bacteria in mixed liquor, together with the lower biodegradation rates or SMP biodegradability expected under anaerobic conditions (Ince et al. 2000). There is also evidence of high molecular weight polymeric material of up to 1,000 kDa being retained by the cake layer in AnMBRs, which presumably decreases permeability and potentially increases rejection capability (Harada et al. 1994). Feed temperature is also a factor, where decreased temperature from 25 to 15°C in an AnMBR increased SMP COD from 50 to 150 mg L À1 (Ho & Sung 2010).

FOULING CONTROL STRATEGIES
While much of the early research on AnMBR for industrial applications sought to pursue externally configured MBR, the development of AnMBR for municipal applications has been generally more focused on immersed MBR technology, which is presumably due to the lower energy penalty that this configuration can achieve (Judd 2011).

Specific gas demand and operation flux
The operational costs related to membrane operation in immersed MBRs are mainly determined by the relationship between the specific gas demand (SGD m ) and operating flux, with the SGD m being the gas flow rate per unit membrane area (the specific aeration demand, SAD m , for aerobic systems). This reflects the relationship between the convective flow towards the membrane produced by permeate suction and the back transport induced by the gas sparging and tangential shear at the boundary layer (Liu et al. 2003). In AeMBRs, sustainable or critical fluxes have been reported to increase by increasing gas intensity up to a certain threshold value beyond which no further increase in flux is observed for flat sheet , hollow fibre (Guglielmi et al. 2007) and multi-tubular membranes (Le-Clech et al. 2006). For instance, Chen et al. (2016) reported an increase in critical flux from 23 to 47 LMH for an AeMBR when SAD m increased from 4.5 to 9.0 m 3 m À2 h À1 . Analysis of full scale immersed AeMBRs indicated an operational SAD m range of 0.21-0.88 m 3 m À2 h À1 , corresponding to fluxes between 24 and 31 LMH (Verrecht et al. 2008).
Increasing membrane flux has been shown to increase fouling rates and decrease the duration of the slow fouling phase in aerobic MBRs under conditions of sub-critical flux operation, prior to the widely reported 'TMP jump' (Pollice et al. 2005;Le Clech et al. 2006). For example, under sub-critical conditions, a decrease in flux from 10 to 2 LMH caused an exponential decrease in the fouling rate, from 19.8 to 0.46 kPa h À1 , and prolonged the time before the TMP jump to up to 8 days (Brookes et al. 2006). Results from a pilot scale study suggest that the duration before the TMP jump decreases linearly at fluxes close to the critical flux and that an asymptote exists at a certain flux below which operation can be extended to long filtration cycles (Guglielmi et al. 2007).
Research into immersed AnMBR for municipal wastewater treatment (Table 1)  Operational fluxes for AnMBR (Table 1) between 2.5 and 25 LMH have been reported with the higher fluxes achieved using synthetic wastewater (Wu et al. 2009;Gouveia et al. 2015a). Similar to AeMBRs, an increase in membrane flux, albeit below the critical flux, also leads to an increased fouling rate in AnMBRs (Vallero et al. 2005). An increase in gas sparging intensity appears to be effective in extending membrane operation in both AeMBRs and AnMBRs, but does not obviously enhance permeability. For instance, Weinrich & Grélot (2008) reported sustained permeability for an AeMBR of 600 LMH bar À1 for around two months at a flux of 25 LMH and SAD m of 0.20-0.35 m 3 m À2 h À1 . Robles et al. (2013b) reported sustained permeability above 100 LMH bar À1 in an AnMBR for over 30 days while operating at a sub-critical flux of 13.3 LMH and SGD m of 0.33 m 3 m 2 h À1 . Three years of AnMBR operation has been reported, by adopting an operating flux of 12-14 LMH, which sustained TMP between 0.35 and 0.6 bar, without extra physical and chemical cleaning (Gouveia et al. 2015b). Overall, the permeability is still lower than for AeMBRs which for full scale municipal wastewater treatment plants is between 150 and 250 LMH bar À1 (Judd 2011), even when the quantity of gas provided to the membrane is up four times higher and fluxes between two and three times lower.
Physical and chemical cleaning: reversible, irreversible and irrecoverable fouling Membrane fouling can be generally classified as external fouling (cake and gel formation) and internal fouling (pore clogging) (Judd 2011;Wang et al. 2014). In AeMBRs, cake and gel layer fouling are Cru CSTR 37 -/-/100 i Pumped 0.5-10 j 3 --9-13 100-200 --0.056 4,080 Con. -- Set UASB/G -PVDF/Tub/0.03 Gas-lift --0.02-0.14 4 11-12 -1-2 6 -12 --Con.    generally considered to govern membrane fouling (Liu et al. 2012a). Dominant cake layer resistance has similarly been reported by a number of authors in AnMBR for municipal wastewater (Liu et al. 2012b;Martinez-Sosa et al. 2012;Ozgun et al. 2015b), which is therefore comparable to the general trend for AeMBR. Chu et al. (2005) reported that 90% of total resistance for an AnMBR corresponded to cake resistance while only 9% was internal fouling. Ozgun et al. (2015b) reported that the relative contribution of cake layer resistance in their AnMBR study decreased with a decrease in operating temperature, which suggests transients in the bulk particulate and colloidal characteristics will influence cake deposition. Resistance to filtration due to membrane fouling can also be classified as reversible, irreversible or irrecoverable, depending on whether it can be removed physically during operation (by relaxation or backflushing), chemically, or if it remains after chemical cleaning ( Judd 2011). In immersed AeMBRs, physical cleaning procedures like relaxation and backwashing have shown to be effective in extending membrane operation compared to continuous filtration reducing the chemical cleaning frequency (Judd 2011). By applying physical cleaning procedures, it is possible to apply fluxes that result in an increase of resistance to filtration as long as the cake layer deposited onto the membrane surface can be removed by relaxation or backflushing. For instance, operational cycles of 10 minutes of filtration followed by 1 minute relaxation at fluxes between 22.3 and 28.5 LMH resulted in fouling rates during filtration cycles of 1.39 and 1.8 mbar min À1 respectively, while the irreversible fouling rates were two orders of magnitude lower (Guglielmi et al. 2007). Similarly, in AnMBRs, relaxation and backwashing have also been applied and proved to be effective for membrane fouling mitigation (Tables 5 and 6). Smith et al. (2013) also showed that a 4 min backwash every 4 h reduced membrane fouling from 0.045 to below 0.001 mbar min À1 , with similar observations made for side-stream AnMBR . A four times decrease in fouling rate was observed following implementation of backwash at double the permeate flux, which indicates backwash flux specification is also important.
Once a tenacious cake layer is formed, it is hard to remove just with physical cleaning which is evidenced by cake layer growth on membranes exposed to high shear flow (Imasaka et al. 1993) or high shear in combination with relaxation and depressurization (Choo & Lee 1996). Permeability recovery can be variable following chemical cleaning in both aerobic and anaerobic MBRs, which is illustrative of irrecoverable fouling. Gouveia et al. (2015a) reported a chemical cleaning protocol for an AnMBR comprising 1,000 mg L À1 sodium hypochlorite (NaClO), which achieved recovery of only 61% of initial permeability. However, more typically only around 7-20% of the total resistance is regarded irrecoverable following incorporation of chemical cleaning policies in AnMBR (Chu et al. 2005;Ozgun et al. 2015b;Dong et al. 2016a).

ANAEROBIC MBR FOR INDUSTRIAL VERSUS MUNICIPAL WASTEWATER TREATMENT
Industrial wastewaters are generally of higher organic strength and comparatively smaller flows than for sewage (van Lier et al. 2008). Organics in industrial wastewater are predominantly soluble and readily assimilable (Liao et al. 2006); whereas sewage comprises a considerable particle fraction, with a proportion of the organics known to be recalcitrant . Due to the smaller flows and higher organic strength, higher methane yield (m 3 CH 4 m À3 permeate) in industrial applications is sufficient for heating the influent to mesophilic temperatures, which improves biodegradation kinetics, and enables application of higher organic loading rates than for sewage van Lier et al. 2008). The combination of a more complex organic feed composition (coarser particle fraction), comprising a large recalcitrant contribution, coupled with insufficient heat to improve biodegradation kinetics, implies that sewage presents the higher fouling potential. The higher methane yield indicates more electrical power is available to industrial AnMBR making   (Lens et al. 1998;Giménez et al. 2012). u Only consider the methane from AFBR. v Directly reported the energy production. w Literature reported the dissolved CH 4 (directly test or estimate from Henry's law).
x Use Henry's law to for dissolved CH4 calculation (assume saturation index is 1.00) (Giménez et al. 2012). sustainable membrane operation easier to achieve with adoption of high cross-flow velocities in sidestream (Lin et al. 2013), and the application of conventional CSTR configured AnMBR with high biomass concentrations (Skouteris et al. 2012) commonly reported that can still deliver energy neutral membrane operation. This practice is not readily translatable into sewage applications, subsequently demanding the development of engineered solutions to facilitate energy neutral operation (Ozgun et al. 2015a;Wang et al. 2018).

ESTABLISHING THE ENERGY PROFILE OF ANMBR FOR MUNICIPAL APPLICATION Towards energy neutral wastewater treatment
The key facet that drives commercial interest for AnMBR is in the potential to achieve energy neutral wastewater treatment through a reduction in net energy demand for treatment, coupled with an Figure 1 | Combined headspace and dissolved methane production rates from anaerobic MBR treating: (a) settled sewage and (b) crude sewage. Where dissolved methane data was not provided, the dissolved fraction was estimated using Henry's law. Energy data was normalised assuming 40% CHP conversion efficiency.  (Giménez et al. 2011;Robles et al. 2013aRobles et al. , 2013b. Downstream of the AnMBR, the decision to have biological or physical separation processes for polishing and nutrient removal is a further consideration Sutton et al. 2011). The merit of these various approaches is beyond the scope of this review. Nevertheless, it is implicit in each of these cases that the AnMBR itself needs to be energy neutral, or better, energy positive (to provide residual energy for the ancillary processes), if overall energy-neutrality is to be achieved. Typical headspace methane yield from the literature is between 0.02 and 0.27 L CH 4 . g COD À1 for both settled and crude wastewater (Table 7). This is below the expected stoichiometric conversion of 0.35 L CH 4 . g COD À1 . The difference between measured and expected values can be explained by: (i) the fact that yield does not take into consideration the solids which are retained (and accumulate) within the bioreactor (Uemura & Harada 2000;Lester et al. 2013); (ii) differences in HRT or OLR between literature studies which can influence specific yield, and (iii) the dissolved methane fraction which is released with the AnMBR effluent. Several studies have now demonstrated this dissolved fraction can comprise over 50% of the dissolved methane balance (Cookney et al. 2012), with losses exacerbated at lower temperature (Cookney et al. 2016). Downstream membrane technology has been demonstrated that can recover this fraction (Bandara et al. 2011;Cookney et al. 2016) in sufficient concentration for reuse in power generation. Collectively the average dissolved and headspace methane yield from the literature provides around 0.34 kWh m À3 of permeate produced (0.80 kWh m À3 maximum), which is the typical membrane operation energy consumption of fullscale aerobic MBR (0.19-0.70 kWh m À3 ). Consequently, AnMBR should be delivered at equivalent or better energetic performance to aerobic MBR to achieve 'energy-neutral' conditions. Comparison of specific energy production with membrane energy demand estimated from published AnMBR literature ( Figure 1) demonstrates that several AnMBR studies can achieve membrane operation within 'energy-neutral' conditions ( Figure 2). Since the shear promoted at laboratory scale is generally a conservative estimate of that attained at scale (Delgado et al. 2004), such evidence is encouraging. While  Several authors have now directly compared different AnMBR reactor configurations and have identified that when configured as a UASB, membrane fouling and energy demand is constrained (van Voorthuizen et al. 2008;Martin-Garcia et al. 2011). The authors ascribed the enhanced performance to the lower particle concentration within the downstream membrane tank (Martin- . Martin-  also screened a host of membrane geometries and were able to demonstrate a specific gas demand of 0.3 kWh m À3 for immersed hollow fibre membranes compared to 3.7 kWh m À3 for externally configured membranes. It has been shown that further reduction in specific energy demand can be attained for immersed membranes through inclusion of novel gas sparging strategies (Cerón-Vivas et al. 2012;Martin Garcia et al. 2013;Wang et al. 2018). Shin et al. (2016) introduced the anaerobic fluidised bed MBR which incorporates granulated activated carbon (GAC) into the reactor to scour the membrane. Significantly, the authors demonstrated sustainable fluxes could be achieved with an immersed membrane without the need for gas sparging, provided a critical upflow velocity could be achieved to sufficiently fluidise the GAC (Wang et al. 2017). Kim et al. (2011) reported a specific energy demand of 0.028 kWh m À3 for this configuration, which evidences a substantive energy reduction when compared to classically commercially configured aerobic MBR technology. One of the future challenges with GAC is to identify a compatible membrane that can withstand the abrasion introduced by the GAC (Shin et al. 2016;Wang et al. 2017). Ruigómez et al. (2016aRuigómez et al. ( , 2016b integrated rotating membrane technology into a CSTR configuration as an alternative method of shear (Zsirai et al. 2016). The authors demonstrated an improvement in critical flux with membrane rotation when compared to conventional gas sparging; specific energy demand for disk rotation is around 0.104 kWh m À3 .
Within the context of scaling-up AnMBR, membrane cost must also be considered as the sustainable fluxes identified at present of around 10 L m À2 h À1 (Table 1) are markedly below those nominally achieved in commercial aerobic MBR. For large scale installations, investment decisions are strongly influenced by initial capital investment. As such, there is a need to establish new paradigms that can access cost efficient water productivities whilst still delivering to the energy neutral objective. One innovative example is anaerobic dynamic membrane bioreactors (Alibardi et al. 2014;Ersahin et al. 2016) which employ woven and non-woven filter substrates as a low cost alternative (less than 13€ m À2 (Ersahin et al. 2016)) to conventional polymeric membranes. Mesh sizes of 10-200 μm have been tested which provide substantial clean water permeability. The developed cake provides the rejection (Alibardi et al. 2014) with .99% COD rejection noted for several prospective applications, coupled with sustainable fluxes, which evidences their considerable potential to making AnMBR both capitally and energetically efficient (Alibardi et al. 2014;Ersahin et al. 2016).

CONCLUSIONS
The present literature review reveals significant differences with respect to biomass characteristics and fouling behavior between aerobic and anaerobic MBRs which can be summarized as follows: 1. The ratio of proteins to carbohydrates in eEPS is higher in AnMBRs than in AeMBRs. However, the total eEPS appears slightly higher in aerobic MBR. There is a general lack of knowledge regarding the relationship between eEPS concentration and composition and surface properties such as charge and hydrophobicity for AnMBR.
2. Although similar median particle sizes have been reported in aerobic and anaerobic MBRs, the presence of a population of fine solids with particle sizes ranging from 1 to 10-15 microns have been widely reported in the latter and associated to biomass of high fouling propensity. 3. The SMP COD concentration contained within bulk sludge is an order of magnitude higher in AnMBR than in AeMBR. The main operational parameters that have shown to enhance SMP production in AnMBRs are low temperature and extended SRT. 4. The effect of turbulent gas sparging on membrane performance in immersed AnMBRs seems to be limited as compared to aerobic systems, indicating that fouling is more determined by sludge properties than by membrane operational conditions 5. AnMBR fluxes are between one-third and one-half of those reported in AeMBRs. Reported permeabilities in AnMBR are around 50% below those of AeMBR and employ specific gas demands between 50 and 300% higher. 6. Although as with aerobic systems the predominant fouling mechanism in AnMBRs has been reported to be cake filtration, contradictory results with respect to the effectiveness of membrane backwashing at reducing membrane fouling and permeability recovery after chemical cleaning have been reported and would require further research. 7. Whilst membrane fouling remains a critical challenge, several research groups have already demonstrated the potential to achieve energy self-sustained conditions at pilot scale. Whilst AeMBR is comparatively mature, continued investment in commercial module and aeration (gas scouring) engineering is still enabling radical reductions in specific energy demand to be realised at scale. Consequently, the potential to achieve energy neutral wastewater treatment at a commercial scale appears inherently viable.