The reactors were seeded with approximately one-third of the reactor’s volume as seed sludge. The VSS had an initial concentration of 7.44 gVSS.L^-1. The sludge of both reactors was adapted for approximately 40 days (Phase I) with an OLR of 2 kg COD. m^-3.d^-1. After this period, the OLR was stepwise increased after COD removal efficiencies above 90% were achieved. Figures 1A, B show the COD removal efficiencies of the AnMBR 1 and 2, respectively, along their operation. As the AnMBR effluent samples passed through the membrane with a nominal pore size of 30 nm (i.e., < 0.45 μm pore size for filtered samples), it was not necessary to perform soluble COD analysis. Thus, from a technological point of view, the reactors were compared in terms of total COD.

The COD removal efficiency in both reactors remained above 95% except during days 40 and 60 when the first OLR increase was applied, with concomitant bicarbonate reduction. The OLR was increased according to Table 2. Studies reporting the anaerobic treatment of vinasse using other types of reactors obtained lower COD removal efficiencies working with similar or higher OLR. Aquino et al. (2017) reported a COD removal efficiency of 86% ± 3.2% treating sugarcane vinasse in a fixed-bed anaerobic reactor at OLR of 5.5 kg COD.m^-3.d^-1 and 84% ± 1.3% at OLR of 10.2 kg COD.m^-3.d^-1. In another study, Fuess et al. (2018) performed sugarcane vinasse treatment in an acidogenic reactor followed by an upflow anaerobic sludge blanket reactor (UASB). OLRs of 15, 20, and 25 kg COD. m^-3.d^-1 were applied and COD removal efficiencies of up to 70% were achieved. Mota et al. (2013) and Santos et al. (2017) used two-stage submerged anaerobic membrane bioreactor for the treatment of sugarcanne vinasse. The studies reported similar COD removal efficiencies of approximately 97% at OLRs of 2.5 and 6 g COD.L^-1.d^-1, respectively.

Sulfate removal was observed in both reactors, indicating the occurrence of sulfate reduction. Figures 2A, B show the sulfate removal during the operation of AnMBR1 and 2. The sulfate removal was higher than 70%, surpassing 95% during the last phases of operation. In the present study, the applied COD/SO₄ ^2- ratio was ≈10 and high COD and high sulfate removal efficiencies were obtained unlike what was reported in previous studies (Silvia et al., 2010). Quantification of the H₂S content in the biogas was not possible due to a lack of analytical equipment.

Sulfate reducing bacteria (SRB) reduce sulfate to sulfide that can dissolve in the effluent (HS⁻, S^2-, H₂S) or becomes part of the biogas (H₂S) (Jimenez et al., 2017). Sulfide generation was monitored on the permeate since Phase II. It is assumed that the sulfide concentration in the reactor was the same as in the permeate as it is not anaerobically converted. The sulfide concentration observed in AnMBR1 was 96.2 ± 20.6 mg.L^-1, 104 ± 0.8 mg.L^-1 and 103.2 ± 0.01 mg.L^-1 during Phases II, III and IV, respectively. In AnMBR2 the sulfide concentrations were 127 ± 0.4 mg.L^-1, 81 ± 23.7 mg.L^-1 and 52.4 ± 12 mg.L^-1, at Phases II, III and IV, respectively.

The presence of sulfate stimulates SRB populations, reducing the methanogenesis because of competition for substrates (Fuess and Garcia, 2015). Methanogenesis is affected by the COD/SO₄ ^2- ratio (Sarti and Zaiat, 2011); at low ratios sulfate reduction predominates, while methanogenesis prevails at higher values (Madden et al., 2014). In the present study, with a COD/SO₄ ^2- ratio of 10, sulfide was detected in the permeate; however, the COD removal efficiency was not affected. Different threshold concentrations for H₂S inhibition in literature can be found as the sulfide inhibition in anaerobic reactors depends on several factors such as pH, substrate characteristics, COD/SO₄ ^2- ratio, presence of metals, organic and hydraulic load, sludge type and biomass acclimation (Turker and Dereli, 2021).

VFA was monitored weekly. The results showed a similar trend as the COD concentrations. A maximum of 90% reduction in VFA concentration was observed in the permeate of the reactors (Supplementary Figure S1). Acetic acid accumulation was observed in AnMBR1 and AnMBR2 during the adaptation phase and the first OLR increase indicating an impairment of the methanogenic activity (Supplementary Figure S2). Propionic acid accumulation was observed in the AnMBR2 during the first days of operation and it could be related to an unbalance in the acetogenesis step, which is expected to occur at the start-up phase of an anaerobic process. During the first increase in OLR (days 40–60) the VFA removal efficiencies dropped due to acetic and propionic acid accumulation. As discussed previously, the increase in OLR and the decrease in bicarbonate supply likely caused stress in the sludge. It took approximately 20 days to get both reactors stabilized again. Interestingly, during subsequent OLR increases, the reactor performance remained stable and no VFA accumulation was observed which could be related to the adaptation and/or selection of the microbial community. In subsequent periods including OLR increments, the VFA concentration in the permeate remained below 250 mg. L^-1 with VFA concentrations reaching almost 0 mg. L^-1, mainly after the Phase II.

During the operation of both reactors, the pH of the feed, permeate and the reactors’ matrix were continuously monitored (Supplementary Figure S3). Due to the low pH of the vinasse, NaHCO₃ was used to neutralize the influent during the acclimation period at a ratio of 0.7 g bicarbonate. g^-1COD. After this period, the alkalinity was weekly measured (Supplementary Table S2) and the ratio was decreased to 0.3 g bicarbonate. g^-1COD and finally NaHCO₃ was completely removed from the influent. The pH in the reactor was maintained approximately at 7.0 during the whole operational period. Even when bicarbonate was no longer supplied, the reactor pH remained stable at neutral values.

The VSS concentration in both reactors was monitored during the reactors’ operation (Supplementary Figure S4). No sludge disposal was performed during the operation with the SRT being approximately equal to the operating time. The VSS concentration gradually increased in both reactors after the adaptation period and the first OLR increase, indicating sludge growth. At the end of the operation the VSS concentration in the AnMBR1 was 39 ± 0.03 g VSS. L^-1 and in the AnMBR2 was 49 ± 0.02 g VSS.L^-1. AnMBR2 presented higher VSS concentrations than AnMBR1, which might be attributable to the fact that 70% of the solids in the feed of AnMBR1 were reduced after centrifugation of the influent vinasse. Apparently, another (non-active) fraction of organic solids was accumulating in AnMBR2.

The biogas production and composition measured during the different stages of the reactor’s operation can be found in the (Supplementary Table S3). The biogas production was very similar in both reactors except in Phase IV when AnMBR2 presented lower biogas (12 ± 6 L d^-1) production than AnMBR1 (27 L d^-1), probably due to the operation issues observed during that phase (Section 3.2.1). The methane yield observed remained around 0.28 CH₄ L. g COD_removed ^-1 in AnMBR1 and 0.24 CH₄ L. g COD_removed ^-1 in AnMBR2.

Figure 3 shows membrane flux monitored during the reactor’s operation. Other membrane parameters can be found in the Supplementary Figure S5.

The flux of the AnMBR2 began to decline after Phase III, at a flux of 6 LMH, as illustrated in Figure 3. Notably, filtration resistance (Supplementary Figure S5) started to rise on day 80, concomitant with the flux increase to 6 LMH. The resistance continued increasing despite the chemical cleaning procedure before the flux was increased to 8 LMH to restore membrane permeability, leading to complete membrane clogging during Phase IV. The flux was decreased to 4 LMH and the crossflow velocity was increased from 1 m s^-1–2 m s^-1 after performing the membrane cleaning procedure. However, despite the cleaning procedures, the membrane permeability did not recover (Supplementary Figure S5), the flux was not recovered and the reactor showed process instability and subsequent membrane clogging. The observed decreased performance of the membrane prompted the decision to cease the reactor operation. It's noteworthy that COD removal efficiencies remained unaffected by these membrane-related operational issues during that period, as depicted in Figure 1B.

AnMBR1’s flux started to decrease after the increase to 8 LMH, in Phase IV (Figure 3). The highest values for the filtration resistance were observed on day 120 (Supplementary Figure S5). Subsequently, a chemical cleaning procedure was performed, the CFV was increased to 2 m s^-1, and the system was restarted at a flux of 4 LMH followed by an incrementally raise to 8 LMH. A tentative increase to 10 LMH resulted in the membrane clogging. The use of centrifuged vinasse showed a better filtration performance as it contained fewer solids that contribute to the membrane fouling.

Figure 4 shows the microbial composition at genera level of the inoculum and sludge samples collected from the AnMBR 1 and 2 at the end of Phases I and III. Phase I was the adaptation phase, in which the inoculum had the first contact with the vinasse. Phase III was chosen to be analyzed, because it was the phase in which the highest removal efficiencies (95%) were achieved, and the membrane flux could be maintained (6 LMH).

The inoculum presented high relative abundance of Methanosaeta, unidentified-SBR1031 and Methanobacterium. Methanosaeta is an acetate-converting methane-producing archaea and one of the key populations in methanogenic bioreactors. Most of the COD is eventually converted via acetate to methane, agreeing with the pre-dominant relative abundance of the acetoclastic Methanosaeta in the AnMBR sludge of our present study. On the other hand, species belonging to the genus Methanobacterium are hydrogenotrophic methanogens that utilize H₂ and CO₂ as substrates for methane production (Whitman, et al., 2014). Their ability to grow autotrophically under strict anaerobic conditions makes them integral to the hydrogenotrophic methane pathway, complementing the acetoclastic pathway of Methanosaeta. The unidentified SBR1031 group appears to play a role in acetogenic dehydrogenation, potentially using ethanol as a carbon source (Xia et al., 2016).

After the adaptation period, the relative abundance of both unidentified SBR1031 and Methanobacterium decreased in both reactors, while the relative abundance of Methanosaeta decreased only in AnMBR1, suggesting distinct community dynamics between the reactors. Aditionally, in AnMBR1, other microbial populations of interest emerged: Mesotoga, Macellibacterioides, unidentified Spirochaetaceae, and Acetobacterium.

Mesotoga, a mesophilic acetogen, is common in hydrocarbon-rich anaerobic environments. Besides being mesophilic, Mesotoga displays lineage-specific phenotypes, such as no or little H₂ production and dependence on sulfur-compound reduction, which may influence its ecological role (Nesbø et al., 2019). In this sense, Mesotaga may have influenced the system by reducing sulfur compounds rather than producing H₂, which could impact electron flow in the bioreactor. Mesotoga was found in all the other AnMBR sludge samples analyzed in this study. Macellibacteroides, a fermentative bacterium, produces lactic, acetic, and butyric acids as fermentation products, contributing to the pool of substrates available for methanogenesis (Jabari et al., 2012).

Acetobacterium plays an important role as an acetogen, reducing CO₂ to acetate and other multiple-carbon compounds using H₂ as an electron donor, further supporting the acetate-driven methane production pathway (Balch et al., 1977). Interestingly, after adaptation, AnMBR1 exhibited a high relative abundance of Candidatus Methanoplasma, an obligate anaerobe that produces methane via methylotrophic metabolism, relying on methyl donors such as methanol and monomethylamine and hydrogen dependent for growth (Lang et al., 2015).

The genera unidentified Spirochaetaceae was found in both AnMBRs after adaptation. The family Spirochaetaceae comprises a wide range of phylogenetically diverse genera, including Alkalispirochaeta, Marispirochaeta, Ocean ispirochaeta, Pleomorphochaeta, Rectinema, Salinispira, Sediminispirochaeta, Sphaerochaeta, Spirochaeta, Spironema, Treponema, and several genera containing uncultured species. Among these, Sphaerochaeta was detected in both AnMBRs, though with higher relative abundance in AnMBR2.

Sphaerochaeta comprises free-living, chemotrophic, organotrophic, anaerobic, mesophilic, and neutrophilic bacteria with fermentative metabolism. These bacteria have been isolated from various anaerobic environments, such as freshwater sediments, termite hindguts, and methanogenic consortia (Bidzhieva et al., 2020). Fermentative growth is observed with carbohydrates including pentose and hexose monosaccharides, disaccharides, and soluble starch. End products of glucose fermentation include acetate, formate, and ethanol and growth is stimulated by yeast extract (Ritalahti et al., 2012), which could explain its higher abundance in AnMBR2 treating non-centrifuged vinasse. This suggests that Sphaerochaeta plays an important role in breaking down complex organic matter in the system, contributing to the pool of fermentation products that feed into methane production. The presence of this genus, alongside other fermentative and methanogenic organisms, highlights its role in the overall process of anaerobic digestion by maintaining metabolic flexibility and supporting the conversion of organic matter into biogas.

An increased relative abundance of the genera Methanosaeta, Sphaerochaeta, and Mesotoga was observed in both reactors in the samples collected in Phase III. The relative abundance of other genera decreased or remained the same. This community composition highlights a complex interplay between acetoclastic, hydrogenotrophic, and methylotrophic methanogens, alongside acetogens and fermentative bacteria, each contributing to the overall stability and efficiency of the anaerobic digestion process. The obtained results showed that the communities developed in both reactors are capable of degrading beet vinasse. Similar microorganisms were found in both reactors, although in different relative abundances.

The reactor was inoculated with 3.4 L of the sludge from the reactor treating centrifuged vinasse from AnMBR1 (39 gVSS.L^-1). Resulting in an initial VSS concentration of approximately 22 g.L^-1. The AnMBR2 was fed with the centrifuged beet vinasse. The specific methanogenic activity (SMA) of the sludge used to inoculate each reactor was 0.25 g COD-CH₄. g^-1VSS.d^-1. As the reactor had an initial solid concentration of 22 g VSS. L^-1, the expected COD conversion rate was 36 g COD.d^-1, which corresponded to an OLR of 5.5 g COD. L^-1d^-1. It was then decided to start the reactor with an OLR below that value and the OLR was stepwise increased after COD removal efficiencies above 90% were achieved, according to Table 3.

The COD removal efficiency was above 95% from day 0 to day 97 of operation (Figure 5A). On day 98 the COD removal efficiency started to be decreased after the OLR was increased to 10 kg COD. m^-3.d^-1 and flux to 12 LMH. The COD removal efficiency decreased to its lowest value (62%) on day 110 and then started to increase again, reaching 80% at the end of the reactor’s operation.

Published research using other reactor configurations for the sugarcane vinasse treatment resulted in lower or equal removal efficiencies at the same OLR as applied in this study (Aquino et al., 2017; Fuess et al., 2018). However, the current studies reporting the use of AnMBR for the treatment of sugarcane vinasse did not reached OLRs higher than 6 g COD.L^-1.d^-1 (Mota et al., 2013; Santos et al., 2017). Turker and Dereli, (2021) investigated the performance of a pilot scale AnMBR treating beet molasses based industrial wastewater. The reactor achieved a COD removal efficiency between 48% and 92% up to a volumetric load of 10 g COD.L^-1.d^-1. When the system operated at an OLR of 10 g COD.L^-1.d^-1 the COD removal was approximately 60%.

Biogas production and composition obtained in each phase of the second experiment are presented in Supplementary Table S4. The results are similar to those obtained in the first experiment. Biogas production increased while the CH₄ portion decreased with the OLR increasing. The low methane content in the biogas produced from vinasse anaerobic degradation is primarily attributed to its complex composition (Fuess and Garcia, 2015; Parsaee et al., 2019) which promotes high acidification in the reactor, especially in a one-stage reactor configuration. The organic acids (normally produced at higher rates by fermenting bacteria than consumed by methanogenic archaea) lower the medium pH and might inhibit the methanogenesis, (Silva et al., 2020). This occurs especially because the AnMBR used in this study is a one-stage (methanogenic) reactor. Two-stage anaerobic reactors provide better conditions for the separate development of acidogenic and methanogenic microorganisms, especially in the treatment of high OLR and low pH effluents, such as vinasse in which acidifying conditions can be clearly separated from the methanogenic ones (Fuess et al., 2017; Santos et al., 2017).

Sulfate removal efficiencies (Figure 5B) remained above 95% throughout the entire operation, except on days 90–105, when the removal efficiency decreased to 90%, probably due to the OLR increase. Mota et al. (2013) and Santos et al. (2017) obtained sulfate removal efficiencies of approximately 87%, using a two-stage membrane anaerobic reactor at OLR lower than the one applied in this study.

VFA were monitored weekly and showed a similar trend as the COD (Supplementary Figure S6). A VFA removal efficiency of 90% was observed in the permeate when compared to the inlet of the AnMBR. The exception occurred after the OLR increased to 10 kg COD. m^-3. d^-1 in phases IV and V (from day 98–119), when mainly acetic and propionic acid were accumulated, indicating an unbalance in the acetogenic and methanogenic steps. In general, the VFA concentration in the permeate remained below 100 mg. L^-1 in almost the entire reactor’s operation.

Different strategies to achieve higher fluxes were tested, such as the increase in CFV and the decrease in VSS concentration in the reactor. Figure 6A shows the concentration of solids in the reactor during operation. In Figure 6B, the theoretical flux (flux applied in the system) and the measured flux are compared. Figure 6C shows the variation in TMP according to the membrane flux. Figure 6D illustrates how the permeability and TMP correlated during reactor operation.

In Phases I (days 1–14) and II (days 15–32), a flux of 6 and 8 LMH were applied, respectively. There were no problems maintaining the intended flux in the membrane (Figure 6B). However, when the theorical flux was increased to 10 LMH in phase III, a decrease in the flux and an increase in the TMP were observed. Therefore, on day 39, it was decided to test the effect of increasing the CFV from 1 to 2 m/s. However, there were no improvements observed in the filtration process. Hereafter, it was decided to reduce the VSS concentration in the reactor from 38 to 21 g VSS.L^-1 on day 72 (Figure 6A). Despite a decrease in the TMP, there were no improvements in the flux. As the COD removal efficiency remained high (≈ 95%), it was decided to further reduce the solids concentration to 17 g VSS. L^-1 (day 93). Due to technical complications, the CFV was reduced from 2 to 1.5 m/s. Those changes did not affect the COD removal efficiency (Figure 5A) and the eventual measured flux achieved 10 LMH. The TMP reduced significantly due to the decrease in VSS concentration and the membrane cleaning procedures (Figure 6C).

On day 98, the membrane flux was then increased to 12 LMH; however, a decrease in the COD removal efficiency was observed. Nevertheless, on day 110, the flux was further increased to 14 LMH (Phase V) since the main objective of this experiment was to evaluate the filtration efficiency of the membrane at higher fluxes despite the bioconversion performance.

The sludge was continuously wasted to maintain the concentration between 16 and 17 g VSS.L^-1 to keep the TMP below 1,000 mbar (Figure 6C). The OLR to the reactor did not change despite the increase in the flux due to a lower concentration of the vinasse used in comparison to the vinasse used in Phase IV (Table 3). At this stage, the results showed that a flux of 14 LMH could be maintained; however, the COD removal efficiency dropped to values close to 60%. Using 2S-AnMBR with submerged membrane Mota et al. (2013) and Santos et al. (2017) applied flux of 4.4 and 5.1 LMH, respectively. In other study, Mota et al. (2019) evaluated the anaerobic treatment of a synthetic acidified effluent using an AnMBR with the same configuration as the present study with a flux set as 9.8 LMH. Flux of 14 LMH was also achieved using an AnMBR consisting of an upflow anaerobic bioreactor coupled to two side-stream ultrafiltration membrane modules connected in parallel treating food wastewater (Ariunbaatar et al., 2021). However, no studies about vinasse treatment in AnMBR in laboratory scale have been published yet applying flux higher than 10 LMH.

It was then decided not to further increase the membrane flux so that the associated OLR would not cause the reactor to collapse. In the last days of operation, a trend towards increased COD removal efficiency was observed. It was not possible to determine whether the efficiencies would again reach similar values as those observed until phase III (96%). However, most likely after a longer period of operation under those conditions, the reactor would have shown higher efficiencies.

The results show that reducing the VSS concentration to 16–17 g/L is essential for reducing the TMP, which leads to higher permeability (Figure 6D) and consequently, the maintenance of the desired flux.