Adding Species to the Mix: Refining Models of Anaerobic Digestion

Anaerobic digestion is a microbial mediated process which occurs both in natural environments such as lake sediments or animal and human guts and in biotechnological settings such as waste water treatment systems. Its ability to produce methane (biogas) from organic waste material makes it an important component of an energy economy that no longer depends on fossil fuels. The process itself is a multi-stage process which is mediated by mixed microbial communities. The comprehensive characterization of the overall process has allowed for the mathematical modelling of the whole process, with the Anaerobic Digestion Model No. 1 (ADM1) being the most prominent example (Batstone et al., 2002). While this differential-equation-based dynamic model captures all relevant process steps, it summarizes the microbial activity in only few functional groups, neglecting the true diversity of microbial communities in terms of composition and metabolic capacity. These simplifications are in contrast to recent developments which enable the reconstruction of genome-scale metabolic models for single species from genomic data (Feist et al., 2009). To harness these developments for the modelling of anaerobic digestion, we set out to test whether dynamic models can be augmented by such detailed network models describing metabolic activity within microbial cells. We selected methanogenesis as the first process to be simulated by a detailed metabolic network model. Methanogenesis is the last step of anaerobic digestion and is performed by archaea following two distinct pathways. In acetoclastic methanogenesis, acetate is converted to methane while in hydrogenotrophic methanogenesis, CO₂ and H₂ are used as alternative substrates. As metabolic network models, we selected a published genome-scale model of Methanosarcina barkeri which is capable of both pathways and contains 815 reactions (Gonnerman et al., 2013), an essential model of M. barkeri with 111 reactions, and an essential model of Methanothermobacter thermoautotrophicus only capable of hydrogenotrophic methanogenesis with 110 reactions. A comparison between model predictions regarding methane production and biomass growth under different substrate concentration levels revealed a general qualitative agreement between ADM1 and the metabolic network models. Only for low acetate concentrations for which the maintenance requirement could not be fulfilled in the metabolic network models, the network models predicted no growth and no productivity, while ADM1 still did. ADM1 predicts a lower biomass production and a slightly higher methane production compared to the metabolic network models. Whether these differences have an impact on dynamic simulations was tested by coupling the models following the dynamic Flux Balance Analysis approach (dFBA, Antoniewicz, 2013) using a Matlab implementation of ADM1 (Weinrich et al., 2015) and CellNetAnalyzer (Klamt et al., 2007) for FBA computations. Overall, we encountered no principle obstacles when augmenting ADM1’s representation of metabolic activity by detailed metabolic network models for single species. Hence, this approach is a promising step towards a new generation of models which better reflect the diversity in composition and metabolic capacity within mixed microbial communities. In particular, this approach opens the way for a better integration of next-generation meta-omics data for improved predictive modeling of anaerobic digestion.