Two pairs of reactors with continuous gas recirculation were assembled for this study. One pair consisted of stirred tank reactors (STR-control and STR-test) and the other of bubble column reactors (BCR-control and BCR-test). Assembly and configuration of the STRs were described in detail by Baleeiro et al. (2021b). BCRs were assembled and operated with the following differences in relation to the STRs: 1) the BCR vessels consisted of bubble columns made of glass with a working volume of 1.2 L each; 2) the systems had no oxidation-reduction potential (ORP) monitoring; 3) pH monitoring and correction was done manually three times a week; 4) temperature regulation was carried out via the water jacket of the vessels and a thermal bath; 5) gas recirculation was carried out with micro-diaphragm gas pumps NMP 830 (KNF Neuberger GmbH, Freiburg, Germany) at a flow of ca. 1.5 L min⁻¹; and 6) an internal vertical hollow glass with holes of 1–2 mm was used to bubble the gas into the broth. In all other aspects, BCRs were operated similarly to STRs. The reactors were operated at 32°C, at pH values of 6.0 ± 0.1 (STRs) and 6.1 ± 0.3 (BCRs), and with a hydraulic retention time (HRT) of 14 days. The basal medium contained 133 mM lactate (12 g L⁻¹) and 200 mM acetate (12 g L⁻¹) as organic carbon sources and the gas reservoirs were periodically refilled with 10 L H₂:CO₂ (80:20), 240 mL ethylene (methanogenesis inhibitor), and 120 mL He (tracer gas). Basal medium composition, reactor start-up and reactor operation were identical for STRs and BCRs and were as described by Baleeiro et al. (2021b), except that lactate was fed ten times per day along with the basal medium.

Figure 1 shows the two reactor types used in the study and the timelines with the oxygen contamination events. The four reactors were inoculated with the broth harvested from the H₂/CO₂/ethylene recirculation reactors described by Baleeiro et al. (2021b). For STRs, the broths of the two reactors were mixed before the start of the experiment and before the second comparison period to ensure that both reactors started with identical microbial and chemical compositions for each comparison period. The same was done with the broths of the BCRs. Startup and broth mixing of STR-test and STR-control occurred on operation days 0–27 and the second broth mixture occurred on days 111–115 (Figure 1A). Startup and acclimatization of the community to a BCR occurred during the 86 days preceding the start of this study. Mixing of the BCR-control and BCR-test broths occurred on days 0–4 and 36–39 (Figure 1B). The broths were mixed without opening the reactors using Hei-Flow Precision peristaltic pumps (Heidolph Instruments GmbH, Schwabach, Germany), PVC tubes Tygon^® LMT-55, and three-way valves. The major air contamination events were detected in STR-test on days 27–59 and 115–119 and in BCR-test on days 11–36 and 39–50 (Figure 1).

The reactors that remained air-tight all the time were used as control reactors. Reactors STR-test and BCR-test presented detectable air intrusion at certain periods due to imperfect air-tight conditions. With the help of a H₂ leak detector (GLD-100, Coy Laboratory Products, Grass Lake, USA), tightness problems that caused gas to leak out of the reactor systems were located and promptly solved. The same was not true for tightness problems that only allowed air to leak into the system and that were solved by trial and error interventions.

High performance liquid chromatography with a refractive index detector (HPLC Prominence-i RID, Shimadzu Europa GmbH, Duisburg, Germany) was operated under the conditions described by Apelt (2020) with adaptations described by Baleeiro et al. (2021a) for monitoring of the following chemicals in the aqueous phase: formate, acetate, ethanol, lactate, propionate, n-propanol, n-butyrate, i-butyrate, n-butanol, n-valerate, i-valerate, n-pentanol, n-caproate, i-caproate, n-hexanol, n-heptanoate, and n-caprylate. Biomass concentration was determined by measuring optical density at 600 nm (spectrophotometer Genesys 10 S, Thermo Scientific Inc., Waltham, United States). Conversion factors for optical density at 600 nm and biomass are described previously (Baleeiro et al., 2021a). In the gas phase, H₂, CO₂, CH₄, He, O₂, N₂, and ethylene were monitored by gas chromatography as described by Logroño et al. (2020).

Microbial communities were analyzed by 16S rRNA gene amplicon sequencing with taxonomic assignments done with the SILVA 138 reference database (Yilmaz et al., 2014). DNA extraction, PCR, and library preparation for Illumina MiSeq sequencing were described previously (Baleeiro et al., 2021a). Trimming, filtering, and denoising of amplicon data as well as visualization of microbiome census data and Spearman correlations were done as described by Baleeiro et al. (2021a). All samples were rarified to an equal depth according to the sample with the lowest read number in the dataset (4,977 counts). Raw sequence data for this study was deposited at the European Nucleotide Archive (ENA) under the study accession number PRJEB44209 (http://www.ebi.ac.uk/ena/data/view/PRJEB44209).

Assumptions and calculation steps used for the component balances in the gas recirculation reactors were described previously (Baleeiro et al., 2021b). Direct measurements of O₂ concentration (such as by using dissolved oxygen probes) are not adequate to monitor micro-aerobic environments (Krayzelova et al., 2015). Therefore, O₂ contamination was determined indirectly via the N₂ concentration in the recirculating gas, using the N₂:O₂ ratio in air of 3.73 according to the following equation: O 2   contamination   rate  [ mL d − 1 ] =   ( Y 2 V g a s   2 − Y 1 V g a s   1 ) [ 3.73   ( t 2 − t 1 ) ] (1) for sampling points 1 and 2: Y is the volumetric fraction of N₂ in the gas phase; V _gas is the gas volume of the system in mL; and t is the sampling time in days. Eq. 1 holds true if no N₂ is formed or consumed in the reactor, if O₂ is below the detection limit, and if the increase in the system’s gas volume due to air leaking in can be neglected. Calculations of contamination rates in the control reactors (STR-control and BCR-control) were used to determine standard errors for the O₂ contamination rate (Supplementary Figure S1). ORP measurements did not show a clear relation with O₂ contamination rates (Supplementary Figure S2) and were, hence, not used to quantify rates.

The main carboxylates produced were propionate, n-butyrate, i-butyrate, n-valerate, n-caproate, and n-caprylate, with maximum concentrations presented in Table 1. The STRs reached higher n-caproate maxima and were the only ones in which n-caprylate was detected, whereas the BCRs reached higher propionate, n-valerate, and i-butyrate maxima. These differences can also be seen in terms of specific production and consumption rates given in mmol L⁻¹ d⁻¹ in Supplementary Table S1. For all reactors, most of the lactate fed was consumed and differences in its consumption rates were due to washout of unconsumed substrate (Supplementary Table S1). Except for the period when O₂ was present in detectable amounts, net consumption of acetate occurred in all reactors, ranging from 1.4 to 4.4 mmol L⁻¹ d⁻¹, which corresponded to a small fraction of the total acetate fed (14.3 mmol L⁻¹ d⁻¹). With the exception of STR-control, where very little i-butyrate was produced, i-butyrate production rates ranged from 0.23 to 1.08 mmol L⁻¹ d⁻¹ and showed no clear relation to O₂ contamination (Supplementary Table S1).

According to previous experience with the H₂/CO₂ recirculation reactors, ethylene was used to inhibit CH₄ production. Even though the partial pressure of ethylene was higher than 1 kPa at all times, methanogens gradually acclimatized to the inhibitor. Methanogenesis was observed first in the reactors that remained anoxic throughout the experimental time: in BCR-control from day 0 and in STR-control from day 31. Later on, methanogenesis was also observed in STR-test from day 48 and in BCR-test from day 60. Methane production rates were similar in the control reactors STR-control and BCR-control (16.5 and 15.9 mmol L⁻¹ d⁻¹, respectively) and the highest rate observed over a sustained period was 19.5 mmol L⁻¹ d⁻¹ during an anoxic operation period between days 59 and 111 of STR-test (Supplementary Table S1).

With 200 mM acetate (12 g L⁻¹) originally present in the growth medium, no net acetate formation was observed and no clear signs of homoacetogenic activity were found. After discounting the H₂ used for CH₄ formation (assuming 1 mol CH₄ produced from 4 mol H₂), almost no additional H₂ consumption was seen in the control reactors STR-control and BCR-control. STR-control had a net H₂ formation of 3.73 mmol L⁻¹ d⁻¹, whereas BCR-control showed a net H₂ consumption of 0.50 mmol L⁻¹ d⁻¹ (Supplementary Table S1). In the periods with O₂ contamination, H₂ consumption rates remained relatively high, despite low methanogenic activity. STR-test showed additional H₂ consumption of 26.3 mmol L⁻¹ d⁻¹ between days 27 and 59 and of 64.7 mmol L⁻¹ d⁻¹ between days 115 and 119 (Supplementary Table S1). This consumption of H₂ corresponded from 3.0 to 3.4 times the molar consumption of O₂ in the same period. In aerated periods in BCR-test, H₂ consumption after discounting methane production ranged from 1.1 to 3.6 times the oxygen consumption.

Electron balances encompassing the whole period of fermentation had errors of -0.63% (i.e., 0.63% of the monitored pool of electron equivalents had unexplained consumption) for STR-test, −0.73% for STR-control, −0.93% for BCR-control, and −1.61% for BCR-test.

Figure 2 shows the profiles of the cumulated amounts of carboxylates and O₂ as well as the microbial community composition at class level for the first period comparing the STRs. Between days 27 and 59, O₂ concentration in the gas phase remained below the detection limit in STR-test (Supplementary Figure S3), although a contamination rate of 220 ± 33 mL O₂ L⁻¹ d⁻¹ was detected. Between days 59 and 111, STR-test had an anoxic operation period although small O₂ contaminations occurred between days 104 and 111 (Figure 2) and were the reason why the reactor could not reach perfectly anoxic conditions (21 ± 33 mL O₂ L⁻¹ d⁻¹).

Even though n-butyrate and n-caproate were the main carboxylates produced in both reactors, n-caproate production in STR-test was, with 0.56 mmol L⁻¹ d⁻¹, 72% lower than in STR-control (2.12 mmol L⁻¹ d⁻¹) during the contamination period (days 27–59, Figure 2). Under O₂ stress, STR-test produced more propionate (0.76 mmol L⁻¹ d⁻¹, 6.4 times that of the control) but n-butyrate production was similar in both reactors. Moreover, O₂ contamination in STR-test caused 63% less methane production (6.05 of 16.5 mmol L⁻¹ d⁻¹) and 38% less acetate consumption (1.80 of 2.91 mmol L⁻¹ d⁻¹, Supplementary Table S1). After anoxic conditions in STR-test had been restored (days 59–111), methane production was 18% higher than in the control reactor (19.5 of 16.5 mmol L⁻¹ d⁻¹) and propionate production decreased slightly to 0.67 mmol L⁻¹ d⁻¹. Under anoxic conditions in STR-test, propionate production slowed down from day 66 on, coinciding with an acceleration of n-caproate production (Figure 2), which was still 16% lower than in the control (1.78 of 2.12 mmol L⁻¹ d⁻¹).

O₂ contamination caused differences in microbial community composition that were visible up to the class level (Figure 2). Clostridia and Methanobacteria predominated in the reactor that remained completely anoxic. In the other reactor, Actinobacteria and Coriobacteriia were the main classes found in periods when O₂ contamination occurred. The community of STR-test converged to a composition similar to that of the control reactor after anoxic conditions had been restored.

For the same operation period, community composition with resolution to the genus level and concentrations of the main carboxylates are shown in Supplementary Figure S4. Clostridia were mainly represented by the genera Caproiciproducens, Clostridium sensu stricto 12, and Oscillibacter, while the Actinobacteria belonged to the genera Acidipropionibacterium and Actinomyces. Methanogens were from the genus Methanobrevibacter and the Coriobacteriia were from the family Eggerthellaceae.

A second O₂ contamination event occurred in STR-test between days 115 and 119 (Figure 3), this time for a shorter period but at double rate (471 ± 33 mL O₂ L⁻¹ d⁻¹) (Supplementary Table S1). With the new contamination event in STR-test, CH₄ and n-caproate production rates fell to 32% (5.33 of 16.5 mmol L⁻¹ d⁻¹) and 50% (1.06 of 2.12 mmol L⁻¹ d⁻¹) of those of the control reactor, respectively (Supplementary Table S1). This time, the O₂ contamination coincided with an increase in n-butyrate formation rate from 1.51 to 2.76 mmol L⁻¹ d⁻¹, and instead of increasing propionate production, n-valerate production reached four times that of the control reactor (0.29 in relation to 0.07 mmol L⁻¹ d⁻¹). The short period of intense O₂ contamination had a smaller impact on the microbial community composition and coincided with an increase in relative abundance of Prevotella belonging to the Bacteroidia (Supplementary Figure S5). In the last 20 days of operation of STR-test, O₂ contamination was reduced but not completely stopped (39 ± 33 mL O₂ L⁻¹ d⁻¹, Supplementary Table S1) as shown between days 125 and 128 (Figure 3). In this period, propionate production decreased to its lowest value in STR-test and n-butyrate production increased once more to 4.77 mmol L⁻¹ d⁻¹. Although CH₄ formation increased again to the level observed during anoxic operation, n-caproate production could not be restored (Figure 3).

The earlier operation phase of the BCRs is shown in Figure 4 and in Supplementary Figure S6. For this comparison period, the mixing of broths was not enough to ensure equal community compositions in BCR-test and BCR-control. The O₂ contamination period between days 11 and 36 in BCR-test differed from the other contamination events since O₂ concentrations up to 18% were detected in the gas phase between days 11 and 28 (Supplementary Figure S3). Considering that the assumptions for balance calculations in the reactors do not account for high O₂ concentrations, the O₂ contamination rate determined for this period (97 ± 28 mL O₂ L⁻¹ d⁻¹) might be inaccurate. When O₂ was detected in the system, the microbial community in the BCR-test showed a strong dominance of Rummeliibacillus (Supplementary Figure S6) accompanied by Actinobacteria and Gammaproteobacteria (Figure 4).

After mixing the broths of the bubble columns to start a new comparison period, BCR-test received 129 ± 28 mL O₂ L⁻¹ d⁻¹ between days 39 and 50 whereas BCR-control remained completely anoxic (Figure 5). During the contamination period, BCR-test produced virtually no n-caproate (0.03 mmol L⁻¹ d⁻¹ in comparison to 0.92 mmol L⁻¹ d⁻¹ in control) and propionate was produced instead (1.78 mmol L⁻¹ d⁻¹, 2.8 times that of the control). With O₂ contamination, BCR-test produced 46% less n-butyrate (1.22 of 2.25 mmol L⁻¹ d⁻¹) and 91% less methane (1.41 of 15.9 mmol L⁻¹ d⁻¹) than the control (Supplementary Table S1). After O₂ contamination stopped in BCR-test (from day 50 on), n-caproate production recovered with a rate of 1.40 mmol L⁻¹ d⁻¹ while propionate formation decreased to 0.34 mmol L⁻¹ d⁻¹. n-Butyrate and methane production recovered partially and remained 18% (1.84 of 2.25 mmol L⁻¹ d⁻¹) and 25% (11.9 of 15.9 mmol L⁻¹ d⁻¹) lower than that of the control, respectively.

Similarly to what was seen in the STRs, Actinobacteria increased their relative abundances at the cost of Clostridia and Methanobacteria during the O₂ contamination period in BCR-test (Figure 5). In addition, Gammaproteobacteria also became more abundant during the contamination. Visualization of the community development at genus level in Supplementary Figure S7 reveals that the main genera of Actinobacteria were the same as those found in the STRs (i.e., Acidipropionibacterium and Actinomyces). Besides the Clostridia genera that dominated in the STRs, a transient presence of Eubacterium was also observed in the BCRs. Gammaproteobacteria were represented by Sutterella and Burkholderia (Supplementary Figure S7).

Figure 6 shows Spearman correlations between the most abundant genera, O₂ contamination, and the formation or consumption rates of the main chemicals. In Supplementary Table S2, the correlation coefficients and their p-values are listed. Positive correlations of n-caproate formation and relative abundances of Caproiciproducens, unclassified Peptostreptococcales, and Methanobrevibacter were found, whereas Clostridium sensu stricto 12, unclassified Micrococcales, Acidipropionibacterium, Burkholderia, Rummeliibacillus, Dialister, and Sutterella correlated negatively. Propionate production correlated positively to abundances of Acidipropionibacterium, Burkholderia, and Proteiniphilum. O₂ contamination correlated negatively with Methanobrevibacter, whereas positive correlations were found with abundances of unclassified Eggerthellaceae, unclassified Actinomycetaceae, Actinomyces, and Proteiniphilum. Abundances of Clostridium sensu stricto 12, Acidipropionibacterium, Acidaminococcus, and Dialister correlated positively with i-butyrate and n-valerate production. H₂ consumption after discounting methane production (i.e. non-CH₄ H₂ consumption) correlated positively with Acidipropionibacterium, Actinomyces, and unclassified Micrococcales. i-Butyrate production correlated negatively with relative abundances of Methanobrevibacter, Caproiciproducens, and unclassified Peptostreptococcales.

Among the most abundant amplicon sequence variants (ASVs), Clostridium sensu stricto 12 and Acidipropionibacterium had ASVs that were identical to isolated species in the Silva 138 database. Within the genus Clostridium sensu stricto 12, ASVs were assigned to C. luticellarii, C. tyrobutyricum, or C. ljungdahlii (the latter was also ambiguously assigned to C. autoethanogenum, C. ragsdalei, and C. coskatii). ASVs within Acidipropionibacterium were assigned to A. acidipropionici.

Hydrogenotrophic methanogenesis was likely the main pathway for CH₄ production. This is supported by the results of previous experiments where Methanobrevibacter was one of the predominating community members under similar conditions with H₂/CO₂ (Baleeiro et al., 2021a). Notably, methanogens partially overcame the inhibition by ethylene. The acclimatization occurred after 42 days of successful inhibition reported by Baleeiro et al. (2021b). To the best of our knowledge, such acclimatization has not been reported before. With ethylene, CH₄ production was relatively strong (up to 19.5 mmol L⁻¹ d⁻¹) but still about one third lower than the rates observed in the absence of the inhibitor in a similar gas recirculation system (up to 32.8 mmol L⁻¹ d⁻¹) (Baleeiro et al., 2021b). We hypothesize that Methanobrevibacter, the main methanogenic genus found in our reactors, may have acclimatized to ethylene by expressing [Fe]-hydrogenases. For hydrogenotrophic methanogens, [Fe]-hydrogenases have less favorable kinetics than [NiFe]-hydrogenases. Still, some methanogens, including Methanobacteria, can express [Fe]-hydrogenases to grow under nickel-limiting conditions (Thauer et al., 2010). As postulated by Baleeiro et al. (2021b), ethylene might not exert an inhibitory effect on nickel-free hydrogenases as it does on [NiFe]-hydrogenases of methanogens. Considering that [Fe]-hydrogenases are not inhibited by O₂ (Thauer et al., 2010; Stiebritz and Reiher, 2012), micro-aerobic conditions and broth mixing between reactors could have caused further selection of methanogens expressing [Fe]-hydrogenases even if nickel was not limiting. Further studies using transcriptome or proteome analyses of pure methanogenic cultures are needed to test this hypothesis.

When the O₂ concentration in the gas phase increased up to 18% between days 11 and 28, the aerobic genus Rummeliibacillus (Vaishampayan et al., 2009; Her and Kim, 2013) flourished in the bubble column reactor and the concentrations of n-butyrate and propionate decreased (Supplementary Figure S6). With an O₂ concentration below the detection limit by day 28 (Supplementary Figure S3), Actinobacteria abundance increased and propionate production reached its highest rates (Supplementary Table S1).

Methanogenesis and n-caproate production were strongly inhibited by O₂ intrusion. After the contamination stopped, CH₄ production recovered on every occasion but n-caproate production did not (Supplementary Table S1), indicating that the O₂ contamination had particularly strong detrimental effects on C4-to-C6 chain elongation. In the period between days 119 and 139 in STR-test, n-caproate production did not increase after O₂ contamination rate changed from 474 ± 33 mL O₂ L⁻¹ d⁻¹ to 39 ± 33 mL O₂ L⁻¹ d⁻¹, instead, the highest rates of n-butyrate production and acetate consumption were achieved. This was observed even with high relative abundances of Caproiciproducens in the community (Supplementary Figure S5).

Relatively low O₂ contamination rates were found to favor propionate formation in lactate-based fermentation. However, the relationship between O₂ and propionate accumulation was not as straightforward as the inhibitory effect of O₂ on chain elongation and methanogenesis. One possible reason is that propionate is not only a product of lactate fermentation, but also a substrate for n-valerate production by chain-elongating bacteria (Angenent et al., 2016). This could explain what was observed in the micro-aerobic period between days 115 and 119 in the STR-test, when propionate production did not increase but n-valerate production was relatively high (Supplementary Table S1).

Among the microorganisms enriched in the O₂ contamination periods, Acidipropionibacterium correlated to propionate production (p < 0.01). Other Actinobacteria enriched during O₂ contamination did not correlate with propionate production and may have diverted electrons from lactate to products other than propionate, which could be another reason why STR-test showed lower propionate production rates (Supplementary Table S1). Actinomyces, a facultative anaerobe (Rao et al., 2012), was particularly enriched in STR-test during an O₂ contamination period (Supplementary Figure S4). Actinomyces can grow anaerobically or aerobically on lactate and produces acetate, formate, and CO₂ during fermentative growth (Takahashi et al., 1994; Takahashi and Yamada, 1999; Rao et al., 2012). In agreement with the reported aerobic growth of Actinomyces naeslundii on lactate (Takahashi and Yamada, 1999), Actinomyces was likely not responsible for propionate production in the STR. The Coriobacteriia (Eggerthellaceae) observed in O₂ contamination periods belong to a family of strict anaerobes that are not reported to produce propionate (Gupta et al., 2013). Proteiniphilum (as well as Dialister) are genera with microaerophilic species that produce propionate, although it is not clear if from lactate (Tomazetto et al., 2018; Sakamoto et al., 2020). In our study, the abundance of Proteiniphilum correlated to O₂ contamination and to propionate formation whereas no significant correlation was found between Dialister, propionate, and O₂ contamination (Figure 6).

Fermentation of lactate by propionate producing bacteria commonly leads to a 2:1:1 stoichiometry of propionate to acetate to CO₂ (Eq. 2). Gammaproteobacteria and Actinobacteria species that produce propionate are known to use methylmalonyl-CoA pathways rather than the acrylate pathway (Seeliger et al., 2002; Gonzalez-Garcia et al., 2017). In particular, Acidipropionibacterium spp. are among the most efficient propionate producers thanks to the highest energy efficiency of their methylmalonyl-CoA pathway (also known as Wood-Werkman cycle, a succinate pathway involving methylmalonyl-CoA:pyruvate transcarboxylase) (Scholz and Kilian, 2016; Gonzalez-Garcia et al., 2017). 3   C H 3 C H O H C O O − → 2   C H 3 C H 2 C O O − + C H 3 C O O − + C O 2 + H 2 O (2) Even though it can express O₂-sensitive enzymes for fermentative growth similar to those in Clostridium, Acidipropionibacterium also has aerotolerant enzymes with similar functions (Piwowarek et al., 2018; McCubbin et al., 2020). Members of this genus are not only more tolerant to O₂ than clostridia, they have also been found to increase propionate and energy yields when exposed to O₂ (McCubbin et al., 2020).

It should be taken into account that propionibacteria, unlike many clostridia, do not form endospores (Gonzalez-Garcia et al., 2017). Hence, if exploration of propionate production is desired, shock treatments of the inoculum (e.g., with pH or heat), common techniques for starting non-methanogenic anaerobic bioprocesses (Baleeiro et al., 2019), should be avoided.

The phenomenon that lactate is diverted to propionate in chain elongation reactors under micro-aerobic conditions may have been overlooked in former studies. In one notable case, Kucek et al. (2016) observed the competitive production of propionate in a lactate-based chain elongation reactor. Although the possibility of O₂ contamination was not discussed, the study detected high abundances of Acinetobacter, strictly aerobic Gammaproteobacteria (Smet et al., 2014) commonly found in micro-aerated reactors (Krayzelova et al., 2015). To explain the propionate production observed in certain time periods, Kucek et al. (2016) suggested the residual concentration of lactate in the reactor to be a determining factor. Although not discussed in the study, O₂ presence could have played a role in propionate production.

A common way for the reduction of O₂ in the presence of H₂ is shown in Eq. 3 and is realized even by obligate anaerobes such as methanogens (Thauer et al., 2010), Negativicutes (Boga et al., 2007), and sulfate reducers (Dannenberg et al., 1992; Chen et al., 1993). 2 H 2 + O 2 → 2 H 2 O (3) H₂ consumption that was not attributed to methane formation was particularly high during O₂ contamination periods. In STR-test, it ranged from 3.0 to 3.4 fold molar O₂ consumption, while in BCR-test it ranged from 1.1 to 3.6 fold. Considering the 2:1 ratio of H₂ to O₂ during H₂ oxidation (Eq. 3), H₂ consumption not linked to methane formation in this study may have been related to other reactions during O₂ intrusion. Interestingly, similar consumption ratios of H₂ to O₂ (between 3.2 and 3.4) have been observed in communities dominated by hydrogen-oxidizing bacteria during autotrophic growth (Matassa et al., 2016). Nevertheless, we did not observe the presence of Sulfuricurvum (the genus enriched by Matassa et al. (2016)) and Burkholderia, a possible aerobic autotroph (Takors et al., 2018) found in our study, did not correlate positively with H₂ consumption nor with O₂ contamination. Besides, if aerobic hydrogen-oxidizing bacteria played a major role in the micro-aerobic reactors in our study, signs of biomass formation and carbon source (e.g., CO₂) consumption should have accompanied H₂ oxidation with O₂. However, no clear relation between O₂ contamination, biomass formation, and CO₂ consumption rates was found.

Communities enriched with propionate-producing bacteria in anaerobic reactors, such as Acidipropionibacterium spp., often correlate with high H₂ consumption or low H₂ production (Cabrol et al., 2017). In the presence of exogenous H₂, some propionate producers such as Propionispira arboris are able to perform homopropionate fermentation of lactate (Eq. 4) producing neither CO₂ nor acetate (Thompson et al., 1984). C H 3 C H O H C O O − + H 2 → C H 3 C H 2 C O O − + H 2 O (4) We did not observe Propionispira spp. in our reactors and its closest relative found in our system (Dialister) is only related at the order level (Veillonellales-Selenomonadales). Besides, Eq.4 alone cannot explain the high H₂ consumption during most of the micro-aerobic periods in this study. H₂ consumption not linked to methane was much higher than propionate formation. In fact, the period between days 39 and 50 of BCR-test had the lowest H₂ to O₂ consumption ratio and was the period with the highest propionate productivity (Supplementary Table S1). Lastly, it is not clear if the correlation found between abundance of Acidipropionibacterium and H₂ consumption not linked to methane (Figure 6) is a direct one. H₂ consumption by isolated members of Propionibacterium is not observed during fermentative growth (Seeliger et al., 2002).

Homoacetogens consume H₂ and CO₂ (Eq. 5) and, among them, at least C. ljungdahlii was shown to have some resistance against O₂ exposure (Whitham et al., 2015). Here, similar clostridia were detected in the reactors and Clostridium sensu stricto 12 was still present during some O₂ contamination events (Supplementary Figures S3, S7). Therefore, homoacetogenic activity could be considered to explain H₂ consumption during O₂ contamination. Nevertheless, no further evidence for this hypothesis was found. H₂ consumption was not accompanied by net CO₂ consumption and the net acetate production was unfavorable because acetate was fed in excess with the growth medium (12 g L⁻¹ acetate) to favor chain elongation as an acetate-consuming reaction. 4 H 2 + 2 C O 2 → C H 3 C O O − + H + + 2 H 2 O (5) Another way that H₂ presence might have influenced the micro-aerated community is by amplifying the effects of O₂ contamination. The activation of O₂ into ROS by hydrogenases and reduced electron carriers might be accentuated by H₂ recirculation (Misra and Fridovich, 1971; Krab et al., 1982). Since we did not have a control reactor for the presence of H₂, we could not test this hypothesis.