Three CSTRs (R_ctrl, R_NH3, R_{NH3, HCl}) with working volumes of 6 L were operated in parallel at 37°C at a hydraulic retention time (HRT) of 5.5 d. All CSTRs were fed continuously with a synthetic, liquid substrate containing a mixture of VFAs as the only carbon source (45% acetic, 10% propionic, and 45% butyric acid based on chemical oxygen demand, COD) with a total concentration of 37.2 gCOD L⁻¹ in a mineral medium containing all necessary trace elements, macronutrients, and vitamins (see Supplementary Material A, Table S1). The CSTRs were inoculated from a lab-scale digester operated at a HRT of 5.5 d with the same synthetic substrate. The ammonia concentration in the substrate was 52 mM at the start for all CSTRs, and was increased to 277 mM for R_NH3 and R_{NH3, HCl} on day 21 until the end of the experiment (see Figure 1). For R_{NH3, HCl}, 100 mM HCl were added starting on day 38. On day 55, 62, and 69, R_NH3 and R_{NH3, HCl} were fed not continuously but with the volume of one daily feeding at once as deliberate disturbances. On the day after each disturbance, continuous feeding was resumed as usual. On day 73, HCl addition to the medium of R_NH3HCl was ceased until the end of the experiment on day 79. Biogas composition (CH₄, CO₂, O₂, H₂, and H₂S content), biogas production rate, pH, VFA concentrations, total solids content (TS), and volatile solids content (VS) were determined as described previously (Mulat et al., 2016). The TS and VS content measurements were modified in a way that a reactor sample was not added directly to a crucible, but only the pellet after centrifuging 100 mL reactor content (10,000 × g, 10 min, 10°C) to increase the amount of biomass per analysis and washing the pellet with phosphate buffered saline (PBS) solution (140 mM NaCl, 10 mM Na₂HPO₄ × 2 H₂O, 2.7 mM KCl, 1.8 mM KH₂PO₄, pH 7.4) to reduce the potential error of VFAs and ammonia in the reactor sample.

The degradation efficiencies E_i (t) for each individual VFA i were calculated for each sampling date based on mass balances over 1 day considering the VFA input via the influent over the last day S_i,_in/HRT·Δt, the VFA output via the effluent over the last day ΔS_{i, out} (t), the amount of VFAs in the reactor at the sampling date minus the amount at the previous day ΔS_i(t) and, for acetic acid degradation only, the production of acetic acid from propionic and butyric acid oxidation over the last day ΔS_{i, p} (t). VFA mass degraded during 1 day is then given by S_i,in/HRT.Δt + ΔS_i,p(t) − ΔS_i,out(t) − ΔS_i(t). Based on a constant reaction volume, the masses of each component can be calculated based on the respective concentrations in the influent and reactor digestate so that degradation efficiency can be expressed as: (1a)Ei(t)=Si,inHRT⋅Δt−ΔSi,out(t)+ΔSi,p(t)−ΔSi(t)Si,inHRT⋅Δt+ΔSi,p(t) (1b)ΔSi,out(t)=Si(t)+Si(t−Δt)2⋅HRT⋅Δt (1c)ΔSi(t)=Si(t)−Si(t−Δt) (1d)ΔSac,p(t)=Spro,inHRT⋅Δt⋅Epro(t)⋅fac,pro                        +Sbu,inHRT⋅Δt⋅Ebu(t)⋅fac,bu (1e)ΔSpro,p(t)=ΔSbu,p(t)=0 with f_ac,pro and f_ac,bu being the stoichiometric conversion factors of propionate oxidation (1 mol acetate per mol propionate) and of butyrate oxidation (2 mol acetate per mol of butyrate), respectively. Between sampling dates, VFA concentrations were linearly interpolated. E values of 1 correspond to a complete degradation in steady state. E values can reach values higher than 1 when accumulated VFAs are degraded (ΔS_i(t) < 0).

We modified the original ADM1 model structure (Batstone et al., 2002) in various aspects. Mass balances were closed by amending the ordinary differential equations for inorganic carbon and inorganic nitrogen. Biogas production rate was calculated considering the overpressure in the headspace. If not indicated otherwise, parameter values from the benchmark model by Rosen and Jeppsson (2006) were used.

The acetoclastic methanogens (X_ac) were replaced by two competing populations: Methanosarcina (X_ac,1, see Equation 2a) and Methanosaeta (X_ac,2, see Equation 2b). Accordingly, the differential equation for acetic acid was adapted to Equation 2c. The original ADM1 inhibition functions (Batstone et al., 2002) were used with different parameter values for X_ac,1 and X_ac,2 concerning the ammonia inhibition function (Equation 2d, Table 1). (2a)dXac,1dt=1HRT⋅(Xin,ac,1−Xac,1)                         +(Yac,1⋅km,ac,1⋅SacSac+KS,ac,1⋅Iac,1−kdec)⋅Xac,1 (2b)dXac,2dt=1HRT⋅(Xin,ac,2−Xac,2)                   +(Yac,2⋅km,ac,2⋅SacSac+KS,ac,2⋅Iac,2−kdec)⋅Xac,2 (2c)dSacdt=1HRT⋅(Sac,in                   −Sac)−km,ac,1⋅SacSac+KS,ac,1⋅Iac,1⋅Xac,1                  −km,ac,2⋅SacSac+KS,ac,2⋅Iac,2⋅Xac,2+∑j=511v7,j⋅ρj, (2d)Inh3, i=11+Snh3KI,nh3,i with Y being the biomass yield, k_m the maximum substrate uptake rate, K_S the half saturation constant, k_dec the decay rate, I the inhibition functions, S the substrate concentration, HRT the hydraulic retention time, ac the index for acetic acid, ac1 the index for Methanosarcina, ac2 the index for Methanosaeta, ∑j=511υ7,j·ρj the acetic acid producing reactions (see Peterson Matrix, Supplementary Material B), I_nh3,i the inhibition ammonia inhibition function, S_nh3 the ammonia concentration and K_I,nh3,i the empirical ammonia inhibition parameter with different values for Methanosarcina (i = ac1) and Methanosaeta (i = ac2).

The propionic acid degraders (X_pro) were replaced by two competing populations: an unknown taxon (X_pro,1) and Syntrophobacter (X_pro,2) following the same approach as described above for the acetoclastic methanogens (Supplementary Material A, Section Implementation of two competing propionic acid oxidizers in ADM1). Ammonia inhibition was introduced for both populations following the same inhibition equation that is used for acetic acid degraders in the original ADM1 structure with different parameter values for X_pro,1 and X_pro,2 (see Table 1). A detailed description of the extended model structure is included in the Supplementary Material (Peterson Matrix in Supplementary Material B). All differential and algebraic equations were implemented and solved in the software environment Matlab R2014b (Mathworks).

Cation and anion concentrations in the influent were calculated based on the composition of our synthetic media to 78.3 and 19.8 mM, respectively. Inorganic carbon concentrations were equal to inorganic nitrogen concentrations in the influent with 52 mM before and 277 mM after day 21 based on the ammonium bicarbonate concentration in the synthetic media. Initial concentrations for all state variables (except X_ac,1 and X_pro,1) were approximated based on simulation results for steady state concentrations for the inorganic nitrogen concentration of 52 mM in the influent (ADM1 Simulation in Supplementary Material B).

Only parameter values for acetic acid and propionic acid degraders were changed compared to the benchmark model parameter values by Rosen and Jeppsson (2006). The initial concentrations and half-saturation constants of the ammonia tolerant populations (X_ac1(t = 0), X_pro1(t = 0), K_S,ac1, K_S,pro1) and ammonia inhibition constants of all acetic acid and propionic acid degraders (K_I,nh3,ac1, K_I,nh3,ac2, K_I,nh3,pro1, K_I,nh3,pro2) were estimated using the solver fmincon (algorithm: interior point) in Matlab R2014b (Mathworks) minimizing the sum of squared error between experimental and simulation results, first for propionic acid concentration and second for acetic acid concentration.

On day 21, after 3 HRTs under low ammonia concentrations of 52 mM, all three CSTRs (see Figure 2) had total VFA concentrations of maximal 1.5 g L⁻¹ and pH values of maximal 7.35. Increasing the ammonia concentration in the influent to 277 mM in R_NH3 and R_{NH3, HCl} starting on day 21 was followed by an accumulation of VFAs and decline in methane production rates. The strongest accumulation was observed for acetic acid with up to 8.86 g L⁻¹ (R_NH3 on day 42). Propionic acid accumulated up to 1.63 g L⁻¹ (R_NH3 on day 49) and butyric acid showed the smallest accumulation with up to 0.52 g L⁻¹ (R_NH3 on day 42). On day 55, acetic, propionic, and butyric acid concentrations had declined to 0.85, 1.26, and 0.02 g L⁻¹ for R_NH3 as well as 0.69, 0.48, and 0.01 g L⁻¹ for R_{NH3, HCl}, respectively. This decline was accompanied by increases in pH values to 7.7 and 7.53 for R_NH3 and R_{NH3, HCl}, respectively. Starting on day 73, after HCl was omitted from the medium of R_{NH3, HCl}, the pH value in R_{NH3, HCl} rose to 7.7 toward the end of the experiment, equal to R_NH3, while total VFA concentrations remained at about 1 gCOD L⁻¹. Hydrogen partial pressures in the biogas were hardly affected by the increasing ammonia concentration and remained below 17 ppm in all CSTRs between day 21 and 55.

On days 55, 62, and 69, R_NH3 and R_{NH3, HCl} were given the amount of one daily feeding within 20 min instead of continuous feeding. Both reactors recovered within 3 days to pre-disturbance acetic and butyric acid degradation efficiencies except for R_{NH3, HCl} on day 55 when an acetic acid accumulation was observed followed by a recovery over about one HRT. In both reactors, propionic acid accumulation was observed after a disturbance followed by a recovery over about two HRTs.

During the start of the experiment, the degradation efficiencies of acetic and propionic acid in R_NH3 and R_{NH3, HCl} showed declines but recovered again until day 21. On day 21, the degradation efficiencies of all VFAs were higher than 86% in all reactors (Figure 3). After the increase in ammonia concentration on day 21, the lowest degradation efficiencies were 31% for propionic and 65% for acetic acid. Butyric acid degradation was almost not affected showing efficiencies of higher than 94% throughout the experiment.

After increasing the ammonia concentration, the formation of aggregates was observed, which reached a size of up to one mm in diameter toward the end of the experiment (see Supplementary Material A, Figure S1).

On day 21, all CSTRs showed a similar methanogenic community composition on DNA level, which was dominated by the acetoclastic Methanosaeta (47–56%) and the hydrogenotrophic Methanomicrobiaceae (25–30%). Methanosarcina, which can perform both acetoclastic and hydrogenotrophic methanogenesis, reached a relative abundance of up to 13% (Figure 4).

After increasing the ammonia concentration in the influent to 277 mM in R_NH3 and R_{NH3, HCl} starting on day 21, a strong increase in the relative abundance of Methanosarcina with a simultaneous decrease of Methanosaeta was detected for R_NH3 and R_{NH3, HCl}, while Methanosaeta continued to dominate over Methanosarcina in R_ctrl (Figure 4A). Methanomicrobiaceae remained in all reactors throughout the experiments and reached a relative abundance of about 20% in R_NH3 and R_{NH3, HCl} at the end of the experiment.

On cDNA level, the shift from Methanosaeta to Methanosarcina as the dominant methanogen was even more pronounced (Figure 4B). For R_NH3, Methanosarcina reached a relative abundance of 90% already on day 36 on cDNA level compared to day 49 on DNA level. Relative T-RF abundances were in general higher on cDNA level compared to DNA level for Methanosarcina in R_NH3 and R_{NH3, HCl} and for Methanosaeta in R_ctrl. No major impacts of the deliberate disturbances on the relative T-RF abundances of mcrA transcripts were observed.

On day 21, 96 to 110 OTUs were detected in the three CSTRs. All CSTRs were dominated by the OTUs Syntrophomonas (8–27% of bacterial reads), Syntrophobacter (13–16%), Thermovirga (15–22%), and Blvii28 wastewater-sludge group (6–38%).

After the increase in ammonia concentration, the community composition of R_NH3 and R_{NH3, HCl} showed strong changes. Syntrophobacter, Thermovirga and Blvii28 wastewater-sludge group decreased in relative abundance to < 0.1% of all bacterial reads in both R_NH3 and R_{NH3, HCl} toward the end of the experiment (Figure 5). Aminobacterium and Lutispora increased in both reactors from below 1.9% to 4.4–12.6% and from below 0.1% to 1.8–13.9% of all bacterial reads, respectively. The OTU Lineage 1 (Endomicrobia) strongly increased in relative abundance shortly after the ammonia inhibition in R_NH3 and decreased again toward the end of the experiment. A similar behavior was observed for Parcubacteria in R_{NH3, HCl}. Syntrophomonas remained one of the most abundant bacterial genera throughout the experiment in all CSTRs.

Our primer pair also led to the amplification of 16S rRNA genes from methanogens despite the lower coverage of archaea compared to bacteria (Klindworth et al., 2013). The ratio of bacterial to total reads varied only slightly throughout the experiment. On day 21, bacterial reads made up 18% and 16% of total reads for R_NH3 and R_{NH3, HCl}, respectively, compared to 13 and 17%, respectively, on day 79.

Simulations using the default ADM1 implementation of ammonia inhibition did not resemble our observed VFA concentrations (Figure 6A, racetic2 _acid = 0.05, rpropionic2 _acid = 0.07). Only acetic acid accumulation was predicted, but not the acclimatization toward the end of the experiment. After implementing our model modifications, i.e., second populations for acetic and propionic acid degradation, respectively, the simulations represented the observed VFA concentrations better (Figure 6B, racetic2 _acid = 0.98, rpropionic2 _acid = 0.56). After the increase of ammonia concentration in the influent on day 21, the ammonia sensitive populations in the simulations (X_ac2, X_pro2) became strongly inhibited and were washed out, while accumulations of 8.2 and 2.5 gCOD L⁻¹ occurred for acetic and propionic acid, respectively. At the same time, ammonia tolerant populations (X_ac1, X_pro1) increased in relative abundance (Figures 6C,D). On day 55, X_ac1 reached a relative abundance to total methanogens of 0.72 and acetic acid accumulation was reduced to 0.5 gCOD L⁻¹. On day 62, X_pro1 reached a relative abundance to total bacteria of 0.12 and propionic acid accumulation was reduced to 0.4 gCOD L⁻¹.

The inhibition of VFAs other than acetic acid are neglected in ADM1 (Batstone et al., 2002). However, looking at VFA degradation efficiencies (Figure 3), it became clear that propionic acid degradation was even more inhibited than acetic acid degradation. We also observed a stronger or similarly strong inhibition of propionic acid degradation compared to acetic acid degradation in several additional experiments we conducted (see Supplementary Material A). This suggests that ammonia inhibition of propionic acid degradation should receive more attention and be included in anaerobic digestion models.

Calculating VFA degradation efficiencies is only possible in synthetic systems like ours and not for more complex substrates such as manure, because the amounts of individual VFAs produced in acidogenesis from complex substrates are hard to quantify and thus, the VFA degradation efficiencies cannot be calculated. Therefore, the impact of ammonia inhibition on propionic acid degradation might have appeared weaker than it actually was in many studies because lower propionic acid concentrations than acetic acid concentrations were reached.

Concerning the mechanism of ammonia inhibition of propionic acid degradation, an indirect inhibition mechanism was suggested by Wiegant and Zeeman (1986) who argued that a strong ammonia inhibition of hydrogenotrophic methanogenesis can lead to accumulation of hydrogen, which inhibits propionic acid degradation by increasing the Gibbs energy change of catabolism to near or above zero (Figure 7A). However, this indirect route seemed not to play a major role in our experiment since we did not observe any hydrogen accumulation. Furthermore, butyric acid degradation was almost not inhibited in our experiment, which would have been the case if hydrogen accumulated. Therefore, we assumed direct ammonia inhibition of propionic acid degradation as the working hypothesis for our study. Based on our observations, acetic and propionic acid degradation were strongly inhibited in our experiment while butyric acid degradation and hydrogen conversion were hardly affected (Figure 7B).

After the increase in ammonia concentration started, it took about 5 HRTs until VFA concentrations reached values similar to the start of the experiment in R_NH3 and R_{NH3, HCl}. These increases and decreases in VFA concentrations were accompanied by strong shifts in the microbial community.

The propionic acid oxidizing Syntrophobacter was washed out of the reactors after ammonia inhibition started. Also in several other experiments we conducted at elevated ammonia concentrations, Syntrophobacter was washed out (see Supplementary Materials A,C). The absence of Syntrophobacter at high ammonia concentrations was also observed in a mesophilic reactor treating household waste at TAN concentrations of 386–414 mM (Westerholm et al., 2015), supporting our observation that Syntrophobacter is an ammonia sensitive taxon. The only other known syntrophic propionic acid degrading genus in our experiments was Pelotomaculum, which was not competitive with a relative abundance of about 0.1% independent of the ammonia concentration.

Since propionic acid oxidation continued after these two taxa reached negligible relative abundances, one or several yet unknown ammonia tolerant propionic acid oxidizing taxa must be among the detected bacteria. The OTUs that increased in relative abundance after the ammonia inhibition such as Aminobacterium and Lutispora are possible candidates; however, they could grow on butyric acid or decaying biomass as well. Other studies support the ammonia tolerance of these two taxa. Lutispora was found to correlate with the recovery of biogas production in an ammonia inhibited biogas reactor digesting wastewater treatment plant sludge (Chen et al., 2018). Aminobacterium increased in abundance after an increase in ammonia concentration in an anaerobic digester treating chicken manure and feathers with wood shavings (Belostotskiy et al., 2015).

The butyric acid degrading Syntrophomonas tolerated the increase in ammonia concentration in our experiment. A tolerance to TAN of up to 6.5 g L⁻¹ was observed earlier (Lee et al., 2018). Poirier et al. (2017) even observed relative abundances to total bacteria of 22% for ammonia concentrations as high as 25 g L⁻¹, supporting our observation that Syntrophomonas is ammonia tolerant.

The bacterial taxa Thermovirga and Blvii28 wastewater sludge group became abundant with up to 30% relative abundance to total bacteria each during the first 21 days. It is unlikely that they were mainly propionic acid degraders in our reactors because of their high abundance compared to the low share of propionic acid in the feed and it is unlikely that they were mainly acetic acid degraders because at the low free ammonia concentrations during the first 21 days in all reactors (<0.028 g L⁻¹), acetoclastic methanogenesis commonly dominates over SAO (Luo et al., 2016). Therefore, they were most likely syntrophic butyric acid degraders. Since both taxa were washed out after the increase in ammonia concentration in our experiment as well as in an experiment by Lee et al. (2018), they likely are ammonia sensitive.

Concerning acetoclastic methanogens, Methanosaeta dominated all CSTRs at the start of the experiments when ammonia and acetic acid concentrations were low. Its advantage over Methanosarcina at low acetic acid concentrations can be explained by its higher substrate affinity and minimum concentration threshold for acetic acid (Jetten et al., 1992). After the increase in ammonia concentration, Methanosaeta was washed out during the ammonia inhibition and replaced by Methanosarcina. The sensitivity of Methanosaeta to ammonia has been shown in a pure culture study (Steinhaus et al., 2007). Methanosarcina is known to be more tolerant against ammonia inhibition than Methanosaeta (De Vrieze et al., 2012). This tolerance has been connected with their ability to form aggregates (De Vrieze et al., 2012). Aggregates with diameters up to 1 mm were also observed in our experiment. After Methanosarcina completely dominated the acetoclastic methanogens, acetic acid degradation efficiencies recovered up to almost 100% toward the end of the experiment, showing that the applied ammonia concentrations were not inhibitory for Methanosarcina.

The hydrogenotrophic methanogens of the family Methanomicrobiaceae were also tolerant to the increase in ammonia concentration and remained abundant until the end of the experiment. We observed this tolerance in several other experiments on ammonia inhibition that we conducted (see Supplementary Material A). Methanomicrobiaceae have been found previously to be tolerant to TAN concentrations of about 3 g L⁻¹ in pure culture (Schnürer et al., 1999; Nettmann et al., 2010; Wang et al., 2015) and mixed culture studies (Angenent et al., 2002; Westerholm et al., 2011, 2012).

Theoretically, it is possible that all acetic acid was converted via SAO and hydrogenotrophic methanogenesis after day 21. In that case, Methanosarcina would be hydrogenotrophic instead of acetoclastic as assumed above. However, this appears to be unlikely. First, a switch to SAO should have led to an increase in relative abundance of bacteria compared to methanogens that we did not observe. Second, genera known for SAO were observed only in negligible abundances in our reactors. Five species capable of SAO have been cultured belonging to the genera Pseudothermotoga, Thermacetogenium, Clostridium, Syntrophaceticus, and Tepidanaerobacter (Müller et al., 2016). From them, only Clostridium and Syntrophaceticus were detected with maximum relative abundances of 0.2% to total bacteria. Third, anaerobic digesters dominated by SAO have been observed to be dominated by Methanoculleus or Methanobrevibacter instead of Methanosarcina (Luo et al., 2016). Finding acetoclastic Methanosarcina dominant in ammonia inhibited biogas plants is not unusual. Luo et al. (2016) observed the dominance of acetoclastic methanogenesis in industrial biogas plants with up to 0.46 g L⁻¹ of free ammonia which is almost double the maximum free ammonia concentration of 0.25 g L⁻¹ that we observed. Schnürer and Nordberg (2008) observed a dominance of acetoclastic methanogenesis over SAO at free ammonia nitrogen concentrations of up to about 0.3 g L⁻¹ in lab-scale digesters. Therefore, we assumed in our ADM1 simulations that Methanosarcina replaces Methanosaeta as an acetoclastic methanogen. Still, it cannot be excluded that Methanosarcina additionally took up hydrogen and that some SAO occurred in our reactors.

The relationship between microbial community dynamics and ammonia as well as VFA concentrations could be successfully reproduced with our extended ADM1 model. However, major changes to the original ADM1 structure were necessary. First of all, inhibition of propionic acid degradation was not implemented in the original ADM1 structure. However, we could show in our experiment that propionic acid degradation was even stronger inhibited than acetic acid degradation, and therefore, the addition of an ammonia inhibition term for propionic acid degradation was necessary. Secondly, simulations based on the original ADM1 structure did not (and in principle cannot) lead to a process recovery after the increase in VFA concentrations following ammonia inhibition because only one taxon of each functional group is part of the original structure. For example, only one taxon is capable of acetoclastic methanogenesis. As a consequence, our simulations using the original ADM1 structure could only lead to no VFA accumulation, to a constant elevated VFA concentration, or process breakdown. However, our experimental data clearly showed an increase in VFA concentration due to ammonia inhibition, followed by a decrease in VFA concentration toward the end of the experiment after the microbial community adapted to the new conditions. Therefore, the addition of at least a second taxon of acetic and propionic acid degraders was necessary in ADM1 to reflect the reactor performance dynamics as a consequence of the microbial community dynamics.

When fitting the simulations to the experimental results, we only changed the ammonia inhibition constants of X_ac,1, X_ac,2, X_pro,1, and X_pro2, and the initial concentration and half saturation constants of X_ac,1 and X_pro,1 compared to the benchmark model parameter values by Rosen and Jeppsson (2006). The fitting resulted in higher substrate affinities (lower half saturation constants) for the ammonia sensitive populations (X_ac,2 and X_pro,2) compared to the ammonia tolerant populations (X_ac,1 and X_pro,1). This reflects the higher substrate affinities previously observed for Methanosaeta compared to Methanosarcina (Jetten et al., 1992; Straub et al., 2006). However, the advantage of Syntrophobacter over its competitors is not known. Other advantages than substrate affinity are possible, such as higher maximum substrate uptake rates, higher specific growth yields or a combination thereof. A high sensitivity for the parameters K_I,nh3,ac,2, K_S,ac,1, K_I,nh3,pro,2, K_S,ac,2 and the initial concentrations of X_ac,1 and X_pro,1 was found (see Sensitivity Analysis in Supplementary Material A). By contrast, increasing the values of the ammonia inhibition constants of the ammonia resistant taxa (K_I,nh3,ac,1 and K_I,nh3,pro,1), i.e., decreasing the ammonia sensitivity of these populations, did not change the simulation results. This indicates that higher ammonia concentrations are necessary to unambiguously determine these parameter values (see Supplementary Material A, Figure S6). Hydrogenotrophic methanogenesis and butyric acid oxidation were not changed in ADM1 because their performance was hardly impacted by the increase in ammonia concentration in our experiment.

While our model helps to illustrate the relationship between microbial community dynamics and VFA accumulation, extending it for the use as predictive model remains a challenge, in particular regarding parameter identification. In communities with competing taxa, the quantification of taxon-specific substrate uptake rates is a major obstacle. Fitting these rates based on gross consumption likely leads to non-unique solutions and a loss of generality and predictive power. A predictive model would require the physiological characterization of all relevant taxa in pure cultures or defined co-cultures on defined media. Isolating prokaryotes was not always successful in the past (Pelletier et al., 2008). However, the advent of metaomics techniques gives hope that suitable cultivation conditions can be inferred more easily in the future (Overmann et al., 2017). Success in cultivation would be rewarded with high benefits: A predictive multi-taxa ADM1 model could be an essential resource in gaining deterministic control over the microbial community composition, for example by changing process parameters and/or bioaugmentation, in order to increase the productivity of biogas plants suffering from ammonia and other inhibitions.

In conclusion, ammonia inhibition is a major challenge for the biogas process, resulting in economic losses and even process failures. Both our experimental and simulation results showed the importance of ammonia-sensitive taxa, such as Methanosaeta and Syntrophobacter, and ammonia-tolerant taxa, such as Methanomicrobiaceae and Syntrophomonas, for understanding the reactor performance as a result of microbial community dynamics of anaerobic digesters impacted by high ammonia concentrations.

The 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.