Research Collection

Doctoral Thesis

Development of non-stationary ¹³C flux analysis methods to quantify microbial and higher cell

Author(s): Hörl, Manuel

Publication Date: 2015

Permanent Link: https://doi.org/10.3929/ethz-a-010532835

Rights / License: In Copyright - Non-Commercial Use Permitted

This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use.

ETH Library DISS ETH NO. 22699

Development of non-stationary 13C flux analysis methods to quantify microbial and higher cell metabolism

A thesis submitted to attain the degree of

DOCTOR OF SCIENCES of ETH ZURICH (Dr. sc. ETH Zurich)

presented by

Manuel Hörl Dipl.-Ing., Technische Universität München born on September 18th, 1984 citizen of Germany

accepted on the recommendation of

Prof. Dr. Uwe Sauer Dr. Nicola Zamboni Prof. Dr. Julia Vorholt

2015

„Der Grund war nicht die Ursache, sondern der Auslöser.“

Franz Beckenbauer

Table of contents

Abstract 1

Zusammenfassung 4

Chapter 1 9 General Introduction

Chapter 2 37 Non-stationary 13C metabolic flux ratio analysis

Chapter 3 71 Non-stationary 13C metabolic flux ratio analysis of embryonic fibroblasts with impaired pyruvate metabolism

Chapter 4 105 The malic enzyme YtsJ could enable redox homeostasis in Bacillus subtilis by an NADPH-dependent switch in enzyme activity

Conclusions and outlook 133

List of Abbreviations 144

Acknowledgements 147

Curriculum Vitae 149

!

Abstract ! Abstract

Metabolism fuels the vital processes of every living cell by transforming available nutrients into building blocks for growth while simultaneously generating energy and redox equivalents to drive biosynthetic reactions. All these processes are facilitated by more than a thousand enzyme-mediated, metabolite-converting reactions that constitute the . The combined in vivo operation of these transformations is the major determinant of cellular physiology and this activity is defined by the material fluxes of metabolites through the reactions. Therefore, experimental assessment of intracellular fluxes is pivotal to identify metabolic phenotypes in the context of systems biology, metabolic engineering and biomedical research.

Unlike other omics measurements, metabolic fluxes cannot be measured directly but have to be inferred by integrating experimental data into computational models of metabolism. In the most elaborate version, metabolic fluxes are inferred from 13C tracer experiments, where the conversion of 13C labeled substrates leads to pathway specific labeling patterns and labeling dynamics, which provide constraints for a computational model of metabolism to iteratively estimate the fluxes. For this purpose, an ensemble of experimental methods has been developed under the general title of 13C metabolic flux analysis, which allow a routine estimation of fluxes in central carbon metabolism of well-studied model organisms grown in minimal media with a single carbon source.

However, cellular phenotypes and evolution are best understood under real-life conditions, which typically feature transient growth and heterogeneous nutrients ± conditions where current 13C flux methods cannot be rigorously applied. Therefore, it would be valuable to have a framework for estimating specific fluxes in metabolic networks from short-term experiments, independent of the media composition.

This thesis attends to this matter and expands the existing repertoire of 13C flux methods by non-stationary 13C metabolic flux ratio analysis ± a novel approach that has the potential to resolve fluxes during transitional processes and in complex media. The novel strategy is to overcome the limitations of previous methods by focusing on specific metabolic nodes where

1 Abstract ! pathways converge and to estimate local fluxes. This is achieved with dynamic labeling data of a small set of metabolites, which are in close proximity to the node of interest, by formal parameter fitting with a small-sized system of ordinary differential equations that describe the local atom mapping. This allows estimating fluxes in extremely short time-scale while neglecting the majority of reactions in metabolism.

In Chapter 2, we first established a generalized workflow for data processing, modeling, and parameter and error estimation, which takes into account all possible metabolic reaction types and the availability of mass spectrometric data on molecular ions or fragments. As a proof-of- principle, we demonstrated the approach by analyzing fluxes at key metabolic nodes in central metabolism of Bacillus subtilis and obtained estimates that were in good agreement with results from previous approaches. However, compared to these methods, which require at least 30±60 minutes of labeling, our results were derived from dynamic labeling data collected within less than a minute. This study illustrated how non-stationary 13C metabolic flux ratio analysis can improve the maximum achievable temporal resolution for flux estimation.

Similar results were obtained in Chapter 3, where we applied our framework to investigate flux changes upon disruption of mitochondrial pyruvate transport in the metabolism of mammalian cells. Deletion of the mitochondrial pyruvate carrier in mouse embryonic fibroblasts completely abolished the import of -derived pyruvate into mitochondria, which was mainly compensated by increased incorporation of the second major carbon source glutamine via the reductive TCA cycle pathway. Interestingly, we also revealed partial compensation of the impaired pyruvate transport by an increase in anaplerosis, which most probably occurred within the cytosol. Since the usual anaplerotic enzyme pyruvate carboxylase is located within the mitochondria, we speculate that a reversal of the cytosolic malic enzyme reaction explains our observation, which will have to be investigated by deletion studies in the future.

Our results from Chapter 2 indicated that such reversibility of the malic enzyme reaction also occurs in B. subtilis. Since B. subtilis contains four malic enzyme isoforms with distinct redox cofactor specificities, it has been suggested that they might contribute to NADPH metabolism by the formation of a transhydrogenation cycle, i.e. several isoenzymes with different cofactor

2 Abstract ! specificities that operate in reverse directions and thereby interconvert NADH and NADPH.

Combining results from non-stationary 13C metabolic flux ratio analysis and global 13C flux analysis, we quantified cellular NADPH metabolism of exponentially growing B. subtilis malic enzyme deletion mutants in Chapter 4. We show that transhydrogenation via the malic enzyme reaction is able to balance NADPH overproduction, but found no evidence for the operation of a conventional redox cycle between different isoforms. On the contrary, one isoform alone seemed to ensure redox homeostasis. We further investigated the potential mechanism by in vitro assays, and found that the enzyme indeed oxidized excess NADPH.

Surprisingly, we detected a concomitant switch in enzyme function towards redox neutral lactate formation, which would represent a novel way to allow the oxidation of excess NADPH without its re-formation.

In our studies, non-stationary 13C metabolic flux ratio analysis lead to the identification of unexpected modes of metabolic operation. We demonstrated that it is applicable in bacterial, as well as higher cell metabolism. It is capable of providing local flux estimates during short- term phenomena, in complex media and with the measurement of only few metabolites. Due to its targeted nature, it is ideally suited to verify hypothesis generated in large-scale metabolomics and non-targeted 13C fluxome profiling screens.

3 Zusammenfassung ! Zusammenfassung ! Der Stoffwechsel bildet die Grundlage zentraler Lebensprozesse einer jeden Zelle, indem er verfügbare Nährstoffe in zelleigene Bausteine umwandelt und gleichzeitig die benötigte

Energie und Reduktionsäquivalente für die Biosynthese und das Zellwachstum bereitstellt.

Diese Vorgänge ermöglichen mehr als 1000 enzymatisch katalysierte chemische Reaktionen zwischen Metaboliten, welche das Stoffwechselnetzwerk bilden. Der kombinierte in vivo-

Ablauf dieser Reaktionen zählt zu den Haupteinflussgrössen auf die Physiologie einer Zelle.

Bestimmt wird er durch die Materialflüsse von Metaboliten durch das Reaktionsnetzwerk. Da diese intrazellulären Flüsse den metabolischen Phänotyp einer Zelle widerspiegeln, ist ihre experimentelle Erschliessung von grundlegender Bedeutung für die Systembiologie, das

Metabolic Engineering sowie die medizinische Forschung.

Im Unterschied zu andHUHQ³Omik´-Messungen können metabolische Stoffflüsse nicht direkt gemessen werden, sondern müssen durch Computermodell-basierte Dateninterpretation messbarer Grössen ermittelt werden. Die fortschrittlichsten Varianten verwenden hierfür 13C-

Markierungsexperimente. Der Abbau 13C-markierter Substrate führt zu Markierungsmustern oder Markierungsdynamiken in intrazellulären Metaboliten, welche charakteristisch für bestimmte Stoffwechselwege sind. Diese können zur Beschränkung eines

Stoffwechselmodells verwendet werden, um die Stoffflüsse in einem iterativen Verfahren zu berechnen. Hierfür existieren bereits mehrere Methoden, welche sich unter dem Begriff

13C-Stoffflussanalyse zusammenfassen lassen. Sie ermöglichen eine routinemässige

Berechnung von Stoffflüssen innerhalb des Zentralstoffwechsels gut charakterisierter

Modellorganismen, welche mit nur einer Kohlenstoffquelle und in Minimalmedien kultiviert werden können.

Erfahrungsgemäss lassen sich Entwicklung sowie Phänotyp einer Zelle am besten unter natürlichen Bedingungen erforschen. Allerdings sind diese gekennzeichnet durch transientes

Wachstum und eine Vielzahl von Substraten ± Bedingungen, unter denen die bisherigen 13C-

Flussmethoden keine exakten Ergebnisse liefern. Daher wäre es von grossem Wert, eine

Methode zu entwickeln, welche die Bestimmung von zumindest individuellen, spezifischen

4 Zusammenfassung ! Flüssen in metabolischen Netzwerken durch kurze Experimente erlaubt ± unabhängig von der

Medienzusammensetzung. Genau hierin liegt die Zielsetzung der vorliegenden Doktorarbeit:

Sie ergänzt die existierenden 13C-Flussmethoden um eine nicht-stationäre 13C-

Flussverhältnisanalyse ± und somit um einen neuen Ansatz, der das Potenzial hat, Flüsse auch während dynamischer Prozesse und in komplexen Medien zu untersuchen.

Grundlegende Strategie ist es, die Limitierungen der bisherigen Methoden dadurch zu

überwinden, dass man sich auf bestimmte metabolische Knotenpunkte fokussiert, an denen mehrere Stoffwechselwege zusammenfliessen. Hier wiederum werden relative

Flussverhältnisse aus dynamischen Markierungsdaten einer geringen Anzahl von Metaboliten in unmittelbarer Nähe des untersuchten Knotenpunktes bestimmt. Dies geschieht durch formelle Parameterberechung mit kleinen Systemen gewöhnlicher Differentialgleichungen, welche die lokalen Atomübergänge beschreiben. Hieraus könnte die Möglichkeit hervorgehen,

Stoffflüsse während Phänomenen extrem kurzer Dauer zu bestimmen, ohne dem Grossteil von Reaktionen im metabolischen Netzwerk Beachtung schenken zu müssen.

In Kapitel 2 haben wir zuerst einen allgemeingültigen Ablauf für die Datenbearbeitung,

Modellierung, Parameter- sowie Fehlerbestimmung aufgestellt, welcher alle möglichen metabolischen Typen von Stoffwechselreaktionen und die Verfügbarkeit von

Massenspektrometriedaten molekularer Ionen oder Fragmente berücksichtigt. Um die

Funktionalität der Methode zu zeigen, wurde sie zur Analyse von drei zentralen metabolischen

Knotenpunkten im Zentralstoffwechsel von Bacillus subtilis angewendet. Die berechneten

Ergebnisse stimmten sehr gut mit denen bisheriger Methoden überein. Während diese jedoch eine experimentelle Zeitdauer von mindestens 30 bis 60 Minuten erfordern, konnten wir unsere

Ergebnisse aus experimentellen, dynamischen Daten von weniger als einer Minute bestimmen. Diese Studie zeigt, wie nicht-stationäre 13C-Flussverhältnisanalyse die maximal erreichbare zeitliche Auflösung drastisch verbessern kann.

Vergleichbare Ergebnisse wurden in Kapitel 3 erzielt: Hier wurde unsere Methode auf den

Stoffwechsel von Säugerzellen angewendet um Flussänderungen aufgrund des Verlustes des mitochondrialen Pyruvattransporters zu untersuchen. Die Deletion des mitochondrialen

5 Zusammenfassung ! Pyruvattransporters in embryonalen Maus-Fibroblasten führte zur vollständigen Unterbindung des Imports des aus Glukose entstandenen Pyruvats in die Mitochondrien. Dieser Verlust wurde hauptsächlich durch erhöhte Einbindung der zweiten Haupt-Kohlenstoffquelle Glutamin

über den reduktiven Citratzyklus kompensiert. Interessanterweise stellten wir zudem eine teilweise Kompensation des unterbundenen Pyruvattransports durch erhöhten anaplerotischen Fluss fest, welcher sehr wahrscheinlich im Cytosol auftrat. Da Pyruvat

Karboxylase ± das Enzym, welches normalerweise die anaplerotische Reaktion in höheren

Zellen katalysiert ± innerhalb der Mitochondrien vorkommt, nehmen wir an, dass der anaplerotische Fluss durch Rückwärtsreaktion des cytosolischen Malatenzyms zustande kam.

Diese Hypothese muss noch durch Deletionsstudien verifiziert werden.

Unsere Ergebnisse aus Kapitel 2 deuteten an, dass die Malatenzymreaktion in B. subtilis ebenfalls reversibel ist. Da B. subtilis vier Isoenzyme des Malatenzyms enthält, welche unterschiedlich spezifisch gegenüber den Redoxkofaktoren NADH und NADPH sind, wird bereits angenommen, dass diese eine Rolle im Redoxstoffwechsel einnehmen könnten.

Genauer gesagt wird spekuliert, dass verschiedene Isoenzyme mit unterschiedlichen

Spezifitäten gegenüber Redoxkofaktoren einen Transhydrogenase-Zyklus bilden, indem sie in unterschiedliche Richtungen arbeiten und dabei NADH und NADPH ineinander umwandeln können. Durch Kombination der Resultate aus nicht-stationärer 13C-Flussverhältnisanalyse und globaler Flussanalyse, quantifizierten wir den zellulären NADPH Stoffwechsel exponentiell wachsender B. subtilis Malatenzym Deletionsmutanten in Kapitel 4. Wir zeigen, dass eine

Malatenzym-katalysierte Reaktion die Transhydrogenierung ermöglicht und dadurch die zelluläre NADPH Bilanz schliesst. Jedoch konnten wir die Existenz eines Redoxzyklus mehrerer Isoenzyme nicht nachweisen. Im Gegenteil: unsere Ergebnisse deuten an, dass ein

Isoenzym alleine die Redoxhomöostase aufrechterhält. Durch genauere Untersuchungen des

Mechanismus in in vitro-Enzymassays, konnten wir zeigen, dass das Enzym tatsächlich im

Stande ist, überschüssiges NADPH zu oxidieren. Überaschenderweise stellten wir zudem eine

Änderung der Enzymaktivität, d.h. die Redoxkofaktor-unabhängige Bildung von Laktat, fest.

6 Zusammenfassung ! Dies wäre ein neuer Mechanismus um die Oxidation überschüssigen NADPHs ohne erneute

Bildung zu ermöglichen.

In unseren Untersuchungen konnten durch nicht-stationäre 13C-Flussverhältnisanalyse neue, unerwartete Funktionsweisen des Stoffwechsels aufgedeckt werden. Wir zeigen, dass die entwickelte Methode sowohl für den Stoffwechsel von Bakterien, als auch für höhere Zellen angewendet werden kann. Sie ermöglicht die Bestimmung lokaler Flüsse während

Phänomenen kurzer Dauer sowie in komplexen Medien und benötigt hierfür nur wenige

Metabolitmessungen. Sie ist zielgerichtet und daher ideal geeignet um Hypothesen gross angelegter Screenings, z.B. durch ungezielte Metabolomanalyse oder 13C Fluxome Profiling, zu bestätigen.

! ! ! ! ! ! ! ! ! ! ! !

7 8 ! Chapter 1 !

General Introduction

9 Chapter 1 ! Cellular metabolism

Cellular metabolism is a fundamental process that enables life even in the most extreme habitats on our planet and under adverse conditions. It allows organisms to obtain chemical energy by capturing solar energy or degrading energy-rich nutrients from the environment. It converts the simplest compounds LQWR WKH FHOO¶V RZQ FKDUDFWHULVWLF PROHFXOHV, including macromolecules such as lipids, nucleic acids, proteins and polysaccharides that determine cellular structure, contain its genetic information and allow the storage of nutrients for periods of starvation1. Additionally, metabolism synthesizes and degrades molecules required for special functions, such as intracellular2±5 and extracellular messengers6,7, which allow the functional coordination of processes between unicellular organisms, populations, and cells in multicellular organisms8,9.

The elementary constituents of metabolism are enzymes, i.e. proteins with catalytic properties that interconvert small molecules (i.e. metabolites). Enzymes are crucial for enabling biochemical reactions as they lower the activation energy of such reactions and also realize energetically unfavorable reactions by coupling them to the breakdown of energy-rich bonds of co-substrates, such as ATP10. In every living cell, several hundred to thousands of enzymes exist and form metabolic pathways, altogether constituting what is called metabolism. While was the first pathway to be fully elucidated in the 1930s by Warburg, von Euler-

Chelpin, Emden and Meyerhof1, more and more pathways followed till the mid-1950s and the first attempts to chart this information was done by Donald Nicholson in 195511. Since then and mostly on the basis of genome sequencing and sequence similarities, information on the metabolic network of more than 3000 organisms has been collected, including archaea, bacteria, protists, fungi, plants and animals12±15. In spite of the extreme variety of lifestyles and habitats that exist between those kingdoms of life, the biochemistry that governs metabolism is surprisingly similar for all organisms.

Metabolism can be separated into two phases. In the degradative phase, called , organic nutrient molecules are converted into smaller, simpler end products (such as , or CO2). Catabolic pathways release energy, which is conserved in the formation

10 Chapter 1 ! of ATP and reduced electron carriers (NADH, NADPH)1. In the constructive phase, called or biosynthesis, small, simple precursors are assembled into larger and more complex molecules, including lipids, polysaccharides, proteins, and nucleic acids. Anabolic reactions in turn require an input of energy, generally in the form of ATP, and the reducing power of NADH and NADPH1.

In organotrophic organisms, catabolism and anabolism are coupled via a core metabolic network, constituted by the roughly sixty reactions of central carbon metabolism, which are highly conserved from microbes to human1. Central carbon metabolism consists of the three pathways (i) glycolysis, (ii) pentose phosphate pathway (PPP) and (iii) the tricarboxylic acid

(TCA) cycle and additional interconnecting reactions. In glycolysis, the major carbon source glucose is catabolized into pyruvate with simultaneous net formation of two molecules of ATP.

Glucose-6-phosphate, the second intermediate of glycolysis, is also partially converted via the

PPP to generate NADPH for biosynthetic reductions and pentoses as building blocks for the synthesis of DNA and RNA. In aerobic organisms, most energy can be generated if pyruvate is further converted to acetyl-CoA and fully oxidized in the TCA cycle into carbon dioxide. In this case, NADH is produced, which mainly serves as an electron donor for the respiratory chain. Thereby, one molecule of glucose is catabolized to generate 36 net ATP molecules through combination of glycolysis and the TCA cycle.

Central metabolism not only couples catabolism and anabolism concerning energy and reducing power, but also provides precursors of both processes that can be used to fuel one another. In other words, the high interconnectivity of central metabolism allows to generate most precursors from each of its individual intermediates. Acetyl-CoA produced from carbohydrates is not only oxidized in the TCA cycle, but also forms the initial precursor for fatty acid and lipid synthesis. Alternatively, fatty acids can be degraded via beta-oxidation into acetyl-CoA, which can be used to fuel the TCA cycle or be converted again for lipid synthesis1.

Similarly, central metabolism provides precursors for amino acids in the form of Į-keto acids, which can be interconverted by transamination reactions and thus fuel both catabolism and anabolism. To compensate for the withdrawal of these precursors, mostly from the TCA cycle,

11 Chapter 1 ! their metabolite pools need to be refilled by anaplerotic reactions, e.g. through phosphoenolpyruvate or pyruvate carboxylation to oxaloacetate.

In central metabolism, the cell needs to make critical decisions on how to coordinate the optimal use of available resources with the concomitant generation of energy and the formation of biosynthetic precursors. Therefore, central metabolism is tightly controlled by a system of molecular regulatory mechanisms that sense external (and internal) conditions to mount appropriate cellular responses5,16. This regulation ensures a robust metabolic operation which only requires minimal adjustments if environmental conditions change17. This is a key component of evolutionary success, since it maintains a balance between immediate

(maximum growth) and long-term requirements (survival) of the cell.

Importance of metabolism for medical science and biotechnology

Although being fundamental for life18,19, metabolism was for a long time only seen as WKH³GXOO workhorse process20³ that runs in the background and simply provides the cell with the necessary energy and resources21. However, more recent studies have changed this picture by showing that metabolism is a cellular activity that autonomously controls itself and other cellular processes and therby takes an active role in cellular decision making and determining cellular physiology20,21. For example, metabolism can actively adapt metabolic operation to fluctuations, without requiring any classical sensing and signalling, by metabolites that modulate the activity of regulators and metabolic enzymes2,5,22,23. Furthermore, an increasing number of metabolic enzymes are recognized as so-called moonlightning proteins. These enzymes can exert secondary functions beyond their well-characterized metabolic function that is not of metabolic nature and control other cellular processes, such as gene expression24,

DNA replication, mRNA processing, apoptosis, and transcriptional regulation25.

Metabolic alterations are increasingly recognized as key mediators and drivers in the development of human diseases such as diabetes26,27 and especially cancer28,29. While aerobic glycolysis of cancerous tissues30,31 was one of the first metabolic abnormalities related to disease, more recently an increasing number of additional aberrations have been linked to

12 Chapter 1 ! cancer. Mutations in enzymes have been found to change their activity, resulting in the accumulation of metabolites (e.g. oncometabolites32,33), which promote malignant transformation by deregulating several cellular processes through epigenetics34±36, post- translational modifications37 and cellular signaling33,38. Many tumors appear to have a benefit over healthy tissues by employing distinct metabolic pathways, enzymes and substrates39±45.

Thus, by understanding the metabolic phenotype of diseases, strategies may be applied to design therapies that specifically target the illness without affecting healthy tissues.

From an applied biotechnology perspective, a proper understanding of metabolism also forms the basis for a rational design of production hosts for molecules that are otherwise difficult or expensive to synthesize, such as fine chemicals and drugs46,47. Classical metabolic engineering efforts incorporate knowledge of the connectivity of the metabolic network to potentially enhance product yields by deleting enzymes that compete for precursors and by overexpressing enzymes that reroute flux towards the desired compound48. However, this strategy had only moderate success, since it neglected the regulatory programs of metabolism, which allows the host to maintain a robust metabolic performance. Examples are the compensation of enzyme deletions through the employment of alternative pathways, and the down-regulation of flux through the (overexpressed) production pathway by feedback inhibition by the desired endproduct49,50. The success of more recent studies that investigated such regulation and subsequently incorporated these findings into the metabolic engineering approach51±55 highlights the importance of understanding metabolic operation for biotechnology.

Systems biology of metabolism

To shed light into the complexity of metabolism, methods are required to investigate how metabolic operation evolves from the interplay of various cellular components20,21,56 and methods of Systems Biology raise great expectation. Through genomics, we know the blueprint of an organism which allows the reconstruction of its metabolic network. The genomes of more than 3000 organisms have been analyzed and are publicly available57, and

13 Chapter 1 ! they allow the automated annotation of the growing number of newly sequenced genomes58.

However, the functional annotation has not even been completed for the genomes of well- studied model-organisms, leaving several genes without known function and subsequently incomplete metabolic networks. Additionally, the genetic code is a rather static, condition- independent component of the cell and therefore provides no direct information on how certain metabolic phenotypes are established dynamically.

Therefore, further omics methods try to infer how metabolic operation emerges by measuring the concentrations of particular condition-dependent system variables, such as transcripts59±

61, proteins62±64 and metabolites65±68. However, these static state-dependent variables, albeit extremely informative, only represent a snapshot of a particular cellular state and do not contain the information to fully predict how metabolic operation evolves from the concentrations of the individual components56. Metabolic operation emerges from the complex interplay of all these variables, including cellular regulation of enzyme capacities, the environment and metabolite pools69,70. While gene expression59±61,71, enzyme levels62±64 and even post-transcriptional modifications through several mechanisms72,73 can be detected, we have no model to translate these events into their actual metabolic output. This becomes even more complex with the additional influence of metabolite pools of reaction substrates and products that determine the catalytic activity of enzymes based on kinetic properties56,74.

Furthermore, metabolites that are not directly involved in the catalyzed reaction can also modify their activity by allosteric protein-metabolite interactions and methods for their systematic detection in vivo are still scarce2,23,43.

While the complex interaction of the individual components does not allow to predict how metabolic activity evolves, their integrated network response is given by the in vivo rates of reactions within the metabolic network (i.e. in vivo molecular fluxes or ³WKH IOX[RPH´ 56,75.

Therefore, the experimental assessment of metabolic fluxes is pivotal to identify the factors governing metabolic responses to environmental and genetic changes, and to describe the actual metabolic phenotype.

!

14 Chapter 1 ! Experimental assessment of metabolic fluxes

Different from other omics measurements, metabolic fluxes cannot be measured directly, but have to be inferred by integrating experimental data into computational models of metabolism.

Extracellular fluxes, i.e. substrate uptake and product secretion rates, are determined by regression of metabolite concentration changes in the supernatant against the increase in biomass. However, intracellular fluxes cannot be determined in this way, since the intracellular concentrations are constant at metabolic steady state. Historically, one of the first methods to systematically estimate intracellular fluxes of central metabolism was

(FBA)76,77. In FBA, the metabolic network is represented as stoichiometric a model of chemical reactions, following the principle of conservation of mass. Metabolic fluxes are calculated by constraining these metabolite balances with measured substrate uptake and product secretion rates and potentially adding thermodynamic78,79 or regulatory80±82 restrictions on intracellular fluxes. The method requires a metabolic steady state, with constant uptake and secretion fluxes, as well as steady growth rate and non-varying intracellular metabolite levels.

In principle, FBA is perfectly suited to calculate fluxes even within genome-scale metabolic models76,83. However, the constraints given by extracellular fluxes are insufficient to resolve the intracellular fluxes of the numerous alternative pathways existing in the entire metabolic network, cycling of metabolites in circular pathways, or the exchange of intermediates over bidirectional reactions. When the number of degrees of freedom exceed the measurable entities, the system is said to be underdetermined and, therefore, virtually infinite flux maps exist that fullfill all system and experimental constraints.

To overcome this empasse, an objective function has often been used to select a most-likely solution from the space of feasible ones. Objective functions were justified on the basis of evolutionary arguments, e.g. maximization of growth rate, minimization of ATP surplus, minimization of total absolute fluxes, etc.. While growth and energy use have been shown to be pertinent cellular objectives for flux estimation in some evolved wild-type bacteria84,85, it has been pointed out that estimated fluxes are strongly biased towards the chosen objective function86±88. Furthermore, the optimality principle of metabolism is dubious for mutant strains

15 Chapter 1 ! whose fluxes are objectively suboptimal89 and in cases where rapid adjustment to environmental perturbations is needed17.

Therefore, additional, measureable constraints on in vivo operation of metabolic pathways are required to obtain an unbiased estimate of the actual intracellular flux distribution. Such information can be obtained by using carbon substrates enriched in the stable 13C isotope in metabolic flux studies and by incorporating the resulting labelling patterns of intracellular metabolites into the flux calculation. On the one hand, this analysis exploits the fact that alternative pathways scramble and rearrange metabolite carbon backbones differently before they converge to the same metabolite. This allows to differentiate them based on particularly enriched positions in the common product90±94. On the other hand, the dynamics of 13C propagation, together with metabolite concentration measurements, can be used to quantify particular fluxes in the network and to include this data as constraints95,96. Over the past two decades, an ensemble of such experimental methods has been developed under the general label of 13C metabolic flux analysis (13C MFA).

13C metabolic flux analysis ! All 13C MFA approaches start with a labeling experiment, where cells are cultured in the presence of a 13C-enriched carbon source. For microbial systems, which can be cultivated in minimal medium with a single carbon source, glucose is the most common substrate. Labeled glucose is administered as either a positionally enriched variant to resolve specific pathways90,97,98 or as a uniformly and naturally labeled mixture to investigate pathways that produce characteristic labeling patterns by merging and scrambling of carbon backbones90,99.

During the labeling experiment, cells must exhibit a metabolic (pseudo-) steady state (Fig. 1), which is achieved in mid-exponential phase of batch or continuous cultivations. The minimum duration of the labeling experiment strongly depends on (i) the methodology applied to calculate fluxes (stationary56,92,93 or non-stationary100 13C MFA) (Fig. 1) and (ii) the analytes used to extract the labeling information, which can be cellular constituents, such as protein- bound or free amino acids or free metabolic intermediates. In the most conventional

16 Chapter 1 ! framework, i.e. stationary 13C MFA, the labeling patterns are evaluated at the end-point when they become invariant over time, i.e. at isotopic steady state (Fig. 1). Conventionally, labeling patterns of protein-bound amino acids are then acquired by gas chromatography-mass spectrometry (GC-MS)90,99 or nuclear magnetic resonance (NMR) spectroscopy101,102, which represent eight key intermediates of central carbon metabolism102,103 and are obtained in large quantities from biomass pellets through hydrolysis. Steady state labeling information of free amino acids104,105 and primary intermediates measured with liquid chromatography-mass spectrometry (LC-MS)65,106±108 was also already employed for 13C MFA, which results in shorter durations to attain isotopic stationarity, but requires more sensitive instrumentation due to lower abundances109,110. Alternatively, non-stationary 13C MFA100 uses the initial transients of

13C propagation before isotopic equilibrium is reached (Fig. 1). Here, multiple samples are taken over time after addition of the 13C tracer. Since the labeling dynamics depend both on metabolic fluxes and on intracellular metabolite pools, the latter have to be additionally measured and included in the flux estimation100,109.

Figure 1!Metabolic and isotopic dynamics in stationary and non-stationary 13C metabolic flux analysis. Non-stationary

13C MFA requires multiple measurements during the transient phase of isotope enrichment to extract the labeling information, while stationary 13C MFA uses a single measurement during isotopic steady state. For canonical non- stationary 13C MFA, but not for stationary 13C MFA, steady state metabolite concentrations have to be measured additionally.

! 17 Chapter 1 ! To quantify fluxes from 13C data, a set of complementary approaches with different characteristics in terms of spatial and temporal resolution exist90,92,95,99,111±114 (Fig. 2). From isotopic stationary data, local ratios of relative pathway fluxes can be estimated by local interpretation of isotope patterns of a few metabolites at specific converging metabolic nodes90,111,113. Kinetic flux profiling95,96 also represents a local analysis tool, which uses non- stationary 13C data and metabolite concentrations to quantify individual, irreversible fluxes based on a small ordinary differential equation (ODE) system. Both approaches yield local flux constraints, which do not require the measurement of extracellular rates. However, if rates are available, these local flux constraints can be used to restrict the degrees of freedom of stoichiometric metabolic models and to calculate network-wide fluxes.

Alternatively, global isotopomer balancing and iterative fitting can be used to calculate network- wide absolute intracellular fluxes92,93,115. Here, the network is represented as a system of balance equations that describe the propagation of carbon atom transitions for each metabolic reaction116±119. Conceptually, the approach builds on balancing of metabolites such as in FBA, but includes the information of all possible 13C states of metabolites. This model, together with measured extracellular rates and knowledge about the biomass composition for the determination of growth rate-dependent biosynthetic fluxes, is then used to calculate in vivo fluxes in an iterative fitting procedure, where in silico 13C labeling patterns are simulated and compared to measured data until a satisfactory match is achieved. While this principle applies both for stationary and non-stationary 13C MFA, the methods differ markedly in the complexity of their equation systems. More specifically, with the isotopic stationary method, time- dependent terms and metabolite pool sizes can be cancelled from the initial ODE system to give a system of algebraic equations120, while both items remain with the dynamic labeling data of non-stationary 13C MFA. This brings about the requirement of metabolite concentration measurements and additionally affects the parameter estimation part, which is computationally significantly more demanding for global isotopomer balancing of non-stationary 13C data. The potentials and shortcomings of each approach are summarized in Figure 3.

18 Chapter 1 !

Figure 2 Overview of approaches to quantify fluxes from 13C labeling data. Flux ratio estimation with stationary labeling data uses the fact that converging metabolic routes lead to pathways-specific labeling patterns in the common product. Global isotopomer balancing incorporates labeling patterns of all metabolites within the network and tries to map a global flux solution that explains the measurements best by iteratively fitting simulated to measured data. Global isotopomer balancing can be applied with stationary and non-stationary labeling data, but also requires measured extracellular rates and, with the non-stationary approach, also metabolite concentrations. Kinetic flux profiling estimates individual fluxes of irreversible reactions from non-stationary labeling data and metabolite concentrations of the reaction product with small ODE models. Non-stationary 13C metabolic flux ratio analysis is the approach developed within this thesis. It allows the quantification of local, relative fluxes with dynamic labeling data and small ODE models, without the requirement of metabolite concentrations.

19 Chapter 1 !

!

Figure 3 Comparison of 13C flux approaches in respect of their potentials and shortcomings. Compared are (i) the requirement of measured extracellular rates, (ii) the amount of labeling patterns that need to be measured, (iii) the complexity of the computational parameter estimation process, (iv) the requirement of concentration measurements, (v) the globality in respect to the amount of fluxes that can be simultaneously estimated, (vi) the experimental duration required for the labeling experiment and (vii) the general applicability in respect to the location within the metabolic network. A value of 1 indicates rather high demands or poor performance of the approach, while a value of 3 indicates little requirements or good performance.

Achievements and shortcomings of current 13C MFA approaches ! Since its original development roughly two decades ago, 13C MFA has found numerous applications in several research areas, including metabolic engineering121±124, systems biology20,125 and biomedical research126. 13C MFA has contributed to the elucidation of metabolic networks of microbial model-organisms and less-characterized species, allowed the recognition of unexpected operation of known metabolic routes127±132, the identification of entirely novel pathways133,134 and also to distinguish the usage of specific pathways among several alternatives135,136. Large scale flux studies have demonstrated an unexpected flexibility and robustness of central microbial metabolism to genetic and environmental alterations17,91,137 and allowed to identify regulatory mechanisms at the transcriptional16,138 and posttranslational139 level. Furthermore, 13C MFA has been used to study cellular energetics and redox balancing in bacteria140,141.

20 Chapter 1 ! Additional improvements of 13C MFA in recent years involved mainly analytical developments, together with an increasing number of manually and automatically curated, innovative labeling strategies142±144. This enabled to quantitatively estimate certain fluxes with some approximations in higher cells. These studies provided unprecedented insights into basic functionality of metabolism in plants125,145, animals146 and human142,147, revealed metabolic alterations related to diseases39,42,44,148±150, and allowed the identification of potential drug targets112. Nevertheless, a universal application of 13C MFA to all cellular systems is still far from being practical because of several challenges110.

First, eukaryotes are compartmentalized. Intracellular compartments such as mitochondria, cytoplasm, or vacuoles hamper the interpretation of labeling patterns due to the parallel existence of several pathways and metabolite pools within one cell151. Since metabolites cannot be individually extracted with current protocols, the only way for discrimination is the existence of compartment-specific reactions that lead to distinct 13C signatures152,153, albeit this is unrealistic for most pathways. Therefore, flux solutions with good confidence are usually only possible by reducing the degrees of freedom in the compartmented models by pooling of compartmentalized reactions and metabolites154 and by assumptions on the activity of certain reactions155. Although it might be possible to partially validate the latter assumptions by feeding specific tracers134, this drastically increases the number of labeling experiments, which have to be performed for every cell-type and condition. An alternative strategy is to incorporate non- stationary 13C MFA, which has the potential to estimate subcellular concentrations from dynamic labeling data. However, non-stationary 13C MFA is still hampered by the fact that publicly available software tools to rigorously calculate fluxes from dynamic 13C data are scarce110, with the recently developed software package INCA as the sole exception156.

Second, current 13C MFA approaches do not really enable flux estimation in real-life environments, since they are mostly restricted to minimal media with one or two carbon sources. This is mainly due to the fact that the uptake of each additional carbon source influences the labeling patterns of intracellular metabolites and thereby increases the degrees of freedom in the network. If possible at all, this can only be compensated by measuring

21 Chapter 1 ! labeling patterns of intracellular metabolites from multiple experiments, with each individual carbon source being replaced by its 13C labeled analog. Additionally, the uptake rate of all carbon sources has to be determined to close the carbon balance. Meeting both requirements is far from practical for mammalian cell culture, because media contain a plethora of amino acids, nucleotides and lipids in addition to glucose. Furthermore, the complex composition of the media increases the risk of sequential carbon usage and consequent non-stationary metabolism.

Third, the aforementioned potential transient growth under-real-life conditions, together with the slow doubling times of mammalian cells, complicates the requirement of metabolic steady state until isotopic steady state is reached. This imposes a general limitation to the applicability of stationary 13C MFA, also for the study of short transient biological changes and non-growing cells. Strategies to solve this problem, are to extract the labeling information from metabolic intermediates that reach isotopic steady state faster than proteinogenic amino acids, such as free amino acids104,105,157 and primary intermediates106,107,158. Moreover, non-stationary

13C MFA additionally reduces the duration of the labeling experiment from several days or hours to the range of few hours, minutes or even seconds, respectively100,110.

As already depicted, non-stationary 13C MFA generally has the potential to overcome most of the aforementioned challenges. Additionally non-stationary 13C MFA allows the expansion of flux analysis beyond central metabolism. In the peripheral (anabolic) network, pathways are mostly branched, linear or large metabolic cycles, so that fluxes can only be inferred from dynamic labeling data, since at isotopic steady state the 13C patterns are independent of the flux110. Noteworthy, another feature of non-stationary 13C MFA is the possibility to estimate

159,160 fluxes in organisms that metabolize monocarbon compounds, e.g. CO2 in plants and other autotrophic organisms161 or methanol in yeast162 and methylotrophs163. Nevertheless, canonical, global non-stationary 13C MFA is still hampered by the requirement of (i) a closed carbon balance, (ii) knowledge on the activity of reactions in the considered network, (iii) measurements of metabolite concentrations and (iv) access to high performance computers

22 Chapter 1 ! to solve the large ODE systems. Although kinetic flux profiling of local fluxes partially circumvents these issues, it still has to meet the aforementioned requirements (ii) and (iii).

To enable a quantitative, in vivo analysis of the functional intracellular metabolism under real- life conditions, a generally applicable approach would be valuable that allows the estimation of, at least individual, specific fluxes in metabolic networks. Ideally such a framework would (i) allow flux estimation from very short time-scale labeling experiments, which would offer the potential to study dynamic systems where metabolic steady states are brief and (ii) be rigorously applicable to any part of the network even if (iii) multiple substrates are co- metabolized. In this PhD thesis we developed an approach that allows the estimation of local fluxes from non-stationary 13C labeling data, paving the road to targeted 13C MFA during transitional processes and in complex media.

!

Thesis outline ! The main contribution of this PhD thesis was the development and evaluation of non-stationary

13C metabolic flux ratio analysis (non-stationary 13C MFRA) ± a novel method for the quantification of relative fluxes from dynamic 13C labeling data in bacteria (Chapter 2) and higher cells (Chapter 3).

In Chapter 2, we present the generalized workflow for non-stationary 13C MFRA for data processing, modeling, parameter and error estimation, which takes into account all metabolic reaction types and the availability of mass spectrometric data on molecular ions and fragments.

As a proof-of-principle, we demonstrated the approach by analyzing fluxes at key metabolic nodes in central metabolism of B. subtilis and obtained estimates that were in good agreement with stationary 13C MFA results, but from drastically shorter labeling experiments.

In Chapter 3, we evaluated to which extent the established procedures of non-stationary 13C

MFRA could be applied to calculate fluxes in the more complex metabolic networks of higher cells. Therefore, we first estimated relative flux changes in mouse embryonic fibroblasts with impaired mitochondrial pyruvate transport, using existing stationary methods and compared

23 Chapter 1 ! these estimates to values obtained with our framework. This analysis revealed that non- stationary 13C MFRA was also able to detect flux changes in higher cell metabolism, which reduced both the experimental duration and the number of required 13C tracers.

The analysis with B. subtilis in Chapter 2 indicated the presence of reversible malic enzyme flux, which was previously hypothesized to constitute a transhydrogenation mechanism for

NADPH balancing. We investigated this possibility in more detail in Chapter 4. Here, we integrated our non-stationary flux constraints with stationary flux data for a global flux analysis and NADPH balancing with B. subtilis malic enzyme mutants, which revealed a distinct role of isoform YtsJ in NADPH balancing. We further investigated the potential mechanism by in vitro assays, and found that the enzyme indeed oxidized excess NADPH. Surprisingly, we detected a concomitant switch in enzyme function towards redox neutral lactate formation, which would represent a novel mechanism to allow the oxidation of excess NADPH.

Additional projects

These projects were also part of this work, but are not included in this thesis.

HIF-driven SF3B1 induces KHK-C to enforce and heart disease

Peter Mirtschink, Jaya Krishnan, Fiona Grimm, Alexandre Sarre, Manuel Hörl, Melis Kayikci,

Niklaus Fankhauser, Yann Christinat, Cédric Cortijo, Owen Feehan, Ana Vukolic, Samuel

Sossalla, Sebastian N. Stehr, Jernej Ule, Nicola Zamboni, Thierry Pedrazzini and Wilhelm

Krek.

Nature 522, 444-9 (2015).

Own contribution: design of 13C labeling experiments, metabolomics and labeling measurements and data analysis.

Dynamic modeling of E. coli in a complex medium reveals coordination of amino acids and glucose catabolism.

Mattia Zampieri*, Manuel Hörl*, Florian Hotz, Nicola Müller and Uwe Sauer.

*equally contributed to this work

24 Chapter 1 ! Manuscript in preparation

Own contribution: physiological characterization, quantitative metabolomics experiments, enzymatic assays and data analysis. Contributed to designing the study and writing the manuscript.

References

1. Nelson, D., Lehninger, A. & Cox, M. Lehninger principles of biochemistry. (W. H. Freeman; 5th edition, 2008).

2. Doucette, C. D., Schwab, D. J., Wingreen, N. S. & Rabinowitz, J. D. Į-Ketoglutarate coordinates carbon and nitrogen utilization via enzyme I inhibition. Nat. Chem. Biol. 7, 894±901 (2011).

3. Chubukov, V., Gerosa, L., Kochanowski, K. & Sauer, U. Coordination of microbial metabolism. Nat. Rev. Microbiol. 12, 327±40 (2014).

4. You, C. et al. Coordination of bacterial proteome with metabolism by cyclic AMP signalling. Nature 500, 301±6 (2013).

5. Kotte, O., Zaugg, J. B. & Heinemann, M. Bacterial adaptation through distributed sensing of metabolic fluxes. Mol. Syst. Biol. 6, 355 (2010).

6. Randle, P. J. Endocrine control of metabolism. Annu Rev Physiol. 25, 291±324 (1963).

7. Saltiel, A. R. & Kahn, C. R. Insulin signalling and the regulation of glucose and . 414, 799±806 (2001).

8. Waters, C. M. & Bassler, B. L. Quorum sensing: cell-to-cell communication in bacteria. Annu. Rev. Cell Dev. Biol. 21, 319±346 (2005).

9. Gerdes, H.-H. & Pepperkok, R. Cell-to-cell communication: current views and future perspectives. Cell Tissue Res. 352, 1±3 (2013).

10. Harold, F. M. The Vital Force: A Study of Bioenergetics. (Freeman and Company, New York, NY., 1986).

11. Dagley, S. & Nicholson, D. E. An introduction to Metabolic Pathways. (Blackwell Scientific Publications, 1970).

12. Caspi, R. et al. The MetaCyc database of metabolic pathways and enzymes and the BioCyc collection of Pathway/Genome Databases. Nucleic Acids Res. 42, d459±71 (2014).

13. .DQHKLVD0 *RWR6.(**ௗ.\RWR(QF\FORSHGLDRI*HQHVDQG*HQRPHVNucleic Acids Res. 28, 27±30 (2000).

25 Chapter 1 ! 14. Kanehisa, M. et al. Data, information, knowledge and principle: back to metabolism in KEGG. Nucleic Acids Res. 42, D199±205 (2014).

15. Michal, G. & Schomburg, D. Biochemical Pathways: An Atlas of Biochemistry and Molecular Biology, 2nd edition. (2012).

16. Haverkorn van Rijsewijk, B. R. B., Nanchen, A., Nallet, S., Kleijn, R. J. & Sauer, U. Large-scale 13C-flux analysis reveals distinct transcriptional control of respiratory and fermentative metabolism in . Mol. Syst. Biol. 7, 477 (2011).

17. Schuetz, R., Zamboni, N., Zampieri, M., Heinemann, M. & Sauer, U. Multidimensional Optimality of Microbial Metabolism. Science. 336, 601±604 (2012).

18. Koshland, D. The Seven Pillars of Life. Science. 64±65 (2002).

19. McKay, C. P. What is life - and how do we search for it in other worlds? PLoS Biol. 2, E302 (2004).

20. Heinemann, M. & Sauer, U. Systems biology of microbial metabolism. Curr. Opin. Microbiol. 13, 337±43 (2010).

21. Ray, L. Metabolism is not boring. Science. 330, 2010 (2010).

22. Grüning, N.-M., Lehrach, H. & Ralser, M. Regulatory crosstalk of the metabolic network. Trends Biochem. Sci. 35, 220±7 (2010).

23. Link, H., Kochanowski, K. & Sauer, U. Systematic identification of allosteric protein- metabolite interactions that control enzyme activity in vivo. Nat. Biotechnol. 31, 357±61 (2013).

24. Commichau, F. M. & Stülke, J. Trigger enzymes: bifunctional proteins active in metabolism and in controlling gene expression. Mol. Microbiol. 67, 692±702 (2008).

25. Commichau, F. M. et al. Novel activities of glycolytic enzymes in Bacillus subtilis: interactions with essential proteins involved in mRNA processing. Mol. Cell. Proteomics 8, 1350±60 (2009).

26. Fisher-Wellman, K. H. & Neufer, P. D. Linking mitochondrial bioenergetics to insulin resistance via redox biology. Trends Endocrinol. Metab. 23, 142±53 (2012).

27. Kwak, S. H., Park, K. S., Lee, K.-U. & Lee, H. K. Mitochondrial metabolism and diabetes. J. Diabetes Investig. 1, 161±9 (2010).

28. Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646± 74 (2011).

29. Ward, P. S. & Thompson, C. B. Metabolic reprogramming: a cancer hallmark even warburg did not anticipate. Cancer Cell 21, 297±308 (2012).

30. Warburg, O. On the origin of cancer cells. Science. 123, 309±314 (1956).

31. Warburg, O. Über den Stoffwechsel der Carcinomzelle. Naturwissenschaften 309±344 (1924).

26 Chapter 1 ! 32. Dang, L. et al. Cancer-associated IDH1 mutations produce 2-hydroxyglutarate. Nature 462, 739±44 (2009).

33. Selak, M. a et al. Succinate links TCA cycle dysfunction to oncogenesis by inhibiting HIF-alpha prolyl hydroxylase. Cancer Cell 7, 77±85 (2005).

34. Turcan, S. et al. IDH1 mutation is sufficient to establish the glioma hypermethylator phenotype. Nature 483, 479±83 (2012).

35. Killian, J. K. et al. Succinate dehydrogenase mutation underlies global epigenomic divergence in gastrointestinal stromal tumor. Cancer Discov. 3, 648±57 (2013).

36. Letouzé, E. et al. SDH mutations establish a hypermethylator phenotype in paraganglioma. Cancer Cell 23, 739±52 (2013).

37. Blatnik, M., Frizzell, N., Thorpe, S. & Baynes, J. Inactivation of glyceraldehyde-3- phosphate dehydrogenase by fumarate in diabetes formation of S-(2-Succinyl) cysteine, a novel chemical modification of protein and possible biomarker of mitochondrial stress. Diabetes 57, 41±49 (2008).

38. Sullivan, L. B. et al. The proto-oncometabolite fumarate binds glutathione to amplify ROS-dependent signaling. Mol. Cell 51, 236±48 (2013).

39. Possemato, R. et al. Functional genomics reveal that the serine synthesis pathway is essential in breast cancer. Nature 476, 346±50 (2011).

40. Zhang, W. C. et al. Glycine decarboxylase activity drives non-small cell lung cancer tumor-initiating cells and tumorigenesis. Cell 148, 259±72 (2012).

41. Ren, J.-G. et al. Knockdown of malic enzyme 2 suppresses lung tumor growth, induces differentiation and impacts PI3K/AKT signaling. Sci. Rep. 4, 5414 (2014).

42. Ye, J. et al. Pyruvate kinase M2 promotes de novo serine synthesis to sustain mTORC1 activity and cell proliferation. Proc. Natl. Acad. Sci. U S A. 109, 6904±6909 (2012).

43. Chaneton, B. et al. Serine is a natural ligand and allosteric activator of pyruvate kinase M2. Nature 491, 458±62 (2012).

44. DeBerardinis, R. J. et al. Beyond aerobic glycolysis: transformed cells can engage in glutamine metabolism that exceeds the requirement for protein and nucleotide synthesis. Proc. Natl. Acad. Sci. U. S. A. 104, 19345±50 (2007).

45. Jain, M. et al. Metabolite profiling identifies a key role for glycine in rapid cancer cell proliferation. Science 336, 1040±4 (2012).

46. Matsuoka, Y. & Shimizu, K. Importance of understanding the main metabolic regulation in response to the specific pathway mutation for metabolic engineering of Escherichia coli. Comput. Struct. Biotechnol. J. 3, (2012).

47. Keasling, J. D. Manufacturing molecules through metabolic engineering. Science 330, 1355±8 (2010).

27 Chapter 1 ! 48. Burgard, A. P., Pharkya, P. & Maranas, C. D. OptKnock: A Bilevel Programming Framework for Identifying Gene Knockout Strategies for Microbial Strain Optimization. Biotechnol. Bioeng. 84, 647±657 (2003).

49. Park, J. H., Lee, K. H., Kim, T. Y. & Lee, S. Y. Metabolic engineering of Escherichia coli for the production of L-valine based on transcriptome analysis and in silico gene knockout simulation. 104, 7797±7802 (2007).

50. Lee, K. H., Park, J. H., Kim, T. Y., Kim, H. U. & Lee, S. Y. Systems metabolic engineering of Escherichia coli for L-threonine production. Mol. Syst. Biol. 3, 149 (2007).

51. Xu, P., Li, L., Zhang, F., Stephanopoulos, G. & Koffas, M. Improving fatty acids production by engineering dynamic pathway regulation and metabolic control. Proc. Natl. Acad. Sci. U. S. A. 111, 1±6 (2014).

52. Vemuri, G. N. & Aristidou, A. a. Metabolic Engineering in the -RPLFV(UDௗ(OXFLGDWLQJ and Modulating Regulatory Networks. Microbiol. Mol. Biol. Rev. 69, 197±216 (2005).

53. Na, D. et al. Metabolic engineering of Escherichia coli using synthetic small regulatory RNAs. Nat. Biotechnol. 31, 170±4 (2013).

54. Dahl, R. H. et al. Engineering dynamic pathway regulation using stress-response promoters. Nat. Biotechnol. 31, 1039±1046 (2013).

55. Zhang, F., Carothers, J. M. & Keasling, J. D. Design of a dynamic sensor-regulator system for production of chemicals and fuels derived from fatty acids. Nat. Biotechnol. 30, 354±9 (2012).

56. Sauer, U. Metabolic networks in motion: 13C-based flux analysis. Mol. Syst. Biol. 2, 62 (2006).

57. Reddy, T. B. K. et al. The Genomes OnLine Database (GOLD) v.5: a metadata management system based on a four level (meta)genome project classification. Nucleic Acids Res. 43, D1099±D1106 (2014).

58. Moriya, Y., Itoh, M., Okuda, S., Yoshizawa, A. C. & Kanehisa, M. KAAS: An automatic genome annotation and pathway reconstruction server. Nucleic Acids Res. 35, 182±185 (2007).

59. Donaldson, K. J., Kricka, L. J., Surrey, S. S. & Fortina, P. Multispot Array Technologies. 1±11. In eLS (John Wiley & Sons Ltd, Chichester) (2009).

60. Nolan, T., Hands, R. E. & Bustin, S. a. Quantification of mRNA using real-time RT-PCR. Nat. Protoc. 1, 1559±1582 (2006).

61. Nagalakshmi, U. et al. The transcriptional landscape of the genome defined by RNA sequencing. Science 320, 1344±1349 (2008).

62. Gstaiger, M. & Aebersold, R. Applying mass spectrometry-based proteomics to genetics, genomics and network biology. Nat. Rev. Genet. 10, 617±27 (2009).

63. Gillet, L. C. et al. Targeted Data Extraction of the MS/MS Spectra Generated by Data- independent Acquisition: A New Concept for Consistent and Accurate Proteome Analysis. Mol. Cell. Proteomics 11, (2012).

28 Chapter 1 ! 64. Aebersold, R. & Mann, M. Mass spectrometry-based proteomics. Nature 422, 198±207 (2003).

65. Buescher, J. M., Moco, S., Sauer, U. & Zamboni, N. Ultrahigh Performance Liquid Chromatography - Tandem Mass Spectrometry Method for Fast and Robust Quantification of Anionic and Aromatic Metabolites. Anal. Chem. 82, 4403±4412 (2010).

66. Fuhrer, T., Heer, D., Begemann, B. & Zamboni, N. High-throughput, accurate mass metabolome profiling of cellular extracts by flow injection-time-of-flight mass spectrometry. Anal. Chem. 83, 7074±80 (2011).

67. Beckonert, O. et al. Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts. Nat. Protoc. 2, 2692± 2703 (2007).

68. Wishart, D. S. Quantitative metabolomics using NMR. Trends Anal. Chem. 27, 228±237 (2008).

69. Gerosa, L. & Sauer, U. Regulation and control of metabolic fluxes in microbes. Curr. Opin. Biotechnol. 22, 566±75 (2011).

70. .LWDQR+6\VWHPV%LRORJ\ௗ$%ULHI2YHUYLHZScience. 295, 1662±1664 (2002).

71. Zaslaver, A., Bren, A., Ronen, M. & Itzkovitz, S. A comprehensive library of fluorescent transcriptional reporters for Escherichia coli. Nat. Methods 3, 623±628 (2006).

72. Oliveira, A. P. & Sauer, U. The importance of post-translational modifications in regulating Saccharomyces cerevisiae metabolism. FEMS Yeast Res. 12, 104±117 (2012).

73. Wang, Q. et al. Acetylation of metabolic enzymes coordinates carbon source utilization and metabolic flux. Science 327, 1004±7 (2010).

74. Gerosa, L. & Sauer, U. Regulation and control of metabolic fluxes in microbes. Curr. Opin. Biotechnol. 22, 566±75 (2011).

75. Sauer, U. High-throughput phenomics: experimental methods for mapping fluxomes. Curr. Opin. Biotechnol. 15, 58±63 (2004).

76. Orth, J. D., Thiele, I. & Palsson, B. Ø. What is flux balance analysis? Nat. Biotechnol. 28, 245±8 (2010).

77. Majewski R A, D. Simple constrained optimization view of overflow in E. coli. Biotechnol bioeng 35, 731±738 (1990).

78. Zamboni, N., Kümmel, A. & Heinemann, M. anNET: a tool for network-embedded thermodynamic analysis of quantitative metabolome data. BMC Bioinformatics 9, 199 (2008).

79. Beard, D. A., Babson, E., Curtis, E. & Qian, H. Thermodynamic constraints for biochemical networks. J. Theor. Biol. 228, 327±33 (2004).

29 Chapter 1 ! 80. Covert, M. W. & Palsson, B. Ø. Constraints-based models: Regulation of Gene Expression Reduces the Steady-state Solution Space. J. Theor. Biol. 221, 309±325 (2003).

81. Shlomi, T., Cabili, M. N., Herrgård, M. J., Palsson, B. Ø. & Ruppin, E. Network-based prediction of human tissue-specific metabolism. Nat. Biotechnol. 26, 1003±10 (2008).

82. Becker, S. a & Palsson, B. Ø. Context-specific metabolic networks are consistent with experiments. PLoS Comput. Biol. 4, e1000082 (2008).

83. Bordbar, A., Monk, J. M., King, Z. A. & Palsson, B. Ø. Constraint-based models predict metabolic and associated cellular functions. Nat. Rev. Genet. 15, 107±20 (2014).

84. Ibarra, R. U., Edwards, J. S. & Palsson, B. Ø. Escherichia coli K-12 undergoes adaptive evolution to achieve in silico predicted optimal growth. 420, 20±23 (2002).

85. Carlson, R. & Srienc, F. Fundamental Escherichia coli biochemical pathways for biomass and energy production: identification of reactions. Biotechnol. Bioeng. 85, 1± 19 (2004).

86. Schuetz, R., Kuepfer, L. & Sauer, U. Systematic evaluation of objective functions for predicting intracellular fluxes in Escherichia coli. Mol. Syst. Biol. 3, 119 (2007).

87. Molenaar, D., van Berlo, R., de Ridder, D. & Teusink, B. Shifts in growth strategies reflect tradeoffs in cellular economics. Mol. Syst. Biol. 5, 323 (2009).

88. Harcombe, W. R., Delaney, N. F., Leiby, N., Klitgord, N. & Marx, C. J. The ability of flux balance analysis to predict evolution of central metabolism scales with the initial distance to the optimum. PLoS Comput. Biol. 9, e1003091 (2013).

89. Segrè, D., Vitkup, D. & Church, G. M. Analysis of optimality in natural and perturbed metabolic networks. Proc. Natl. Acad. Sci. U. S. A. 99, 15112-7 (2002).

90. Fischer, E. & Sauer, U. Metabolic flux profiling of Escherichia coli mutants in central carbon metabolism using GC-MS. Eur. J. Biochem. 270, 880±891 (2003).

91. Fischer, E. & Sauer, U. Large-scale in vivo flux analysis shows rigidity and suboptimal performance of Bacillus subtilis metabolism. Nat. Genet. 37, 636±40 (2005).

92. Wiechert, W. 13C metabolic flux analysis. Metab. Eng. 3, 195±206 (2001).

93. Stephanopoulos, G. Metabolic fluxes and metabolic engineering. Metab. Eng. 1, 1±11 (1999).

94. Sauer, U. et al. Metabolic flux ratio analysis of genetic and environmental modulations of Escherichia coli central carbon metabolism. J. Bacteriol. 181, 6679±88 (1999).

95. Yuan, J., Bennett, B. D. & Rabinowitz, J. D. Kinetic flux profiling for quantitation of cellular metabolic fluxes. Nat. Protoc. 3, 1328±40 (2008).

96. Yuan, J., Fowler, W. U., Kimball, E., Lu, W. & Rabinowitz, J. D. Kinetic flux profiling of nitrogen assimilation in Escherichia coli. Nat. Chem. Biol. 2, 529±30 (2006).

30 Chapter 1 ! 97. Lee, W. P. et al. Mass isotopomer study of the nonoxidative pathways of the pentose 13 cycle with [1,2- C2] glucose. Am. J. Physiol. 274, 843±851 (1998).

98. Holleran, A. Quantitative analysis of acetoacetate metabolism in AS-30D hepatoma cells with 13C and 14C isotopic techniques. Am. J. Physiol. E945±951 (1997).

99. Zamboni, N., Fendt, S.-M., Rühl, M. & Sauer, U. 13C-based metabolic flux analysis. Nat. Protoc. 4, 878±92 (2009).

100. Nöh, K. et al. Metabolic flux analysis at ultra short time scale: isotopically non-stationary 13C labeling experiments. J. Biotechnol. 129, 249±67 (2007).

101. Szyperski, T. et al. Bioreaction network topology and metabolic flux ratio analysis by biosynthetic fractional 13C labeling and two-dimensional NMR spectroscopy. Metab. Eng. 1, 189±97 (1999).

102. Szyperski, T. Biosynthetically directed fractional 13C-labeling of proteinogenic amino acids. An efficient analytical tool to investigate intermediary metabolism. Eur. J. Biochem. 232, 433±48 (1995).

103. Nanchen, A., Fuhrer, T. & Sauer, U. Determination of Metabolic Flux Ratios From 13C- Experiments and Gas Chromatography±Mass Spectrometry Data. Methods Mol. Biol. 358, 177±197 (2007).

104. Rühl, M., Zamboni, N. & Sauer, U. Dynamic flux responses in riboflavin overproducing Bacillus subtilis to increasing glucose limitation in fed-batch culture. Biotechnol. Bioeng. 105, 795±804 (2010).

105. Drysch, a, El Massaoudi, M., Wiechert, W., de Graaf, a a & Takors, R. Serial flux mapping of Corynebacterium glutamicum during fed-batch L-lysine production using the sensor reactor approach. Biotechnol. Bioeng. 85, 497±505 (2004).

106. Van Winden, W. a et al. Metabolic-flux analysis of Saccharomyces cerevisiae CEN.PK113-7D based on mass isotopomer measurements of 13C-labeled primary metabolites. FEMS Yeast Res. 5, 559±68 (2005).

107. Rühl, M. et al. Collisional fragmentation of central carbon metabolites in LC-MS/MS increases precision of 13C metabolic flux analysis. Biotechnol. Bioeng. 109, 763±71 (2012).

108. Rühl, M., Le Coq, D., Aymerich, S. & Sauer, U. 13C-flux analysis reveals NADPH- balancing transhydrogenation cycles in stationary phase of nitrogen-starving Bacillus subtilis. J. Biol. Chem. 287, 27959±70 (2012).

109. Nöh, K. & Wiechert, W. The benefits of being transient: isotope-based metabolic flux analysis at the short time scale. Appl. Microbiol. Biotechnol. 91, 1247±65 (2011).

110. Zamboni, N. 13C metabolic flux analysis in complex systems. Curr. Opin. Biotechnol. 22, 103±8 (2011).

111. Zamboni, N., Fischer, E. & Sauer, U. FiatFlux - a software for metabolic flux analysis from 13C-glucose experiments. BMC Bioinformatics 6, 209 (2005).

31 Chapter 1 ! 112. Munger, J. et al. Systems-level metabolic flux profiling identifies as a target for antiviral therapy. Nat. Biotechnol. 26, 1179±86 (2008).

113. Rantanen, A. et al. An analytic and systematic framework for estimating metabolic flux ratios from 13C tracer experiments. BMC Bioinformatics 9, 266 (2008).

114. Fischer, E., Zamboni, N. & Sauer, U. High-throughput metabolic flux analysis based on gas chromatography±mass spectrometry derived 13C constraints. Anal. Biochem. 325, 308±316 (2004).

115. Weitzel, M. et al. 13CFLUX2 - high-performance software suite for 13C-metabolic flux analysis. Bioinformatics 29, 143±5 (2013).

116. Antoniewicz, M. R., Kelleher, J. K. & Stephanopoulos, G. Elementary metabolite units (EMU): a novel framework for modeling isotopic distributions. Metab. Eng. 9, 68±86 (2007).

117. Van Winden, W. a, Heijnen, J. J. & Verheijen, P. J. T. Cumulative bondomers: a new concept in flux analysis from 2D [13C,1H] COSY NMR data. Biotechnol. Bioeng. 80, 731± 45 (2002).

118. Zupke, C. & Stephanopoulos, G. Modeling of Isotope Distributions and Intracellular Fluxes in Metabolic Networks Using Atom Mapping Matrices. Biotechnol. Prog. 10, 489± 498 (1994).

119. Wiechert, W., Mo, M., Isermann, N., Wurzel, M. & de Graaf, A. A. Bidirectional Reaction 6WHSVLQ0HWDEROLF1HWZRUNVௗ,,,([SOLFLW6ROXWLRQDQG$QDO\VLVRI,VRWRSRPHU/DEHOLQJ Systems. Biotechnol. Bioeng. 66, 69-85 (1999).

120. Noack, S., Nöh, K., Moch, M., Oldiges, M. & Wiechert, W. Stationary versus non- stationary 13C-MFA: a comparison using a consistent dataset. J. Biotechnol. 154, 179± 90 (2011).

121. Lau, W., Fischbach, M. a., Osbourn, A. & Sattely, E. S. Key Applications of Plant Metabolic Engineering. PLoS Biol. 12, 1±5 (2014).

122. Kim, H. U., Kim, T. Y. & Lee, S. Y. Metabolic flux analysis and metabolic engineering of microorganisms. Mol. Biosyst. 4, 113±120 (2008).

123. Boghigian, B. a., Seth, G., Kiss, R. & Pfeifer, B. a. Metabolic flux analysis and pharmaceutical production. Metab. Eng. 12, 81±95 (2010).

124. Otero, J. M. & Nielsen, J. Industrial systems biology. Biotechnol. Bioeng. 105, 439±60 (2010).

125. Kruger, N. J. & Ratcliffe, R. G. Insights into plant metabolic networks from steady-state metabolic flux analysis. Biochimie 91, 697±702 (2009).

126. Hiller, K. & Metallo, C. M. Profiling metabolic networks to study cancer metabolism. Curr. Opin. Biotechnol. 24, 60±8 (2013).

127. Amador-Noguez, D. et al. Systems-level metabolic flux profiling elucidates a complete, bifurcated tricarboxylic acid cycle in Clostridium acetobutylicum. J. Bacteriol. 192, 4452± 61 (2010).

32 Chapter 1 ! 128. Fischer, E. & Sauer, U. A novel metabolic cycle catalyzes glucose oxidation and anaplerosis in hungry Escherichia coli. J. Biol. Chem. 278, 46446±51 (2003).

129. Olszewski, K. L. et al. Branched tricarboxylic acid metabolism in Plasmodium falciparum. Nature 466, 774±778 (2010).

130. Fuhrer, T., Fischer, E. & Sauer, U. Experimental Identification and Quantification of Glucose Metabolism in Seven Bacterial Species. J. Bacteriol. 187, 1581±1590 (2005).

131. Tang, Y. J. et al. Flux analysis of central metabolic pathways in Geobacter metallireducens during reduction of soluble Fe(III)-nitrilotriacetic acid. Appl. Environ. Microbiol. 73, 3859±3864 (2007).

132. Del Castillo, T. et al. Convergent peripheral pathways catalyze initial glucose catabolism in Pseudomonas putida: Genomic and flux analysis. J. Bacteriol. 189, 5142±5152 (2007).

133. Gunnarsson, N., Mortensen, U. H., Sosio, M. & Nielsen, J. Identification of the Entner- Doudoroff pathway in an antibiotic-producing actinomycete species. Mol. Microbiol. 52, 895±902 (2004).

134. Metallo, C. M. et al. Reductive glutamine metabolism by IDH1 mediates under hypoxia. Nature 481, 380±4 (2012).

135. Feng, X. et al. Characterization of the central metabolic pathways in Thermoanaerobacter sp. strain X514 via isotopomer-assisted metabolite analysis. Appl. Environ. Microbiol. 75, 5001±5008 (2009).

136. Tang, Y. J. et al. Investigation of carbon metabolism in Dehalococcoides ethenogenes strain 195 by use of isotopomer and transcriptomic analyses. J. Bacteriol. 191, 5224± 5231 (2009).

137. Ishii, N. et al. Multiple high-throughput analyses monitor the response of E. coli to perturbations. Science 316, 593±597 (2007).

138. Nanchen, A., Schicker, A., Revelles, O. & Sauer, U. Cyclic AMP-dependent catabolite repression is the dominant control mechanism of metabolic fluxes under glucose limitation in Escherichia coli. J. Bacteriol. 190, 2323±2330 (2008).

139. Oliveira, A. P. et al. Regulation of yeast central metabolism by enzyme phosphorylation. Mol. Syst. Biol. 8, 623 (2012).

140. Rühl, M., Le Coq, D., Aymerich, S. & Sauer, U. 13C-flux analysis reveals NADPH- balancing transhydrogenation cycles in stationary phase of nitrogen-starving Bacillus subtilis. J. Biol. Chem. 287, 27959±70 (2012).

141. Fuhrer, T. & Sauer, U. Different biochemical mechanisms ensure network-wide balancing of reducing equivalents in microbial metabolism. J. Bacteriol. 191, 2112±21 (2009).

142. Metallo, C. M., Walther, J. L. & Stephanopoulos, G. Evaluation of 13C isotopic tracers for metabolic flux analysis in mammalian cells. J. Biotechnol. 144, 167±174 (2009).

33 Chapter 1 ! 143. Buescher, J. M. et al. A roadmap for interpreting 13C metabolite labeling patterns from cells. Curr. Opin. Biotechnol. 34, 189±201 (2015).

144. Antoniewicz, M. R. 13C metabolic flux analysis: Optimal design of isotopic labeling experiments. Curr. Opin. Biotechnol. 24, 1116±1121 (2013).

145. Kruger, N. J., Masakapalli, S. K. & Ratcliffe, R. G. Strategies for investigating the plant metabolic network with steady-state metabolic flux analysis: Lessons from an Arabidopsis cell culture and other systems. J. Exp. Bot. 63, 2309±2323 (2012).

146. Ahn, W. S. & Antoniewicz, M. R. Towards dynamic metabolic flux analysis in CHO cell cultures. Biotechnol. J. 7, 61±74 (2012).

147. Vacanti, N. M. et al. Regulation of Substrate Utilization by the Mitochondrial Pyruvate Carrier. Mol. Cell 56, 425±435 (2014).

148. Grassian, A. R. et al. IDH1 mutations alter metabolism and increase dependence on oxidative mitochondrial metabolism. Cancer Res. 74, 3317±31 (2014).

149. Vo, T. D., Paul Lee, W. N. & Palsson, B. O. Systems analysis of energy metabolism HOXFLGDWHV WKH DIIHFWHG UHVSLUDWRU\ FKDLQ FRPSOH[ LQ /HLJK¶V V\QGURPH Mol. Genet. Metab. 91, 15±22 (2007).

150. Noguchi, Y. et al. Effect of anaplerotic fluxes and availability on hepatic lipoapoptosis. J. Biol. Chem. 284, 33425±33436 (2009).

151. Masakapalli, S. K. et al. Subcellular flux analysis of central metabolism in a heterotrophic Arabidopsis cell suspension using steady-state stable isotope labeling. Plant Physiol. 152, 602±619 (2010).

152. Allen, D. K., Shachar-Hill, Y. & Ohlrogge, J. B. Compartment-specific labeling information in 13C metabolic flux analysis of plants. Phytochemistry 68, 2197±210 (2007).

153. Lewis, C. a et al. Tracing Compartmentalized NADPH Metabolism in the Cytosol and Mitochondria of Mammalian Cells. Mol. Cell 55, 253±263 (2014).

154. Goudar, C. et al. Metabolic flux analysis of CHO cells in perfusion culture by metabolite balancing and 2D [13C, 1H] COSY NMR spectroscopy. Metab. Eng. 12, 138±49 (2010).

155. Quek, L.-E., Dietmair, S., Krömer, J. O. & Nielsen, L. K. Metabolic flux analysis in mammalian cell culture. Metab. Eng. 12, 161±71 (2010).

156. Young, J. D. INCA: A computational platform for isotopically non-stationary metabolic flux analysis. Bioinformatics 30, 1333±1335 (2014).

157. Iwatani, S. et al. Determination of metabolic flux changes during fed-batch cultivation from measurements of intracellular amino acids by LC-MS/MS. J. Biotechnol. 128, 93± 111 (2007).

158. Costenoble, R. et al. 13C-Labeled metabolic flux analysis of a fed-batch culture of elutriated Saccharomyces cerevisiae. FEMS Yeast Res. 7, 511±526 (2007).

34 Chapter 1 ! 159. Schwender, J. Metabolic flux analysis as a tool in metabolic engineering of plants. Curr. Opin. Biotechnol. 19, 131±137 (2008).

160. Allen, D. K., Libourel, I. G. L. & Shachar-Hill, Y. Metabolic flux analysis in plants: Coping with complexity. Plant, Cell Environ. 32, 1241±1257 (2009).

161. Shastri, A. a. & Morgan, J. a. A transient isotopic labeling methodology for 13C metabolic flux analysis of photoautotrophic microorganisms. Phytochemistry 68, 2302±2312 (2007).

162. Charoenrat, T., Ketudat-Cairns, M., Stendahl-Andersen, H., Jahic, M. & Enfors, S. O. Oxygen-limited fed-batch process: An alternative control for Pichia pastoris recombinant protein processes. Bioprocess Biosyst. Eng. 27, 399±406 (2005).

163. Peyraud, R. et al. Demonstration of the ethylmalonyl-CoA pathway by using 13C metabolomics. Proc. Natl. Acad. Sci. U. S. A. 106, 4846±4851 (2009).

!

!

35

36 Chapter 2

Non-stationary 13C metabolic flux ratio analysis

Manuel Hörl, Julian Schnidder, Uwe Sauer and Nicola Zamboni

Published in: Biotechnology & Bioengineering 2013 Dec; 110(12):3164-76.

Manuel Hörl designed the study, performed all experiments and data analysis and wrote the manuscript. Julian Schnidder contributed to the computational implementation. Uwe Sauer and Nicola Zamboni conceived and supervised the study.!

37 Chapter 2

Abstract

13C metabolic flux analysis (13C MFA) has become a key method for metabolic engineering and systems biology. In the most common methodology, fluxes are calculated by global isotopomer balancing and iterative fitting to stationary 13C labeling data. This approach requires a closed carbon balance, long-lasting metabolic steady state, and the detection of 13C patterns in a large number of metabolites. These restrictions mostly reduced the application of 13C MFA to the central carbon metabolism of well-studied model organisms grown in minimal media with a single carbon source. Here we introduce non-stationary 13C metabolic flux ratio analysis as a novel method for 13C MFA to allow estimating local, relative fluxes from ultra-short 13C labeling experiments and without the need for global isotopomer balancing. The approach relies on the acquisition of non-stationary 13C labeling data exclusively for metabolites in the proximity of a node of converging fluxes and a local parameter estimation with a system of ordinary differential equations. We developed a generalized workflow that takes into account reaction types and the availability of mass spectrometric data on molecular ions or fragments for data processing, modeling, parameter and error estimation. We demonstrated the approach by analyzing three key nodes of converging fluxes in central metabolism of Bacillus subtilis.

We obtained flux estimates that are in agreement with published results obtained from steady state experiments, but reduced the duration of the necessary 13C labeling experiment to less than a minute. These results show that our strategy enables to formally estimate relative pathway fluxes on extremely short time scale, neglecting cellular carbon balancing. Hence this approach paves the road to targeted 13C MFA in dynamic systems with multiple carbon sources and towards rich media.

38 Chapter 2

Introduction

The activity of cellular metabolism is defined by the material fluxes through the network that result from the integrated interplay of nutrient availability, enzyme kinetics, and cellular regulation1±3. Experimental assessment of intracellular fluxes is pivotal to identify the factors governing metabolic responses to external and internal cues, for example in systems biology4,5 or metabolic engineering6±8. In vivo intracellular fluxes are not directly measureable, but have to be inferred from isotopic tracer experiments, typically involving 13C enriched substrates9±12.

For this purpose, an ensemble of experimental methods has been developed under the general label of 13C metabolic flux analysis (13C MFA). When 13C labeled nutrients are administered to cells, the isotopic label propagates through the metabolic network according to the intracellular fluxes. Depending on the activity of different metabolic pathways, the carbon backbone is rearranged to produce characteristic 13C patterns in metabolic intermediates and cellular components. These labeling patterns are detectable by mass spectrometry or nuclear magnetic resonance spectrometry. Finally, metabolic fluxes are inferred mathematically from the measured 13C patterns, the uptake rates of substrates, and the production rates of all products within the constraints given by the stoichiometry of the enzymatic reaction network.

The most used and flexible approach for calculating fluxes is global isotopomer balancing and iterative flux fitting within a detailed model of atomic transitions for all compounds in the network of interest8,11. This approach seeks to find the set of intracellular fluxes that complies with measured physiological rates, fulfills metabolic balances, and would generate 13C labeling patterns that best match the measured ones. The methodology has allowed the estimation of intracellular fluxes in central carbon metabolism of bacteria13±18, yeast19,20, plants21,22 and with some approximations also in mammalian cells23±26. To apply global isotopomer balancing, however, a number of aspects must be addressed. First, the method requires a closed carbon balance. This implies the measurement of all substrate uptake and product secretion rates, which is particularly challenging for cells growing in rich media. Additionally, knowledge about the biomass composition of the organism is required because fluxes in biosynthetic pathways leading to the formation of cellular building blocks are not estimated from 13C data but inferred

39 Chapter 2

from the assumed biomass composition. Second, the labeling experiment should be designed to provide characteristic labeling information for many metabolites in order to resolve most fluxes in the network and to obtain a distinct flux solution. For complex metabolic networks, this is not possible from a single experiment, which is the major problem with mammalian cells27,28. Third, global isotopomer balancing is susceptible to ill-defined networks, because all fluxes are fitted at once.

These limitations mostly restricted 13C MFA to conditions that allowed to attain isotopic steady- state after long labeling time and thereby limited the estimation of fluxes to central carbon metabolism of well-studied organisms grown on a single carbon source in defined minimal media. However, a comprehensive understanding of cellular physiology requires the possibility to also estimate metabolic fluxes under real-life conditions, which typically feature transient growth and multiple carbon sources. An ideal framework would therefore (i) allow flux estimation from very short time-scale labeling experiments to also investigate dynamic systems where metabolic steady states are brief and (ii) be rigorously applicable to any part of the network even if (iii) multiple substrates are co-metabolized. Although there has already been substantial progress reducing the issue of long experimentation times20,29±32, by now there is no approach that combines all of the aforementioned aspects.

We introduce here non-stationary 13C metabolic flux ratio analysis as a novel strategy for 13C

MFA. To overcome the limitations that arise from global isotopomer balancing and fitting, we focused on specific metabolic nodes where pathways converge and estimated the relative fluxes locally at the nodes. For this purpose, non-stationary 13C data was collected for a small set of metabolites which are in proximity to the node of interest. A small-sized system of ordinary differential equations (ODEs) that describe the local atom mapping was used for formal parameter estimation. To provide a generally applicable framework, we first defined a workflow that takes into account varying reaction types and the availability of mass spectrometry data on molecular ions or fragments. The principles of the workflow are presented in the theory section. The framework was then systematically validated for three key nodes of converging fluxes between glycolysis and tricarboxylic acid (TCA) cycle of B. subtilis.

40 Chapter 2

Our results demonstrate that non-stationary 13C metabolic flux ratio analysis enables to correctly estimate relative pathway fluxes on extremely short time-scale while neglecting the majority of reactions in central carbon metabolism.

41 Chapter 2

Materials and Methods

Strains and media

All experiments were performed with wild-type B. subtilis BSB168 trp+33. Frozen glycerol stocks were used to inoculate 5 mL of Luria-Bertani (LB) medium. After 5 h of incubation at

37°C and 300 rpm on a gyratory shaker, 5 mL of M9 minimal medium were inoculated at 1000- to 4000-fold dilutions as precultures. Mid-exponentially M9 precultures at optical densities at

600 nm (OD600) of 1-2 were then used to inoculate a 70 mL M9 batch culture in a 1 L shake flask to an OD600 of 0.03. The M9 medium contained, per liter of deionized water: 8.5 g of

Na2HPO4 2H2O, 3.0 g KH2PO4, 1 g NH4Cl, 0.5 g NaCl and was adjusted to pH 7 before autoclaving. The following components were filter sterilized separately and then added (per liter of final medium): 1 mL of 1 M MgSO4, 1 mL of 0.1 M CaCl2, 1 mL 0.05 M FeCl3 containing

0.1 M citric acid and 10 mL of a trace element solution containing (per liter) 170 mg ZnCl2, 100 mg MnCl2 4H2O, 60 mg CoCl2 6H2O, 60 mg Na2MoO4 2H2O, and 43 mg CuCl2 2H2O.

Autoclaved glucose solution was added to a final concentration of 5 g L-1.

13C labeling experiments

13C labeling experiments were performed on filters as described by Link et al.34. B. subtilis was grown in shake flask culture to mid-exponential phase (OD600 between 0.8 and 1.2). 2 mL of the culture broth was transferred onto a 0.45 µm pore size nitrocellulose filter (Millipore) and vacuum-filtered, followed by a 10 seconds washing phase with preheated 50% M9 medium containing 2 g L-1 naturally labeled glucose. 13C labeling was initiated by changing the washing solution to 50% M9 medium containing 2 g L-1 of a mixture of 50% (w/w) uniformly labeled and

50% naturally labeled glucose. The exact experimental setup and washing profile is illustrated in Fig. 4a and Fig. 4b, respectively. 13C labeling was performed in triplicates with cells from different shake flasks. Experiments were done in a temperature room kept at 37°C.

42 Chapter 2

13C pattern analysis of metabolic intermediates by LC-MS/MS

After each 13C labeling step, the nitrocellulose filter was immediately transferred into 4 mL of

60% (v/v) ethanol/water kept 3 min at 78°C for extraction. The metabolite extract was separated from cell debris and nitrocellulose by centrifugation at 14,000 g at 4°C for 10 min.

The supernatants were dried at 0.12 mbar in a SpeedVac composed of an Alpha 2-4 LD plus cooling trap, a RVC 2-33 rotational vacuum concentrator and a RC-5 vacuum chemical hybrid pump (Christ, Osterode am Harz, Germany). Dry metabolite extracts were stored at -80°C until further analysis.

Dried extracts were resuspended in 100 µL deionized water, 10 µL of which were injected into a Waters Acquity UPLC with a Waters T3 column (150 x 2.1mm x 1.8 µm; Waters Corporation,

Milford, MA, USA) coupled to a Thermo TSQ Quantum Ultra triple quadrupole instrument

(Thermo Fisher Scientific, Waltham, MA, USA) with electrospray ionization. Compound separation was achieved by a gradient of two mobile phases (i) 10 mM tributylamine, 15 mM acetic acid, 5% (v/v) methanol and (ii) 2-propanol35. Acquisition of mass isotopomer distributions (MIDs) of intact and fragmented carbon backbones was done as described by

Rühl et al.36. Peak integration was performed by an in-house software (Begemann and

Zamboni, unpublished).

Model construction

Local atom mapping at metabolic merging nodes was modeled with mass isotopomer balances and mass isotopomer/isotopomer hybrid balances, depending on reaction stoichiometry at the node and on availability of mass spectrometric data for the balanced intermediates. The principles are detailed in the theory section.

Preprocessing of data

MIDs were calculated from measured peak areas as previously described12. The resulting

MIDs were corrected for naturally occurring 13C37. Average and standard deviations of the measured mass fractions were calculated using three biological replicates. Measured MID time courses of the input metabolites (substrates) were perturbed by a Monte Carlo procedure,

43 Chapter 2

where all measurements were superimposed with random noise sampled from a normal distribution with standard deviation equivalent to the experimentally measured deviation. The resulting MIDs were then normalized, interpolated by cubic splines and used to simulate the carbon labeling over time with the ODE model. The measured dynamics and standard deviations of the output MIDs (products) were used as observed data and applied for weighted least square fitting.

Generation of artificial isotopomer and mass isotopomer distributions

Mass isotopomer distributions for non-measureable metabolites were randomly sampled assuming Hill kinetics (Eq 1).

௧೙ (ሺݐሻ ൌ (1ܦܫܯ ௧೙ା௄

This kinetic law was chosen to flexibly mimic the initial dynamics following a switch from unlabeled to labeled substrate. This switch typically leads to a non-linear increase of the relative abundance of all labeled isomers and a decrease of the [U-12C] form. Hence, for a metabolite A with c carbon atoms, c time curves with random n and K were generated using

Eq 1. These were normalized such that their sum at isotopic steady state was equal to the

13 fractional labeling of the C tracer, and assigned to the Am+1«$m+c mass isotopomers. The profile of the fraction Am+0 was calculated such that at any time point the sum of all mass isotopomer fractions equals 1.

Isotopomer distributions necessary to model cleavage reactions were randomly sampled to ensure that they comply to the measured mass isotopomer distributions. For each mass isotopomer fraction, the measured values were randomly subdivided in the isotopomer fractions of isomers with matching mass. The resulting time courses were superimposed with noise as above, again normalized to ensure that the sum of all isotopomers at any time point is 1, and interpolated with cubic splines. The mass isotopomer fractions calculated by summation of randomly sampled isotopomer fractions are identical to the measured values within the tolerance given by the experimental noise.

44 Chapter 2

Parameter estimation and statistical analysis

The flux ratio between converging fluxes and the product pool size were estimated by fitting simulated to measured mass distributions with PottersWheel 2.2.1338. At each iteration, we added random noise within the experimentally determined standard deviations to the measurements of input metabolites and sampled artificial data whenever necessary. These data were used to simulate the time-dependent labeling of the output metabolite, i.e. the compound at the metabolic merging node. Parameters were fitted by minimizing the weighted sum of squared residuals (Ȥ2) between measured and simulated labeling kinetics. Estimated

IOX[ UDWLR SRRO VL]H DQG Ȥ2-values were saved after each iteration. A total of at least 5000 independent fitting iterations were performed for each node. The noising procedure frequently lead to cases in which the model could not properly fit the data. These estimates were rejected

EDVHG RQ Ȥ2-values with significance level below 0.05, as specified in the PottersWheel documentation38. Because of the randomized noising, some iterations led to a transformation of the input data to biologically implausible trajectories which still fell into the group of estimates with a significance level above 0.05. Since we expected these fittings to give results with poor statistical significance, we calculated the flux ratio as median of the distribution of the 20% of estimates with highest significance levels. We excluded that this cutoff lead to a bias of the calculated flux ratio, by also testing a more restrictive (10%) and a less constrained (30%) cutoff, which gave equivalent results. Standard deviations were calculated from the 68% confidence interval of the distribution. All calculations were performed with Matlab 7.12.0 (The

Matworks Inc., Natick, MA).

A two-stage estimation procedure was adopted in the case of systems that required artificial input data. Since the missing information was sampled using Hill kinetics, we first screened for meaningful ranges of Hill parameters testing a large space of combinations in a grid search algorithm and a modest number of iterations (up to 1000 per combination). Reasonable ranges for Hill parameters were identified based on estimations that resulted in a Ȥ2-value with a significance level above 0.05. The identified range of Hill parameters compatible with the

45 Chapter 2

measured labeling data was then used to estimate the flux ratio with a larger amount of

Monte Carlo samples.

46 Chapter 2

Theory

Workflow of non-stationary 13C metabolic flux ratio analysis

Non-stationary 13C metabolic flux ratio analysis aims at estimating the relative contribution of pathways to the biosynthesis of a common, downstream product. In contrast to the well- established approach of global isotopomer balancing, we focus on a local problem and formally model 13C label propagation only in proximity of a metabolic node of interest. Each node of converging fluxes is modeled with a small scale ordinary differential equation (ODE) system to simulate the dynamic labeling patterns of metabolites from the labeling patterns of its precursors. Model construction depends on the type of biochemical reactions composing the metabolic node of interest and the data available from mass spectrometry. The basic principles are detailed in the following sections and illustrated in Fig. 1. The model is then used to identify the set of parameters, including the flux ratio of interest, that are compatible with the measured

13C patterns. Parameter estimation is repeated thousands of times varying input data according to the experimentally measured variance, to finally obtain a distribution for the flux ratio of interest based on several hundreds of individual estimates that passed all quality criteria.

Model construction

Given a metabolic node of interest with two or more converging pathways, a model should be constructed that precisely reflects the reaction topology. We distinguish between three basic reaction types: unimolecular, condensation and cleavage reactions (Fig. 2a) that together can produce basic network motifs (Fig. 2b-f) or more complex networks. The construction of an

ODE system depends on the type of reactions that constitute the network. Whenever possible, we construct the models to rely solely on measurable data, i.e. mass isotopomer distributions of metabolites. Mass isotopomers differ only in the number of 13C atoms contained in the molecule and so the MID of a metabolite with n carbon atoms represents the relative amounts of the n+1 mass isotopomers. However, depending on the topology of the reactions and the mapping of carbon atoms between substrates and products, the model might require the inclusion of non-measurable entities such as non-detectable compounds or isotopomer

47 Chapter 2

distributions (IDs), which additionally discriminate between the positions of 13C atoms and are not completely measureable by mass spectrometry, and therefore complicate computation and flux estimation. In the optimal case, MIDs are detectable for the metabolites directly reacting at the converging node of interest. If not all of these MIDs are measureable, there is the possibility to substitute the missing information by the measureable MID of a closely connected metabolite. However, one has to be aware that this can impact the flux ratio calculation since the labeling enrichment will be temporarily shifted for the surrogate, compared to the unobservable metabolite. The extent of the difference depends on (i) the rate of equilibration of the connecting reaction between both metabolites and (ii) their pool sizes, therefor it has to be evaluated considering such a replacement.

In the following sections we describe the underlying principles for the basic network motifs.

More complex networks can be modeled following the same rules. For each motif, we assume that metabolite Z is solely produced by the fluxes v1 and v2. If these reactions are reversible, v1 and v2 are the forward fluxes leading to Z formation. The effect of potential reverse fluxes from Z on the labeling patterns of the educts (e.g. A and B in Fig. 2b) is implicitly accounted for by the availability of experimental data for their mass distribution that integrate the contribution of all biosynthetic sources for each educt. In our example motifs, v3 is a unidirectional forward flux converting Z into its downstream product. If this reaction is reversible, the backward reaction constitutes an additional flux into the Z pool, which has to be included in the model. Although the presented principles are still valid for this reversible case, it was not further investigated within this study.

48 Chapter 2

Figure 1. Workflow for non-stationary 13C metabolic flux ratio analysis. MID: mass isotopomer distribution; ID: isotopomer distribution; ODE: ordinary differential equation.

Metabolic nodes formed uniquely by unimolecular reactions

The simplest motif of a metabolic node of converging fluxes is composed of two unimolecular reactions converging to the same metabolite, which is further processed by an irreversible reaction of any type (Fig. 2b). In a tracer experiment, the temporal dynamics of the mass isotopomer distribution of the common product Z depend on (i) the mass isotopomer distribution vectors of its precursor metabolites A and B, (ii) the pool size PZ of metabolite Z, and (iii) the fluxes v1 and v2 forming metabolite Z (Eq. 2).

ௗ௓೘శబ ଵ (௠ା଴ െ ݒଷ ή ܼ௠ା଴ሻ (2ܤ ௠ା଴ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శభ ଵ ௠ାଵ െ ݒଷ ή ܼ௠ାଵሻܤ ௠ାଵ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శమ ଵ ௠ାଶ െ ݒଷ ή ܼ௠ାଶሻܤ ௠ାଶ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శయ ଵ ௠ାଷ െ ݒଷ ή ܼ௠ାଷሻܤ ௠ାଷ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

49 Chapter 2

At metabolic steady state, the flux leaving the node (v3) equals the sum of fluxes v1 and v2. We obtain a system of ordinary differential equations with v1, v2 and PZ as unknowns.

ௗ௓೘శబ ଵ (௠ା଴ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ା଴ሻ (3ܤ ௠ା଴ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శభ ଵ ௠ାଵ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଵሻܤ ௠ାଵ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శమ ଵ ௠ାଶ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଶሻܤ ௠ାଶ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శయ ଵ ௠ାଷ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଷሻܤ ௠ାଷ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

This system allows to simulate the time-dependent mass isotopomer distribution of Z from

௩భ those of A and B assuming values for v1, v2 and PZ are known. Conversely, the flux ratio ௩భା௩మ and ଵ can be estimated by fitting simulated to measured values. The ଵ term is the inverse of ௉ೋ ௉ೋ the Z pool and must be handled as unknown. Knowledge of the pool size and the total flux v3 allows, in addition, to estimate the absolute fluxes around the Z pool.

Metabolic nodes containing condensation reactions

Condensation reactions merge two or more carbon-containing substrates into a larger product.

If the mass distribution of all reactants can be measured by mass spectrometry, the ODE system for parameter fitting can be constructed based on MIDs as in the previous case for unimolecular reactions. The mass distribution of a condensation product is obtained by the

Cauchy product between the mass isotopomer fractions of the educts. For the example in

Fig. 2c, the ODEs describing the temporal dynamics of metabolite Z are given by Eq. 4.

ௗ௓೘శబ ଵ (௠ା଴ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ା଴ሻ (4ܧ ௠ା଴ ήܦ ௠ା଴ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శభ ଵ ௠ାଵሻ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଵሻܧ ௠ା଴ ήܦ ௠ା଴ ൅ܧ ௠ାଵ ήܦ௠ାଵ ൅ ݒଶ ή ሺܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శమ ଵ ௠ାଵሻ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଶሻܧ ௠ାଵ ήܦ ௠ା଴ ൅ܧ ௠ାଶ ήܦ௠ାଶ ൅ ݒଶ ή ሺܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శయ ଵ ௠ାଵ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଷሻܧ ௠ାଶ ήܦ ௠ାଷ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

50 Chapter 2

Figure 2. (a) Basic reaction types for modeling nodes of converging fluxes. (b-f) Example motifs of nodes of converging fluxes. We distinguish between nodes containing only unimolecular reactions (b), one unimolecular and one condensation reaction (c,d) and one unimolecular and one cleavage reaction (e,f). Grey fillings refer to carbon backbones for which the MID is measureable, white fillings indicate non-measureable compounds. Dashed lines represent molecule parts split during fragmentation in the MS.

A typical case of condensation reaction is the incorporation of CO2. Since the time-dependent

13 C enrichment of intracellular CO2 is virtually not accessible by experimental means, we lack an essential component for parameter fitting. A possible solution for this impasse is provided by the use of collisionally induced fragmentation in mass spectrometry. With this technique, it is possible for some compounds to obtain the mass distribution for fragments thereof (Rühl et al. 2011). In case matching fragments are accessible by mass spectrometry, the flux estimation can be performed based on the mass distribution of the measurable moieties as in the aforementioned case of unimolecular reactions. For the example shown in Fig. 2d, an amended ODE system independent of the non-measurable compound E can be designed based on the two carbon portions of A (A12), Z (Z12) and D:

51 Chapter 2

ௗ௓భమǡ೘శబ ଵ (௠ା଴ െ ሺݒଵ ൅ ݒଶሻ ή ܼଵଶǡ௠ା଴ሻ (5ܦ ଵଶǡ௠ା଴ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓భమǡ೘శభ ଵ ௠ାଵ െ ሺݒଵ ൅ ݒଶሻ ή ܼଵଶǡ௠ାଵሻܦ ଵଶǡ௠ାଵ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓భమǡ೘శమ ଵ ௠ାଶ െ ሺݒଵ ൅ ݒଶሻ ή ܼଵଶǡ௠ାଶሻܦ ଵଶǡ௠ାଶ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

Unfortunately, the fragmentation patterns of metabolites depend on the molecular structure and can hardly be controlled to selectively break carbon bonds at will. In case matching fragments in substrates and common product are not experimentally accessible, the ODE system in Eq. 6 cannot be used for parameter estimation. Instead, the balance must be constructed based on the intact reactants as in Eq. 4. The lack of experimental information on

CO2 (or any other input metabolite) is overcome by performing the parameter estimation with artificial, randomly sampled labeling data. The procedure is detailed in the methods, and ensures that the estimation of fluxes is constrained primarily by experimental data. This complicates parameter estimation and, depending on the available labeling data, negatively affects confidence intervals.

Metabolic nodes containing cleavage reactions

Cleavage reactions at merging nodes produce multiple products (Fig. 2e). The propagation of label from substrates to products can be correctly modeled at the level of mass distributions only if matching fragments can be obtained by mass spectrometry. For the example depicted in Fig. 2f, assuming that we can obtain the mass distribution vector of the fragment C1-C3 of substrate F (F123) by mass spectrometry, the balance for the merging node is given by

ௗ௓೘శబ ଵ (ଵଶଷǡ௠ା଴ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ା଴ሻ (6ܨ ௠ା଴ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శభ ଵ ଵଶଷǡ௠ାଵ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଵሻܨ ௠ାଵ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శమ ଵ ଵଶଷǡ௠ାଶ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଶሻܨ ௠ାଶ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శయ ଵ ଵଶଷǡ௠ାଷ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଷሻܨ ௠ାଷ ൅ ݒଶ ήܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

This balance is independent of the cleaved molecule part F4 that does not contribute to Z, and can be used to estimate the flux ratio. However, matching fragments are rarely available. This

52 Chapter 2

precludes the option to model the node based purely on mass distributions. Instead, positional information has to be included to correctly map the transition of atoms from substrates to the common metabolite. For this purpose, the labeling information of the substrates of cleavage reactions must be modeled listing all isotopomer fractions individually. For the example in

Fig. 2e, the balances of the mass distribution for metabolite Z are

ௗ௓೘శబ ଵ (଴଴଴ଵሻ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ା଴ሻ (7ܨ ଴଴଴଴ ൅ܨ௠ା଴ ൅ ݒଶ ή ሺܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శభ ଵ ଴଴ଵଵሻ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଵሻܨ ଴ଵ଴ଵ ൅ܨ ଵ଴଴ଵ ൅ܨ ଴଴ଵ଴ ൅ܨ ଴ଵ଴଴ ൅ܨ ଵ଴଴଴ ൅ܨ௠ାଵ ൅ ݒଶ ή ሺܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శమ ଵ ଴ଵଵଵሻ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଶሻܨ ଵ଴ଵଵ ൅ܨ ଵଵ଴ଵ ൅ܨ ଴ଵଵ଴ ൅ܨ ଵ଴ଵ଴ ൅ܨ ଵଵ଴଴ ൅ܨ௠ାଶ ൅ ݒଶ ή ሺܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

ௗ௓೘శయ ଵ ଵଵଵଵሻ െ ሺݒଵ ൅ ݒଶሻ ή ܼ௠ାଷሻܨ ଵଵଵ଴ ൅ܨ௠ାଷ ൅ ݒଶ ή ሺܣ ൌ ή ሺݒଵ ή ௗ௧ ௉ೋ

Where Am+0«$m+n are the fractions of the mass distribution of metabolite A and e.g. F0001 is the fraction of metabolite F which is 13C labeled only at position C4. The labeling content of the substrate A and the product Z can be still modeled with (pooled) mass isotopomer fractions.

The use of mass isotopomer fractions is generally preferred whenever possible, because they reflect the labeling data measurable by mass spectrometry. In contrast, the isotopomer fractions necessary to simulate carbon propagation in cleavage reactions are only partly measurable. Therefore, time-dependent trajectories of isotopomer fractions are randomly sampled under the constraint that the sum of the fraction of isotopomers with same mass is consistent with the measured mass isotopomer fractions (see also methods).

Parameter estimation

Flux ratios and pool sizes of a converging node are estimated by simulating the labeling patterns at the node, starting from the measured or artificial labeling patterns and using the

ODE model. The best set of parameters is identified by an iterative fitting procedure. To estimate confidence intervals that account for experimental error and are possibly independent of assumptions in artificial data, the fitting has to be repeated many times following a

Monte Carlo method39.

53 Chapter 2

Results

Non-stationary 13C metabolic flux ratio analysis of B. subtilis central metabolism

To test our method, we analyzed three key metabolic converging nodes in the central metabolism of B. subtilis (Fig. 3). We focused on (i) the formation of phosphoenolpyruvate from glycolysis vs. the gluconeogenic reaction catalyzed by phosphoenolpyruvate carboxykinase,

(ii) the generation of oxaloacetate from TCA cycle vs. the anaplerotic reaction from pyruvate and (iii) the production of pyruvate through malic enzymes vs. pyruvate kinase. These nodes are of key importance in metabolism as relative fluxes through them characterize the operation of glycolysis, and the oxidative TCA cycle. Furthermore, these nodes comprise the three basic reaction types where flux estimates obtained with canonical isotopic stationary flux ratio analysis and global isotopomer balancing with iterative fitting were available32,40±42.

Figure 3. Investigated flux ratios in B. subtilis central metabolism. The phospoenolpyruvate node, constituted by the unimolecular reaction of enolase (ENO) and the cleavage reaction of phosphoenolpyruvate carboxy kinase (PCK) is highlighted in white. Pyruvate formation occurs via the unimolecular reaction of pyruvate kinase (PYK) and the cleavage reaction catalyzed by malic enzymes (MAE) (highlighted in black). The oxaloacetate node consists of the unimolecular reaction of malate dehydrogenase (MDH) and the condensation reaction of pyruvate carboxylase (PYC) and is highlighted in grey. Detectable 13C patterns of carbon backbones are indicated in grey, numbers refer to carbon atom positions.

54 Chapter 2

Non-stationary 13C labeling experiment

To measure non-stationary 13C labeling dynamics during exponential growth, we grew B. subtilis in glucose batch cultures. Since extensive overflow metabolism to acetate and pyruvate occurs under this condition, accumulation of non-labeled extracellular acetate and pyruvate could compromise the measurement of intracellular 13C labeling kinetics, i.e. of pyruvate.

Therefore, we used a variant of the filter cultivation method34,37 to perform the labeling experiment. This experimental setup allows for thorough washing before initiating 13C labeling without perturbing intracellular metabolite levels, and a precisely timed cell harvest34. B. subtilis was initially grown in shake flasks on naturally labeled glucose (Fig. 4a). When the culture reached an OD600 between 0.8 and 1.2, an aliquot of cells was transferred to a membrane filter, vacuum-filtered, and washed with fresh medium. Immediately after washing, the 13C labeling experiment started by changing the washing solution to a medium containing

50% (w/w) [U-13C]glucose. Rapid and frequent sampling was obtained by transferring the filter with the cell film in hot 60% ethanol/water solution. Labeling kinetics were monitored by varying the duration of washing cells with 13C medium (Fig. 4b). Mass isotopomer distributions of free intracellular metabolites of central carbon metabolism were recorded by targeted LC-MS/MS analysis36. With this experimental setup, we were able to reproducibly measure the 13C labeling kinetics at intervals of as little as 5 sec for a duration of up to one minute (Fig. 5). The 13C dynamics obtained with this system were similar or faster to those obtained from a bioreactor experiment (data not shown), demonstrating the validity of the employed setup. The resulting

13C data favorably matches published data from other bacteria29,43 with most glycolytic metabolites reaching isotopic steady state already within one minute, and slower label enrichment in TCA cycle intermediates.

55 Chapter 2

Figure 4. (a) Experimental setup to determine 13C labeling kinetics of intracellular metabolites. Cells from a shake flask culture were transferred onto a membrane filter, vacuum filtered and washed with fresh medium before onset of 13C labeling. (b) 13C labeling kinetics were estimated by washing with medium containing [U-13C]glucose for different durations.

Figure 5. 13C labeling data and standard deviations of intracellular metabolites and metabolite fragments used for the estimation of metabolic flux ratios.

56 Chapter 2

Phosphoenolpyruvate from glycolysis vs. gluconeogenesis

We first analyzed the ratio between glycolytic and gluconeogenic production of phosphoenolpyruvate. Phosphoenolpyruvate can be formed by enolase in a unimolecular reaction from phosphoglycerate or via cleavage of oxaloacetate by PEP carboxykinase. Since oxaloacetate cannot be detected due to degradation during sample processing and ionization in the mass spectrometer, we used the labeling kinetics of aspartate that originates from oxaloacetate by transamination. Rapid propagation of 13C was confirmed in our experiment by the rapid accumulation of 13C in aspartate comparable to that of malate, which is also one reaction step away from oxaloacetate. By collisional fragmentation, we could directly measure the mass isotopomer distribution of the aspartate C2-C3-C4 fragment (aspartate234), which corresponds to the oxaloacetate234 fragment that fuels the phosphoenolpyruvate pool via PEP carboxykinase. This node is equivalent to the motif shown in Fig. 2f. Therefore, the ODE model was constructed entirely with mass isotopomer distributions as detailed in Supplementary

Table I.

Table I. Ratios of converging glucose fluxes in batch grown B. subtilis obtained with different 13C MFA methods. lb: lower bound; ub: upper bound.

Flux ratios (%)

Oxaloacetate from Method TCA cycle PEP from glycolysis Pyruvate from malate

Non-stationary metabolic flux ratio 50 ± 9 100 ± 2 27 ± 4 analysis

Stationary metabolic flux ratio analysis 56a 99a n.d.a 50 ± 2b 97 ± 3b 6 ± 3 (lb); 11 ± 5 (ub)b 49 ± 4c 94 ± 2c 1 ± 2 (lb); 3 ± 5 (ub)c

Global isotopomer balancingd 51 ± 6 97 ± 2 8 ± 6 aFischer and Sauer, 2005 bLerondel et al., 2006 cRühl et al., 2010 dKleijn et al., 2010

We performed the parameter estimation using the 13C data of phosphoenolpyruvate, phosphoglycerate and aspartate234 measured over the entire duration of the labeling experiment (1 min). The estimated ratio was 100 ± 2 % (Fig. 6a) and favorably matched

57 Chapter 2

previously reported results for wild-type B. subtilis with absent gluconeogenesis in the formation of phosphoenolpyruvate (Table I). To investigate the impact of the labeling experiment duration on the flux ratio estimation, we determined the parameter with a progressive decrease of data points, to identify the minimal experiment duration for a reliable parameter estimation. As expected, the accuracy of the estimation increased with the number of data points (Fig. 6b). With six data points collected over 35 sec, we obtained results comparable with solutions calculated from stationary 13C MFA which would require at least

30-60 minutes of labeling32.

Oxaloacetate from TCA cycle vs. anaplerosis

Next we attempted to estimate the relative contribution of the TCA cycle versus the anaplerotic reaction catalyzed by pyruvate carboxylase in the formation of oxaloacetate. In the TCA cyle, oxaloacetate is produced from malate by malate dehydrogenase. The reaction is coupled to

NADH, but in terms of carbon atoms it is equivalent to a unimolecular reaction. In contrast, anaplerosis occurs by condensation between pyruvate and CO2. The labeling patterns of malate and pyruvate are directly measurable by LC-MS/MS and instead of oxaloacetate we

13 used aspartate. The labeling state of intracellular CO2 was not available. Its C enrichment depends both on biotic factors, such as decarboxylating enzymes in the TCA cycle and the oxidative pentose phosphate pathway, and on abiotic factors as the exchange with water- soluble carbonate and ambient CO2. Therefore, artificial time courses were generated for CO2 enrichments by picking randomly from a heterogeneous population of trajectories. The CO2 labeling curves that lead to an acceptable fit as qualified by the Ȥ2-value are shown in

Supplementary Fig. 1. Most of them showed enrichment dynamics slower or comparable to those of TCA cycle metabolites, which seems plausible with the TCA cycle as the major source of CO2 production. The ratio of oxaloacetate originating from the TCA cycle was determined to be 50 ± 9 % (Fig 6c). Although the median was consistent with the results obtained with stationary methods (Table I), we wondered what could be the reason of the poor confidence.

58 Chapter 2

13 To test whether the randomly sampled CO2 kinetics caused the large deviation, we investigated the sensitivity of the flux ratio to this component. First, we performed the

13 parameter estimation with 1¶ random artificial CO2 labeling curves and measured

13 pyruvate and malate kinetics to identify the artificial CO2 time course that best fits the data.

To exclude the influence of measurement errors in this process, mean measured values were

13 used for pyruvate and malate without superimposing experimental noise. The best CO2 time

FRXUVH ZDV WKHQ IL[HG DQG ZH UHSHDWHG WKH SDUDPHWHU ILWWLQJ ZLWK ¶ QRLVH-sampled

13 pyruvate, malate, and CO2 data. The magnitude of pyruvate and malate noise was set to

13 reflect the experimental variation. The relevance of artificial CO2 data was tested by adding normally distributed noise with up to 20% variance. Large errors substantially modified the

13 13 shape of the original CO2 profile. If the artificial CO2 dynamics had a major impact on the

13 confidence interval of the flux ratio, increasing the amplitude of variations in the CO2 time course would decrease the confidence of the flux ratio. Since this was not observed (data not shown), we concluded that the large confidence interval was not due to the artificial sampling of CO2. Therefore, in comparison to stationary methods, our result at this node is probably less precise because of the data source used for estimation. Stationary methods use the labeling state of both free and protein-bound amino acids to infer the ratio, which are highly abundant compared to free metabolites. Due to the high abundance, labeling measurements of amino acids are more accurate and contain smaller standard deviations. Also the poorer data quality for pyruvate and malate labeling dynamics compared to labeling measurements of metabolites involved at the phosphoenolpyruvate node, further show that data quality is the reason for the lower precision in the estimate.

59 Chapter 2

Figure 6. (a, c, e) Distribution of statistically significant flux ratio estimates for metabolic nodes considered in this study using the entire dataset consisting of 8 data points collected within one minute. The median of the distribution is indicated as black line, grey areas indicate the 68% confidence interval. (b, d, f) Flux ratio distributions, estimated with datasets progressively reduced in data points. Whiskers extend to the 68% confidence interval.

60 Chapter 2

A stable flux ratio estimate could be obtained with the data measured in the first 35 sec of 13C labeling (Fig. 6d). Interestingly, the average estimate obtained with shorter durations drifts from very high values to the steady-state. This effect might be attributed to the existence of the intermediary pool of oxaloacetate that in our models was neglected and replaced by aspartate.

In the initial seconds upon 13C supplementation, oxaloacetate and aspartate might not be isotopically equivalent as we had previously assumed. Based on our results, the model simplification seems to be justified only after 30 seconds, i.e. when the flux ratio estimate becomes stable and comparable to the stationary estimate.

Pyruvate from malic enzyme vs. glycolysis

Next we tackled the quantification of the relative fluxes through the malic enzyme and pyruvate kinase, both producing pyruvate. The malic enzyme cleaves CO2 from malate at position C1.

Since neither CO2 nor the fragment malate234 were measurable by mass spectrometry, we had to model malate with all 24 isotopomers (Supplementary Table I). The isotopomer fractions were sampled within the boundaries set by the measured mass isotopomer distributions in each iteration. The flux ratio was estimated from the entire labeling experiment to be 27 ± 4%

(Fig. 6e). A similar estimate could be achieved with the 13C data pertinent to the first 35 sec of the labeling experiment (Fig. 6f).

The flux ratio estimate obtained here from non-stationary data is higher than the value of

0-16 % obtained by stationary metabolic flux ratio analysis (Table I). These results are not necessarily contradictory. The value obtained by stationary methods is likely a better indicator of the net flux carried by the malic enzyme44, and it is reproducibly found also when global isotopomer balancing and iterative fitting is used to calculate net fluxes41. In contrast, our non- stationary analysis reflects the ratio of forward fluxes to pyruvate. The two estimates suggest that the forward flux from malate to pyruvate is larger than the net flux, and a fraction of pyruvate is converted to malate by reversed malic enzyme reaction. These results are supported by a recent study which demonstrated that a substantial reversible flux through malic enzymes can also occur to create a transhydrogenation cycle in B. subtilis45.

61 Chapter 2

Discussion

We present non-stationary 13C metabolic flux ratio analysis as a generalized approach to estimate the relative contribution of pathways to the synthesis of a common metabolite. The methodology is based on local modeling of 13C label propagation in close proximity to the metabolic node of interest and formal parameter estimation by iterative weighted least square fitting. Analyzing three key metabolic merging nodes in central metabolism of B. subtilis with our framework, we were able to obtain estimates that are in good agreement with published data obtained from isotopically stationary methods. However, comparing isotopic stationary methods with our approach, the experimental duration of the labeling experiment could be reduced from at least 30-60 min12,32 to less than a minute while maintaining a comparable accuracy for the flux ratio estimates. Such extremely short time scale measurements were mainly possible because of the high fluxes in central carbon metabolism that rapidly enrich its intermediates with 13C DQG WKH PHWDEROLWHV¶ SUR[LPLW\ WR WKH HQWU\ SRLQW RI WKH WUDFHU

[U-13C]glucose. Even if the labeling dynamics were slower in other parts of metabolism, the results demonstrate that our approach has the potential to strongly shorten the duration required for a labeling experiment. Therefore, it can provide access to systems where a metabolic (pseudo)-steady state can only be obtained for very short time periods, such as e.g. during metabolic transients or in media with multiple carbon substrates that are consumed sequentially in distinct metabolic steady states.

The presented approach for non-stationary flux ratio analysis is highly complementary to the more classic global isotopomer balancing and iterative fitting. The latter requires a closed carbon balance and a metabolic model that accounts for all reactions between cellular substrates, products, and biomass components. The method performs perfectly with simple systems, e.g. microbes grown on single carbon sources, to generate comprehensive flux maps for all reactions in the model. However, its demand for data and computing power increases drastically for the non-stationary case and becomes almost prohibitive for the analysis of cells in rich media46. Non-stationary flux ratio analysis, in contrast, is a merely local analysis that relies only on the time-resolved 13C profiles of intermediates in proximity of the metabolic node

62 Chapter 2

of interest. The approach is independent of measuring any extracellular uptake or production rate. Even in absence of a global metabolite balancing, absolute net fluxes can be estimated from dynamic 13C data if the concentration of the metabolite at the merging node is known.

Compared to the approach of global isotopomer balancing and iterative fitting, our approach provides less flux resolution, because of the missing stoichiometric balancing within the entire network. This makes global isotopomer balancing less susceptible if labeling data is entirely missing for few metabolites compared to our local analysis. In theory, additional flux variables can be fitted with larger models that describe more extensive portions of the metabolic network.

However, this comes at the cost of additional variables that have to be fitted and demand for additional data. In summary, we believe that global isotopomer balancing is better suited to generally investigate differences in the response of an organism to environmental or genetic perturbations. In this respect, only the calculation of net fluxes over the entire network allows to draw overall balances on redox and energy metabolism. In contrast, our framework of non- stationary flux ratio analysis primarily provides a tool to formally validate specific hypotheses on fluxes in a possibly targeted and quantitative way.

The main drawback of the presented approach is that it is very difficult to assess a priori the calculability of a given (relative) flux by in silico simulations. An optimal tracer labeling strategy that maximizes the differences on MIDs of the precursors can be devised from simulating isotopically stationary experiments27,28,47. In contrast, the labeling dynamics of all metabolites depend both on the fluxes that lead to their formation, their pool size, and the labeling states of their precursors ± factors which can hardly be assessed without experimental data. In particular, it is virtually impossible to predict the minimal duration of the labeling experiment necessary to attain a stable solution. A second important shortcoming is that in models comprising reversible reactions, our approach only allows to estimate the ratio of forward fluxes, while the net contribution of each pathway might be a more relevant entity to describe metabolic operation.

The locality underpinning non-stationary flux ratio analysis offers several advantages. First, it reduces the number of MIDs to be measured in favor of analytical sensitivity, precision of

63 Chapter 2

labeling data, and time necessary for acquisition by mass spectrometry. Second, the ODE systems are very simple and allow rapid and exhaustive parameter estimation in short time.

Moreover, they are much less prone to converge in local minima than large models. Finally, the key advantage of the presented approach is that all parts of the network not included in the local ODE model can be fully neglected. Flux calculation becomes fully independent of any measurements of extracellular rates and of any (partial) knowledge of the metabolic network outside the boundaries encoded in the ODE model. The only precondition for flux estimation is that the input metabolites of the model feature measurable, characteristic, and time- dependent 13C patterns. This aspect has major consequences, since it paves the road to targeted 13C MFA in systems with multiple carbon sources and towards rich media, which are not tractable by other approaches, i.e. global isotopomer balancing and stationary flux ratio analysis.

Acknowledgements

We gratefully acknowledge Mattia Zampieri for helpful discussions and critical remarks on the manuscript. We are grateful for financial support through the Sinergia program of the Swiss

National Foundation and the SystemsX.ch project MetaNetX.

64 Chapter 2

References

1. Gerosa, L. & Sauer, U. Regulation and control of metabolic fluxes in microbes. Curr. Opin. Biotechnol. 22, 566±75 (2011).

2. Nielsen, J. It Is All about Metabolic Fluxes. J. Bacteriol. 185, 7031±7035 (2003).

3. Fernie, A. R., Geigenberger, P. & Stitt, M. Flux an important, but neglected, component of functional genomics. Curr. Opin. Plant Biol. 8, 174±182 (2005).

4. Sauer, U. Metabolic networks in motion: 13C-based flux analysis. Mol. Syst. Biol. 2, 62 (2006).

5. Zamboni, N. & Sauer, U. Novel biological insights through metabolomics and 13C-flux analysis. Curr. Opin. Microbiol. 12, 553±8 (2009).

6. Kohlstedt, M., Becker, J. & Wittmann, C. Metabolic fluxes and beyond-systems biology understanding and engineering of microbial metabolism. Appl. Microbiol. Biotechnol. 88, 1065±1075 (2010).

7. Otero, J. M. & Nielsen, J. Industrial systems biology. Biotechnol. Bioeng. 105, 439±60 (2010).

8. Stephanopoulos, G. Metabolic fluxes and metabolic engineering. Metab. Eng. 1, 1±11 (1999).

9. Kruger, N. J. & Ratcliffe, R. G. Insights into plant metabolic networks from steady-state metabolic flux analysis. Biochimie 91, 697±702 (2009).

10. Sauer, U. High-throughput phenomics: experimental methods for mapping fluxomes. Curr. Opin. Biotechnol. 15, 58±63 (2004).

11. Wiechert, W. 13C metabolic flux analysis. Metab. Eng. 3, 195±206 (2001).

12. Zamboni, N., Fendt, S.-M., Rühl, M. & Sauer, U. 13C-based metabolic flux analysis. Nat. Protoc. 4, 878±92 (2009).

13. Dauner, M. From fluxes and isotope labeling patterns towards in silico cells. Curr. Opin. Biotechnol. 21, 55±62 (2010).

14. Haverkorn van Rijsewijk, B. R. B., Nanchen, A., Nallet, S., Kleijn, R. J. & Sauer, U. Large-scale 13C-flux analysis reveals distinct transcriptional control of respiratory and fermentative metabolism in Escherichia coli. Mol. Syst. Biol. 7, 477 (2011).

15. Kiefer, P., Heinzle, E., Zelder, O. & Wittmann, C. Comparative Metabolic Flux Analysis of Lysine-Producing Corynebacterium glutamicum Cultured on Glucose or Fructose. Appl. Environ. Microbiol. 70, 229±239 (2004).

16. Marx, A., de Graaf, A. A., Wiechert, W., Eggeling, L. & Sahm, H. Determination of the fluxes in the Central Metabolism of Corynebacterium glutamicum by nuclear Magnetic Resonance Spectroscopy combined with metabolite balancing. Biotechnol. Bioeng. 49, 111±129 (1996).

65 Chapter 2

17. Wittmann, C. & Heinzle, E. Genealogy Profiling through Strain Improvement by Using Metabolic Network Analysis: Metabolic Flux Genealogy of Several Generations of Lysine-Producing Corynebacteria. Appl. Environ. Microbiol. 68, 5843±5859 (2002).

18. Wittmann, C., Kiefer, P. & Zelder, O. Metabolic Fluxes in Corynebacterium glutamicum during Lysine Production with Sucrose as Carbon Source. Appl. Environ. Microbiol. 70, 7277±7287 (2004).

19. Gombert, A. K., Moreira dos Santos, M., Christensen, B. & Nielsen, J. Network Identification and Flux Quantification in the Central Metabolism of Saccharomyces cerevisiae under Different Conditions of Glucose Repression. J. Bacteriol. 183, 1441± 1451 (2001).

20. Van Winden, W. a et al. Metabolic-flux analysis of Saccharomyces cerevisiae CEN.PK113-7D based on mass isotopomer measurements of 13C-labeled primary metabolites. FEMS Yeast Res. 5, 559±68 (2005).

21. Alonso, A. P., Dale, V. L. & Shachar-Hill, Y. Understanding fatty acid synthesis in developing maize embryos using metabolic flux analysis. Metab. Eng. 12, 488±97 (2010).

22. Schwender, J. Metabolic flux analysis as a tool in metabolic engineering of plants. Curr. Opin. Biotechnol. 19, 131±137 (2008).

23. Goudar, C. et al. Metabolic flux analysis of CHO cells in perfusion culture by metabolite balancing and 2D [13C, 1H] COSY NMR spectroscopy. Metab. Eng. 12, 138±49 (2010).

24. Quek, L.-E., Dietmair, S., Krömer, J. O. & Nielsen, L. K. Metabolic flux analysis in mammalian cell culture. Metab. Eng. 12, 161±71 (2010).

25. Strigun, A. et al. Metabolic flux analysis gives an insight on verapamil induced changes in central metabolism of HL-1 cells. J. Biotechnol. 155, 299±307 (2011).

26. Metallo, C. M. et al. Reductive glutamine metabolism by IDH1 mediates lipogenesis under hypoxia. Nature 481, 380±4 (2012).

27. Crown, S. B., Ahn, W. S. & Antoniewicz, M. R. Rational design of 13C-labeling experiments for metabolic flux analysis in mammalian cells. BMC Syst. Biol. 6, 43 (2012).

28. Metallo, C. M., Walther, J. L. & Stephanopoulos, G. Evaluation of 13C isotopic tracers for metabolic flux analysis in mammalian cells. J. Biotechnol. 144, 167±174 (2009).

29. Nöh, K. et al. Metabolic flux analysis at ultra short time scale: isotopically non-stationary 13C labeling experiments. J. Biotechnol. 129, 249±67 (2007).

30. Wiechert, W. & Nöh, K. From Stationary to Instationary Metabolic Flux Analysis. Adv. Biochem. Eng. Biotechnol. 92, 145±172 (2005).

31. Noh, K. & Wiechert, W. Experimental Design Principles for Isotopically Instationary 13C Labeling Experiments. Biotechnol. Bioeng. 92, 234±251 (2006).

66 Chapter 2

32. Rühl, M., Zamboni, N. & Sauer, U. Dynamic flux responses in riboflavin overproducing Bacillus subtilis to increasing glucose limitation in fed-batch culture. Biotechnol. Bioeng. 105, 795±804 (2010).

33. Jules, M., Le Chat, L., Aymerich, S. & Le Coq, D. The Bacillus subtilis ywjI (glpX) gene encodes a class II fructose-1,6-bisphosphatase, functionally equivalent to the class III Fbp enzyme. J. Bacteriol. 191, 3168±71 (2009).

34. Link, H., Kochanowski, K. & Sauer, U. Systematic identification of allosteric protein- metabolite interactions that control enzyme activity in vivo. Nat. Biotechnol. 31, 357±61 (2013).

35. Buescher, J. M., Moco, S., Sauer, U. & Zamboni, N. Ultrahigh Performance Liquid Chromatography - Tandem Mass Spectrometry Method for Fast and Robust Quantification of Anionic and Aromatic Metabolites. Anal. Chem. 82, 4403±4412 (2010).

36. Rühl, M. et al. Collisional fragmentation of central carbon metabolites in LC-MS/MS increases precision of 13C metabolic flux analysis. Biotechnol. Bioeng. 109, 763±71 (2012).

37. Yuan, J., Bennett, B. D. & Rabinowitz, J. D. Kinetic flux profiling for quantitation of cellular metabolic fluxes. Nat. Protoc. 3, 1328±40 (2008).

38. Maiwald, T. & Timmer, J. Dynamical modeling and multi-experiment fitting with PottersWheel. Bioinformatics 24, 2037±2043 (2008).

39. Fishman, G. S. Monte Carlo concepts, algorithms, and applications. (Springer New York, 2003).

40. Fischer, E. & Sauer, U. Large-scale in vivo flux analysis shows rigidity and suboptimal performance of Bacillus subtilis metabolism. Nat. Genet. 37, 636±40 (2005).

41. Kleijn, R. J. et al. Metabolic fluxes during strong carbon catabolite repression by malate in Bacillus subtilis. J. Biol. Chem. 285, 1587±96 (2010).

42. Lerondel, G., Doan, T., Zamboni, N., Sauer, U. & Aymerich, S. YtsJ has the major physiological role of the four paralogous malic enzyme isoforms in Bacillus subtilis. J. Bacteriol. 188, 4727±36 (2006).

43. Noack, S., Nöh, K., Moch, M., Oldiges, M. & Wiechert, W. Stationary versus non- stationary 13C-MFA: a comparison using a consistent dataset. J. Biotechnol. 154, 179± 90 (2011).

44. Fischer, E. & Sauer, U. Metabolic flux profiling of Escherichia coli mutants in central carbon metabolism using GC-MS. Eur. J. Biochem. 270, 880±891 (2003).

45. Rühl, M., Le Coq, D., Aymerich, S. & Sauer, U. 13C-flux analysis reveals NADPH- balancing transhydrogenation cycles in stationary phase of nitrogen-starving Bacillus subtilis. J. Biol. Chem. 287, 27959±70 (2012).

46. Zamboni, N. 13C metabolic flux analysis in complex systems. Curr. Opin. Biotechnol. 22, 103±8 (2011).

67 Chapter 2

47. Möllney, M., Wiechert, W., Kownatzki, D. & de Graaf, a a. Bidirectional reaction steps in metabolic networks: IV. Optimal design of isotopomer labeling experiments. Biotechnol. Bioeng. 66, 86±103 (1999).

68 Chapter 2

Supplementary Figure

Supplementary Figure 1. All simulated CO2 kinetics (grey) and CO2 kinetics of statistically significant estimates (blue).

69 Chapter 2

Supplementary Table

Supplementary Table I. Model equations for the metabolic nodes considered in this study.

Phosphoenolpyruvate from glycolysis vs. gluconeogenesis

݀ܲܧܲ௠ା଴ ͳ ௠ା଴ሻܲܧܲ ଶଷସǡ௠ା଴ െ ሺݒாேை ൅ ݒ௉஼௄ሻ ήܲܵܣ ௠ା଴ ൅ ݒ௉஼௄ ήܩܲ ൌ ή ሺݒாேை ή ݐ ܿ௉ா௉݀

݀ܲܧܲ௠ାଵ ͳ ௠ାଵሻܲܧܲ ଶଷସǡ௠ାଵ െ ሺݒாேை ൅ ݒ௉஼௄ሻ ήܲܵܣ ௠ାଵ ൅ ݒ௉஼௄ ήܩܲ ൌ ή ሺݒாேை ή ݐ ܿ௉ா௉݀

݀ܲܧܲ௠ାଶ ͳ ௠ାଶሻܲܧܲ ଶଷସǡ௠ାଶ െ ሺݒாேை ൅ ݒ௉஼௄ሻ ήܲܵܣ ௠ାଶ ൅ ݒ௉஼௄ ήܩܲ ൌ ή ሺݒாேை ή ݐ ܿ௉ா௉݀

݀ܲܧܲ௠ାଷ ͳ ௠ାଷሻܲܧܲ ଶଷସǡ௠ାଷ െ ሺݒாேை ൅ ݒ௉஼௄ሻ ήܲܵܣ ௠ାଷ ൅ ݒ௉஼௄ ήܩܲ ൌ ή ሺݒாேை ή ݐ ܿ௉ா௉݀

Oxaloacetate from TCA cycle vs. anaplerosis

݀ܣܵܲ௠ା଴ ͳ ௠ା଴ሻܲܵܣ ௠ା଴ െ ሺݒெ஽ு ൅ ݒ௉௒஼ሻ ήʹܱܥ ௠ା଴ ൅ ݒ௉௒஼ ή ܻܴܲ௠ା଴ ήܮܣܯ ൌ ή ሺݒெ஽ு ή ݐ ܿ஺ௌ௉݀

݀ܣܵܲ௠ାଵ ͳ ௠ାଵ െ ሺݒெ஽ு ൅ ݒ௉௒஼ሻʹܱܥ ௠ା଴ ൅ ܻܴܲ௠ା଴ ήʹܱܥ ௠ାଵ ൅ ݒ௉௒஼ ή ሺܻܴܲ௠ାଵ ήܮܣܯ ൌ ή ሺݒெ஽ு ή ݐ ܿ஺ௌ௉݀

ή ܣܵܲ௠ାଵሻ

݀ܣܵܲ௠ାଶ ͳ ௠ାଵ െ ሺݒெ஽ு ൅ ݒ௉௒஼ሻʹܱܥ ௠ା଴ ൅ ܻܴܲ௠ାଵ ήʹܱܥ ௠ାଶ ൅ ݒ௉௒஼ ή ሺܻܴܲ௠ାଶ ήܮܣܯ ൌ ή ሺݒெ஽ு ή ݐ ܿ஺ௌ௉݀

ή ܣܵܲ௠ାଶሻ

݀ܣܵܲ௠ାଷ ͳ ௠ାଵ െ ሺݒெ஽ு ൅ ݒ௉௒஼ሻʹܱܥ ௠ା଴ ൅ ܻܴܲ௠ାଶ ήʹܱܥ ௠ାଷ ൅ ݒ௉௒஼ ή ሺܻܴܲ௠ାଷ ήܮܣܯ ൌ ή ሺݒெ஽ு ή ݐ ܿ஺ௌ௉݀

ή ܣܵܲ௠ାଷሻ

݀ܣܵܲ௠ାସ ͳ ௠ାସሻܲܵܣ ௠ାଵ െ ሺݒெ஽ு ൅ ݒ௉௒஼ሻ ήʹܱܥ ௠ାସ ൅ ݒ௉௒஼ ή ܻܴܲ௠ାଷ ήܮܣܯ ൌ ή ሺݒெ஽ு ή ݐ ܿ஺ௌ௉݀

Pyruvate through malic enzyme vs. glycolysis

ܻܴ݀ܲ௠ା଴ ͳ ଵ଴଴଴ሻ െ ሺݒ௉௒௄ ൅ ݒெ஺ாሻ ή ܻܴܲ௠ା଴ሻܮܣܯ ଴଴଴଴ ൅ܮܣܯ௠ା଴ ൅ ݒெ஺ா ή ሺܲܧܲ ൌ ή ሺݒ௉௒௄ ή ݐ ܿ௉௒ோ݀

ܻܴ݀ܲ௠ାଵ ͳ ܯܣܮ଴ଵ଴଴൅ܯܣܮ଴଴ଵ଴ ൅ ܯܣܮ଴଴଴ଵ ൅ ௠ାଵ൅ݒெ஺ா ή ൬ ൰ െ ሺݒ௉௒௄ ൅ ݒெ஺ாሻ ή ܻܴܲ௠ାଵሻܲܧܲ ൌ ή ሺݒ௉௒௄ ή ଵ଴଴ଵܮܣܯ ଵ଴ଵ଴ ൅ܮܣܯଵଵ଴଴ ൅ܮܣܯ ݐ ܿ௉௒ோ݀

ܻܴ݀ܲ௠ାଶ ͳ ܯܣܮ଴ଵ଴ଵ ൅ ܯܣܮ଴଴ଵଵ൅ܯܣܮ଴ଵଵ଴ ൅ ௠ାଶ൅ݒெ஺ா ή ൬ ൰ െ ሺݒ௉௒௄ ൅ ݒெ஺ாሻ ή ܻܴܲ௠ାଶሻܲܧܲ ൌ ή ሺݒ௉௒௄ ή ଵଵଵ଴ܮܣܯଵ଴ଵଵ ൅ܮܣܯ ଵଵ଴ଵ ൅ܮܣܯ ݐ ܿ௉௒ோ݀

ܻܴ݀ܲ௠ାଷ ͳ ଵଵଵଵሻ െ ሺݒ௉௒௄ ൅ ݒெ஺ாሻ ή ܻܴܲ௠ାଷሻܮܣܯ ଴ଵଵଵ ൅ܮܣܯ௠ାଷ ൅ ݒெ஺ா ή ሺܲܧܲ ൌ ή ሺݒ௉௒௄ ή ݐ ܿ௉௒ோ݀

70 Chapter 3

Non-stationary 13C-metabolic flux ratio analysis of embryonic

fibroblasts with impaired pyruvate metabolism

Manuel Hörl, Dimitris Christodoulou, Petra Krznar and Nicola Zamboni

Institute of Molecular Systems Biology, ETH Zürich, Zürich, Switzerland

Manuel Hörl designed the study, performed metabolomics and labeling experiments, conducted the computational analysis and wrote the manuscript. Dimitris Christodoulou contributed to the implementation of the computational framework in the SimBiology format.

Petra Krznar maintained cells in culture and contributed to metabolomics experiments. Nicola

Zamboni conceived and supervised the study.

71 Chapter 3

Abstract

13C metabolic flux analysis has become a key methodology to investigate metabolic fluxes in microorganisms growing in minimal medium with a single carbon source. Nevertheless, the application of current flux estimation approaches to higher cells under real-life conditions is still hampered by their more complex metabolism, which features compartmentation, requires multiple nutrients and is characterized by slow growth. Here, we investigate if our previously established non-stationary 13C metabolic flux ratio analysis has the potential to overcome the aforementioned challenges and can be rigorously applied to estimate relative fluxes also in mammalian cells. Therefore, we first estimated relative flux changes in mouse embryonic fibroblasts with deletion of the mitochondrial pyruvate carrier (MPC) by incorporating steady state labeling information from multiple experiments with specifically and uniformly labeled glucose and glutamine tracers. This analysis revealed that pyruvate transport into mitochondria was abolished upon MPC deletion, which was compensated by an increased relative contribution of partially anaplerotic, but mainly reductive TCA cycle flux. Using short-term dynamic 13C labeling data from [U-13C]glucose and [U-13C]glutamine experiments, we then evaluated if similar flux information could be extracted from the individual dynamic datasets and found the anticipated results that glucose alone provided good estimates of anaplerotic fluxes, while glutamine allowed to infer relative TCA cycle fluxes with good confidence.

However, only the combination of both datasets in the estimation process yielded results that were in good agreement with the results obtained by steady state analysis. In summary, our approach reduced the duration of the necessary labeling experiment from at least 12 to 2 hours and minimized the amount of required tracers. These results show that our strategy also enables to formally estimate relative pathway fluxes in the more complex metabolic networks of higher cells and might be used to validate specific flux changes also in such systems in extremely short time scales.

72 Chapter 3

Introduction

13C metabolic flux analysis (13C MFA) has emerged as a powerful tool for characterizing intracellular metabolic fluxes especially in microbes, where it has been extensively applied to investigate metabolism in the context of metabolic engineering1±4 and systems biology5,6.

Canonical (stationary) 13C MFA approaches require knowledge of the metabolic network, the measurement of all carbon uptake and product secretion rates, as well as labeling experiments with substrates enriched in 13C. The labeling experiments have to be performed at metabolic steady state, where intracellular metabolite concentrations, fluxes and the growth rate are invariant in time and have to be allowed to reach time-invariant labeling patterns in intracellular metabolites within the duration of the experiment (i.e. isotopic steady state)7,8. These requirements can be met with little effort for bacteria, which can be grown at metabolic steady state in minimal medium on a single carbon source for short periods that still allow to attain isotopic stationarity within all cellular constituents. Therefore, 13C MFA has become a routine tool to study central metabolism of microbial model-organisms over the past two decades8±10.

This progress was achieved mainly thanks to the possibility of deducing the labeling information of central metabolites from highly abundant proteinogenic amino acids11. However, recent advances in analytical techniques facilitate extracting this information also directly from metabolic intermediates12±14, and this paved the road to flux analysis in higher organisms.

The aforementioned analytical developments, together with an increasing number of manually and automatically curated, innovative labeling strategies15±17, made it also possible to quantitatively estimate certain fluxes with some approximations in higher cells. This allowed to quantify fluxes in cell lines important for biopharmaceutical production18 and provided unprecedented insights into basic functionality of metabolism in plants19 and human15 and metabolic alterations related to cancer20±23 and other diseases24.

However, a universal application of 13C-MFA to higher cells in the fashion of microbial 13C-MFA is still far from being practical because of several challenges. First, eukaryotic cells are compartmented, which hampers the interpretation of labeling patterns due to the co-existence of several metabolite pools within one cell. These pools cannot be individually extracted with

73 Chapter 3 current methodologies and the only way to distinguish them is by the existence of compartment-specific reactions that lead to distinct 13C signatures25,26. Second, the requirement of multiple nutrients, together with the higher complexity of mammalian metabolic networks does not allow for obtaining a distinct flux solution from a single labeling experiment15,27. Third, maintaining a steady growth until reaching isotopic steady state is challenging, since higher cells feature slow doubling times and transient growth under-real-life conditions.

An alternative strategy to stationary 13C MFA that could potentially overcome the current limitations of higher cell flux analysis is to employ the dynamics of 13C propagation for flux estimation. This so-called non-stationary 13C MFA drastically reduces the temporal duration of labeling experiments and theoretically also has the power to calculate metabolite concentrations within compartments28,29. Furthermore, the incorporation of several, time- resolved measurements instead of a single data point in the isotopically equilibrated state yields more information30, which might reduce the number of tracer experiments that need to be performed. On the downside, non-stationary 13C-MFA additionally requires knowledge of most intracellular metabolite concentrations and access to high performance computing to estimate fluxes from large ordinary differential equation (ODE) systems.

We previously introduced non-stationary 13C metabolic flux ratio analysis31 (13C MFRA) for the estimation of local, relative fluxes as an alternative to traditional non-stationary 13C MFA. To avoid the aforementioned challenges, this method relies on the labeling dynamics of only few metabolites around a metabolic converging node and parameter estimation with small, computationally less demanding ODE models. Additionally, the methodology is independent of measuring extracellular uptake and secretion rates and does not imply knowledge of intracellular metabolite concentrations.

Although we demonstrated the validity of non-stationary 13C-MFRA for the estimation of relative fluxes in bacterial metabolism in Chapter 231, an investigation of its performance with higher cell metabolism is still missing. In this study, we addressed this point by comparing relative flux estimates obtained with steady state data and non-stationary 13C MFRA. Specifically, we

74 Chapter 3 compared mouse embryonic fibroblasts (MEFs) wild-type (WT) and cells with deletion of the mitochondrial pyruvate carrier (Mpc1KO), where distinct metabolic flux rearrangements are anticipated32. For both cell lines, we performed non-stationary 13C labeling experiments with glucose and glutamine as tracers until isotopic steady state was reached and used this consistent dataset for both types of analysis. In a thorough investigation of relative fluxes in the formation of aspartate, we show that deletion of the MPC completely abolishes mitochondrial pyruvate transport, which was compensated by increased contribution of the reductive TCA cycle and a hitherto unknown anaplerotic flux. Furthermore, we evaluated what information could be extracted from the dynamic labeling data of individual tracer experiments and show that the incorporation of several datasets allowed to correctly predict relative fluxes from dynamic, short-term labeling data.

75 Chapter 3

Materials and Methods

Cell lines and cultivation

Mouse embryonic fibroblasts were derived from homozygous Mpc1 +/+ (WT) and -/- (Mpc1KO) mouse embryos in the lab of J.-C. Martinou. &HOOOLQHVZHUHFXOWXUHGLQ'XOEHFFR¶VPRGLILHG

(DJOH¶V PHGLXP '0(0  ZLWK  GLDO\]HG IHWDO ERYLQH VHUXP )%6  ERWK IURP /LIH

Technologies) at physiological oxygen level (21 %) and 5 % CO2 in a humidified incubator at

37 °C. For tracer and MFA studies, DMEM was formulated by replacing the substrate of interest with 13C labeled glucose or glutamine (both from Cambridge Isotopes) with other components unlabeled. Cultures were washed with PBS before adding tracer media.

Metabolomics and 13C labeling experiments

For metabolomics and 13C labeling experiments, 6 to 8 x 104 cells/well (ca. 50 % confluency) were seeded in a 6 well-plate (Nunc) and allowed to attach for ca. 12 h in the presence of unlabeled DMEM. Then medium was thoroughly removed, cells washed 1x with PBS and fresh

DMEM added. For metabolomics experiments, cells were incubated for at least another 24 h until metabolite sampling. For labeling experiments, DMEM containing the 13C labeled substrate was added after washing with PBS. The labeling time course was performed by incubating cells with the labeling medium for the duration of 1, 5, 10, 20, 30, 45 and 60 min and 2, 4, 8, 12, 18, and 24 h. For each condition and time point three replicates were collected.

Metabolite extraction and LC/MS analysis

At the conclusion of a metabolomics or tracer experiment, the media was removed from the culture wells, the cells were washed twice with 75 mM ammonium carbonate buffer (pH 7.4), and quenched by shock freezing the plate in liquid N2. Plates were stored at -80°C until metabolite extraction.

Metabolites were extracted by adding 1.8 mL acetonitrile/methanol/water (2:2:1) (-20°C) to each well and, in the case of metabolite quantification experiments, 200 µL of a uniformly 13C labeled E. coli metabolite extract was added as internal standard33. For pyruvate measurements, the acetonitrile/methanol/water mixture contained 25 µM phenylhydrazine for

76 Chapter 3 derivatization and 100 µL of a 5 µM [U-13C]pyruvate solution was added as internal standard.

After 1 h incubation at -20°C, the bottom of each well was scraped with a 1 ml pipette tip and the extract was collected in a 2 ml tube. To remove the cell debris, tubes were centrifuged

(4°C, 10,000 rpm, 10 min), the supernatant was transferred to a new tube and dried to complete dryness. The remaining cell debris pellets were incubated with CelLytic lysis reagent

(Sigma) and used to quantify the individual protein content with a Bradford assay34, in order to normalize metabolite concentrations to cellular protein.

For LC/MS analysis, dried extracts were resuspended in 100 µL deionized water of which 10 µL were injected into a Waters Acquity UPLC (Waters Corporation, Milford, MA) with a Waters

Acquity T3 column coupled to a Thermo TSQ Quantum Ultra triple quadrupole instrument

(Thermo Fisher Scientific) with negative-mode electrospray ionization. Compound separation was achieved by a gradient of two mobile phases (A) 10 mM tributylamine, 15 mM acetic acid, and 5 % (v/v) methanol and (B) 2-propanol35,36. Acquisition of mass isotopomer distributions

(MIDs) of intact and fragmented carbon backbones was done as previously described37. Peak integration was performed by an in-house software (Begemann and Zamboni, unpublished).

Data processing and parameter estimation

Metabolite concentrations were calculated by normalizing the peak area of each compound to the respective signal from the internal 13C standard and comparing these numbers to the ratios obtained from a calibration curve with known metabolite concentrations. MIDs and fractional labeling were calculated from measured peak areas as previously described38 and corrected for naturally occurring 13C39. Average and standard deviations of the measured mass fractions were calculated using three biological replicates.

Procedures of parameter estimation, model construction and statistical analysis with non- stationary 13C MFRA are explained in great detail in Chapter 2 of this thesis31. Briefly, flux ratios and product pool sizes were estimated by weighted least square fitting of simulated to measured non-stationary MIDs with ODE models implemented in the SimBiology toolbox format and Matlab 2014b. A total of 10,000 fittings were performed for each condition, where random noise within the experimentally determined standard deviations was added to the

77 Chapter 3 dynamics of input metabolite MIDs at each iteration. Because of the randomized noising, iterations frequently lead to a transformation of the input data to biologically implausible trajectories. We previously rejected these non-significant estimates based on weighted sum of squared residuals values Ȥ2) with significance level below 0.05. Since no estimate met this

VWULFWFXWRIIZHXVHGWKHGLVWULEXWLRQRIȤ2-values of each fitting round to define a new cutoff.

Specifically, we calculated the mean of the estimated Ȥ2-distribution for each round of 10,000 iterations and considered only those estimates with a Ȥ2-value below an arbitrarily chosen cutoff of three standard deviations from the mean as statistically significant. Only these estimates were used to calculate the mean and standard deviation of the flux ratio. The procedure is also described in the results section.

78 Chapter 3

Results

Metabolite changes upon Mpc1 deletion

To investigate how the metabolism of embryonic fibroblasts is affected upon deletion of the

MPC, we first performed a metabolomics analysis of intracellular metabolites with WT and

Mpc1KO. Therefore, the polar metabolome was extracted from growing MEFs after 24h of cultivation. In accordance to previously reported results for other cell lines with MPC knockdown32, intracellular pyruvate levels were ~8 times higher in Mpc1KO (Fig.1), indicating that the loss of mitochondrial pyruvate transport lead to build-up of pyruvate. This accumulation of pyruvate seemed to spread through lower glycolysis, since also phosphoenolpyruvate and

2/3-phosphoglycerate levels were increased in Mpc1KO. A second group of metabolites also showing increased levels were aspartate and the connected TCA cycle intermediates malate and fumarate. This most likely indicates that the citrate synthase reaction is drastically decreased in Mpc1KO due to the loss of pyruvate transport into the mitochondria and concomitant shortage of acetyl-CoA (usually produced from pyruvate in the mitochondria), which leads to a built up of oxaloacetate (OAA), the second substrate of this reaction. This hypothesis is additionally supported by the fact that citrate levels were drastically decreased in Mpc1KO.

Figure 1. Mpc1 knockout leads to changes in metabolites around reactions where pyruvate is involved. Relative metabolite abundance in Mpc1KO compared to WT measured by targeted LC-MS/MS analysis35,36.

79 Chapter 3

Stationary flux analysis of embryonic fibroblasts

To identify variations in metabolic operation that lead to the observed metabolite changes, we assessed the metabolic fate of the two main carbon sources glucose and glutamine in central metabolism of WT and Mpc1KO cells. We measured the steady state isotopic labeling patterns of free intracellular metabolites in central carbon metabolism after 24 h of cultivation with either

[U-13C]glucose or [U-13C]glutamine by targeted LC-MS/MS analysis36,37. The main pathways potentially involved in mitochondrial glucose and glutamine metabolism are illustrated in Fig. 2.

The use of uniformly labeled substrates allows assessing their total contribution to any detectable intracellular metabolite. This relative amount is recapitulated by the fractional labeling. Consistent with the expected reduction of glucose-derived pyruvate entering the mitochondria in Mpc1KO, the fractional labeling of TCA-cycle intermediates was significantly decreased with [U-13C]glucose as substrate (Fig. 3a). In the presence of [U-13C]glutamine,

TCA-cycle intermediates and connected amino acids aspartate and glutamate were significantly enriched in labeling in Mpc1KO (Fig. 3b), indicating increased glutamine incorporation upon restriction of mitochondrial pyruvate transport. Noteworthy, the sum of glucose and glutamine derived fractional labeling in TCA cycle metabolites, e.g. malate or aspartate, was between 90 and 100 % in WT, while in Mpc1KO fractional labeling was ca. 5% lower and originated from different, unlabeled substrates. Most likely, these sources are amino acids and fatty acids, which are present in the medium and have been previously shown to support mitochondrial metabolism upon Mpc deletion32.

To gain additional insight into TCA cycle metabolism, we focused on the mass isotopomer distribution of the TCA cycle intermediates succinate, malate and aspartate (as approximation for its precursor OAA) (Fig. 3c-h). In WT, [U-13C]glucose-derived pyruvate enters the mitochondria and is incorporated into the TCA cycle via (PDH) as acetyl-CoA (Fig. 2), resulting in an enrichment of the m2 fraction (Supplementary Fig. 1). While this fraction was present in WT, Mpc1KO showed a significant decrease, confirming the impaired pyruvate transport (Fig. 3c-e).

80 Chapter 3

Figure 2. Overview of potential pathways involved in glucose and glutamine catabolism in MEFs. Notably, the initial steps of the reductive TCA-cycle are depicted as mitochondrial, while also cytosolic isoenzymes of the pathway exist. Therefore, the pathway might also run outside the mitochondria.

Surprisingly, in Mpc1KO cells both malate and aspartate still showed significant m3 labeling

(Fig. 3d, e), which was not derived from the operation of the oxidative TCA cycle, indicated by the complete absence of this fraction in succinate (Fig. 3c). Normally, the m3 isotopomer in

[U-13C]glucose experiments reflects the anaplerotic carboxylation of pyruvate to OAA.

However, the only known anaplerotic enzyme pyruvate carboxylase localizes in the mitochondrion40. In a Mpc1KO background, the contribution of this reaction should be abolished as much as the m2 fraction because no [U-13C]pyruvate can enter mitochondria. The same holds true if the accumulation of pyruvate in the cytosol would promote its passive diffusion into the mitochondria. Alternatively, it might be that the build-up of pyruvate in the cytosol promotes the reversal of a normally gluconeogenic decarboxylation reaction. Potential candidates are the reactions catalyzed by PEP carboxykinase Pck1, which might perform the carboxylation of PEP into OAA or the cytosolic malic enzyme ME1, which could carboxylate pyruvate and generate malate. The reversal of PEP carboxykinase flux has already been

81 Chapter 3 shown for certain bacterial mutants, where a strong increase in PEP concentrations modified the reaction thermodynamics and facilitated a reversal of flux41. However, the ratio between aspartate (as approximation for OAA) and PEP remained constant in Mpc1KO, so that reversed Pck1 activity seems improbable. Therefore, we speculated that a reversal of the cytosolic malic enzyme reaction, favored by the drastic increase in pyruvate levels, caused the increase in malate and aspartate m3.

To verify that the m3 enrichment indeed originated through the anaplerotic reaction, we used positionally enriched 13C-glucose. Specifically, we cultivated WT and Mpc1KO with

13 13 13 [3,4- C2]glucose, which in glycolysis is converted to [1- C]pyruvate. C at position C-1 of

13 pyruvate is lost as CO2 if pyruvate is converted into acetyl-CoA through the pyruvate dehydrogenase, but remains if pyruvate is metabolized anaplerotically (Supplementary

Fig. 2a). A high activity of alternative glucose catabolizing pathways such as the PPP could

13 also lead to the formation of m0 and m2 pyruvate from [3,4- C2]glucose, which we excluded by showing that intermediates of lower glycolysis, e.g. 1,3-biphospho glycerate, contained exclusively m1 (Supplementary Fig. 2c). We found that the fraction of anaplerotically generated aspartate and malate was slightly, but significantly elevated in the Mpc1KO (Fig. 3i) and matched the fraction of [U-13C]glucose-derived labeling of these metabolites (Fig 3a).

Analyzing the isotopic enrichment of succinate, malate and aspartate with [U-13C]glutamine as substrate, the m4 fraction was the most abundant in all three metabolites (Fig. 3f-h). This fraction originates via flux through the oxidative branch of the TCA cycle

(Supplementary Fig.3). Although glutamine-derived malate and aspartate were elevated in the

Mpc1KO (Fig. 3b), the m4 fraction of these metabolites showed no increase compared to WT, indicating no compensation in their formation by increased contribution of the oxidative TCA cycle. In contrast, both metabolites exhibited significantly increased m3 fractions, suggesting formation via reductive glutamine metabolism (Fig. 2; Supplementary Fig. 3). To test this hypothesis, we grew cells in the presence of [1-13C]glutamine, which allows to distinguish oxidative from reductive TCA-cycle activity (Supplementary Fig. 4). Indeed, the fraction of aspartate and malate originating via reductive mode of the TCA cycle was significantly

82 Chapter 3 increased in Mpc1KO (Fig. 3j) and perfectly matched the aforementioned increase in the m3 fraction observed with [U-13C]glutamine feeding.

Figure 3. Increased glutamine incorporation supports the TCA-cycle upon loss of the MPC. (a and b) Fractional labeling of TCA intermediates glutamate, succinate, malate and aspartate resulting from culture with (a) [U- 13C]glucose (UGlc) and (b) [U-13C]glutamine (UGln). (c, d and e) Mass isotopomer distribution of (c) succinate, (d) malate and (e) aspartate from UGlc. (f, g and h) Mass isotopomer distribution of (f) succinate, (g) malate and (h) 13 aspartate from UGln. (i) Percentage of anaplerotically derived aspartate and malate, using [3,4- C2]glucose. P <  3DLUHG6WXGHQW¶V7-test, unequal variance). (j) Contribution of reductive TCA cycle to aspartate, malate and succinate estimated by feeding [1-13C]glutamine.

To also get a quantitative understanding on the flux rerouting upon MPC deletion, we used the steady state labeling information in aspartate from all four experiments and calculated relative metabolic flux ratios (Fig. 4, Table 1). Specifically, the relative contribution of oxidative vs. reductive TCA cycle can be estimated by subtracting [1-13C]glutamine derived m1 from the fractional labeling originating from [U-13C]glutamine. Similarly, subtraction of the percentage of

13 13 [3,4- C2]glucose-derived aspartate m1 from [U- C]glucose-derived fractional labeling of

83 Chapter 3 aspartate allows to calculate the relative flux through anaplerosis vs. PDH. In summary, Mpc1 deletion prevented pyruvate metabolism through PDH, which was mainly circumvented by an increase in the relative reductive TCA cycle flux (8% in WT vs. 18% in Mpc1KO), while relative flux through the oxidative branch was decreased (89% in WT vs. 76% in Mpc1KO). However, the block of pyruvate transport was also partially compensated by incorporation of glucose- derived carbon through an anaplerotic reaction, resulting in a shift of the relative contribution of anaplerosis vs. PDH from 18% in WT to 91% in Mpc1KO (Fig. 4a). Relative to both TCA cycle fluxes, the contribution to aspartate formation by anaplerosis was 3% in WT and 6% in

Mpc1KO (Fig. 4b).

Collectively, the stationary labeling with differently 13C-enriched glucose and glutamine tracers allowed to identify that the loss of the MPC in mouse embryonic fibroblasts completely abolished the transport of glucose derived pyruvate into the mitochondria. This was compensated by the increase in the relative reductive TCA cycle flux and an anaplerotic flux, which we hypothesize to be facilitated by a reversal of the malic enzyme reaction, catalyzed by cytosolic ME1.

Figure 4. MPC knockout is mainly compensated by increased reductive TCA cycle. (a) Relative flux of anaplerosis vs. PDH and (b) relative fluxes of anaplerosis, oxidative (ox TCA) and reductive (red TCA) TCA cycle in aspartate formation in WT and Mpc1KO.

84 Chapter 3

Non-stationary 13C labeling experiment

While the stationary labeling experiments provided comprehensive information on the origin of aspartate, they required the use of several tracers to resolve all relative fluxes and labeling experiments of at least 12 hours to reach isotopic steady state. Non-stationary 13C approaches have the potential to drastically shorten the experiment duration by incorporating the initial label propagation after adding 13C tracer for flux estimation. Additionally, the dynamic data contain more information than a single measurement at isotopic steady state, and thus may reduce the amount of required tracers. Therefore, we set out to investigate if similar flux estimates could be obtained using non-stationary 13C metabolic flux ratio analysis. To obtain

13C labeling dynamics of free intracellular metabolites, we cultured both WT and Mpc1KO in the presence of [U-13C]glucose and [U-13C]glutamine for varying durations. Specifically, we collected up to 13 time points between 1 minute and 24 hours and measured the MIDs of intracellular metabolites at each time point.

Cultivation with [U-13C]glucose resulted in both cell lines in rapid labeling of glycolytic intermediates, which reached steady state already within 5 to 15 minutes (Fig. 5). An exception was pyruvate, which reached steady state only after 30 minutes in WT (Fig. 6a) and 50 minutes in MPC1KO (Fig. 6b). The slightly slower pyruvate labeling dynamics in Mpc1KO are most reasonably explained by the drastic intracellular accumulation of pyruvate upon Mpc deletion.

In WT, dynamics of most intermediates in the TCA cycle were dramatically slower than in glycolysis and reached steady state after roughly 8 to 12 hours (Fig. 6c, e, g). Similar results were already observed for bacterial metabolism28 and might be explained by high exchange fluxes of (unlabeled) amino acids from the medium and subsequent transfer of amino groups between these amino acids DQG Į-keto acids of the TCA cycle (e.g. Į-ketoglutarate and

OAA)28,42. In Mpc1KO, most TCA intermediates GLGQ¶WUHSRUW DQ\ measureable 13C labeling from glucose (Fig. 6d, f). Only aspartate (Fig. 6h) and malate (not shown) showed a slight increase in the m3 trace over time, again indicating the activity of an anaplerotic reaction.

Supplementation with [U-13C]glutamine resulted in labeling dynamics of TCA cycle intermediates at similar time scales as with [U-13C]glucose tracing (Fig. 7c-h), but no labeling

85 Chapter 3 was detectable in pyruvate and other glycolytic intermediates (Fig. 7a, b). Notably, the immediate and simultaneous increase of m3 and m4 in aspartate of Mpc1KO (Fig. 7h) confirmed our previous hypothesis of simultaneous flux through the oxidative and the reductive branch of the TCA cycle and proved that m3 labeling in aspartate was not just an artifact due

13 to e.g. CO2 loss by multiple cycling of molecules in the TCA cycle.

In summary, the labeling dynamics qualitatively confirmed our two hypothesis that (i) glucose carbon was partially incorporated into aspartate and malate through an anaplerotic reaction in

Mpc1KO and that (ii) Mpc1KO exhibited an increased relative flux through the reductive TCA cycle. The fact that [U-13C]glutamine labeling did not lead to any labeling of glycolytic intermediates indicates that no gluconeogenic reactions were appreciably active in any of the cell lines.

Figure 5. 13C labeling dynamics of glycolytic intermediates glucose-6-phosphate (G6P) and 1,3-biphosphoglycerate

(BPG) and 24 h steady state labeling in (a, b) WT and (c, d) Mpc1KO upon feeding [U-13C]glucose.

86 Chapter 3

Non-stationary 13C MFRA of relative fluxes in aspartate formation

To quantify the relative contribution of anaplerosis, and the oxidative and reductive TCA cycle to the formation of aspartate (and thus OAA) from dynamic labeling data, we used non- stationary 13C MFRA. To describe the influence of anaplerosis on aspartate labeling, we used pyruvate labeling dynamics, while succinate dynamics were used to reflect the oxidative TCA cycle. Since our labeling measurements of the direct pathway substrate citrate were error- prone (data not shown), we used glutamate to approximate the reductive TCA cycle contribution. We described the influence of the three pathways on aspartate labeling by small

ODE models, both for the [U-13C]glucose and [U-13C]glutamine tracer experiment, as detailed in Supplementary Table 1. We assumed fast equilibrium between cytosolic and mitochondrial pools and therefore neglected cellular compartmentation in the model. The formation of aspartate via anaplerosis occurs through pyruvate and the oxidative TCA cycle proceeds via succinate. Both reactions result in no rearrangements of the carbon backbone, so that each individual mass isotopomer will lead to the same mass isotopomer in aspartate and can be directly employed for the fitting. In the case of anaplerosis, we assumed that CO2 incorporated by the carboxylation reaction contained only 12C. For aspartate production from glutamate via the reductive TCA cycle, carbon backbone rearrangements have to be considered

(Supplementary Fig. 3). While the m0 isotopomer of glutamate generates m0 in aspartate

13 independent of the C tracer, the carboxylation with unlabeled CO2 will introduce a change in the labeling patterns, which additionally depends on the tracer. Specifically, in the case of

[U-13C]glucose, glutamate m2 originates from m2 acetyl-CoA, which is cleaved again in the reductive TCA cycle by the cytosolic ATP citrate lyase (ACL), so that glutamate m2 will lead to m0 in aspartate. With [U-13C]glutamine as tracer, reductive glutamate m5 carboxylation will result in aspartate m3, because of the incorporation of unlabeled CO2, which remains in aspartate upon cleavage of acetyl-CoA by ACL (Supplementary Fig. 3). Similarly, glutamate m3 will lead to aspartate m3, since glutamate m3 mainly originates through unlabeled, glucose- derived acetyl-CoA.

87 Chapter 3

Figure 6. 13C labeling dynamics and 24 h steady state labeling upon [U-13C]glucose feeding. (a, b) Pyruvate,

(c, d) glutamate, (e, f) succinate and (g, h) aspartate in WT (a, c, e, g) and Mpc1KO (b, d, f, h). The labeling

dynamics until 120 min were used for non-stationary 13C metabolic flux ratio analysis.

88 Chapter 3

Figure 7. 13C labeling dynamics and 24 h steady state labeling upon [U-13C]glutamine feeding. (a, b) Pyruvate,

(c, d) glutamate, (e, f) succinate and (g, h) aspartate in WT (a, c, e, g) and Mpc1KO (b, d, f, h). The labeling dynamics until 120 min were used for non-stationary 13C metabolic flux ratio analysis.

89 Chapter 3

The labeling dynamics of all considered metabolites showed fast transients within the first two hours of labeling and subsequent slower enrichment until reaching steady state after roughly

8 to 12 hours. Therefore, we only considered the labeling data of the first two hours for the parameter fitting process. To assess the information content of each individual labeling experiment, we performed the parameter estimation with only [U-13C]glucose data, only

[U-13C]glutamine data, and both datasets combined. For each condition, we performed 10,000 independent fittings, where the mass isotopomer dynamics of pyruvate, succinate and glutamate were first perturbed by a Monte Carlo approach to incorporate experimental measurement errors and then used as input data to fit the labeling dynamics of aspartate with concomitant estimation of the flux ratio. As explained LQWKHPHWKRGVZHXVHGWKHȤ2-value of each fit to identify the set of statistically most significant flux ratio estimates. Specifically, we considered only those estimates, which were three standard deviations lower than the mean

RIWKHGLVWULEXWLRQRIDOOȤ2-values, as significant.

With only [U-13C]glucose labeling data, the estimated relative contribution of anaplersois vs. oxidative vs. reductive TCA cycle was 2:87:11% in WT and 5:0:94% in Mpc1KO (Fig. 8 &

Table 1). For the WT the estimate was in agreement with values obtained from the stationary analysis (Fig. 8a, c, e), however, estimates for relative TCA cycle fluxes had only poor confidence intervals of ca. ±10% (see Table 1). For Mpc1KO, the estimates of relative TCA cycle fluxes diverged quite significantly from the stationary estimates (Fig. 8b, d). Notably, the estimated relative contribution of anaplerosis perfectly matched the previous estimates for both cell lines (Fig. 8e, f). For Mpc1KO, we anticipated such uncertain results for the relative TCA cycle fluxes with [U-13C]glucose as tracer, because labeling in the metabolites considered for the fitting is virtually absent and leaves the model with too many possible ratios that explain the data. Although estimates for the WT were less confident than with the stationary analysis, we were surprised that the relatively low incorporation of glucose in TCA cycle intermediates of ca. 20% (Fig. 3a) seemed to provide sufficient information to still obtain correct estimates.

90 Chapter 3

Table 1. Ratios of converging fluxes in the formation of aspartate (Asp) obtained with different 13C flux ratio methods and the incorporation of different dynamic and stationary labeling datasets

Method

Stationary MFRA Non-stationary MFRA

Model Flux ratio Dataset used WT Mpc1KO WT Mpc1KO

UGlc - - 87 (76;97) 0 (0;8) Asp from UGln - - 88 (82;89) 70 (55;72) ox TCA UGlc & UGln 89a,c (86;92)b 76 (74;78) 87 (86;88) 69 (68;70)

UGlc - - 11 (0;23) 94 (87;96) Asp from Ana UGln - - 12 (10;12) 29 (18;30) red TCA UGlc & UGln 8 (5;11) 18 (16;20) 11 (11;12) 21 (21;28)

UGlc - - 2 (1;3) 5 (4;7) Asp from UGln - - 0 (0;8) 0 (0;27) ana UGlc & UGln 3 (0;6) 6 (4;8) 2 (1;2) 9 (4;10)

a median of flux estimates and b lower and upper boundaries of the 68% confidence interval. c 13 13 To estimate this ratio with stationary MFRA, additional information with [1- C]glutamine and [3,4- C2]glucose labeling was required. Ana, anaplerosis; ox TCA, oxidative TCA cycle; red TCA, reductive TCA cycle.

Using only [U-13C]glutamine data, the TCA cycle flux ratio estimates for the WT pretty much resembled the values obtained from the fitting with [U-13C]glucose data (Fig. 8a-d), while the relative anaplerotic flux was estimated with very poor confidence (Fig. 8e, f). The estimates for relative TCA cycle contribution in Mpc1KO were closer to the estimates obtained using stationary data, but were still a bit higher and had relatively poor confidence intervals, while estimates for anaplerosis were just as poor as for WT (Fig. 8b, d, f). These results confirmed our expectation that the higher label incorporation of glutamine in TCA cycle metabolites would translate into a higher information content on relative TCA cycle fluxes, indicated by the narrowed confidence interval for the WT and by more reasonable estimates for Mpc1KO.

While the parameter fitting with the individual tracers provided acceptable estimates only for partial fluxes, performing the analysis with both datasets combined allowed to obtain values for the relative contribution of anaplerosis, oxidative and reductive TCA cycle that were almost identical to the steady state results both for WT and Mpc1KO (Fig. 8). From these results, we concluded that non-stationary 13C MFRA has, in principal, the potential to estimate relative flux

91 Chapter 3 estimates with good confidence also within the more complex metabolism of higher cells.

However, the prerequisite is that dynamic labeling data is obtainable for all substrates that participate in the flux ratio that is to be estimated.

Figure 8. Distribution of statistically significant estimates of relative flux through (a, b) oxidative TCA cycle, (c, d) reductive TCA cycle and (e, f) anaplerosis, estimated with the individual [U-13C]glucose and [U-13C]glutamine datasets and their combination ((a, c, e) WT; (b, d, f) Mpc1KO). The median of each distribution is indicated by a red line and the edges of each box represent the 68% confidence interval.

92 Chapter 3

Discussion

We characterized the metabolic phenotype of mouse embryonic fibroblasts upon deletion of the mitochondrial pyruvate transporter by targeted metabolomics and stationary and non- stationary 13C flux methods. Using stationary 13C labeling data from several experiments with specifically and uniformly labeled glucose and glutamine tracers, we could find that pyruvate transport was absent upon deletion of MPC, which was indicated by a strong decrease in the relative contribution of PDH and subsequently citrate synthase in the formation of TCA cycle metabolites. This was consistent with the observations from metabolomics measurements, which revealed a drastic increase in pyruvate levels and organic acids of the oxidative TCA cycle branch in Mpc1KO, together with strongly decreased citrate levels. Through labeling experiments with glucose, we detected slightly increased incorporation of glucose-derived pyruvate into aspartate and malate in Mpc1KO via a presumed anaplerotic reaction. However, the impaired pyruvate transport into mitochondria was mainly compensated by an elevated incorporation of glutamine carbon into TCA cycle metabolites. This increase was mainly due to an increased contribution of the reductive TCA cycle. Relative flux estimates using dynamic labeling data and non-stationary 13C metabolic flux ratio analysis confirmed that compared to

WT, aspartate formation via anaplerosis and the reductive TCA cycle was increased in

Mpc1KO to compensate the loss of mitochondrial pyruvate transport.

Collectively, the combination of stationary and non-stationary 13C flux ratio analysis revealed novel responses of mammalian cells towards disruption of mitochondrial pyruvate transport.

First, a hitherto uncharacterized cytosolic anaplerotic reaction seems to convert glucose derived pyruvate into OAA or malate. We excluded the activity of pyruvate carboxylase, the only known anaplerotic enzyme in mammalian cells, since it is located within the mitochondria.

Thus, the anaplerotic flux seems to be attributed to a reversal of a normally gluconeogenic reaction in the cytosol. Due to the elevated pyruvate levels in the cytosol in Mpc1KO, we speculate that this might have caused the reversal of cytosolic malic enzyme ME1. Although the malic enzyme reaction is generally considered to be reversible, to our knowledge this would be the first study to show the reverse reaction in vivo in mammalian metabolism. One might

93 Chapter 3 speculate that this anaplerotic flux could be a way to still allow shuttling of glucose-derived carbon into the mitochondria upon MPC loss, however, we found no evidence for this, since mitochondrial metabolites, such as succinate, showed no labeling in the characteristic isotopomer traces. To confirm our hypothesis on malic enzyme, the effect of simultaneous deletion of Mpc1 and Me1 on aspartate formation has to be investigated.

Second, the increased contribution of the reductive TCA cycle mainly compensated for the lack of mitochondrial pyruvate. This pathway is usually active to contribute acetyl-CoA for lipogenesis in cells proliferating under hypoxia or those with a compromised respiratory chain43±45, but has not been linked to impaired mitochondrial pyruvate transport. On the contrary, previous studies with C2C12 myoblasts found that knockdown and inhibition of the

MPC did not affect reductive TCA cycle activity, but was rather compensated by oxidative glutaminolysis, i.e. acetyl-CoA formation via the combined activity of mitochondrial malic enzyme and PDH32. Unfortunately, we could not directly estimate the activity of this pathway in MEFs upon deletion of Mpc1, since this information is derived from citrate labeling patterns upon [U-13C]glutamine labeling, which were error prone in our measurements. However, the increased contribution of reductive TCA cycle in Mpc1KO raises the question how important is the MPC for respiratory activity. Previous studies gave no clear answer to this question, since it was both shown that oxygen consumption was unaffected in C2C12 myoblasts upon knock- down of Mpc32, but increased when MPC was overexpressed in naturally low MPC expressing colon cancer cell lines46. Therefore, we can only speculate that the increased contribution of reductive TCA cycle in MEFs might be an indicator for impaired respiration resulting from deletion of Mpc1 in this cell line, which apparently has to be investigated in the future, by measuring pyruvate oxidation in WT and Mpc1KO.

From a methodological viewpoint, this is the first study that successfully shows that non- stationary 13C MFRA allows to provide relative flux estimates from short-term labeling data in the complex metabolism of higher cells. Incorporating dynamic labeling information from

[U-13C]glucose and [U-13C]glutamine labeling experiments of 2 hours, we were able to obtain similar results as with stationary labeling information of 24 hours, which required labeling with

94 Chapter 3

13 13 13 13 [3,4- C2]glucose, [U- C]glucose, [1- C]glutamine and [U- C]glutamine to fully resolve the same relative fluxes. We found that already the dynamic labeling information with the individual tracers allowed to get satisfactory results on relative fluxes connected to the labeled tracer, i.e. anaplerotic flux with [U-13C]glucose and relative oxidative vs. reductive TCA cycle fluxes with

[U-13C]glutamine. However, only incorporation of both datasets in the parameter estimation process allowed resolving all fluxes at once with good confidence.

Several recent publications have emphasized the importance of considering cellular compartments, especially with non-stationary 13C MFA approaches42,47,48. Therefore, we were surprised that our approach allowed to correctly estimate relative fluxes despite neglecting any compartmentalization. Since the considered central metabolic fluxes are relatively high, which should feature fast transport of metabolites across compartments, we did not expect any issues due to missing transport reactions in our models. However, we anticipated that the mixing of the separate pools during the extraction might lead to falsification of the compartment specific labeling dynamics. The fact that we did not observe such falsification, indicates that our assumption on fast equilibrating mitochondrial and cytosolic pools of aspartate was justified, since this would result in similar labeling dynamics of both pools, but still retain their characteristic labeling dynamics. However, this cannot be assumed for all compartment specific fluxes and future applications of non-stationary 13C MFA to fluxes where no steady state information for validation of the estimates can be obtained should both consider compartmentalized pools, as well as labeling dynamics. Because this cannot be achieved experimentally hitherto, one solution could be to randomly split the measured dynamics into two pools in silico and perform the parameter estimation with this data. The correct split ratio should yield the statistically most significant estimates, which would additionally allow to unravel compartment specific metabolite concentrations.

Compared to other local non-stationary 13C metabolic flux analysis approaches, our methodology has certain advantages. Kinetic flux profiling (KFP) employs the labeling dynamics of reaction substrates and products to estimate absolute, local fluxes39,49. While the major benefit of this methodology is the absolute, and not only relative, quantification of flux, it

95 Chapter 3 additionally requires the measurement of absolute metabolite concentrations and is less straightforward with multiple substrates and converging fluxes. A more recent extension of

KFP, called rKFP50, allows to estimate relative flux changes between two conditions from non- stationary labeling data and only relative, but not absolute, metabolite changes. In our study, this method might have potentially also allowed to investigate the flux differences between WT and Mpc1KO. However, our method is superior in analyzing relative fluxes for only one condition, since no reference state is required.

In summary, we demonstrate that non-stationary 13C MFRA has the potential to strongly shorten the time required for a labeling experiment while allowing precise estimates of relative fluxes. Therefore, it can provide access to systems where a metabolic (pseudo)- steady state can only be obtained for very short time periods, such as, for example during metabolic transitions or in media with multiple carbon substrates that are consumed sequentially in distinct metabolic steady states.

96 Chapter 3

References

1. Lau, W., Fischbach, M. a., Osbourn, A. & Sattely, E. S. Key Applications of Plant Metabolic Engineering. PLoS Biol. 12, 1±5 (2014).

2. Kim, H. U., Kim, T. Y. & Lee, S. Y. Metabolic flux analysis and metabolic engineering of microorganisms. Mol. Biosyst. 4, 113±120 (2008).

3. Boghigian, B. a., Seth, G., Kiss, R. & Pfeifer, B. a. Metabolic flux analysis and pharmaceutical production. Metab. Eng. 12, 81±95 (2010).

4. Sanford K., Soucaille P., Whited G. and Chotani G. Genomics to fluxomics and physiomics - pathway engineering. Curr. Opin. Microbiol. 5, 318±322 (2002).

5. Heinemann, M. & Sauer, U. Systems biology of microbial metabolism. Curr. Opin. Microbiol. 13, 337±43 (2010).

6. Kruger, N. J. & Ratcliffe, R. G. Insights into plant metabolic networks from steady- state metabolic flux analysis. Biochimie 91, 697±702 (2009).

7. Stephanopoulos, G. N., Aristidou, A. A. & Nielsen, J. Metabolic engineering: principles and methodologies. (Academic press, 1998).

8. Zamboni, N., Fendt, S.-M., Rühl, M. & Sauer, U. 13C-based metabolic flux analysis. Nat. Protoc. 4, 878±92 (2009).

9. Zamboni, N. 13C metabolic flux analysis in complex systems. Curr. Opin. Biotechnol. 22, 103±8 (2011).

10. Fischer, E. & Sauer, U. Large-scale in vivo flux analysis shows rigidity and suboptimal performance of Bacillus subtilis metabolism. Nat. Genet. 37, 636±40 (2005).

11. Nanchen, A., Fuhrer, T. & Sauer, U. Determination of Metabolic Flux Ratios From 13C- Experiments and Gas Chromatography±Mass Spectrometry Data. Methods Mol. Biol. 358, 177-97 (2007).

12. Antoniewicz, M. R. Tandem mass spectrometry for measuring stable-isotope labeling. Curr. Opin. Biotechnol. 24, 48±53 (2013).

13. Rühl, M. et al. Collisional fragmentation of central carbon metabolites in LC-MS/MS increases precision of 13C metabolic flux analysis. Biotechnol. Bioeng. 109, 763±71 (2012).

14. Rühl, M., Zamboni, N. & Sauer, U. Dynamic flux responses in riboflavin overproducing Bacillus subtilis to increasing glucose limitation in fed-batch culture. Biotechnol. Bioeng. 105, 795±804 (2010).

15. Metallo, C. M., Walther, J. L. & Stephanopoulos, G. Evaluation of 13C isotopic tracers for metabolic flux analysis in mammalian cells. J. Biotechnol. 144, 167±174 (2009).

16. Buescher, J. M. et al. A roadmap for interpreting 13C metabolite labeling patterns from cells. Curr. Opin. Biotechnol. 34, 189±201 (2015).

97 Chapter 3

17. Antoniewicz, M. R. 13C metabolic flux analysis: Optimal design of isotopic labeling experiments. Curr. Opin. Biotechnol. 24, 1116±1121 (2013).

18. Goudar, C. et al. Metabolic flux analysis of CHO cells in perfusion culture by metabolite balancing and 2D [13C, 1H] COSY NMR spectroscopy. Metab. Eng. 12, 138±49 (2010).

19. Kruger, N. J., Masakapalli, S. K. & Ratcliffe, R. G. Strategies for investigating the plant metabolic network with steady-state metabolic flux analysis: Lessons from an Arabidopsis cell culture and other systems. J. Exp. Bot. 63, 2309±2323 (2012).

20. Possemato, R. et al. Functional genomics reveal that the serine synthesis pathway is essential in breast cancer. Nature 476, 346±50 (2011).

21. DeBerardinis, R. J. et al. Beyond aerobic glycolysis: transformed cells can engage in glutamine metabolism that exceeds the requirement for protein and nucleotide synthesis. Proc. Natl. Acad. Sci. U. S. A. 104, 19345±50 (2007).

22. Grassian, A. R. et al. IDH1 mutations alter citric acid cycle metabolism and increase dependence on oxidative mitochondrial metabolism. Cancer Res. 74, 3317±31 (2014).

23. Ye, J. et al. Pyruvate kinase M2 promotes de novo serine synthesis to sustain mTORC1 activity and cell proliferation. Proc. Natl. Acad. Sci. U. S. A. 109, 6904-6909 (2012).

24. Guzmán, S. et al. 13C metabolic flux analysis shows that resistin impairs the metabolic response to insulin in L6E9 myotubes. BMC Syst. Biol. 8, 109 (2014).

25. Allen, D. K., Shachar-Hill, Y. & Ohlrogge, J. B. Compartment-specific labeling information in 13C metabolic flux analysis of plants. Phytochemistry 68, 2197±210 (2007).

26. Lewis, C. a et al. Tracing Compartmentalized NADPH Metabolism in the Cytosol and Mitochondria of Mammalian Cells. Mol. Cell 55, 253±263 (2014).

27. Crown, S. B., Ahn, W. S. & Antoniewicz, M. R. Rational design of 13C-labeling experiments for metabolic flux analysis in mammalian cells. BMC Syst. Biol. 6, 43 (2012).

28. Nöh, K. et al. Metabolic flux analysis at ultra short time scale: isotopically non- stationary 13C labeling experiments. J. Biotechnol. 129, 249±67 (2007).

29. Noack, S., Nöh, K., Moch, M., Oldiges, M. & Wiechert, W. Stationary versus non- stationary 13C-MFA: a comparison using a consistent dataset. J. Biotechnol. 154, 179±90 (2011).

30. Nöh, K. & Wiechert, W. The benefits of being transient: isotope-based metabolic flux analysis at the short time scale. Appl. Microbiol. Biotechnol. 91, 1247±65 (2011).

31. Hörl, M., Schnidder, J., Sauer, U. & Zamboni, N. Non-stationary 13C-metabolic flux ratio analysis. Biotechnol. Bioeng. 110, 3164±3176 (2013).

98 Chapter 3

32. Vacanti, N. M. et al. Regulation of Substrate Utilization by the Mitochondrial Pyruvate Carrier. Mol. Cell 56, 425±435 (2014).

33. Wu, L. et al. Quantitative analysis of the microbial metabolome by isotope dilution mass spectrometry using uniformly 13C-labeled cell extracts as internal standards. Anal. Biochem. 336, 164±171 (2005).

34. Zor, T. & Selinger, Z. Linearization of the Bradford protein assay increases its sensitivity: theoretical and experimental studies. Anal. Biochem. 236, 302±8 (1996).

35. Buescher, J. M., Moco, S., Sauer, U. & Zamboni, N. Ultrahigh Performance Liquid Chromatography - Tandem Mass Spectrometry Method for Fast and Robust Quantification of Anionic and Aromatic Metabolites. Anal. Chem. 82, 4403±4412 (2010).

36. Zimmermann, M., Sauer, U. & Zamboni, N. Quantification and Mass Isotopomer Profiling of Į-Keto Acids in Central Carbon Metabolism. Anal. Chem. 86, 3232-3237 (2014).

37. Rühl, M. et al. Collisional fragmentation of central carbon metabolites in LC-MS/MS increases precision of 13C metabolic flux analysis. Biotechnol. Bioeng. 109, 763±71 (2012).

38. Nanchen, A., Fuhrer, T. & Sauer, U. Determination of metabolic flux ratios from 13C- experiments and gas chromatography-mass spectrometry data: protocol and principles. Methods Mol. Biol. 358, 177±197 (2007).

39. Yuan, J., Fowler, W. U., Kimball, E., Lu, W. & Rabinowitz, J. D. Kinetic flux profiling of nitrogen assimilation in Escherichia coli. Nat. Chem. Biol. 2, 529±30 (2006).

40. Mootha, V. K. et al. Integrated analysis of protein composition, tissue diversity, and gene regulation in mouse mitochondria. Cell 115, 629±640 (2003).

41. Zamboni, N., Maaheimo, H., Szyperski, T., Hohmann, H.-P. & Sauer, U. The phosphoenolpyruvate carboxykinase also catalyzes C3 carboxylation at the interface of glycolysis and the TCA cycle of Bacillus subtilis. Metab. Eng. 6, 277±84 (2004).

42. Nicolae, A., Wahrheit, J., Bahnemann, J., Zeng, A.-P. & Heinzle, E. Non-stationary 13C metabolic flux analysis of Chinese hamster ovary cells in batch culture using extracellular labeling highlights metabolic reversibility and compartmentation. BMC Syst. Biol. 8, 50 (2014).

43. Metallo, C. M. et al. Reductive glutamine metabolism by IDH1 mediates lipogenesis under hypoxia. Nature 481, 380±4 (2012).

44. Wise, D. R. et al. Hypoxia promotes isocitrate dehydrogenase-dependent carboxylation of Į-ketoglutarate to citrate to support cell growth and viability. Proc. Natl. Acad. Sci. U. S. A. 108, 19611±19616 (2011).

45. Mullen, A. R. et al. Reductive carboxylation supports growth in tumour cells with defective mitochondria. Nature 481, 385±8 (2012).

99 Chapter 3

46. Schell, J. C. et al. A role for the mitochondrial pyruvate carrier as a repressor of the Warburg Effect and colon cancer growth. Mol. Cell 56, 400-13 (2014).

47. Wahrheit, J., Nicolae, A. & Heinzle, E. Eukaryotic metabolism: Measuring compartment fluxes. Biotechnol. J. 6, 1071±1085 (2011).

48. Schryer, D. W., Peterson, P., Paalme, T. & Vendelin, M. Bidirectionality and compartmentation of metabolic fluxes are revealed in the dynamics of isotopomer networks. Int. J. Mol. Sci. 10, 1697±1718 (2009).

49. Yuan, J., Bennett, B. D. & Rabinowitz, J. D. Kinetic flux profiling for quantitation of cellular metabolic fluxes. Nat. Protoc. 3, 1328±40 (2008).

50. Huang, L., Kim, D., Liu, X., Myers, C. R. & Locasale, J. W. Estimating Relative Changes of Metabolic Fluxes. PLoS Comput. Biol. 10, e1003958 (2014).

100 Chapter 3

Supplementary Figures

Supplementary Figure 1. Schematic of carbon transitions in the TCA-cycle in a [U-13C]glucose labeling experiment adapted from Metallo et al. (2012)43.

13 Supplementary Figure 2. (a) Schematic of carbon transitions in the [3,4- C2]glucose labeling experiment. (b and c) Control that both metabolites in upper (e.g. (b) glucose-6-phosphate) and lower glycolysis (e.g. (c) 1,3-biphospho glycerate) showed the expected label enrichment.

101 Chapter 3

Supplementary Figure 3. Schematic of carbon transitions in the TCA-cycle in a [U-13C]glutamine labeling experiment adapted from Metallo et al. (2012)43.

Supplementary Figure 4. Schematic of carbon transitions in a [1-13C]glutamine labeling experiment adapted from Metallo et al. (2012)43.

102 Chapter 3

Supplementary Table 1. ODE model equations

ODE systems for anaplerosis vs oxidative vs reductive TCA cycle flux estimation

Glucose labeling

݀ܣݏ݌௠଴ ͳ ݌௠଴ ή ሺݒ௢௫்஼஺ ൅ ݒ௥௘ௗ்஼஺ ൅ ݒ௔௡௔ሻ൯ݏܣ ௠଴ ή ݒ௔௡௔ െݎݑݐ௠ଶሻ ή ݒ௥௘ௗ்஼஺ ൅ ܲݕ݈ܩ ݑݐ௠଴ ൅݈ܩൌ ή ൫ܵݑܿܿ௠଴ ή ݒ௢௫்஼஺ ൅ ሺ ݐ ܿ஺௦௣݀

݀ܣݏ݌௠ଶ ͳ ݌௠ଶ ή ሺݒ௢௫்஼஺ ൅ ݒ௥௘ௗ்஼஺ ൅ ݒ௔௡௔ሻ൯ݏܣ ௠ଶ ή ݒ௔௡௔ െݎൌ ή ൫ܵݑܿܿ௠ଶ ή ݒ௢௫்஼஺ ൅ ܲݕ ݐ ܿ஺௦௣݀

݀ܣݏ݌௠ଷ ͳ ݌௠ଷ ή ሺݒ௢௫்஼஺ ൅ ݒ௥௘ௗ்஼஺ ൅ ݒ௔௡௔ሻ൯ݏܣ ௠ଷ ή ݒ௔௡௔ െݎൌ ή ൫ܵݑܿܿ௠ଷ ή ݒ௢௫்஼஺ ൅ ܲݕ ݐ ܿ஺௦௣݀

Glutamine labeling

݀ܣݏ݌௠଴ ͳ ݌௠଴ ή ሺݒ௢௫்஼஺ ൅ ݒ௥௘ௗ்஼஺ ൅ ݒ௔௡௔ሻ൯ݏܣ ௠଴ ή ݒ௔௡௔ െݎݑݐ௠଴ ή ݒ௥௘ௗ்஼஺ ൅ ܲݕ݈ܩ ൌ ή ൫ܵݑܿܿ௠଴ ή ݒ௢௫்஼஺ ൅ ݐ ܿ஺௦௣݀

݀ܣݏ݌௠ଷ ͳ ݌௠ଷ ή ሺݒ௢௫்஼஺ ൅ ݒ௥௘ௗ்஼஺ ൅ ݒ௔௡௔ሻ൯ݏܣ ௠ଷ ή ݒ௔௡௔ െݎݑݐ௠ଷሻ ή ݒ௥௘ௗ்஼஺ ൅ ܲݕ݈ܩ ݑݐ௠଴ ൅݈ܩൌ ή ൫ܵݑܿܿ௠ଷ ή ݒ௢௫்஼஺ ൅ ሺ ݐ ܿ஺௦௣݀

݀ܣݏ݌௠ସ ͳ ݌௠ସ ή ሺݒ௢௫்஼஺ ൅ ݒ௥௘ௗ்஼஺ ൅ ݒ௔௡௔ሻ൯ݏܣ ൌ ή ൫ܵݑܿܿ௠ସ ή ݒ௢௫்஼஺ െ ݐ ܿ஺௦௣݀

103

104 Chapter 4

7KHPDOLFHQ]\PH

%DFLOOXVVXEWLOLVE\DQ1$'3+GHSHQGHQWVZLWFKLQHQ]\PH

DFWLYLW\

Manuel Hörl, Tobias Fuhrer and Nicola Zamboni

Institute of Molecular Systems Biology, ETH Zürich, Zürich, Switzerland

Manuel Hörl designed the study, performed all experiments and data analysis and wrote the manuscript. Tobias Fuhrer advised the flux estimation and NADPH balancing. Nicola Zamboni supervised the study.

105 Chapter 4

Abstract

The redox cofactor NADPH is required as a reducing equivalent in about 100 anabolic reactions, but only a handful of reactions in central carbon metabolism generate this cofactor during catabolism of glucose. To sustain growth, anabolic demand and catabolic production of

NADPH must be tightly balanced and genetically or environmentally induced perturbations of this equilibrium have been shown to be lethal. For B. subtilis, 13C metabolic flux analysis predicts a catabolic excess production of NADPH of up to ~70%, demonstrating that our current understanding of metabolic operation in this organism lacks the mechanisms that ensure NADPH rebalancing.

In this study, we investigated the NADPH-balancing capacities of the four malic enzymes that coexist in B. subtilis. These enzymes have been suggested to be involved in redox homeostasis by cyclic operation of isoforms with different cofactor specificities, allowing for the transhydrogenation of NADPH into NADH. We used 13C metabolic flux analysis to quantify the

NADPH-production of malic enzyme mutants and identified a distinct role of isoform YtsJ in redox balancing, which could potentially be explained by the high reversibility of the enzyme, which would allow a transhydrogenation if specific cofactor usage was assumed in each direction. Through in vitro assays with the purified enzyme, we show for the first time that YtsJ is the sole isoform capable of catalyzing NADPH oxidation. Further investigations on the mechanism through mass spectrometry-based in vitro transhydrogenation assays revealed that the presence of physiological concentrations of NADPH unexpectedly modified the activity of the enzyme from a pyruvate producing malic enzyme to a lactate generating malolactic enzyme. Lactate formation is not coupled to NADPH generation and would, combined with the identified NADPH oxidation by YtsJ, allow its net consumption. Finally, metabolomics data suggested that this mechanism might also occur in vivo and could, together with lactate dehydrogenase, endow B. subtilis with a transhydrogenation cycle to flexibly balance the cellular redox state. In summary, our study extends the known redox cofactor balancing mechanisms, by providing for the first time evidence that a metabolic enzyme changes its function depending on the redox state.

106 Chapter 4

Introduction

Bacteria grow in various environments by metabolizing a plethora of small carbon compounds, such as sugars, alcohols and organic acids. Catabolism of the majority of these molecules proceeds through the roughly sixty reactions of central carbon metabolism, which convert them into the building blocks for about a hundred anabolic reactions to synthesize the macromolecular components of a cell. Concomitantly, central carbon metabolism also provides energy and reducing power via the cofactors ATP, NADH and NADPH. In aerobic conditions, the primary role of NADH is the transfer of electrons to oxygen, driving the oxidative phosphorylation of ADP to ATP1,2. The chemically very similar redox cofactor NADPH, in contrast, serves as the reducing equivalent for anabolic reductions.

The primary NADPH-producing reactions are the oxidative pentose phosphate pathway, and isocitrate dehydrogenase in the TCA cycle. Additionally, most organisms also contain a

NADPH-producing malic enzyme3±6. The production of NADPH is directly coupled to the intracellular fluxes through these catabolic pathways, which strongly depend on the available nutrients. Consumption of NADPH depends primarily on anabolic reactions and is coupled to the rate of biomass formation7. For steady growth, organisms must therefore be capable of establishing a perfect balance between catabolic supply and anabolic demand of NADPH.

While yeast are indeed able to adjust their intracellular rates accordingly8,9, bacteria often produce more NADPH in carbon catabolism than is required for their anabolism10. To establish homeostasis, microbes hence require mechanisms to decouple catabolic NADPH formation and anabolic NADPH demand. Two such biochemical mechanisms have been reported. First, organisms may possess nicotinamide nucleotide transhydrogenases, which catalyze the reversible transfer of electrons between NAD+ and NADP+11. Their physiological role in NADPH metabolism has been studied in detail in Escherichia coli, where the membrane-bound, energy-dependent transhydrogenase PntAB reduces NADP+ during shortage, while the soluble transhydrogenase UdhA oxidizes NADPH under excess production12. Second, biochemical redox cycles consisting of two isoenzymes with different cofactor specificities that operate in a cyclic manner, can realize a net conversion of NADPH into NADH without affecting

107 Chapter 4

the net catabolic fluxes. Examples of such redox cycles include the simultaneous operation of isoforms of isocitrate dehydrogenase in animal mitochondria13 and alcohol and glyceraldehyde-3-phosphate dehydrogenases in Kluyveromyces lactis14.

During exponential growth on glucose, the Gram±positive bacterium Bacillus subtilis exhibits extensive NADPH-overproduction10,15, but the mechanism of NADPH reoxidation is not yet understood. Without an annotated gene for a transhydrogenase, redox cycles were suggested to perform this task in B. subtilis16. Two potential cycles exist in central metabolism of B. subtilis between the two glyceraldehyde-3-phosphate dehydrogenases and the four malic enzymes with different NAD+ and NAPD+ specificities. However, since the NADP+-dependent glyceraldehyde-3-phosphate dehydrogenases GapB appears to be inactive during growth on glucose17, only malic enzymes remain as a potential balancing mechanism under this condition.

Here, we set out to investigate a potential role of B. subtilis malic enzymes in balancing of catabolic NADPH formation with anabolic demand. Specifically, we analyzed intracellular steady state fluxes and NADPH balances of exponentially growing B. subtilis malic enzyme deletion strains from stationary and non-stationary 13C labeling data. These estimates suggested the oxidation of excess NADPH by one malic enzyme isoform YtsJ, which we confirmed by in vitro enzymatic assays. We further investigated the specific mechanism in vitro, by combining enzymatic assays and untargeted metabolite profiling. This led to the discovery of a secondary, redox-neutral reaction of YtsJ, which, in combination with lactate dehydrogenase, would allow a net transhydrogenation cycle in vivo. Finally, based on metabolomics data, we provide evidence for the existence of the demonstrated mechanism in vivo.

108 Chapter 4

Materials and Methods

Bacterial strains, growth conditions and media

3 The E. coli strains used for overproduction of His6-MaeA, His6-MalS, His6-MleA and His6-YtsJ were grown in Luria-Bertani (LB) medium supplemented with 100 mg L-1 ampicillin and

25 mg L-1 kanamycin.

The B. subtilis strains used in this study are listed in Table 1. Frozen glycerol stocks were used to inoculate 5 mL of LB medium. Antibiotics for selection were added at 5 mg L-1

(chloramphenicol), 0.4 mg L-1 (erythromycin), 5 mg L-1 (kanamycin), 100 mg L-1 (ampicillin) or

100 mg L-1 (spectinomycin). After 5 h of incubation at 37°C and 300 rpm on a gyratory shaker,

5 mL of M9 minimal medium were inoculated at 1000- to 8000-fold dilutions as precultures.

Mid-exponentially M9 precultures at optical densities at 600 nm (OD600) of 1-2 were used to inoculate 35 mL M9 batch cultures in 500 mL shake flasks to an OD600 of 0.03. The M9 medium contained, per liter of deionized water: 8.5 g of Na2HPO4 2H2O, 3.0 g KH2PO4, 1 g NH4Cl, 0.5 g NaCl and was adjusted to pH 7 before autoclaving. The following components were filter sterilized separately and then added (per liter of final medium): 1 mL of 1 M MgSO4, 1 mL of

0.1 M CaCl2, 1 mL 0.05 M FeCl3 and 10 mL of a trace element solution containing (per liter)

170 mg ZnCl2, 100 mg MnCl2 4H2O, 60 mg CoCl2 6H2O, 60 mg Na2MoO4 2H2O, and 43 mg

-1 CuCl2 2H2O. Autoclaved carbon source solutions were added to a final concentration of 5 g L .

Table 1. B. subtilis strains used in this study.

Strain Genotype Source or reference 168CA wild-type, trpC2 Laboratory stock GM1608 ytsJ'>S087,1¨ lacZ-ery)::kan] (Lerondel et al., 2006)3 GTD102 maeA'::kan (Doan et al., 2003)18 GTD110 ¨malS'::spc (Doan et al., 2003)18 GM1632 mleA'::cat (Lerondel et al., 2006)3 GM1655 maeA::pEC23 (kan) malS::spec mleA::pMutin3 (ery) (Lerondel et al., 2006)3

109 Chapter 4

Physiological parameters

Cell growth was determined spectrophotometrically at 600 nm. Glucose, acetate, fumarate, pyruvate, lactate and succinate concentrations in the supernatant were estimated by the signals of a refractive index and diode array detector on a HPLC (Agilent 1100), using an

Aminex HPX-87H column at a temperature of 60°C with 5 mM H2SO4 as eluent. Supernatant samples were prepared by centrifugation of 1 mL culture broth for 5 min at 4°C and 14000x g.

Specific growth rates were calculated by linear regression of logarithmic OD600 over time.

Specific uptake and secretion rates were estimated by linear regression of substrate or product concentration against biomass concentration.

Stationary 13C-flux analysis

Fluxes in central metabolism of B. subtilis were estimated according to Zamboni et al. (2009)19.

Specifically, cultures were inoculated to an OD600 of 0.03 in M9 medium containing 100%

[1-13C]glucose and a mixture of 50% (w/w) uniformly labeled and 50% naturally labeled glucose, respectively. During mid-exponential growth phase, 1mL of cell broth was harvested by centrifugation (2 min, 23000x g, 4°C), washed with 0.9% NaCl and stored at -20°C untill further analysis. The pellets were hydrolyzed with 6 M HCl at 105 °C for 18 h and dried at 95°C under constant air stream. Hydrolysates were dissolved in 20 µL of dimethylformamide (Sigma

Aldrich) and transferred to GC-MS vials. After addition of 20 µL N-tertbutyldimethylsilyl-N- methyltrifluoroacetamide with 1% (wt/wt) tertbutyldimethyl-chlorosilane (Sigma-Aldrich), the mixture was incubated at 85 °C for 1 h. Subsequently, mass isotopomer distributions of protein- bound amino acids were determined on a 6890N GC system (Agilent Technologies) combined with a 5875 Inert XL MS system (Agilent Technologies).

After correction for naturally occurring stable isotopes, amino acid mass isotopomer distributions were used to calculate ratios of converging metabolic fluxes. These ratios together with extracellular fluxes and a stoichiometric model of B. subtilis central metabolism20 were then used as constraints to calculate absolute intracellular fluxes. All calculations were performed using FiatFlux21 and Matlab 7.12.0.

110 Chapter 4

Free metabolite measurements

B. subtilis strains were grown in shake flask culture to mid-exponential phase (OD600 between

0.5 and 1). An amount proportional to 1 mL*OD of the culture broth was transferred onto a

0.45 µm pore size Durapore filter (Millipore) and vacuum-filtered and the filter was immediately transferred into 4 mL of -20°C acetonitrile/methanol/water (2:2:1) to quench metabolism and kept at -20°C for 1 h for extraction. For pyruvate measurements, 2 mL*OD were similarly sampled by filtration and added to the acetonitrile/methanol/water-mixture containing 25 µM phenylhydrazine for derivatization of Į-keto acids22. Experiments were performed in triplicates with cells from different shake flasks.

The supernatants were dried at 0.12 mbar in a SpeedVac composed of an Alpha 2-4 LD plus cooling trap, a RVC 2-33 rotational vacuum concentrator and a RC-5 vacuum chemical hybrid pump (Christ, Osterode am Harz, Germany). Dried extracts were resuspended in 100 µL deionized water, 10 µL of which were injected into a Waters Acquity UPLC with a Waters T3 column (150 x 2.1mm x 1.8 µm; Waters Corporation, Milford, MA, USA) coupled to a Thermo

TSQ Quantum Ultra triple quadrupole instrument (Thermo Fisher Scientific, Waltham, MA,

USA) with electrospray ionization. Compound separation and acquisition was achieved as described22,23 and peak integration was performed by an in-house software (Begemann and

Zamboni, unpublished).

Malic enzyme expression and purification

His6-tagged proteins were overexpressed using the QIAexpress kit (QIAGEN). After induction with 1 mM IPTG for 5-7 h, cells were harvested by centrifugation at 4°C, washed with 0.9%

NaCl, resuspended in lysis buffer (100 mM Tris-HCl, pH 7.5, 5 mM MgCl2, 1 mM dithiothreitol

(DTT) and 4 mM PMSF) and disrupted by three passages through a French press cell at 4°C.

Cell-free lysates were obtained by centrifugation for 10 min at 23000x g and 4°C. His6-tagged proteins were purified using Ni2+-charged nitrilotriacetic acid (Ni-NTA) affinity columns (GE

Healthcare) DFFRUGLQJWRPDQXIDFWXUHU¶VLQVWUXFWLRQVProteins were eluted from the column with elution buffer (500 mM imidazole) which was subsequently replaced by storage buffer (50 mM Tris pH 8, 150 mM NaCl, 1mM DTT, and 0.5 mM EDTA)3 using ultrafiltration columns with

111 Chapter 4

10 kDa size cutoff (Millipore). Proteins were stored at 4° C at most 1 day before performing activity assays.

Spectrophotometrical-based in vitro malic enzyme assays

Reverse malic enzyme activities were tested at 37°C by spectrophotometrically monitoring

NAD(P)H oxidation (pyruvate plus NAD(P)H into malate plus NAD(P)+) at 340 nm. The reaction mixture consisted of 100 mM Tris-HCl pH 7.8, 5 mM MgCl2, 50 mM KCl, 50 mM NaHCO3,

24 20mM pyruvate and 0.2 mM NAD(P)H . For determination of the Km for pyruvate, both cofactors NADH and NADPH were added to the reaction mixture at 0.2 mM. The Km for cofactors was estimated at a pyruvate concentration of 20 mM. For both cofactors, a molar

6 2 -1 extinction coefficient of 6.22 x 10 cm mol was used for calculations. To determine Km and kcat, the initial reaction rates were fitted to a Michaelis-Menten relationship by least-squares analysis. Enzyme concentrations were estimated by a Bradford protein assay25.

Mass spectrometry-based in vitro malic enzyme transhydrogenation assays

Purified YtsJ was incubated at room temperature in 200 ȝL of 10 mM potassium phosphate

+ buffer pH 7.4, 2.5 mM MgCl2, 50 mM KCl and 1mM NaHCO3 with a mixture of NAD , malate and pyruvate at physiological concentrations and varying amounts of NADPH. The enzyme reaction samples were assayed by direct online flow injection into a TOF MS (6520 Series

QTOF, Agilent Technologies) operated in the negative ionization mode. High-precision mass spectra were recorded from 50±1,000 m/z and analyzed as described previously26.

112 Chapter 4

Results

The cellular NADPH balance changes upon deletion of malic enzyme YtsJ

To investigate the potential role of malic enzymes for NADPH balancing in B. subtilis, we first quantified NADPH production and consumption in wild-type and malic enzyme single deletion mutants ¨maeA, ¨malS, ¨mleA, and ¨ytsJ. We estimated absolute intracellular net carbon fluxes by 13C metabolic flux analysis. Specifically, we harvested exponentially growing cells that were cultured either with 100% [1-13C]glucose or a mixture of 50% naturally labeled and

50% [U-13C]glucose and determined 13C labeling patterns of protein-bound amino acids by GC-

MS analysis to calculate ratios of converging metabolic fluxes19. These ratios together with physiological rates were then used to constrain a stoichiometric model of B. subtilis central metabolism20 using the FiatFlux software21.

The estimated flux distribution for wild-type B. subtilis favorably matched previously reported results10,15, with glucose being catabolized ~70%:30% by glycolysis and the pentose phosphate pathway, respectively (Fig. 1a). Glucose was further catabolized in the TCA cycle and mainly incorporated into biomass formation, with only a small fraction being secreted as acetate. Fluxes in glycolysis and the pentose phosphate pathway were almost identical in malic enzyme single deletion mutants and wild-type (Fig. 1b-e). Deletion of each malic enzyme isoform lead to an increase in acetate secretion by 17% to 40% of the wild-type value and lower flux through the TCA cycle and the malic enzyme reaction. The two latter effects were most pronounced in the ¨ytsJ mutant, with ~40% decreased TCA cycle flux compared to wild- type and virtually zero net flux from malate to pyruvate.

To estimate the NADPH balancing capacities of each strain, we summed the contribution of major NADPH-producing reactions including the first two steps of the pentose phosphate pathway and isocitrate dehydrogenase in the TCA cycle. We also included potential NADPH formation by malic enzyme, since YtsJ has a strong preference for NADP+ and was shown to be the most active isoenzyme during growth on glucose3.

113 Chapter 4

Figure 1. Metabolic flux distribution in B. subtilis (a) wild-type and (b-f) malic enzyme deletion mutants, estimated

13 by C flux ratio analysis and metabolite balancing. Flux values are normalized to the glucose uptake rate (qglc, mmol

-1 -1 gCDW h ) of each mutant and arrow sizes scale with flux magnitudes. (h) NADPH-balances of wild-type and malic

enzyme deletion strains. NADPH production is calculated by summing the rates through NADPH-dependent glucose-

6-phosphate dehydrogenase (green), 6-phosphogluconate dehydrogenase (yellow), isocitrate dehydrogenase (red)

and potentially malic enzymes (blue), and comparing it to the normalized growth dependent NADPH-consumption

(grey). Excess NADPH production is indicated in white. (g) Metabolic flux distribution and (h) NADPH-balance in B.

subtilis wild-type with hypothetical, NADPH-consuming transhydrogenation cycle (purple).

We then compared NADPH production to the growth rate-dependent NADPH-demand for biomass formation (Fig. 1h). Consistent with previous studies24, wild-type B. subtilis exhibited an NADPH overproduction of 67% of the biomass requirements. While the NADPH

114 Chapter 4

overproduction was reduced by up to 40% of the wild-type value in ¨maeA, ¨malS and ¨mleA mutants, overproduction in the ¨ytsJ mutant was completely absent, partially due to the missing malic enzyme flux but mainly because of the drastic decrease of flux through isocitrate dehydrogenase in the TCA cycle. This result suggested that the loss of YtsJ had to be compensated by an accurate adjustment of NADPH-producing fluxes that meet the demands for biomass formation. From this we concluded that YtsJ is usually involved in redox homeostasis of B. subtilis by consuming (excess) NADPH in wild-type.

We first hypothesized about the existence of a redox cycle formed by NADPH-consuming YtsJ and either of the three NAD-dependent malic enzymes, which seemed to be individually compensated upon deletion by the remaining isoforms. If this was true, the triple deletion mutant of NAD-dependent malic enzymes ¨maeA ¨malS ¨mleA should show similar fluxes and overall NADPH balance to the YtsJ deletion strain. However, intracellular fluxes and

NADPH balance in the triple deletion were comparable to the wild-type (Fig. 1f, h), indicating that the presence of any of the NAD+-dependent isoenzymes is not necessary to bring about the hypothesized transhydrogenation associated to YtsJ and that no redox cycling between

YtsJ and the other malic enzyme isoforms exists under the investigated condition.

Transhydrogenation by YtsJ alone could theoretically be accomplished, if we assume that the reaction is reversible and simultaneously performs NAD-dependent malate decarboxylation and NADPH-dependent pyruvate carboxylation. Results from stationary 13C flux analysis do not provide information on such reversibility, but only estimates on net fluxes. On the other hand, the estimation of relative fluxes with non-stationary 13C metabolic flux ratio analysis provides the ratio of sheer forward fluxes27. A discrepancy in this forward to net flux ratio of pyruvate formation from malic enzyme, as revealed in our previous analysis in Chapter 2, indicates the existence of a reversible malic enzyme reaction and, included as an additional constraint in the 13C flux fitting, allows to calculate the reversibility of the reaction. We included this additional constraint for the flux fitting of the wild-type and assumed NAD-specific malate decarboxylation and NADPH-specific pyruvate carboxylation. The estimated flux reversibility

115 Chapter 4

in the malic enzyme reaction lead to a closed NADPH balance (Fig. 1g, h), suggesting that

YtsJ alone could indeed facilitate NADPH homeostasis in B. subtilis.

YtsJ catalyzes NADPH consumption in vitro

Even though the specific mechanism by which YtsJ would accomplish redox homeostasis in

B. subtilis is not known, it implies the ability of this malic enzyme to catalyze the oxidation of excess NADPH through reductive carboxylation of pyruvate. This goes against the traditional view of bacterial malic enzymes, which are rather believed to catalyze the oxidative decarboxylation of malate coupled with the reduction of NAD(P)+ to produce pyruvate and

NAD(P)H, in order to supply acetyl-CoA from malate during growth on TCA cycle intermediates and to generate NADPH for anabolism4,28±31.

To verify if YtsJ is capable of catalyzing NADPH oxidation, we performed in vitro biochemical assays with purified malic enzymes. For this purpose, we overexpressed all B. subtilis malic enzyme isoforms in E. coli3 and purified them by His-tag affinity chromatography. Correct identity was verified on an SDS-PAGE and by testing each isoform in the well-known oxidative decarboxylation reaction direction from malate to pyruvate with its preferred cofactor3 using a spectrophotometric assay24 (Fig. 2). Similarly, the reversed reaction of reductive pyruvate carboxylation was assayed for all isoforms with both NADH and NADPH by monitoring their oxidation at 340 nm. While there was no detectable reductive pyruvate carboxylation with

NADH in the presence of any of the four malic enzymes (Fig. 3c), with NADPH the signal declined when YtsJ was present in the assay (Fig 3a, b). The NADPH consumption showed a clear dependence on the YtsJ concentration (Fig 3a, b), reflecting the exclusive ability of this malic enzyme to catalyze the reductive pyruvate carboxylation with concomitant oxidation of

NADPH to NADP+.

To get a quantitative understanding of the NADPH oxidation capacity of YtsJ, we then determined Km and kcat values for pyruvate and NADPH (see Materials and Methods &

Supplementary Table 1). The Km values for both pyruvate (Km 5.3±1.3 mM) and NADPH

(Km 0.8+/-0.5 mM) were in the range of the physiological concentrations of B. subtilis growing

116 Chapter 4

on glucose (pyruvate: 8.9+/-5.75 mM; NADPH: 0.36+/-0.25 mM)24,32. This data suggested that the reaction could indeed operate in vivo and additionally indicated that its in vivo reaction rate can be flexibly adjusted by changes in the substrate concentrations of pyruvate and NADPH33.

Figure 2. In vitro spectrophotometric assays to confirm malic enzyme activity (oxidative malate decarboxylation) for

all purified isoforms. According to the previously determined cofactor preferences3, (a) MaeA, (b) MalS and (c) MleA

were assayed with NAD+ and (d) YtsJ with NADP+.

Figure 3. In vitro spectrophotometric assays to characterize NADH and NADPH oxidation through reductive

pyruvate carboxylation by B. subtilis malic enzymes. (a) Time courses of NADH (unfilled circles) and NADPH

(filled circles) oxidation at different YtsJ concentrations recorded at 340 nm. Initial consumption rates of NADPH

(b) and NADH (c) as a function of MaeA, MalS, MleA and YtsJ concentration.

117 Chapter 4

Increasing NADPH levels transform YtsJ into a malolactic enzyme

The observation that pyruvate carboxylation solely occurred in the presence of YtsJ and exclusively with NADPH as cofactor in the in vitro assays, together with the identified YtsJ- dependent net flux from malate towards pyruvate in vivo by 13C flux analysis (Fig 1a), led us to hypothesize that YtsJ alone could potentially reoxidize cellular excess NADPH based on its unique enzymatic properties. More precisely, YtsJ falls into the EC 1.1.1.40 category as a malic enzyme with a dual specificity for NAD+ and NADP+ for malate decarboxylation3. Since we showed that the carboxylation of pyruvate is NADPH-specific, YtsJ could itself theoretically constitute a transhydrogenation mechanism to convert NADPH into NADH. This would imply that the intracellular concentrations of all reaction partners thermodynamically drive a NAD+- dependent malate decarboxylation together with NADPH-dependent pyruvate carboxylation within the cell. Notably, this hypothesis would be consistent with the fact that the two redox couples are generally not in thermodynamic equilibrium and the NADPH-to-NADP+ ratio is in a more reduced state than the NADH-to-NAD+ ratio34.

To theoretically assess this possibility, we first performed a thermodynamic analysis35 and calculated the Gibbs energies (¨GR) of the NAD-dependent malate decarboxylation and the

NADPH-dependent pyruvate carboxylation, based on measured metabolite concentrations during growth on glucose (Table 2). For the CO2 level, we assumed a concentration of 10 µM,

36 which represents the amount of dissolved CO2 under atmospheric conditions . This concentration can be rather considered as a lower bound, since the cellular CO2 production should feature even higher concentrations within the cell. If the minimal and maximum ¨GR for a reaction ranges between -10 and 10 kJ/mol, the reaction is defined as reversible32. According to this definition, the thermodynamic analysis suggested that the in vivo conditions result in an irreversible NAD+-dependent malate decarboxylation, while the NADPH-dependent reaction is reversible (Table 3).

118 Chapter 4

Table 2 Physiological concentrations of metabolites involved in the malic enzyme reaction. These values were both used for the thermodynamic analysis and to set the concentrations for the mass spectrometry based enzyme assays.

metabolite Concentration (mM) Standard deviation Malate 0.57a 0.14 Pyruvate 8.90a 5.75 NAD+ 1.09b 0.52 NADH 0.43b 0.21 NADP+ 0.19b 0.14 NADPH 0.36b 0.25 a estimates from Kleijn et. al. (2010)32. b Concentrations calculated from the determined cofactor ratios in Fuhrer et al. (2009)24 and absolute pool sizes of combined NAD+/NADH and NADP+/NADPH32.

Table 3 Gibbs energies of malate decarboxylation and pyruvate decarboxylation calculated based on in vivo concentrations (Table 2).

¨*NAD+-dependent malate decarboxylation ¨*1$'3+-dependent pyruvate carboxylation

-11a,c kJ/mol 6.1b,c kJ/mol

Standard deviations due to concentration measurement errors were a±6 kJ/mol and b8 kJ/mol. c Standard errors due to assumptions in the equilibrator software35 were ±6.2 kJ/mol.

Next, we set out to experimentally validate this finding and determined the ability of purified

YtsJ to compensate the presence of increasing concentrations of NADPH by generation of

NADH. Since NADH and NADPH cannot be distinguished spectrophotometrically, transhydrogenation assays were performed using flow injection electrospray-time-of-flight mass spectrometry26. Purified YtsJ was incubated at room temperature with a mixture of NAD+, malate and pyruvate at physiological concentrations (Table 2) and varying NADPH amounts above and below its in vivo concentration, and the metabolite changes followed dynamically by direct injection of the assay mixture into the mass spectrometer. Since all reactants have distinct mass-to-charge (m/z) ratios, we could follow their evolution over time.

Upon addition of YtsJ, we observed distinct temporal patterns, which were dependent on the concentration of NADPH. Within the first 15 minutes, malate levels decreased relatively fast in the presence of NADPH (Fig. 4a) while the redox cofactor concentrations remained constant

(Fig. 4b-d). After 15 minutes, once malate was almost consumed, NADPH levels started to

119 Chapter 4

decline (Fig. 4c) and NADP+ levels synchronously increased (Fig. 4d), suggesting NADPH oxidation via reductive pyruvate carboxylation. In the absence of NADPH, the decline in malate was much slower (Fig. 4a) and, although a decrease of pyruvate and NAD+ could not be measured within the monitored time range in any of the assays due to saturation of the respective signals (Supplementary Figure 1), this data provides strong evidence that malate was converted into pyruvate, because of the observed associated increase in NADH (Fig. 4b).

The rapid decrease in malate in the presence of NADPH was not coupled to NADH generation

(Fig. 4b), opening the question how malate conversion could proceed. Therefore, we searched for potential reaction products by using the data from the assay with 1 mM NADPH and performed unsupervised k-means clustering to analyze which metabolite ions showed increasing fold-changes over time (Supplementary Fig. 2). We specified three clusters, whereby cluster 1 contained ions strongly increasing over time and cluster 3 contained strongly decreasing ions. While cluster 3 contained only malate, cluster 1 contained NADP+ and, surprisingly, an ion with m/z of 89.03, annotated as lactate, which we subsequently confirmed by targeted LC-MS/MS analysis of the reaction mixture at the end of the assay23

(Supplementary Fig. 3). The lactate increase occurred also in the assays with less NADPH and showed a clear dependence on the NAPDH level, with highest lactate accumulation with the highest NADPH amount tested and absent lactate formation when no NADPH was present

(Fig. 4e).

120 Chapter 4

Figure 4. Mass spectrometry-based in vitro transhydrogenation assays with malic enzyme YtsJ. Time courses of metabolites (a) malate, (b) NADH, (c) NADP+, (d) NADPH and (e) lactate at different NADPH concentrations recorded by direct flow injection analysis. The dashed line indicates the two aforementioned phases of metabolite dynamics. Relative intensities represent intensities of the respective ion, normalized to the initial intensity. (f) Time courses of significantly changing metabolites in in vitro control assays where direct pyruvate to lactate conversion by YtsJ was tested. Error bars represent s.d. calculated using two independent assays.

121 Chapter 4

Catalytic malate decarboxylation to lactate and CO2 is typically found in lactic acid bacteria, where so-called malolactic enzymes catalyze the reaction without release of NAD(P)H37. To confirm that lactate formation occurred via malolactic enzyme activity and not through pyruvate reduction via potential lactate dehydrogenase side activity of YtsJ, we also performed the assay without addition of malate. In these assays, pyruvate was rapidly transformed to malate and lactate accumulation only started once malate had been formed (Fig. 4f). Additionally, the malate decrease and lactate increase showed again synchronous time courses, demonstrating that lactate formation occurred via malolactic enzyme and not lactate dehydrogenase activity of YtsJ. Interestingly we also detected simultaneous NADPH oxidation to NADP+ in the course of malate to lactate conversion in this assay, confirming that NADPH-oxidation and lactate formation occur in parallel, which allows for net degradation of NADPH, uncoupled from the formation of another redox cofactor.

A transhydrogenation cycle between YtsJ and lactate dehydrogenase could explain cellular NADPH homeostasis in vivo

The observed shift from malic enzyme to malolactic enzyme activity of YtsJ in the presence of physiological concentrations of NADPH, together with the result that NADPH oxidation and lactate formation occurred at a certain point in parallel (Fig. 4), suggested that YtsJ could similarly operate in vivo and thereby reoxidize the predicted NADPH excess in wild-type

B. subtilis. To test if YtsJ is indeed involved in lactate formation in vivo, we compared steady state metabolite levels of wild-type and the ytsJ deletion mutant during exponential growth on glucose minimal medium. While we detected no significant change in most metabolites around malate and pyruvate, lactate levels were significantly decreased in the ytsJ mutant (Fig. 5), indicating that YtsJ indeed contributes to cellular lactate formation. Since we did not observe lactate secretion into the culture supernatants of exponentially growing B. subtilis, the lactate produced by YtsJ seems to be reincorporated into metabolism. The most plausible route is depicted by lactate oxidation to pyruvate via lactate dehydrogenase (Ldh), which would be primarily coupled to NADH formation38. The fact that Ldh is expressed during growth on

122 Chapter 4

glucose17, would allow B. subtilis to employ this transhydrogenation cycle between YtsJ and

Ldh to reoxidize the predicted NADPH excess. Additionally, the operation of this cycle would yield a flux distribution similar to the one depicted in (Fig. 1g), since direct malate to pyruvate conversion results in the same labeling patterns as the malate-lactate-pyruvate route. Notably,

NADPH production and consumption were balanced in this situation.

Figure 5. Relative abundance and standard deviations of intracellular metabolites during mid-exponential growth

phase of B. subtilis wild-type and the ytsJ deletion mutant estimated by targeted LC-MS/MS analysis. Concentration

values are normalized to concentrations in the wild-W\SH3  3DLUHG6WXGHQW¶V7-test, unequal variance).

123 Chapter 4

Discussion

Since our current understanding of B. subtilis metabolism lacks an important mechanism for redox homeostasis, we investigated the previously suggested role of malic enzymes.

Quantifying in vivo NADPH-formation with stationary 13C metabolic flux analysis, we found that a predicted NADPH-overproduction of wild-type was absent upon deletion of malic enzyme ytsJ, due to a reshuffling of NADPH producing fluxes, indicating that this enzyme constitutes a mechanism for reoxidation of excess NADPH in the wild-type. The previously proposed operation of a redox cycle between YtsJ and any of the three other B. subtilis malic enzyme isoforms cannot explain this effect, since their deletion had no decisive impact on the cellular

NADPH-balance. Incorporating non-stationary flux data, we show that YtsJ operation is highly reversible and that this mechanism could close the redox balance if one assumes distinct cofactor specificities of the forward and the backward reaction. Specifically, the reductive pyruvate carboxylation would have to be NADPH specific, which we were able to confirm by in vitro enzymatic assays. Additionally, thermodynamic data supports the hypothesis that this

NADPH-oxidation mechanism might also occur in vivo. Notably, pyruvate carboxylation by YtsJ seems not to play a significant role in anaplerosis, since it is not able to rescue the lethality of a pyruvate carboxylase mutant39.

Further investigating the redox balancing mechanism in vitro, we found that YtsJ changes its activity to a non-redox cofactor forming malolactic enzyme in the presence of NADPH.

Together with the identified NADPH-oxidation, this mechanism would avoid additional accumulation of NADPH and thereby allow for its net oxidation in vivo. To our knowledge, this represents the first demonstration of a biochemical mechanism by which a bacterial enzyme changes its function, depending on the redox state. Furthermore, metabolomics data support the hypothesis that YtsJ also performs this mechanism in vivo and suggest that B. subtilis employs a redox cycle between YtsJ and lactate dehydrogenase to sustain the cellular redox balance.

The reason why B. subtilis exhibits NADPH-overproduction and therefore requires this reoxidation mechanism in the first place could be due to several reasons. First, it might be a

124 Chapter 4

way to allow higher, energy generating flux through the TCA-cycle, by partially decoupling

NADPH-production of isocitrate dehydrogenase. This hypothesis is supported by the finding that the ytsJ deletion was overcome by reduced TCA-cycle fluxes in the mutant. Second, the high NADPH generating flux might be required to maintain the cellular redox state and to ensure a high NADPH/NADP+-ratio40. Since it has been shown that sporulation of B. subtilis is accompanied by a drop in redox cofactors NAD(H) and NADP(H)41 and that a major initiator of spore formation, KinA, contains domains to sense changes in redox potential42, the high

NADPH generation might be a safety mechanism to ensure no initiation of sporulation during unlimited growth.

Our findings lead us to hypothesize that YtsJ is not important as a sheer malic enzyme to produce pyruvate during gluconeogenesis, but that it might rather act as a ³valve´ in redox homeostasis. Therefore, we propose the following model of YtsJ-dependent redox homeostasis in B. subtilis (Fig. 6). During exponential growth, NADPH-production exceeds the anabolic demands and the NADP+/NADPH pool is in a more reduced state1,24, which would favor NADPH oxidation by YtsJ and at the same time change its activity, by a yet unknown mechanism, from a malic to a malolactic enzyme to avoid further NADPH accumulation. On the contrary, this mechanism would also provide the potential to compensate for abrupt drops in the NADPH/NADP+ ratio, as they might occur for instance upon oxidative stress43, since this in turn would favor YtsJ malic enzyme activity with concomitant NADPH production. This hypothesis is favored by the finding that ytsJ is constitutively expressed during growth on several carbon sources, which result in varying NADPH-producing fluxes44. However, the fact that deletion of ytsJ causes a marked growth defect with malate as carbon source, which is probably because of its missing malic enzyme activity and NADPH-generation3, indicates that the NADP+/NADPH ratio is not the only regulator of YtsJ enzyme activity. Most likely, the high malate concentration drives the malic enzyme activity under this condition, indicating that the activity of YtsJ is potentially determined by the concentrations of all its substrates and products.

The observation that deletion of ytsJ only led to a moderate growth defect during growth on glucose might reflect the general flexibility of B. subtilis metabolism to perturbations of the

125 Chapter 4

redox balance. Although B. subtilis contains no transhydrogenase, this activity was detected in cell extracts, even upon deletion of ytsJ16. Without an annotated gene for a transhydrogenase, this indicates that further transhydrogenation cycles exist, most probably formed by continuous cycling between NADPH-requiring anabolism and NADH-producing catabolism of amino acids, which unfortunately is still virtually impossible to quantify on a global scale with current 13C metabolic flux approaches. However, non-stationary 13C flux ratio analysis might provide the potential to validate the existence of certain hypothesized cycles.

NADPH balances of several distinct bacterial species have been investigated and most of them show an imbalance towards catabolic overproduction of NADPH24. Since the majority of these organisms do not contain transhydrogenases, it becomes clear that other balancing mechanisms must exist in these species. It is tempting to speculate that the enzymatic properties of YtsJ may be a well-conserved feature within bacterial metabolism. This hypothesis would be supported by the fact that all of these species contain malic enzymes and also feature relatively high malic enzyme fluxes during growth on glucose10. Therefore, future studies are required to investigate if the identified mechanism is a common feature within certain bacteria.

Figure 6. Proposed model of YtsJ-dependent redox homeostasis. Exponential growth on glucose features high

NADPH-production, which leads to NADPH concentrations that favor NADPH oxidation by YtsJ and at the same time change its activity to a malolactic enzyme to avoid further NADPH accumulation. At the same time, this mechanism would be able to provide NADPH during sudden drops in the NADPH/NADP+ ratio, since these conditions favor the malic enzyme activity of YtsJ with concomitant NADPH production.

126 Chapter 4

References

1. Nelson, D., Lehninger, A. & Cox, M. Lehninger principles of biochemistry. (W. H. Freeman; 5th edition, 2008).

2. Haddock, B. A. Microbial energetics. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 290, 329± 339 (1980).

3. Lerondel, G., Doan, T., Zamboni, N., Sauer, U. & Aymerich, S. YtsJ has the major physiological role of the four paralogous malic enzyme isoforms in Bacillus subtilis. J. Bacteriol. 188, 4727±36 (2006).

4. Bologna, F. P., Andreo, C. S. & Drincovich, M. F. Escherichia coli malic enzymes: two isoforms with substantial differences in kinetic properties, metabolic regulation, and structure. J. Bacteriol. 189, 5937±46 (2007).

5. Zelle, R. M., Harrison, J. C., Pronk, J. T. & van Maris, A. J. Anaplerotic role for cytosolic malic enzyme in engineered Saccharomyces cerevisiae strains. Appl. Environ. Microbiol. 77, 732±8 (2011).

6. Boles, E., de Jong-Gubbels, P. & Pronk, J. T. Identification and characterization of MAE1, the Saccharomyces cerevisiae structural gene encoding mitochondrial malic enzyme. J. Bacteriol. 180, 2875±82 (1998).

7. Neidhardt, F. C., Ingraham, J. L. & Schaechter, M. Physiology of the Bacterial Cell: A Molecular Approach. (Sinauer Associates Inc., Sunderland, MA, 1990).

8. Blank, L. M., Lehmbeck, F. & Sauer, U. Metabolic-flux and network analysis in fourteen hemiascomycetous . FEMS Yeast Res. 5, 545±58 (2005).

9. Bakker, B. M. et al. Stoichiometry and compartmentation of NADH metabolism in Saccharomyces cerevisiae. FEMS Microbiol. Rev. 25, 15±37 (2001).

10. Fuhrer, T., Fischer, E. & Sauer, U. Experimental Identification and Quantification of Glucose Metabolism in Seven Bacterial Species. J. Bacteriol. 187, 1581±1590 (2005).

11. Hoek, J. B. & Rydström, J. Physiological roles of nicotinamide nucleotide transhydrogenase. Biochem. J. 254, 1±10 (1988).

12. Sauer, U., Canonaco, F., Heri, S., Perrenoud, A. & Fischer, E. The soluble and membrane-bound transhydrogenases UdhA and PntAB have divergent functions in NADPH metabolism of Escherichia coli. J. Biol. Chem. 279, 6613±9 (2004).

13. Sazanov, L. a & Jackson, J. B. Proton-translocating transhydrogenase and NAD- and NADP-linked isocitrate dehydrogenases operate in a substrate cycle which contributes to fine regulation of the tricarboxylic acid cycle activity in mitochondria. FEBS Lett. 344, 109±16 (1994).

14. Overkamp, K. M., Bakker, B. M., Steensma, H. Y., van Dijken, J. P. & Pronk, J. T. Two mechanisms for oxidation of cytosolic NADPH by Kluyveromyces lactis mitochondria. Yeast 19, 813±24 (2002).

127 Chapter 4

15. Tännler, S., Decasper, S. & Sauer, U. Maintenance metabolism and carbon fluxes in Bacillus species. Microb. Cell Fact. 7, 19 (2008).

16. Rühl, M., Le Coq, D., Aymerich, S. & Sauer, U. 13C-flux analysis reveals NADPH- balancing transhydrogenation cycles in stationary phase of nitrogen-starving Bacillus subtilis. J. Biol. Chem. 287, 27959±70 (2012).

17. Muntel, J., Fromion, V., Goelzer, A. & Hecker, M. Comprehensive Absolute Quantification of the Cytosolic Proteome of Bacillus subtilis by Data Independent, Parallel Fragmentation in Liquid Chromatography/Mass Spectrometry. Mol. Cell. Proteomics 1008±1019 (2014).

18. Doan, T. The Bacillus subtilis ywkA gene encodes a malic enzyme and its transcription is activated by the YufL/YufM two-component system in response to malate. Microbiology 149, 2331±2343 (2003).

19. Zamboni, N., Fendt, S.-M., Rühl, M. & Sauer, U. 13C-based metabolic flux analysis. Nat. Protoc. 4, 878±92 (2009).

20. Zamboni, N., Mouncey, N., Hohmann, H.-P. & Sauer, U. Reducing maintenance metabolism by metabolic engineering of respiration improves riboflavin production by Bacillus subtilis. Metab. Eng. 5, 49±55 (2003).

21. Zamboni, N., Fischer, E. & Sauer, U. FiatFlux - a software for metabolic flux analysis from 13C-glucose experiments. BMC Bioinformatics 6, 209 (2005).

22. Zimmermann, M., Sauer, U. & Zamboni, N. Quantification and Mass Isotopomer Profiling of Į-Keto Acids in Central Carbon Metabolism. Anal. Chem. (2014).

23. Buescher, J. M., Moco, S., Sauer, U. & Zamboni, N. Ultrahigh Performance Liquid Chromatography - Tandem Mass Spectrometry Method for Fast and Robust Quantification of Anionic and Aromatic Metabolites. Anal. Chem. 82, 4403±4412 (2010).

24. Fuhrer, T. & Sauer, U. Different biochemical mechanisms ensure network-wide balancing of reducing equivalents in microbial metabolism. J. Bacteriol. 191, 2112±21 (2009).

25. Zor, T. & Selinger, Z. Linearization of the Bradford protein assay increases its sensitivity: theoretical and experimental studies. Anal. Biochem. 236, 302±8 (1996).

26. Fuhrer, T., Heer, D., Begemann, B. & Zamboni, N. High-throughput, accurate mass metabolome profiling of cellular extracts by flow injection-time-of-flight mass spectrometry. Anal. Chem. 83, 7074±80 (2011).

27. Hörl, M., Schnidder, J., Sauer, U. & Zamboni, N. Non-stationary 13C-metabolic flux ratio analysis. Biotechnol. Bioeng. 110, 3164±3176 (2013).

28. Gourdon, P., Baucher, M., Lindley, N. D. & Guyonvarch, A. Cloning of the Malic Enzyme Gene from Corynebacterium glutamicum and Role of the Enzyme in Lactate Metabolism. Appl. Environ. Microbiol. (2000).

29. Hansen, E. & Juni, E. Isolation of mutants of Escherichia coli lacking NAD-and NADP- linked malic enzyme activities. Biochem. Biophys. Res. Commun. 65, 559±566 (1975).

128 Chapter 4

30. Dunn, M. F. Tricarboxylic acid cycle and anaplerotic enzymes in rhizobia. FEMS Microbiol. Rev. 22, 105±23 (1998).

31. Driscoll, B. T. & Finan, T. M. NADP+-dependent malic enzyme of Rhizobium meliloti. J. Bacteriol. 178, 2224±2231 (1996).

32. Kleijn, R. J. et al. Metabolic fluxes during strong carbon catabolite repression by malate in Bacillus subtilis. J. Biol. Chem. 285, 1587±96 (2010).

33. Fendt, S.-M. et al. Tradeoff between enzyme and metabolite efficiency maintains metabolic homeostasis upon perturbations in enzyme capacity. Mol. Syst. Biol. 6, 356 (2010).

34. Harold, F. M. The Vital Force: A Study of Bioenergetics. (Freeman and Company, New York, NY., 1986).

35. Flamholz, A., Noor, E., Bar-Even, A. & Milo, R. EQuilibrator - The biochemical thermodynamics calculator. Nucleic Acids Res. 40, 770±775 (2012).

36. Falkowski, P. G. & Raven, J. A. Aquatic . (Princeton University Press, 2007).

37. Caspritz, G. & Radlert, F. Malolactic enzyme of Lactobacillus plantarum. J. Biol. Chem. 258, 4907±4910 (1983).

38. Romero, S., Merino, E., Bolívar, F., Gosset, G. & Martinez, A. Metabolic engineering of Bacillus subtilis for ethanol production: Lactate dehydrogenase plays a key role in fermentative metabolism. Appl. Environ. Microbiol. 73, 5190±5198 (2007).

39. Diesterhaft, M. D. & Freese, E. Role of pyruvate carboxylase, phosphoenolpyruvate carboxykinase, and malic enzyme during growth and sporulation of Bacillus subtilis. J. Biol. Chem. 248, 6062±6070 (1973).

40. Fuhrer, T. & Sauer, U. Different biochemical mechanisms ensure network-wide balancing of reducing equivalents in microbial metabolism. J. Bacteriol. 191, 2112±21 (2009).

41. Chubukov, V. & Sauer, U. Environmental dependence of stationary-phase metabolism in Bacillus subtilis and Escherichia coli. Appl. Environ. Microbiol. 80, 2901±2909 (2014).

42. Piggot, P. J. & Hilbert, D. W. Sporulation of Bacillus subtilis. Curr. Opin. Microbiol. 7, 579±586 (2004).

43. Kühne, A. et al. Acute activation of oxidative pentose phosphate pathway as first-line response to oxidative stress in human skin cells. Mol. Cell. 59, 359-71 (2015).

44. Chubukov, V. et al. Transcriptional regulation is insufficient to explain substrate-induced flux changes in Bacillus subtilis. Mol. Syst. Biol. 9, 709 (2013).

129 Chapter 4

Supplementary Figures

Supplementary Figure 1. Time courses of metabolites pyruvate (a) and NAD+ (b) in mass spectrometry-based in vitro transhydrogenation assays.

Supplementary Figure 2. K-means clustering of the log-transformed annotated ion responses of both assays with

1mM NADPH, normalized to the initial time point. Three clusters were specified and ions were assigned to the clusters based on squared Euclidean distance. The black line represents the centroid of each cluster. Cluster 1 of strongly increasing ions contained lactate and NADP+, cluster 3 of strongly decreasing ions contained malate as deprotonated (m/z = 133.01) ion and ion with loss of water (m/z = 115).

130 Chapter 4

Supplementary Figure 3. Confirmation of ion m/z = 89.03 in enzyme assay mixture as lactate by MRM analysis using targeted LC-MS/MS.

Supplementary Tables

Supplementary Table 1 Enzymatic parameters of the four B. subtilis malic enzymes.

Pyruvatea NADHa NADPHa Malateb NAD+b NADP+b Protein Km kcat Km kcat Km kcat Km kcat Km kcat Km kcat

MaeA ndc nd nd nd nd nd 1.55 54 2.80 3 0.055 4 MalS nd nd nd nd nd nd 1.56 47 1.00 9 nd nd MleA nd nd nd nd nd nd 3.52 56 5.50 495 7.3 66 YtsJ 5.3 4.3 nd nd 0.8 7.8 3.95 75 5.30 159 6.7 54 a this study b Lerondel et al. (2006)3 c not detected

131 132 Conclusions and Outlook

Conclusions and Outlook

133 Conclusions and Outlook

Technical developments

The concerted interplay of factors governing metabolic responses to environmental and genetic changes is reflected by the in vivo molecular fluxes through the metabolic network.

Existing methods of 13C flux analysis allow the routine estimation of fluxes in central carbon metabolism of well-studied model organisms grown in minimal media with a single carbon source, but do not enable flux estimation during metabolic transitions and in complex media.

In this thesis, we developed non-stationary 13C metabolic flux ratio analysis, an approach that has the potential to overcome the aforementioned limitations by estimating the relative contribution of pathways to the synthesis of a common metabolite from dynamic 13C labeling data. The approach offers unique advantages compared to the existing flux analysis methods.

First, it follows a generalized workflow that is widely applicable to all cellular systems and any part of the network. Second, its locality principle permits flux estimation without topological knowledge of the metabolic network outside the considered metabolic node. Thereby, it avoids the measurement of extracellular rates, decreases the amount of 13C measurements to a small number of intermediates and reduces the computational cost of the parameter estimation process by relying on relatively small systems of ordinary differential equations. Third, it allows flux estimation from very short time-scale labeling experiments, which provides the possibility to study dynamic systems where metabolic steady states are brief.

We successfully demonstrated our approach as a valid tool to estimate specific relative fluxes in central metabolism of bacterial and mammalian cells from dynamic labeling data of only few intermediates. This was possible by rigorously following our predefined workflow for data processing and parameter estimation, which is universal and flexible to be applied all over metabolism. Although, we cannot precisely determine how applicable non-stationary 13C

MFRA is outside central metabolism with the data generated in this thesis, there is no concrete reason to expect a poor performance. In peripheral metabolism canonical, stationary methods fail to estimate fluxes, since pathways consist of long linear pathways or large cycles, where steady state patterns are independent of the flux. Here, non-stationary 13C flux methods are the only possibility to estimate fluxes and our approach has the potential for both the relative

134 Conclusions and Outlook quantification of fluxes at merging points of several pathways and, if concentration measurements are available, also for the estimation of absolute fluxes in linear pathways in the fashion of kinetic flux profiling1. An interesting case to demonstrate our approach in peripheral metabolism would be the investigation of coexisting anabolic and catabolic fluxes in the formation of certain amino acids. Continuous cycling between NADPH-dependent de novo biosynthesis and NAD+-dependent degradation has been suggested to rebalance cellular

NADPH overproduction in non-growing B. subtilis2, but could not be investigated with existing methodologies. Our approach would allow to quantify NADPH consumption by these mechanisms, given that anabolic and catabolic pathways proceed through different substrates.

Since there are no 13C flux methods that would allow for validation of the estimates obtained with our approach, its validity would have to be verified by showing that a deletion of the anabolic/catabolic enzymes reveals itself in a change of the estimated ratio.

In summary, we are convinced that our method also has the potential to resolve nodes in the peripheral network, given that the considered metabolic merging nodes comply with the requirements of our approach, i.e. the availability of labeling measurements of intermediates in close proximity to the node and knowledge of the local reaction topology. If these preconditions can be met, it is even conceivable to extend our approach beyond isolated ratios and build quantitative models also for smaller subnetworks within peripheral metabolism.

The proof of principle studies performed in this thesis mainly demonstrate how the method can improve the maximum achievable temporal resolution for flux estimation and its potential to greatly shorten the experimental duration required for labeling experiments. Specifically, we were able to obtain flux estimates for B. subtilis that were in good agreement with the results from previous, stationary methods, from data collected within less than one minute, compared to at least 30 to 60 minutes with previous approaches3. Also with mammalian cells, the experimental duration could be reduced from roughly 12 to 2 hours and even shorter experimental durations should be possible with more data points being sampled in the initial phase of tracer propagation. To fully exploit this temporal gain, the method can be applied to investigate flux rearrangements during slow transitional metabolic processes that might even

135 Conclusions and Outlook occur on a time-scale of only a few minutes, such as transcriptionally controlled activities. It would be interesting for instance, to investigate which dynamic flux rearrangements take place in B. subtilis during the mainly transcriptionally controlled switch from gluconeogenic operation with malate to co-metabolism of malate and glucose4. From a medical perspective, alterations in metabolic operation due to transcriptional deregulation have been increasingly recognized in cancer5, which makes transcription a promising drug target6,7. Here, our method could be used to study how metabolism in malignant cell lines responds upon exposure to these drugs.

The design of non-stationary 13C MFRA as a tool to formally validate specific hypotheses on local fluxes in a possibly targeted and quantitative way imposes certain limitations on the approach. First, our method is not suited for high-throughput screening and discovery of global differences in the response of an organism to environmental or genetic perturbations. Here, only global strategies, such as isotopomer balancing or comparative fluxome profiling8, provide the information to draw conclusions, e.g. on the cellular redox and energy state. Second, assessing the calculability of a given (relative) flux a priori by in silico calculations is in praxis difficult. The labeling dynamics of metabolites depend both on fluxes and pool sizes and the labeling states of their precursors, which in turn depend on the chosen tracer as shown in

Chapter 3. All these factors will affect flux calculability, but can hardly be assessed without experimental data. This issue can partly be addressed by performing preliminary labeling experiments, where only few samples are collected during the transients of 13C propagation.

This dataset should allow to identify the optimal sampling times to coarsely describe the labeling transients in the metabolites of interest and to select the ideal set of tracers that provide most labeling information. Third, our method only allows to estimate the ratio of sheer forward fluxes, while the net contribution of pathways might be a more relevant entity to describe metabolic operation. However, together with data on net fluxes, our approach provides information on flux reversibility, which initiated the discoveries of Chapter 4.

Compared to other local non-stationary 13C metabolic flux analysis approaches, our methodology provides certain advantages. Kinetic flux profiling (KFP) employs the labeling dynamics of reaction substrates and products to estimate absolute, local fluxes1,9. While the

136 Conclusions and Outlook major benefit of this methodology is the absolute, and not only relative, quantification of flux, it additionally requires the measurement of absolute metabolite concentrations and is less straightforward with multiple substrates and converging fluxes. A more recent extension of

KFP, called rKFP10, partially overcomes these challenges by estimating relative flux changes between two conditions from non-stationary labeling data and only relative, measured metabolite changes. Similar to our framework, rKFP has the advantage that it does not rely on absolute metabolite concentrations, however, it requires a reference state to contrast flux changes between different conditions, which cannot always be defined and requires additional measurements. In contrast, our method also allows to estimate relative fluxes for individual organisms and conditions.

How does non-stationary 13C MFRA extend 13C flux analysis of mammalian cells? Based on the results of this thesis, we found that both our framework and stationary approaches allowed for the accurate estimation of relative major fluxes in central metabolism of mammalian cells.

Using non-stationary 13C MFRA, the amount of tracers could be reduced, most reasonably explained by the additional information given by the labeling dynamics, but this comes at the cost of taking several time-resolved measurements. We concluded that if cells can be grown at metabolic steady state until isotopic stationarity is reached and if tracers are available that allow to distinguish the activity of distinct pathways, stationary methods seem to be advantageous, since they are less demanding in experimentation. However, for flux estimation during short-term biological phenomena and outside central metabolism, these methods provide no information, while non-stationary 13C MFRA can still give estimates. This is of significant importance, since the usage of secondary pathways to fuel intuitively not directly associated reactions, is a more and more recognized feature exploited by cancer cells. Specific examples are serine and glycine biosynthesis, which promote purine and Į-ketoglutarate formation in rapidly proliferating cancer types11,12. Both findings were based on initial large- scale functional genomics or metabolite profiling screens to generate these hypotheses, and subsequently verified by co-feeding specific 13C tracers. Similarly, our method can be applied to verify such hypothesized shifts towards secondary pathways in the formation of a common

137 Conclusions and Outlook product, by formally integrating all possible metabolic routes for the formation of the intermediate.

The compartmentation of higher cell metabolism can impose additional complexity on non- stationary 13C flux approaches, since multiple metabolite pools can coexist within one cell. For the system considered in this thesis, we did not find evidence of flux estimate artifacts, caused by ignoring compartments in the model. However, this was most likely due to the high fluxes in central metabolism, which feature rapid equilibration of mitochondrial and cytosolic pools, and cannot be treated as being general. Therefore, future applications of non-stationary 13C

MFA to metabolic nodes within different compartments should also consider compartmentalized pools and labelling dynamics. Since this information cannot be experimentally measured hitherto, it might be generated from the measured (convoluted) data in silico by Monte Carlo methods and concomitantly used to perform the parameter estimation.

A correct flux ratio would result in statistically highly significant estimates.

Based on the theoretical concepts underpinning our method and the experience accumulated in this thesis, non-stationary 13C MFRA is mostly suited for targeted flux analyses of selected metabolic nodes. It can be used to verify hypotheses on flux alterations, generated by large- scale screening techniques, such as non-targeted metabolomics and 13C fluxome profiling.

Additionally, it can be used to validate how effective are the desired flux changes through a manipulation of the cell, as for example by drug treatment or metabolic engineering. Future applications should further exploit the FDSDELOLWLHV RI WKH PHWKRG ,W¶V H[WUHPHO\ VKRUW experimental times combined with the small, computationally inexpensive ODE models, could be further combined with kinetic modeling strategies, to provide access to flux estimation in metabolically fully dynamic systems, such as fast post-translationally or allosterically regulated responses to nutrient changes4,13 or stresses14. Further developments might even make the method applicable to systems that would never reach isotopic stationarity, such as entire organs and even multicellular organisms.

138 Conclusions and Outlook

Biological findings

In addition to the technical developments, we also gained biological insights through the application of non-stationary and stationary 13C MFRA. We contributed novel insights into metabolic rearrangements caused by the disruption of mitochondrial pyruvate transport in mammalian cells (Chapter 3) and identified a novel potential mechanism to sustain the redox balance in B. subtilis (Chapter 4).

Deletion of the mitochondrial pyruvate carrier in mouse embryonic fibroblasts completely abolished the import of glucose derived pyruvate into mitochondria, which was mainly compensated by increased contribution of glutamine via the reductive TCA cycle. The fact that this pathway is usually active during hypoxia or in cells with an impaired respiratory chain15,16 indicates that respiratory activity was affected in this cell line, which will have to be shown by measuring pyruvate oxidation and determining absolute TCA cycle fluxes using the estimated ratios and measured extracellular rates. Notably, not the reductive TCA cycle, but the glutaminolysis pathway, i.e. the formation of pyruvate via malic enzyme and the concomitant generation of acetyl-CoA via pyruvate dehydrogenase was found to compensate the impaired mitochondrial pyruvate transport in another cell line17. Unfortunately, our current data does not allow to determine the activity of this pathway, because measurements of the required metabolites were poor due to low abundance. Repeating the labeling experiment with more biomass might solve this issue. Nevertheless, the discrepancy between previous and our results indicates that the metabolic response to the disruption of the MPC might be rather cell type specific and could depend on factors, such as the kinetic properties of the tissue-specific enzymes or their expression levels in the different cell lines.

Interestingly, we also revealed partial compensation of the impaired pyruvate transport by an increase in anaplerosis. Because of the missing pyruvate transport, mitochondrial pyruvate carboxylase, which appears to be the only known anaplerotic enzyme in mammalian cells hitherto18, seems not to be responsible and we speculate that a reversal of the reaction catalyzed by cytosolic malic enzyme ME1 explains our observation. To our knowledge, this

139 Conclusions and Outlook would be the first time such flux reversal of malic enzyme has been observed in mammalian cells in vivo. Further investigation is required to confirm the reverse activity of cytosolic malic enzyme, by simultaneously deleting MPC and ME1 in MEFs, which should lead to a disruption of the anaplerotic flux. If the reversal of malic enzyme could be confirmed, its general anaplerotic role in other cell lines should be explored. In this case, more elaborate deletion studies should only be performed after showing that the kinetic properties of ME1 allow the reversal of the reaction in the presence of all substrates at intracellular concentrations by in vitro assays.

In combination with stationary flux methods, we also identified reverse activity of one malic enzyme isoform in B. subtilis. However, here its role was not to contribute to anaplerosis, but to provide a mechanism that allows the bacterium to balance its catabolic NADPH overproduction under batch growth. Combining our non-stationary 13C MFRA results with flux analysis by global balancing, in vitro enzymatic assays and in vivo metabolomics measurements, we provide first evidence for a novel transhydrogenation mechanism for cellular NADPH balancing. Specifically, we found that malic enzyme YtsJ responds to the increasing NADPH levels by a change in function towards a secondary reaction, which produced lactate and would allow the net oxidation of excess NADPH. Although the kinetics of metabolic enzymes are known to be modified depending on the cellular NADPH level19, a change in enzyme function has not been reported hitherto, not even in the context of enzyme promiscuity20. We also found a reduction in intracellular lactate levels in the ytsJ deletion strain, which indicates that the reaction occurs in vivo. The fact that lactate is not secreted by

B. subtilis made us hypothesize that lactate dehydrogenase reincorporates the lactate produced by YtsJ into pyruvate, which would result in a net redox cycle between the two enzymes.

The question arises how the identified mechanism impacts our general understanding of bacterial redox metabolism. While our data strongly supports the existence of the mechanism in B. subtilis in vivo, we still miss the final evidence. Specifically, fluxes in the lactate dehydrogenase mutant should reveal the existence of our hypothesized redox cycle. In case

140 Conclusions and Outlook these results will be negative, YtsJ has to be tested for lactate dehydrogenase activity, to see if the enzyme itself facilitates the transhydrogenation cycle. If we anticipate positive results, further studies will be required to investigate if also other bacteria make use of the mechanism.

It is still unclear how several bacterial species maintain their redox balance and neither wrong assumptions on cofactor specificities, nor the existence of transhydrogenases seem to be the answer21. Other balancing mechanisms must exist in these species and it is tempting to speculate that the identified properties of the B. subtilis malic enzyme might be a well- conserved feature within bacterial metabolism. This hypothesis would be supported by the fact that all the investigated species contain malic enzymes and also feature relatively high malic enzyme fluxes during growth on glucose22. Additionally, it has to be explored, if the identified mechanism is a common feature within a certain group of bacteria and, if so, whether it arises from (i) specific enzymatic properties of certain malic enzymes, out of (ii) environmental factors or (iii) from features of the network. Therefore, the kinetic properties of malic enzymes of organisms that (i) share high sequence similarity to YtsJ or (ii) populate similar habitats to B. subtilis need to be characterized in vitro and intracellular metabolic fluxes estimated in wild- type vs. malic enzyme deletion strains. Comparing metabolic fluxes in bacteria with a similar inventory of metabolic reactions, e.g. no glyoxylic shunt etc., and equal redox cofactor specificities of enzymes, might reveal if the network structure induces the mechanism.

141 Conclusions and Outlook

References

1. Yuan, J., Bennett, B. D. & Rabinowitz, J. D. Kinetic flux profiling for quantitation of cellular metabolic fluxes. Nat. Protoc. 3, 1328±40 (2008).

2. Rühl, M., Le Coq, D., Aymerich, S. & Sauer, U. 13C-flux analysis reveals NADPH- balancing transhydrogenation cycles in stationary phase of nitrogen-starving Bacillus subtilis. J. Biol. Chem. 287, 27959±70 (2012).

3. Rühl, M., Zamboni, N. & Sauer, U. Dynamic flux responses in riboflavin overproducing Bacillus subtilis to increasing glucose limitation in fed-batch culture. Biotechnol. Bioeng. 105, 795±804 (2010).

4. Buescher, J. M. et al. Global Network Reorganization During Dynamic Adaptations of Bacillus subtilis Metabolism. Science. 335, 1099±1104 (2012).

5. Jiang, P., Du, W., Mancuso, A., Wellen, K. E. & Yang, X. Reciprocal regulation of p53 and malic enzymes modulates metabolism and senescence. Nature 493, 689±93 (2013).

6. Redmond, A. M. & Carroll, J. S. Defining and targeting transcription factors in cancer. Genome Biol. 10, 311 (2009).

7. Vassilev, L. T. et al. In vivo activation of the p53 pathway by small-molecule antagonists of MDM2. Science 303, 844±848 (2004).

8. Zamboni, N. & Sauer, U. Model-independent fluxome profiling from 2H and 13C experiments for metabolic variant discrimination. Genome Biol. 5, R99 (2004).

9. Yuan, J., Fowler, W. U., Kimball, E., Lu, W. & Rabinowitz, J. D. Kinetic flux profiling of nitrogen assimilation in Escherichia coli. Nat. Chem. Biol. 2, 529±30 (2006).

10. Huang, L., Kim, D., Liu, X., Myers, C. R. & Locasale, J. W. Estimating Relative Changes of Metabolic Fluxes. PLoS Comput. Biol. 10, 1±11 (2014).

11. Possemato, R. et al. Functional genomics reveal that the serine synthesis pathway is essential in breast cancer. Nature 476, 346±50 (2011).

12. Jain, M. et al. Metabolite profiling identifies a key role for glycine in rapid cancer cell proliferation. Science 336, 1040±4 (2012).

13. Link, H., Kochanowski, K. & Sauer, U. Systematic identification of allosteric protein- metabolite interactions that control enzyme activity in vivo. Nat. Biotechnol. 31, 357±61 (2013).

14. Kühne, A. et al. Acute activation of oxidative pentose phosphate pathway as first-line response to oxidative stress in human skin cells. Mol. Cell. 59, 359-71 (2015).

15. DeBerardinis, R. J. et al. Beyond aerobic glycolysis: transformed cells can engage in glutamine metabolism that exceeds the requirement for protein and nucleotide synthesis. Proc. Natl. Acad. Sci. U. S. A. 104, 19345±50 (2007).

142 Conclusions and Outlook

16. Metallo, C. M. et al. Reductive glutamine metabolism by IDH1 mediates lipogenesis under hypoxia. Nature 481, 380±4 (2012).

17. Vacanti, N. M. et al. Regulation of Substrate Utilization by the Mitochondrial Pyruvate Carrier. Mol. Cell 56, 425±435 (2014).

18. Owen, O. E., Kalhan, S. C. & Hanson, R. W. The key role of anaplerosis and cataplerosis for citric acid cycle function. J. Biol. Chem. 277, 30409±30412 (2002).

19. Holten, D., Procsal, D. & Chang, H. L. Regulation of pentose phosphate pathway dehydrogenases by NADP+/NADPH ratios. Biochem. Biophys. Res. Commun. 68, 436± 441 (1976).

20. Hult, K. & Berglund, P. Enzyme promiscuity: mechanism and applications. Trends Biotechnol. 25, 231±238 (2007).

21. Fuhrer, T. & Sauer, U. Different biochemical mechanisms ensure network-wide balancing of reducing equivalents in microbial metabolism. J. Bacteriol. 191, 2112±21 (2009).

22. Fuhrer, T., Fischer, E. & Sauer, U. Experimental Identification and Quantification of Glucose Metabolism in Seven Bacterial Species. 187, 1581±1590 (2005).

143 Abbreviations !

Abbreviations

6PG 6-phosphogluconate

13C MFA 13C metabolic flux analysis

13C MFRA 13C metabolic flux ratio analysis

AKG alpha-ketoglutarate

ATP

B. subtilis Bacillus subtilis

C carbon c concentration

CDW cell dry weight

CoA coenzyme A

DHAP dihydroxyacetone phosphate

DNA deoxyribonucleic acid

E4P erythrose-4-phosphate

ENO enolase

F6P fructose-6-phosphate

FBA flux balance analysis

FL fractional labeling

G6P glucose-6-phosphate

GAP glyceraldehyde-3-phosphate

GC gas chromatography h hour

HPLC high-performance liquid chromatographie

ID isotopomer distribution

LB Luria-Bertani lb lower bound

LC liquid chromatography

144! Abbreviations ! Ldh lactate dehydrogenase

MAE malic enzyme

MaeA malic enzyme MaeA

MAL malate

MalS malic enzyme MalS

MDH malate dehydrogenase

MEF mouse embryonic fibroblast

MID mass isotopomer distribution min minute

MleA malic enzyme MleA

Mpc1KO mouse embryonic fibroblast cell line with deletion of mitochondrial pyruvate

carrier unit 1 mRNA messenger ribonucleic acid

MS mass spectrometry

N nitrogen

NAD+ nicotinamide adenine dinucleotide oxidized form

NADH nicotinamide adenine dinucleotide reduced form

NADP+ nicotinamide adenine dinucleotide phosphate oxidized form

NADPH nicotinamide adenine dinucleotide phosphate reduced form

OAA oxaloacetate

OD optical density

ODE ordinary differential equation

P pool size

PC pyruvate carboxylase (higher cells)

PCK phosphoenolpyruvate carboxykinase

PDH pyruvate dehydrogenase

PEP phosphoenolpyruvate

PPP pentose phosphate pathway

145 Abbreviations ! PYC pyruvate carboxylase (bacteria)

PYK pyruvate kinase qX uptake or secretion rate of compound X

Ru5P ribulose-5-phosphate

R5P ribose-5-phosphate

S7P sedoheptulose-7-phosphate

SDS-page sodium dodecyl sulfate polyacrylamide gel electrophoresis

TBDMS tert-butyldimethylsilyl

TCA cycle tricarboxylic acid cycle ub upper bound

UPLC ultra performance liquid chromatography

WT wild-type

X5P xylulose 5-phosphate

µ specific growth rate

Ȥ2 weigthed sum of squared residuals

146! Acknowledgements ! Acknowledgements

The presented work would have never been possible without the contributions and continuous support by many people, whom I would like to thank at this point.

Dr. Nicola Zamboni, who gave me the opportunity to work in his group and for guiding my ideas in the right directions. Your power of endurance in telling me that I am an engineer and that I should work and think accordingly made this thesis happen.

Prof. Dr. Uwe Sauer, whose guidance had a large share in my development throughout the project. Thanks for running the lab the way you do and for sharing my humor (especially about the fact that Martin doesn’t know what Leberkäs is).

Prof. Dr. Julia Vorholt, who was a member of my PhD committee, for her scientific input and her encouragement, especially after the second PhD committee meeting.

The „Zambonis“ Sébastien Dubuis, Maria Kogadeeva, Petra Krznar, Andreas Kühne and

Leila Alexander for being great colleagues. Of course also the entire Sauer lab, especially

Dimitris Christodoulou, Tobias Fuhrer and Mattia Zampieri, who contributed to this work.

I hope and believe that each individual of you Zambonis and Sauers knows what she/he means to me and what a great time I had with all of you the last five years.

My students Florian Hotz, Dan Mossing and Nicola Müller, for their good work and the opportunity to teach and learn.

Prof. Dr. Johann Kohmann, who was my scientific mentor from childhood on and paved the road to where I am now by exciting my curiosity for biotechnology. Danke Hans!

My mom and my dad, for their unlimited support throughout my entire life, their believe in me and for being the best parents in the world.

Mama, Papa, Danke für Eure Unterstützung seit mehr als 30 Jahren, dass Ihr an mich glaubt und dafür, dass Ihr die Besten seid!

147 Acknowledgements

! My brother Dominik, for showing me that what I do is cool by studying the same field and for being the best brother I can imagine.

All my friends from home, especially Hannes Lechner, Christoph Ostermair and Raimund

Strauss, for distraction from the scientific world when I escaped to the Free State.

Lisa, for following me to Zurich, for being who you are and for being there for me – in general and especially during the last months of my PhD.

148 Curriculum Vitae !

Curriculum Vitae !

Manuel Hörl Limmattalstrasse 209 8049 Zürich

PERSONAL DATA

Date of Birth: 18.09.1984 Nationality: German

EDUCATION

Since 08.2010 Institute of Molecular Systems Biology, ETH Zürich PhD candidate in Systems Biology with Dr. Nicola Zamboni

03.2009 – 08.2009 Energy Biosciences Institute, UC Berkeley Research project

10.2005 – 05.2010 Diploma in Chemical Engineering, TU München Specialization: Biotechnology

09.1995 – 05.2004 Ignaz-Taschner-Gymnasium, Dachau Abitur; orientation: mathematics and science

PUBLICATIONS

Hörl, Schnidder, Sauer, Zamboni Biotechnology and Bioengineering, 110(12), 3164-3176, 2013 Non-stationary 13C-metabolic flux ratio analysis.

Mirtschink, Krishnan, Grimm, Sarre, Hörl, Kayikci, Fankhauser, Christinat, Cortijo, Feehan, Vukolic, Sossalla, Stehr, Ule, Zamboni, Pedrazzini, Krek Nature, 522, 444-9, 2015 HIF-driven SF3B1 induces KHK-C to enforce fructolysis and heart disease.

149 Curriculum Vitae

! PUBLIC PRESENTATIONS

Dechema conference: Trends in Metabolomics, Frankfurt a.M., Germany, June 2014. Combination of stationary and non-stationary 13C flux methods reveals a novel redox cofactor balancing mechanism in Bacillus subtilis. (short talk and poster)

International Conference of the Metabolomics Society, Washington DC, USA, June 2012. Generalized and exact 13C metabolic flux ratio analysis for complex systems. (talk)

Dechema conference: Trends in Metabolomics, Frankfurt a.M., Germany, May 2011. Estimating metabolic flux ratios from dynamic 13C labeling data. (poster)

150