Mechanistic understanding of mixed-culture fermentations by metabolic modelling

TESE DE DOUTORAMENTO MECHANISTIC UNDERSTANDING OF MIXED-CULTURE FERMENTATIONS BY METABOLIC MODELLING Alberte Regueira López ESCOLA DE DOUTORAMENTO INTERNACIONAL PROGRAMA DE DOUTORAMENTO EN ENXEÑARÍA QUÍMICA E AMBIENTAL SANTIAGO DE COMPSOTELA 2020 DECLARACIÓN DO AUTOR DA TESE Mechanistic understanding of mixed-culture fermentations by metabolic modelling D. Alberte Regueira López Presento a miña tese, seguindo o procedemento axeitado ó Regulamento, e declaro que: 1) A tese abarca os resultados da elaboración do meu traballo. 2) De selo caso, na tese faise referencia ás colaboracións que tivo este traballo. 3) A tese é a versión definitiva presentada para a súa defensa e coincide coa versión enviada en formato electrónico. 4) Confirmo que a tese non incorre en ningún tipo de plaxio doutros autores nin de traballo presentados por min para a obtención doutros títulos. Santiago de Compostela, a 16 de marzo de 2020 Asdo. Alberte Regueira López AUTORIZACIÓN DOS DIRECTORES DA TESE Mechanistic understanding of mixed-culture fermentations by metabolic modelling D. Juan M. Lema Rodicio, Profesor Emérito de Enxeñaría Química e Miguel Mauricio Iglesias, Investigador Posdoutoral do Departamento de Enxeñaría Química INFORMAN Que a presente tese, correspóndese co traballo realizado por D. Alberte Regueira López , baixo a nosa dirección, e autorizamos a súa presentación, considerando que reúne os requisitos esixidos no Regulamento de Estudos de Doutoramento da USC, e que como directores desta non incorre nas causas de abstención establecidas na Lei 40/2015. Santiago de Compostela, 16 de marzo de 2020 Asdo. Juan M. Lema Rodicio Asdo. Miguel Mauricio Iglesias From elephant to butyric acid bacterium, it is all the same! Albert Kluyver Abstract The biorefinery carboxylate platform is set to have a key role in the establishment of a more circular economy since in this production scheme organic wastes are transformed into added-value products such as bioplastics or biofuels. The first step is the production in anaerobic mixed-culture fermentations (MCF) of volatile fatty acids (VFA), a family of valuable compounds that are also used as substrate in subsequent processes to yield the final biorefinery products. One of the main issues of MCFs is their characteristic poor and variable product selectivity. The product spectrum is substantially influenced by environmental conditions without a systematic and mechanistic understanding or control. As a consequence, designing MCF processes targeting specific products is a challenging task and limits significantly the progress of the carboxylate platform. Therefore, the aim of this thesis is to develop predictive tools to gain insight of the mechanisms governing the stoichiometry of MCF under different environmental conditions and to use that knowledge to design MCF processes that allow for the selective production of the desired products. The work performed in this thesis led to a more comprehensive and wider knowledge of MCF metabolism. In the case of glucose MCF, the inclusion of the biochemical mechanism electron bifurcation in the metabolic network and the consideration of the resource allocation theory in the mathematical models allowed to fully understand its metabolism and stoichiometry. For the first time, lactate production occurring only under a narrow operational condition space could mechanistically understood. The stoichiometry of VFA production from protein mono-fermentation and from its co-fermentation with carbohydrates can now be predicted at different operational conditions. The mathematical models contributed in gaining insight of the mechanisms linking the environmental conditions of the reactor and the product selectivity of the MCF processes. The substrate composition and the reactor configuration and pH were identified as the operational conditions with the highest influence on the product spectrum. The models shown that the different substrates may interact in the fermentation and compete for shared resources. These relationships are directly correlated with the concentration of the substrates involved and therefore the proportions between the substrates have an impact on the global stoichiometry of the process. The reactor configuration proved to exert an influence on the outcome of microbial competition. Discontinuous 8.1.1. Substrate composition influences MCF outcome ............................ 224 8.1.2. Operational conditions determine the product spectrum of MCF 227 8.1.3. MCF selectivity can be controlled with design parameters............. 229 8.2. Limitations of the bioenergetic models ....................................... 231 8.2.1. Enzyme soup approach ........................................................................ 231 8.2.2. Optimisation strategy ............................................................................ 233 8.3. Implications of this thesis for process design ............................. 234 8.3.1. The BIOCHEM project ....................................................................... 234 8.3.2. Enhanced process configurations ....................................................... 235 8.3.3. Early-stage bioprocess design methodology ..................................... 236 8.4. Future perspectives on computer-aided design .......................... 239 REFERENCES .......................................................................................... 241 List of publications................................................................................. 269 1 ABBREVIATIONS AND ACRONYMS AA Amino acid ADP Adenosine 5’-diphosphate Ala Alanine Arg Arginine Asn Asparagine Asp Aspartate ATP Adenosine 5’-triphosphate BPB Butyrate-producing bacteria COD Chemical oxygen demand CSTR Continuous stirred tank reactor CV Coefficient of variation Cys Cysteine D Dilution rate E Redox potential EB Electron bifurcation EC Electron carrier FBA Flux balance analysis Fdox Oxidised ferredoxin Fdred Reduced ferredoxin ΔG Gibbs free energy of reaction Gln Glutamine Glu Glutamate Gly Glycine HA Homoacetogenesis His Histidine HRT Hydraulic retention time Ile Isoleucine LAB Lactic acid bacteria Leu Leucine 2 Lliq Liter of liquid (excluding biomass) Lr Liter of reactor LX Liter of biomass Lys Lysine MCF Mixed-culture fermentation Met Methionine NAD(H) Nicotinamide adenine dinucleotide OD Optic density OLR Organic loading rate PHA Polyhydroxyalkanoates pmf Proton motive force Pro Proline Ri Reaction rate (mol/L·h) RMSD Root-square-mean deviation RT Transport rate (mol/L·h) SBR Sequential batch reactor (SBR) Ser Serine Sex Extracellular concentration (mol/L) Sin Intracellular concentration (mol/L) SIR Substrate-to-inoculum ratio SLP Substrate-level phosphorylation SRT Solids retention time SX Biomass concentration (mol/L) TAN Total ammonia nitrogen Thr Threonine Val Valine VFA Volatile fatty acid RESUMO Resumo 5 Capítulo 1. Introdución e contexto O actual modelo económico baseado no uso masivo de recursos fósiles está comezando a súa transición cara un esquema socioeconómico máis circular en resposta a un futuro escenario de altos e inestables prezos de petróleo e á demanda da sociedade de apostar por este tipo de modelos. Neste contexto enmárcase o concepto de biorrefinería, un esquema de produción no que a biomasa (residuos orgánicos ou madeira) é transformada nun espectro de produtos e enerxía con valor de mercado como plásticos, combustibles ou electricidade. As biorrefinerías pretenden ser tan flexibles como as tradicionais refinerías de petróleo e producir baixo demanda un abano de produtos dende un número reducido de materias primas. Dentro dos diferentes tipos de biorrefinería, destaca a plataforma dos ácidos carboxílicos xa que se centra en empregar residuos orgánicos como substratos. Neste esquema de biorrefinería o substrato é transformado nunha primeira etapa de fermentación anaerobia a unha mestura de ácidos carboxílicos ou ácidos graxos volátiles (AGV), coma o acético ou butírico, e alcois que son finalmente convertidos en produtos de alto valor engadido como bioplásticos ou biocombustibles. Os AGV actúan de nexo entre os residuos orgánicos e os compostos finais e, polo tanto, o proceso de fermentación anaerobia é primordial para o esquema da plataforma dos ácidos carboxílicos. Porén, un dos principais problemas das fermentacións anaerobias é que o seu espectro de produtos adoita ser unha mestura de AGV cunha composición que depende tanto dos parámetros de operación como do substrato empregado. No contexto da plataforma dos ácidos carboxílicos, as fermentacións anaerobias adoitan ser realizadas por unha comunidade microbiana mixta e non definida e xeralmente denomínanse fermentacións en cultivo mixto (FCM). En contraposición co uso de cultivos puros, os cultivos mixtos permiten empregar substratos non esterilizados como os residuos orgánicos, o que diminúe notablemente os custos de operación e permite procesos en réxime continuo. Ademais, os cultivos mixtos conteñen unha ampla variedade de funcións metabólicas, o que lles permite consumir substratos complexos que precisan varias transformacións e ser máis resistentes a cambios nas condicións ambientais. Nas FCM, un cambio nas condicións ambientais adoita provocar un cambio no espectro de produtos. Isto abre a porta a poder dirixir que AGV se producen no Resumo 6 proceso variando parámetros de operación. Na literatura científica téñense identificado certos parámetros que inflúen notablemente no espectro de produtos. Entre eles destacan a composición do substrato e en maior medida o pH do reactor. As proporcións entre os distintos substratos dunha fermentación inflúen sobre as estequiometrías individuais dos substratos xa que existen interaccións entre eles. O pH do reactor ten un alto impacto no espectro de produtos e un valor pode facer desaparecer produtos que son maioritarios noutras condicións. Ademais, esta variable ten o atractivo adicional de que é moi sinxela de manipular en calquera montaxe experimental. Polo tanto, unha das principais preguntas que trata de responder esta tese é ata que punto se pode controlar e dirixir a selectividade do proceso coas condicións de operación. Dado que o coñecemento mecanístico sobre a influencia destes parámetros é limitado, cómpre desenvolver ferramentas preditivas para coñecer os mecanismos que ligan as condicións ambientais e a estequiometría do proceso. Este coñecemento, ademais, proporcionaranos a capacidade de deseñar procesos altamente selectivos naqueles produtos desexados. Os modelos matemáticos son ferramentas que permiten interpretar e predicir diversos fenómenos naturais. Poden ser empregados para validar novas hipóteses ou para predicir e deseñar procesos, un aspecto cun gran atractivo xa que reducen notablemente a necesidade de traballo experimental. Deste xeito, o seu potencial para seren empregados en sistemas biolóxicos e, en particular, nas FCM é moi elevado. Na literatura científica xa existe un coñecemento axeitado, aínda que non completo, sobre a modelaxe da FCM de carbohidratos como a glicosa. Porén, o coñecemento sobre a FCM doutros tipos de substratos como as proteínas é moito máis limitado. O obxectivo principal desta tese é comprender os mecanismos que controlan as FCM e como están afectadas polas condicións ambientais empregando modelos matemáticos. O coñecemento adquirido e as ferramentas desenvolvidas tamén se empregarán co obxectivo de deseñar procesos de FCM cunha alta selectividade nos compostos desexados para facilitar o establecemento da biorrefinería como un esquema de produción viable técnica e economicamente. En particular, nesta tese, preténdese responder ás seguintes preguntas científicas: • Cal é o efecto da composición do substrato nas FCM? Cal é o espectro de produtos na fermentación de proteínas? Quedaría afectado no caso de cofermentar a proteína con outros substratos? Resumo 7 • Como afectan os parámetros de operación no resultado das FCM? Como se comporta o espectro de produtos a diferentes condicións de operación? • É posible dirixir o espectro de produtos das FCM cara produtos específicos? Ata que punto se pode modificar o espectro de produtos variando as condicións de operación e a composición dos substratos? Capítulo 2. Desenvolvemento do modelo Este capítulo céntrase na descrición detallada dos modelos matemáticos desenvolvidos nesta tese. Pódense distinguir pola súa estrutura e obxectivo dous tipos de modelo: modelos bioenerxéticos e modelos cinéticos. Os modelos bioenerxéticos teñen como obxectivo a predición da estequiometría do sistema a diferentes condicións de operación. O seu cometido é determinar cal é o espectro de produtos máis probable tendo en conta o substrato e as condicións ambientais do reactor. Para seleccionar os produtos óptimos, asumen que os microorganismos producen aqueles AGV que lles permiten obter a maior enerxía posible do substrato en forma de ATP. Son modelos baseados en primeiros principios e polo tanto teñen un número moi limitado de parámetros que depende do sistema en particular a simular. Os modelos cinéticos son aqueles que se centran en describir as dinámicas do proceso (a velocidade de produción dos diferentes AGV, as posibles inhibicións por produtos, etc.). Ó contrario que os modelos bioenerxéticos, a maioría dos seus parámetros son específicos para cada proceso e polo tanto necesitan ser calibrados con datos experimentais en cada sistema a simular (p. ex. se a proteína a simular é diferente ou mesmo se o pH do reactor é modificado). Capítulo 3. O mecanismo de bifurcación de electróns e a homoacetoxénese explican a produción en fermentacións anaerobias en cultivo mixto. A fermentación de glicosa en cultivos mixtos está estudada en profundidade na literatura científica tanto dende un punto de vista experimental como de modelaxe. En particular, o efecto do pH do reactor no espectro de produtos resulta de interese xa que exerce unha gran influencia. Os modelos matemáticos máis recentes para a predición da estequiometría do proceso son capaces de capturar correctamente os cambios observados experimentalmente no espectro Resumo 8 de produtos co pH. Deste xeito, predín unha transición dende ácido butírico a etanol e ácido acético cando o pH cambio de ácido a alcalino. Porén, aínda non son capaces de predicir con precisión o espectro observado experimentalmente. A pH baixos os modelos predín que a gran maioría da glicosa é transformada a ácido butírico, mentres que os experimentos mostran un espectro dominado por unha mestura equimolar de ácido butírico e acético. Este desfase nas predicións pode estar debido a que a rede metabólica que empregan os modelos non estea completa e actualizada. Neste capítulo inclúese na rede metabólica o mecanismo bioquímico bifurcación de electróns co obxectivo de comprobar se permite aumentar a precisión das predicións do modelo. Unha nova rede incluíndo este mecanismo comparouse coa rede que empregaban os modelos actuais. Para isto, desenvolveuse unha metodoloxía baseada en realizar balances de NADH a diferentes series de resultados experimentais na literatura, xa que unha rede correcta debe producir balances pechados de NADH. Os resultados indican inequivocamente que a inclusión da bifurcación de electróns na ruta de produción do butírico permite pechar mellor o balance de NADH en todos os datos analizados, indicando claramente que a nova rede é máis completa e precisa. A inclusión da bifurcación de electróns na rede da glicosa implica tamén que a relación predito polo modelo entre H2 e CO2 deixe de ser equimolar, xa que hai unha produción adicional de H2 na nova ruta do ácido butírico. Para conciliar este feito coas medidas experimentais gasosas que mostran unha relación equimolar, proponse que o proceso de homoacetoxénese (o consumo de H2 e CO2 para producir ácido acético) é a causa máis probable para explicar cuantitativamente os datos experimentais. Os resultados deste traballo indican que todas as rutas de produción de ácido butírico, independentemente do substrato do que partan, deben considerar a inclusión deste mecanismo bioquímico para poder predicir correctamente o espectro de produtos. Capítulo 4. A teoría de asignación de recursos explica a produción de ácido láctico en fermentacións anaerobias en cultivo mixto. O traballo deste capítulo céntrase na predición da produción en FCM de glicosa de ácido láctico, un dos ácidos carboxílicos máis producidos con aplicacións no eido da industria farmacéutica e alimentaria. Experimentalmente só se observou Resumo 9 a súa produción en FCM en condicións de operación moi específicas. O reactor ten que ser operado en modo descontinuo e no medio de cultivo cómpre que haxa aminoácidos e certas vitaminas presentes, xa que as bacterias acidolácticas son auxotróficas para estes compostos. Os modelos de predición de estequiometría en FCM asumen que os microorganismos xeran aqueles produtos que lles permiten obter a máxima produción de ATP por unidade de substrato. O lactato só produce 2 moléculas de ATP por cada molécula de glicosa convertida en comparación coas 3 ou 4 que se producen cando o produto é o ácido butírico ou acético, respectivamente. Dado o seu baixo rendemento en ATP, os modelos non o predín baixo ningunha circunstancia e non son capaces de reproducir o observado experimentalmente. Polo tanto, é obvio que ademais de motivos enerxéticos (rendemento en ATP) se necesitan outros argumentos para explicar mecanisticamente por que se produce lactato nesas condicións experimentais. Neste capítulo desenvolveuse un modelo matemático que inclúe conceptos da teoría de asignación de recursos (usualmente coñecida pola súa denominación en inglés: resource allocation). Esta teoría establece que os distintos procesos celulares (crecemento ou catabolismo) compiten por unha cantidade finita de recursos, entre os cales o máis limitado é a capacidade proteica ou encimática. O modelo predí que a produción de ácido láctico se favorece naqueles escenarios nos que a selección microbiana se fai de acordo á velocidade máxima de crecemento, é dicir, naqueles ambientes nos que hai unha concentración elevada de substrato, como en reactores en descontinuo ou en reactores en continuo cun tempo de dilución moi elevado, sempre e cando o medio de cultivo conteña aminoácidos e vitaminas. En calquera outras condicións ambientais, como nun reactor en continuo limitado por substrato, a produción doutros AGV como o ácido acético ou butírico é o resultado esperado da FCM de glicosa. O modelo é quen de identificar que a auxotrofía para aminoácidos das bacterias acidolácticas resulta ser unha vantaxe competitiva xa que lles permite atinxir unha velocidade de consumo de substrato específica máis elevada ó poder reasignar os recursos da produción de aminoácidos a outras tarefas. As conclusións deste capítulo permiten delimitar as condicións nas que optimizar exclusivamente en base ó rendemento en ATP dos produtos é unha estratexia correcta. Introduction 17 1.1. BIOREFINERIES FOR A CIRCULAR ECONOMY The economic and social era based on the massive use of fossils fuels that started in the industrial revolution is beginning its transformation in response to a future scenario of high prices, scarcity and geopolitical instability related with oil reserves (Landeweerd et al., 2011). In addition, social awareness of the effects of indiscriminate use of fossil fuels in the climate is steadily increasing and the society is already demanding alternatives. It is clear then that our production model should shift towards long-term sustainable options, among which, microbial production processes could play an important and primordial role. The current socioeconomical model is also eminently linear: raw materials are transformed into products that are disposed to the environment at the end of their lifespan, at best after treatment. Disposing endlessly wastes represents an equally worrying problem and it also concerns society. New models aiming at using fewer resources and at reusing wastes are desired and demanded. The biorefinery concept was created in this context of demand of a more circular society and economy. A biorefinery can be defined as “the sustainable processing of biomass into a spectrum of marketable products and energy” (de Jong and Jungmeier, 2015). In this production scheme, organic wastes or renewable feedstocks (e.g. wood) are used as substrates in different microbial processes to produce the final products (e.g. plastics, fuels, chemicals, electricity, etc.). In this sense, renewable raw materials are used creating effectively a matter and energy loop and imitating one of the many biogeochemical cycles of nature. Moreover, using organic wastes instead of crops or other agricultural products avoids competition with food purposes, avoiding any ethical conflict of the biorefinery concept. Including the word refinery in the term envisions one of the target characteristics of the process: flexibility. Biorefineries are conceived to be as flexible as oil refineries and produce a variety of final products from a reduced array of feedstocks through different processes upon demand. Several types of biorefinery can be distinguished among which the three most noteworthy are: the sugar biorefinery in which biomass is firstly hydrolysed chemically or enzymatically to sugars, the thermochemical biorefinery, in which synthesis gas is produced by biomass thermal gasification and the so-called carboxylate platform, in which biomass is firstly converted to short carboxylates by anaerobic fermentation. After an initial start, the sugar and thermochemical biorefinery are now focused on the use of lignocellulosic biomass as substrate (Octave and Thomas, 2009; Rasmussen et al., 2014), while the most recent Chapter 1 18 carboxylate platform is centred in treating organic wastes such as food waste or the organic fraction of the municipal waste (Lee et al., 2014; Strazzera et al., 2018; Venkata Mohan et al., 2016). In this sense, the carboxylate platform is a very attractive option since i) it can valorise those organic wastes and thus close the matter loop and ii) it can complement the other platforms by using their wastes, increasing thus their efficiency (Bastidas-Oyanedel et al., 2015). However, because of its shorter lifespan and due to the variability of the substrates used, the carboxylate platform is still a less mature technology than the other biorefinery types. Figure 1.1. Carboxylate platform biorefinery scheme. Organic wastes are converted to short carboxylates via anaerobic fermentation (gold arrow), which are further converted to the final high added-value products in subsequent processes (green arrows). In the carboxylate platform (Figure 1.1), the catabolic products of anaerobic fermentation can be further transformed in subsequent microbial processes to produce higher added-value products. For instance, VFA are used as substrate for the production of polyhydroxyalkanoates (PHA), a bioplastic family with some members (polyhydroxybutyrate) presenting mechanical properties similar to polypropylene (Kumar et al., 2019). Another example is the medium chain fatty acids (carboxylic acids of 6-12 carbon atoms) production in chain elongation processes using VFA and alcohols or lactate (Carvajal-Arroyo et al., 2019; Spirito Acetate Propionate Butyrate Valerate Ethanol Lactate Methane Electricity Medium-chain fatty acids Bioplastics Biodiesel Chain elgongation Organic wastes Introduction 19 et al., 2014). These carboxylates are valuable products and have application that include antimicrobial agents (Reddy et al., 2018), precursors of biofuels (Shilling et al., 2013) or monomers in bioplastics (medium-chain PHA) that can be used in medical applications (Rai et al., 2011). Moreover, the resultant longer carboxylates are easier to separate from the fermentation broth than VFA as they can be produced in concentrations close to the solubility of their undissociated form in water (Steinbusch et al., 2011). In this regard, short carboxylates act as a nexus between the organic wastes used as feedstock and the final biorefinery high added-value products. The process of anaerobic fermentation is pivotal within the carboxylate platform and forms part of the core of the biorefinery philosophy, as it allows the transformation of a waste into valuable chemicals which could be further transformed in subsequent processes to high added-value products. One of the main drawbacks of anaerobic fermentation is that usually the product spectrum consists of a mixture of VFA with a variable composition depending on both the operational conditions and the substrate composition. Therefore, developing predictive tools to understand the connection between operational conditions and stoichiometry and to design processes targeting specific VFA is a key element for the success of the carboxylate platform and is the main objective of this thesis. 1.2. ANAEROBIC FERMENTATIONS TO PRODUCE ENERGY-DENSE COMPOUNDS Anaerobic fermentation is a process forming part of the anaerobic food chain transforming organic matter into methane, carbon dioxide and ammonia, the final products of decomposition (Figure 1.2). In particular, anaerobic fermentation transforms the monomers (i.e. amino acids, glucose and long-chain fatty acids) of the organic polymers that compose organic matter (i.e. proteins, carbohydrates and lipids, respectively) into volatile fatty acids (VFA), alcohols, ammonia, hydrogen and carbon dioxide. Chapter 1 20 Figure 1.2. Anaerobic food chain. In aerobic processes microorganisms generally oxidise the substrate completely in their catabolism to carbon dioxide and water and transfer the electrons to oxygen, a strong electron acceptor (Figure 1.3). This electron transfer is very exergonic (i.e. releases Gibbs free energy) which allows microorganisms to harvest large amounts of energy from the substrate. As a result, biomass production is high and up to half of the substrate can be converted into biomass (Kleerebezem et al., 2015). Usually, there are no energetic barriers in the catabolism of aerobic microorganisms and the substrate is transformed to nonvaluable products. Therefore, aerobic processes are used when the desired product is produced in their anabolism, as for example alginate (Sabra et al., 2001), exopolysaccharides (Lin et al., 2010; Satpute et al., 2010), biosurfactants (Mnif et al., 2011; Satpute et al., 2010) or antibodies (Lee and Jeong, 2015). Biopolymers (carbohydrates, proteins and lipids) Monomers (monosacharydes, amino acids, long-chain fatty acids) VFA and alcohols AcetateH2, CO2 CH4, CO2 and NH4+ NH4+ Hydrolysis Acidogenesis Acetogenesis Homoacetogenesis Acetoclastic methanogenesis Hydrogenotrophic methanogenesis Introduction 21 Figure 1.3. Anaerobic and aerobic processes differ in the biomass yield from substrate due to the difference in energy production in catabolism. For a same substrate conversion flux, aerobic processes biomass flux is much higher than that of anaerobic processes. On the contrary, anaerobic fermentation processes are characterised by the lack of external and strong electron acceptors, which forces the substrate to be both the electron donor and acceptor. The driving force of the reactions is thus smaller, as the redox potential of the catabolic reactions is lower, and therefore microorganisms can harvest much less energy from the same substrate. In consequence, a larger amount of substrate is needed to harvest the energy required for biomass production (Figure 1.3). Catabolism in anaerobic fermentations takes place usually at conditions not far from the thermodynamic equilibrium, which leads to partial transformations of the substrate in which limited energy can be harvested (Jackson and McInerney, 2002a; Oh and Martin, 2007). These strong constraints led evolution to produce a plethora of possible catabolic reaction combinations to extract energy from the surroundings in the most efficient possible way. Therefore, anaerobic processes are more attractive when the targeted compounds form part of its catabolism (Borodina and Nielsen, 2014; Dharmadi et al., 2006; Jansen et al., 2017; Mans et al., 2018; Shen et al., 2011). Specifically, in anaerobic fermentation, the typical catabolic end-products are a mixture of energy-dense compounds consisting mainly of VFA and also alcohols and hydrogen (Agler et al., 2011; Angenent et al., 2004; Kleerebezem and van Loosdrecht, 2007). Substrate Energy-dense organic compounds Substrate Biomass Energy (ATP) Substrate CO2+ H2O Substrate Biomass Energy (ATP) Anaerobic fermentations Aerobic processes O2 CATABOLISM CATABOLISM ANABOLISM ANABOLISM Chapter 1 22 1.2.1. Volatile fatty acids as main products of anaerobic fermentations Anaerobic fermentation main products range from alcohols such as ethanol or butanol to medium-chain fatty acids as for example caproate; however, the main products are fundamentally VFA (Atasoy et al., 2018; Hoelzle et al., 2014). The term VFA refers to the carboxylic acids with a carbon-chain length between two and five carbons, that could also be referred as short carboxylates. Currently at an industrial level, VFA are usually produced from oil in the petrochemical industry and feature a wide variety of applications from polymer production to food preservatives and fuel additives (Table 1.1). The market price of the different anaerobic fermentation products is roughly proportional to the carbon chain length, being butyrate and caproate the most valuable products. In comparison, producing methane in anaerobic digestion from the same substrate appears as a less attractive option due to its lower market price and for being a gas (it hinders its transport). It is worth mentioning that anaerobic fermentation by pure cultures of microorganisms is already the principal method for producing at industrial level lactate and ethanol, highlighting the industrial and economic viability of microbial production processes. In the case of lactate, 90% of its production comes from bacterial fermentation (Alves de Oliveira et al., 2018) while more than 96% of the ethanol global market was fulfilled by fermentation in 2008 (Taherzadeh and Keikhosro, 2008). VFA production in anaerobic fermentation processes faces two main issues. First of all, the selectivity of the process for specific products is often low as VFA are usually produced in a mixture and at variable proportions with different operational conditions. In second place, separating the products from the broth is a challenging task due to typical low VFA concentrations encountered in anaerobic fermentations and because of the very similar chemical properties of VFAs (Zacharof and Lovitt, 2014). In fact, the separation and purification steps are considered to be the main contributors to the production cost of VFA in biological processes (Alkaya et al., 2009; Singhania et al., 2013). Introduction 23 Table 1.1. Market price and size, applications and production methods of typical products of anaerobic processes. Main production methods Natural deposits Petrochemical derivative (carboxylation of methanol)b,c Industrial fermentation (>95%)b and petrochemical derivate (ethylene oxidation, <5%) Petrochemical derivative (oxidation of butyraldehyde)c Industrial fermentation (pure culture)b Petrochemical derivate (ethylene hydro formylation, carboxylation of ethylene)b, c Applications Fuel, chemical industry. Production of vinyl acetate and derivates, pharmaceuticalsb, food additivec Solvent, antifreeze, fuel (supplement), raw chemicalb Cellulose-based plastics, food preservativeb, precursor of chemicals Polylactate productiond Cellulose-based plastics, feed preservative Overall market sizeb (ktonne/y) 14,000-17,000c 51,000 350-470c 1,220b 90-105c Market Pricea (USD/tonne) 200-600 400-800 700-900 1500-1700 1000-1600 2000-2500 Compound Methane Acetic acid Ethanol Propionic acid Lactic acid Butyric acid Chapter 1 24 Table 1.1 (continued). Market price and size, applications and current production method of typical products of anaerobic processes. Current production methods Petrochemical derivative (from propylene in the oxo process)f Petrochemical derivative (oxidation of n -pentanal)h Oleo-chemistry (from coconut or palm kernel oil)h a Bastidas-Oyanedel and Schmidt (2018); b Bastidas-Oyanedel et al. (2015); c Atasoy et al. (2018); d Alves de Oliveira et al. (2018), e (Liu et al., 2015), f (Dürre, 2008), g (Jänisch et al., 2019), h (Kubitschke et al., 2014) Applications Solvent (paints, dyes), chemical industry (methacrylate esters)f Ester-type lubricant, plasticizers, perfumes and cosmeticsg, synthetic flavourh Precursor of pharmaceuticals and flavoursb Market sizeb (ktonne/y) 2,800e - 25 Market Pricea (USD/tonne) 1800e - 2000-2500 Compound Butanol Valeric acid (pentanoic acid) Caproic acid (hexanoic acid) Introduction 25 1.3. MIXED-CULTURE FERMENTATIONS FOR VALORISING WASTES Anaerobic fermentations can be performed by pure cultures of microorganisms (commonly just one single species) or use a mixed microbial consortium. In this last case, the process is usually open (i.e. the system is open to the entry of microorganisms e.g. in the feeding), the microbial community is undefined and is referred to as mixed-culture fermentation (MCF). Pure single culture processes are characterised for their high product selectivity and good control with operational parameters (Bastidas-Oyanedel et al., 2015) (Figure 1.4). Owing to these advantages, well-established industrial bioprocesses usually employ pure of microorganisms. For example, in the ethanol and lactate industrial bioproduction pure cultures of the yeast Saccharomyces cerevisiae (Klein et al., 2019) and bacteria of the genus Lactobacillus or Lactococcus (Alves de Oliveira et al., 2018) are employed, respectively. Figure 1.4. Main characteristic of processes using pure and mixed cultures of microorganisms from an operational perspective. Adapted from Smid and Lacroix (2013). On the contrary, mixed cultures of microorganisms are more difficult to control accurately using operational parameters and the process outcome usually Pure culture Mixed culture Control Predictable Robustness Metabolic flexible Cost Chapter 1 32 control the energy and matter exchanges with the surroundings. In this way, cells isolate and modify aliquots of the environment to allow reactions that otherwise would not be possible or to avoid harmful conditions incompatible with life. The task of a bioprocess engineer is somehow a bit similar to that of cell membranes: to create specific environments to modify and guide intentionally cellular metabolism towards the production of specific products. The only way of domesticating cells in such a way without genomic modifications, and especially in open systems as MCF, is to modify the boundary conditions where cells live. In this way, the optimal strategy of survival may differ and could involve the production of other products. Following the well-known postulate of Bas Beckel “everything is everywhere but, the environment selects” (O’Malley, 2008), it is clear that the focus should be set in the operational conditions rather than in the microbial inoculum. Hence, to successfully engineer microbial communities the desired product must have a metabolic function or a selective trait that increases cellular fitness under some environmental conditions. If so, it will be theoretically possible to design a process in which the community produces substantially that desired product. For example, the driving force to storage carbon in the form of a biopolymer (i.e. bioplastic) is to consume, before other, all the available substrate when the conditions do not allow growth. In this sense, when the conditions allow again to grow, all the substrate is captured intracellularly and not available for competitors (Jiang et al., 2011). If bioplastic is the target compound, the system should be designed to enrich the mixed-culture with bioplastic-forming bacteria, with can be accomplished with feast and famine cycles, as bacteria storing the bioplastic have a fitness advantage. To establish the biorefinery scheme as a sound technology alternative, we need to understand and guide the microbial community of anaerobic fermentations at our own will. Therefore, one of objectives of this thesis is to develop predictive tools to i) understand the mechanisms linking environmental conditions and cellular metabolism (i.e. its stoichiometry) and with that insight ii) design processes targeting the production of the desired compounds. Introduction 33 1.5. MATHEMATICAL MODELS FOR UNDERSTANDING AND PREDICTING MICROBIAL METABOLISM A mathematical model is an abstract description of a system using mathematical language, aiming at understanding a system or predicting its behaviour (Ningthoujam et al., 2018). The use of mathematical models has spread in all branches of science as they are powerful tools for interpreting and predicting diverse natural phenomena and empirical observations (Gombert and Nielsen, 2000). Mathematical models can be employed to validate new hypotheses or can be used for their predictive capacity, in other words, for their design capabilities. In the latter case, mathematical models are especially attractive since they may be helpful to screen a priori operational conditions, thus reducing substantially the need of experimental work, which is usually time-consuming and expensive. All models are simplified visions of the reality, but this is a matter of necessity and convenience. In the case of mathematical models of living systems, this is already intuitive even for the simplest life forms (e.g. bacteria) as it is not possible to capture all possible processes due to our limited knowledge. However, the key element is that although models are conceptually wrong, in the sense that they cannot describe reality perfectly, they are built with the purpose of being useful for a particular task. Models should be developed for specific ranges of environmental conditions and, very importantly, with the adequate level of accuracy for each occasion, since adding additional layers complexity could difficult the identification of the cause-effect relationships. It is the task of the model developer to find a balance between model complexity and the applicability of the model and decide what mechanisms should be include depending on the purpose of the model. In this regard, the statistician George Box stated (Box, 1976): Since all models are wrong the scientist cannot obtain a "correct" one by excessive elaboration. On the contrary following William of Occam he should seek an economical description of natural phenomena. Just as the ability to devise simple but evocative models is the signature of the great scientist so over-elaboration and overparameterization is often the mark of mediocrity. He then perfectly synthesised his own words to create the well-known statement: “All models are wrong, but some are useful” (Box, 1979). Naturally, model use should be accompanied by a strong conscience of their range of application, Chapter 1 34 especially when extrapolating. Failing to identify the limits of model application imposed by assumptions or the non-inclusion of certain mechanism could lead to far greater inaccuracies than the oversimplification of reality themselves. Again, in words of George Box (Box, 1976): Since all models are wrong the scientist must be alert to what is importantly wrong. It is inappropriate to be concerned about mice when there are tigers abroad. 1.5.1. Modelling cellular metabolism In general, cellular metabolism can be schematically divided into anabolism and catabolism (Figure 1.5). In anabolism, new biomass is formed from the substrates in an energy-consuming process which decreases the entropy of the system as complex and ordered molecules are formed from simpler building blocks. The needed energy is provided in form of ATP by catabolism, which is the process in which cells obtain energy by converting the substrate to other simpler and more disordered compounds. Figure 1.5. Cellular metabolism is composed of catabolism and anabolism. The ATP produced in catabolism is consumed in anabolism to produce biomass from the substrate. Adapted from Kleerebezem and Van Loosdrecht (2010). The word anabolism means literally in Greek to throw upwards and was coined by physiologist W. H. Gaskell for the first time in 1876 with the interpretation of building up biomass compounds or a constructive metabolism (Gaskell, 1886). Catabolism, on the contrary, means to throw down and should be interpreted as breaking down the substrate or a destructive metabolism. Therefore, the Introduction 35 energetic and entropic nature of both processes is already perfectly captured in their etymology. Catabolism destroys or breaks down the substrate in smaller units to gain the energy that is needed to construct the complex biomass compounds from the substrate. In anaerobic environments, catabolic reactions are thermodynamically, and hence energetically, limited which leads to reduced biomass production, as mentioned in section 1.2. Additionally, different substrate conversions are usually very similar in terms of energy (González-Cabaleiro et al., 2015), which could explain the quite high variability of catabolic end-products found in MCF under different environmental conditions. This provides a great flexibility to the system as it allows it to produce different products from the same substrate. If a given end-product, e.g. propionate, provided significantly more energy to microorganisms, it would probably be more difficult to steer microbial production towards other end-products. Therefore, when modelling MCF the main focus should be set in catabolism and more specifically in the mechanisms responsible of the changes in production strategies. 1.5.2. Cells face an optimisation problem Cellular metabolism is assumed to be the result of adaptation to the environmental conditions and cellular surroundings. It could be assumed that only those microbial groups performing the optimal metabolic strategy prevail as only the fittest survive (Koonin and Wolf, 2012). Focusing on MCF, it could be argued then that the end-products representing the optimal catabolic strategy are likely to dominate the product spectrum of the process, since they allow the microorganisms that produce them to dominate the microbial community. Empirical observations indeed indicate that MCF global catabolism adapts to changing environmental conditions and, as a result, different end-products are produced by the mixed microbial community (section 1.3.1). The community dominant species may adapt to the new conditions and change their catabolic strategy or different species with a more suitable strategy can outcompete them and become the new dominant species. The adaptation and selection of microbial groups has been represented in mathematical models as an optimisation problem where the objective is to maximise the growth rate, or an equivalent expression, and the decision variable are the metabolic fluxes performed (Table 1.3). Physical and biological restrictions would act as constraints of the model and limit the possible solutions. For example, thermodynamics dictate what processes are Chapter 1 36 possible to occur (physical constraint) or an overall maximum intracellular concentration could be set to reflect a limit to osmotic pressure (biological constraint). Table 1.3. Cellular metabolism can be translated into a mathematical optimisation problem with a maximisation objective (to maximise growth rate) and additional constraints. Empirical observation Mechanistic interpretation Mathematical expression The fittest survive Growth rate is maximised max(µ) Biomass yield is maximised max(YX) ATP production is maximised while catabolic fluxes minimised max(rATP/∑ri) Physical constraints No external electron acceptor Electron balance in the system is zero ∑(rELECTRON) = 0 Second law of thermodynamics Globally endergonic reactions (j) are not possible If ΔGj ≥ 0, then rj = 0 Biological constraints Some intracellular compounds are at quasi steady state conditions Net fluxes of those compounds (i) are null Ni·r = 0 Intracellular osmotic pressure has a maximum Maximum intracellular concentration allowed If ∑Ci>Cmax rTRANSPORT = rPRODUCTION Kinetics dictate a minimum substrate concentration for a reaction to occur Minimum concentration of reactants If Ci<Cmin rj = 0 Usually the strategy maximising growth rate is interpreted to also provide the highest ATP yield with respect to the substrate and it is commonly used as optimisation factor (González-Cabaleiro et al., 2015; Rodriguez et al., 2006; Zhang et al., 2013). Nevertheless, there are other interpretations of what provides the highest possible growth rate (Schuetz et al., 2007). Biomass yield on the substrate is also proposed under the reasoning that evolution would have driven Introduction 37 cells to optimise the production of biomass per unit of substrate. Other approaches consider also kinetic arguments and propose that pathway length should be minimised (Melendez-Hevia, 1990) or that cells maximise the ATP production rate while minimising the overall catabolic fluxes (Dauner and Sauer, 2001). 1.5.3. Types of metabolic models Metabolic models can be classified depending on the description of the microbial community as unsegregated or segregated (Figure 1.6). Most of the models are unsegregated and considered each microbial group as a homogeneous population. On the contrary, segregated models, such as individual-based models, contemplate differences among individuals from the same population in, for example, size, shape or phenotype (Li et al., 2019; Millat et al., 2013).The application of these models is of interest when spatial distribution may influence metabolism as in biomass aggregates (e.g. biofilms) or when cells can adopt different states (e.g. vegetative or sporulated) as in ABE fermentations. Figure 1.6. Classification of metabolic models according to their structure and biomass description. Taken from Gernaey (2015) with permission from Elsevier Books. If in the system to model, more than one microbial species is present, there are distinct ways of describing the community and the interactions between species. Models can describe the mixed community as different compartments representing each group (with individual metabolic networks), in which case the Chapter 1 38 optimisation strategy is usually referred to the global community (Zhuang et al., 2011; Zomorrodi and Maranas, 2012). This approach is well suited to describe microbial communities consisting of up to a handful of different species and it particularly well-suited to capture microbial interactions beyond competition such as commensalism, parasitism or mutualism (Perez-Garcia et al., 2016; Zomorrodi et al., 2014). However, a precise knowledge about the metabolism (i.e. identifying the substrates and possible products of each of the species) is mandatory for applying a compartmentalised approach, which makes it an unattractive option for describing MCFs. An enzyme soup or lumped network approach ignores completely species boundaries and it assumes that all metabolites are shared equally among all community members (Biggs et al., 2015). This is equivalent to describing the microbiome as a single virtual microorganism owning all its metabolic functions and capabilities, which are condensed in a single lumped metabolic network. For its characteristics, this method was identified as being specifically appropriate for those occasions in which there is scarce knowledge about the community and the main objective is to explore the metabolic capacity of a specific microbial community and the interactions among community members are irrelevant (Biggs et al., 2015; Perez-Garcia et al., 2016; Taffs et al., 2009). These characteristics make the enzyme soup approach to be much more in line with the system of study, open MCFs, and with the main objectives of this work. Metabolic models can also be classified according to the level of detail with which microbial metabolism is described (Figure 1.6). On the one hand, unstructured models consider cells as black boxes and treat microorganisms just as another chemical compounds, albeit with (auto)catalytic properties. A good example is the well-known and widely-applied Anaerobic Digestion Model No. 1 (ADM1) for the simulation of the anaerobic digestion process (Batstone et al., 2002a). On the other hand, structured models describe intracellular compounds and kinetic processes separately from the reactor dynamics and may provide a greater level of mechanistic knowledge. However, as discussed at the beginning of this section, each model should have the needed degree of complexity for the purpose it was developed. Unstructured models are computationally much simpler and their level of detail is usually enough to describe accurately the kinetic behaviour of many microbial processes of interest such as anaerobic digestion. Introduction 39 1.5.4. State of the art The existing modelling works in literature concerning MCF can be classified into bioenergetic or kinetics models, according to the complexity and aim of the model Bioenergetics models have as central assumption that cellular metabolism is strongly shaped by bioenergetics. These models consider that cells pursue the maximisation ATP yield per substrate unit, as anaerobic fermentations are energy-scarce environments (González-Cabaleiro et al., 2013), and tune their catabolism accordingly. Their main objective is to predict the stoichiometry of the system under different environmental conditions and to understand the mechanisms that promote them. Consequently, they give great emphasis to describing with detail the different intracellular mechanisms (e.g. transport, homeostasis or energy conservation) and thus their predictions could be in principle extrapolated since they require a limited number of environmentspecific parameters. They are unsegregated but structured and describe the microbial community following the enzyme soup approach, as it is an appropriate description of poorly defined microbial communities as the microbiome of MCF. Kinetic models are unsegregated, unstructured and their objective is to describe and predict the general kinetic dynamics of a system. These models usually describe cellular metabolism in a simple way and do not consider possible changes in the stoichiometry with the operational conditions. Contrary to bioenergetic models, different parameters for different environments are needed and consequently they depend on calibration with experimental information in each condition. Scientific literature on MCF modelling (Table 1.4) has dedicated much more attention to carbohydrates, in special to glucose, and the number of bioenergetics models is significantly more reduced than kinetic models. While all bioenergetics models available in literature, to the best of my knowledge, are listed in Table 1.4, the kinetic models featured are just a selection. Chapter 1 40 Table 1.4. Previous modelling works. Bioenergetic models. Ref A B C D Comments pH effect on VFA and H2 selectivity is not well captured. Only one EC is considered. VFA undissociated form is used for transport calculations. Gaseous compounds prediction is improved. Butyrate shift at acidic pH is well captured but it is predicted to be the only products, which is not seen experimentally. It captures well the shift towards butyrate at acidic pH but ethanol production is never predicted at all, while reported experimentally. Transport energy cost considers only the dissociated form of VFA. It captures the observed shift from acetate-ethanol towards butyrate from alkaline to acidic pH values. However, at low pH butyrate is predicted to be the only product, when experiments show an equimolar production of butyrate and acetate Description First and seminal work on glucose MCF bioenergetic model ling for stoichiometry prediction. Modification of the previous model considering correctly the different EC. Modification of the first model (Ref. A) with updated metabolic network, including electron bifurcating mechanism. EC ratios are set by the H2 partial pressure. Most recent model for stoichiometry prediction of glucose MCF. Introduction 41 Table 1.4 (continued). Previous modelling works. Kinetic models. Ref E F G H I EC: Electron carrier. A: (Rodriguez et al., 2006); B: (Kleerebezem et al., 2008), C: (Zhang et al., 2013); D: (González-Cabaleiro et al., 2015); E: (Shi et al., 2019); F: (Bai et al., 2017); G: (Tommaso et al., 2013); H: (Infantes et al., 2012); I: (Fernández et al., 2011). Comments Only one EC is considered. It uses kinetic parameters estimated in methanogenic conditions. Methanogenesis cannot be ruled out. Operational conditions are poorly defined. Methanogenesis was not successfully inhibited. The stoichiometry is not considered to vary with the operational conditions Product selectivity is not considered Description Modification of ADM1 to simulate sucrose MCF with variable stoichiometry, which depends on the H2 partial pressure. Modification of ADM1 to simulate high-solid sludge MCF, focusing on ammonia inhibition. Kinetic model for the MCF of protein and its cofermentation with carbohydrates. Kinetic model for glucose MCF studying the effect of pH, total undissociated acid concentration and temperature. Kinetic model for glucose and fructose MCF to analyse the differences in kinetic behaviour Model development 49 2.1. INTRODUCTION In this section, the main predictive tools developed throughout the thesis are described. These mathematic models respond to different objectives: bioenergetic models have the purpose of predicting the process stoichiometry under different environmental conditions and kinetic models are responsible of describing the process the kinetic behaviour. Bioenergetic models are used in Chapter 5 and Chapter 6 and kinetic models in Chapter 7. In both cases the system to model is a continuous stirred tank reactor (CSTR) in which up to 3 compartments can be distinguished. The kinetic model considers the reactor bulk and gas headspace of the reactor. Bioenergetic models consider additionally the intracellular space, as they model in detail intracellular processes and transport processes between the bulk reactor and the intracellular matrix (Figure 2.1). Figure 2.1. System modelled with the four considered compartments: reactor headspace (gaseous compounds), bulk reactor, biomass and intracellular space. In the bioenergetic model the following intracellular processes are modelled: 1: Substrate transport from the bulk reactor to the intracellular space; 2: Active transport of products; 3: Homeostasis via sodium-proton pump; 4: Passive transport of products; 5: ATP production via proton translocation. The development of the bioenergetic models of this thesis (Chapter 5 and Chapter 6) follows the framework established in a previous contribution of a bioenergetic model for glucose MCF (González-Cabaleiro et al., 2015). The Gas compounds Qgas Qliq Qliq Bulk reactor Biomass Intracellular matrix Chapter 2 50 description of the main cellular processes (e.g. metabolite transport, reaction rate determination) is based on this previous approach with some substantial modifications in the implementation in MATLAB. The building of the metabolic network for protein fermentation and all the consideration regarding the structure of the microbial community and interaction among substrates in cofermentation scenarios are novel contributions of this thesis. In this section, firstly the bioenergetic model is described in detail. The mass balances of the system are presented followed by a discussion of the most relevant hypotheses of the model. Then the metabolic network for glucose and AAs is presented and its construction (e.g. electron carrier election, energy conservation methods) explained. Finally, the architecture of the model is described and the optimisation problem defined. The processes included in the kinetic model and their description are discussed at the end of this section. 2.2. BIOENERGETIC MODEL 2.2.1. Model structure and mass balances The model development is based on the approach used by González-Cabaleiro et al. (2015) for building a glucose fermentation model. The model is built on the mass balances in a continuous stirred tank reactor (CSTR) of the different compounds (states) (Eq. 2.1-2.4). The model defines three compartments: intracellular (where the reactions occur), extracellular (i.e. reactor bulk) and gas phase. The intracellular and extracellular compartments exchange compounds through passive and active transport, and the transfer between the bulk reactor and the gas phase occurs via liquid-gas transfer. Among the states, 3 are moieties related with ATP (ATP, ADP and Pi). The rest represent the concentration of different intracellular compounds (24), extracellular compounds in the bulk reactor (40), gaseous compounds (3) and biomass. NAD+ to NADH ratio is set fixed to a value of 10 and the intracellular substrate concentrations are assumed constant at a value of 0.1 mM and therefore they are not states. The mass balances of the system are defined by the following equations (Eq. 2.12.4). Model development 51 Intracellular compounds 𝑑𝑆𝑖𝑛 𝑑𝑡 =𝑅𝑖+𝑅𝑇,𝑖𝑛 (2.1) where Sin is the intracellular concentration (mol LX-1), Ri and RT,i are, respectively, the reaction rate and intra-extra cellular transport rate (mol Lx-1 h-1). Intracellular concentrations and processes are referred with respect to the biomass volume (LX). Extracellular compounds 𝑑𝑆𝑒𝑥 𝑑𝑡 =𝐷𝑙𝑖𝑞·(𝑆𝑒𝑥,𝑖𝑛𝑙𝑒𝑡−𝑆𝑒𝑥)+𝑅𝑇,ex (2.2) where Sex is the extracellular concentration (mol Lliq-1), Dliq is the dilution rate of the liquid fraction (h-1), Sex,inlet is the concentration on the inlet (mol L-1) and RT,j is the intra-extra cellular transport rate (mol Lliq-1 h-1). Extracellular concentrations and processes are referred with respect to the liquid volume of the reactor (Lliq), i.e. not considering the biomass volume. Biomass 𝑑𝑆𝑋 𝑑𝑡=−𝐷𝑙𝑖𝑞·𝑆𝑋+𝑅𝑎𝑛𝑎−𝑅𝑑𝑒𝑐𝑎𝑦 (2.3) where SX is the biomass concentration (mol Lr-1), Rana is the anabolism rate and Rdecay the decay rate. The biomass-related concentration and processes are referred with respect to the complete working reactor volume (Lr). Gas compounds 𝑑𝐺𝑚 𝑑𝑡 =−𝐷𝑔𝑎𝑠·𝐺𝑚+𝑅𝑇,𝑚 (2.4) where Gm is the concentration (mol Lgas-1), Dgas is the gas space dilution rate (h-1) and RT,m is liquid-gas transport rate (mol Lgas-1 h-1). Gas-related concentrations and processes are referred with respect to the headspace volume of the reactor (Lgas). Hereafter, in the next section, the determination of the reaction and transport rates is described in detail. 2.2.2. Model Hypotheses 2.2.2.1. Fermenters are efficiency-driven systems Anaerobic fermentations carried out in conditions of substrate scarcity (as in a properly operated continuous reactor) are low-energy environments (GonzálezCabaleiro et al., 2013; Hoehler and Jørgensen, 2013; Jackson and McInerney, Chapter 2 52 2002b; LaRowe et al., 2012). Under these conditions, microorganisms behave as efficient energy scavengers and microbial groups capable of harvesting the most energy from the substrate (as ATP) will likely dominate the community. In other words, the microbial competition in substrate limitation conditions is set on efficiency rather than on speed. It is expected then that kinetic differences among the branches consuming a particular substrate (e.g. glucose or the different amino acids of a protein) do not play an important role in cellular metabolism and, in consequence, their kinetic description is alike. 2.2.2.2. The microbial community of mixed-culture fermentations is described as an enzyme soup It is considered that there is a population of one virtual microorganism capable of performing all the theoretical metabolic pathways. This approach assumes that all intracellular metabolites are always available for all routes or, equivalently, that the ability of performing determined pathways is equally distributed across the microbial populations. This approach was termed as enzyme soup in opposition to compartmentalized approaches that model the different microorganisms separately and where the boundaries between community members shape the model solution (Bauer and Thiele, 2018; Biggs et al., 2015). The enzyme soup approach is appropriate for those systems in which there is limited a priori knowledge about the microbial consortia, such as MCF. Moreover, the communities of such systems are changing continuously (even when the system is at macroscopic steady state) as a result of function redundancy among the species and due to the supply of new microorganisms in the feeding (Carballa et al., 2015; Fernández et al., 1999). In this model, the emphasis is set on exploring the metabolic potential of complex microorganism consortia and not on the interactions between species or with the environment. 2.2.2.3. Some amino acids may be consumed incompletely Substrate conversion can be limited when its consumption is not energetically feasible or beneficial to the microbial consortia. Although in glucose fermentation the substrate is usually fully converted, protein conversion may be incomplete. Some AA may reach thermodynamic barriers and their degradation pathways result in endergonic reactions under typical intracellular conditions. Experimental evidence indeed shows that it is frequent that proteins are not fully degraded in fermentations (Breure et al., 1985; Breure and van Andel, 1984; Fang and Yu, 2002; Ramsay, 1997; Yin et al., 2016). Consequently, the model can choose to not Model development 53 consume specific AA completely or partially. Cells will not consume a particular AA if all degradation pathways are overall endergonic. Also, an AA could be not completely consumed even if its degradation is exergonic just because cells cannot conserve energy from its degradation. 2.2.3. Metabolic network 2.2.3.1. Electron carriers Reduction and oxidation reactions happening in the metabolic network involve different electron carriers (ECs). In the proposed network we consider two of them: ferredoxin (Fdred/Fdox) and NADH (NADH/NAD+). The couple Fdred/Fdox is characterized by its low redox potential –E0’ ≈ - 400 mV1 and E’ ≈ - 500 mV2 (Buckel and Thauer, 2013)– and because it is the only EC capable of reducing protons to H2. It is only considered to be reduced in high exergonic reactions due to its low reduction potential as, for example, decarboxylations. To regenerate the oxidised form (Fdox) two options are possible: either the Fdred is oxidised reducing protons to produce H2 by cytoplasmic ferredoxin:proton reductases (Ech) or formate is produced when CO2 is reduced instead by ferredoxin:CO2 oxidoreductases. The production of H2 and CO2 versus formate is in a thermodynamic equilibrium ruled only by the pH (González-Cabaleiro et al., 2015; Hoelzle et al., 2014; Temudo et al., 2007). The couple NADH/NAD+ has a higher redox potential than Fdred/Fdox –E0’ = - 320 mV (White et al., 2012), E’ = - 280 mV (Buckel and Thauer, 2013)– and it is involved in most of the redox reactions occurring in the metabolic network (González-Cabaleiro et al., 2015; Kleerebezem et al., 2008). 2.2.3.2. Energy conservation It is assumed that microorganisms conserve energy in two ways. Through substrate level phosphorylation (SLP) microorganisms can yield ATP, transferring a phosphoryl group (PO43-) from a metabolite to ADP. Alternatively, microorganisms can extrude one proton outside the cells against its electrochemical gradient -also termed proton motive force (pmf), when coupled to sufficiently exergonic reactions. That same proton will be used to produce 1 E0’ is the redox potential of a reaction at biological standard conditions, i.e. at pH 7 and metabolic concentrations of 1 mM. 2 E’ is the redox potential of a reaction corrected to the typical concentrations found in microorganisms. Chapter 2 54 ATP when returning to the cytoplasm down its electrochemical gradient. Nevertheless, extruding a proton requires a complex enzymatic machinery, meaning that there is not a proton extrusion in all the steps that energetics would theoretically allow for it. Proton extrusions are only allocated in the metabolic network is places where experimental evidence was found. Usually, to introduce a phosphoryl group in an organic molecule, it must be previously activated with the cofactor coenzyme-A (CoA-SH), forming a thioester. Then the CoA cofactor is swapped by a phosphoryl group and is eventually transferred to an ADP molecule to give ATP (the SLP process). However, the activation reaction is endergonic (+30-50 kJ/mol) and therefore must be linked to exergonic reactions, as for example decarboxylations. When there is no exergonic reaction to link the CoA activation, the CoA group is transferred directly from other metabolite containing it. In this case the donor molecule loses the possibility of conserving energy by SLP. This consideration is important when constructing the network as the number of ATP molecules yielded in each pathway has a big impact in the model solution. 2.2.3.3. Glucose metabolic network The metabolic pathways of the major products observed when glucose is fermented by open microbial communities are included in the network (Figure 3.1). The products considered in the glucose metabolic network are: acetate, ethanol, propionate, lactate and butyrate (Angenent et al., 2004; Fang and Liu, 2002; González-Cabaleiro et al., 2015; Hoelzle et al., 2014; Horiuchi et al., 2002; Lengeler et al., 1999; Mohd-Zaki et al., 2016; Rodriguez et al., 2006; Temudo et al., 2008b, 2007; Zhang et al., 2013; Zoetemeyer et al., 1982). All these products derive from a common glycolytic process (glucose conversion to pyruvate), in which the conversion of glucose starts with the Embden-Meyerhof-Parnas pathway (EMP) producing two moles of pyruvate per mole of glucose. Two other degradation pathways of glucose can be found in prokaryotes microorganisms: Entner-Doudoroff (ED) and Pentose-Phosphate pathway (PPP). ED pathway is reported to be related with aerobic environments and to be used to mainly metabolise sugar acids (Peekhaus and Conway, 1998). PPP pathway is mainly used for anabolic purposes (i.e. to obtain the needed metabolites in biomass generation) (Kruger and Von Schaewen, 2003). But the fact that EMP pathway has an ATP yield twice as big as the ED pathway strongly suggests selecting EMP as the glycolysis route in our metabolic network. These reasons, and the fact that in literature the EMP pathway for describing glycolysis is ubiquitous (Buckel and Model development 55 Thauer, 2013; González-Cabaleiro et al., 2015; Hoelzle et al., 2014; Kleerebezem et al., 2008; Mosey, 1983; Rodriguez et al., 2006; Temudo et al., 2009, 2008b, 2007; Zhang et al., 2013; Zoetemeyer et al., 1982), led us to only consider the EMP pathway. The three-carbon compounds, propionate and lactate, are directly yielded from pyruvate. Acetate, ethanol and butyrate (compounds with an even number of carbons) need a first decarboxylation of pyruvate to yield acetyl-CoA, in which an equimolar mixture of H2 and CO2 or formate is produced via Fdred. Figure 2.2. Selected metabolic network of glucose fermentation for mixed cultures under fermentative environmental conditions. ►CoA Succinate ATP ▼ ►ATP NADH ► NADH ► 2NADH ► H2 Ethanol Glucose Acetate Propionate Propionyl-CoA Acetoacetyl-CoA Crotonyl-CoA Butyryl-CoA Acetaldehyde Pyruvate Fumarate Acetyl-CoA Oxaloacetate Lactate Acetyl-P 2x Butyrate CO2 CO2 CO2 0.5x ►2NADH ►2 NADH ►2 ATP ►ATP ►ATP CoA ► ►CoA ►CoA ►CoA NADH ► ▲CoA NADH ► NADH ▼ ►Fdred Fdred ◄ CoA ► Chapter 2 56 Crotonyl-CoA is an intermediate in the butyrate yielding pathway. The crotonylCoA reduction with NADH to butyryl-CoA is a very exergonic step (ΔG’m=-57 kJ/mol), indicating that it is probable that energy is conserved in this step. Literature indeed reports energy conservation via proton translocation (Buckel and Thauer, 2013; González-Cabaleiro et al., 2015; Li et al., 2008), but also indicates that another biochemical mechanism might take place: electron bifurcation (EB). EB is a mechanism by which cells can couple endergonic reactions with sufficiently exergonic ones (Buckel and Thauer, 2018; Peters et al., 2016). It was experimentally detected that in this step the exergonic crotonylCoA reduction is coupled to the endergonic reduction of Fdox by NADH, which eventually will yield H2 (Li et al., 2008). Both possibilities are included in all the crotonyl-CoA reductions featured in the metabolic network. In chapter 3 the inclusion of EB in the metabolic network is studied and the results prove indeed that the consideration of EB reduces the discrepancies between the predicted and experiments results of glucose MCF. Fumarate reduction to succinate is a similar metabolic step to crotonyl-CoA reduction in terms of energetics and reaction mechanisms (a hydrogenation of a double carbon bond). However, only energy conservation via proton translocation is reported in literature and therefore EB is not placed as an option in this step (Buckel, 2001; Herrmann et al., 2008). 2.2.3.4. Protein metabolic network The protein bioenergetic model is the first attempt to describe in detail the fermentation of all the AA that conform proteins, to the best of my knowledge. As a consequence, its metabolic network had to be built from scratch integrating many different literature sources for each considered AA. In all cases, only reactions likely to occur in fermentative environments were included in the metabolic network The metabolic network used in the model is formed by the degradation pathways of 17 amino acids (AAs): alanine, arginine, asparagine, aspartate, cysteine, glutamate, glutamine, glycine, histidine, isoleucine, leucine, lysine, methionine, proline, serine, threonine and valine. AA containing aromatic side chains were not included in the metabolic network (phenylalanine, tyrosine and tryptophan) because their degradation pathways are not sufficiently clear on literature (Andreesen et al., 1989; Barker et al., 1987). Besides, these AA do not account for more than 10% (molar basis) in the usual proteins found in wastes (9.2% in Model development 57 casein, 9.4% in gelatine, 8.8% in albumin, 7.4% in gluten, 3.7% in keratin and 6.9% in zein). The products covered are fatty acids from C1 to C6, ethanol, formate, methanethiol, hydrogen sulphide, CO2 and H2. Butyrate and valerate are present in both their linear and branched form and in the case of caproate only the branched appears as a possible product in the metabolic network. Most of the routes are adapted from Andreesen et al. (1989), Barker (1981) and Fonknechten et al. (2010). Some AAs are interconverted to others instead of being converted to VFA. This is the case, for example, of glutamine and asparagine, that are the amides of glutamate and aspartate, respectively. In this case we considered that the common AA acts like a node, and the degradation pathways of these AA end in another AA and not in the final products (i.e. VFA). Other AAs have pyruvate as an intermediate in their conversion to VFA. In these cases, the conversion of pyruvate is assumed to follow the metabolic network of glucose from pyruvate on (Figure 2.2). Alanine (Ala) Alanine is deaminated to pyruvate by direct oxidation via NAD-dependent alanine dehydrogenase. Pyruvate is then converted following Figure 2.2. Aspartate (Asp) Aspartate can be oxidatively deaminated with NAD+ to yield oxaloacetate, which is then decarboxylated to pyruvate. Aspartate can be as well deaminated to yield fumarate, which is further reduced with NADH to succinate (Unden et al., 2013). Succinate can be either an end product or further catabolised to yield propionate and CO2 with concomitant ATP formation. Arginine (Arg) Arginine is first deaminated to citrulline, which is decomposed into carbamoyl-P and L-ornithine (Figure 2.3). Carbomoyl-P is a compound that releases bicarbonate, ammonium and ATP when decomposed enzymatically, providing the cell with ATP in the hydrolytic step. L-ornithine is as well an AA but, as it only acts as an intermediate in Arginine degradation, it is not considered as a starting AA in the network. L-ornithine has two possible degradation pathways. The first one consists in a deamination to L-proline, racemization to D-proline and then it follows the proline usual degradation pathway (described later). The other option yields D-alanine, acetyl-CoA and NADH (Uematsu et al., 2003). Chapter 2 64 Acid 𝐻𝐴 ↔𝐴−+𝐻+ Base 𝐵+𝐻+↔𝐵𝐻+ The charge contribution of each compound is added up in a global charge balance (Eq. 2.5). The resulting non-linear equation is solved following the Newton-Raphson method, thus calculating the proton concentration and pH. 𝐹(𝐻+)=∑𝑓𝑖(𝐻+)=0 𝑖 (2.5) where, F is the charge balance and fi are rational functions describing the equilibrium and charge balance of each acid/base compound. 2.2.4.2. Thermodynamic feasibility The thermodynamic feasibility of the pathways should be addressed, as some pathways may be endergonic under particular environmental conditions. In these cases, the model prevents such pathways using as indicator their change of Gibbs free energy. The standard Gibbs energy of a reaction (ΔG0, at standard conditions) is determined with the Gibbs energy of formation of the compounds involved, which are obtained from literature (Alberty, 2010, 2006; Hanselmann, 1991; Thauer et al., 1977) or by using the group contribution method (Flamholz et al., 2012). This value is corrected in each simulation step with the current chemical activities of the compounds involved in the reaction, obtaining thus the actual ΔG’ (Eq. 2.6). To determine the activity the Debye-Hückel law is followed. ∆𝐺′=∆𝐺0+𝑅𝑇·ln𝑄 (2.6) where ΔG’ is the actual Gibbs energy of reaction, ΔGº is the standard Gibbs energy of reaction, R is the gas constant, T is the temperature and Q is the reaction quotient. For each of the degradation pathways and at each time step their ΔG is calculated and their feasibility factor (f) is determined. This factor is a step function that varies between 0, when the reaction is endergonic and should not happen, and 1, when the reaction is sufficiently exergonic and can run without limitations. A minimum feasible value for ΔG’ of -2 kJ/mol is assumed to consider a reaction to run, which is a value used previously (González-Cabaleiro et al., 2015). All step function used in the model are expressed as derivable functions and follow the next general form: Model development 65 𝑓(𝑥)=𝑎·tanh(𝑥+𝑏)+𝑐 (2.7) where a, b and c are constants to modify the shape and limits of the resultant curve and x is the function input. Given that this factor varies continuously between zero and one, there are values of ΔG close to -2 kJ/mol that result in intermediate values, meaning that the reaction does run but at a lower rate. This is in accordance with LaRowe et al. (LaRowe et al., 2012). 2.2.4.3. Kinetics As previously stated (section 2.2.2.1), kinetic differences in the pathways of the same kind of substrate (e.g. carbohydrates or protein), are not expected to shape cellular metabolism in MCF. A tight intracellular regulation is assumed and hence all catabolic reactions arising from a substrate are assumed to have the same rate as the consumption rate of that substrate. The uptake rate of each individual substrate (i.e. glucose and AA) is modelled with a Monod-like equation (Eq. 2.8) with common parameter values for each kind of substrate. Therefore, there are as many independent substrate uptake rates as number of individual substrates in the reactor. 𝑟𝑆,i=𝑟𝑆,i𝑚𝑎𝑥·𝑆𝑒𝑥,𝑖 𝐾𝑆,𝑖+𝑆𝑒𝑥,𝑖 ( 𝑚𝑜𝑙 𝑖 𝑚𝑜𝑙 𝑋·ℎ) (2.8) where rSi is the uptake rate of the ith individual substrate, rSmax is its maximum uptake rate, Sex,i is the total bulk concentration of the ith individual substrates and KS,i is the affinity constant. Subsequent reactions within the pathways of each individual substrate are modelled to have the same rate as the consumption rate of the substrates they originated (Eq. 2.9). 𝑅𝑖=𝑓𝑖·𝜈𝑖·𝑟𝑆,i·𝑧𝑖·𝑆𝑋·𝑉𝑅 𝑉𝑋 (𝑚𝑜𝑙 𝑖 𝐿𝑥·ℎ) (2.9) where, for the ith reaction, Ri is the reaction rate, fi is the feasibility factor of the reaction, νi is the stoichiometry factor between the reaction and the substrate uptake, zi is the reaction selection parameter (described in section 2.2.5), SX is the biomass concentration, VR is the reactor volume and VX is the biomass volume in the reactor. Chapter 2 66 2.2.4.4. Anabolism and Decay The biomass is considered homogeneous and described with a general molecular formula of CH1.8O0.5N0.2. It may be formed from glucose, AA and ammonia (when glucose is the only substrate of the system). Anabolism for glucose follows the stoichiometry of Eq. 2.10. A lumped stoichiometry with a certain degree of decarboxylation is used to describe anabolism (González-Cabaleiro et al., 2015; Tobajas and Garcia-Calvo, 1999). The degree of decarboxylation varies with the substrate and is assumed to be dependent on the heat of combustion of the substrate (Gommers et al., 1988), which in turn can be easily correlated with the degree of reduction (Gary et al., 1995). For glucose the degree of reduction value is 4, which leads to a carboxylation degree of 15% (in terms of moles of carbon of the substrate). The ATP needed to form biomass is set to 2 mol ATP/C-mol biomass, as in previous approaches (González-Cabaleiro et al., 2015). 𝐺𝑙𝑢𝑐𝑜𝑠𝑒+1.02 𝑁𝐻4++9.45 𝐻2𝑂+10.2 𝐴𝑇𝑃→ →5.0 𝑋+0.90 𝐶𝑂2+ 12.1 𝐻++10.2 𝐴𝐷𝑃+10.2 𝑃𝑖 (2.10) When protein is the sole substrate, biomass is assumed to be formed from an equimolar mixture of the 17 different AA because any more specific information about AA proportions in typical biomass was not available (Eq. 2.11). To make a general anabolism description that could be used for different proteins as substrate, a mean degree of reduction value of 4 for all AAs was chosen, which leads to the same decarboxylation degree as in the anabolism from glucose (15%). The ATP needed to form biomass is set to 2 mol ATP/C-mol biomass, as in anabolism from glucose, because specific information for biomass growth on protein could not be found (González-Cabaleiro et al., 2015; Tobajas and GarciaCalvo, 1999). ∑(1 17·𝐴𝐴𝑖) 𝑖=𝐴𝐴 +9.34 𝐻2𝑂+7.80 𝐴𝑇𝑃→ → 3.90 𝑋+0.69 𝐶𝑂2+0.69 𝑁𝐻4++10.0 𝐻+ +7.80 ATP+7.80 𝑃𝑖 (2.11) For decay processes the considered stoichiometry is the same but in the forwards direction. If protein is the only substrate, biomass decay is also considered to produce glucose (Eq. 2.10 in the forward direction), since polysaccharides are as well one of the main constituents of biomass (Stouthamer, 1973). Model development 67 The participation of electron carriers in anabolism and decay is neglected to simplify the modelling procedure but this is not expected to compromise the results of the model due to the generally modest biomass yields in anaerobic processes and because the degree of reduction of biomass and the substrates are very similar. Anabolism and decay rates are modelled to depend on energy availability, determined by the Gibbs energy of ATP formation (Eq. 2.6) in each simulation step, as an indirect indicator of the availability of ATP. A value higher than 50 kJ/mol ATP indicates that cells have enough energy to grow as it indicates that the ATP concentration is high. If the value is lower than 50 kJ/mol ATP, decay processes take place to regain energy and raise ATP concentration. Anabolism rate is described with a Monod-like equation (Eq. 2.13), in which the maximum rate term is variable and dependent on energy availability (Eq. 2.12). In the case of anabolism from protein, the substrate is considered to be the sum of the 17 AA in carbon molar basis. 𝑘𝑎𝑛𝑎=∆𝐺𝐴𝑇𝑃−50 5 (𝑚𝑜𝑙 𝑋 𝑚𝑜𝑙 𝑋·ℎ) (2.12) 𝑅𝑎𝑛𝑎=𝑘𝑎𝑛𝑎·∑𝑆𝑒𝑥,𝑖𝑖 𝑀𝑎𝑛𝑎+∑𝑆𝑒𝑥,𝑖𝑖 ·𝑆𝑋 (𝑚𝑜𝑙 𝑋 𝐿𝑟·ℎ) (2.13) where kana is the anabolism maximum rate, ΔGATP is the Gibbs energy of ATP formation, Rana is the anabolism rate, Sex,i is the extracellular concentration of the ith substrate (in carbon molar basis) and Mana is the anabolism affinity constant (6·10-3 Cmol/L). In cofermentation scenarios, anabolism occurs from glucose and protein and their anabolism rates are determined by Eq. 2.13 using the same parameters. A different rate can only be provoked by a different availability of the substrates through the Monod term. Decay rate is only controlled by the Gibbs energy of ATP formation (Eq. 2.14). 𝑅𝑑𝑒𝑐𝑎𝑦=50−∆𝐺𝐴𝑇𝑃 5·𝑆𝑋 (𝑚𝑜𝑙 𝑋 𝐿𝑟·ℎ) (2.14) 2.2.4.5. Transport Cells have evolved semi-permeable membranes that allow for both uncontrolled passive transport and controlled active transport. Uncharged molecules diffuse freely through the membrane, as for example the protonated form of VFA. Charged molecules and big neutral molecules (e.g. glucose or AA) cannot diffuse Chapter 2 68 across the membrane and must be transported actively by a wide set of channel proteins that are controlled by the cell (White et al., 2012). Passive transport (Eq. 2.15) is energetically uncoupled to microorganisms and is governed by Fick’s Law (i.e. transport follows the concentration gradient of each compound). There is little data available in literature about diffusion coefficients (kDiff) for the different compounds present in the model but values used in previous modelling works are in the same order of magnitude (GonzálezCabaleiro et al., 2015; Rodriguez et al., 2006; Zhang et al., 2013). Following González-Cabaleiro et al. (2015) approach, all diffusion coefficient are set to 100 L/molX·h, as the main divergencies among the transport rates will be caused by the differences in their acidification degree (i.e. proportion between the charged and uncharged form of a molecule following acid-base equilibrium), which are already accounted for in the model (section 2.2.4.1). 𝑅𝐷𝑖𝑓𝑓,𝑖=𝑘𝐷𝑖𝑓𝑓·(𝑆𝑒𝑥,𝑖𝑢𝑛𝑐ℎ𝑎𝑟𝑔𝑒𝑑−𝑆𝑖𝑛,𝑖𝑢𝑛𝑐ℎ𝑎𝑟𝑔𝑒𝑑)·𝑋·𝑉𝑅 𝑉𝑋 (𝑚𝑜𝑙 𝑖 𝐿𝑥·ℎ) (2.15) where, for the ith compound, RDiff,i is the passive transport rate, kDiff is the diffusion coefficient, Sex,iuncharged is the extracellular uncharged concentration and Sin,iuncharged is the intracellular uncharged concentration. Active transport, on the contrary, is coupled energetically to metabolism and can be performed against or in favour of the electrochemical gradient and hence be an energy expenditure or gain, respectively. These channel proteins (or ports) are modelled to be coupled to proton translocations. Negatively charged molecules (e.g. anions of organic acids) are transported to the extracellular matrix by a symporter with protons and positively charged molecules (e.g. ammonium) are transported in antiport with protons, to make the process transport electrically neutral. As an enzymatically controlled process, active transport is also modelled with a Monod-like equation (Eq. 2.16). We assumed that the maximum active transport ratio is equal to the maximum production rate of that compound being transported (Eq. 2.17). Finally, Eq. 2.18 determines the active transport rate. 𝑟𝐴𝑐𝑡,𝑖=𝑟𝐴𝑐𝑡,𝑖𝑚𝑎𝑥·𝑆𝑖𝑛,𝑖 𝐾𝑇+𝑆𝑖𝑛,𝑖 ( 𝑚𝑜𝑙 𝑖 𝑚𝑜𝑙 𝑋·ℎ) (2.16) 𝑟𝐴𝑐𝑡,𝑖𝑚𝑎𝑥=∑𝜈𝑖,𝑗·𝑟𝑆,𝐴𝐴𝑚𝑎𝑥 𝑗=𝐴𝐴 ( 𝑚𝑜𝑙 𝑖 𝑚𝑜𝑙 𝑋·ℎ) (2.17) Model development 69 𝑅𝐴𝑐𝑡,𝑖=𝑟𝐴𝑐𝑡,𝑖·𝑋·𝑉𝑅 𝑉𝑋 (𝑚𝑜𝑙 𝑖 𝐿𝑥·ℎ) (2.18) where, for the ith compound, rAct,i is the active transport rate, rAct,imax is the maximum active transport rate, Sin,i is the intracellular concentration, KT is the active transport affinity constant (0.15 M) and νi,j is the stoichiometric coefficient of compound i in the degradation reaction of AA j. Intracellular metabolic concentrations above 10 mM are considered not physiologically compatible (González-Cabaleiro et al., 2013) and rarely measured above this value (Bar-Even et al., 2012). When the concentration of one actively transported compound reaches this value, active transport rate is calculated with Eq. 2.19, preventing thus a higher accumulation. 𝑅𝐴𝑐𝑡,𝑖=𝑅𝑖−𝑅𝐷𝑖𝑓𝑓,𝑖 (𝑚𝑜𝑙 𝑖 𝐿𝑥·ℎ) (2.19) Finally, total transport is the sum of both transport mechanisms (Eq. 2.20). 𝑅𝑇,𝑖=𝑅𝐷𝑖𝑓𝑓,𝑖+𝑅𝐴𝑐𝑡,𝑖 (𝑚𝑜𝑙 𝑖 𝐿𝑥·ℎ) (2.20) where RT,i is the total transport rate. Substrate molecules are modelled to be transported inwards by active transport because they are either electrically charged and/or are not small enough to freely diffuse across cell membranes. However, it is not clear how their transport is coupled to the energetics of microorganisms. Studies are more focused on describing the characteristic of the ports and how they are controlled than on how transport is coupled with the energetic part of metabolism (Berger, 1973; Guidotti et al., 1978; Heyne et al., 1991; Meister, 2016; Oxender and Christensen, 1963; Poole, 1978). For example, some AA transport mechanisms appear to be stoichiometrically linked to Na+ or proton intrusion but it is not clear whether they use the energy of those movements or not (Poole, 1978). Substrate intake mechanisms are therefore left uncoupled to cell energetics in the model due to lack of information. Abiotic transport of H2 and CO2 between the liquid phase and the gas head space is modelled assuming phase equilibrium (Henry’s Law) and constant atmospheric pressure in the reactor head. Chapter 2 70 2.2.4.6. Energetics evaluation The model considers that cell energetics revolve around ATP: the energy gained in catabolism or transport is stored in form of phosphate bonds in ATP and all energetic needs are satisfied by hydrolysing those phosphate bonds to generate ADP, Pi and useable energy. This energy can then be invested in growth, maintenance or fuelling active transport against the electrochemical gradient. The balance of ATP in catabolism is composed of five terms, which are described down below. Substrate-level phosphorylation (SLP) ATP can be directly generated by catabolic reactions. The contribution of SLP to the global ATP balance can hence be calculated straightforwardly from the reaction rates and the stoichiometry (Eq. 2.21) 𝑅𝐴𝑇𝑃,𝑆𝐿𝑃=∑𝑅𝑖·𝜈𝐴𝑇𝑃,𝑖 𝑖 (2.21) where RATP,SLP is the production rate of ATP by SLP. Proton translocation The cell membrane acts like a capacitor. In some exergonic metabolic steps, microorganisms can extrude a proton from the cytoplasm to the medium, storing thus energy on the membrane as its electric potential difference rises. When that proton re-enters the cell, it does it through a channel protein (ATPase) and lowers the membrane potential as a result. The energy released is stored in ATP molecules, following the chemiosmotic theory (White et al., 2012). Conversely, an ATP molecule can be broken to release energy and extrude protons that will fuel endergonic processes when re-entering the cell. The energy needed to extrude a single proton and the number of protons needed to yield an ATP molecule depend on the proton motive force (pmf, i.e. the electrochemical potential energy of a proton). The energy needed for translocating a proton is calculated using Eq. 2.22. ∆𝜇𝐻+=𝐹·∆𝜓+𝑅·𝑇·𝑙𝑛(𝑆𝑒𝑥,𝐻+ 𝑆𝑖𝑛,𝐻+) ( 𝑘𝐽 𝑚𝑜𝑙 𝐻+) (2.22) where ΔµH+ is the pmf, F is the Faraday constant, Δψ is the difference between the extracellular and intracellular electric potential, R is the gas constant, T the temperature, Sex,H+ is the extracellular proton concentration and Sin,H+ the intracellular proton concentration. Model development 71 The electric potential of the membrane is assumed to be 0 V on the outside and -0.2 V on the inside. In this model, the electric potential is considered to be constant, meaning that processes increasing membrane electrical potential are balanced with processes decreasing it. Intracellular pH is also considered constant at a value of 7, leaving extracellular pH as the only variation source for the pmf. In those reactions where the possibility of a proton extrusion is considered, the model evaluates in each simulation step whether the energy available (i.e. ΔG’) is high enough for extruding a proton (i.e. if it is higher that the pmf). In case the reaction is not exergonic enough, the same reaction without a proton extrusion is considered. The energy released by proton intrusion is converted into ATP, as explained before. The rate at which ATP is produced by proton translocations is calculated with Eq. 2.23. 𝑅𝐴𝑇𝑃,𝑝𝑚𝑓=𝑅𝐻+,𝑝𝑚𝑓·∆𝜇𝐻+ ∆𝐺𝐴𝑇𝑃 (2.23) where RATP,pmf is the ATP formation rate due to proton translocations and RH+,pmf is the proton rate across the membrane due to proton translocations. Transport Active transport of end products is modelled to occur concomitantly with proton transport inwards or outwards, depending on the electric charge of the product being transported. Logically, protons being transported across the membrane exchange energy with the cell and, accordingly, it should be accounted for. The transport of an end product can be against or in favour of its concentration gradient, consuming or releasing energy, respectively. This energy exchange is coupled to proton extrusion as well, resulting in a net production or consumption of ATP (Eq. 2.24). When the active transport of a compound is an exergonic process, energy is conserved by proton translocations to the cytoplasm, which is converted to ATP as previously explained. On the contrary, if the active transport of a compounds is an endergonic process it is fuelled by ATP (through a proton translocation coupled to ATP hydrolysis). 𝑅𝐴𝑇𝑃,𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡=(𝑅·𝑇·𝑙𝑛(𝑆𝑖𝑛,𝑖 𝑆𝑒𝑥,𝑖)−𝑧𝑖·𝐹·∆𝜓 ∆𝜇𝐻++𝑧𝑖)·∆𝜇𝐻+ ∆𝐺𝐴𝑇𝑃·𝑅𝐴𝑐𝑡,𝑖 (2.24) where RATP,Transport is the ATP formation rate associated to active transport Chapter 2 72 Maintenance Energy requirements for cell maintenance are defined to be directly correlated with the biomass concentration (4.5 kJ/molCx h). Therefore, the ATP consumption is calculated with Eq. 2.25. 𝑅𝐴𝑇𝑃,𝑚𝑎𝑖𝑛𝑡𝑒𝑛𝑎𝑛𝑐𝑒=− 4.5 ∆𝐺𝐴𝑇𝑃 (2.25) where RATP,maintenance is the ATP consumption rate due to maintenance Homeostasis Intracellular pH is maintained around a value of 7 with a Na+/H+ antiporter (Padan et al., 1981). It is modelled like a proportional controller with a setpoint of an intracellular pH of 7. Extracellular Na+ concentration is set equal to its intracellular concentration at each simulation step as we consider that microorganisms would choose the cation (e.g. Na+ or K+) that requires less energy to be transported (i.e. that has the minimum concentration gradient). To calculate its ATP cost the equation for active transport (Eq. 2.26) is used considering that the intra and extracellular Na+ concentrations are equal, resulting in Eq. 2.26. 𝑅𝐴𝑇𝑃,ℎ𝑜𝑚𝑒𝑜𝑠𝑡𝑎𝑠𝑖𝑠=𝑅·𝑇·𝑙𝑛(𝑆𝑒𝑥,𝐻+ 𝑆𝑖𝑛,𝐻+) ∆𝜇𝐻+·𝑅𝑁𝑎+/𝐻+ (2.26) where RATP,homeostasis is the ATP spent in homeostasis. 2.2.5. Solution strategy The different terms of the balances (section 2.2.1) are determined following the flowchart of Figure 2.7. The initial state values and the feeding properties (flow rate and compound concentrations) are the initial inputs of the model. Firstly, the thermodynamic limitation factor is calculated with the current state values (thermodynamic limitation step). Model development 73 Figure 2.7. Workflow diagram for model solution Then, in the reaction selection step, the different degradation pathways of the different substrates are first evaluated and then selected by an optimisation procedure (Eq. 2.29-2.30). The reaction evaluation step is divided into several Feeding characteristics Sk,in Initial conditions (states value) Thermodynamic limitation Reaction selection Kinetics Reaction rates determination Ri Transport Passive and active transport rates determination RT,i Energetics Active transport and pmf related ATP rate RATP,pmf, RATP,transport Kinetics Reaction rates determination Ri Transport Active transport rates determination RT,i Energetics Active transport and pmf related ATP rate RATP,pmf, RATP,transport Alternatives evaluation and optimization ATP, NADH and CO2 rates assessment Mass balances Steady state? Pseudo-time stepping Final states values NO YES Chapter 2 80 The kinetic parameters used are taken from the ADM1 model (Batstone et al., 2002a) and shown in Table S3. Table 2.4. Kinetic parameter values used for simulating methanogenesis. Parameter Value used qVal,max (gCOD Val/gCOD BM·h) 0.8 qBut,max (gCOD But/gCOD BM·h) 0.8 qPro,max (gCOD Pro/gCOD BM·h) 0.5 qAc,max (gCOD Ac/gCOD BM·h) 0.4 qH2,max (gCOD H2/gCOD BM·h) 1.0 KS,Val (gCOD Val/L) 0.2 KS,But (gCOD But/L) 0.2 KS,Pro (gCOD Pro/L) 0.1 KS,Ac (gCOD Ac/L) 0.15 KS,H2 (gCOD H2/L) 6·10-4 YC4-C5 (gCOD BM/gCOD C4-C5) 0.06 YPro (gCOD BM/gCOD Pro) 0.04 YAc (gCOD BM/gCOD Ac) 0.05 YH2 (gCOD BM/gCOD H2) 0.06 All three processes (acetogenesis and the two methanogenesis) performance is affected negatively by low pH values. The rate of the processes is multiplied by an inhibition function (Eq. 2.32) with two parameters representing the lower and upper limit of the inhibition. The inhibition function and the parameters for each process are taken from the ADM1 model (Table 2.5). Table 2.5. Parameters for the pH inhibition function of the three processes affected by pH inhibition pHLL pHUL Acetogenesis 4 5.5 Acetate consumption 6 7 Hydrogen consumption 5.8 7 𝐼= 1+2·100.5·(𝑝𝐻𝐿𝐿−𝑝𝐻𝑈𝐿) 1+10(𝑝𝐻−𝑝𝐻𝑈𝐿)+10(𝑝𝐻𝐿𝐿−𝑝𝐻) (2.32) Model development 81 Hydrogen concentration only inhibits acetogenesis and follows the form of a non-competitive inhibition (Eq. 2.33). The values for the three acetogenesisrelated processes are shown in Table 2.6 and were taken from the ADM1 model. 𝐼𝐻2=1 1+𝑆𝐻2 𝐾𝐼 (2.33) Table 2.6. Parameters of the hydrogen inhibition function for the different acetogenic processes. KI (gCOD-H2/L) Butyrate and valerate consumption (C4-C5) 1·10-5 Propionate consumption (Pro) 3.5·10-6 Additional processes (e.g. chain elongation) could be easily added to the model if needed. Also, the kinetic description of the current processes may be modified to reflecting ongoing phenomena such as product inhibition or ammonia toxicity. CHAPTER 3 Electron bifurcation mechanism and homoacetogenesis explain products yields in mixed-culture anaerobic fermentations 84 Summary Available mathematical models for the prediction of glucose MCF stoichiometry cannot reproduce the experimental results under some pH conditions. In particular, under acidic conditions the experimentally reporter equimolar production of acetate and butyrate is not well captured, which highlights our incomplete knowledge of the process. The discrepancies between the model and experimental results could be connected to the use of an incomplete or inaccurate metabolic network. To address this issue, in this chapter a new metabolic network of glucose fermentation by microbial mixed cultures incorporating electron bifurcation and homoacetogenesis is proposed. Using a methodology based on NADH balances to analyse published experimental data the new stoichiometry proposed is evaluated. This work proves for the first time that including electron bifurcation in the metabolic network allows for a better description of the experimental results. Homoacetogenesis has been used to explain the discrepancies between observed and theoretically predicted yields of gaseous H2 and CO2 and it appears as the best solution among other options studied. Overall, the conclusions of this chapter support the consideration of electron bifurcation as an important biochemical mechanism in microbial mixed cultures fermentations and underline the importance of considering homoacetogenesis when analysing anaerobic fermentations. EB and HA explain products yields in MCF 85 3.1. INTRODUCTION Experimentation on MCFs show that there is a tight relationship between the operational conditions and the products obtained (Fang and Liu, 2002; Horiuchi et al., 2002; Temudo et al., 2007; Zoetemeyer et al., 1982). How this relationship works is not clear enough and for this reason, directing product formation in MCFs is still a challenging task. Most part of the literature is purely an experimental description of the system and the mechanisms causing the shifts in product spectrum are rarely matter of discussion. Only a handful of studies, mostly mathematical models, are truly centred in understanding the connection between the environmental conditions and the changes observed in the product spectrum in glucose MCF (González-Cabaleiro et al., 2015; Groeger et al., 2017; Kleerebezem et al., 2008; Mosey, 1983; Rodriguez et al., 2006; Zhang et al., 2013). However, even though their predictions have improved over time and are able now to capture the main changes observed in the stoichiometry with some operational conditions, there are still some discrepancies when compared with the experiments. For example, in glucose MCF at low pH butyrate is predicted to be the main product in González-Cabaleiro et al. (2015) or Kleerebezem et al. (2008) but experimentally an equimolar mixture of butyrate and acetate is reported (Fang and Liu, 2002; Temudo et al., 2007; Zoetemeyer et al., 1982). These discrepancies might be due to the use of an incomplete metabolic network that does not describe accurately the stoichiometry of the process. In this chapter, modifications to the most commonly used metabolic network for glucose in MCFs are proposed, with the aim of improving the prediction of experimental results reported in literature. The novel biochemical mechanism of electron bifurcation (EB), fully described elsewhere, was included (Buckel and Thauer, 2013; Herrmann et al., 2008; Li et al., 2008; Peters et al., 2016). Homoacetogenesis (HA) (Dinamarca et al., 2011) is also proposed to be the most reasonable hypothesis for correctly explaining the gaseous species yields. 3.2. MATERIALS AND METHODS A metabolic network defines the global stoichiometry of the process as a compilation of the pathways that a single microorganism or a microbial population can catalyse. These pathways are described with all the intermediate metabolic steps, including the energetic coupling sites and the coenzymes involved (e.g. coenzyme A or NADH). Chapter 3 86 To check if including EB in a metabolic network for glucose fermentation helps improving modelling prediction capacity, a reference metabolic network is used in this study (Figure 3.1). The construction of this metabolic network is described in detail in chapter 2. This network is equal to the one presented in section 2.2.3.3, but it does not include EB in the butyrate-forming branch and is compared throughout this study with a metabolic network incorporating EB mechanism. Figure 3.1. Reference metabolic network of glucose fermentation with mixed cultures used in this study. ►CoA NADH ► Succinate ◄CoA ► ATP ▼ ►ATP NADH ► NADH ► Ethanol Glucose Acetate Propionate Propionyl-CoA Acetoacetyl-CoA Crotonyl-CoA Butyryl-CoA Acetaldehyde Pyruvate Fumarate Acetyl-CoA Oxaloacetate Lactate Acetyl-P Formate 2x Butyrate CO2 CO2 CO2 0.5x H2 ►2 NADH ►2 ATP ►ATP ►ATP CoA ► ►CoA ►CoA ►CoA NADH ► ▲CoA NADH ► NADH ▼ NADH ► Biomass ►CoA ATP ► NH4+► EB and HA explain products yields in MCF 87 3.2.1. Anabolism The metabolic network is closed by including the consumption and production of chemicals in the anabolic process. Following McCarty (2007), it was considered anabolism as a process that uses acetyl-CoA as substrate to produce biomass, which has a lumped chemical formula of C5H7O2N (Eq. 3.1). This biomass formula has the same oxidation state as acetyl-CoA (γ=4), therefore no further redox reactions are needed. However, producing acetyl-CoA from glucose implies that the anabolism carries an extra production of H2, CO2 and NADH (Figure 3.1) and this must be considered in the global stoichiometry. 2.5 Acetyl-SCoA + NH4+ → C5H7O2N +H+ + 0.5 H2O + 2.5 CoA-SH (3.1) 3.2.2. Methodology At steady state, electron carriers must maintain the balance between oxidised and reduced forms to keep the redox neutrality of the system. Therefore, when an electron carrier is reduced or oxidised, it needs to be regenerated. In fermentations all Fdred is assumed oxidised to produce H2 (section 2.2.3.1). NADH, on the contrary, cannot produce H2 directly as the redox potential of the NADH/NAD+ couple is higher than that of the proton reduction even in the most favourable fermentative conditions (Kleerebezem et al., 2008). Therefore, the NADH produced during glycolysis must be consumed in other places of the metabolic network. In consequence, a complete and accurate metabolic network results in a neutral NADH balance of the data analysed. To verify the proposed metabolic network, it was compared with the experimental data by calculating the NADH balance (Eq. 3.2). With the stoichiometry given by the network, the amount of corresponded NADH formed and consumed was determined according to the experimental products yields (moles of product formed per mole of substrate consumed in the system). If the NADH balance is neutral, this means that the metabolic network represents accurately the stoichiometry of the process. This can be used to compare different metabolic networks or to check whether a modification in a network improves its accuracy. ∑𝜗𝑁𝐴𝐷𝐻,𝑖·𝑦𝑖=0 𝑖=𝑁 (3.2) where νNADH, i is the NADH stoichiometric coefficient associated with the product i and yi is the yield of the product i (moles of product i per mole of glucose consumed). N is the total number of products. The result of the summation at steady state must be zero. Chapter 3 88 3.2.3. Experimental data used The experimental work of Temudo et al. (2007) (hereafter termed Temudo experiments) was selected, as it is the most comprehensive data set available on glucose fermentation using mixed cultures. It consists in a series of experiments using glucose as substrate (4 g/L) at pH values from 4 to 8.5. The hydraulic retention time (HRT) is 8 h for pH 5.5 to 8.5 and 20 h for pH from 4 to 5.5 (pH 5.5 was tested twice at different HRT values). The volume of the reactor is 2 L with 1 L of heading space and it was operated in continuous mode at 30ºC. To keep an anaerobic environment, N2 was flushed in the liquid phase at a 200 ml/min. They measured the yields of volatile fatty acids (acetate, propionate, lactate, and butyrate), ethanol, formate, CO2, H2, biomass and other minor products, closing the electron and carbon balances within a 10% of confidence (the values of the experimental yields are presented in section 3.5.1). At neutral and high pH, the proportion of inorganic carbon in the form of bicarbonate ion (HCO3-) should be considered and it was estimated assuming liquid-gas equilibrium and acid-base equilibrium. The experimental setup was designed to ensure full substrate consumption and that steady state is truly reached. To use these data in this study, formate production was considered equivalent to a sum of H2 and CO2 production. 3.3. RESULTS AND DISCUSSION 3.3.1. Incorporation of the electron bifurcation into the metabolic network EB is an enzymatic mechanism by which an exergonic electron transfer reaction is coupled with a sufficiently endergonic one. In this way, the energy surplus of the exergonic reaction is used to drive the endergonic one (Peters et al., 2016). EB was first hypothesised (Herrmann et al., 2008) in the crotonyl-CoA reduction with NADH in the butyrate synthesis pathway as a possible way for additional energy conservation. The reduction of crotonyl-CoA to butyryl-CoA (E’=-37 mV) when combined with the oxidation of NADH (E’=-280 mV), creates a highly exergonic step that could be used to drive endergonic reactions. The mechanism was then detected (Li et al., 2008) linking the highly exergonic and irreversible NADH-mediated reduction of crotonyl-CoA with the endergonic Fdox reduction by NADH. Overall, when one mole of crotonyl-CoA is reduced to butyryl-CoA, one mole of Fdox is reduced and two moles of NADH are oxidized (Figure 3.2). EB and HA explain products yields in MCF 89 Figure 3.2. Electron Bifurcation mechanism in NADH mediated Crotonyl-CoA reduction in butyrate production pathway The set of reactions is catalysed by a cytoplasmic enzyme complex (butyryl-Coa dehydrogenase/Etf complex) containing flavoproteins (Buckel and Thauer, 2013). The Fdred yielded is re-oxidised producing H2 (section 2.2.3.1), increasing thus the global H2 yields of the system to a theoretical maximum of 2.7 moles of H2 per mole of glucose (35% more than without EB). Therefore, EB could explain why higher ratios than 2 moles of H2 per mole of glucose are found in some works (Davila-Vazquez et al., 2008; Hallenbeck and Ghosh, 2009; Jungermann et al., 1973; Kapdan and Kargi, 2006; Petitdemange et al., 1976; Ren et al., 2016). So far, EB was only observed using purified enzymes from C. kluyveri (Buckel and Thauer, 2013; Li et al., 2008). However, there is no fundamental impediment to its occurrence in living C. kluyveri or C. pasteurianum, as proposed by Buckel and Thauer (2013), in MCFs. For these reasons, in the new metabolic network EB is included in the butyrate pathway (Figure 3.3). The inclusion of EB implies the consumption of one extra mole of NADH. This helps cells to decrease the reducing potential generated by the initial glycolysis. At the same time, it modifies the global stoichiometry of the process, as now the butyrate yielded from glucose requires the production of more oxidised products to close the electron balance of the fermentation. Without considering EB, the two moles of NADH formed in glycolysis were consumed in the butyrate pathway, closing in this way the NADH balance (Figure 3.1). 2 NADH 2 NAD+ 4 eFdox Fdred Crotonyl-CoA Butyryl-CoA 2 e2 eH2 FLAVOPROTEIN CONTAINING ENZYME COMPLEX 280 mV -37 mV ca. -500 mV Chapter 3 96 Figure 3.6. Absolute error for the NADH/NAD+ balance with the different stoichiometry considered for the experimental data presented by Temudo et al. (2007). NADH/NAD+ balances considering: ■ the metabolic network with electron bifurcation ■ the metabolic network with electron bifurcation and assuming homoacetogenesis. For some of the experiments (especially those at pH 4 and the first at pH 5.5), this consideration implies an increment of the error in the NADH balance. However, these differences might be very well attributed to experimental deviations. For example, carbon recovery in the first experiment at pH 5.5 is around 110%. At the same time, the ratio between butyrate and acetate yields in this experiment is higher than it is in the rest of low-pH experiments (yields are available in section 3.5.1), which might imply that the measurement of the butyrate yield could be deviated. As butyrate consumes NADH, its consumption could be overestimated and explain the negative deviation at this pH (Stoichiometric coefficients are available in section 3.5.2). At pH 4 the butyrate to acetate ratio also seems to be overestimated and a negative deviation in the NADH balance is again observed. Nevertheless, these deviations can be considered relatively small because for every glucose oxidation to pyruvate two mole of NADH are yielded. 3.3.3. EB in the propionate pathway Fumarate reduction to succinate in the propionate pathway is a similar step to crotonyl-CoA reduction to butyryl-CoA. They are mechanistically alike (hydrogenation of a double carbon bond) and in both cases their energetics -0.4 -0.2 0 0.2 0.4 0.6 0.8 4 4.75 5 5.5 5.5 6.25 7 7.75 8.5 NADH Balance (mol NADH/mol glucose) pH EB and HA explain products yields in MCF 97 suggest an extra energy conservation (E’Crotonyl-CoA= - 37 mV; E’Fumarate = - 5 mV). However, the EB mechanism was reported in the crotonyl-CoA reduction of the butyrate pathway but not in the fumarate reduction (Buckel and Thauer, 2013; Li et al., 2008). The energy surplus of this metabolic step in the propionate pathway is reported to be conserved extruding a proton from the cytoplasm and therefore creating an electrochemical proton gradient (Buckel, 2001; Herrmann et al., 2008). Even though EB was never detected in this step, it was assumed as feasible since there is no theoretical impediment for it. Using the same methodology as above, the NADH balances were calculated assuming EB in both butyrate and propionate pathways and HA (see section 3.5.7). However, the results show a worse fit when EB is included in the propionate pathway than when it is only included in the butyrate branch (the NADH balances increase the error by 23 %) and therefore the consideration of EB in the fumarate reduction was discarded. 3.4. CONCLUSIONS This study considers for first time electron bifurcation in mixed-culture fermentation processes yielding butyrate, which provides a better fit between the predicted stoichiometry and the experimental yields measured. Therefore, it is recommended to consider this mechanism in the reduction of crotonyl-CoA to butyryl-CoA in butyrate formation pathways. Including electron bifurcation in the metabolic network allows for the prediction of the production of butyrate accompanied by acetate. This was experimentally observed but not theoretically predicted before. The addition of electron bifurcation revealed an unbalance between the stoichiometry and the observation in the production of the gaseous components, H2 and CO2. In this study, it is proposed that the unbalance of H2 and CO2 could be explained by the occurrence of homoacetogenesis in the mixed-culture fermentation of glucose. Chapter 3 98 3.5. ANNEXES 3.5.1. Experimental yields from Temudo et al. (2007) In the Table A3.1 the yields reported by Temudo et al. (2007) are presented. The Hydraulic Retention Time (HRT) was of 20 hours at pH values more acidic than 5.5 and of 8 hours at more basic pH values than 5.5. This avoided biomass washing at low pH. At pH 5.5 the experiment was repeated at both HRT. Table A3.1. Experimental yields as reported in Temudo et al. (2007). pH HRT: 20h HRT: 8h 4 4.75 5 5.5 5.5 6.25 7 7.75 8.5 Product Butyrate 0.580 0.416 0.546 0.655 0.513 0.036 0.001 0.185 0.034 Acetate 0.418 0.455 0.366 0.370 0.403 0.619 0.700 0.578 0.680 Propionate 0.000 0.054 0.000 0.044 0.040 0.007 0.000 0.041 0.020 Glycerol 0.185 0.116 0.059 0.129 0.094 0.134 0.063 0.058 0.174 Lactate 0.068 0.165 0.056 0.022 0.005 0.004 0.004 0.098 0.062 Succinate 0.000 0.000 0.000 0.003 0.004 0.087 0.106 0.097 0.065 Ethanol 0.000 0.099 0.133 0.048 0.028 0.636 0.689 0.558 0.587 Biomass 0.385 0.533 0.878 0.653 1.102 0.709 0.625 0.902 0.577 CO2 1.249 1.398 1.657 1.481 1.210 1.315 1.414 1.398 1.420 H2 1.398 1.508 1.652 1.713 1.276 1.283 1.451 1.315 1.439 3.5.2. NADH, H2 and CO2 stoichiometric coefficients Table A3.2 contains the NADH and H2 stoichiometric coefficients for each of the pathways considered. Table A3.2. NADH stoichiometric coefficients for each of the final products Stoichiometry NADH H2 Without EB With EB Without EB With EB Product Butyrate 0 -1 2 3 Acetate 1 1 1 1 Propionate -1 -2 0 1 Glycerol -1 -1 0 0 Lactate 0 0 0 0 Succinate -1 -2 0 1 Ethanol -1 -1 1 1 Biomass (C-mole) 0.5 0.5 0.5 0.5 EB and HA explain products yields in MCF 99 3.5.3. Calculation example of a NADH balance Equation A3.1 shows an example of a NADH balance calculation. The product yields used are the ones of the experiment at pH 4 (Table A3.1). No EB is considered. ∑𝜈𝑖∙𝑦𝑖=0.580∙0+0.418∙1+0∙(−1)+0.185∙(−1)+0.068∙0 +0∙(−1)+ 0·(−1)+0.385∙0.5= =0.4255 𝑚𝑜𝑙 𝑁𝐴𝐷𝐻 𝑚𝑜𝑙 𝑔𝑙𝑢𝑐𝑜𝑠𝑒 (A3.1) where νi is the NADH stoichiometric factor for each compound (e.g. butyrate, acetate, etc.) and yi is the yield of that compound per moles of glucose consumed. 3.5.4. Results from other data sets The analysis with the methodology explained in the main text was repeated using data obtained from other experimental acidogenic fermentations besides the data presented in the main text from Temudo et al. (2007). Nevertheless, the other data sets are not as comprehensive and, for example, the mass balances are not closed in most cases. In general, these experiments do not report the yields of minor products (e.g. succinate) therefore the carbon balance is not fully closed, which may have a substantial impact on the NADH balances. In all the cases, it is observed that the NADH balances fit better the experimental data when electron bifurcation (EB) is considered. Figure A3.1 to Figure A3.6 show the balance results considering and not considering EB in the stoichiometry of the process. Specific comments on the possible causes for the deviations in the balances of some data sets are discussed below. In some experiments of the data set of Figure A3.1 (Fang and Liu, 2002), glucose consumption was not complete (representing around 20% of the carbon content of the effluent), especially at low pH. At high pH, on the contrary, methane was detected in the gas effluent. This could explain the high NADH balances because part of the reductive equivalents was transferred to methane, which was not considered in the analysis. Chapter 3 100 Figure A3.1. NADH balances without and with EB stoichiometry considered for the experimental data presented in Fang and Liu (2002). ■ without EB ■ with EB Figure A3.2. NADH balances without and with EB stoichiometry considered for experimental data presented in Horiuchi et al. (2002). ■ without EB ■ with EB 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 4 4.5 5 5.5 6 6.5 7 NADH Balance (mol NADH/mol glucose) pH -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 555678 NADH Balance (mol NADH/mol glucose) pH EB and HA explain products yields in MCF 101 Figure A3.3. NADH balances without and with EB stoichiometry considered for experimental data presented in Zoetemeyer et al. (1982). ■ without EB ■ with EB In some cases, carbon balances are higher than 100% –e.g. at pH 7.9, (Zoetemeyer et al., 1982) –, which could indicate that some of the concentration values could be overestimated. Figure A3.4. NADH balances without and with EB stoichiometry considered for experimental data presented in de Kok et al. (2013) ■ without EB ■ with EB 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 4.5 5.0 5.7 6.0 6.4 6.9 7.9 NADH Balance (mol NADH/mol glucose) pH 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 A B C D NADH Balance (mol NADH/mol glucose) Experiment Chapter 3 102 Figure A3.5. NADH balances without and with EB stoichiometry considered for experimental data presented in ■ without EB ■ with EB. In the data set of Figure A3.5 (Mohd-Zaki et al., 2016) and mostly in those cases where NADH balances are high, carbon recovery is higher than 100%, which would indicate uncertainty in the experimental measurements. Figure A3.6. NADH balances without and with EB stoichiometry considered for experimental data presented in Temudo et al. (2008) and Temudo et al. (2009) ■ without EB ■ with EB. -0.05 0.00 0.05 0.10 0.15 0.20 0.25 4.5 5 5.5 6 6.5 7 7.5 8 NADH Balance (mol NADH/mol glucose) pH 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Xylose 4 g/L Xylose 10 g/LGlucose 4 g/LGlycerol 4 g/LGlucose 4 g/L NADH Balance (mol NADH/mol substrate) Experiment EB and HA explain products yields in MCF 103 3.5.5. Energetics of homoacetogenesis Figure A3.7 shows the Gibbs energy for HA at different hydrogen partial pressures. The pH is considered constant at 7 and concentrations of acetate and bicarbonate assumed 10 and 2 mM respectively. Figure A3.7. Gibbs energy of the HA function of the H2 partial pressure in the headspace. Assumed ideal liquid-gas equilibrium. 3.5.6. Example of the calculation to include homoacetogenesis For this example, a simplified fermentation experiment with the following yields is assumed: butyrate (0.6 mol/mol glucose), acetate (0.45 mol/mol glucose), biomass (0.4 C-mol/mol glucose), H2 (1.7 mol/mol glucose) and CO2 (1.5 mol/mol glucose). Using the stoichiometric factors provided in Table A3.2, the theoretical H2 and CO2 yields are calculated (Eq. A3.2-A3.3). ∑ 𝜈𝑖,𝐻2∙𝑦𝑖=0.6·3+0.45·1+0.4·0.5=2.45 𝑚𝑜𝑙 𝐻2 𝑚𝑜𝑙 𝑔𝑙𝑢𝑐𝑜𝑠𝑒 𝑖=𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠 (A3.2) ∑ 𝜈𝑖,𝐶𝑂2∙𝑦𝑖=0.6·2+0.45·1+0.4·0.5=1.85 𝑚𝑜𝑙 𝐶𝑂2 𝑚𝑜𝑙 𝑔𝑙𝑢𝑐𝑜𝑠𝑒 𝑖=𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠 (A3.3) -100 -80 -60 -40 -20 0 1.E+02 1.E+03 1.E+04 1.E+05 ΔG (kJ/mol acetate) PH2 (Pa) 4 H2+ 2 HCO3-→ Acetate + 4 H2O Chapter 3 104 The difference between the theoretical and the measured yields is 0.75 and 0.35 mole for H2 and CO2, respectively. It is assumed that HA consumes all the CO2 yield difference (0.35 mol). Therefore, the amount of extra H2 consumed (0.7 mol) and extra acetate produced (0.18 mol) are calculated straightforwardly from the stoichiometry of the HA (Eq. 3.2). The new theoretical yields are 1.75 mole for H2 and 1.5 mole for CO2 (Table A3.3) Table A3.3. First iteration of the calculation including HA Theoretical yield without HA Measured yield Yield difference Consumed yield in HA H2 2.45 1.70 0.75 0.70 CO2 1.85 1.50 0.35 0.35 However, this result is not completely correct since it was assumed for the calculation of the H2 and CO2 theoretical yields that acetate production was contributing to them (0.45 mol/mol glucose). Now, part of this acetate is assumed to come from HA and therefore is not accompanied with H2 and CO2 yield. In consequence, the acetate produced by HA (0.18 mol/mol glucose) should be subtracted from the overall yield measured (0.45 mol/mol glucose) recalculating the theoretical yields of H2 and CO2 (Eq. A3.4-A3.5). ∑ 𝜈𝑖,𝐻2∙𝑦𝑖=0.6·3+(0.45−0.18)·1+0.4·0.5 𝑖=𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠 =2.28 𝑚𝑜𝑙 𝐻2 𝑚𝑜𝑙 𝑔𝑙𝑢𝑐𝑜𝑠𝑒 (A3.4) ∑ 𝜈𝑖,𝐶𝑂2∙𝑦𝑖=0.6·2+(0.45−0.18)·1+0.4·0.5 𝑖=𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠 =1.68 𝑚𝑜𝑙 𝐶𝑂2 𝑚𝑜𝑙 𝑔𝑙𝑢𝑐𝑜𝑠𝑒 (A3.5) With the new theoretical yields, the calculation procedure is repeated (Table A3.4). Table A3.4. Second iteration of the calculation including HA New theoretical yield Measured yield Yield difference Consumed yield in HA H2 2.28 1.70 0.58 0.35 CO2 1.68 1.50 0.18 0.18 The iterations continue until the yield difference between the theoretical values and the measured ones is minimum. EB and HA explain products yields in MCF 105 3.5.7. NADH balances considering EB in butyrate and propionate pathways Figure A3.8. Absolute error for the NADH balance with the different metabolic networks considered for the experimental data presented by (Temudo et al., 2007). NADH balances are from: ■ the metabolic network without electron bifurcation and ■ the metabolic network with electron bifurcation in Butyrate and Propionate pathways ■ the metabolic network with electron bifurcation and assuming homoacetogenesis -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 4 4.75 5 5.5 5.5 6.25 7 7.75 8.5 NADH Balances (mol NADH/mol glucose) pH Chapter 4 112 limiting (i.e. at low specific growth rates), pathways with higher enzymatic concentration requirements but also higher ATP yields may be more beneficial and generate a higher growth rate. Thus, there is a trade-off between being efficient in ATP yield and being efficient in terms of enzyme requirements and resource allocation models can predict which strategy prevails under a given conditions and therefore the product spectrum. For example, in the case of the Crabtree effect resource allocation models predict that, at certain environmental conditions, strategies providing a lower ATP yield but low enzymatic requirements (such as ethanol production in presence of oxygen) can sustain a higher specific growth rate than high ATP yield but high enzymatic requirements options (complete respiration of the substrate). 4.2.2. Model assumptions ▪ Biomass is divided in two fractions: i) raw biomass (RBM) representing the carbohydrates, lipids, DNA and RNA and ii) protein. The molecular composition of RMB is considered to be constant regardless of the specific growth rate. This has the advantage that the concentration of the different protein fractions (Fig. 1) can be expressed over RBM in the form of a mass ratio (gPROT/gRBM) as a growth-independent measurement. The elemental composition of the two fractions and its ATP requirements are assumed as reported in (Schumacher, 2018). ▪ Protein is modelled to constitute up to 50% of the total cell weight (i.e. for each gram of protein in the cell, there is one gram of raw biomass resulting in a concentration of 1 gprotein/gRBM). Values in the same order of magnitude were measured and used in previous resource allocation models (Basan et al., 2015; Mori et al., 2016; Nilsson and Nielsen, 2016; Schumacher, 2018). ▪ Proteins can be located in the cytoplasm (ribosomes and catabolic enzymes) or in the membrane (transport proteins). The maximum content of membrane proteins is 20% of the total protein pool as experimentally measured in different bacterial species (Liu and Rost, 2001) ▪ Half of the proteins are not used in growth-related processes and their fraction is constant regardless independent of the specific growth rate. In consequence, of the maximum protein concentration of 1 gPROT/gRBM, 0.5 are not used for growth and the remaining 0.5 may be used for growth purposes, of which a maximum of 0.1 gPROT/gRBM can be located in the membrane. This value is in line with the values used in similar approaches (Basan et al., 2015; Mori et al., 2016). Resource allocation explains lactic acid production 113 ▪ The flux catalysed by enzymes follows a first-order kinetics with respect to the enzyme concentration and activity (Eq. 4.1) (Goelzer and Fromion, 2011; Hui et al., 2015; Scott et al., 2014). This assumption was used as well in the previous modelling approaches using the resource allocation theory (Basan et al., 2015; Mori et al., 2016; Nilsson and Nielsen, 2016; Zeng and Yang, 2019). 𝑖=𝑝𝑖·𝜎𝑖·𝑎𝑖 (4.1) where vi is the flux, pi is the protein (i.e. enzyme) concentration, σi is the saturation coefficient and ai is the activity of the enzyme catalysing flux i. ▪ All enzymes are considered to be half-saturated (σi = 0.5), except for the glucose transporters whose kinetics are governed by the extracellular glucose concentration according to the Michaelis-Menten equation. This assumption is based on the previously hypothesised trade-off between low metabolite concentrations and high enzyme saturation (Tepper et al., 2013) and on the observation that most glycolytic enzymes are half saturated in E. coli (Bennett et al., 2009). Previous resource allocation models also followed this hypothesis (Nilsson & Nielsen, 2016; Schumacher, 2018) ▪ LAB are considered auxotrophic in amino acids (i.e. they cannot synthesise them and need to be transported). The main implication is that LAB do not need to allocate enzymes (ribosomes) to amino acid production from inorganic nitrogen and can dedicate the freed protein fraction to other growth tasks. Experimental measurements revealed that in E. coli, the share of amino acid biosynthesis enzymes drops from 20% to 6% when switching from rich to mineral growth medium while the share of other synthetic enzymes increases (Li et al., 2014). This may be interpreted as the enzymatic activity of ribosomes (i.e. the flux of new biomass formed per unit of mass of ribosomal protein) is higher as amino acids are not synthesised. In the model, this observation is represented by a 15% higher enzymatic activity of the ribosomes in LAB with respect to BPB. Information on this subject for LAB or other prokaryotic microorganisms was not available in literature. ▪ Energy requirements for transport of substrates (glucose, inorganic nitrogen or amino acids) and end-products is not considered as it depends highly on the intracellular concentrations of the substrates and products to be transported, which are not simulated in the model. This assumption does not affect the competition between auxotrophic LAB and BPB since the Chapter 4 114 energy required for transporting amino acids or inorganic nitrogen is similar (Rombouts et al., 2020; Stouthamer, 1973). 4.2.3. Model description The model follows the formulation of a flux balance analysis (FBA), in which the different cellular fluxes are optimised by linear programming to maximise the specific growth rate (Eq. 4.2). The optimisation has common FBA constraints of steady state condition (as in a CSTR, Eq. 4.3) and positive fluxes (Eq. 4.4). Additionally, two inequality constraints related with resource allocation are included. As previously stated, the total concentration of growth-related protein (i.e. ribosomes and catabolic and transport enzymes) has an upper limit of 0.5 gPROT/gRBM (Eq. 4.5), of which 20% can be located in the membrane (i.e. transport enzymes), which is expressed in the model by an upper limit on membrane protein concentration of 0.1 gPROT/gRBM (Eq. 4.6). The total and membrane protein concentration are determined by addition of the concentration of all growth-related or membrane enzymes, respectively, for a given set of metabolic fluxes using Eq. 4.1, which relates the needed enzyme mass to catalyse a flux. max 𝑣𝜇 h-1 (4.2) 𝑁· =0 mol/molBM·h (4.3) 𝑖>0 mol/molBM·h (4.4) ∑ 𝑖 𝜎𝑖·𝑎𝑖≤𝑝𝑚𝑎𝑥,𝑡𝑜𝑡𝑎𝑙=0.5 gPROT/gRBM (4.5) ∑ 𝑗 𝜎𝑗·𝑎𝑗≤𝑝𝑚𝑎𝑥,𝑚𝑒𝑚𝑏𝑟𝑎𝑛𝑒=0.1 gPROT/gRBM (4.6) where N is the stoichiometric matrix, v are the fluxes, σ is the saturation degree, which has a value of 0.5 for all enzymes except the glucose transporter, a is the enzymatic activity and pmax is the maximum protein concentration. There are i fluxes of which j correspond to membrane-related (i.e. transport) processes. Since steady state is assumed and glucose is the only substrate, all fluxes depend on the glucose import flux, which in turn depends on the activity of its transporter (Eq. 4.1). As stated in the previous section, the enzymatic activity of this transporter is governed by the Michaelis-Menten equation (Eq. 4.7), depending thus the glucose import flux on the external glucose concentration (Eq. 4.8). Resource allocation explains lactic acid production 115 𝜎𝐺𝑇= [𝑙𝑢𝑐𝑜𝑠𝑒] 𝐾𝑚+[𝐺𝑙𝑢𝑐𝑜𝑠𝑒] (4.7) 𝐺𝑙𝑢𝑐𝑜𝑠𝑒 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡= 𝑎𝐺𝑇·𝑝𝐺𝑇·[𝐺𝑙𝑢𝑐𝑜𝑠𝑒] 𝐾𝑚+[𝐺𝑙𝑢𝑐𝑜𝑠𝑒] (4.8) where σGT is the glucose transporter saturation degree, Km is the half saturation constant, aGT is the activity of the glucose transporter and pGT is its concentration. The form of Eq. 4.8 is equivalent to a Michaelis-Menten equation in which the maximum rate is represent by aGT· pGT. Three model microorganisms with different catabolic and anabolic characteristics are simulated to discern whether catabolism or anabolism is responsible for the different behaviours of LAB and BPB: i) An amino acid auxotrophic LAB, i.e. with a higher ribosomal activity. ii) A hypothetical prototrophic (i.e. non auxotrophic) LAB. iii) A prototrophic BPB. 4.2.4. Metabolic network The experiments of Rombouts et al. (2020) show that there is duality in the ecological domination of the microbial consortia in MCF operated in a glucosefed sequential batch reactor. LAB of the genus Lactobacillus and Lactococcus dominate the microbial community in rich medium fermentations and bacteria belonging to the genus Clostridium dominate the community when mineral medium is used. For the sake of simplicity, the metabolic network used in this model only considers one possible catabolism for each of the microorganisms dominating the culture: acetate and butyrate in the case of BPB and lactate in the case of LAB (Table 4.1). The glycolytic part of the network (conversion of glucose to pyruvate) is considered equal for both microorganisms and follow the EMP pathway. To produce acetate and butyrate at an equimolar ratio, as observed experimentally with BPB (Temudo et al., 2007), the electron bifurcation mechanism must be included in the butyrate pathway to allow for a closed electron balance in the conversion reaction from glucose (Chapter 3). Biomass is considered to be built from glucose including 10% molar conversion of carbon substrate to carbon dioxide in anabolism and without consumption of electron equivalents. The influx of amino acids or inorganic nitrogen is omitted from the metabolic Chapter 4 116 network and the model, as glucose is considered to be the only limiting substrate in the system. Table 4.1. Summarised catabolic reactions considered in the metabolic network. Microorganism Conversion reaction ATP yield LAB Glucose → 2 lactate- + 2 H2O 2 BPB Glucose → 0.67 acetate- + 0.67 butyrate- + 0.67 H2O + 2.67 H2 + 2 HCO33.33 4.2.5. Enzymatic activities The activities of the enzymes involved in the steps described in the metabolic network were collected in the BRENDA database (Schomburg et al., 2002) and in literature (Table 4.2). Glycolysis activity values are referred to enzymes found in E. coli as values for the genus of the microorganisms considered in the metabolic network could not be found. These glycolysis enzymatic activity values are used in the catabolism of the two microorganisms considered in the model and, in consequence, the interference in the competition between them is considered to be negligible. For BPB, activity values related with different species belonging to the genus Clostridium were chosen. The activity value of the lactate dehydrogenase (the only enzyme needed for converting pyruvate into lactate) was selected from the species L. plantarum, a LAB found in dairy, meat and vegetables fermentations involved in food spoilage (de Vries et al., 2006). Due to lack of available information in literature and databases, some approximations had to be made. Information on the enzymatic activity of any monocarboxylate transporter in bacteria was not available in literature. For the acetate, butyrate and lactate exporter it was decided to use for the three of them the activity value used in Schumacher (2018) for the lactate exporter in S. cerevisiae. Glucose importer activity could not be found either and its value was set to reflect a maximum glucose uptake rate of 40 mmol/gBM·h (Batstone et al., 2002a), assuming that glucose is converted to two lactate molecules that have to be transported out of the cell by a system that also occupies space in the membrane. Ribosomes activity was also set to reflect a LAB maximum specific growth rate of 0.35 h-1 similar to that observed experimental in lactate producers in Rombouts et al. (Rombouts et al., 2020). Resource allocation explains lactic acid production 117 Table 4.2. Enzymes considered in the metabolic network and their enzymatic activity. Enzyme Name EC Number Specific activity Microorganism Ref Glycolysis HXK 2.7.1.2 158 E. coli PGI 5.3.1.9 212.6 E. coli PFK 2.7.1.11 190 E. coli FBA 4.1.2.13 477 E. coli 1 TPI 5.3.1.1 55 E. coli 2 GLD 1.2.1.12 40 E. coli 3 PGK 2.7.2.3 480 E. coli GPM 5.4.2.11 124 E. coli ENO 4.2.1.11 260 E. coli CDC 2.7.1.40 55 E. coli 4 Lactic acid LDH 1.1.1.27 2350 L.plantarum Butyric acid PFOR 1.2.7.1 25 C. acetobutylicum 5 Acetyl-CoA C-acetyltransferase 2.3.1.9 216 C. acetobutylicum BHBD 1.1.1.157 445 C. kluyveri 3-hydrobutyrate dehydratase 4.2.1.17 C. acetobutylicum Etf 1.3.1.109 12 C. difficile Phosphate butyryltransferase 2.3.1.19 1380 C. acetobutylicum Butyrate kinase 2.7.2.7 402 C. acetobutylicum Acetic acid PAT 2.3.1.8 7140 C. kluyveri ACK 2.7.2.1 1087 C. acetobutylicum Transporters Glucose transporter 16.7 Parametrised Acid transporter 106 Parametrised Anabolism Ribosomes 1.7 Parametrised 1: (Baldwin et al., 1978), 2: (Garza-ramos et al., 1996), 3: (D´Alessio and Josse, 1971), 4: (Waygood and Sanwal, 1974), 5: (Meinecke et al., 1989) Chapter 4 118 4.2.6. Uncertainty and sensitivity analysis Given that the model relies on a number of assumptions on parameter values, a global uncertainty and sensitivity analysis of the parameter space was performed. The parameters selected for this analysis (Table 4.3) are assumed to have a uniform probability distribution bounded 25% around the default value, an adequate value for parameters with an intermediate uncertainty (Vangsgaard et al., 2012). Table 4.3. Parameters whose uncertainly is assessed in the Monte Carlo procedure and their default values. Parameter Default value Total protein concentration 1 gPROTEIN/gRBM Growth-related protein 50% Protein located in the membrane 20% Decrease in ribosomal activity in prototrophic BPB 15% Glucose transporter 1.01 mgPROT/(mmolGlucose·h) Glycolytic enzymes 2.52 mgPROT/(mmolGlucose·h) Butyrate production enzymes 2.91 mgPROT/(mmolButyrate·h) Acetate production enzymes 0.68 mgPROT/(mmolAcetate·h) Lactate production enzymes 0.007 mgPROT/(mmolLactate·h) Butyrate transporter 0.16 mgPROT/(mmolButyrate·h) Acetate transporter 0.16 mgPROT/(mmolAcetate·h) Lactate transporter 0.16 mgPROT/(mmolLactate·h) Ribosomes 9.30 mgPROT/(mmolBiomass·h) The first four parameters relate to assumptions about the cellular structure and the differences in proteomics between auxotrophic and prototrophic microorganisms. The remaining correspond to the enzyme requirements of different sections of the metabolic network and growth. All the parameters are considered to be uncorrelated to avoid underestimating the uncertainty of the model solution. The parameter space was sampled using the Latin Hypercube Sampling methods to ensure a maximal coverage of the parameter space (Helton and Davis, 2003). Following a Monte Carlo procedure, a total of 1000 samples were taken and used to solve the model at different residual glucose concentration values. The model outputs (specific growth rate values) are the basis for the subsequent sensitivity analysis. Resource allocation explains lactic acid production 119 A global sensitivity analysis was carried out to determine what parameters (and assumptions) exert a higher influence in the model output and check the need of refining the model hypotheses. The method of standardised regression coefficients (SRC) was chosen, whereby a first order linear multivariable model relating the model outputs taken from the Monte Carlo procedure (yk) to the parameter values (θi) is fitted with the least squares method (Saltelli et al., 2008): 𝑦 =𝑏 ,0+∑𝑏 ,𝑖·𝜃𝑖 𝑖 (4.9) where yk are the model outputs, bk,0 and bk,i are the linear regression coefficients and θi the parameters. The subscript k denotes each of the model outputs and the subscript i refers to the parameters analysed. In this case, only the ratio of the specific growth rates of LAB and BPB is used as model output (denoted as Y) and the parameters included in Table 4.3 were considered. If Eq. 4.9 is made dimensionless by mean-centred sigma-scaling, to make the coefficients directly comparable regardless of their absolute value, the standardised linear regression coefficients, βk,I, are obtained: 𝑦 −𝜇𝑦𝑘 𝜎𝑦𝑘=∑(𝛽 ,𝑖·𝜃𝑖−𝜇𝜃𝑖 𝜎𝜃𝑖) 𝑖 (4.10) where µyk and µθi are the mean values and σyk and σθi are the standard deviation of the model outputs and input parameters, respectively. If the multivariable model of Eq. 4.10 is linearly additive, then ∑𝛽𝑖2=1 𝑖 for each of the model outputs should be fulfilled. The parameter βi2 represents the contribution to the variance of parameter i and can be used as a measure of its importance on the output of the model. To assume the model linear, the squared coefficient of correlation (R2) between the Monte Carlo simulation output (Y) and the values produced with the regression model with the estimated SRC (Eq. 4.10) regressed linear output should be above 0.7 (Vangsgaard et al., 2012). 4.3. RESULTS AND DISCUSSION 4.3.1. Evaluation of the enzymatic activities for the different strategies The enzyme requirements (i.e. the mass of enzyme needed to catalyse one unit of flux) for the different sections of the metabolic network and of the different catabolic strategies were determined (Table 4.4). Glycolysis requires a high enzyme mass, similar to butyrate production and, to a lower extent, acetate. In comparison, yielding lactate from pyruvate needs much less enzyme mass Chapter 4 120 because lactate is only one metabolic step away from pyruvate and the enzyme catalysing that conversion, lactate dehydrogenase, has a remarkable high activity. As a consequence, lactate fermentations only need 3.85 mg of enzyme to catalyse one flux unit while butyrate fermentations need 6.13 mg, which is 40% more, reinforcing the hypothesis that shorter pathways are faster due to lower enzymatic requirements (Kappler et al., 1997; Pfeiffer and Bonhoeffer, 2004). However, lactate formation is energetically less favourable, as the ATP yield is 40% lower (Table 4.1). As a result, the mass of enzymes needed to produce a unit of ATP flux in catabolism, which is proportional to the specific growth rate, is very similar for both strategies (Table 4.4). Table 4.4. Enzymatic requirements for metabolic fluxes of different pathways. Pathway Enzyme requirement Glucose → 2 Pyruvate 2.52 mg Prot/(mmol glucose/h) Pyruvate → Lactate 0.007 mg Prot/(mmol pyruvate/h) Pyruvate → ½ Butyrate 1.46 mg Prot/(mmol pyruvate/h) Pyruvate → Acetate 0.68 mg Prot/(mmol pyruvate/h) Complete catabolism Glucose → 2 Lactate 3.85 mg Prot/(mmol glucose/h) Glucose → 2/3 Butyrate + 2/3 Acetate 6.13 mg Prot/(mmol glucose/h) Catabolism in terms of ATP flux Lactate (2 ATP) 1.93 mg Prot/(mmol ATP/h) Butyrate fermentation (3.33 ATP) 1.86 mg Prot/(mmol ATP/h) Lactate fermentations cannot be explained through the Crabtree effect, where the differences between fermentation and respiration in terms of enzymatic requirements are noticeable higher than in this case. In this sense, the Crabtree effect can be already explained by only the ATP flux enzymatic requirements. Moreover, if only producing lactate was the key to attain a very high specific growth rate, fermentations dominated by Clostridium bacteria potentially could be producing lactate as well, since they have the genes to synthetize lactate dehydrogenase and lactate transporters (Biddle et al., 2014; Geer et al., 2010) and were reported to produce lactate under specific environmental conditions (Payot et al., 1999). 4.3.2. LAB auxotrophic anabolism explains its fast growth The growth of the three microorganisms presented in section 4.2.3 is modelled at different glucose concentrations and the predicted specific growth rates are Resource allocation explains lactic acid production 121 shown in Figure 4.2. Glucose concentration values are divided by the half saturation constant (Km, Eq. 4.7) to isolate the influence of the value of this parameter on the analysis. Two areas can be clearly identified in the graph. At high glucose concentrations, auxotrophic LAB are the fastest growers and, on the contrary, at low glucose concentrations, BPB can achieve a higher specific growth rate and will likely dominate the microbial community under these conditions. LAB grow slower and at a very similar rate regardless of their anabolism (auxotrophic or prototrophic). Figure 4.2. Predicted specific growth rate of three model microorganisms at different glucose concentrations. The horizontal axis represents the dimensionless glucose concentration determined by division by Km ■ Auxotrophic lactic acid bacteria (LAB), ■ prototrophic lactic acid bacteria (LAB) and ■ butyrate-producing bacteria (BPB). The dashed line represents specific growth rate value of 0.28 h-1. LAB display a maximum specific growth rate of 0.35 h-1, which is 9% higher than that of BPB. This result suggests that auxotrophic LAB outcompete BPB in environments that select for microorganisms with the highest maximum specific growth rate, as for example in sequencing batch reactors. The estimated kinetic parameters of the two microbial communities in the experiments of Rombouts et al. (2020) also show that LAB in rich medium also developed a higher maximum growth rate (0.23 h-1) than BPB in mineral medium. (0.19 h-1), which represent a 20% increase. Chapter 4 128 Figure 4.7. Uncertainty analysis results for the ratio between specific growth rate of LAB and BPB at different non-dimensional glucose concentrations. The line represents the mean value of the Monte Carlo simulations and the shaded area the region of ratio values within a 95% confidence interval (percentiles 2.5%-97.5%). The sensitivity of the model was analysed at two glucose concentrations (0.5 and 10 times Km) to identify in each case the mechanisms that affect the most the specific growth rate of each microbial group. Table 4.5 features the squared standardised linear coefficients, which represent the contribution of each parameter to the variance of the model outcome at the two points analysed. At low glucose concentrations, where BPB dominate, membrane-related parameters have the highest influence on the model outcome. The share of the total protein that is located in the membrane represents almost 50% of the variance observed in the model outcome, followed by the enzymatic requirements for glucose transport with 25%. The high sensitivity of these parameters on the model output is in close agreement with section 4.3.3 as transport was already identified to limit growth under these conditions. Resource allocation explains lactic acid production 129 Table 4.5. Squared standardised linear coefficients for the parameters varied in the uncertainty analysis. Parameter βi2 [Glucose]=0.5 [Glucose]=10 Glucose transporter 0.25 0.03 Glycolytic enzymes 0.01 0.36 Butyrate production enzymes 0.03 0.37 Acetate production enzymes 0.00 0.02 Lactate production enzymes 0.00 0.00 Butyrate transporter 0.02 0.04 Acetate transporter 0.00 0.00 Lactate transporter 0.00 0.00 Ribosomes 0.10 0.02 Total protein concentration 0.06 0.02 Growth-related protein 0.01 0.00 Protein located in the membrane 0.47 0.01 Decrease in ribosomal activity in prototrophic BPB 0.00 0.05 Model linearity (R2) 0.949 0.935 At high glucose concentrations, where LAB are predicted to dominate the microbial community, the most sensitive parameters are related with cytoplasmatic processes. The enzymatic requirements of glycolysis and butyrate yielding explain 36% and 37% of the variance, respectively. This is in line with the conclusions of section 4.3.3, where it was shown that cytoplasmatic protein capacity was the factor limiting growth at high substrate concentrations. 4.3.6. Resource allocation and process design Microorganisms strive for energy in their pursuit to grow and dominate their ecosystem. In this sense, strong selective pressures are believed to have been exerted to the ATP-producing pathways (Pfeiffer et al., 2001). Apparently, based on the results of this model, these selective pressures might have led LAB to lose the ability to synthesise amino acids from inorganic compounds as it gives them a competitive advantage, which is in agreement with a previous study regrading auxotrophies prediction. Several auxotrophies for amino acids were predicted in gram-negative bacteria using genome-scale metabolic reconstruction and some of them were postulated to confer a fitness in in vivo experiments depending on the environmental conditions (Seif et al., 2020). Chapter 4 130 According to the results of this work, LAB can dominate the microbial community when the system selects on specific growth rate (as in discontinuous processes or very high dilution rate CSTR) because they are capable of attaining a very high substrate consumption rate. Their anabolism is more enzymatically efficient allowing LAB to allocate a higher share of the proteome to catabolic processes. In consequence, LAB are common in habitats with available peptides as for example wine, milk or grass (Carr et al., 2002), which is in complete agreement with the conclusions of this work. Resource allocation might also help explaining other metabolic behaviours such as polymer accumulation in some bacteria. In environmental conditions in which growing is limited (e.g. nitrogen is the limiting factor), these bacteria store the substrate in the form of a polymer (e.g. polyhydroxyalkanoate) to use it afterwards when growth is not hampered. Under such conditions, the accumulating bacteria outcompete other non-storing bacteria because they can develop a much higher substrate uptake rate (Jiang et al., 2011; Johnson et al., 2009). From a resource allocation perspective, it could be hypothesised that the hampered anabolism leaves room for additional catabolic enzymes, allowing in this way consuming substrate at a higher rate. Cross feeding between species is another candidate to be analysed using resource allocation. The most accepted interpretation is that dividing the catabolism in several microbial groups allows for attaining a higher consumption rate and therefore outcompete microorganisms that perform the conversion completely but at a slower rate. In fact, lactate production could also be considered as a case of cross feeding as lactate can be further metabolised to typical fermentation end products. In the experiments of Rombouts et al. (Rombouts et al., 2020) lactate is consumed by a non-glucose consumer microbial group (bacteria belonging to the genus Megasphaera mainly) producing a mixture of acetate, propionate, butyrate and valerate. In this sense, resource allocation is in line with the most-recent cross feeding interpretations, as it also predicts that in high substrate concentration conditions an association of two microbial groups consuming the substrate faster and in two steps is fitter than one single microbial group converting the substrate to the end products by its own. The faster consumption rate can be given by a more enzymatically efficient catabolism (Crabtree effect or acetate overflow) or anabolism, as shown in this work for lactate fermentation. Resource allocation explains lactic acid production 131 This mechanistic knowledge can be applied, for example, in the design of processes aiming at the intermediate compounds in two-step microbial conversions. It was already shown in this work that lactate production is promoted in SBR configurations and butyrate production is favoured in a CSTR (except at very high dilution rates) (Figure 4.3). To avoid lactate consumption the operational conditions should be tuned to avoid the presence of lactate consumers (e.g. at a dilution rate incompatible with the survival of these microbial populations). A similar design case could be the partial nitrification of ammonia to nitrite. If the goal is to achieve a complete partial nitrification, the environmental conditions should favour a two-step process and should avoid nitrite oxidation. Following the resource allocation theory, the longer complete nitrification to nitrate by one microbial group (comammox) would be favoured only at very low substrate concentrations. Indeed experimental evidences show that commamox has a higher biomass yield but lower specific growth rate than a two-step nitrification process and is only observed at low substrate environments (Hu and He, 2017). Thus complete partial nitrification can be achieved with a high ammonia load and low dissolved oxygen concentration as in this way i) the high specific growth rate strategy of two steps is favoured and ii) the second step is discouraged by a low dissolved oxygen concentration, as proved experimentally (Wei et al., 2014). 4.4. CONCLUSIONS A possible mechanism triggering lactate production in anaerobic mixed-culture fermentations was identified by means of a resource allocation FBA model. Simulation results indicate that the characteristic anabolic auxotrophy on amino acids of lactate acid bacteria is advantageous and enables a higher maximum specific growth rate than butyrate-producing bacteria, provided peptides are available. Maximum total and membrane protein concentration constraints explain the different metabolic strategies and proteome regulation behaviours observed experimentally. The model is in line with empirical observations and predicts lactate production only in rich cultivation medium and at high dilution rates in a CSTR or a (repeated) batch process. On the contrary, prototrophic butyrate-producing bacteria are predicted to dominate the community under any other operational condition. To fully validate the predictions of the model, additional experiments are needed, with a special focus in analysing the metaproteomics of the microbial community to corroborate the predicted link between proteome and the change in production strategy. In particular, a rich Chapter 4 132 cultivation medium CSTR experiment at low dilution rate is lacking in the literature to verify that anabolic auxotrophies are only advantageous at high specific growth rates. The model described in this work for the first time uses resource allocation theory to identify and explain an ecological niche for a specific group of microorganisms in a mixed microbial consortium. This chapter also defines boundaries on the application of the ATP maximisation optimisation strategy for stoichiometry prediction. The results shown that when the substrate concentration in the reactor is low (below the affinity constant), ATP efficient pathways dominate and therefore the ATP yield maximisation approach is enough to describe correctly the product spectrum. On the contrary, under high substrate concentrations, as in a batch reactor, being efficient is not enough and optimisations approaches considering also kinetic limitations, as resource allocation, might help to predict the process stoichiometry. CHAPTER 5 Bioenergetic modelling for stoichiometry prediction of protein fermentation 134 Summary Protein-rich organic effluents are suitable substrates to produce VFA in anaerobic mixed-culture fermentations. In these processes, the stoichiometry depends significantly on operational conditions such as pH or feeding characteristics but there are still no predictive tools to predict the stoichiometry of process under different conditions. In this chapter a bioenergetic metabolic model is developed for the prediction of the stoichiometry of VFA production from proteins. In particular, the effect of pH on the product yields is analysed and, for the first time, the observed changes are mechanistically explained. The model reproduces experimental results at both neutral and acidic pH and it is also capable of predicting the tendencies in product yields observed with a pH drop. It also offers mechanistic insight into the interaction among the different amino acids of a particular protein and how an amino acid might yield different products depending on the relative abundance of other amino acids. Particular emphasis is placed on the utility of this mathematical model as a process design tool and different examples are given on how to use the model for this purpose. Protein bioenergetic modelling 135 5.1. INTRODUCTION Suitable organic effluents for mixed-culture fermentations (MCF) at industrial scale include the organic fraction of urban waste or agro-industrial residual streams (e.g. cheese whey or canning industry waste). These organic wastes contain carbohydrates, proteins and lipids. While short carbohydrates have been extensively tackled from an experimental (Temudo et al., 2007) and modelling (González-Cabaleiro et al., 2015; Rodriguez et al., 2006) point of view, protein anaerobic fermentation has been barely addressed. Ramsay and Pullammanappallil (2001) proposed a product spectrum predictor for the MCF of proteins, with the objective of better understanding its anaerobic digestion to methane. In that work it was assumed that the outcome of protein MCF is unaltered by changes in operational conditions (e.g. pH) and that the different amino acids (AA) are degraded always through the same pathway. Protein conversion is also assumed to be complete in all the cases. That means that only the protein composition in AA would affect the product spectrum as their degradation pathways are fixed. However, experimental evidence contradicts most of these assumptions. Protein degradation is not complete and the degradation extent can be affected by pH (Breure and van Andel, 1984; Yu and Fang, 2003), temperature (Yu and Fang, 2003) or dilution rate (Breure et al., 1986b). Moreover, the resulting product spectrum is dependent on operational parameters such as pH (Breure et al., 1986a; Breure and van Andel, 1984). The objective of this chapter is to develop a bioenergetic metabolic model for the production of VFA from proteins in anaerobic fermentation processes using mixed cultures of microorganisms. This work pretends to give insight for the first time on the degradation mechanisms of the different AA and to predict the stoichiometry of VFA production in protein MCF, the protein conversion and how they are affected by the environmental conditions of the reactor. The influence of pH in the process outcome is specially studied because it is one of the most manipulable design variables and due to its high impact on the energetics of the system. 5.2. MODEL DESCRIPTION The model is built on the mass balances in a CSTR of the different compounds (states). There are 68 states, of which 3 are moieties related with ATP (ATP, ADP and Pi). The rest represent the concentration of different intracellular compounds (24), extracellular compounds in the bulk reactor (40), gaseous Chapter 5 136 compounds (3) and biomass. For simplicity, protein hydrolysis is omitted in the model and is directly considered a mixture of AA, as the limiting step in protein fermentation is AA fermentation (Duong et al., 2019). NAD+ to NADH ratio is set fixed to a value of 10 and the intracellular AA concentrations are assumed constant at a value of 0.1 mM and therefore are not states. Regarding the kinetic parameters, the maximum uptake rate is set to 0.75 mol AA LX-1 h-1 and the affinity constant as 1 mM for all AA. There are 99 possible reactions, resulting in a 68x99 metabolic network matrix. Amongst all the reaction rates, 22 of them are independent, i.e. depending solely on extracellular concentrations. The model balances and processes are described in detail in the model development chapter (Chapter 2). 5.2.1. Metabolic network The metabolic network used in the model is formed by the degradation pathways of 17 AAs: alanine, arginine, asparagine, aspartate, cysteine, glutamate, glutamine, glycine, histidine, isoleucine, leucine, lysine, methionine, proline, serine, threonine and valine. AA containing aromatic side chains were not included in the metabolic network (phenylalanine, tyrosine and tryptophan). The products included in the network are fatty acids from C1 to C6, ethanol, formate, methanethiol, hydrogen sulphide, CO2 and H2. Butyrate and valerate are present in both their linear and branched form and in the case of caproate only the branched appears as a product. The considered possible end-products of each AA are shown in Table 5.1. The metabolic network construction and the pathway selection criteria are discussed thoroughly in Chapter 2. Protein bioenergetic modelling 137 Table 5.1. Summarized amino acid metabolic network. Amino acid End products Comments Ref. Alanine (Ala) Pyruvate, NADH Arginine (Arg) Proline, ATP, CO2 Via ornithine Alanine, acetyl-CoA, ATP, NADH, CO2 Via ornithine 1 Aspartate (Asp) Pyruvate, NADH, CO2. Via oxaloacetate Succinate, NAD+ Via fumarate 2 Propionate, NAD+, CO2. Via fumarate and succinate Cysteine (Cys) Pyruvate, H2S 3 Glutamate (Glu) Pyruvate, acetate 4, 5 Butyrate, NAD+, CO2 Via glutaconyl-CoA and crotonyl-CoA with two PT 4, 5 Glycine (Gly) Acetate, ATP, NAD+ 6 Histidine (His) Glutamate, formamide Lysine (Lys) Butyrate, acetate, ATP 7 Proline (Pro) ½ acetate, ½ propionate, ½ nvalerate, ½ ATP, ½ NAD+ Via 5-aminovalerate 8 Serine (Ser) Pyruvate, ATP 9 Threonine (Thr) Propionate, ATP, Fdred. Via 2-oxobutyrate 9 Glycine, acetyl-CoA, NADH Via 2-amino-3oxobutyrate 9 Valine (Val) Isobutyrate, ATP, NADH, Fdred 10 Isoleucine (Ile) Isovalerate, ATP, NADH, Fdred 10 Leucine (Leu) Isovalerate, ATP, NADH, Fdred Oxidative pathway 10 Isocaproate, NAD+ Reductive pathway 11 Methionine (Met) Propionate, methanethiol, ATP, Fdred 12 Butyrate, methanethiol, NAD+ Either H2 production or a PT 12 Glutamine (Gln) Glutamate Asparagine (Asn) Aspartate Fdred: Reduced ferredoxin; PT: proton translocation; 1: Uematsu et al. (2003); 2:Unden et al. (2013); 3: Loddeke et al. (2017); 4: Buckel (2001); 5: Buckel and Barker (1974); 6: Andreesen (1994); 7: Kreimeyer et al. (2007); 8: Barker et al. (1987); 9: Sawers (1998); 10: Elsden and Hilton (1978); 11: Simon et al. (1985); 12: Bonnarme et al. (2001). Chapter 8 240 with a lactate pulse at a specific time, determined by simulation, would help to have a better description of the process. Also, an exogenous additional of lactate could alter the product selectivity of the processes and this strategy could be optimised with the models to steer the product spectrum. Figure 8.5. Proposed information flows between experimental information and model development in the envisioned design methodology. This methodology has the additional advantage of continuous improving the accuracy of the model predictions, as with each new operational condition or strategy proposed, the model is tested in a different environmental condition. In this way, it is avoided that both the model and the experimental setup are centred in a very concise region of the environmental conditions space. The examples of strategies given before would be a result of using the modelling tools and it is very likely that without using models they would not be envisioned. Likewise, the additional rounds of experiments would provide the needed confirmation and the opportunity to keep refining the model and of finding new strategies to accomplish the design objective. The strategies sketched here are just an example of the potential we envision of a design methodology that combines the richness of the experimental information with the power of predictive tools. EXPERIMENTS MODELLING ✓Propose new operational conditions ✓Propose validation experiments ✓Test new operational strategies (e.g. pulses, feeding regime) Final design ✓Kinetic parameters ✓Microbiological and proteomic information ✓Model validation REFERENCES References 243 Agler, M.T., Wrenn, B.A., Zinder, S.H., Angenent, L.T., 2011. Waste to bioproduct conversion with undefined mixed cultures: the carboxylate platform. Trends Biotechnol 29, 70–78. Alberty, R.A., 2010. Biochemical thermodynamics and rapid-equilibrium enzyme kinetics. J. Phys. Chem. B 114, 17003–17012. Alberty, R.A., 2006. Biochemical Thermodynamics: Applications of Mathernatica. Wiley, Hoboken. Alibardi, L., Cossu, R., 2016. Effects of carbohydrate, protein and lipid content of organic waste on hydrogen production and fermentation products. Waste Manag. 47, 69–77. Alkaya, E., Kaptan, S., Ozkan, L., Uludag-Demirer, S., Demirer, G.N., 2009. Recovery of acids from anaerobic acidification broth by liquid-liquid extraction. Chemosphere 77, 1137–1142. Alves de Oliveira, R., Komesu, A., Vaz Rossell, C.E., Maciel Filho, R., 2018. Challenges and opportunities in lactic acid bioprocess design—From economic to production aspects. Biochem. Eng. J. 133, 219–239. Alves, M.M., Pereira, M.A., Sousa, D.Z., Cavaleiro, A.J., Picavet, M., Smidt, H., Stams, A.J.M., 2009. Waste lipids to energy: How to optimize methane production from long-chain fatty acids (LCFA). Microb. Biotechnol. 2, 538–550. Andreesen, J.R., 1994. Glycine metabolism in anaerobes. Antonie Van Leeuwenhoek 66, 223–237. Andreesen, J.R., Bahl, H., Gottschalk, G., 1989. Introduction to the Physiology and Biochemistry of the Genus Clostridium, in: Clarke, N.P.M. and D.J. (Ed.), Clostridia. Springer Science + Business Media New York, pp. 27–62. Angelidaki, I., Ellegaard, L., Ahring, B.K., 1999. A comprehensive model of anaerobic bioconversion of complex substrates to biogas. Biotechnol. Bioeng. 63, 363–372. Angenent, L.T., Karim, K., Al-Dahhan, M.H., Wrenn, B.A., DomíguezEspinosa, R., 2004. Production of bioenergy and biochemicals from industrial and agricultural wastewater. Trends Biotechnol 22, 477–485. Angenent, L.T., Kleerebezem, R., 2011. Crystal ball. Microb. Biotechnol. 4, 109– 137. References 244 Arslan, D., Steinbusch, K.J.J., Diels, L., Hamelers, H.V.M., Strik, D.P.B.T.B., Buisman, C.J.N., De Wever, H., 2016. Selective short-chain carboxylates production: A review of control mechanisms to direct mixed culture fermentations. Crit. Rev. Environ. Sci. Technol. 46, 592–634. Atasoy, M., Owusu-Agyeman, I., Plaza, E., Cetecioglu, Z., 2018. Bio-based volatile fatty acid production and recovery from waste streams: Current status and future challenges. Bioresour. Technol. 268, 773–786. Bachmann, H., Molenaar, D., Branco dos Santos, F., Teusink, B., 2017. Experimental evolution and the adjustment of metabolic strategies in lactic acid bacteria. FEMS Microbiol. Rev. 41, S201–S219. Bai, J., Liu, H., Yin, B., Ma, H., 2015. Modeling of enhanced VFAs production from waste activated sludge by modified ADM1 with improved particle swarm optimization for parameters estimation. Biochem. Eng. J. 103, 22– 31. Bai, J., Liu, H., Yin, B., Ma, H., Chen, X., 2017. Modified ADM1 for modeling free ammonia inhibition in anaerobic acidogenic fermentation with highsolid sludge. J. Environ. Sci. (China) 52, 58–65. Baldwin, S.A., Perham, R.N., Stribling, D., 1978. Purification and characterization of the class-II D-fructose 1,6-bisphosphate aldolase from Escherichia coli (Crookes’ strain). Biochem. J. 169, 633–641. Bar-Even, A., Flamholz, A., Noor, E., Milo, R., 2012. Thermodynamic constraints shape the structure of carbon fixation pathways. Biochim. Biophys. Acta - Bioenerg. 1817, 1646–1659. Barker, H.A., 1981. Amino Acid Degradation by Anaerobic Bacteria. Annu. Rev. Biochem. 50, 23–40. Barker, H.A., D’Ari, L., Kahn, J., 1987. Enzymatic reactions in the degradation of 5-aminovalerate by Clostridium aminovalericum. J. Biol. Chem. 262, 8994–9003. Basan, M., Hui, S., Okano, H., Zhang, Z., Shen, Y., Williamson, J.R., Hwa, T., 2015. Overflow metabolism in Escherichia coli results from efficient proteome allocation. Nature 528, 99–104. Bastidas-Oyanedel, J.R., Bonk, F., Thomsen, M.H., Schmidt, J.E., 2015. Dark fermentation biorefinery in the present and future (bio)chemical industry. Rev. Environ. Sci. Biotechnol. 14, 473–498. References 245 Bastidas-Oyanedel, J.R., Schmidt, J.E., 2018. Increasing profits in food waste biorefinery-a techno-economic analysis. Energies 11. Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S. V., Pavlostathis, S.G., Rozzi, A., Sanders, W.T.M., Siegrist, H., Vavilin, V.A., 2002a. Anaerobic Digestion Model No.1 (ADM1). IWA Publishing, London. Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S. V, Pavlostathis, S.G., Rozzi, A., Sanders, W.T., Siegrist, H., Vavilin, V.A., 2002b. The IWA Anaerobic Digestion Model No 1 (ADM1). Water Sci Technol 45, 65–73. Bauer, E., Thiele, I., 2018. From Network Analysis to Functional Metabolic Modeling of Human Gut Microbiota. Nov. Syst. Biol. Tech. 3, e00209-17. Berger, E.A., 1973. Different mechanisms of energy coupling for the active transport of proline and glutamine in Escherichia coli. Proc. Natl. Acad. Sci. U. S. A. 70, 1514–8. Bevilacqua, R., Regueira, A., Mauricio-iglesias, M., Lema, J.M., Carballa, M., 2020a. Protein composition determines the preferential consumption of amino acids during anaerobic mixed-culture fermentation. Water Res. Under review. Bevilacqua, R., Regueira, A., Mauricio-iglesias, M., Lema, J.M., Carballa, M., 2020b. Steering protein fermentation with pH to target the production of specific volatile fatty acids. In preparation. Biddle, A.S., Leschine, S., Huntemann, M., Han, J., Chen, A., Kyrpides, N., Markowitz, V., Palaniappan, K., Ivanova, N., Mikhailova, N., Ovchinnikova, G., Schaumberg, A., Pati, A., Stamatis, D., Reddy, T., Lobos, E., Goodwin, L., Nordberg, H.P., Cantor, M.N., Hua, S.X., Woyke, T., Blanchard, J.L., 2014. The complete genome sequence of Clostridium indolis DSM 755T. Stand. Genomic Sci. 9, 1089–1104. Biesterveld, S., Kok, M.D., Dijkema, C., Zehnder, A.J., Stams, A.J., 1994. Dxylose catabolism in Bacteroides xylanolyticus X5-1. Arch Microbiol 161, 521–527. Biggs, M.B., Medlock, G.L., Kolling, G.L., Papin, J.A., 2015. Metabolic network modeling of microbial communities. Wiley Interdiscip. Rev. Syst. Biol. Med. 7, 317–334. Bonnarme, P., Lapadatescu, C., Yvon, M., Spinnler, H.E., 2001. L-methionine degradation potentialities of cheese-ripening microorganisms. J. Dairy Res. 68, 663–674. References 246 Bonnarme, P., Psoni, L., Spinnler, H.E., 2000. Diversity of L-Methionine catabolism pathways in cheese-ripening bacteria. Appl. Environ. Microbiol. 66, 5514–5517. Booth, I.R., 1985. Regulation of cytoplasmic pH in bacteria. Microbiol. Rev. 49, 359–378. Borodina, I., Nielsen, J., 2014. Advances in metabolic engineering of yeast Saccharomyces cerevisiae for production of chemicals. Biotechnol. J. 9, 609–620. Bosdriesz, E., Molenaar, D., Teusink, B., Bruggeman, F.J., 2015. How fastgrowing bacteria robustly tune their ribosome concentration to approximate growth-rate maximization. FEBS J. 282, 2029–2044. Box, G.E.P., 1979. Robustness in the Strategy of Scientific Model Building, Robustness in Statistics. ACADEMIC PRESS, INC. Box, G.E.P., 1976. Science and Statistics. J. Am. Stat. Assoc. 71, 791–799. Breure, A.M., Beeftink, H.H., Verkuijlen, J., Andel, J.G. Van, 1986a. Acidogenic fermentation of protein / carbohydrate mixtures by bacterial populations adapted to one of the substrates in anaerobic chemostat cultures. Appl. Microbiol. Biotechnol. 23, 245–249. Breure, A.M., Mooijman, K.A., van Andel, J.G., 1986b. Protein degradation in anaerobic digestion: influence of volatile fatty acids and carbohydrates on hydrolysis and acidogenic fermentation of gelatin. Appl. Microbiol. Biotechnol. 24, 426–431. Breure, A.M., van Andel, J.G., 1984. Hydrolysis and acidogenic fermentation of a protein, gelatin, in an anaerobic continuous culture. Appl. Microbiol. Biotechnol. 20, 40–45. Breure, A.M., van Andel, J.G., Burger-Wiersma, T., Guijt, J., Verkuijlen, J., 1985. Hydrolysis and acidogenic fermentation of gelatin under anaerobic conditions in a laboratory scale upflow reactor. Appl. Microbiol. Biotechnol. 21, 50–54. Broer, S., kramer, R., 1990. Lysine uptake and exchange in Corynebacterium glutamicum. J. Bacteriol. 172, 7241–7248. Buchanan, B.B., Arnon, D.I., 1970. Ferredoxins-chemistry and function in photosynthesis, nitrogen fixation, and fermentative metabolism, in: Adv Enzymol Relat Subj Biochem. Interscience Publishers, New York, pp. 119– 176. References 247 Buckel, W., 2001. Unusual enzymes involved in five pathways of glutamate fermentation. Appl. Microbiol. Biotechnol. 57, 263–273. Buckel, W., 1998. Anaerobic Energy Metabolism, in: Joseph W. Lengeler, G.D. and H.G.S. (Ed.), Biology of the Prokaryotes. Blackwell Science Ltd, Oxford, UK, pp. 278–326. Buckel, W., Barker, H.A., 1974. Two Pathways of Glutamate Fermentation by Anaerobic Bacteria. J. Bacteriol. 117, 1248–1260. Buckel, W., Thauer, R.K., 2018. Flavin-based electron bifurcation, ferredoxin, flavodoxin, and anaerobic respiration with protons (Ech) or NAD+(Rnf) as electron acceptors: A historical review. Front. Microbiol. 9. Buckel, W., Thauer, R.K., 2013. Energy conservation via electron bifurcating ferredoxin reduction and proton/Na+ translocating ferredoxin oxidation. Biochim. Biophys. Acta - Bioenerg. 1827, 94–113. Bundhoo, M.A.Z., Mohee, R., 2016. Inhibition of dark fermentative biohydrogen production: A review. Int. J. Hydrogen Energy 41, 6713–6733. Carballa, M., Regueiro, L., Lema, J.M., 2015. Microbial management of anaerobic digestion: exploiting the microbiome-functionality nexus. Curr. Opin. Biotechnol. 33, 103–111. Carr, F.J., Chill, D., Maida, N., 2002. The lactic acid bacteria: A literature survey. Crit. Rev. Microbiol. 28, 281–370. Carvajal-Arroyo, J.M., Candry, P., Andersen, S.J., Props, R., Seviour, T., Ganigué, R., Rabaey, K., 2019. Granular fermentation enables high rate caproic acid production from solid-free thin stillage. Green Chem. 21, 1330–1339. Cavaleiro, A.J., Pereira, M.A., Guedes, A.P., Stams, A.J.M., Alves, M.M., Sousa, D.Z., 2016. Conversion of Cn-Unsaturated into Cn-2-Saturated LCFA Can Occur Uncoupled from Methanogenesis in Anaerobic Bioreactors. Environ. Sci. Technol. 50, 3082–3090. Ceze, M., Fidkowski, K.J., 2015. Constrained pseudo-transient continuation. Int. J. Numer. Methods Eng. 102, 1683–1703. Chen, Y., Jiang, S., Yuan, H., Zhou, Q., Gu, G., 2007. Hydrolysis and acidification of waste activated sludge at different pHs. Water Res. 41, 683– 689. Costa, E., Pérez, J., Kreft, J.U., 2006. Why is metabolic labour divided in References 248 nitrification? Trends Microbiol. 14, 213–219. D´Alessio, G., Josse, J., 1971. Glyceraldehyde Phosphate Dehydrogenase of Escherichia coli. J. Biol. Chem. 246, 4326–4333. Dauner, M., Sauer, U., 2001. Stoichiometric growth model for riboflavinproducing Bacillus subtilis. Biotechnol. Bioeng. 76, 132–143. Davila-Vazquez, G., Arriaga, S., Alatriste-Mondragón, F., de León-Rodríguez, A., Rosales-Colunga, L.M., Razo-Flores, E., 2008. Fermentative biohydrogen production: trends and perspectives. Rev. Environ. Sci. Bio/Technology 7, 27–45. de Jong, E., Jungmeier, G., 2015. Biorefinery Concepts in Comparison to Petrochemical Refineries, Industrial Biorefineries and White Biotechnology. Elsevier B.V. de Kok, S., Meijer, J., van Loosdrecht, M.C., Kleerebezem, R., 2013. Impact of dissolved hydrogen partial pressure on mixed culture fermentations. Appl Microbiol Biotechnol 97, 2617–2625. de Vries, M.C., Vaughan, E.E., Kleerebezem, M., de Vos, W.M., 2006. Lactobacillus plantarum-survival, functional and potential probiotic properties in the human intestinal tract. Int. Dairy J. 16, 1018–1028. Dennis, P.P., Bremer, H., 2008. Modulation of Chemical Composition and Other Parameters of the Cell at Different Exponential Growth Rates. EcoSal Plus 3. Dharmadi, Y., Murarka, A., Gonzalez, R., 2006. Anaerobic fermentation of glycerol by Escherichia coli: A new platform for metabolic engineering. Biotechnol. Bioeng. 94, 821–829. Dinamarca, C., Bakke, R., 2009. Apparent hydrogen consumption in acid reactors: observations and implications. Water Sci. Technol. 59, 1441– 1447. Dinamarca, C., Ganan, M., Liu, J., Bakke, R., 2011. H2 consumption by anaerobic non-methanogenic mixed cultures. Water Sci Technol 63, 1582–1589. Driessen, A.J.M., Kodde, J., De Jong, S., Konings, W.N., 1987. Neutral amino acid transport by membrane vesicles of Streptococcus cremoris is subject to regulation by internal pH. J. Bacteriol. 169, 2748–2754. Driessen, A.J.M., Van Leeuwen, C., Konings, W.N., 1989. Transport of basic References 249 amino acids by membrane vesicles of Lactococcus lactis. J. Bacteriol. 171, 1453–1458. Duong, T.H., Grolle, K., Nga, T.T.V., Zeeman, G., Temmink, H., Van Eekert, M., 2019. Protein hydrolysis and fermentation under methanogenic and acidifying conditions. Biotechnol. Biofuels 12, 1–10. Dürre, P., 2008. Fermentative butanol production: Bulk chemical and biofuel. Ann. N. Y. Acad. Sci. 1125, 353–362. Elbehti, A., Brasseur, G., Lemesle-Meunier, D., 2000. First evidence for existence of an uphill electron transfer through the bc1 and NADH-Q oxidoreductase complexes of the acidophilic obligate chemolithotrophic ferrous ion-oxidizing bacterium Thiobacillus ferrooxidans. J. Bacteriol. 182, 3602–3606. Elsden, S.R., Hilton, M.G., 1978. Volatile acid production from threonine, valine, leucine and isoleucine by clostridia. Arch. Microbiol. 117, 165–172. Excherichia, I.N., 1972. Mechanisms of Active Transport Bacterial Membrane Vesicles in Isolated. J. Biol. Chem. 248, 7858–7863. Fang, H.H.P., Liu, H., 2002. Effect of pH on hydrogen production from glucose by a mixed culture. Bioresour. Technol. 82, 87–93. Fang, H.H.P., Yu, H., 2002. Mesophilic acidification of gelatinaceous wastewater. J. Biotechnol. 93, 99–108. Fernández, A., Huang, S., Seston, S., Xing, J., Hickey, R., Criddle, C., Tiedje, J., 1999. How stable is stable? Function versus community composition. Appl. Environ. Microbiol. 65, 3697–3704. Fernández, F.J., Villaseñor, J., Infantes, D., 2011. Kinetic and stoichiometric modelling of acidogenic fermentation of glucose and fructose. Biomass and Bioenergy 35, 3877–3883. Flamholz, A., Noor, E., Bar-Even, A., Milo, R., 2012. EQuilibrator - The biochemical thermodynamics calculator. Nucleic Acids Res. 40, 770–775. Flotats, X., Palatsi, J., Ahring, B.K., Angelidaki, I., 2006. Identifiability study of the proteins degradation model, based on ADM1, using simultaneous batch experiments. Water Sci. Technol. 54, 31–39. References 256 Loddeke, M., Schneider, B., Oguri, T., Mehta, I., Xuan, Z., Reitzer, L., 2017. Anaerobic Cysteine Degradation and Potential Metabolic Coordination in Salmonella enterica and Escherichia coli. J. Bacteriol. 199, 1–16. Mahadevan, R., Edwards, J.S., Doyle, F.J., 2002. Dynamic flux balance analysis of diauxic growth. Biophys. J. 83, 1331–1340. Mans, R., Daran, J.M.G., Pronk, J.T., 2018. Under pressure: evolutionary engineering of yeast strains for improved performance in fuels and chemicals production. Curr. Opin. Biotechnol. 50, 47–56. Martin, W.F., 2012. Hydrogen, metals, bifurcating electrons, and proton gradients: The early evolution of biological energy conservation. FEBS Lett. 586, 485–493. McCarty, P.L., 2007. Thermodynamic electron equivalents model for bacterial yield prediction: Modifications and comparative evaluations. Biotechnol. Bioeng. 97, 377–388. Meinecke, B., Bertram, J., Gottschalk, G., 1989. Purification and characterization of the pyruvate-ferredoxin oxidoreductase from Clostridium acetobutylicum. Arch. Microbiol. 152, 244–250. Meister, A., 2016. On the Enzymology of Amino Acid Transport Published by : American Association for the Advancement of Science Stable URL : http://www.jstor.org/stable/1735289 Linked references are available on JSTOR for this article : - On tlle Enzymology of Amino Acid Tr 180, 33– 39. Melendez-Hevia, E., 1990. The game of the pentose phosphate cycle: A mathematical approach to study the optimization in design of metabolic pathways during evolution. Biomed. Biochim. Acta 49, 903–916. Millat, T., Janssen, H., Thorn, G.J., King, J.R., Bahl, H., Fischer, R.J., Wolkenhauer, O., 2013. A shift in the dominant phenotype governs the pH-induced metabolic switch of Clostridium acetobutylicumin phosphatelimited continuous cultures. Appl. Microbiol. Biotechnol. 97, 6451–6466. Mnif, S., Chamkha, M., Labat, M., Sayadi, S., 2011. Simultaneous hydrocarbon biodegradation and biosurfactant production by oilfield-selected bacteria. J. Appl. Microbiol. 111, 525–536. Mohd-Zaki, Z., Bastidas-Oyanedel, J., Lu, Y., Hoelzle, R., Pratt, S., Slater, F., Batstone, D., 2016. Influence of pH Regulation Mode in Glucose References 257 Fermentation on Product Selection and Process Stability. Microorganisms 4, 2. Molenaar, D., Van Berlo, R., De Ridder, D., Teusink, B., 2009. Shifts in growth strategies reflect tradeoffs in cellular economics. Mol. Syst. Biol. 5, 1–10. Mori, M., Hwa, T., Martin, O.C., De Martino, A., Marinari, E., 2016. Constrained Allocation Flux Balance Analysis. PLoS Comput. Biol. 12, 1–24. Mosey, F.E., 1983. Mathematical modelling of the anaerobic digestion process: Regulatory mechanisms for the formation of short-chain volatile acids from glucose. Water Sci. Technol. 15, 209–232. Nielsen, J., Villadsen, J., 1994. Bioreaction Engineering Principles, First. ed. Springer US, New York. Nilsson, A., Nielsen, J., 2016. Metabolic Trade-offs in Yeast are Caused by F1F0ATP synthase. Sci. Rep. 6, 1–11. Ningthoujam, S.S., Talukdar, A.D., Sarker, S.D., Nahar, L., Choudhury, M.D., 2018. Prediction of Medicinal Properties Using Mathematical Models and Computation, and Selection of Plant Materials, Computational Phytochemistry. Elsevier Inc. Noor, E., Haraldsdóttir, H.S., Milo, R., Fleming, R.M.T., 2013. Consistent Estimation of Gibbs Energy Using Component Contributions. PLoS Comput. Biol. 9, e1003098. O’Malley, M.A., 2008. “Everything is everywhere: but the environment selects”: ubiquitous distribution and ecological determinism in microbial biogeography. Stud. Hist. Philos. Sci. Part C Stud. Hist. Philos. Biol. Biomed. Sci. 39, 314–325. Octave, S., Thomas, D., 2009. Biorefinery: Toward an industrial metabolism. Biochimie 91, 659–664. Oh, S.T., Martin, A.D., 2007. Thermodynamic equilibrium model in anaerobic digestion process. Biochem. Eng. J. 34, 256–266. Ohnishi, A., 2015. Megasphaera as lactate-utilizing hydrogen-producing bacteria, in: Kalia, V.C. (Ed.), Microbial Factories: Biofuels, Waste Treatment: Volume 1. Tokyo, pp. 47–71. References 258 Oxender, D.L., Christensen, H.N., 1963. Distinct Mediating Amino Systems for the Transport Acids by the Ehrlich Cell * of Neutral 238. Padan, E., Zilberstein, D., Schuldiner, S., 1981. pH homesstasis in bacteria. BBA - Rev. Biomembr. 650, 151–166. Pavlostathis, S.G., Giraldo-Gomez, E., 1991. Kinetics of anaerobic treatment: A critical review. Crit. Rev. Environ. Control 21, 411–490. Payot, S., Guedon, E., Gelhaye, E., Petitdemange, H., 1999. Induction of lactate production associated with a decrease in NADH cell content enables growth resumption of Clostridium cellulolyticum in batch cultures on cellobiose. Res. Microbiol. 150, 465–473. Peekhaus, N., Conway, T., 1998. What ’ s for Dinner ?: Entner-Doudoroff Metabolism in Escherichia coli. Society 180, 3495–3502. Perez-Garcia, O., Lear, G., Singhal, N., 2016. Metabolic network modeling of microbial interactions in natural and engineered environmental systems. Front. Microbiol. 7. Peters, J.W., Miller, A.F., Jones, A.K., King, P.W., Adams, M.W.W., 2016. Electron bifurcation. Curr. Opin. Chem. Biol. 31, 146–152. Peters, V., Janssen, P., Conrad, R., 2006. Efficiency of hydrogen utilization during unitrophic and mixotrophic growth of Acetobacterium woodii on hydrogen and lactate in the chemostat. Petitdemange, H., Cherrier, C., Raval, R., Gay, R., 1976. Regulation of the NADH and NADPH-ferredoxin oxidoreductases in clostridia of the butyric group. Biochim Biophys Acta 421, 334–337. Pfeiffer, T., Bonhoeffer, S., 2004. Evolution of cross-feeding in microbial populations. Am. Nat. 163. Pfeiffer, T., Schuster, S., Bonhoeffer, S., 2001. Cooperation and competition in the evolution of ATP-producing pathways. Science (80-. ). 293, 1436. Poole, R.J., 1978. Energy Coupling for Membrane Transport. Annu. Rev. Plant Physiol. 29, 437–460. Poolman, B., Driessen, A.J., Konings, W.N., 1987. Regulation of solute transport in streptococci by external and internal pH values. Microbiol. Rev. 51, 498– 508. Prusiner, S., Milner, L., 1970. A rapid radioactive assay for glutamine synthetase, References 259 glutaminase, asparagine synthetase, and asparaginase. Anal. Biochem. 37, 429–438. Rafrafi, Y., Trably, E., Hamelin, J., Latrille, E., Meynial-Salles, I., Benomar, S., Giudici-Orticoni, M.T., Steyer, J.P., 2013. Sub-dominant bacteria as keystone species in microbial communities producing bio-hydrogen. Int. J. Hydrogen Energy 38, 4975–4985. Rai, R., Keshavarz, T., Roether, J.A., Boccaccini, A.R., Roy, I., 2011. Medium chain length polyhydroxyalkanoates, promising new biomedical materials for the future. Mater. Sci. Eng. R Reports 72, 29–47. Ramsay, I.R., 1997. Modelling and Control of High-Rate Anaerobic Wastewater Treatment Systems. The University of Queensland. Ramsay, I.R., Pullammanappallil, P.C., 2001. Protein degradation during anaerobic wastewater treatment: Derivation of stoichiometry. Biodegradation 12, 247–257. Rasmussen, H., Sørensen, H.R., Meyer, A.S., 2014. Formation of degradation compounds from lignocellulosic biomass in the biorefinery: Sugar reaction mechanisms. Carbohydr. Res. 385, 45–57. Reddy, M.V., Mohan, S.V., Chang, Y.C., 2018. Medium-Chain Fatty Acids (MCFA) Production Through Anaerobic Fermentation Using Clostridium kluyveri: Effect of Ethanol and Acetate. Appl. Biochem. Biotechnol. 185, 594–605. Ren, N.-Q., Zhao, L., Chen, C., Guo, W.-Q., Cao, G.-L., 2016. A review on bioconversion of lignocellulosic biomass to H2: Key challenges and new insights. Bioresour. Technol. 215, 92–99. Rodriguez, J., Kleerebezem, R., Lema, J.M., van Loosdrecht, M.C., 2006. Modeling product formation in anaerobic mixed culture fermentations. Biotechnol Bioeng 93, 592–606. Rombouts, J.L., Kranendonk, E., Regueira, A., Weissbrodt, D., Kleerebezem, R., van Loosdrecht, M.C.M., 2020. Selecting for lactic acid producing and utilising bacteria in anaerobic enrichment cultures. Biotechnol Bioeng 117, 1281-1293. Rombouts, J.L., Mos, G., Weissbrodt, D.G., Kleerebezem, R., van Loosdrecht, M.C.M., 2019a. Diversity and metabolism of xylose and glucose fermenting microbial communities in sequencing batch or continuous culturing. FEMS References 260 Microbiol. Ecol. 95. Rombouts, J.L., Mos, G., Weissbrodt, D.G., Kleerebezem, R., Van Loosdrecht, M.C.M., 2019b. The impact of mixtures of xylose and glucose on the microbial diversity and fermentative metabolism of sequencing-batch or continuous enrichment cultures. FEMS Microbiol. Ecol. 95, 1–10. Saady, N.M.C., 2013. Homoacetogenesis during hydrogen production by mixed cultures dark fermentation: Unresolved challenge. Int. J. Hydrogen Energy 38, 13172–13191. Sabra, W., Zeng, A.P., Deckwer, W.D., 2001. Bacterial alginate: Physiology, product quality and process aspects. Appl. Microbiol. Biotechnol. 56, 315– 325. Sakai, S., Nakashimada, Y., Inokuma, K., Kita, M., Okada, H., Nishio, N., 2005. Acetate and ethanol production from H2 and CO2 by Moorella sp. using a repeated batch culture. Saltelli, A., Ratto, M., Andres, T., Campolongo, F., Cariboni, J., Gatelli, D., Saisana, M., Tarantola, S., 2008. Global Sensitivity Analysis: The Primer. John Wiley & Sons, Ltd, West Sussex. Satpute, S.K., Banat, I.M., Dhakephalkar, P.K., Banpurkar, A.G., Chopade, B.A., 2010. Biosurfactants, bioemulsifiers and exopolysaccharides from marine microorganisms. Biotechnol. Adv. 28, 436–450. Saum, S.H., Müller, V., 2007. Salinity-dependent switching of osmolyte strategies in a moderately halophilic bacterium: Glutamate induces proline biosynthesis in Halobacillus halophilus. J. Bacteriol. 189, 6968–6975. Sawers, G., 1998. The anaerobic degradation of L-serine and L-threonine in enterobacteria: Networks of pathways and regulatory signals. Arch. Microbiol. 171, 1–5. Schomburg, I., Chang, A., Hofmann, O., Ebeling, C., Ehrentreich, F., Schomburg, D., 2002. BRENDA: A resource for enzyme data and metabolic information. Trends Biochem. Sci. 27, 54–56. Schuetz, R., Kuepfer, L., Sauer, U., 2007. Systematic evaluation of objective functions for predicting intracellular fluxes in Escherichia coli. Mol. Syst. Biol. 3. Schumacher, R., 2018. Metabolic trade-offs arising from increased free energy conservation in Saccharomyces cerevisiae. Delft University of Technology. References 261 Scott, M., Klumpp, S., Mateescu, E.M., Hwa, T., 2014. Emergence of robust growth laws from optimal regulation of ribosome synthesis. Mol. Syst. Biol. 10, 747. Seeliger, S., Janssen, P.H., Schink, B., 2002. Energetics and kinetics of lactate fermentation to acetate and propionate via methylmalonyl-CoA or acrylylCoA. FEMS Microbiol. Lett. 211, 65–70. Seif, Y., Sonal, K., Hefner, Y., Anand, A., Yang, L., 2020. Metabolic and genetic basis for auxotrophies in Gram-negative species. Shen, C.R., Lan, E.I., Dekishima, Y., Baez, A., Cho, K.M., Liao, J.C., 2011. Driving forces enable high-titer anaerobic 1-butanol synthesis in Escherichia coli. Appl. Environ. Microbiol. 77, 2905–2915. Shi, E., Li, J., Zhang, M., 2019. Application of IWA Anaerobic Digestion Model No. 1 to simulate butyric acid, propionic acid, mixed acid, and ethanol type fermentative systems using a variable acidogenic stoichiometric approach. Water Res. 161, 242–250. Shilling, M., Matt, L., Rubin, E., Visitacion, M.P., Haller, N.A., Grey, S.F., Woolverton, C.J., 2013. Antimicrobial effects of virgin coconut oil and its medium-chain fatty acids on clostridium difficile. J. Med. Food 16, 1079– 1085. Siegrist, H., Vogt, D., Garcia-Heras, J.L., Gujer, W., 2002. Mathematical model for mesoand thermophilic anaerobic sewage sludge digestion. Environ. Sci. Technol. 36, 1113–1123. Simon, H., Bader, J., Günther, H., Neumann, S., Thanos, J., 1985. Chirale Verbindungen durch biokatalytische Reduktionen. Angew. Chemie 97, 541–555. Sims, G.K., Sommers, L.E., Konopka, A., 1986. Degradation of pyridine by Micrococcus luteus isolated from soil. Appl. Environ. Microbiol. 51, 963– 968. Singhania, R.R., Patel, A.K., Christophe, G., Fontanille, P., Larroche, C., 2013. Biological upgrading of volatile fatty acids, key intermediates for the valorization of biowaste through dark anaerobic fermentation. Bioresour. Technol. 145, 166–174. Smid, E.J., Lacroix, C., 2013. Microbe-microbe interactions in mixed culture food fermentations. Curr. Opin. Biotechnol. 24, 148–154. Spirito, C.M., Richter, H., Rabaey, K., Stams, A.J.M., Angenent, L.T., 2014. Chain References 262 elongation in anaerobic reactor microbiomes to recover resources from waste. Curr. Opin. Biotechnol. 27, 115–122. Stams, A.J.M., Plugge, C.M., 2009. Electron transfer in syntrophic communities of anaerobic bacteria and archaea. Nat. Rev. Microbiol. 7, 568–577. Steinbusch, K.J.J., Hamelers, H.V.M., Plugge, C.M., Buisman, C.J.N., 2011. Biological formation of caproate and caprylate from acetate: Fuel and chemical production from low grade biomass. Energy Environ. Sci. 4, 216– 224. Stouthamer, A.H., 1973. A theoretical study on the amount of ATP required for synthesis of microbial cell material. Antonie Van Leeuwenhoek 39, 545– 565. Straub, M., Demler, M., Weuster-Botz, D., Dürre, P., 2014. Selective enhancement of autotrophic acetate production with genetically modified Acetobacterium woodii. J. Biotechnol. 178, 67–72. Strazzera, G., Battista, F., Garcia, N.H., Frison, N., Bolzonella, D., 2018. Volatile fatty acids production from food wastes for biorefinery platforms: A review. J. Environ. Manage. 226, 278–288. Taffs, R., Aston, J.E., Brileya, K., Jay, Z., Klatt, C.G., McGlynn, S., Mallette, N., Montross, S., Gerlach, R., Inskeep, W.P., Ward, D.M., Carlson, R.P., 2009. In Silico approaches to study mass and energy flows in microbial consortia: A syntrophic case study. BMC Syst. Biol. 3, 1–16. Taherzadeh, M.J., Keikhosro, K., 2008. Bioethanol: Market and Production Processes, in: Nag, A. (Ed.), Biofuels Refining and Performance. The McGraw-Hill Companies, Inc., pp. 69–106. Tan, R., Miyanaga, K., Uy, D., Tanji, Y., 2012. Effect of heat-alkaline treatment as a pretreatment method on volatile fatty acid production and protein degradation in excess sludge, pure proteins and pure cultures. Bioresour. Technol. 118, 390–398. Temudo, M.F., Kleerebezem, R., van Loosdrecht, M., 2007. Influence of the pH on (open) mixed culture fermentation of glucose: A chemostat study. Biotechnol. Bioeng. 98, 69–79. Temudo, M.F., Mato, T., Kleerebezem, R., Van Loosdrecht, M.C.M., 2009. Xylose anaerobic conversion by open-mixed cultures. Appl. Microbiol. Biotechnol. 82, 231–239. References 263 Temudo, M.F., Muyzer, G., Kleerebezem, R., Van Loosdrecht, M.C.M., 2008a. Diversity of microbial communities in open mixed culture fermentations: Impact of the pH and carbon source. Appl. Microbiol. Biotechnol. 80, 1121–1130. Temudo, M.F., Poldermans, R., Kleerebezem, R., Van Loosdrecht, M.C.M., 2008b. Glycerol fermentation by (open) mixed cultures: A chemostat study. Biotechnol. Bioeng. 100, 1088–1098. Thauer, R.K., Jungermann, K., Decker, K., 1977. Energy conservation in chemotrophic anaerobic bacteria. Bacteriol. Rev. 41, 100–180. Thomas, T.D., Ellwood, D.C., Longyear, V.M.C., 1979. Change from homoto heterolactic fermentation by Streptococcus lactis resulting from glucose limitation in anaerobic chemostat cultures. J. Bacteriol. 138, 109–117. Tobajas, M., Garcia-Calvo, E., 1999. Determination of biomass yield for growth of Candida utilis on glucose: Black box and metabolic descriptions. World J. Microbiol. Biotechnol. 15, 431–438. Tommaso, G., Domingues, M.R., Ribeiro, R., Varesche, M.B.A., Zaiat, M., Foresti, E., 2013. Anaerobic degradation of protein: Simplified kinetic modelling and microbial dynamics. Water. Air. Soil Pollut. 224. Tuck, C.O., Pérez, E., Horváth, I.T., Sheldon, R.A., Poliakoff, M., 2012. Valorization of Biomass : Deriving More Value from Waste. Science (80-. ). 337, 695–699. Uematsu, H., Sato, N., Hossain, M.Z., Ikeda, T., Hoshino, E., 2003. Degradation of arginine and other amino acids by butyrate-producing asaccharolytic anaerobic Gram-positive rods in periodontal pockets. Arch. Oral Biol. 48, 423–429. Unden, G., Strecker, A., Kleefeld, A., Kim, O. Bin, 2013. C4-Dicarboxylate Utilization in Aerobic and Anaerobic Growth. EcoSal Plus. Vangsgaard, A.K., Mauricio-Iglesias, M., Gernaey, K. V., Smets, B.F., Sin, G., 2012. Sensitivity analysis of autotrophic N removal by a granule based bioreactor: Influence of mass transfer versus microbial kinetics. Bioresour. Technol. 123, 230–241. Vazquez, A., Oltvai, Z.N., 2011. Molecular crowding defines a common origin References 264 for the warburg effect in proliferating cells and the lactate threshold in muscle physiology. PLoS One 6, 1–9. Venkata Mohan, S., Nikhil, G.N., Chiranjeevi, P., Nagendranatha Reddy, C., Rohit, M. V., Kumar, A.N., Sarkar, O., 2016. Waste biorefinery models towards sustainable circular bioeconomy: Critical review and future perspectives. Bioresour. Technol. 215, 2–12. Wang, Y., Tashiro, Y., Sonomoto, K., 2015. Fermentative production of lactic acid from renewable materials: Recent achievements, prospects, and limits. J. Biosci. Bioeng. 119, 10–18. Waygood, B.E., Sanwal, B.D., 1974. The Control of Pyruvate Kinases of Escherichia. J. Biol. Chem. 249, 265–274. Wei, D., Xue, X., Yan, L., Sun, M., Zhang, G., Shi, L., Du, B., 2014. Effect of influent ammonium concentration on the shift of full nitritation to partial nitrification in a sequencing batch reactor at ambient temperature. Chem. Eng. J. 235, 19–26. White, D., Drummond, J., Fuqua, C., 2012. The physiology and biochemistry of prokaryotes, 4th ed. Oxford University Press, New York. Yin, J., Yu, X., Wang, K., Shen, D., 2016. Acidogenic fermentation of the main substrates of food waste to produce volatile fatty acids. Int. J. Hydrogen Energy 41, 21713–21720. Yu, H.G., Fang, H.H., 2002. Acidogenesis of dairy wastewater at various pH levels. Water Sci. Technol. 45, 201–206. Yu, H.Q., Fang, H.H.P., 2003. Acidogenesis of gelatin-rich wastewater in an upflow anaerobic reactor: influence of pH and temperature. Water Res. 37, 55–66. Yu, H.Q., Fang, H.H.P., Gu, G.W., 2002. Comparative performance of mesophilic and thermophilic acidogenic upflow reactors. Process Biochem. 38, 447–454. Yu, H.Q., Mu, Y., 2006. Biological hydrogen production in a UASB reactor with granules. II: Reactor performance in 3-year operation. Biotechnol. Bioeng. 94, 988–995. References 265 Zacharof, M.P., Lovitt, R.W., 2014. Recovery of volatile fatty acids (VFA) from complex waste effluents using membranes. Water Sci. Technol. 69, 495–503. Zeng, H., Yang, A., 2019. Modelling overflow metabolism in Escherichia coli with flux balance analysis incorporating differential proteomic efficiencies of energy pathways. BMC Syst. Biol. 13, 1–18. Zhang, F., Zhang, Y., Chen, M., van Loosdrecht, M.C., Zeng, R.J., 2013. A modified metabolic model for mixed culture fermentation with energy conserving electron bifurcation reaction and metabolite transport energy. Biotechnol Bioeng 110, 1884–1894. Zhang, M., Wu, H., Chen, H., 2014. Coupling of polyhydroxyalkanoate production with volatile fatty acid from food wastes and excess sludge. Process Saf. Environ. Prot. 92, 171–178. Zhuang, K., Izallalen, M., Mouser, P., Richter, H., Risso, C., Mahadevan, R., Lovley, D.R., 2011. Genome-scale dynamic modeling of the competition between Rhodoferax and Geobacter in anoxic subsurface environments. ISME J. 5, 305–316. Zoetemeyer, R.J., van den Heuvel, J.C., Cohen, A., 1982. pH influence on acidogenic dissimilation of glucose in an anaerobic digestor. Water Res. 16, 303–311. Zomorrodi, A.R., Islam, M.M., Maranas, C.D., 2014. D-OptCom: Dynamic Multi-level and Multi-objective Metabolic Modeling of Microbial Communities. ACS Synth. Biol. 3, 247–257. Zomorrodi, A.R., Maranas, C.D., 2012. OptCom: A multi-level optimization framework for the metabolic modeling and analysis of microbial communities. PLoS Comput. Biol. 8.