What Happens in Yeast During the Crabtree Effect? an Investigation of Work Are Discussed

8 APPLIED MICROBIOLOGY

Volume 8

In this thesis a metabolic model of S. cerevisiae was enhanced with the biochemistry of volatile metabolites that have been found by a literature investigation. Not only the volatile metabolites in question have been added, but also substances and reactions that Christoph Halbfeld connect them to metabolites already present in the model. In total, 225 metabolites and 219 reactions were added to the model. Furthermore, 12 metabolic reactions could be verified by physiological and enzyme assays of knockout mutants. What happens in yeast during

Additionally, volatile metabolite dynamics during the induction of the Crabtree effect in a fully respiratory growing continuous culture of S. cerevisiae were explored with real- the Crabtree effect? time analyses of the fermentation off-gas. Secondary electrospray ionization-Orbitrap- mass spectrometry was used for this endeavor. In these measurements, I detected about An investigation of S. cerevisiae’s 2,500 signals of which 16 showed a response to the perturbation of the metabolic state prior to the detection of ethanol. volatile space Furthermore, the possibility of online volatile metabolite monitoring due to multi capillary column–ion mobility spectrometry (MCC-IMS) analysis of yeast fermentation off-gas was established. The MCC-IMS used was developed for the detection of volatiles in human breath, and several technical adaptations were applied to allow robust detection of volatiles in the headspace of yeast fermentations. volatile space The MCC-IMS in its optimized configuration was applied to monitor volatile metabolite changes of a laboratory and an industrial yeast strain during the Crabtree effect. In addition, metabolic differences in this setting were examined on transcriptional level using a cDNA microarray. The metabolic shift could be observed in the volatile space of both S. cerevisiae’s strains and in all tested conditions. The transcriptome showed differences in the leucine and isoleucine pathways, as well as in genes related to the TCA cycle and the respiratory pathways. Interestingly, the expression data indicated that the industrial strain upregulated its respiration during the shift, while it was downregulated in the laboratory strain.

Lastly, possible applications for the knowledge gained and methods developed in this What happens in yeast during the Crabtree effect? An investigation of work are discussed. This thesis provides a blueprint for studies of the volatile space in other organisms. For example, the knowledge gained about the changes in the volatilome during metabolic transitions could be used to online determine and potentially control the metabolic state of a yeast. Christoph Halbfeld

“What happens in yeast during the Crabtree effect? An investigation of S. cerevisiae's volatile space”

Von der Fakultät für Mathematik, Informatik und Naturwissenschaften der RWTH Aachen University zur Erlangung des akademischen Grades eines Doktors der Naturwissenschaften genehmigte Dissertation

vorgelegt von

M. Sc. Biologie Christoph Halbfeld

aus

Moers, Nordrhein-Westfalen, Deutschland

Berichter: Universitätsprofessor Dr.-Ing. Lars M. Blank

Universitätsprofessor Dr. rer. nat. Jörg Ingo Baumbach

Tag der mündlichen Prüfung: 12.04.2018

Diese Dissertation ist auf den Internetseiten der Universitätsbibliothek online verfügbar.

Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über https://portal.dnb.de abrufbar.

Christoph Halbfeld:

What happens in yeast during the Crabtree effect? An investigation of S. cerevisiae’s volatile space

1. Auflage, 2018

Gedruckt auf holz- und säurefreiem Papier, 100% chlorfrei gebleicht.

Apprimus Verlag, Aachen, 2018 Wissenschaftsverlag des Instituts für Industriekommunikation und Fachmedien an der RWTH Aachen Steinbachstr. 25, 52074 Aachen Internet: www.apprimus-verlag.de, E-Mail: [email protected]

Printed in Germany

ISBN 978-3-86359-635-4

D 82 (Diss. RWTH Aachen University, 2018)

Danksagung

Zuerst möchte ich mich bei Prof. Lars M. Blank für die Erstkorrektur meiner Arbeit und all die Unterstützung, die ich von ihm erhalten habe bedanken. Außerdem möchte ich mich für die Möglichkeit meine Arbeit am iAMB durchführen zu dürfen bedanken. Mein Dank gilt ebenfalls Prof. Jörg-Ingo Baumbach für die Zweitkorrektur meiner Arbeit. Außerdem möchte ich mich dafür bedanken, dass ich in der B&S Analytik die Möglichkeit hatte erste Erfahrungen mit dem IMS zu sammeln und jederzeit für weitere Messungen willkommen war. Außerdem möchte ich mich für die vielen aufschlussreichen Gespräche über das IMS bedanken. Ich möchte mich ebenfalls bei Prof. Alan Slusarenko für die Übernahme der Drittprüferschafft bedanken. Ich danke ebenfalls meiner Betreuerin Dr. Birgitta E. Ebert an die ich mich jederzeit wenden konnte. Durch ihre Unterstützung war ich in der Lage die Art, wie ich Experimente plane und Texte schreibe wesentlich zu verbessern. Vielen Dank ebenfalls an alle Projektpartner für die gute Zusammenarbeit. Jessica Kuhlmann danke ich für die gute Zusammenarbeit bei der Verbesserung des IMS und der IMS-Analytik. Ann-Kathrin Sippel möchte ich für die Zusammenarbeit bei der IMS-Analytik und der Vermessung von GC- Proben danken. Prof. Sven Rahmann und Elias Kuthe möchte ich für die Hilfe bei der Auswertung der cDNA-Microarrays danken. Sven Wegerhoff und Prof. Sebastian Engell möchte ich für die interessanten Diskussionen während den Projekttreffen danken. Dr. Michael Quantz und Dr. Erik Pollmann möchte ich für die immerwährende Hilfe in der Erkundung der unendlich großen Hefeliteratur, für die Hilfe bei Fermentationsproblemen sowie für die gute Zusammenarbeit und die Unterbringung bei der VH-Berlin danken. Mein Dank gilt ebenfalls Prof. Pablo Martinez-Lozano Sinues für die Möglichkeit Experimente an der SESI-Orbitrap-MS in Zürich durchzuführen. Ebenfalls danke ich der Zenobi-Gruppe für die freundliche Arbeitsatmosphäre. Außerdem danke ich allen Studenten, die im Verlauf meiner Arbeit mit mir zusammengearbeitet haben: Christina Redmers, Daniela Dey, Christiane Sonntag, Sandrine Nankia Tatepo, Kai Büchner, Mathis Wolter, Niklas Kitschen und Birthe Halmschlag. Durch sie wurde meine Arbeit bereichert, da mehr Arbeitspakete bearbeitet werden konnten, als ich alleine hätte schaffen können. Ebenfalls danke ich allen, die während meiner Zeit am iAMB gearbeitet haben, durch alle wurde eine einzigartige Arbeitsatmosphäre geschaffen, die die oftmals langen Tage teils sehr unterhaltsam gemacht haben. Besonders möchte ich hierbei Dr. Thiemo Zambanini, Benedikt Wienands, Hamed Tehrani, Mathias Lehnen, Vanessa Bayer und Birthe Halmschlag danken, durch sie wurde die Zeit unvergesslich. Ebenfalls möchte ich mich bei Eik Czarnotta für die Starthilfe bei Hefefermentationen bedanken. Ein besonderer Dank gilt Dr. Martin Zimmermann, der mich schon während meines Studiums, aber auch in der Phase der Promotion immer wieder mit motivierenden Gesprächen auf dem Weg begleitet hat. Ich möchte mich auch bei den Mitarbeitern der Mechanischen und Elektrotechnik Werkstatt bedanken. Mit Hilfe des Teams war ich dazu in der Lage Ideen zu realisieren, die ich alleine nicht hätte umsetzen können. Ich möchte ebenfalls meiner Freundin Dr. Sandra Jumpertz danken, sie hat mich auch in schwierigen Phasen der Promotion immer aufgebaut und mir die Stärke gegeben die Promotion erfolgreich

abzuschließen. Zuletzt möchte ich meiner Familie danken, die es mir ermöglicht hat diesen Weg zu beschreiten und mir immer Rückhalt gegeben haben.

Funding

The project was funded by the German Federal Ministry of Education and Research (BMBF).

Eidesstattliche Erklärung

„Hiermit erkläre ich, Christoph Halbfeld, an Eides statt, dass ich die vorliegende Dissertation selbstständig verfasst und keine anderen als die angegebenen Quellen und Hilfsmittel benutzt habe.“

Table of Contents

SUMMARY ...... V

ZUSAMMENFASSUNG ...... VII

LIST OF ABBREVIATIONS ...... IX

LIST OF FIGURES ...... XIII

LIST OF TABLES ...... XV

GENERAL INTRODUCTION ...... 5 SUMMARY ...... 5 INTRODUCTION ...... 5 VOLATILE ORGANIC COMPOUNDS ...... 5 BREAD ...... 6 WINE ...... 6 Wine flavors...... 6 Off-flavors ...... 7 FUNGAL VOCS ...... 8 METABOLIC PATHWAYS OF YEAST ...... 9 VOCs derived from carbohydrate catabolism and fermentation ...... 9 VOCs produced by amino acid synthesis and degradation ...... 10 VOCs from fatty acids biosynthesis and degradation...... 11 VOCs derived from terpene biosynthetic pathways ...... 12 ANALYTICAL METHODS FOR THE IDENTIFICATION OF VOCS ...... 14 ION MOBILITY SPECTROMETRY ...... 14 SECONDARY ELECTROSPRAY IONIZATION MASS SPECTROMETRY ...... 17 PROTON-TRANSFER-REACTION TIME-OF-FLIGHT MASS SPECTROMETRY ...... 17 GAS-CHROMATOGRAPHY BASED METHODS ...... 18 OFFLINE-SAMPLING ...... 19 PROS AND CONS OF METHODS FOR VOC DETECTION ...... 21 METHODS FOR THE DELINEATION OF VOC METABOLIC PATHWAYS ...... 22 BOTTOM-UP APPROACHES ...... 23 VERIFICATION OF PREDICTED PATHWAYS...... 24 AIMS ...... 25

RESULTS ...... 33 EXPANDING THE CONSENSUS YEAST MODEL BY VOLATILE METABOLISM ...... 33 SUMMARY ...... 33 INTRODUCTION ...... 34 MATERIAL AND METHODS ...... 35 Software tools for pathway prediction ...... 35 Strains, media and plasmids used ...... 35

I Table of Contents

Enzymes and PCR components ...... 36 Commercial kits ...... 39 Creating multiple knockout mutants ...... 39 Biotransformation assay ...... 40 Data evaluation ...... 40 RESULTS AND DISCUSSION ...... 41 CONCLUSION AND OUTLOOK ...... 50 YEAST VOLATILOME DYNAMICS DURING METABOLIC SHIFTS ...... 55 SUMMARY ...... 55 INTRODUCTION ...... 56 Dynamics of (volatile) compounds ...... 56 MATERIAL AND METHODS ...... 58 Yeast strains and growth conditions ...... 58 Online SESI-Orbitrap MS measurements ...... 60 Data evaluation: ...... 61 RESULTS AND DISCUSSION ...... 61 Interpretation of the results ...... 62 CONCLUSION AND OUTLOOK ...... 65 MULTICAPILLARY COLUMN-ION MOBILITY SPECTROMETRY OF VOLATILE METABOLITES EMITTED BY SACCHAROMYCES CEREVISIAE ...... 71 SUMMARY ...... 71 INTRODUCTION ...... 72 MATERIAL AND METHODS ...... 73 Yeast strains and growth conditions ...... 73 Analytics ...... 74 SPME GC-MS measurements for validation of volatile metabolites ...... 74 MCC-IMS measurements ...... 74 Volatile metabolite identification ...... 75 RESULTS AND DISCUSSION ...... 75 Experimental setup ...... 75 MCC-IMS monitoring of batch cultures...... 76 MCC-IMS monitoring of glucose-limited chemostat fermentations ...... 80 Identification of volatile metabolites ...... 84 CONCLUSION AND OUTLOOK ...... 85 THE CRABTREE EFFECT REVISITED ...... 91 A COMPARATIVE STUDY ON THE TRANSCRIPTIONAL CHANGES DURING THE CRABTREE EFFECT USING LABORATORY AND INDUSTRIAL STRAINS AND CONDITIONS ...... 91 SUMMARY ...... 91 INTRODUCTION ...... 92 MATERIAL AND METHODS ...... 93 Yeast strains and growth conditions ...... 93 MCC-IMS analysis ...... 94 cDNA microarray analysis ...... 94 RESULTS AND DISCUSSION ...... 95 II Table of Contents

MCC-IMS and microarray analysis of two yeast strains ...... 95 CONCLUSION AND OUTLOOK ...... 100

GENERAL DISCUSSION ...... 105 POSSIBLE IMPACT OF THIS THESIS ...... 105 IMPACT ON FUTURE STUDIES ...... 105 STRAIN ENGINEERING & POSSIBLE USE OF VOLATILE PRODUCTS ...... 106 THE USE OF DIFFERENT METABOLIC CONDITIONS ...... 107 ONLINE FLUX ANALYSIS ...... 107 CONCLUSION ...... 107 THE CONSENSUS YEAST METABOLIC MODEL...... 107 SESI-ORBITRAP ...... 108 MCC-IMS ...... 108 LABORATORY VS. INDUSTRIAL CONDITIONS ...... 108

OUTLOOK...... 113

APPENDIX ...... 115

REFERENCES ...... 123

CURRICULUM VITAE ...... 145

PUBLICATIONS ...... 147

ORAL PRESENTATIONS ...... 147

POSTER PRESENTATIONS ...... 148

III

Summary

The baker’s yeast Saccharomyces cerevisiae is one of the best investigated organisms and widely used in science. This scientific interest is partially based on the yeast’s use in the industrial production of pharmaceuticals, beverages and food. Even though the taste of food and beverages is clearly influenced by volatiles, the yeast’s volatilome, i.e., the entirety of the volatile metabolites produced, is to date largely uncharted, with ethanol and acetaldehyde being prominent exceptions. In this thesis, in chapter 2.1, a metabolic model of S. cerevisiae was enhanced with the biochemistry of volatile metabolites that have been found by a literature investigation. Not only the volatile metabolites in question have been added, but also substances and reactions that connect them to metabolites already present in the model. In total, 225 metabolites and 219 reactions were added to the model. Furthermore, 12 metabolic reactions could be verified by physiological and enzyme assays of knockout mutants. These mutants were created using the CRISPR-Cas9 method and contained only one gene coding for a protein with alcohol dehydrogenase activity. In chapter 2.2, volatile metabolite dynamics during the induction of the Crabtree effect in a fully respiratory growing continuous culture of S. cerevisiae were explored with real-time analyses of the fermentation off-gas. SESI (secondary electrospray ionization)-Orbitrap-mass spectrometry (MS) was used for this endeavor. In these measurements, we detected about 2,500 signals of which 16 showed a response to the perturbation of the metabolic state prior to the detection of ethanol. These observations not only revealed the extent of the yeast’s volatilome, but also indicated that volatile metabolite dynamics correlate with the metabolic state of the cell culture and hence might be a noninvasive and fast online analytical method to monitor and later control fermentation processes. In chapter 2.3, to evaluate the possibility of online volatile metabolite monitoring, multi capillary column–ion mobility spectrometry (MCC-IMS) analysis of yeast fermentation off-gas was established. This analytical device was chosen because it runs at ambient temperature and pressure resulting in lower investment and operating costs compared to SESI-Orbitrap-MS. The MCC-IMS used was developed for the detection of volatiles in human breath, and several technical adaptations were required to allow robust detection of volatiles in the headspace of yeast fermentations. In chapter 2.4, the MCC-IMS in its optimized configuration was applied to monitor volatile metabolite changes of a laboratory and an industrial yeast strain during the transition from fully respiratory to respiro-fermentative metabolism (Crabtree effect). In addition, metabolic differences in this setting were examined on transcriptional level using a cDNA microarray. The metabolic shift could be observed in the volatile space of both strains and in all tested conditions. The transcriptome showed differences in the leucine and isoleucine pathways, as well as in genes related to the TCA cycle and the respiratory pathways. Interestingly, the expression data indicated that the industrial strain upregulated its respiration during the shift, while it was downregulated in the laboratory strain. Lastly, possible applications for the knowledge gained and methods developed in this work are discussed. This thesis provides a blueprint for studies of the volatile space in other organisms. The extended metabolic model could be used to generate yeast strains with special flavors or for the production of fragrances, perfumes or precursors of pharmaceuticals. Also, the knowledge gained about the changes in the volatilome during metabolic transitions could be used to online determine and potentially control the metabolic state of a yeast. Finally, the analytical methods developed in this work might be used for online flux analysis, if some more of the detected volatiles can be identified.

Zusammenfassung

Die Bäckerhefe Saccharomyces cerevisiae ist einer der am besten untersuchten Organismen und wird häufig in der Wissenschaft genutzt. Dieses wissenschaftliche Interesse beruht teilweise auf der industriellen Nutzung von Hefe bei der Produktion von Medikamenten, Getränken und Nahrungsmitteln. Obwohl der Geschmack von Nahrungsmitteln und Getränken deutlich von flüchtigen Metaboliten beeinflusst wird, ist das Volatilom, also die Gesamtheit der produzierten flüchtigen Metabolite, bei Hefen bisher noch weitgehend unerforscht, mit Ausnahme der prominenten Ausnahmen Ethanol und Acetaldehyd. In dieser Arbeit wurde in Kapitel 2.1 die Literatur nach flüchtigen Metaboliten durchsucht und ein metabolisches Modell von S. cerevisiae durch deren Integration verbessert. Zusätzlich zu den gefundenen Metaboliten wurden Substanzen und Reaktionen in das Modell eingepflegt, um die hinzugefügten volatilen Metaboliten mit dem bereits vorhandenen Netzwerk zu verbinden. Insgesamt wurden 225 Metaboliten und 219 Reaktionen zu dem Modell hinzugefügt. Zudem konnten 12 metabolische Reaktionen durch physiologische und enzymatische Versuche mit knockout-Mutanten verifiziert werden. Die Mutanten wurden durch Anwendung des CRISPR-Cas9-Systems erzeugt und enthalten jeweils nur ein Gen, welches für ein Protein mit Alkoholdehydrogenaseaktivität kodiert. In Kapitel 2.2 wurden die flüchtigen Metabolite einer kontinuierlichen, respiratorischen Saccharomyces cerevisiae- Kultur während der Induktion des Crabtree-Effekts untersucht. Dafür wurden die Abluft der Fermentation direkt in ein SESI (Sekundäre Elektrospray-Ionisation)-Orbitrap Massenspektrometer (MS) geleitet, wo sie in Echtzeit analysiert wurde. Pro Experiment wurden ca. 2.500 Signale aufgenommen, 16 von diesen zeigten nach der Perturbation des metabolischen Zustandes, noch bevor Ethanol detektiert wurde, eine Änderung in der Signalintensität. Diese Ergebnisse verdeutlichen nicht nur die Größe des Hefe-Volatiloms, sondern zeigen ebenfalls, dass die Dynamiken der volatilen Metabolite im Gasraum über Hefefermentationen mit dem metabolischen Zustand der Zellen korrelieren. Daher könnte eine nicht-invasive und schnelle online-Analysemethode für das Überwachen von Hefekulturen dazu geeignet sein, Fermentationsprozesse zu kontrollieren. Genau dies wurde in Kapitel 2.3 untersucht, dazu wurde die Analyse des Gasraums über einer Hefefermentation mit einem Multikapillarsäulen-Ionen-Mobilitäts-Spektrometer (MCC-IMS) etabliert. Dieses analytische Gerät wurde ausgewählt, da es bei Normaldruck und Raumtemperatur betrieben werden kann. Im Vergleich zur SESI- Orbitrap-MS sind für das MCC-IMS geringere Wartungskosten aufzubringen und auch die Anschaffungskosten des MCC-IMS liegen deutlich unter denen der SESI-Orbitrap-MS. Das hier verwendete MCC-IMS wurde für die Detektion von Metaboliten in menschlicher Ausatemluft konzipiert, so dass einige technische Anpassungen vorgenommen werden mussten, um eine robuste Detektion der volatilen Metaboliten von Hefefermentationen zu ermöglichen. In Kapitel 2.4 wurde das optimierte MCC-IMS verwendet, um Metabolitkonzentrations-Veränderungen im Gasraum über Hefefermentationen während dem Übergang von ausschließlich respirativem hin zu respirofermentativem Wachstum (Crabtree-Effekt) zu untersuchen. Unter diesen Bedingungen wurde auch mithilfe eines cDNA-Microarrays die Änderungen auf transkriptioneller Ebene untersucht. Für diese Versuche wurde ein Laborstamm und ein industriell genutzter Stamm verwendet. Die metabolischen Veränderungen spiegelten sich im Gasraum der beiden unterschiedlichen Hefestämme deutlich wieder. Im Transkriptom konnten Unterscheide in der Regulation der Gene, die mit dem Leucin- und Isoleucin-Metabolismus zusammenhängen, sowie bei den für die Atmung und den Citratzyklus zuständigen Genen festgestellt werden. Überraschend war, dass die mit der Atmung assoziierten Gene im Industriestamm mit Änderung des metabolischen Zustandes hochreguliert wurden, während die gleichen Gene im Laborstamm herunterreguliert wurden. Im letzten Teil der Arbeit werden mögliche Anwendungen für das in dieser Arbeit generierte Wissen diskutiert. So könnte diese Dissertation als Vorlage für die Untersuchung von Volatilen anderer Organismen genutzt werden. Das erweiterte metabolische Modell könnte dazu verwendet werden, Hefestämme mit besonderen Aromen für die Produktion von Duftstoffen, Parfums oder als Vorläufer für Medikamente, zu entwerfen. Darüber hinaus könnte das hier generierte Wissen über die Veränderung flüchtiger Metabolite während einer metabolischen Zustandsänderung ebenfalls dazu verwendet werden, um eine Fermentation mittels online-Analytik zu kontrollieren. Letztlich könnten die hier entwickelten analytischen Methoden zur online-Flussanalyse genutzt werden, falls noch weitere der hier detektierten flüchtigen Metabolite identifiziert werden.

VII

List of Abbreviations

List of Abbreviations Abbreviations % percent (NH4)2HPO4 diammonium phosphate (NH4)2SO4 ammonium sulfate μF microfarad μL microliter μm micrometer ACO aconitase ADH alcohol dehydrogenase AIMS aspiration IMS BAT branched-chain amino acid transmaniase BNICE biochemical network integrated computational explorer CaCl2 2H2O calcium chloride diydrate Ca-pantheonate calcium-pantheonate Cas9 CRISPR associated protein 9 cDNA copy desoxyribonucleic acid CDW cell dry weight CFME continuous flow microextraction CHA1 catabolic L-serine deaminase CIT citrate synthase cm centimeter CO2 carbondioxide CoA conenzyme A COB cytochrome b CoCl2 7H2O cobalt(II) chloride heptahydrate COR core protein of QH2 cytochrome c reductase COX cytochrom oxidase CRISPR clustered regulatory interspaced short palindromic repeats CuSO4 5H2O copper(II) sulfate pentahydrate CYT1 cycochrome c1 D dimer DI direct immersion DMA differential mobility analyzer DNA desoxyribonucleic acid DO dissolved oxygen DTIMS drift time IMS DTT dithiothreitol E. coli Escherichia coli e.g. exempli gratia EDTA ethylendiamintetraacetic acid ESI electrospray ionization et al. et alia FAEE fatty acid ethyl esters FAIMS high field asymmetric waveform IMS IX List of Abbreviations

FBA flux balance analysis FBA flux balance analysis FDR false discovery rate FeSO4 7H2O ferrous sulfate heptahydrate FI flame ionization FMM from metabolite to methabolite g gram GEM genome scale model GENRE genome scale network reconstruction GRE 3-methylbutanal reductase h hour H2O water H3BO3 boric acid HMG-CoA 3-hydroxy-3-methylglutaryl CoA HPLC high-performance liquid chromatography HS headspace i.e. id est IDH isocitrate dehydrogenase ILV acetolactate synthase IMS ion mobility spectrometry /spectrometer K ion mobility K0 reduced ion mobility KEGG Kyoto encyclopedia of genes and genomes KGD alpha-ketoglutarate dehydrogenase KH2PO4 potassium phosphate KI potassium iodide kV kilovolt L liter LB lysogeny broth LEU alpha-isopropylmalate synthase LPD lipoamide dehydrogenase LSC ligase of succinyl-CoA M molar M monomer m/z mass to charge ratio m/z mass per charge ratio MALDI matrix assisted laser desorption ionization MCC multi capillary column MCCase 3-methylcrotonyl-CoA MDH malate dehydrogenase mg miligram MgSO4 7H2O magnesium sulfate heptahydrate min minutes mL milliliter mM millimolar X List of Abbreviations

MnCl2 4H2O manganese(II) chloride tetrahydrate MS mass spectrometry /spectrometer N2 nitrogen NAD nicotinamid adenin dinucleotide NADP nicotinamid adenin dinucleotid phosphate NaMoO4 2H2O sodium molybdate dihydrate ND not detected NDE NADH dehydrogenase, external NDI NADH dehydrogenase, internal ng nanogram nL nanoliter NMR nuclear magnetic resonance OD optical density OLIMS open loop IMS OMS overtone mobility analyzer ORF open reading frame ppb parts per billion ppm parts per million ppt parts per trillion PTM posttranslational modification PTR proton transfer reaction PYC pyruvate carboxylase Q quadrupole QCR ubiquinol-cytochom c oxidoreductase RIP reaction ion peak RIP rieske iron sulfur protein RNA ribonucleic acid rpm rounds per minute s seconds S. cerevisiae Saccharomyces cerevisiae SBML systems biology markup language SDH succinate dehydrogenase SDME single drop microextraction SESI secondary electrospray ionization SFA bifunctional alcohol and formaldhyde dehydrogenase SIM selected ion monitoring SPME solid phase microextraction TCA tricarboxylic acid TDS thermal desorption spectroscopy TFBA thermodynamics-based flux analysis TIC total ion current TIMS trapped IMS TMIMS transversal modulation IMS TNT 2,4,6-trinitrotoluene TOF time of flight XI List of Abbreviations

TWIMS traveling-wave IMS USER uracil-specific excission reagent VOC volatile organic compound vvm gasvolume per fermentervolume per minute w/w weight per weight ZnSO4 7H2O zinc sulfate heptahydrate Ω ohm

XII List of Figures

List of Figures

Figure 1: Overview of volatile metabolites released by yeast. Depicted are few representatives and their origin from central carbon metabolism [1]...... 9 Figure 2: The Ehrlich pathway. Biochemistry and main genes involved in amino acid degradation (A) and intermediates and products derived from involved amino acids (B). Figure adapted from [2, 3]. Figure previously published in [1]...... 11 Figure 3: Mitochondrial monoterpene biosynthesis via the leucine catabolic MCC pathway and possible relationships with sterol formation in Saccharomyces cerevisiae as proposed by Carrau et al. [4]. Figure adapted from [4]. Previously published in [1]...... 13 Figure 4: Working principle of a drift tube-IMS; adapted with permission of Journal of Physiology and Pharmacology from [5], previously published in [1, 6]...... 15 Figure 5: Schematic representation of a secondary electrospray ionization source. The here indicated DMA is interchangeable with an Orbitrap. DMA= differential mobility analyzer. Reprinted with permission from [300]. Copyright 2012 American Chemical Society...... 57 Figure 6: Influence of mass resolution on peak separation. Taken from [133]...... 58 Figure 7: Schematic view (A) and picture (b) of the mini-bioreactor setup connected to the SESI-Orbitrap-MS (Bioreactor constructed by Eik Czarnotta and Suresh Sudarsan, RWTH Aachen University)...... 60 Figure 8: Data evaluation of SESI Orbitrap analyses of the off-gas of a continuous glucose-limited yeast culture at steady state which was disrupted with a 27 mM glucose pulse. The filtered data is displayed as a clustered heatmap in which the scaled intensity of single masses (m/z value) are plotted over time (A), zoom-in excerpt of the heatmap shown in A (B). Smoothed data of interesting signal traces (blue line) identified in the heatmap are plotted against time together with the signal of ethanol (black line). The left black vertical line indicates the timepoint of the glucose pulse injection, while the right line indicates the timepoint of the first detection of ethanol (C-F)...... 64 Figure 9: Experimental set-up for the online MCC-IMS measurements of fermenter off-gas. Previously published in [185]...... 76 −2 Figure 10: MCC-IMS topographic plot of sterile Verduyn medium. The reaction ion peak (1/K0 = 0.5 Vs cm ) was compensated by the software VisualNow...... 77 Figure 11: MCC-IMS topographic plot of S. cerevisiae during the early stationary phase of a batch fermentation. Boxes indicate signals that showed the most significant changes during the growth (perturbation) experiments...... 78 Figure 12: (A) Fermentation profile of S. cerevisiae during batch growth in glucose minimal medium, (B) trends in intensity, (C) heat map of selected peaks detected by MCC-IMS analysis of the fermentation off-gas. Areas in the heat map show the detected signal peak and the surrounding area; DO = dissolved oxygen...... 79 Figure 13: (A) Fermentation profile of S. cerevisiae adh1Δ during batch growth in glucose minimal medium and (B) trends in intensity and (C) heat map of selected peaks detected by MCC-IMS analysis of the fermentation off-gas. The areas in the heat map show the detected signal peak and the surrounding area; DO = dissolved oxygen...... 80 Figure 14: MCC-IMS topographic plot of the off-gas of a glucose-limited continuous cultivation of S. cerevisiae. Boxes indicate signals that showed the most significant changes during the growth (perturbation) −2 experiments. The reaction ion peak (1/K0 = 0.5 Vs cm ) was compensated for by the VisualNow software...... 81 Figure 15: (A) Fermentation profile of S. cerevisiae during growth in a glucose-limited chemostat and (B) trends in intensity and (C) heat map of selected peaks detected by MCC-IMS measurements of the fermentation off-gas after perturbation of the metabolic steady state with a pulse of 22 mmol glucose; DO = dissolved oxygen...... 82 Figure 16: (A) Fermentation profile of S. cerevisiae during growth in a glucose-limited chemostat and (B) trends in intensity and (C) heat map of selected peaks detected by MCC-IMS measurements of the fermentation off-gas during transition to anaerobic conditions; DO = dissolved oxygen...... 83 Figure 17: Volatile signals of Saccharomyces cerevisiae CEN.PK 113-7D (A) and Saccharomyces cerevisiae DHW (B) in a fed-batch that was slowly overfed to trigger the Crabtree effect. The grey background represents XIII List of Figures

the feed rate. The commercially available Alcoline sensor form Biotechnology Kempe was used for ethanol quantification in the fermentation broth along with the MCC-IMS Breath Discovery from B&S Analytik for the monitoring of volatile metabolites in the fermentation head space. The black dotted line indicates the start of the overfeeding process, the red dotted lines indicate the division into early (left) mid (middle) and late (right) phase. D and M behind the substance names describe the measured dimer and monomer respectively...... 96 Figure 18: Transcriptional alterations of the most differing pathways based on expressional changes between Saccharomyces cerevisiae CEN.PK 113-7D (left) and Saccharomyces cerevisiae DHW (right). The two strains were run in fed-batch fermentations and overfeed gradually. The respiration pathway (A), the TCA-cycle (B) and the branched-chain amino acid production (C) are depicted. The genes without color coding are not displayed, since it had a false discovery rate of more than 5% in at least one of the tested strains...... 99

XIV List of Tables

List of Tables

Table 1. Features of different analytical devices for volatile compound measurement. Data on resolving power taken from [181-184], all other data taken from product sheets of current devices; LOD, limit of detection. Previously published in [1]...... 22 Table 2: Microbial strains ...... 36 Table 3: Primers (table continues over multiple pages) ...... 37 Table 4: Oligonucleotides for homologous recombination ...... 39 Table 5: Volatile organic compounds emitted from S. cerevisiae fermentations. Compounds that are included in the latest yeast genome scale metabolic reconstruction (Yeast 7.11) and in the Saccharomyces Genome Database SGD [185] are categorized. (Table continues over multiple pages.) ...... 41 Table 6: Volatile metabolites added to the metabolic yeast model. The column reference indicates, whether the pathway was taken from KEGG, a scientific publication, that the pathway was inferred by pathway prediction tools (blank space), or predicted by established yeast enzymes with reactions similar to those needed to establish the pathway (also blank space). (Table continues over multiple pages.) ...... 44 Table 7: Enzymatic capacity of sADH mutants. The strains were tested for their ability to convert alcohols, ketones, and aldehydes in enzymatic assays. The mutants showed either clear conversion (+), slight conversion (+/-) or no conversion of the substrate (-). Green indicates accordance to literature data, red indicates contrariety...... 49 Table 8: Mass to charge ratio (m/z) of the most interesting signals recorded during the metabolic transient state in a continuous glucose-limited fermentation after setting a glucose pulse. + and # indicate masses whose signal intensity linear correlated with each other and which probably originate from the same metabolite. m/z 69.0701 and 89.0962 have been tentatively identified as 3-methyl-1-butanol and isoprene, respectively...... 63 Table 9: Volatile organic compounds detected in fermentations of S. cerevisiae via GC-MS measurements growing in glucose minimal salt medium. ND, not detected...... 85 Table 10: Profiles of fermentation parameters of the fed-batch fermentations...... 94

Chapter 1

General introduction

Exploration and Exploitation of the Yeast Volatilome

Partially published as: Ebert, B.; Halbfeld, C.; Blank, L. Exploration and exploitation of the yeast volatilome. Current Metabolomics 2017, 5, 102-118.

Contributions: The introduction was written by Christoph Halbfeld. Birgitta E. Ebert contributed 1.2.4 to 1.3 and 1.4. Lars Blank critically read the text.

Chapter 1

General introduction Summary Volatile organic compounds (VOCs) are small molecular mass substances, which exhibit high-vapor pressures, low boiling points, and lipophilic character. VOCs are produced by all organisms including eukaryotic microbes like yeast. Volatile metabolites are for centuries exploited, for examples as flavors in bread, beer, and wine. Notably, while the applications of VOCs are many, the knowledge on their biochemical synthesis is still limited. This introduction reviews the current information of yeast volatile metabolites and techniques to further explore the VOC landscape made possible by improvements of the analytical possibilities, regarding sampling frequency, identification, and quantification and the development to computationally interpret (high-throughput) data. Especially possibilities for online and even real- time analysis should trigger new experimental approaches that elucidate the biochemistry as well as the regulation of VOC synthesis. Baker’s yeast is here the organism of choice as the genetic inventory can be linked to VOC formation and with this in hand improved applications can be envisaged. The physical, chemical or biological properties make many VOCs interesting targets for different industrial sectors, while their natural function as semiochemicals or in defense mechanisms can be exploited to engineer synthetic microbial communities or to develop new antibiotics. VOCs produced by microbes including yeast are a chemical diverse group of compounds with highly different applications. The new analytical techniques briefly summarized here will enable the use of VOCs in even broader applications including human health monitoring and bioprocess control. We envisage a bright future for VOC research and for the resulting applications.

Introduction Volatile organic compounds Volatile organic compounds (VOCs) are small molecular mass substances (<300 Da), which exhibit high-vapor pressures, low boiling points and lipophilic character. VOCs encompass various chemical classes, e.g., low molecular weight fatty acids and their derivatives (hydrocarbons, alcohols, aldehydes and ketones), terpenoids, aromatic compounds, nitrogen containing compounds, and volatile sulphur compounds. VOCs are ubiquitous in nature and play an important role in all domains of life, especially in the intra- and interspecies communication and self-protection. Plants, for example, use VOCs as direct or indirect defense mechanism to repel herbivores or attract carnivores that exterminate herbivore populations [7]. Likewise, animals use pheromones as means of communication or behavior-altering agents, while humans use VOCs formed by microbes as an indicator of spoiled food, another form of interspecies communication [5][8]. Taste perception is also determined by volatiles since the papillae in the human mouth can only distinguish between salty, sweet, bitter, sour and, umami and what we experience as flavor is created through the binding of volatiles to sensory receptors [6]. It is for this reason that volatiles play an important role in the food and beverage industry, where they are often

5 General introduction produced by microorganisms. A prominent example is the baker’s yeast Saccharomyces cerevisiae which is used for everyday products like bread, wine, beer, and yeast extract and whose endogenous metabolites contribute to the flavor of these products. Accordingly, aroma as well as unpleasant odor components formed by yeast during food fermentation have been extensively investigated. The aroma of food fermentation is explained in further detail in the following paragraphs.

Bread Bread has been consumed by humankind for millennia as it is essential food and is well known for its rich flavor. Its odor and flavor depend on a multitude of factors like dough-composition, fermentation and baking parameters. While wheat flour contributes only little to the aroma of bread, enzymes and yeast contribute a better part to it [9-12]. To investigate VOC pattern in dough fermentation, Frasse et al. analyzed the headspace of first yeasted, secondly yeasted and fermented, and thirdly non-yeasted doughs. The experiments showed that alcohols, esters, ketones, lactones and sulfur compounds were produced during yeast fermentation, while aldehydes were consumed in the process [12]. The main compounds contributing to the flavor prior to baking were mostly fusel alcohols (isoamyl alcohol, 2-methyl-1-butanol, 2-Phenylethanol), but also aldehydes (diacetyl, methional) due to a high odor threshold [12]. During the baking process the flavor changes, the caramelization reaction of the sugars, and especially the Maillard reaction play an important role in changing the flavor profile [9, 13-15]. Depending on the composition of amino acids in the dough, the Maillard reaction will lead to the production of acetaldehyde, phenylacetaldehyde, 2-methylebutanal and other carbonyl compounds. Leucine, isoleucine and lysine lead to a pleasing aroma, while phenylalanine and methionine lead to an unpleasant odor in a model Browning system, while phenylalanine led to a rose oil like odor in baked bread. Especially lysine, arginine, histidine, and tryptophan lead to the browning of the crumb [16, 17]. After the baking process, the aroma of fresh baked bread unfolds. This aroma has been investigated in versatile studies, it is created mainly by a mixture of aldehydes (e.g., 3- methylbutanal, 2-methylpropanal, (2E,4E)-deca-2,4-dienal, hexanal, 2-phenylacetaldehyde, 3- methylsulfanylpropanal, 2-methylbutanal, non-2-enal), ketones (butane-2,3-dione, oct-1-en-3-one) and acids (e.g., acetic acid, 2-methylbutanoic acid, 3-methylbutanoic acid) [18-22]. The aroma of the baked bread will change over time and the desirable flavor will disappear, this is also caused by volatiles that slowly fade (e.g., 2-acetyl-1-pyrroline, 3-methylbutanal), while other compounds formed through lipid peroxidation are formed [20, 23].

Wine Wine flavors Many studies have been conducted in the field of volatile wine metabolites, since these compounds are responsible for the rich wine aroma. Welke et al. investigated the volatile profile of Merlot wines and tentatively identified a total of 334 compounds; the article also gives a good overview of volatile metabolites that were identified in other studies [24]. The aroma is influenced by the different stages of wine making: first the choice and processing of the grapes (pre-fermentative flavor), then the 6 Chapter 1 fermentative conditions, choice of the yeast and amount of inoculum (fermentative flavor) and finally the post-fermentative flavor that comes from enzymatic or chemical reactions while storage in the barrel and bottle [25-30]. The wine grapes differ in a multitude of aroma compounds; especially important roles play hereby the monoterpenes. There are about 70 monoterpenes known at this time and differences in their concentration can be used for varietal characterization. The most prominent monoterpenes in grape aromas are linalool, geraniol, nerol, citronellol, 3,6-dimethyl-1,5-octadien- 1,7-diol, and α-terpineol [30]. In 2001, Mateo et al. showed that a different inoculum concentration of the same yeast strain can influence the volatile composition of wine. They showed that the quantity of the yeast inoculum is positively correlated to the amount of higher alcohols that can be found later in the wine, while higher alcohols above 400 mg/L are regarded as negative factor in the final product [27, 31]. Other than higher alcohols the amount of long chain esters is strain dependent which has to be considered because the amount of higher alcohols and long chain esters relate to each other by the equation:

ͶͲͲ (1)[27] ሿݔݏݎݐ݁ݏሾ݈݋݄݊݃ܿܽ݅݊݁ ሾ݄݄݅݃݁ݎ݈ܽܿ݋݄݋݈ݏሿ

The choice of the correct strain is also important because of different amount of alcohols, organic acids and esters that are also important for the bouquet in the resulting wines [27]. During the aging in a bottle different reactions take place that have an influence on the aroma: the ester content changes, the acetate concentration decreases, mono- and dicarboxylic acid ethyl esters increase, carotenes and carbohydrates break down and monoterpenes react in the sour conditions. Especially the acetates decrease over a period of about six years until an equilibrium between alcohol, acetic acid and acetate is reached. All these factors have an influence to the taste difference between young and aged wine, e.g., the acetates distribute the fresh and fruity note of young wine [30]. Also a connection between D,L-piperitone [32] and 1,8-cineole [33] and the positive aging effects of red wine could be associated.

Off-flavors The following section will focus on the negative flavor compounds in wine. These compounds are perceived as negative, if their concentration exceeds a certain sensory threshold. The first off-flavor was detected in Europe, when European cultivars of Vitis vinifera were cross-bread with wild American wine plants to increase the fungal resistance. This off flavor can be tracked back to 2,5- dimethyl-4-hydroxy-2,3-dihydro-3-furanone (furaneol) [30, 34]. The sensorial detection limit for furaneol lies in the range between 30 and 300 ppb [30, 35, 36]. Another undesirable off-flavor is 2- iso-butyl-3-methoxy-pyrazine that has a low perception threshold of about 0.002 to 0.4 ppb. It smells like herbs or potatoes and is enriched in ripening Sauvignon grapes [30, 37-39]. 2- ethyltetrahydropyridine, 2-acetyl-tetrahydropyridine and 2-acetylpyroline are responsible for a mousy (taste of mouse urine) like flavor in wine [40-46]. An amount of more than 800 μg/L of the pure substances or a mixture of 4-vinylguajacol and 4-vinylphenol leads to a medicinal or Elastoplast like off-flavor. This flavor is developed, especially if the grapes were exposed directly to sunlight in 7 General introduction warmer regions. A similar effect occurs if the sum of more than 400 μg/L of 4-ethylphenol and 4- ethylguajacol are exceeded, these compounds however have leathery and respectively horse sweat aroma [30, 47]. The corky off-note that some wines develop during the aging process are associated with several components such as 2,3,6-trichloroanisole, 2,3,4-trichloroanisole, 2,3,5,6- tetrachloroanisole, pentachloroanisole, 2,4,6-tribromoanisole, 2-methylisoborneol, ethylenguaiacol, 4-ethylenphenol and 2,4,6-trichloroanisole. The single compounds show earthy, mushroom and cardboard-like aromas. The olfactory threshold for 2,4,6-trichlooanisole is about 4-10 ng/L and thus way lower than that of 4-vinylguajacol, 4-vinylphenol, 4-ethylphenol and 4-ethylguajacol [48-52]. It is not possible to sum up the complete literature of wine flavors in this thesis. To learn more about wine flavors, the research of Prof. Rapp and the book “Flavour and Fragrances” from 2007 provides a useful basis [30, 53, 54].

Fungal VOCs On the contrary to the general capability of microorganisms, the production of volatiles has not been thoroughly explored so far but is receiving increased attention because of a growing awareness of the potential of volatile metabolites as antimicrobial agents, biofuels or plant growth promoting compounds. Also, it has been shown that the volatile footprint is specific for different organisms, allowing VOC analysis to be used for diagnostic purposes, the detection of food spoilage, and hidden growth of molds in buildings [7, 55-60]. By the same token, VOC synthesis depends on growth condition, carbon source and carbon source availability [61, 62], hence can be used in fermentation process monitoring. The synthesis of microbial VOCs has long been seen as a cellular detoxification or waste disposal mechanism, but analogous to plant hormones or pheromones they have important ecological roles and are released into the environment, to impact metabolism and growth of competing or symbiotic organisms. Research on these microbial semio- or infochemicals is of interest in basic research but also in applied sciences, e.g., for the design of synthetic microbial communities tailored for the production of chemicals, degradation of pollutants, and wine manufacturing [55]. Still, the spectrum of fungal volatile metabolites, also referred to as volatilome [63], is today not broadly and systematically explored. This becomes apparent when comparing known fungal VOCs with the number identified from the plant and animal kingdom. While databases for the latter list up to 8000 compounds [64, 65], the microbial specific VOC database, mVOC [66], contains 1174 metabolites, out of which 500 are of fungal origin (as of May 2016). The current knowledge about the VOC biosynthesis pathways and their regulation is even scarcer. This might be attributed to challenges in detection and identification of these mainly low abundant metabolites and their chemical diversity which requires complimentary sampling and analytical techniques to capture the complete VOC space. Furthermore, the synthesis of VOCs originating from secondary metabolism might only be activated under special growth conditions different from those the cells encounter during cultivation in laboratory settings. Advances in system-wide analytical techniques [67] and methods for the identification and awakening of secondary biosynthetic pathways pave the way for comprehensive elucidation of yeast’s metabolic potential for volatile metabolite synthesis [68-70].

8 Chapter 1

Metabolic pathways of yeast Yeasts release a manifold of volatile metabolites, which contribute to their rich odor. The VOCs are produced from primary and secondary metabolism and include hydrocarbons, heterocycles, aldehydes, ketones, alcohols, phenols, thioalcohols, -esters, terpenes amongst others (Figure 1). VOCs produced from primary metabolism pathways, including glycolysis, proteolysis, and lipolysis are thought of as waste materials and their production as a means of detoxification, while VOCs involved in defense or communication mechanisms are assigned to secondary metabolism [71, 72].

Figure 1: Overview of volatile metabolites released by yeast. Depicted are few representatives and their origin from central carbon metabolism [1].

In the following we give a coarse overview about currently known volatile metabolites and the biosynthetic pathways of yeast. Given the focus of prior volatile research the presented volatile spectrum is biased for VOCs impacting on the flavor of fermented food as previously mentioned.

VOCs derived from carbohydrate catabolism and fermentation Ethanol is the most well-known volatile metabolite of yeast. It is produced under anaerobic conditions or aerobically at high glycolytic fluxes to re-oxidize NADH produced by glucose oxidation. Acetic acid, on the other hand, is formed if alternative pathways do not satisfy the anabolic NADPH demand and is, for example, observed during recombinant protein production [73, 74]. Acetaldehyde is the direct metabolic precursor of ethanol and acetate and at low concentrations contributes a fruity flavor to beer, wine, and other alcoholic beverages. At higher concentrations, however, it possesses a pungent, irritating odor [75]. The physiological role of acetaldehyde secretion is not well understood and it is argued to simply be a leakage product of ethanol fermentation [76]. However, it has also been reported that low amounts of acetaldehyde in the fermentation medium support growth of some yeast strains [77]. 2,3-butanediol is another fermentative volatile byproduct of yeast along with its precursor metabolite acetoin. Their synthesis might also be a means to adjust redox cofactor levels but it has 9 General introduction also been reported that acetoin, in a mix with other yeast VOCs, attracts the fruit fly Drosophila melanogaster with which some yeasts undergo a symbiotic relationship [78].

VOCs produced by amino acid synthesis and degradation The amino acids valine, leucine, isoleucine, methionine, phenylalanine, tryptophan, and tyrosine are catabolized by the Ehrlich pathway [20] (Figure 2). The resulting aliphatic and aromatic alcohols and corresponding acids are known as fusel alcohols and fusel acids [79]. As with acetaldehyde, at high concentrations fusel alcohols impart off-flavors (thus the German term fusel, bad liquor), while low concentrations of these compounds (and their esters) contribute to the flavor and aroma of fermented foods and beverages [80]. During beer and wine fermentation, yeasts utilize the amino acids contained in the wort or grape must as main nitrogen source. In the first step of their degradation the amino acids are deaminated to α-keto acids (Figure 2). The ketoacids are further decarboxylated and then either reduced or oxidized to the respective fusel acid or alcohol. [2]. While proposed already in 1907 [79], the genes encoding the enzymes of this pathway are still to be entirely explored [81]. Also, the physiological role of fusel oil formation and its regulation is not well understood but it is argued that tuning the alternative oxidation and reduction steps of this pathway helps to maintain the NADH/NAD+ ratio and to balance redox cofactors [81, 82]. Moreover, the irreversible decarboxylation of the keto acid acts as a thermodynamic pull as low levels of this intermediate allow shifting the equilibrium of the reversible deamination to the right side, thereby fostering complete degradation of the amino acids and maximal nitrogen supply. The fusel alcohols 3-methyl-1-butanol (isoamyl alcohol) and 2-methyl-1-propanol (isobutanol), the end products of the degradation of the branched amino acids leucine and valine bear potential as biofuels [83], while the aromatic 2-phenylethanol, the catabolic end product of phenylalanine is used as rose-like fragrance. Catabolism of the sulphur containing amino acids, methionine and cysteine via the Ehrlich pathway releases volatile sulphur compounds, e.g., methional and dimethyldisulfide. These VOCs are responsible for the flavor in potato-based snacks and are also formed by the heat- induced Maillard reaction. To a lesser extent the higher alcohols are synthesized via an alternative anabolic pathway from pyruvate [84]. The carboxylic acids are easily converted to acetate esters by condensation with acetyl coenzyme A (CoA), with ethyl acetate (solvent-like aroma), isoamyl acetate (banana aroma), isobutyl acetate (fruity aroma) and phenyl ethyl acetate (roses, honey) being important flavor active examples found in fermented beverages [85]. The physiological role of these esters is not elucidated, yet. One hypothesis is that ester formation serves the recycling of CoA under anaerobic conditions when acetyl-CoA consumption in the TCA cycle is suppressed [86]. On the first sight, production of the volatile esters might be costlier for the cell than acetate secretion, but it has the advantage of preventing the accumulation of this toxic weak acid which at high concentrations entails more detrimental growth defects. Studies, in which acetyl-CoA overproduction increased isoamyl acetate synthesis, while increased acetyl-CoA consumption decreased acetate ester formation, support this theory [87, 88]. Also, antimicrobial effects have been reported for some esters [55].

10 Chapter 1

Figure 2: The Ehrlich pathway. Biochemistry and main genes involved in amino acid degradation (A) and intermediates and products derived from involved amino acids (B). Figure adapted from [2, 3]. Figure previously published in [1].

Diacetyl and 2,3-pentanediol, VOCs with butter or toffee-like flavor are undesired byproducts (off- flavors) in beer fermentation and are formed when 2-acetolactate or 2-oxobutyrate, intermediates of valine and isoleucine biosynthesis, accumulate. When these compounds diffuse into the medium, they are non-enzymatically decarboxylated to diacetyl and 2,3-pentanediol [89]. As long maturation times are required for diacetyl degradation, minimizing the synthesis of this byproduct is desired. This can be achieved by adjusting the valine concentration in the wort as valine uptake results in feedback inhibition of acetohydroxy acid synthase, responsible for the catalysis of the diacetyl and 2,3- pentanediol precursors [90].

VOCs from fatty acids biosynthesis and degradation Another source of volatile metabolites is fatty acid synthesis and degradation. The most abundant free fatty acids released during yeast fermentations are hexanoic, octanoic, and decanoic acid [91]. Besides these medium chain fatty acids propanoic and butanoic acid have been detected in the headspace of yeast cultures, which might, however be rather byproducts from fermentation [91]. Fatty acids are decarboxylated to alkanes, 1-alkenes, and methyl-ketones or reduced to the respective aldehydes and 1-alkanols [92]. Aldehydes can also be formed by autoxidation of fatty acids [93].

11 General introduction

Lactones, e.g., γ-decalactone, γ-butyrolactone (peach flavor), 4-hydroxy cis-6-dodecenoic acid-γ- lactone (buttery flavor) [94] are derived from fatty acids via hydroxylation, β-oxidation and a final intramolecular esterification step that leads to molecule cyclization [95]. The acyl-CoA derivatives of volatile medium chain length fatty acids react with ethanol to form fatty acid ethyl esters (FAEE). Compared to the biosynthetic pathways for the acetate esters described above much less is known about the responsible enzymes and genes for these conversions and the regulation of these pathways. This might be due to the comparable lower levels FAEEs and the accordingly difficult measurements. These esters are, however, of similar industrial interest due to their fruity or floral flavors [85]. FAEE synthesis is hypothesized to be a detoxification mechanism under anaerobic conditions under which fatty acid synthesis is inhibited and medium chain length fatty acids are prematurely released from the fatty acid synthase. Ester formation avoids accumulation of these toxic acids and the diffusion of the volatile FAEE out of the cells further helps to shift the equilibrium of the esterification reaction towards ester formation [96].

VOCs derived from terpene biosynthetic pathways Terpenes are unsaturated hydrocarbons derived from isoprene and are mainly produced by plants. Their oxygenated derivatives, the terpenoids, have a major impact on the sensory profile of many foods and beverages [97]. It is known that yeasts transform terpenes present in the must into terpenoids thereby contributing to the flavor profile of wine. One example is the transformation of the monoterpene α-pinene to verbenol, a secondary allylic alcohol with camphor and mint-like flavor properties [98]. Also, biotransformations of the terpenoids linalool, α-terpineol, nerol, and geraniol by S. cerevisiae and other yeast have been reported [99] but the responsible enzymes for these conversions are ill-defined as only one Saccharomyces cyclase, the lanosterol cyclase Erg7p, is currently known [100]. The analysis of the biocatalytic potential of yeast or other microbes for the synthesis of terpenes and terpenoids is also of interest for the pharmaceutical industry as many of these compounds show promising biological activity [101]. Yeasts were assumed to be incapable of de novo synthesis of terpenes or terpenoids, with the exception of farnesol and geraniol. Yet, Carrau et al. showed the production of a variety of terpenes by wine yeasts of the genus Saccharomyces from simple carbon sources including linalool, α- terpineol, citronellol, nerol, furan linalool oxides, pyran linalool oxides, amongst others [4]. Carrau et al. further observed different responses in the synthesis of the terpenoids to varying supply of nitrogen and oxygen compared to each other and to sterol synthesis and concluded that these compounds are synthesized by two separate pathways. They hypothesized that the cytosolic mevalonate pathway whose end product is drained for sterol synthesis in the endoplasmic reticulum contributes to the synthesis of the sesquiterpenes farnesol and nerolidol, while the synthesis of the monoterpene geraniol might take place in the mitochondrion. They proposed that its synthesis is linked to the catabolism of leucine, which is converted to 3-hydroxy-3-methylglutaryl CoA (HMG- CoA) via 3-methylcrotonyl-CoA carboxylase (MCCase) (Figure 3). This enzyme has not been experimentally confirmed to be existent in Saccharomyces but strong genetic evidence exists for

12 Chapter 1

Aspergillus nidulans [102] and it is known to be present in the mitochondria of plants and mammals [103, 104]. The further conversion of geraniol to linalool, nerol, and citronellol is proposed to take place in the vacuole as the low pH might favor these reactions [105]. Again, no genes have been identified yet [100]. The missing information in the VOC biochemistry is in our opinion also strongly linked to the somewhat cumbersome experimental protocols consisting of sample preparation and analysis. The increasing interest in VOCs develops parallel with the advancements of analytical techniques, which are summarized briefly.

Figure 3: Mitochondrial monoterpene biosynthesis via the leucine catabolic MCC pathway and possible relationships with sterol formation in Saccharomyces cerevisiae as proposed by Carrau et al. [4]. Figure adapted from [4]. Previously published in [1].

13 General introduction

Analytical methods for the identification of VOCs The analysis of all VOCs an organism produces or even more the quantification of these VOCs faces several challenges entailing the so far limited knowledge of this compound class. Samples taken from the headspace of a culture are less complex compared to metabolome samples extracted from cells or culture broth, principally easing separation and measurement, nevertheless the chemical diversity of the analytes complicates the analysis as most techniques do not allow complete coverage. The volatile nature and dilute concentrations of VOCs further aggravate their measurements – both in terms of sampling and analytics – requiring efficient concentration techniques and / or sensitive analytics. Also, the analytical devices suitable for the measurement of volatiles are partially very specific, expensive or require expert knowledge for proper use impeding easy access to VOC measurements. Compound identification by mass spectral analysis is a further challenge because existing databases such as NIST or Wiley contain mostly VOCs of plants, animals and insects and do not cover exotic microbial VOCs. Furthermore, some compounds are so far completely unknown and have to be structurally elucidated, requiring mass spectrometry (MS) analysis of stable isotope labeled metabolites or nuclear magnetic resonance (NMR) spectrometry. Sodorifen, a VOC emitted by the rhizobacterium Serratia plymuthica, is one such novel compound, which has only been discovered recently [106, 107]. However, new or advanced sample extraction and analytical techniques allow today to broadly capture and survey volatile metabolites produced in microbial fermentations, some of which are briefly reviewed in the following.

Ion mobility spectrometry In ion mobility spectrometry (IMS), ionized gas phase molecules are separated based on different mobilities that are determined by the applied electric field, their size, mass, and shape, the IMS has been in use for decades [108]. Through the advancing computational development, ion mobility spectrometry technology could be used out of the lab since the 1970s when it was still called plasma chromatography. However, due to the increasing computation power that developed during the 1990s, major changes occurred in IMS technology and the understanding of its fundamental principles [5]. IMS is used to detect volatile compounds and is already used in military-, medical-, and customs control applications. In military, the IMS is used nowadays to detect chemical warfare agents and to warn soldiers, before the concentrations of chemical warfare agents become harmful [109]. At airports, the IMS is used to detect illegal drugs and explosives in a matter of seconds [110]. It is even possible to detect explosives in concentrations as low as 80 ppt, which has been proven for the explosive 2,4,6-trinitrotoluene (TNT) [108, 111]. In medicine, the detection of lung carcinoma can be achieved within minutes using the IMS [112-114]. Even the possibility of process control has been illuminated previously to this thesis [115]. There have also been other areas of application in the lab, as for example the detection of lipids and characterization of proteins. For more information about applications the reviews by Collins and Lee from 2002 and the reviews by Cumeras et al. from 2015 are recommended [108, 116, 117].

14 Chapter 1

Compared to gas chromatography the ion mobility spectrometry is a relatively unknown technique in life sciences. There are eight main methods that can be used to measure the ion mobility: 1. Drift time IMS (DTIMS), 2. Traveling–wave IMS (TWIMS), 3. High field asymmetric waveform IMS (FAIMS), 4. Trapped IMS (TIMS), 5. Open loop IMS (OLIMS) also called Aspiration IMS (AIMS), 6. Differential mobility analyzers (DMA), 6. Transversal modulation IMS (TMIMS) and 8. Overtone mobility spectrometers (OMS). Since this is not the focus of this thesis, only the basic DTIMS will be illuminated in more detail. For further information concerning ion mobility spectrometry, the review “Review on Ion Mobility Spectrometry” in two parts is recommended [108, 117]. Solid, liquid and gaseous samples can be analyzed by IMS. After injection of the sample, the analytes are ionized in the ion source. Radioactive sources [118], corona dischargers [119], photoionization sources [120, 121], matrix assisted laser desorption ionization (MALDI) sources [122], and electrospray ionizers [123] are most commonly used for analyte ionization. The ionized molecules are guided by an electric field and are usually held back by an ion shutter [124, 125]. The shutter opens periodically and releases the bundled ions into the drift tube. In the classic DTIMS the electric field constantly pulls the ions towards the detector, located at the end of the drift tube. The detector is classically a faraday plate but MS detectors are becoming more common (Figure 4) [116].

Figure 4: Working principle of a drift tube-IMS; adapted with permission of Journal of Physiology and Pharmacology from [5], previously published in [1, 6].

In detail, in the classic DTIMS, the drift tube is constantly flushed by drift gas, opposing the ions. Based on the attributes of an ion more or less collisions occur and some ions travel faster through the drift tube than others. This relationship is called ion mobility and defined as K. The ion mobility is dependent on the electric field intensity E. The drift velocity (ߴ݀ሻ is directly proportional to the ion mobility and the electric field intensity [126]:

ߴ݀ ൌ ܭܧ (2)

15 General introduction

The ion mobility can also be calculated using the length of the drift tube (L), the voltage drop across L, and the time it takes an ion to move through the drift tube (td):

ܮଶ (3) ܭൌ ଶ ݐௗ ܸ

At the molecular level K can be defined as fundamental relationship between ion mobility and the collisions that occur.

͵ݍ ʹߨ ଵ ݉൅ܯ ଵ ͳ (4) כ ሺ ሻଶכ ሺ ሻଶכ ൌܭ ͳ͸ܰ ݇ܶ ݉ܯ π

With q as charge of ions, N as number density of the drift gas, k as Boltzmann’s constant, T as absolute temperature, m as mass of the ion, M as mass of the drift gas, and Ω collision cross section of the ion in the drift gas. Since the temperature and the pressure are often different in different devices, respectively locations, the reduced ion mobility was introduced:

ܲଵ ܶ଴ (5) ܭ଴ ൌܭ ܲ଴ ܶଵ

Here P0 is 101,325 kPa and T0 is 273 K [126-128].

In classical DTIMS the drift gas is also ionized and in combination with residual water, reaction ion peaks (RIP) form. These peaks do not represent pure nitrogen or water, but ion clusters that are dependent on the used drift gas. For nitrogen containing gasses e.g., air, this is how the measured clusters are formed [129]:

+ N2 + e Æ N2 + 2e (1) + + N2 + 2N2 Æ N4 + N2 (2) + + N4 + H2O Æ H2O + 2N2 (3) + + H2O + H2O Æ H3O + OH (4) + + H3O + H2O + N2 Æ H (H2O)2 + N2 (5) + + H (H2O)n-1 + H2O + N2 Æ H (H2O)n + N2 (6)

The choice of drift gas and to some extent the water content can be influenced by the experimenter. Higher water concentrations cause the formation of larger ion clusters that have a lower ion mobility, and hence lead to longer drift times [130]. Thus, for IMS applications such as online process control varying water content has to be avoided. This can be achieved using a chromatographic pre-separation of the sample, for example using a multi capillary column (MCC), which separates water from other analytes [131]. 16 Chapter 1

MCCs consist of about 1000 parallel short columns of 20 cm to 1 m length each. Because of the large cross-sectional area of these parallel columns, it is possible to work with high gas flow rates (up to 250 mL/min) resulting in much shorter separation times of MCCs compared to classical GC capillary columns. The water in the samples does not interact with the column material, it thus directly passes the MCC and is flushed out of the system while other components have longer retention times and are introduced into the IMS. This pre-separation step increases the overall separation performance and is equally suited for online measurements as unhyphenated IMS as it can be used without prior sample preparation [131].

Secondary electrospray ionization mass spectrometry A new and highly efficient way of measuring volatile compounds online is the use of a secondary electrospray ionization (SESI) source coupled to an Orbitrap mass analyzer. Compared to IMS, this MS based system enables the identification of analytes by fragmentation of the compounds and analyses of the resulting mass spectra [132-135]. To prevent condensation of metabolites with low boiling points, analytes are guided through a heated transfer line into the ion source. Sample ionization via SESI proceeds in two steps. First a liquid (usually 0.1% formic acid in water) is ionized by an electrospray in which the liquid is exposed to a high voltage to form an aerosol of charged, minute droplets. Then, the sample gas flow is introduced into this mist and the secondary ionization occurs through ion-molecule interaction. A recent optimization of this system by Barrios-Collando et al. resulted in a 5-fold improvement in sensitivity compared to standard SESI [136]. The ionized sample is guided into the Orbitrap [132], where the ions oscillate along the z-axis of a spindle-like inner electrode. Based on their mass to charge ratio (m/z) the analytes have different frequencies in traveling back and forth along the z-axis. This frequency is detected by the outer barrel like electrode. SESI-MS has successfully been used to discriminate between smokers and non-smokers [133] or to screen the volatile space of plants. For the latter study, plants were cultivated in a closed glass beaker flushed with a constant air flow, which was directly introduced into the SESI-MS device. In these experiments, about 400 VOC species were identified which correlated with the diurnal cycle of Begonia semperflorens and about 1200 VOCs were related to mechanical damages of the plant [134].

Proton-transfer-reaction time-of-flight mass spectrometry Proton-transfer-reaction time-of-flight mass spectrometry (PTR-TOF-MS) is another online and + highly sensitive measurement technique for VOCs. The ionization of VOCs by H3O ions, applied here, is similar to that in IMS except that PTR-MS generally does not use a radioactive, but a hollow + cathode ion source, which generates H3O ions from water through electric discharges. The gaseous sample is introduced into the drift tube containing the ionized water, and analytes with a higher proton affinity than water are protonated. The ionized molecules are periodically guided through a transfer lens system and accelerated into the TOF-MS, where they are electrically deflected onto a specific trajectory. Based on their mass to charge ratio the flight time of the ions differs. Thus, through every pulse created by the lens system a complete mass spectrum can be measured [137]. PTR-TOF-MS 17 General introduction and PTR-MS (usually equipped with a quadrupole mass analyzer) devices have been used to detect microorganisms, including human and plant pathogens [138-143]. Also lactic acid fermentation during the production of yogurt and the storage processes of milk and meat were monitored [139, 144-147]. Moreover, VOC profiles of dough fermentations with various yeasts and flours were used to determine product quality. For example, specific masses of three unidentified compounds were suggested as biomarkers for the optimization of the aromatic quality of bread dough [139, 148-150].

Gas-chromatography based methods The most common method for offline measurements of volatile metabolites is gas-chromatography after appropriate sample extraction (see section 3.5) although in some cases direct sample injection is also possible. When liquid samples are to be measured, they are evaporated in the injector unit, while gaseous samples can directly be injected onto the chromatographic column. The chromatographic separation is based on analyte interaction with the stationary phase of a coated gas chromatographic column. Thin capillary columns of 10 m to 100 m length, an inner diameter between 0.1 and 0.33 mm and a film thickness between 0.1 and 0.33 μm are typically used, which limits the maximal gas flow rates to 1-10 mL/min. The analytes are transported through the column by a carrier gas (e.g., helium or hydrogen) and are separated depending on their interaction with the column material. The interaction strength depends on the chemical properties of the analytes and the conditions in the column, especially temperature and pressure, which in sum lead to different retention times and hence separation of the analytes [151-153]. For increased analyte resolution, two-dimensional gas chromatography can be employed, which connects the first column with a second, shorter column, with a typical length of 1-2 m, in daisy-chain. The second column is usually coated with a different stationary phase, consequently leading to different retention times of the analytes [154-156]. Two- dimensional gas chromatography has already been used to investigate the volatiles in wine in great detail and allowed to analyze hard to separate chemical compounds, such as ethyl carbamate [157]. Common GC detectors are flame ionization (FI) and mass spectrometer (MS). In VOC research, olfactometers are also widely used in conjunction with specially trained persons to directly evaluate the olfactory effect of single or complex mixtures of VOCs [158, 159]. This approach exploits the human sense of smell and is commonly used to determine the sensory activity of VOCs and odor threshold concentrations [160]. Also, product quality can be assessed which is difficult to achieve without an olfactometer, since many substances smell good only in a limited range of concentration and importantly in combination with other substances. GC-olfactometry has been used to analyze the complex VOC mixtures of wine and to link VOCs to the organoleptic properties [161]. A plethora of further GC-based applications for volatile metabolite measurements have been reported and are reviewed, for example in [162].

18 Chapter 1

Offline-sampling IMS, PTR-TOF-MS, SESI-MS and - in limited applications - GC-Q-MS are suited for online measurements without any upstream sample preparation. While often advantageous, depending on the scientific problem, online measurements are not always necessary and sample preparation can be useful to concentrate low abundant analytes. Subject to the extraction method used, it is possible to selectively concentrate compound classes. On the downside, this selective binding leads to loss of analytes that do not interact with the solvent or fiber. A common technique for sample preparation is thermal desorption spectroscopy (TDS) in which a column packed with adsorbent is flushed with the gaseous sample. Depending on gas flow rate and sampling time the amount of analyte adsorbed to the packing material can be adjusted. It is also possible to directly adsorb metabolites during the fermentation process by using special stirring bars, made of adsorbing material. This technique was previously used to measure off-flavor producing components in wine and to profile the metabolites of different fungi [163, 164]. Analyte extraction via purge and trap is achieved by stripping the liquid sample with an inert gas, which is then passed through a sorbent trap to bind the extracted metabolites. These traps are filled with equivalent adsorbent material as used in TDS. The analytes are released by thermal desorption and transferred onto the GC column. This approach has been used to measure volatiles in red wine or to determine the impact of altered alcohol acetyltransferase expression on the synthesis of esters and alcohols of different yeast strains [165, 166]. In solid phase microextraction (SPME), a fiber of the adsorbing material is loaded with volatile metabolites by introducing it into the headspace of the liquid sample. SPME has been used to analyze diverse biological samples including fungi and human samples for cancer marker detection [159, 167, 168]. After loading of the adsorbent, the TDS column or SPME fiber is connected to a gas flow leading to a GC-MS system. For efficient desorption of the bound analytes, the adsorbent is heated. The disadvantage of this sample preparation approach is that all analytes are desorbed and the sample can only be measured once [167]. Common adsorbents are polyacrylate, CARBOWAXTM (polyethylene glycols, methoxypolyethylene glycols), CarboxensTM (carbon molecular sieve), polydimethylsiloxane, Tenax® TA (poly(2,6-diphenyl-p-phenylenoxid)), and silica gel [164, 168]. Direct immersion single drop microextraction (DI-SDME) is a miniaturization of classical liquid- liquid extraction which greatly reduces the extractant to sample ratio and achieves higher analyte concentrations [169-172]. A water-immiscible drop hanging at the tip of a needle is submersed into the liquid sample. The drop is not mixed with the liquid but keeps attached to the needle. When the extraction is at equilibrium the drop is withdrawn back into the needle and directly injected into the GC-MS. The extraction efficiency can be enhanced by continuously pumping the fluid sample alongside the drop as it is done in continuous flow microextraction (CFME) [169, 173]. A variant of this technique for gaseous analytes is headspace single drop microextraction (HS-SDME) in which the drop is placed in the headspace of the liquid sample and only volatile metabolites present in the gas phase are extracted [169, 174]. SDME theoretically allows extraction of all volatile metabolites as the solvents can be freely chosen. Only the following requirements apply: The viscosity of the extraction solvent has to be high enough to form a drop that sticks to the needle. The solvent used in

19 General introduction

DI-SDME and CFME has to be insoluble in the aqueous sample and the extractant for HS-SDME has to have a sufficiently low vapor pressure to prevent evaporation at the applied temperature. Commonly used solvents are toluene, n-octanol, decane and benzyl alcohol. Automated SDME workflows have been reported, but the technique is still not routinely used in many laboratories [169, 175]. For more detailed information about SDME the review by Xu et al. is recommended [169].

20 Chapter 1

Pros and cons of methods for VOC detection The analytical techniques presented here have pros and cons, and the user has to choose the most appropriate method according to the application requirements but also available budget (Table 1). Ion mobility spectrometers combine relatively low costs (acquisition, operation, maintenance) with a low detection limit, and the operation at atmospheric pressure allows hand-held IMS devices [109]. IMS can be used for online measurements as ionization is continuous and no derivatization is required. If the IMS is used without a chromatographic pre-separation, 20 spectra per second can be acquired. Combined with a MCC the acquisition is in the range of few minutes, still fast compared to common GC analyses, which often take 30 minutes or longer. However, IMS alone does not allow compound identification. This is possible with hyphenated IMS-MS devices which however, come with the disadvantage of stationary use and higher price [176]. As GC-MS techniques have been used for a long time and for a variety of analytical applications, comprehensive, commercial databases such as NIST are available, while only comparably small databases exist for IMS applications [177]. Data analysis is further complicated by modifications of ionized analytes that occur by collisions with other ions and uncharged drift gas during passage through the drift tube [178]. This process is influenced by the ion source, the strength of the electric field and as well as the length of the drift tube impeding the set-up of a consistent database. As a consequence reliable compound identification currently requires the measurement of pure standard substances but endeavors exist to establish workable IMS-MS databases [130]. A further disadvantage of IMS is the high gas requirement (carrier gas, drift gas) compared to GC if no air-loops are used. In GC-MS systems often quadrupole (Q) mass analyzers are the detectors of choice. GC-Q-MS systems are usually more expensive than IMS but much cheaper than those with TOF or Orbitrap detectors (Table 1). In GC-MS analyses compounds are identified based on their mass, fragmentation pattern and retention index. It is possible to scan the full mass spectrum or to filter single masses in selected ion monitoring (SIM) mode. The mass resolution of a quadrupole MS is limited to about half a mass unit thus identification of non-fragmented analytes is rarely possible. Since the samples usually need to be water-free before entering the GC column, online measurements are often not possible. Also, the mass spectrometer is not mobile as it requires carrier gas and a turbo pump for the high vacuum of about 1-100 mTorr. TOF detectors are extremely fast and can measure mass spectra in the kHz region. In current devices, however, the actual data acquisition time is limited by the sample inlet time and ionization. Compared to Q-MS, TOF detectors have a higher mass range (up to 36 000 m/z) and a higher mass resolution [179]. The mass resolution of the TOF is sufficient to estimate the elementary composition of most analytes, which can be used to tentatively identify detected compounds. However, if a soft ionization method such as PTR is used isomers cannot be distinguished as the metabolites rarely decompose into informative fragments. Orbitraps are the most expensive but also the most sensitive mass analyzers presented here. They are also the newest detectors on the market, since the first publication describing an Orbitrap was only published in the year 2000 [180]. Furthermore, Orbitraps have the highest mass accuracies, high mass ranges and the highest mass resolutions of all the detectors introduced here. Through the high

21 General introduction mass resolution, it is possible to separate peaks that even in TOF detectors are measured as one peak. The higher mass accuracy results in improved determination of the elemental composition, and consequently more conclusive compound identification. Structural identification can further be achieved by fragmentation of selected compounds. However, since there is usually no pre-separation it is not useful to fragment all metabolites while measuring complex samples.

Table 1. Features of different analytical devices for volatile compound measurement. Data on resolving power taken from [181-184], all other data taken from product sheets of current devices; LOD, limit of detection. Previously published in [1]. SESI- Device IMS GC-Q-MS PTR-TOF Orbitrap Price category $ $$ $$$ $$$$ Resolving power 60 2,8 up to 5,000 60,000 to >100,000 Scanning speed 20 Hz 97 Hz 10 Hz (TOF 20 Hz 200 kHz) Portability limited none none none LOD ppt ppt ppt ppt Application Environmental (Online) Real-time analyses examples monitoring (e.g., VOC analyses of monitoring of of plant VOC water plant or human microbial emission, contamination), samples, pesticide contamination of breath analysis, breath analysis screening [186- food, detection of [133, 134] (biomarker 188] bacterial infections detection), and drug level detection of control via breath chemical warfare analysis [138, 139, agents [109, 113, 141, 144, 147, 131, 185] 189]

Methods for the delineation of VOC metabolic pathways As outlined earlier in the introduction volatile metabolites are used in interspecies communication, can signal metabolic shifts and are of industrial interest for use as fragrances, food flavor enhancer or biofuels. We described tools for the investigation of the volatile spectrum of yeast, but to understand the link between specific VOCs and the metabolic mode and to eventually exploit the metabolic machinery of an organism for the production of valuable VOCs, it is crucial to also know the metabolic and regulatory network involved in the biosynthesis of these compounds. The current knowledge of microbial VOC biosynthesis and regulation is rather scarce and scattered although recent endeavors exist to aggregate microbial VOC data including metabolic pathways in one 22 Chapter 1 database [66]. These knowledge gaps even exist for the industrial workhorse and eukaryotic model organism S. cerevisiae. Having been investigated in numerous scientific studies and 78% of its 6,604 ORFs been assigned to an enzymatic, regulatory or further cellular function [190], its biochemistry is generally well established, documented in several databases [191-193], and well-curated genome- scale models exist [194-196]. Remarkably, however, biosynthetic pathways of volatile metabolites are only moderately covered. This incompleteness of metabolic models and databases became apparent when checking VOC coverage of metabolic pathway databases and models: Of 100 VOCs identified in the literature to be synthesized by yeast only 20% were contained in yeast specific metabolic databases and the S. cerevisiae latest genome-scale model [185]. A suite of computational and experimental methods is available to close this lack of knowledge, i.e., to identify the metabolic network underlying the synthesis of these metabolites. For the in silico identification two approaches exist: The bottom-up network reconstruction aims to predict possible pathways from available metabolic reaction networks or databases, while the top-down approach takes condition-specific omics data and extracts possible networks using statistical and bioinformatics analyses. The latter does not rely on a priori knowledge of network component interactions and is therefore well suited for the prediction of less studied biosynthetic pathways and regulatory networks such as those of volatile metabolites. Also, powerful methods exist that combine both approaches [197-199]. Metabolic networks predicted by either approach still have to be manually pruned and experimentally verified. After integration of the novel pathways into an existing metabolic model, the extended model can be used for the computation of engineering strategies to improve or prohibit VOC synthesis or to interrogate interactions between VOC synthesis and the overall metabolic behavior of the cell. Metabolic models further allow to address questions, such as to why the organism releases volatile metabolites. For example, the occurrence of the Crabtree effect in S. cerevisiae, that is aerobic ethanol formation, was in silico reproduced by stoichiometric models taking into account the investment costs for the synthesis of enzymes or mitochondria. These simulations disclosed a trade-off between these anabolic costs and the metabolic yield of alternative fermentative pathways [200, 201] at high glycolytic fluxes. In this way, experimentally testable hypotheses can be generated and engineering strategies derived to improve the metabolic activity towards a defined target function. For a detailed description of the use of metabolic models, we refer to [202, 203].

Bottom-up approaches The basis of bottom-up approaches for the prediction of biosynthetic pathways is a (genome-scale) metabolic model of the organism under study. Such a metabolic model is generated based on the information provided by the annotated genome and experimental data, gathered from databases and the scientific literature and rationalizes and structures the existing knowledge, scattered in these resources. For metabolic pathway prediction, stoichiometric models, which do not capture kinetic properties, are sufficient. These models store the stoichiometry of the enzymatic repertoire of the cell and the gene-enzyme-reaction relationship and thereby allow to explore the metabolic capabilities of an organism and the effect of environmental or genetic perturbations.

23 General introduction

To comprehend metabolic models with enzymatic capabilities for the synthesis of additional metabolites, a retrosynthetic approach is employed which parses species-unspecific metabolic databases such as KEGG, MetaCyc or BiGG [204-206] for enzymatic transformations that allow connecting the novel metabolite to the existing network. Several publicly available or web-based tools exist for this bottom-up prediction of metabolic pathways [207-209]. The approach presented by Christian et al. [197] differs from these pathway prediction tools in that it directly derives models that allow the synthesis of a novel compound from a defined carbon and energy source instead of isolated pathways. The original network is first appended with all reactions of a universal metabolic database and then stepwise reduced by eliminating previously added but non-essential reactions to eventually derive at a minimally extended model capable of target compound synthesis. As the order of reaction removal impacts the model extension, a huge number of parallel runs is performed. In order to increase the number of biologically meaningful models in this set and to reduce a posteriori pathway selection and verification, the algorithm integrates pathway prioritization into the pathway prediction process. This is enabled by ranking all foreign reactions according to the probability that the catalyzing enzyme is encoded in the organism’s genome and a preferential extension with high-ranked reactions. The Biochemical Network Integrated Computational Explorer, BNICE, [210] goes beyond reconfiguring known enzymatic reactions and uses instead a set of generalized enzymatic reaction rules to generate metabolic networks built of existing and novel, i.e., previously unobserved, biochemical reactions and compounds. The use of reaction rules is organized according to the enzyme classification system but reduced to the first three digits that define the reaction mechanism but do not constrain the substrate. This approach eases the identification of enzyme candidates for de novo biochemical reactions. Not relying on metabolic reaction databases, this approach can also predict non-enzymatic reactions such as decarboxylations or oxgenations. The drawback of BNICE is the extensive pathway extraction and validation procedure, which follows the network inference step. Although mainly applied to predict novel pathways for the heterologous synthesis of industrially interesting products, the general applicability of BNICE allows to equally apply it for the delineation of uncharted metabolic pathways as recently shown on the example of lipid metabolism [211]. The integration of systems-wide omics data, mRNA, protein or metabolite levels into metabolic modeling constrains the model to more physiological relevant solutions that is reaction rates more likely reflecting the in vivo enzymatic activities for the specific experimental setup [212-217] Embedding such constraints into the pathway computation procedure is desirable as this would significantly reduce the number of predicted pathways and aid in filtering out meaningful solutions.

Verification of predicted pathways Top-down and bottom-up network predictions often result in a plethora of pathway candidates, which have to be pruned, ranked and experimentally verified. A basic requirement for a functional pathway is a stoichiometric and thermodynamic feasible synthesis of its product from substrates provided in the growth experiment in which its production

24 Chapter 1 has been observed. This can be verified by (thermodynamics-based) flux balance analysis [218, 219] and models shown to be stoichiometrically or thermodynamically infeasible are to be rejected. Experimental identification of pathway specific intermediates by targeted metabolite profiling gives strong indications for pathway activity. This method is applicable for secondary metabolites, produced by linear pathways, whose intermediates do not participate in other reactions but less informative for pathways involving highly connected metabolites. A powerful alternative for the discrimination of active pathways is an isotope labeling experiment. In the early days of biochemical pathway analysis, the conversion of metabolites has been traced by tracking the incorporation of heavy, radioactive isotopes. Today, stable isotopes are used and the resulting isotope isomers (isotopomers) analyzed by MS or NMR techniques. The introduced tracer has to be carefully designed so that the activity of alternative pathways results in distinct labeling pattern of pathway intermediates or end products. Computational methods exist that aid in the design of informative tracer experiments and data evaluation [220-223]. Top-down approaches integrating transcriptome and/or proteome data directly yield predictions about genes and enzymes involved in VOC synthesis, while for many bottom-up prediction tools, gene candidates have to be defined a posteriori. BridgIT is such a computational framework that builds on BNICE results and suggests enzyme candidates for novel reactions identified in the pathway prediction process [224]. Following this, BLAST searches are applied to identify gene candidates encoding these enzymes. The proposed gene-enzyme-reaction relationships are to be validated experimentally. For this purpose, gene deletion mutants are generated and subsequently evaluated for VOC synthesis or screened for the accumulation of proposed pathway intermediates. While verifying or disproving alternative one-step pathways catalyzed by single enzymes is straightforward, finding gene knockout targets of intertwined and longer pathway alternatives is less intuitive. Computational tools such as the forced coupling algorithm FOCAL support the identification of genetic and environmental conditions for conclusive model discrimination [225].

Aims The overall goal of this thesis was the exploration of the volatile metabolites of baker´s yeast. Although well known to many of us from daily life, as outlined in the introduction, much is unknown and only limited information about the interrelationship between phenotype and genotype exist. To close the gap this thesis shall be listing the known volatile metabolites from the literature and combine this information with the massive information on the biochemistry of known-volatiles. The information gathered shall be used for the enhancement of the consensus metabolic yeast model. The environmental driven production of volatiles shall be investigated, to be able to exploit this noninvasive signal from yeast metabolism in industrial applications. Here, finding a reporter metabolite that indicates the metabolic shift from respiratory to respiro-fermentative metabolism. This reporter shall be used in the YeastScent project, to regulate the feed rate of a yeast fed-batch fermentation to maximize the biomass yield while maximizing the growth rate, hence by operating the fed-batch below the critical growth rate that triggers the Crabtree effect. This transition point was investigated in great detail, but not in respect of volatile production. Here, the transition from respiratory to respiro-fermentative growth (Crabtree effect) was investigation in 25 General introduction industrial-like conditions by volatile and transcriptome analyses. These aims are connected by the YeastScent project that uses a process model based prediction for fed-batch pump control with a volatile reporter metabolite as proxy for yeast metabolism.

Chapter 2

Results

Expanding the consensus yeast model by volatile metabolism

Contributions: The study was designed by Christoph Halbfeld, Birgitta E. Ebert and Lars M. Blank. The experiments were performed by Christoph Halbfeld, Birthe Halmschlag and Niklas Kitschen. The data was evaluated by Christoph Halbfeld and Birgitta E. Ebert.

Chapter 2

Results Expanding the consensus yeast model by volatile metabolism Summary The consensus metabolic yeast model has so far been lacking the representation of the volatile metabolite space. To overcome this shortcoming, a literature investigation of all known volatile metabolites produced by S. cerevisiae was performed and 92 volatile metabolites were found of which 78 did not exist in the latest metabolic model. To connect the newly found metabolites to the metabolic network, 219 reactions were added to the consensus yeast metabolic model using computational tools as well as manual integration of pathways. Stoichiometric and thermodynamic feasibility of steady state production of the amended volatile metabolites was verified with flux balance analysis and thermodynamics-based flux analysis. 12 reactions were further experimentally verified by enzymatic assays of knockout mutants, which contained only one enzyme with alcohol dehydrogenase activity. The new version of the consensus yeast model, Yeast8, covers the yeast volatile metabolite space and represents the necessary basis to shed light on the metabolic constraints underlying volatile metabolite formation, offering new possibilities in modeling for example to create yeasts with a special aroma.

33 Results

Introduction Through computational advancements it became possible to simulate microbial behavior in silico. However, performing these simulations requires comprehensive knowledge about the biochemical reactions that occur inside a cell. For the well investigated baker’s yeast Saccharomyces cerevisiae, such knowledge is already accessible and has been converted into a metabolic model in a standardized informatic language, the systems biology markup language (SBML). The first consensus yeast model was based on two older models the iMM904 [226] and the iLL672 [196]. The new model was created using SBML and MIRIAM standards. The new model introduced name conventions as well as identifiers and is based on S. cerevisiae S288C [227]. The first major enhancements were made, with version 4 of the model, which was appended with lipid metabolism [228]. In version 5 of the model, the sphingolipid metabolism was added. Also, a script to test for anaerobic growth conditions was added. As well in version 5, an apportionment in genome scale models (GEMs) and genome scale network reconstructions (GENRES) was introduced. This enables the possibility to use models based on a specific task. For modeling approaches GEMs are used, where assumptions are made that enable modeling but have no experimental evidence. GENREs can be used if established knowledge that has been verified experimentally is needed [229]. In version 6 of the model general enhancements were included, such as the manual curation of the model [194]. In the 7th and still current model, the fatty acid, glycolipid, and glycerophospholipid metabolism of the yeast model was curated in great detail [230]. Even though the model was constantly updated in the past, the representation of the volatile space was not yet tackled. Except for the well investigated volatiles ethanol and acetaldehyde only few volatile metabolites could be found in the model representation [185]. The lack of volatiles in the model is most likely based on difficulties that come along with detecting volatile organic compounds (VOCs). Recently better analytical techniques have become available and thus more and more detailed information on the volatile space of S. cerevisiae will emerge by time. To fully understand the metabolic processes inside of S. cerevisiae the inclusion of volatile metabolites to the consensus metabolic yeast model is crucial, therefore here the model is enhanced by the VOCs that were found by a literature investigation [185]. The addition of new pathways to a model is always challenging. The information gaps between the added metabolites and the already existing network have to be closed by the most probable reactions. To close the gaps several tools are available for example, FMM, Pathpred and Metaroute [207-209]. These tools search for possible pathways using already known reactions from a database listing data of multiple organisms, e.g., KEGG [231-233]. The Biochemical Network Integrated Computational Explorer (BNICE) is a tool that suggests new routes based on biochemical rules. It does not rely on known reactions and, hence, completely novel pathways can be created [234]. In addition, manual integration of new reactions is an option. Therefore, enzymes catalyzing similar reactions need to be identified as they could potentially act on more than one substrate, thereby contributing to novel pathways. Ideally, new pathways are also experimentally verified, e.g., by 13C-tracer experiments or biochemical assays. Here, we set out to comprehensively append the Yeast 7 model with biosynthetic pathway for volatile metabolites based on extensive scientific literature and database searches as well as computational analyses. Pathway predictions were checked for thermodynamic feasibility and

34 Chapter 2 partially verified by wet-lab experiments.[219, 235] Another way and more solid is experimental verification of added reactions, was partially performed here. We focused on reactions catalyzed by alcohol dehydrogenases (Adhps) as their biocatalytic spectrum of volatile alcohols or aldehydes synthesis has already been investigated. Although industrial alcohol production is a highly valuable yeast process; the single enzymes were biochemically characterized for all tested alcohols.

Material and Methods Software tools for pathway prediction For pathway prediction, the web based tools FMM (http://fmm.mbc.nctu.edu.tw/) [207], PathPred (http://www.genome.jp/tools/pathpred/) [209], and MetaRoute (http://abi.inf.uni- tuebingen.de/Services/MetaRoute/) [208] were used. In addition, BNICE [234] analyses were run at the laboratory of computational systems biology (LCSB, Prof. Vassily Hatzimanikatis). Flux balance analysis (FBA) for the verification of the stoichiometrical feasibility of metabolic pathways was performed using the Yeast 7.5 Cobra yeast model and the COBRA- Toolbox for Matlab [230, 236]. Linear programming problems were solved with the commercially available IBM CPLEX solver (IBM, North Castle, NY, USA). Maximization of the production of the respective volatile pathway product was used as objective function, the carbon source uptake rate was set to 1 mmol/(gCDW x h), in order to relate all other fluxes to the carbon source uptake rate. The lower bound for the ATP production reaction was set to 1 mol/mol glucose uptake. To inhibit falsely added new reactions from disturbing the tests all new reactions but the tested one were set to zero. Thermodynamics-based flux analysis (TFBA) was performed at the Laboratory of Computational Systems Biotechnology (LCSB) in Lausanne Switzerland [219]. A script for performing the TFBA is available there, however it is not publicly accessible.

Strains, media and plasmids used S. cerevisiae CEN.PK 113-17a (Euroscarf, Oberursel, Germany) was cultivated in Verduyn minimal salt medium [237]. The medium contained 20 g/L glucose, 5 g/L (NH4)2SO4, 3 g/L KH2PO4, 0,5 g/L MgSO47H2O, as well as 1 mL 1000x trace elements and 1 mL of 1000x vitamin solution. The vitamin solution contained 0.05 g L−1 D-biotin, 1 g L−1 calcium D- pantothenate, 1 g L-1 nicotinic acid, 25 g L−1 myo-inositol, 1 g L−1 thiamine hydrochloride, 1 g L−1 pyridoxine hydrochloride and 0.2 g L−1 p-aminobenzoic acid. The trace element solution −1 −1 −1 −1 consisted of 15 g L EDTA, 4.5 g L ZnSO4·7H2O, 1 g L MnCl2·4 H2O, 0.3 g L CoCl2·7 −1 −1 −1 −1 H2O, 0.3 g L CuSO4·5 H2O, 0.4 g L NaMoO4·2 H2O, 4.5 g L CaCl2·2 H2O, 3 g L −1 −1 FeSO4·7 H2O, 1 g L H3BO3 and 0.1 g L KI. The strains containing the Cas9 containing plasmid pCfB2312 were cultivated in Verduyn medium containing 200 μg mL-1 geneticin (G418). E. coli DH5α (Invitrogen, Carlsbad, CA, USA) was grown in LB media containing 10 g/L tryptone, 5 g/L yeast extract and 5 g/L NaCl. All strains used are listed in Table 2.

35 Results

Table 2: Microbial strains strain relevant characteristics Origin E. coli supE44 ΔlacU169 (φ80 lacZΔM15) hsdR17 (rk-mk+) DH5α Eurofins recA1 endA1 gyrA96 thi-1 relA1 DH5α pCfB2312 DH5α bearing pCfB2312 [238] DH5α pGRNA- DH5α bearing pGRNA-ADH1 this work ADH1 DH5α pGRNA- DH5α bearing pGRNA-ADH2B this work ADH2B DH5α pGRNA- DH5α bearing pGRNA-ADH3 this work ADH3 DH5α pGRNA- DH5α bearing pGRNA-ADH4C this work ADH4C DH5α pGRNA- DH5α bearing pGRNA-ADH5 this work ADH5 DH5α pGRNA- DH5α bearing pGRNA-ADH6 this work ADH6 DH5α pGRNA- DH5α bearing pGRNA-GRE2C this work GRE2C DH5α pGRNA- DH5α bearing pGRNA-SFA1 this work SFA1 DH5α pUSER- DH5α bearing pUSER-ADH2B this work ADH2B DH5α pUSER- DH5α bearing pUSER-GRE2C this work GRE2C DH5α pUSER- DH5α bearing pUSER-NKO2 this work NKO2

S. cerevisiae MATα; ura3-52; leu2-3_112; TRP1; HIS3; MAL2-8C; CEN.PK113-17a Eurofins SUC2 CEN.PK113-17a CEN.PK-113-17a bearing pCfB2312 this work pCfB2312 sADH1 Δadh2 Δadh3 Δadh4 Δadh5 Δadh6 Δgre2 Δsfa1 this work sADH2 Δadh1 Δadh3 Δadh4 Δadh5 Δadh6 Δgre2 Δsfa1 this work sADH3 Δadh1 Δadh2 Δadh4 Δadh5 Δadh6 Δgre2 Δsfa1 this work sADH4 Δadh1 Δadh2 Δadh3 Δadh5 Δadh6 Δgre2 Δsfa1 this work sADH5 Δadh1 Δadh2 Δadh3 Δadh4 Δadh6 Δgre2 Δsfa1 this work sADH6 Δadh1 Δadh2 Δadh3 Δadh4 Δadh5 Δgre2 Δsfa1 this work sSFA1 Δadh1 Δadh2 Δadh3 Δadh4 Δadh5 Δadh6 Δgre2 this work sGRE2 Δadh1 Δadh2 Δadh3 Δadh4 Δadh5 Δadh6 Δsfa1 this work

Enzymes and PCR components All enzymes used in this work were purchased from New England Biolabs (Frankfurt, Germany), Agilent (Santa Clara, USA) and Fermentas (St. Leon-Rot, Germany). All oligonucleotides (Table 3 and Table 4) were purchased from Eurofins MWG Operon (Ebersberg, Germany). Sequencing was also carried out by Eurofins MWG Operon (Ebersberg,

36 Chapter 2

Germany). Table 3: Primers (table continues over multiple pages) Primer name Primer sequence Properties amplification of adh1 for sequencing. A1 seq fw CTACGAATCCCACGGTAAG Annealing at: 48 °C amplification of adh1 for sequencing. A1 seq rv CTGGCGAAGAAGTCCAAAGC Annealing at: 48 °C amplification of adh2 for knockout A2B_seq fw CGTCTTCAGAGCTCATTG confirmation. Annealing at: 46 °C amplification of adh2 for knockout A2B_seq rv GGCATCCTTGACACATTC confirmation. Annealing at: 46 °C amplification of adh3 for sequencing A3 seq fw GCAATCCACAGCTGCAATC Annealing at: 48 °C amplification of adh3 for sequencing. A3 seq rv CGGTACCACATGGTCTAAC Annealing at: 48 °C amplification of adh4 for knockout A4B_seq fw CGTAGTGCGTTACAGTTC confirmation. Annealing at: 46 °C amplification of adh4 for knockout A4B_seq rv ATATTGCGGCTGGTAAGG confirmation. Annealing at: 46 °C amplification of adh5 for sequencing. A5 seq fw GCCTTCGCAAGTCATTCC Annealing at: 48 °C amplification of adh5 for sequencing. A5 seq rv CAGGACGACAGTACCATTG Annealing at: 48 °C amplification of adh6 for sequencing. A6 seq fw CAATCACACGAAGATTGG Annealing at: 43 °Cv amplification of adh6 for sequencing. A6 seq rv GGAACCTAAAGCACTGTAAG Annealing at: 43 °C amplification of gre2 for knockout Gre2C_seq fw CAACAATTGGCCCTCACCTC confirmation. Annealing at: 52 °C amplification of gre2 for knockout Gre2C_seq rv TGTGGGGAGACGGGTAGAAG confirmation. Annealing at: 52 °C amplification of sfa1 for sequencing. S1 seq fw GCTGCTGTTGCGTATGATG Annealing at: 49 °C amplification of sfa1 for sequencing. S1 seq rv CAGGCTTCCAAAGCATCTC Annealing at: 49 °C

37 Results

gRNA ACTGACTTGCACGCTTGGCAGTTTTAGAGCTA specific 20bp + structural A1 fw GAA part ADH2 CTGGGGTCGTTGTTGCTATCGTTTTAGAGCTA specific 20bp + structural B GAA part gRNA GTCACACCGATTTACATGCTGTTTTAGAGCTA specific 20bp + structural A3 fw GAA part gRNA TTCATAGGCTTTCTTGATAAGTTTTAGAGCTA specific 20bp + structural AHD4 GAA part C gRNA GTCATAGTGACTTGCACGCGGTTTTAGAGCTA specific 20bp + structural A5 fw GAA part gRNA ATGAAGATGCCGCTAGTCGTGTTTTAGAGCTA specific 20bp + structural A6 fw GAA part GRE2 GACCCAGTTAACGCCTACTGGTTTTAGAGCTA specific 20bp + structural C GAA part gRNA TTATCAGGCTCTGATCCAGAGTTTTAGAGCTA specific 20bp + structural S1 fw GAA part gRNA sequencing of single GGTCAAACGCTGTAGAAG seq gRNA-plasmids gRNA TJOS-20 reverse primer for GATCATTTATCTTTCACTGCGGAGAAG all rv all gRNAs [238] amplification of USER P1 fw CGTGCGAUAGGGAACAAAAGCTGGAGCT insert amplification of USER P1 rv CACGCGAUTAACTAATTACATGACTCGA insert amplification of USER P2 fw AGTGCAGGUAGGGAACAAAAGCTGGAGCT insert amplification of USER P2 rv ACCTGCACUTAACTAATTACATGACTCGA insert USER colony PCR to check GTCAGTGAGCGAGGAAG check 2 USER reaction USER colony PCR to check check GTACCGGCCGCAAATTAAAG USER reaction fw USER sequencing after USER GTACCGGCCGCAAATTAAAG seq reaction USER sequencing after USER CCGCTCACAATTCCACAC seq fw reaction USER CTGGCACGACAGGTTTC sequencing of USER site site seq USERi TJOS-21, integration of ATGTTGTGTGGAATTGTGAG nt fw USER sequence [238] USERi GAATGCGTGCGATCGCGTGCATTCATGAGTGA TJOS-68, integration of nt rv GGTAACTCACAT USER sequence [238]

38 Chapter 2

Table 4: Oligonucleotides for homologous recombination Gene Enzyme Repair Oligonucleotide Name CGTTAAATACTCTGGTGTCTGTCACACTGACTTGCACGCTTGGC ADH1 YOL086C ATAAGACTGGCCATTGCCAGTTAAGCTACCATTAGTCGGTGGTC AC ATATCAAGCTACAAAAAGCATACAATCAACTATCAACTATTAAC ADH2 YMR303C TATATCGTAATACACAGCGGATCTCTTATGTCTTTACGATTTATA GTTTTCATTATCAAGCATGCCTATATTAGTA TCAACGTTAAATATTCTGGTGTATGTCACACCGATTTACATGCTT ADH3 YMR083W AAACGGCGATTGGCCATTACCTGTTAAACTACCATTAGTAGGTG G TGCAAAACAACCATCAACAACAAGTTTACATTTGCAACAACTA ADH4 YGL256W ATAGTCAAATAAGAAAAAAAAATCGAACGAACTCATAAACGTC AATTATGCGTGTGCCTTATTTATTTAGTTGTGCG CGTTAAATATTCTGGTGTTTGTCATAGTGACTTGCACGCGTAAA ADH5 YBR145W CGGTGATTGGCCATTTCAATTGAAATTTCCATTAATCGGTGGTC AC CATTGTGCAGCTGGTCATTGGGGCAATATGAAGATGCCGCTAGT ADH6 YMR318C CGTTAACATGAAATCGTTGGTAAAGTTGTCAAGCTAGGGCCCAA GT CGTTGATATAACGTGTACGATTTTCAAACAAACAGATAGCAGTA GRE2 YOL151W TCACACGCCCGTAAATACTTTAAATGAAAATAGATAATATTTAT ATATATTAACGTTATTACAATTATTTTTTATC CTGTATGCCACACTGATGCGTACACTTTATCAGGCTCTGATCCA SFA1 YDL168W GATAACTTTTCCCTTGCGTTCTGGGCCACGAAGGAGCCGGTATC G

Commercial kits DNA fragments were purified using the GenepHlow Gel/PCR Kit from Geneaid (New Taipei City, Taiwan). Plasmid purification was carried out using the QiaPrep Spin Miniprep Kit from Qiagen (Hilden, Germany). Plasmids were extracted using the QiaPrep Miniprep Kit (Qiagen, Hilden, Germany).

Creating multiple knockout mutants E. coli was used for plasmid amplification. The cells were made electrocompetent following the protocol of the Barrick Lab [239]. For electroporation cells were thawed on ice. 1 μL of the circular vector (~100 ng) was added to the cells and the mixture was added to a precooled electroporation cuvette. The electroporation pulse (2.5 kV, 200 Ω, 25 μF) was applied and 1 mL of LB medium was added quickly. The cells were regenerated at 37 °C for 1 h. The gRNA was integrated into the vector pCFB2312 [238] using the USER technology (New England BioLabs GmbH, Frankfurt am Main, Germany). To generate multiple gene knockout, the CRISPR/Cas9 system was used. The Cas9 containing plasmid pCFB2312 [240] was introduced to S. cerevisiae CEN.PK 113-17A via heat shock transformation [241]. In the next step, a plasmid, which was constructed using the USER technology (New England BioLabs GmbH, Frankfurt am Main, Germany) [242-244] and contained 3 gRNAs and 6 μg of 90 or 120 39 Results bp ds oligonucleotides (see Table 4), was transformed into chemically competent S. cerevisiae cells [241]. The oligonucleotides were used as template for homologous recombination. The 90 bp oligonucleotides were used for the first successful disruptions and lead to the introduction of a stop codon into the gene sequences of ADH1,3,5,6 and SFA1. The 120 bp oligonucleotides were used for complete deletion of ADH2, ADH4 and GRE2. The deletions were confirmed using PCR (for the complete deletions) and subsequently by sequencing (see Table 3).

Biotransformation assay The biotransformation of different aldehydes and alcohols was tested using the knockout mutants that contained only one of the seven known Adhp encoding genes. The mutants were cultivated in shake flasks using YEP medium and fed again with a 50 % (w/w) glucose solution, 14 h prior to harvest. Equal amounts of biomass of 0.04 g CDW were harvested from the culture by centrifugation of appropriate volumes of the culture at a timepoint when excess glucose was still present in the medium. The pellet was washed twice with 0.9% NaCl. The cells were disrupted by incubating the cells for 1 h with Zymolyase 20T (5 mg in 500 μL, WAK-Chemie Medical GmbH, Steinbach (Taunus), Germany) at 37 °C in lysis buffer (1 M sorbit, 10 mM DTT, 10 μL protease inhibitor (Protease Inhibitor Cocktail, Merck, Darmstadt, Germany) and subsequent treatment with 0.5 mm glass beads in the VXR basic Vibrax (IKA Wilmington, USA) at 4 °C. After cell disruption, the suspension was filtered through a 4 μm filter and the filtrate was kept on ice until further use. For the enzymatic assay, 10 μL of the extracted proteins were mixed with 170 μL potassium phosphate buffer (pH 8 for Adh3p and Adh4p, pH 7 for all other enzymes) and 10 μL of 5 mM NAD+/NADP+ for alcohols or NADH/NADPH for aldehydes. The assays were performed in 96-well microtiter plates and the consumption or production of the redox cofactors monitored by reading the absorbance change at 340 nm using the well plate reader Synergy MX (BioTek, Friedrichshall, Germany). The final volume of the assay added up to 200 μL in each well. The assay was started when the absorbance stabilized by adding 10 μL of 1 M substrate and stopped when the absorbance stabilized. All assay substrates and the redox cofactors were purchased from Carl Roth (Karlsruhe, Germany), Merck (Kenilworth, NJ, USA), VWR (Radnor, PA, USA) or Alpha Aesar (Haverhill, MA, USA) and were of < 97% purity (for biochemistry).

Data evaluation The absorbance data was corrected with the data of the control experiment, which included everything but the substrate. The control was included on each 96-well plate and for each time point. The absorbance values obtained after the addition of substrate were subtracted from the last value recorded prior to the addition of the substrate. The resulting data was evaluated for values above zero indicating a change in NAD+/NADP+ concentrations higher than that of the control. For values close to zero, the plot of raw absorbance data over time was visually inspected to check if the slope of the assay data was larger than that of the control.

40 Chapter 2

Results and Discussion Mathematical models help to structure the knowledge of an organism’s metabolism. Models can be analyzed to estimate its biochemical capabilities or to predict genetic engineering targets to improve production of a target chemical thereby rationalizing and enhancing metabolic engineering of productions strains [245]. Many volatile metabolites are of industrial interest as they can be used as fragrances or flavor compounds in the cosmetic and food industry. Interestingly, although yeast is known to produce a variety of volatile, aromatic compounds, the current yeast consensus metabolic model, yeast 7.11, is almost completely lacking those volatile metabolites. Here, we are closing this gap by appending the model biochemical pathways for the synthesis of 92 yeast volatile metabolites that were extracted from the scientific literature (Table 5).

Table 5: Volatile organic compounds emitted from S. cerevisiae fermentations. Compounds that are included in the latest yeast genome scale metabolic reconstruction (Yeast 7.11) and in the Saccharomyces Genome Database SGD [185] are categorized. (Table continues over multiple pages.) included in Compound PubChem Compound Yeast Reference # class ID SGD 7.11 (2-phenylcyclopropyl) alcohols 317540 no no [246] methanol 1,2-benzene dicarboxylic acids 1017 no no [66, 247] acid 1,3-butanediol alcohols 6440 no no [246] 1-butanol alcohols 263 no no [246] 1-heptanol alcohols 8129 no no [246] 1-hexanol alcohols 8103 no no [248-251] 1-propanol alcohols 1031 no no [248-251] 2,3-butanediol alcohols 262 yes yes [250, 251] 2,5-dimethylpyrazine * pyrazines 31252 no no [66, 247] 2-ethyl-1-hexanol alcohols 7720 no no [66, 247, 248] 2-furfuraldehyde aldehydes 7362 no no [249] 2-hexanol alcohols 12297 no no [246] 2-methyl-2-butanol * alcohols 6405 no no [248-250] 2-methylbutanal * aldehydes 7284 yes no [66, 247] 2-methylbutanoic acid acids 8314 no no [66, 247] 2-methylbutanol alcohols 8723 yes yes [248-250] 2-pentanone ketones 7895 no no [66, 247] [66, 247, 249- 2-phenylethanol * benzenoids 6054 yes yes 252] 2-phenylethyl acetate esters 7654 no no [249, 252] 2-propane alkanes 6334 no no [248] 2-propanol alcohols 3776 no no [66, 247, 248] 2-xylene benzenoids 7237 no no [66, 247] 3-methylbutanal * aldehydes 11552 yes yes [248] [66, 247, 249, 3-methylbutanoic acid * acids 10430 no no 250] 41 Results

3-methylheptyl acetate esters 537686 no no [248] 5-methyl-2-furfuraldehyde aldehydes 12097 no no [249] acetaldehyde aldehydes 177 yes yes [248-250, 253] acetaldehyde diethylacetal ethers 7765 no no [249] acetic acid acids 176 yes yes [66, 247, 251] 2-propenyl acetate esters 11584 no no [248] ethenyl acetate esters 7904 no no [66, 247] acetoin ketones 179 yes yes [249] acetone ketones 180 no no [66, 247] acetophenone ketones 7410 no no [249] benzaldehyde aldehydes 240 no no [249] benzyl acetate alcohols 8455 no no [249] benzyl alcohol * alcohols 244 no yes [249] butanal aldehydes 261 no no [248] butanone * ketones 6569 no no [66, 247] butyric acid carboxylic acids 264 no no [250, 252] cis-3-hexen-1-ol alcohols 5281167 no no [249, 250] decanoic acid carboxylic acids 2969 no no [249-252] diacetyl * ketones 650 no yes [248, 249] diethyl succinate esters 31249 no no [249-251] dimethyl disulfide * sulfides 12232 no no [66, 247] dodecanoic acid * carboxylic acids 3893 no no [250, 252] ethanol alcohols 702 yes yes [66, 247] ethyl 2-methylbutyrate esters 24020 no no [249] [66, 247-250, ethyl acetate esters 8857 yes no 253] ethyl benzoate esters 7165 no no [249] ethyl butyrate esters 7762 no no [249, 251, 252] ethyl caproate esters 31265 no no [248-252] ethyl caprylate esters 7799 no no [248-252] ethyl decanoate esters 8048 no no [249-252] ethyl furoate esters 11980 no no [249] ethyl heptanoate esters 7797 no no [249] ethyl isobutyrate esters 7342 no no [248, 249] ethyl isovalerate esters 7945 no no [248, 249] ethyl lactate esters 7344 no no [249, 250] ethyl phenylacetate esters 7590 no no [249] ethyl propanoate esters 7749 no no [248, 249] ethyl pyruvate esters 12041 no no [250] ethyl valerate esters 10882 no no [249] ethyl-2-hydroxy propionate esters 545098 no no [251] furfuryl alcohol * alcohols 7360 no no [249] guaiacol * alcohols 460 no yes [249] heptanal * aldehydes 8130 no no [248] heptanoic acid carboxylic acids 8094 no no [249] hexanal aldehydes 6184 no no [249] hexanoic acid * carboxylic acids 8892 no no [249-252] hexyl acetate esters 8908 no no [249, 250, 252] isoamyl acetate esters 31276 yes no [248] isoamyl alcohol alcohols 31260 yes yes [248-252] isobutanal * aldehydes 6561 no yes [248]

42 Chapter 2

[66, 247-251, isobutanol alcohols 6560 yes yes 253] isobutyl acetate esters 8038 yes no [248, 249] limonene terpenes 22311 no no [66, 247] linalyl propionate esters 61098 no no [251] methanol * alcohols 887 no no [249, 250] methyl acetate esters 76214 no no [249] methylpropanoic acid * acids 6590 no yes [66, 247, 250] monoethyl succinate esters 70610 no no [251] n-butyl acetate esters 31272 no no [248, 249] nonanal aldehydes 31289 no no [248] nonanoic acid carboxylic acids 8158 no no [249] n-propyl acetate esters 7997 no no [248, 249] octanoic acid * carboxylic acids 379 no no [249-252] pentanal aldehydes 8063 no no [248] propionic acid carboxylic acids 1032 no no [250] pyrazine pyrazines 9261 no no [66, 247] undecane alkanes 14257 no no [66, 247] α-terpineol terpenes 17100 no no [249] β-phenylethyl formate esters 7711 no no [248] * These compounds were also detected in yeast extract [254]. In reference [247], yeast was cultivated in malt extract and tryptone soya, in reference [248] yeast extract, plus bactopeptone and glucose was used. All other references reported volatiles from wine fermentations.

To find the pathways that connect each of the 92 metabolites to the reactions already existing in the consensus model, the computational tools FMM, PathPred and MetaRoute were used. In addition, a manual search for metabolic pathways in the literature and the manual addition of possible reactions was performed (Table 6) [207-209]. Apart from seven metabolites (acetaldehyde diethyl acetal, acetic acid 2-propenyl ester, acetic acid ethenyl ester, undecane, (2-phenylcyclopropyl)methanol, 2-ethyl-1-hexanol and 2-xylene), all metabolites could be connected to the metabolic pathways using these tools. For the pathway prediction of two of the missing seven metabolites, 2-ethyl-hexanol and 2-xylene, BNICE [234] was used. BNICE predicts possible pathways using generic enzyme reaction rules and here was confined to the use of compounds present in the KEGG database [231, 232, 255]. To make sure that S. cerevisiae is capable of catalyzing the added reactions, gene sequences coding for enzymes known to catalyze those reactions were blasted against the yeast genome of S. cerevisiae S288C. Flux Balance Analysis (FBA) and thermodynamics-based flux analysis (TFBA) [219, 235, 256] were performed on 36 of the newly introduced pathways to further verify stoichiometric and thermodynamic feasibility of the biosynthetic pathways of the volatile metabolites. All 36 tested pathways yielded feasible reactions for both FBA and TFBA. Using TFBA it was possible to correct pathways that would have not been thermodynamically feasible. An example for this is the production of methanol. Methanol was first proposed to be produced during the reduction of formaldehyde, this reaction however, turned out not to be thermodynamically feasible. Instead the decarboxylation of glycolate was added to the model. Stoichiometric and thermodynamic feasibility was confirmed by FBA and TFBA.

43 Results

Table 6: Volatile metabolites added to the metabolic yeast model. The column reference indicates, whether the pathway was taken from KEGG, a scientific publication, that the pathway was inferred by pathway prediction tools (blank space), or predicted by established yeast enzymes with reactions similar to those needed to establish the pathway (also blank space). (Table continues over multiple pages.) PubChe Pathway based Substance KEGG Literature m on 1,2-benzenedicarboxylic 1017 C01606 acid another organism KEGG 1,3-butanediol 6440 C20335 another organism KEGG 1-butanol 263 C06142 literature data [257] pathway 1-heptanol 8129 prediction pathway 1-hexanol 8103 prediction 1-propanol 1031 C05979 literature data [257] 2-furfuraldehyde 7362 C14279 another organism KEGG 2-methylbutanoic acid 8314 C18319 literature data [258] 2-pentanone 7895 C01949 another organism KEGG 2-phenylethyl acetate 7654 C12303 literature data [259, 260] 2-propane 6334 C20783 another organism KEGG pathway 2-propanol 3776 C01845 prediction 2-xylene 7237 C07212 another organism KEGG 3-methylbutanoic acid 10430 C08262 literature data [258] 3-methylheptyl acetate 537686 literature data [259, 260] 5-methyl-2- 12097 C11115 furfuraldehyde another organism KEGG pathway acetone 180 C00207 prediction acetophenone 7410 C07113 another organism KEGG pathway benzaldehyde 240 C00261 prediction benzyl acetate 8785 C15513 literature data [259, 260] pathway benzyl alcohol 244 C00556 prediction butanal 261 C01412 literature data [257] butanone 6569 C02845 another organism KEGG butyric acid 264 C00246 literature data [257] pathway cis-3-hexen-1-ol 5281167 C08492 prediction pathway diacetyl 650 C00741 prediction pathway diethyl succinate Knoll 1994 31249 prediction dimethyl disulfide 12232 C08371 literature data [261] C02679 pathway dodecanoic acid 3893 prediction pathway ethyl 2-methylbutyrate Knoll 1994 7945 prediction 44 Chapter 2

pathway ethyl benzoate Knoll 1994 7165 C01839 prediction pathway ethyl butyrate Knoll 1994 7762 prediction pathway ethyl caproate Knoll 1994 31265 prediction pathway ethyl caprylate Knoll 1994 7799 C12292 prediction pathway ethyl decanoate Knoll 1994 7799 prediction pathway ethyl furoate Knoll 1994 11980 prediction pathway ethyl heptanoate Knoll 1994 7797 prediction pathway ethyl isobutyrate Knoll 1994 7342 prediction pathway ethyl isovalerate Knoll 1994 7945 C12290 prediction pathway ethyl lactate Knoll 1994 7344 prediction pathway ethyl phenylacetate Knoll 1994 7590 prediction pathway ethyl propanoate Knoll 1994 7749 prediction pathway ethyl pyruvate Knoll 1994 12041 prediction pathway ethyl valerate Knoll 1994 10882 prediction ethyl-2-hydroxy pathway propionate 45258 C01013 prediction furfuryl alcohol 7360 C20441 another organism KEGG guaiacol 460 C01502 another organism KEGG pathway heptanal 8130 C14390 prediction pathway heptanoic acid 8094 C17714 prediction pathway hexanal 6184 prediction pathway hexanoic acid 8892 C01585 prediction hexyl acetate 8908 literature data [259, 260] Isobutanal 6561 literature data [258] limonene 22311 C00521 another organism KEGG pathway linalyl propionate 61098 prediction pathway methanol 887 C00132 prediction methyl acetate 6584 literature data [259, 260] methylpropanoic acid 6590 C02632 [258] (isobutyrate) literature data

45 Results

pathway monoethyl succinate 70610 [262] prediction n-butyl acetate 31272 C12304 literature data [259, 260] n-propyl acetate 7997 C14928 literature data [259, 260] propionic acid 1032 C00163 literature data [257] α-terpineol 17100 C16772 another organism KEGG pathway β-phenylethyl formate 7711 prediction

While the positive BLAST and FBA results increased reliability of the introduced pathways, confirmation requires experimental verification. Here, we aimed for biochemical confirmation of those new reactions that are catalyzed by alcohol dehydrogenases but which could not be assigned to specific genes, yet. The Yeast 7.11 model already contained 8 genes encoding alcohol dehydrogenases and 7 Adhp catalyzed reactions. This set of reactions was extended to 17 in this study. Because the ADHs are well investigated, the probability of unknown isoenzymes that need to be knocked out additionally was extremely low. Another reason for choosing the ADHs was their importance, as fermentative growth on hexoses is not possible for yeast lacking Adhs. To do both, verify new reactions and to assign enzymes and genes, we generated S. cerevisiae mutants deficient in all but one of the known Adhp encoding genes. As we used S. cerevisiae CEN.PK113-17a in this study, which in contrast to S. cerevisiae S288c does not possess ADH7, in total 9 mutants were created (Table 2). The nomenclature of these mutants is s, for single, plus the name of the remaining Adhp gene, e.g., sADH1 refers to the mutant possessing only the ADH1 gene; the S. cerevisiae strain void of all 8 Adhp enzymes was named sKO8. The wild type strain and the sKO8 strain were tested twice for their biocatalytic activity. Three alcohols, seven aldehydes, and 3 ketones were tested (Table 7) and depending on the cofactor specificity either NAD(H) or NADP(H) were used in the assay. The wild type S. cerevisiae strain showed activity for all substrates but isovaleraldehyde in the assays complemented with NAD(+/H). Like the wild type, sADH1 also showed a broad substrate specificity and converted all tested substrates except of acetone. However, Adh1p is known mostly for the conversion of acetaldehyde to ethanol [263-265]. Interestingly, sADH1 showed activity in the assay with isovaleraldehyde, while no activity could be detected for the wildtype, although both strains were cultivated under identical conditions and harvested during exponential growth in the presence of glucose excess. sADH2 only showed activity with half of the here tested substrates (Table 7). In accordance with literature, Adh2p oxidized acetaldehyde and propanol [263]. Another mutant with an interesting substrate spectrum was sADH3 which accepted acetaldehyde, 2,3-butanedione, 2-butanone, propanol, and formaldehyde. The discrepancy to Adh1p, Adh2p and Adh5p is surprising as we expected a similar substrate spectrum for these enzymes given Adh3p shares 80% of amino acids with Adh2p and Adh1p, a paralog of Adh5p. The role of these enzymes is reported to be similar, while a big discrepancy in the biocatalytic activity has been shown here. Adh5p is not well investigated and no substrate spectrum has been recorded so far. From the here tested substances only acetaldehyde has been tested before and contrary to the investigation here, it was converted by a mutant lacking Adh1p-Adh4p, however all other enzymes with Adh function were still active. [263, 266-268]. Dickinson et al. tested in 2002 the possibility of 46 Chapter 2 knockout mutants to produce isoamyl alcohol and 2-phenylethanol, however they failed to create knockout mutants carrying only 1 single gene with ADH activity [258]. However, these findings are not in any way contradicting the here created results, since mutants with more than one ADH gene still active were tested. Of the 13 tested compounds, sADH4, solely showed activity with 1,3-butanediol. In the literature ADH4 is described to possess formaldehyde dehydrogenase activity, as well as being able to convert butanol [263, 265], however this could not be confirmed here. sSFA1 and sADH5 showed no activity on either of the tested substrates, while a slight conversion of 2-hexanone was found in both strains, a slightly stronger activity was found in the sKO8 mutant. The latter observation indicates that there were still other enzymes active, which could convert 2-hexanone. In contrast to literature [265, 269], sADH5 was not able to catalyze the reaction from acetaldehyde to ethanol and sSFA1 was not able to convert formaldehyde. However, the published data was collected using knockout strains which still contained other enzymes with ADHAdh6p and Gre2p use NADP(H) instead of NAD(H). Therefore, experiments with NADP(H) were performed with sADH6 and sGRE2 as well as the wild type and the sKO8 mutant (Table 7). With this cofactor, the wild type was able to catalyze the conversion of every tested substrate. Surprisingly, the sADH6 mutant showed a strong conversion of 2,3-butanedione, while in literature it is only reported that Adh6p is involved in fusel alcohol production [270, 271]. This could be explained by upregulation of another enzyme (possibly Bdh1p) that converts 2,3-butanedione since a slightly lower activity is also found in sGRE2 and even in the sKO8. However, the activity with sADH6 and sGRE2 was higher than in sKO8, indicating that these enzymes contribute to the conversion of 2,3-butanedione but are not concentrated high enough in the wild type to make an impact on activity. The sADH6 mutant was also able to convert hexanal, 2-butanone, 1-heptanal, acetone and isobutyraldehyde. The sGRE2 mutant showed lower conversion rates of hexanal and 2-butanone in comparison to the sADH6 mutant and even lower rates than seen with the sKO8 mutant. Therefore, it can be assumed that these two metabolites were converted by other enzymes and not by Gre2p. sGRE2 was also able to convert isovaleraldehyde in accordance with Hauser et al. [272]. Here it was demonstrated that sADH1,2,3,4,6 and sGRE2 showed activity for some of the tested substances. However, the substances are not protein specific but mostly are converted by multiple enzymes. Adh5p and Sfa1p do not seem to be involved in the here tested fusel alcohol/aldehyde metabolism at all. One explanation for this discrepancy could be that in literature other S. cerevisiae strains (S288C, YPH499) have been used whose enzymatic make- up might differ from CEN.PK 113-7D. Another possible explanation is moonlighting or promiscuity of Adhps, the ability of enzymes to harbor separate functions or convert diverse substrates [273, 274]. A moonlighting function could be the inhibition or activation of another enzyme and therefore have a regulatory effect (e.g., S. cerevisiae’s arginase inhibits the ornithine transcarbamylase, when arginine and ornithine are both present [275, 276]). Here it is possible that one of the Adhps is only expressed or active if one of the other ADH genes is being expressed. Since all other ADH genes are knocked out this could explain the results shown for sADH5 and sGRE2. Since Adh5p is an Adh1p paralog it would have been expected to have similar functions to the ones of Adh1p [277]. It is important to notice that all here discussed results have been evaluated for yeast in the exponential growth phase in a shake flask. Therefore, the cells were oxygen limited with excess

47 Results glucose. It is possible that yeasts in a stationary growth phase, or yeasts that have excess oxygen and that are glucose limited, show different results. However, since the here tested conditions lead to yeast fermentation, it is unlikely that alcohol dehydrogenases are expressed higher under aerobic conditions. It was surprising that the main activity not only in the conversion of acetaldehyde to ethanol, but also in the conversion of most of the tested substrates could be detected in S. cerevisiae´s main ADH. Not only Adh1p, but most of the tested enzymes (all but Adh4p, Adh5p and Sfa1p) showed catalytic activity with a number of the tested substrates. Single enzymes are not responsible for the production of specific substances. The conversion of the 13 investigated substrates by alcohol dehydrogenases or enzymes with alcohol dehydrogenase activity, respectively, was confirmed.

48 Chapter 2

------+/- +/- +/- +/- +/- +/- +/- +/- sKO8

2 . Green ------+ + + + + + GRE s NADP(H) NADP(H) 6 ------+ + + + + ADH s + + + + + + + + + + + + + WT ------+/- +/- sKO8 sKO8 1 to convert alcohols, ketones, and aldehydes in ------SFA s 5 ------ADH s 4 ------+ ADH s NAD(H) NAD(H) 3 ------+ + + + + ADH s 2 - - - - - + + + + + + +/- +/- +/- ADH mutants. The strains were tested formutants. The strains their ability s data, red indicates contrariety.

ADH 1 - + + + + + + + + + + + + ADH s

- - + + + + + + + + + + + WT

hexanal acetone acetone cofactor cofactor propanol propanol substrate 1-heptanal 1-heptanal 2-butanone 2-butanone 2-hexanone acetaldehyde acetaldehyde formaldehyde formaldehyde 1,3-butanediol 1,3-butanediol 2-methylbutanal 2-methylbutanal 2,3-butanedione isovaleraldehyde isovaleraldehyde isobutyraldehyde isobutyraldehyde Table 7: Enzymatic capacity of s capacity Table 7: Enzymatic assays. The mutantsenzymatic showed either clear (+), slight conversion conversion (+/-) no conversion or of the substrate (-) indicates accordance to literature

49 Results

Computational models enable us to structure the vast field of knowledge related to an organism, the enhancement of a model helps to close gaps in this accessible knowledge. All here made additions to the model are non-essential for growth of the organism. The majority of amendments made here are based on computational predictions and were only substantiated by BLAST searches and (T)FBA analyses. To increase the reliability of the here expanded model, experimental verification of the new pathways is required. To this end, analyses of pathway intermediates by, e.g., LC or GC combined with mass spectrometry [278, 279], besides enzymatic assays of know-out mutants, as done here, can be applied. The here added reactions could be especially interesting, when a yeast shall be enhanced for the production of volatiles that may alter the taste of the food produced. It would be possible to construct yeasts that have a fruity taste by enhancing the production of some esters. For example, the upregulation of genes producing 2-phenylacetate would lead to a rose like odor. These aroma properties could even be transferred from yeast to the final product, e.g., a bread that smells like apple or banana, or a wine that has a rose-like bouquet. Also, it would be possible to create yeasts that produce industrially relevant products, such as fuels or precursors for fuel production, such as butanol (biofuels) [280] or 2,3-butanediol (bioplastic) [281].

Conclusion and Outlook The consensus yeast metabolic model was lacking the volatile metabolite space nearly completely in earlier versions. Here 225 metabolites were added to cover this part of the yeast metabolism. To connect the metabolites to the rest of the existing network, 219 reactions were added to the model. It represents a structured representation about the current knowledge of volatile metabolite synthesis in yeast. Although stoichiometric and thermodynamic coherent, most of the introduced pathway lack experimental evidence. In further work those pathways are to be substantiated by wet-lab examinations. The resulting, curated model not only represent a knowledgebase of VOC biosynthesis in yeast but can be computationally analyzed to guide rational engineering of, e.g., flavor- overproducing strains or odorless yeast, void of VOC production. It would be also possible to create yeast strains that do not produce a product that is hard or impossible to purify. To monitor the volatile space this thesis concentrated on online measurement methods. In the next chapters a SESI-Orbitrap, as well as an MCC-IMS are illuminated in greater detail.

Yeast volatilome dynamics during metabolic shifts

Contributions: The study was designed by Christoph Halbfeld, Birgitta E. Ebert, Lars M. Blank and Pablo Martinez- Lozano Sinues. The experiments were performed by Christoph Halbfeld and Pablo Martinez-Lozano Sinues. The data was evaluated by Christoph Halbfeld, Henrik Cordes and Birgitta E. Ebert.

Results

Yeast volatilome dynamics during metabolic shifts Summary

S. cerevisiae is an industrially important organism, with applications as different as baking and bioethanol production for which the different growth modi of yeast, aerobic respiration and fermentation are exploited. Notably, while intensely studied, the metabolic transition from aerobic respiration to fermentation is still not fully understood. Also S. cerevisiae’s volatile space, while again important in many applications like wine and beer making, was widely neglected in the past. For example, no study reported the volatilome in dependence of the metabolic state of yeast. Here, we analyzed the off-gas of a continuous glucose-limited cultivation of S. cerevisiae using a SESI- Orbitrap that enabled online measurements in near real-time with high mass accuracy. This resulted in a large amount of data (2,500 signals/sample) that was evaluated using a self-written MATLAB script. The data evaluation revealed 16 signals that changed during the shift from fully respiratory to respiro-fermentative metabolism prior to the detection of ethanol. Both, masses that showed an increase and signals that showed a decrease of signal intensity upon the shift were detected. Two of these masses were tentatively identified as isoprene and 3-methyl-1-butanol. The other volatiles remain yet to be determined and might give valuable insights into the yeast’s volatile metabolism. However, even without the exact knowledge about the volatile metabolites and their biochemical synthesis, it can be concluded that the presented fast and sensitive SESI-Orbitrap off-gas analysis allows to clearly determine the metabolic state (respirative- or fermentative growth) of a microbiological culture in near real time.

55 Chapter 2

Introduction The baker’s yeast Saccharomyces cerevisiae has been in the focus of research and industrial production ever since industrialization has started. It has been used to produce beverages like beer and wine since ancient times [282-286] but has also been in the focus of scientific research and was the first eukaryotic organism for which the complete genome has been sequenced [287, 288]. A specific trait of baker’s yeast is the Crabtree effect, which represents a metabolic shift from respiration to aerobic fermentation when high sugar concentrations are supplied in the growth medium [289]. This shift has significant, economic relevance in baking yeast production since fermentation with high initial sugar concentration or excess glucose feeding mainly leads to the production of ethanol instead of biomass and worsens baking and storage ability of the product (Dr. Quantz, VH-Berlin, personal communication). The Crabtree effect can easily be induced by overfeeding of glucose and has been studied in nearly every aspect. However, the impact of the Crabtree effect on the formation of volatile metabolites has not been investigated so far. While there are plenty of studies covering volatile flavor compounds of yeast products, e.g., bread or wine [9, 290-293], almost no studies exploring volatile metabolite formation during yeast growth in minimal salt medium exist. This trend is changing lately with scientific investigations putting the focus on Saccharomyces’ volatile space [1, 91, 185]. Changing the metabolic state is required for an organism to adapt to altered environmental conditions for example by enzymatic adaptation to new nutrition sources or the production of metabolites that fend off enemies. Here we used the induction of the Crabtree effect, i.e., the transition from respiration to aerobic fermentation, as an example for a metabolic shift. Induction of the Crabtree effect results in an increased flux through glycolysis and reduction of the relative (normalized to the glucose uptake rate) pentose phosphate pathway flux. The major change, however, is a downregulation of the citric acid cycle, the respiratory pathway, and the strong increase in cytosolic pyruvate transformation to ethanol [294]. These changes in the central carbon metabolism are well investigated but not necessarily the only changes that occur in the cell and additional changes of the fluxes through the pathways responsible for the synthesis of amino acids, fatty acid synthesis and degradation as well as terpene biosynthesis involving also volatile metabolites might occur.

Dynamics of (volatile) compounds Until recently it was complicated to measure volatile compounds and nearly impossible to investigate the dynamics of their production or consumption. However, new ionization techniques and improved analytical devices allow the measurement of volatile compounds without sample preparation and in high-throughput enabling to capture fast dynamics [1, 295]. For online measurements of volatile dynamics, the analytical device must have a high mass resolution, short measurement times and be suited for direct sampling. These attributes are combined in secondary electrospray ionization Orbitrap mass spectrometry (SESI-Orbitrap-MS). The SESI ion source allows a continuous ionization of volatile compounds which are then directly injected into the Orbitrap system [296, 297]. The ionization is fast and efficient also for low abundant analytes. The principle can be briefly described as follows: First, electrospray ionization (ESI) is performed with a dilute aqueous formic acid solution (0.1%) in the absence of analyte ions. The formic acid solution is lead through a capillary in which an electric field between the tip of the capillary and a counter

56 Results electrode is applied. The ions in the formic acid solution are pulled towards the negative charge in the electric field and thus form a cone, where all ions are collected at the tip of the cone. When the charge is high enough at the tip of the cone, droplets are formed that dissociate from the rest of the water. The droplets are multiply charged and become smaller over time since the water evaporates until the electrostatic repulsion in the drops becomes larger than the surface tension. The droplets thus undergo Coulomb fission and smaller droplets are formed. This continues until one or no charge is left per droplet [298]. The sample is introduced into the charged droplets. Through collisions of the uncharged volatiles with the charged droplets, charges are transferred and the volatile molecules are ionized. Since this happens instantaneously, a constant flow of ionized molecules can directly be transferred to the Orbitrap [299]. The ionized gas is bundled by an electric field (focusing electrode) and led through the impaction slit. Here the ionized gas collides with a counter flow of clean gas and the charged ions are dragged further along the electric field while uncharged particles are flowing out of the system (Figure 5). Fast and continuous ionization makes SESI-Orbitrap-MS especially interesting for fermentation off-gas analysis as samples can be withdrawn and measured without any sample preparation.

Figure 5: Schematic representation of a secondary electrospray ionization source. The here indicated DMA is interchangeable with an Orbitrap. DMA= differential mobility analyzer. Reprinted with permission from [300]. Copyright 2012 American Chemical Society.

The Orbitrap system analyzes the samples with a frequency of up to 20 Hz. This enables the detection of real time changes and to resolve even very fast dynamics. Because of the high resolving power of the device (100,000 to 1,000,000 depending on the device and m/z) (see Figure 6), no chromatographic pre-separation is necessary [301]. The high mass resolution gives the possibility to distinguish isobaric ions and the fast measurements allow to record a complete mass spectrum in each measurement. Depending on the concentration and stability of the metabolite, single interesting masses can be further fragmented if the machine is equipped with a collision chamber and thus be identified based on their fragment spectrum.

57 Chapter 2

Figure 6: Influence of mass resolution on peak separation. Taken from [133].

SESI-MS has been used to investigate the dynamics of volatile compounds excreted by Begonia semperflorens [297]. These measurements allowed to elucidate diurnal changes of 400 volatile metabolites released by these plants. As consequence of mechanical damage 1200 volatile metabolites were detected that showed changes compared to non-damaged plants. Some of the changes could not have been detected with a lower sampling frequency, hence the direct ionization and high sampling frequency were needed [297]. Similar studies have been conducted for the human exhalome by Sinues et al. who reported that 40% of the 111 detected signals showed significant circadian modulation [302, 303]. In yet another study Gaugg et al. showed that smokers and non-smokers could be distinguished based on breath analysis [133]. In this study, we focused on off-gas analysis of fermentations of baker’s yeast. Here we set out to exploit the SESI-Orbitrap-MS technology to investigate rapid changes in the volatilome of a continuous culture after targeted perturbations of the metabolic steady-state and induction of the Crabtree effect. We show that it is possible to distinguish metabolic states using the SESI-Orbitrap and to monitor the transition from one metabolic state to another. Identification of volatile metabolites correlating with metabolic shifts and their biosynthetic pathways will shed new light on the regulatory mechanisms associated with the Crabtree effect.

Material and Methods Yeast strains and growth conditions In this study, two yeast strains were used: S. cerevisiae CEN.PK 113-7D (Euroscarf, Oberursel, Germany), which was recommended for yeast physiology and S. cerevisiae VH 2200 (Versuchsanstalt der Hefeindustrie e.V., Berlin, Germany), used for industrial baking yeast production. The strains were grown in Verduyn minimal salt medium [237] containing 3 g L−1 −1 −1 −1 glucose, 3 g L KH2PO4, 0.5 g L MgSO4·7 H2O, 20.4 g L potassium hydrogen phthalate as well as 1 mL of vitamin solution and 1 mL of trace elements. The vitamin solution contained 0.05 g L−1 D-biotin, 1 g L−1 calcium D-pantothenate, 1 g L-1 nicotinic acid, 25 g L−1 myo-inositol, 1 g L−1 thiamine hydrochloride, 1 g L−1 pyridoxine hydrochloride and 0.2 g L−1 p-aminobenzoic acid. The −1 −1 −1 trace element solution consisted of 15 g L EDTA, 4.5 g L ZnSO4·7 H2O, 1 g L MnCl2·4 H2O, −1 −1 −1 −1 0.3 g L CoCl2·7 H2O, 0.3 g L CuSO4·5 H2O, 0.4 g L NaMoO4·2 H2O, 4.5 g L CaCl2·2 H2O, −1 −1 −1 3 g L FeSO4·7 H2O, 1 g L H3BO3 and 0.1 g L KI. The medium was buffered by addition of 100 58 Results mM potassium hydrogen phthalate and the pH was adjusted to 5 using KOH prior to autoclavation, while the vitamin solution and the trace elements were sterile filtered. The glucose was separately autoclaved. The pre-culture was incubated at room temperature and stirred with a stirrer bar at low rpm. The continuous cultivation was performed with 4 parallel self-made 10 mL bioreactors (Seele Glasapparatebau & Laborservice, Swisttal-Straßfeld, Germany). Mixing of each reactor was achieved by a stirrer bar (10x3 mm, Karl Roth, Karlsruhe Germany) powered by a second magnet below the glass vessel, which itself was driven by a motor. The fermentations were run at 30 °C; the temperature was controlled by a water jacket connected to a Thermomix 1420 (B. Braun, Bethlehem, PA, USA). The pH was not actively controlled but kept at pH 5 due to buffer and the incoming fresh medium. The dilution rate and thus the specific growth rate, was set to 0.15 h-1 (below the critical dilution rate of both strains in Verduyn minimal medium). For the medium feed an Ismatec Reglo ICC (Cole- Parmer GmbH, Wertheim, Germany) and Ismatec Pharmed tubing with 0.25 mm ID (Cole-Parmer GmbH, Wertheim, Germany) was used. The fermentation volume was kept constant at 10 mL by adjusting the efflux tubing to a predefined height. For the efflux an Ismatec Reglo ICC pump (Cole- Parmer GmbH, Wertheim, Germany) and Ismatec Pharmed tubing with 1.02 mm ID (Cole-Parmer GmbH, Wertheim, Germany) were used. The reactor was aerated with 500 mL/min (50 vvm) of compressed air via a Sho-Rate Series rotameter (Books Instruments, PA, USA). The air was filtered and humidified prior to its introduction into the reactor (Figure 7). Safeflow connectors (B. Braun, Melsungen Germany) were used for sampling. The gas outlet was coupled to a SESI-Orbitrap via a transfer line that was heated to 130 °C. The off-gas of the cultivation was continuously monitored by the SESI-Orbitrap. The measurements were started when the fully respiratory, continuous culture had reached a metabolic steady state. To capture potentially fast dynamics of volatile metabolites, the SESI-Orbitrap measurement frequency was set to one full spectrum every 5 seconds. The S. cerevisiae culture was perturbed by injecting a pulse of a sterile glucose solution (50 mg ؙ 27.7 mM) and measurements were taken until the ethanol signal reached its maximum. After the pulse, the system was left undisturbed for 48 hours to re-establish a metabolic steady state.

59 Chapter 2

Figure 7: Schematic view (A) and picture (b) of the mini-bioreactor setup connected to the SESI- Orbitrap-MS (Bioreactor constructed by Eik Czarnotta and Suresh Sudarsan, RWTH Aachen University).

Online SESI-Orbitrap MS measurements The off-gas of the mini bioreactor was transferred via a heated (130 °C) transfer tube (short Rotilabo silicon tubing (Carl Roth GmbH, Karlsruhe, Germany) directly into the SESI ion source without prior treatment. The SESI ion source was developed at the ETH Zurich in the group of Pablo Martinez-

60 Results

Lozano Sinues and is commercially available (SEADM, Boecillo, Spain). The Thermo Finnigan LTQ Orbitrap (Thermo Fisher Scientific, Waltham, MA, USA) was used as detector. The SESI and Orbitrap settings were similar to those used by Rioseras et al. 2017 [295]. The SESI used a 0.1% formic acid solution that was infused at ~100 nL/min through a 20 μm ID silica capillary. The voltage of the electrospray was set to 5.4 kV. The mass range was 50-500 m/z, with an acquisition rate of 0.20 spectra/s. The pre-set resolution of the device was 30,000, while the mass accuracies were typically within 2 ppm. After connecting the bioreactor off-gas to the SESI-Orbitrap, a quick decline of the total ion current (TIC) could be observed. Measurements were only started when the TIC had stabilized.

Data evaluation: Data was acquired in Thermo Scientific’s .raw format and converted to mzXML files using the ProteoWizards msconvert tool [304]. The data was imported into MATLAB R2012b (Mathworks, Natick, MA, USA) and each dataset was interpolated linearly (106 points in the range of 50-500 Da); interpolated profile mass spectra were then centroided (intensity threshold of 50 a.u.) as described by Rioseras et al. [295]. The mass spectral data was evaluated using a self-written MATLAB script (see Appendix). The data was first processed using the medfilt1 (equation 1) and sgolayfilt filter for smoothing. Those processed datasets were used to detect signals whose concentration changed in the period between glucose pulse injection and the increase of the smoothed ethanol signal above a defined threshold. For visual inspection, two clustergrams were created (see Figure 8 A and B). The first one shows all signals which displayed a maximum intensity between the glucose pulse and the appearance of ethanol. The second one shows those signals having a maximum later than the appearance of ethanol but that changed after the injection of the glucose pulse. In addition, the normalized data was plotted against time, one plot per signal, and overlaid with the normalized ethanol signal (see Figure 8 C). ݔ݅ െ ݔ݉݅݊ Equation 1 ܺൌ ݔ݉ܽݔ െ ݔ݉݅݊

The signals left after filtering, were further evaluated manually by scanning the clustergram for changes in the timeframe between the glucose pulse and ethanol formation. In case the masses showed interesting dynamics in more than 5 of the eight experiments they were considered to correlate with the metabolic changes induced by the glucose pulses. One exception was the mass most likely representing isoamyl alcohol (3-methyl-1-butanol), which was included in the list of relevant masses although it only showed a rapid change in 5 of the 8 performed experiments.

Results and Discussion To elucidate potential changes in the volatilome during the metabolic shift from full respiratory growth to aerobic fermentation, S. cerevisiae was cultivated in an aerated glucose-limited chemostat. The dilution rate of 0.15 h-1 was set to be sufficiently below the critical growth rate of >0.2 h-1 of both strains, thereby guaranteeing full respiratory metabolism. The off-gas of the cultivation was monitored to detect volatile metabolites using a SESI-Orbitrap. First, the metabolic steady state was investigated, later the steady state was disturbed by injecting a concentrated glucose solution. The

61 Chapter 2 volatiles that were either consumed or produced in this transient state prior to the production of ethanol were in the focus of this study. Per experiment about 2,500 signals were recorded (see Figure 8 A and B) that show clearly individual trends during the course of the experiment. Due to the large amount of signals, the data had to be software filtered for interesting signals since manual evaluation of the complete datasets would be cumbersome. The data recorded during the steady state clearly differentiated from the data recorded after the glucose solution was injected, therefore the two metabolic states could be clearly distinguished using the SESI-Orbitrap. After the computational evaluation about 200 to 300 signals of each dataset were left that were subsequently evaluated manually by plotting these signals against the ethanol signal. Most of the here found signals were increasing in intensity after the glucose pulse, while only few were decreasing. Therefore, it can be assumed that the flux rates during aerobic fermentation are higher than during respiratory growth. All interesting signals were added to a list and a score, based on the number of experiments, the signal resulted to be interesting, was generated. The most 16 interesting signals were evaluated further (Table 8).

Interpretation of the results Some of the 16 signals probably represent isotopologues (molecules with a different number of neutrons) of the same metabolite. Most of the identified signals were not unambiguously identified during this thesis (Table 8). The response of these volatile molecules to the glucose pulse indicate that the metabolic changes of yeast are manifold and not just restricted to central carbon metabolism. This conclusion is based on the knowledge that most of the metabolites in central carbon metabolism are not volatile, while the here used detection method only reported signals of volatile compounds. The interesting signals can be distinguished into signals that increased and signals that decreased after the glucose pulse was applied to the system (see Figure 8 C-F). Most of the signals increased and are possibly related to the production of metabolites that are required for the fermentative metabolism but not used during fully respiratory metabolism. Some signals rose quickly and disappeared again in a matter of minutes (see Figure 8 E). These signals are potentially linked to intermediates that shortly accumulate during the transient metabolic state, e.g., metabolites involved in the production of amino acids for protein synthesis or intermediates of metabolic pathways activated by the glucose pulse. Because of the very fast dynamics in the seconds range, it can be concluded that the catalyzing enzymes were already present in the cells before the metabolic shift. We exclude that these quickly changing signals are contaminants introduced with the pulsed glucose solution as these signals could not be reproduced by a glucose injection into sterile medium and did not respond immediately with the pulse but with a delay of about 3 minutes. As an artefact of data- smoothing some signals showed changes already prior to the glucose pulse (Figure 8 F). However, signal changes close after the pulse region can reliably be attributed to the metabolic perturbation. For 16 signals, intensity changes were detected prior to an increase of the ethanol peak (Figure 8 C, D, F). Those metabolites might be precursors of ethanol or other metabolites that accumulate because of an imbalance during the metabolic shift or either arise or deplete due to intracellular flux rearrangements. Increasing signals might also be degradation products of storage compounds whose catabolism is activated during the metabolic shift.

62 Results

Table 8: Mass to charge ratio (m/z) of the most interesting signals recorded during the metabolic transient state in a continuous glucose-limited fermentation after setting a glucose pulse. + and # indicate masses whose signal intensity linear correlated with each other and which probably originate from the same metabolite. m/z 69.0701 and 89.0962 have been tentatively identified as 3- methyl-1-butanol and isoprene, respectively. percentage of experiments in which the m/z signal was detected (%) comment 57.0700 87.5 58.0735 87.5 69.0701 100 (+) isoprene 70.0732 100 (+) isoprene (protonated) 71.0857 100 cyclopentane 72.0887 100 73.0649 87.5 ethoxy ethene 74.0680 87.5 89.0962 75 3-methyl-1-butanol 118.0962 75 119.0885 100 127.1117 75 1-octen-3-one 128.1152 75 166.9860 100 (#) 177.0763 100 196.1123 100 (#)

Most signals have not been identified yet, but we speculate that they have not even been observed and be linked to the Crabtree effect so far as detection of these low abundant molecules and monitoring of the fast dynamics is only possible today with highly sensitive and fast analytics such as SESI-Orbitrap-MS.

63 Chapter 2

Figure 8: Data evaluation of SESI Orbitrap analyses of the off-gas of a continuous glucose-limited yeast culture at steady state which was disrupted with a 27 mM glucose pulse. The filtered data is displayed as a clustered heatmap in which the scaled intensity of single masses (m/z value) are plotted over time (A), zoom-in excerpt of the heatmap shown in A (B). Smoothed data of interesting signal traces (blue line) identified in the heatmap are plotted against time together with the signal of ethanol (black line). The left black vertical line indicates the timepoint of the glucose pulse injection, while the right line indicates the timepoint of the first detection of ethanol (C-F).

Two compounds were tentatively identified as isoprene and 3-methyl-1-butanol (isoamyl alcohol). Isoamyl alcohol is produced during the degradation of leucine via the Ehrlich pathway. Possibly during the metabolic transition from aerobic respiration to aerobic fermentation, amino acids are required in large quantities because of an increased reproduction and production of biomass. Therefore, leucine might have exceeded a threshold that triggers leucine degradation resulting in isoamyl alcohol production. However, the definite role of isoamyl alcohol remains unclear.

64 Results

Isoprene is not known to be naturally produced by S. cerevisiae, however precursors of it are native to S. cerevisiae’s metabolism and metabolic engineering has been performed to create isoprene production strains [305]. It would be more likely that the measured volatile is not isoprene but another volatile with the same molecular weight. However, since the volatile space in yeast fermentations has been neglected thus far widely, it is possible that isoprene has not been detected before. The direct precursor of isoprene, dimethylallyl pyrophosphate, is a part of S. cerevisiae’s mevalonate pathway [306]. The possibility of spontaneous conversion in low quantities is given. While there were studies that have investigated the volatile space of yeast in the recent past, they have not detected the presence of isoprene [91, 307]. This could be due to the used indirect measurement methods that included adsorption prior to the measurement, as isoprene might not be adsorbed due to its high volatility. Nevertheless, as mentioned before it is more likely that the tentatively identified isoprene signal is another compound.

Conclusion and Outlook With fast, high-resolution, accurate mass SESI-Orbitrap-MS measurements, we could finely resolve the dynamics of S. cerevisiae’s volatilome. Those analyses showed not only the expected changes in the ethanol concentration upon induction of the Crabtree effect but revealed changes of several other signals that either correlated positively or negatively with ethanol or showed a short peak and then dropped back to the initial value. This is an interesting observation as so far metabolic changes initiated by the Crabtree effect were primarily associated to central carbon metabolism, i.e., upregulation of the pyruvate decarboxylase pathway and downregulation of the TCA cycle and less to peripheral metabolic pathways that produce volatile metabolites [294]. The information content of the SESI-Orbitrap MS data generated in this study has not been exploited to its full extent. Further research on the identity of these compounds, their biosynthetic pathways and regulatory programs is necessary to understand their dynamics and the relation to the Crabtree effect. To achieve a deeper insight in the metabolism during the shift from respiratory to respiro-fermentative metabolism, besides (volatile) metabolite profiling transcriptome and / or proteome analyses are required ideally complemented with the computation of intracellular reaction rates. We propose that the highly time and mass resolved time series can be used to infer enzymatic reaction and enzyme kinetics. For example, the occurrence of linearly correlating traces differing by 18 Da indicates a dehydration or condensation reaction. Also, the identification of the measured masses can be achieved by further analysis, e.g., fragmentation of interesting masses and could further increase the insight into yeast’s metabolism. Moreover, SESI-Orbitrap analysis might be used for 13C metabolic flux analysis and could allow near real-time determination of intracellular fluxes [308]. Through the high mass resolving power, the high measurement frequency and the low detection limit, it was possible to detect changes in the yeast volatile space in near real time. The only limitation to this work is that solely excreted metabolites can be measured using a SESI-Orbitrap. Here it would be interesting to try out an approach, where whole cells are directly measured using a flow injection system that takes samples out of a stirred bioreactor, comparably to what Link et al. reported [309]. This procedure could give even greater insights into the yeasts dynamic metabolome. However, up to now the number of detected metabolites is small compared to what can be measured using the SESI-Orbitrap. The SESI-Orbitrap is an ideal device for research, but it is not too robust due to a 65 Chapter 2 capillary that feeds the SESI and high vacuum that is needed for measurements. A less costly and more robust system is given with the multicapillary column ion-mobility spectrometer as explained in detail in the next chapter.

Multicapillary column-ion mobility spectrometry of volatile metabolites emitted by Saccharomyces cerevisiae

Partially published as: Halbfeld, C.; Ebert, B.; Blank, L. Multi-capillary column-ion mobility spectrometry of volatile metabolites emitted by Saccharomyces cerevisiae. Metabolites 2014, 4, 751-774.

Contributions: Christoph Halbfeld performed the experiments. Christoph Halbfeld and Birgitta E. Ebert analyzed the results. Christoph Halbfeld and Birgitta E. Ebert wrote the manuscript with the help of Lars M. Blank. We thank Marco Fraatz (Justus-Liebig-University Gießen, Germany) for GC-MS measurements and help with data evaluation and Jörg I. Baumbach (Reutlingen University, Germany) for fruitful discussions. We also thank Peter Kaiser (B&S Analytik, Dortmund, Germany) for technical support and Mathis Wolter (RWTH Aachen, Germany) for assistance with MCC-IMS-measurements and implementation of the experimental set-up.

Results

Multicapillary column-ion mobility spectrometry of volatile metabolites emitted by Saccharomyces cerevisiae Summary This chapter deals with the analysis of the headspace of Saccharomyces cerevisiae cultures using multicapillary column-ion mobility spectrometry (MCC-IMS). The high sensitivity and fast data acquisition of the MCC-IMS enabled online analysis of the fermentation off-gas. Of the 19 specific signals, four were identified as ethanol, 2-pentanone, isobutyric acid, and 2,3-hexanedione by MCC- IMS measurements of pure standards and cross validation with thermal desorption–gas chromatography-mass spectrometry measurements. As shown in the previous chapter, monitoring of VOCs produced during microbial fermentations can give valuable insight into the metabolic state of the organism. Here, it is shown that the MCC-IMS technology is suitable for fast, robust, and noninvasive online measurements. The low maintenance requirements of the MCC-IMS could enable the device for easy to use monitoring of industrial fermentations.

71 Chapter 2

Introduction Yeasts are key model organisms in eukaryotic research and play a significant role in many biotechnological processes. The use of yeast for biotechnological processes dates back to 7000–5000 BC, where it has been used for wine fermentation and in food processing [284, 310-312]. Today, yeast strains are used in the chemical, food, and pharmaceutical industry for the synthesis of a broad range of products from bakery products to bioethanol and pharmaceuticals [313-316]. With 4 million tons of yeast biomass produced for baking worldwide in 2009 and an estimated yearly increase of 7% [317], the yeast biomass market is in the top 10 of biotechnology products. The so far most often used yeast cell factory is the baker’s yeast Saccharomyces cerevisiae. This yeast is not only the most widely used yeast in biotechnology, but is additionally the model system for eukaryotic organisms. Accordingly, much research has been conducted with S. cerevisiae and many of the today state-of- the-art analytical techniques and molecular biology methods have been first developed for and with this yeast. The simple and rapid cultivation, genetic accessibility, and industrial importance were and still are drivers to maintain its lead in the yeast community in many disciplines of science. As mentioned above, the volatile metabolites produced by yeast have been mainly neglected in yeast research with the most prominent exceptions of acetaldehyde and ethanol. This can be attributed to the limited knowledge of the biochemistry due to the complex analysis of volatiles. In addition, genetics involved in the formation of volatile metabolites hinder the general incorporation in systems biology studies. By far, most reports of volatile metabolites from yeast originate from researchers interested in high quality wine making [249-252, 291, 318, 319], as the scent of wine impacts its organoleptic properties [291]. The main classes of volatile metabolites observed are alcohols, aldehydes, and esters (Table 5); while depending on grape and yeast strain used, many others can be found. Notably, most studies were carried out on media that have wine-like compositions. Thus, the media are most often complex, with alternative carbon and nitrogen sources present. The observed volatile metabolites can therefore originate from glucose catabolism or are products from biotransformation, i.e., do not originate from sugar (the carbon and energy source), but rather from grape metabolites that were only modified by yeast enzymes. One of the key bottlenecks in VOC research, and one of the reasons why these have not been studied broadly so far, is that sample preparation of gaseous chemicals requires additional care and that the analysis of volatiles is challenging. Most often, VOCs are extracted and enriched using headspace/solid phase microextraction (HS/SPME) methods and analyzed with gas chromatography coupled to advanced mass spectrometers [5, 159]. In the previous chapter, the use of a SESI-Orbitrap was reported, but because of the high invest cost, this device is unlikely to be used in industrial applications. As described in the next chapter, industrial applicability of the analytical system is of interest, therefore an alternative to the SESI-Orbitrap had to be investigated to monitor the yeast’s volatile space. For this purpose, the ion mobility spectrometry that was already explained broadly in the general introduction was used here. The MCC-IMS’s characteristics, high time resolution, low cost, and low maintenance, together with the high sensitivity perfectly suit this analytical technique for online measurements of dilute volatile metabolites in the headspace of microbial fermentations. GC-MS devices are suited for the identification of metabolites, but come with the disadvantage of longer sampling times, although rapid GC techniques exist. Furthermore, GC-MS instruments require special gases such as helium and high vacuum, hence come with relatively high operating costs and technical expenditure. In contrast, MCC-IMS can be operated with nitrogen or air (not necessarily

72 Results synthetic air), at ambient temperature and pressure. In addition, handheld, mobile devices are available underlining that MCC-IMS can be used flexible as they can be robust. The potential of MCC-IMS analyses for fermentation monitoring has already been shown for measurements of mVOCs produced during batch cultivation of Escherichia coli and Pseudomonas aeruginosa in shake flasks [320, 321]. Also, MCC-IMS measurements of yeast fermentation have been reported in which online measurements of the off-gas of yeast fermentations were performed. While Kotiaho [322] focused on the quantification of one single metabolite, ethanol, Kohlemainen et al. measured patterns of off-gas metabolites without any analyte identification [323]. The potential of IMS analyses for quality control during beer fermentation was shown by Vautz et al. [324] by monitoring the ripening indicators diacetyl and 2,3-pentanedione. In this contribution, we evaluated the capabilities of MCC-IMS for online monitoring of mVOCs produced by S. cerevisiae. Different growth conditions were tested and special emphasis was put on the dynamics of the VOC profiles during transient metabolic conditions, i.e., the shift from respiratory to fermentative metabolism.

Material and Methods Yeast strains and growth conditions The yeast strains used in this study were the reference strain S. cerevisiae CEN.PK 113-7D [325] and the isogenic ADH1 knockout mutant S. cerevisiae CEN.PK 117 YOL086c::kanMX4 [326] devoid of the main alcohol dehydrogenase Adh1p. The yeast strains were grown in Verduyn minimal salt medium −1 −1 −1 −1 [237] containing 20 g L glucose, 3 g L KH2PO4, 0.5 g L MgSO4·7 H2O 20.4 g L potassium hydrogen phthalate as well as 1 mL of vitamin solution and 1 mL of trace elements. The nitrogen −1 source (NH4)2SO4 was replaced by 2 g L urea, as at higher concentrations of ammonia, the water chemistry of the IMS ionization could shift to ammonia chemistry [327]. The vitamin solution contained 0.05 g L−1 D-biotin, 1 g L−1 calcium D-pantothenate, 1 g L-1 nicotinic acid, 25 g L−1 myo- inositol, 1 g L−1 thiamine hydrochloride, 1 g L−1 pyridoxine hydrochloride and 0.2 g L−1 p- −1 −1 aminobenzoic acid. The trace element solution consisted of 15 g L EDTA, 4.5 g L ZnSO4·7 H2O, −1 −1 −1 −1 1 g L MnCl2·4 H2O, 0.3 g L CoCl2·7 H2O, 0.3 g L CuSO4·5 H2O, 0.4 g L NaMoO4·2 H2O, −1 −1 −1 −1 4.5 g L CaCl2·2 H2O, 3 g L FeSO4·7 H2O, 1 g L H3BO3 and 0.1 g L KI. All precultures were performed in 500 mL shake flasks filled with 10% medium at 30 °C and 250 rpm. The bioreactor experiments were run in a Sartorius Biostat A plus bioreactor (Göttingen, Germany) with a working volume of 1 L at 30 °C. The fermentation parameters were controlled by an external computer and the software PC Panel μDCU. The pH was monitored with a Mettler Toledo pH electrode and controlled at pH 5 using 10 M potassium hydroxide and 4 M hydrochloric acid. In aerobic fermentations, the fermenter was aerated with pressurized air with a flow rate of 2.8 L min−1. To reduce impurities, the air was filtered by a Sartorius Midisart 2000 sterile filter (0.2 μm pore size) and passed through water, resulting in water-saturated gas. Dissolved oxygen was monitored with a Hamilton Oxyferm dissolved oxygen electrode. If not mentioned otherwise, the dissolved oxygen tension was maintained at 90% by adjusting the stirrer speed. Bioreactor cultivations were started by inoculating to an OD600 of 0.1. During batch cultivation, sample for OD and HPLC measurements were taken regularly in the first 12 h. Continuous, glucose-limited cultivations were run at a dilution rate of 0.13 h−1 by feeding fresh medium at a flow rate of 2.5 mL min−1. A constant volume of culture broth was maintained by positioning a tube at a predetermined height that corresponded to 1120 mL 73 Chapter 2 volume and connecting it to a separate pump that removed excess fluid. Both batch and continuous cultivations were single experiments. To check for possible contamination during fermentations, samples were examined daily under the microscope (Leica DM750) with a 10X ocular and a 100X oil immersion objective.

Analytics The optical density was determined with an Ultrospec 10 photometer (Amersham Bioscience, Amersham, Switzerland) with a fixed wavelength of 600 nm. When necessary, the samples were diluted using demineralized water. For the determination of glucose and fermentative byproducts, samples were taken directly out of the fermenter using a syringe and a steel pipe. Samples were harvested by centrifugation (Heraeus Megafuge 16R, Thermo Fisher Scientific, Marietta, Ohio, USA) at 5,000 rpm for 5 min at 4 °C. The supernatant was stored at −20 °C until further analysis. Analytes were separated using an organic acid resin column (C-S Chromatography, Langerwehe, Germany) at 50 °C. 5 mM H2SO4 was used as eluent at a flow rate of 0.8 mL min−1 (System Gold 125 Solvent Module). Analytes were detected with a UV detector (166 Detector, (Beckman Coulter, Krefeld, Germany) at a wavelength of 210 nm and a RI detector (Melz Differential Refractometer LDC 201) operated at 50 °C. Standard solutions of the analytes were measured in concentrations of 0.1, 0.5, 1, 5, 10, 20, 40 and 50 g L−1.

SPME GC-MS measurements for validation of volatile metabolites To cross-check the identification of volatile compounds, 10 mL of the supernatant were sampled from the bioreactor and transferred into a headspace vial (20 mL). Metabolites were extracted from the headspace of culture supernatants via solid phase microextraction (CAR/PDMS fibers, Supelco, Steinheim, Germany) and analyzed with GC-MS as described in [159]. Briefly, the samples were incubated at 50 °C for 15 min and after that agitated for 10 s at 250 rpm. Afterwards, the analytes were desorbed from the SPME liner of the GC at 250 °C. GC-MS analyses were carried out with an Agilent 7890A gas chromatograph equipped with an Agilent 7000B triple-quadrupole mass spectrometer (Agilent Technologies, Waldbronn, Germany). The split was set to a ratio of 5:1. The column temperature was increased from 40 to 250 °C with a rate of 12 °C min−1. All other parameters were identical as those described in [159]. For mVOC identification, retention indices and mass spectra were compared with the NIST mass spectral library (Version: 2.0) and data published in the FlavorNet database (flavornet.org, [64]).

MCC-IMS measurements The MCC-IMS used was a BreathDiscovery (B&S Analytics, Dortmund, Germany) with an upstream multi capillary column type OV-5 (Multichrom Ltd. Novosibirsk, Russia) of 17 cm length consisting of approx. 1000 capillaries. The capillaries have an inner diameter of 40 μm and are coated with OV- 5 stationary phase with a film thickness of 0.2 μm. The column temperature was set to 40 °C. Samples were ionized with a 550 MBq 63Ni ion source. The ionized analytes were introduced into the drift column (length, 120 mm) through a shutter that had a pulse frequency of 50 ms and an opening time of 30 μs. Separation in the drift chamber was carried out in a negative coaxial electric field with an intensity of 300 Vcm−1 (positive measurement mode). The MCC-IMS was operated at ambient 74 Results temperature and pressure (i.e., laboratory conditions). Raw data of mVOCs was acquired using VOCan (B&S Analytik) with a frequency of 10 Hz, 5 consecutive single spectra were averaged. A single round of data acquisition required 0.5 s. The program was used to control all parameters of the MCC-IMS like gas flow and temperature, and to program measurement sequences. Nitrogen 5.0 (Westfalen, Münster, Germany) was used both as drift gas in the MCC-IMS and as carrier gas in the MCC. The drift gas flow rate was set to 100 mL min−1; the carrier gas flow rate was set to 150 mL min−1 during batch and 50 mL min−1 during continuous fermentation. The fermenter off-gas was introduced into the system through a 10 mL stainless steel sampling loop coupled to a six-port valve. Between single measurements the MCC-IMS and the sampling line was purged with a nitrogen flow of 100 mL min−1. The MCC-IMS topographic plots were evaluated using the software VisualNow (B&S Analytik). −2 Reduced inverse ion mobilities, 1/K0 (Vs cm ), were calculated by normalizing the measured drift velocities (drift time per drift distance) to the electric field, temperature, and pressure. This reduced ion mobility is characteristic for the ion and independent of the experimental conditions. The program allowed the definition of peak regions and comparison of the peak intensities in different datasets. The intensities of detected peak regions, the reduced ion mobility and retention time were exported to Excel for further data evaluation [328]. The MCC-IMS was connected to the fermenter off-gas with a Teflon tube (ID, 1.58 mm; length, 1000 mm). To prevent overloading of the MCC-IMS, the off-gas was diluted with sterile, moisturized air or nitrogen at a flow rate of 2.4 mL min−1. The air was filtered with a 0.2 μm sterile filter. Mixing was achieved by introducing both gas streams into a 500 mL Schott bottle.

Volatile metabolite identification For mVOC identification, pure standard substance measurements were performed for which the MCC-IMS was operated with identical parameters as during the continuous cultivations. One mL of aqueous standard solutions of a concentration of 0.01 g L−1 was filled in 100 mL Erlenmeyer flasks, which were closed with an aluminum cap. The sampling tube was introduced to the flask and the measurement was started.

Results and Discussion Experimental setup To enable off-gas measurements, the MCC-IMS had to be connected to the bioreactor in a way that guaranteed that all analytes are captured. To achieve this, the off-gas of the bioreactor was connected to a mixing chamber where it was diluted with filtered, compressed air. To ensure proper mixing, the gas inlets were positioned at the bottom, while the gas outlet to the MCC-IMS was positioned in the upper section of the chamber. A second outlet leading through a filter into the environment was positioned at the top, to ensure that no pressure build-up is possible (Figure 9). Between measurements, the MCC-IMS and sampling line were flushed with nitrogen at a flow rate of 100 mL min−1 (Figure 4). To avoid microbial contaminations and the introduction of volatile impurities from the compressed air, the gas used for aeration of the bioreactor was passed through a 0.2 μm filter and a water bath. The overall setup allowed controlled sampling without disturbances

75 Chapter 2 from the environment as no impurities were observed in the MCC-IMS during abiotic operation of the bioreactor.

Figure 9: Experimental set-up for the online MCC-IMS measurements of fermenter off-gas. Previously published in [185].

MCC-IMS monitoring of batch cultures So far, most studies of yeast VOCs focused on the determination of volatiles produced during wine fermentation and their impact on wine aroma. These studies did not discriminate whether the VOCs originated from yeast fermentation or the products of biotransformations of the grape metabolites [249-252]. Here, to explore the mVOCs de novo synthesized by baker’s yeast, minimal salt medium with glucose as the sole carbon source was used. For a first evaluation of the MCC-IMS set-up for real- time fermentation monitoring, S. cerevisiae was grown in batch culture. Before inoculation of the fermenter, signals occurring from the sterile medium were measured under process conditions (stirrer speed, temperature and pH control, aeration). During a period of 12 h, the MCC-IMS measurements (Figure 10) showed four major signals of invariant intensity (data not shown).

76 Results

Figure 10: MCC-IMS topographic plot of sterile Verduyn medium. The reaction ion peak −2 (1/K0 = 0.5 Vs cm ) was compensated by the software VisualNow.

After inoculation to a starting OD600 of 0.1, MCC-IMS analyses were performed throughout the growth experiment in one hour intervals. In addition to the signals detected in the sterile medium, 19 peaks emerged during the batch growth experiment, at retention times between 1 and 190 s (Figure 11). The time course of the six most distinctive peaks is shown in Figure 12 B, C. To correlate the MCC- IMS patterns of the volatile metabolites to the yeast physiology, the optical cell density, carbon source consumption, and byproduct formation (ethanol, glycerol) were quantified in parallel to the MCC- IMS measurements (Figure 12 A).

77 Chapter 2

Figure 11: MCC-IMS topographic plot of S. cerevisiae during the early stationary phase of a batch fermentation. Boxes indicate signals that showed the most significant changes during the growth (perturbation) experiments.

In the single fermentation experiment, the profile of signal C3-5 correlated with the growth rate since its signal increased after the lag phase and reached its maximum during the exponential growth phase. In the stationary phase, the peak intensity of C3-5 decreased again. Peak C3-9 was detected at the same time as ethanol, measured in the fermentation broth, and its signal diminished simultaneously with complete ethanol consumption. The signal of C3-8 showed an increasing trend in the late exponential to early stationary phase, when ethanol was nearly consumed. Signal C3-15 showed a rather peculiar profile. The signal intensity first increased, but rapidly decreased after 5 h and maintained a constant level for about 11 h. After this period, the intensity of C3-15 abruptly increased and reached its prior maximal value followed by a steady decrease at the end of the batch. This behavior might be explained by incomplete ionization of the analyte molecules. The intensity of the two signals decreased when the intensity of peak C3-5 increased. This signal might have a higher proton affinity and therefore be preferably ionized to substance C3-15. This hypothesis is substantiated by the low reaction ion peak (RIP) in the MCC-IMS chromatograms for the period between 5 h–21 h (data not shown). The RIP consists of reaction ion molecules originating from the drift gas, here nitrogen. In the ionization chamber, water molecules react with positively charged + nitrogen ions to a cluster of the type (H2O)nH . These ions form the RIP and transfer the charge to molecules with a higher proton affinity. Hence, with increasing analyte concentration the RIP diminishes. For more detailed information about the charge transfer reactions, the reader is referred to [128]. Note that the data presented here originates from single experiments and are thus not based on statistics. The intention of this work was the development of a set-up for online MCC-IMS measurements of volatile metabolites in the off-gas of yeast fermentations, with which we will generate more comprehensive datasets in future experiments.

78 Results

Figure 12: (A) Fermentation profile of S. cerevisiae during batch growth in glucose minimal medium, (B) trends in intensity, (C) heat map of selected peaks detected by MCC-IMS analysis of the fermentation off-gas. Areas in the heat map show the detected signal peak and the surrounding area; DO = dissolved oxygen.

To elucidate the potential of MCC-IMS analyses for the differentiation of different S. cerevisiae strains or mutants we compared the MCC-IMS chromatograms of S. cerevisiae CEN.PK 117 YOL086c::kanMX4, deficient in the major alcohol dehydrogenase Adh1p, with its isogenic reference strain. Again, we want to stress that in this proof-of-principle work, only single experiments were performed to show the general applicability of MCC-IMS for online measurements of fermentation off- gas. The ADH1 deletion mutant had a reduced growth rate and biomass yield, about 56% and 42% of the reference values. The ethanol formation was clearly reduced (maximal accumulation of 0.8 g L−1 vs. 6 g L−1 for the reference strain), while glycerol production was increased (Figure 13 A). These differences in the strain physiology were also reflected in the MCC-IMS pattern (Figure 13 B,C). While no new peaks compared to the reference strain chromatograms were detected, the profiles of several peaks differed. Peak C3-15 increased much slower compared to the reference strain fermentation. The signal of C3-5 rose in the beginning, stagnated within the 5th–20th h after inoculation and increased afterwards. While the absolute intensity of peak C3-6 was lower in the measurements of the mutant strain compared to the reference strain, the time profile was similar for both cultivations. The profile of peak C3-20 was similar to that in the reference strain cultivation but showed a shallower increase at the start of the fermentation, while the intensity of C3-8 rose faster. Although replicate experiments are required for a statistically valid statement, we hypothesize that these distinct mVOC derived MCC-IMS signals allow differentiation of the two yeast strains. Similarly, species specific volatile footprints or markers have are already used to detect cancer via breath analysis of patients [113] or fungal contaminants in buildings [329].

79 Chapter 2

Figure 13: (A) Fermentation profile of S. cerevisiae adh1Δ during batch growth in glucose minimal medium and (B) trends in intensity and (C) heat map of selected peaks detected by MCC-IMS analysis of the fermentation off-gas. The areas in the heat map show the detected signal peak and the surrounding area; DO = dissolved oxygen.

These measurements gave first valuable information about the volatile metabolite patterns produced by yeast and their dynamics during growth on glucose minimal medium. However, because of the high signal of volatiles, most likely ethanol and acetaldehyde, in the off-gas and probable incomplete ionization of metabolites, data interpretation and thorough metabolite detection under these conditions was limited. To make batch observation possible, other column materials or longer columns would be needed, however, longer columns would increase the analysis time by far [324].

MCC-IMS monitoring of glucose-limited chemostat fermentations High glucose concentrations as in batch fermentations trigger glucose repression. One consequence is the induction of respiro-fermentative metabolism that is aerobic ethanol formation, known as Crabtree effect [289, 330]. This switch of the metabolic mode is undesirable for example during production of yeast biomass or protein production, as ethanol formation reduces product yield and quality [331-334]. Such processes are therefore run as glucose-limited fermentations in which this regulatory mechanism is repressed. While industrial processes are most often run as fed-batch fermentations, in academic research glucose-limited chemostats are favored as these continuous cultivations can be performed under defined, controlled, and constant conditions allowing very reproducible experiments. Moreover, this fermentation mode allows varying one single fermentation parameter, making it ideal for studying the impact of specific perturbations on growth physiology or metabolism. To elucidate possible changes in the volatile metabolites during the transition from

80 Results respiratory to fermentative metabolism, we cultivated S. cerevisiae in a glucose-limited chemostat at a dilution rate of 0.13 h−1. Under these fully oxidative growth conditions, S. cerevisiae produced no ethanol. At metabolic steady state of the single experiment, the MCC-IMS chromatogram showed 13 peaks, which were not detected in the sterile medium, while two peaks observed in the sterile medium were not or with considerably less intensity detected in the chemostat culture (Figure 14). Because of the overloaded chromatogram during batch cultivation, it is difficult to state, which of these are specific for the glucose-limited respiratory growth conditions and which appear generally during growth of S. cerevisiae. The most distinct peaks were identical for both growth conditions.

Figure 14: MCC-IMS topographic plot of the off-gas of a glucose-limited continuous cultivation of S. cerevisiae. Boxes indicate signals that showed the most significant changes during the growth −2 (perturbation) experiments. The reaction ion peak (1/K0 = 0.5 Vs cm ) was compensated for by the VisualNow software.

To induce a shift from respiratory to fermentative metabolism, the steady state culture was perturbed once with a pulse of 22 mmol glucose, which was rapidly injected into the bioreactor. Immediately after the glucose pulse, ethanol accumulated in the fermentation broth (Figure 15 A). Acetate (data not shown) and glycerol were detected as well and showed a similar profile as ethanol. The surplus glucose was consumed within 75 min after which ethanol was catabolized and diminished about 3 h after the pulse. The optical density (OD600) increased from 31–39. The intensity of several signals detected with the MCC-IMS increased after the pulse and decreased again after about 2 h (Figure 15 B,C). The most prominent changes were observed for the signals marked in Figure 14, these are the same peaks as in the batch cultivation of the wild type yeast and the ADH1 mutant. Peak C3-15 showed a strong correlation with the ethanol concentration determined in the fermentation broth. However, other peaks, like C3-6 and C3-8, did not resume the intensities prior to the perturbation, but maintained a higher level within the 7 h of MCC-IMS monitoring, hence, correlated

81 Chapter 2 with the increase in biomass concentration. C3-5 was the only peak whose intensity decreased upon glucose addition. About 3 h after the glucose pulse, its intensity increased again and regained its original value in the next 3 h. C3-20 showed a rapid decrease directly after the pulse, remained at a constant level and showed a decreasing trend after about 2 h and 30 min.

Figure 15: (A) Fermentation profile of S. cerevisiae during growth in a glucose-limited chemostat and (B) trends in intensity and (C) heat map of selected peaks detected by MCC-IMS measurements of the fermentation off-gas after perturbation of the metabolic steady state with a pulse of 22 mmol glucose; DO = dissolved oxygen.

The physiological response of S. cerevisiae to limited oxygen availability is very similar to that of the Crabtree effect. In both cases, the fluxes through glycolysis are upregulated, while the fluxes through the TCA are downregulated and as a consequence ethanol is produced [294, 335]. To elucidate possible differences in the volatile metabolite patterns of yeast cultures responding to a glucose pulse and anaerobiosis, respectively, in a second perturbation experiment gassing was switched from air to nitrogen. In this single perturbation experiment, the dissolved oxygen (DO) concentration dropped to zero within 10 min. Simultaneously with this drop, ethanol and glycerol accumulated in the fermentation broth, while only little acetate which accumulated only after 1 h of anaerobic growth (Figure 16 A). The biomass concentration decreased slowly during anaerobiosis. Note that the data of this perturbation experiment cannot be directly compared to the glucose pulse as the air supply was permanently replaced by the same flow rate of nitrogen.

82 Results

Figure 16: (A) Fermentation profile of S. cerevisiae during growth in a glucose-limited chemostat and (B) trends in intensity and (C) heat map of selected peaks detected by MCC-IMS measurements of the fermentation off-gas during transition to anaerobic conditions; DO = dissolved oxygen.

As in the glucose pulse perturbation experiment, peak C3-15 increased as soon as ethanol was produced (Figure 16 B, C). The increase flattened after about 60 min and decreased after about 2 h. Again, incomplete ionization and possible incomplete evaluation of the peak area might be responsible for this trend. The intensities of peak C3-6 and C3-8 increased rapidly in the first 30 min after the shift to nitrogen gassing, considerably faster than in the glucose pulse experiment. C3-9 showed a slight increase, while peak C3-5 stayed constant over the period of 3 h. Peak C3-20 showed a similar trend as in the glucose pulse perturbation, a rapid decrease after the shift to nitrogen gassing followed by a steady intensity during the 3 h of monitoring. The dynamics of the MCC-IMS signals were very similar to that observed during the imposed Crabtree effect and no new peaks were detected. However, S. cerevisiae responded faster to anaerobic than to glucose excess conditions. Changes in the peak intensities of, for example, C3-15 were already observed 7 min after the shift to nitrogen gassing, i.e., before complete anaerobiosis, while the first changes in the MCC-IMS peak intensities after the rapid glucose pulse were observed only after 17 min. In both perturbation experiments, samples were measured in intervals of 215 s over a period of about 30 min. These rapid measurements show the potential of MCC-IMS analysis for online monitoring and control of bioprocesses.

83 Chapter 2

Identification of volatile metabolites A limitation of MCC-IMS is its poor ability for analyte identification by either retention time or ion mobility, or combination of both. While databases of reference molecules are developed, the resolving power is low. To get first hints of the mVOC spectrum in the off-gas, we performed HS/SPME GC-MS analyses. Volatile metabolites were extracted from 10 mL the headspace of culture supernatant by solid phase microextraction. Peaks detected in the GC-MS analyses were identified by comparing retention indices and spectra with data from the NIST library (Version 2.0) and the FlavorNet database [64]. Although this identification procedure often results in ambiguous results, it narrowed down the standard measurements conducted with the MCC-IMS for peak identification. GC-MS samples were taken from the glucose-limited chemostat cultivation of S. cerevisiae CEN.PK during steady state conditions, after perturbation of the continuous cultivation by a glucose pulse and shift to anaerobic conditions (15 min and 50 min after pulse injection and shift to anaerobic conditions, respectively). Samples were also taken of the batch cultivation of the isogenic ADH1 knock-out mutant during exponential growth and stationary phase. Several peaks in the GC-MS analyses were found in all samples, while some peaks were specific for the different growth conditions or strain used. In total, metabolites could be assigned to 10 peaks by database searches (Table 9). Compounds found in the samples of all culture conditions and the S. cerevisiae adh1∆ mutant were ethanol (in different amounts), 2-pentanone, 2-phenylethanol, 2,3-hexanedione, butyric acid, and isobutyric acid. Benzaldehyde was found in all samples except those from anaerobic cultures. Acetoin and 2,3-butanediol were only detected in the samples of the S. cerevisiae adh1∆ strain. Acetoin is an intermediate of the pathway to 2,3-butanediol and can be derived from pyruvate and acetaldehyde via three different pathways [336-341]. With the reduced flux of acetaldehyde to ethanol in this mutant, conversion of acetaldehyde to acetoin and further to 2,3-butanediol might be induced as has already been reported for a ADH1, ADH3 and ADH5 triple knockout mutant of S. cerevisiae BY4742 [342]. By contrast, isovaleric acid was detected in all samples except in those from the ADH1 knock-out mutant. Besides the 10 metabolites identified in the GC-MS analyses, standards of 1- butanol and acetaldehyde were measured by MCC-IMS. mVOCs were assigned to unknown peaks MCC-IMS signals by comparing the retention time and reduced inverse ion mobility to data of the pure standard substance measurements. As already assumed from the correlation with the ethanol concentration in the fermentation broth, peak C3-15 was identified as ethanol. Compounds found in the GC-MS, which are part of the Ehrlich pathway, were isobutyric acid, isovaleric acid, and 2- phenylethanol and were also measured via MCC-IMS. The Ehrlich pathway is a catabolic pathway which degrades amino acid into aroma compounds such as higher alcohols or volatile fatty acids [246]. The fusel acids (and corresponding alcohols) isobutyric acid, isovaleric acid, and 2- methylbutanoic acid are, for example, derived from the branched chain amino acids valine, leucine, and isoleucine, respectively, by the activity of aldehyde dehydrogenases, encoded by ALD2 to ALD6 [253]. Only isobutyric acid could be assigned to one of the peaks (C3-8) detected in the off-gas analyses of yeast fermentations. Unambiguous identification of 2-phenylethanol was not possible due to overlapping peak regions, which is due to co-elution and similar drift times, with 2,3-butanediol. The pure standard MCC-IMS measurements identified peak C3-5 as 2-pentanone. In Penicillium roqueforti, 2-pentanone is derived from β-oxidation of fatty acids and might be synthesized in S. cerevisiae via the same pathway [343]. However, although 2-pentanone has been found in several S. cerevisiae fermentations [66, 247], a biochemical confirmation of the synthesis via β-oxidation is so far not described. Peak C3-7 (not shown in Figure 10 and Figure 14) was identified as 2,3- 84 Results

hexanedione, which is reported as a metabolite of brewer’s yeast with a cheesy aroma [193] and has recently been detected in the headspace of agar plates cultures of Corynebacterium glutamicum [344].

Table 9: Volatile organic compounds detected in fermentations of S. cerevisiae via GC- MS measurements growing in glucose minimal salt medium. ND, not detected. reduced inverse MCC-IMS GC-MS SESI- MCC- Compound peak ion mobility retention retention Orbitrap IMS −2 1/K0 (Vs cm ) time (s) time (min) 2,3-butanediol x x 0.575 4.5 26.517 2,3-hexanedione ND x C3-7 0.57 19.4 9.633 2-pentanone ND x C3-5 0.554 6.4 5.029 acetoin x x 0.532 8.5 16.932 benzaldehyde ND x 0.566 37.5 25.366 butyric acid ND x 0.63 24.4 30.127 ethanol x x C3-15 0.516 4 4.193 isobutyric acid ND x C3-8 0.618 10.4 27.827 isovaleric acid ND ND − − 31.594 2-phenylethanol ND x 0.578 4.5 39.67

The low recovery rate of metabolites identified by GC-MS might partially be explained by the different sampling procedures. While for the MCC-IMS measurements 10 mL of the off-gas were directly measured, for GC-MS analyses, VOCs were extracted from the culture broth at 50 °C and were preconcentrated. Furthermore, GC-MS peaks were only tentatively identified by comparison with databases and require validation by pure standard analyte measurements. However, 2,3- butanediol and acetoin have also been found in the SESI-Orbitrap measurements and therefore support the designation as identified (Table 9) [295]. To broaden the spectrum of analyte detection in the MCC-IMS, measurements with both negative and positive ion mode are useful. Our future experiments will focus on the identification of unknown MCC-IMS signals, including thorough verification by GC-MS measurements of pure standard substances. Ideally, feeding experiments with labelled precursors should be performed to conclusively prove whether the identified compounds are actually produced by S. cerevisiae, this verification needs to be carried out using a suitable MS.

Conclusion and Outlook Application of MCC-IMS in positive ion mode for online monitoring of fermentation off-gas detected 19 signals produced by S. cerevisiae during growth on glucose minimal salt medium. Although the here presented first analyses probably captured only a minute fraction of S. cerevisiae’s volatile metabolite spectrum, they were sufficient to differentiate S. cerevisiae strains and to reveal the impact of different growth conditions on the production of mVOCs. Also, the capability of the MCC-IMS for online measurements was shown. Four compounds were identified by complementary GC-MS measurements of fermentation broth extracts and pure standard substance measurements.

85 Chapter 2

However, limitations of the device were apparent, when small amounts of ethanol lead to an overloading of the device with problems in overall detection. Therefore, the setup cannot be used in batch fermentations. Another problem is that all signals accumulate at low 1/k0 values and low retention times. To increase the resolvability and to prevent overloading the MCC-IMS configuration has to be optimized by, e.g., installation of a sample loop of lower volume or a more polar MCC. In the following chapter, the investigation of VOCs is taken to another level, by adding a transcriptome analysis to the MCC-IMS measurements. Also, the focus was shifted from solely laboratory to near industrial conditions and a comparison of these two conditions.

The Crabtree effect revisited A comparative study on the transcriptional changes during the Crabtree effect using laboratory and industrial strains and conditions

Contributions: This study was designed by Christoph Halbfeld, Birgitta E. Ebert, Sven Rahman, Erik Pollmann and Lars M. Blank. The experiments were performed by Christoph Halbfeld, Erik Pollmann, Ann-Kathrin Sippel and Michael Quantz. The data was evaluated by Christoph Halbfeld, Ann-Kathrin Sippel, Sven Rahmann and Birgitta E. Ebert. Here the microarray-data was statistically evaluated by Sven Rahmann and checked for biological relevance by Christoph Halbfeld and Birgitta E. Ebert. The MCC-IMS data analysis was performed by Ann-Kathrin Sippel, Christoph Halbfeld and Birgitta E. Ebert evaluated the data. The chapter was written by Christoph Halbfeld.

Results

The Crabtree effect revisited A comparative study on the transcriptional changes during the Crabtree effect using laboratory and industrial strains and conditions Summary S. cerevisiae is broadly used in research and in industrial processes. The yeast strains used in the laboratory and by the industry are related but differ in metabolic activity. However, the two groups and growth conditions have up to now not been compared in relation to pathway usage before and after a metabolic shift. Here the laboratory yeast strain S. cerevisiae CEN.PK 113-7D in minimal salt medium was compared with the industrial strain S. cerevisiae DHW cultivated in industrial molasses medium, for both strains and conditions the Crabtree effect was induced by overfeeding. The strains were investigated using volatile metabolite as well as microarray analyses. No differences could be detected between the two strains’ volatilome, however, differences in the transcription of genes involved in leucine biosynthesis as well as oxidative phosphorylation were observed. This could be especially interesting for the industrial yeast production since the lack of respiratory inhibition in the industrial strain bears the potential of a higher biomass yield compared to the laboratory strain.

91 Chapter 2

Introduction The Crabtree effect describes alcoholic fermentation by Crabtree positive yeasts under aerobic conditions with high sugar concentrations [289]. This metabolic shift causes a rerouting of the relative central carbon metabolism fluxes from the pentose phosphate pathway to the glycolysis [294]. In addition, high glucose concentrations repress parts of the electron transport chain and therefore respiration, an effect called glucose catabolite repression. Beyond respiration, also the non-glucose sugar metabolism as well as the hexose metabolism are specifically up- or downregulated on transcriptional level, depending on the amount of available sugar [345, 346]. However, metabolic effects outside the central carbon metabolism, especially on volatile compound synthesis, have not yet been investigated in detail. To gain insight into pathways that are differentially activated prior to and past a metabolic shift, it is useful to investigate the transcriptome. For this purpose, microarray analysis is a powerful method to generate a broad overview of transcriptional differences. In microarrays, probes consisting of specific DNA sequences, representing the genotype of the specific strain, are spotted on a glass slide. For the microarray, DNA is required, however, mRNA describes the expression state of the cell. To convert mRNA into cDNA a reverse transcriptase is used. In the same step, the cDNA is also marked with a fluorescent dye and subsequently applied to the microarray. The fluorescence intensity of the single spots on the microarray correlates with the amount of bound cDNA, and thus the expression level of the gene [347-349]. To cope with changing environmental conditions, the cells need to quickly adapt their repertoire of enzymes. Here it is possible that the required enzymes are either needed to be produced or are already dormant inside the cell and have to be activated by posttranslational modification (PTM). Phosphorylation is the most used PTM in S. cerevisiae [350, 351]. If the enzymes need to be produced de novo, a large amount of mRNA will be existent in the cell, and even if an enzyme is constitutively present in the cells there is still some mRNA since the cell’s proteome is not static. Therefore, by using DNA microarrays it is possible to get an in-depth insight into the transcriptional state of the cell and based on that an estimation of the metabolic state. The transcriptional changes of yeast during the Crabtree effect have been investigated before [352] as well as the shift from aerobic to anaerobic conditions [353]. Kresnowati et al. found out that transcripts of the TCA cycle genes are decaying one order of magnitude quicker than previously expected leading to a quick metabolic transition [352]. However, all studies on S. cerevisiae’s metabolic shift, the Crabtree effect, previous to this thesis have been based on chemostat experiments, Regenber et al. reported that the growth rate in chemostats has an influence on the transcriptome [354]. Even though a multitude of studies investigated the yeast transcriptome [352, 354, 355] there has not been any study that investigated the Crabtree effect during fed-batch experiments and that compares laboratory and industrial strains, however other comparisons between industrial and laboratory strains have been performed [356]. Because of S. cerevisiae’s importance in both, scientific studies as well as industrial processes, it is important to investigate and compare the yeast metabolism under both conditions. Many scientific studies have been carried out with focus only on laboratory conditions, however in applied microbiology the industrial aspect and therefore industry-near conditions should also be considered because the behavior of the microorganism might change. Therefore, a laboratory strain in minimal medium was compared to an industrial strain in molasses medium in this study.

92 Results

To find out more about the regulation of the volatile metabolism at the onset of the Crabtree effect, custom microarrays based on the here measured samples were run. Samples were taken during carbon-limited fed-batch fermentations while the feed rate was gradually increased to induce a metabolic shift. This experiment was performed with a laboratory strain on minimal salt medium as well as with an industrial strain on industrial molasses medium.

Material and Methods Yeast strains and growth conditions In this study two types of yeast strains were used, the laboratory strain S. cerevisiae CEN.PK 113-7D (Euroscarf, Oberursel, Germany), from here on referred to as laboratory strain, and the industrial strain S. cerevisiae DHW (Deutsche Hefewerke GmbH, Nürnberg, Germany, from here on referred to as industrial strain. The laboratory strain was cultivated in Verduyn minimal salt medium (see chapter 2.2 for the medium composition and preparation), while the industrial strain was cultivated in industrial molasses medium. The strains were first cultivated in shake flasks at 30 °C in a New Brunswick Controlled Environment Incubator shaker (Eppendorf AG, Hamburg, Germany) shaker at 180 rpm. Hereby the medium was complemented with 20 g/L glucose for the minimal medium and 6 g/L mixed sugars for the industrial molasses medium. To increase the biomass for inoculation of the main culture, the strains were further cultivated in the same media (molasses medium with 7 g/L mixed sugars and Verduyn medium with 20 g/L glucose, respectively) in 18 L batch cultivations at 30 °C pH 5 overnight. The biomass was harvested by centrifugation in a Hettich Rotanta 460 centrifuge (Rotor: Typ 5624, Andreas Hettich GmbH & Co.KG, Tuttlingen, Germany) for 5 minutes at 4000 rpm. The cells were washed once with tap water and finally stored at 4 °C until further use. In the next step, the fed-batch cultivations were carried out. 7,5 L of salt solution (34 g (NH4)2HPO4, 10 g MgSO4x7H2O, 0.6 g ZnSO4x7H2O, 30 mg biotine [2%], 13 mg thiamine, 21 mg pyridoxine, 65 mg nicotinamide, 18 mg Ca-pantheonate and 35g (NH4)SO4 were put into the vessel. Then 88 gCDW of cells were diluted in 0.5 L tap water and filled into the reactor, finally the feed was started. The feed consisted of 320 g/L glucose in ten times concentrated Verduyn minimal salt medium for the laboratory strain, respectively, 320 g/L of fermentable sugar equivalents (based on CO2 formation) the in molasses medium for the industrial strain. The feed of the fed-batch was controlled based on the signal of the alcohol sensor (Alcoline, Biotechnologie Kempe, Kleinmachnow, Germany), hereby ethanol was tolerated in the first 3 hours of cultivation, after that the feed was adapted automatically to avoid ethanol formation. Later on, the feed was controlled manually to slightly overfeed the system. The other fermentation parameters were set as given in Table 10. The pH was controlled using 25% phosphoric acid and 20% ammonium hydroxide solution. The fed-batch was run carbon-limited with a growth rate of about 0.13 h-1 and about 0.19 h-1 for the laboratory and the industrial strain, respectively. To induce the Crabtree effect, the feed rate was slowly increased until ethanol formation was detected. After the detection of ethanol, the feed rate was reduced to reestablish carbon-limited conditions. At each sampling timepoint (frequency of 1 sample per minute), three samples were taken. The three samples of 100 μL each were stored in an ice-salt water bath, which had a temperature below 0 °C. When 24 samples were taken, the samples were centrifuged in a Hettich Universal 32 (Rotor: Typ 1624, Andreas Hettich GmbH & Co.KG, Tuttlingen, Germany) at 4000 rpm for 30 seconds. The supernatant was discarded and the pellet 93 Chapter 2 subsequently frozen in liquid nitrogen. These samples were used later for the microarray analysis. When ethanol was detected in the IMS signals, samples were taken every other minute. Based on the ethanol sensor and IMS data, it was decided which samples were to be used for microarray experiments. In total, 40 samples were sent to Oaklabs (Hennigsdorf, Germany), where transcriptome analyses were performed with custom microarrays.

Table 10: Profiles of fermentation parameters of the fed-batch fermentations. time [h] air [L/min] pH temp. [°C] 0 5 4.5 30 1 7 4.7 30 2 9 4.9 31 3 11 5.1 32 4 13 5.3 32 5 15 5.5 32 6 15 5.7 32 7 15 5.8 32 8 15 5.9 32 9 15 6.0 33 10 15 6.0 34 11 15 6.1 35 12 15 6.2 36 13 15 6.3 36 >14 15 6.4 36

MCC-IMS analysis The ion mobility spectrometer Breath Discovery (B&S Analytik, Dortmund Germany) was used for the detection of volatile metabolites in the fermentation off-gas. The MCC-IMS device was enhanced by B&S analytic especially for the measurements in a yeast fermentation headspace. In comparison to the device used in chapter 2.3, the multicapillary column was replaced by the MCC OV-1701 with a film thickness of 0.6 μm. Also, the sample loop was shortened from 10 mL to 50 μL. The MCC- IMS analysis measurements were performed as described in chapter 2.3 with the exception that the gas mixing flask was not used and the device was directly coupled with the bioreactor with a polytetrafluoroethylene tubing connected to the head plate of the bioreactor. Measurements were taken every 5 minutes.

cDNA microarray analysis The initial analysis of the microarray samples was carried out by Prof. Sven Rahman (Chair of Genome Informatics at the Faculty of Medicine at Duisburg-Essen University, Ruhr Professor for Bioinformatics at TU Dortmund). The samples were grouped into early (before a change of the IMS signals), mid (after changes of IMS signals were observed but prior to a change of the ethanol sensor signal) and late (after a change of the ethanol sensor signal) groups. Also, a false discovery rate value (FDR) was calculated based on the method described by Benjamini and Hochberg indicating the significance of the transcriptional change was [357]. Here only data with an FDR value of less than or equal to 5% were considered for further analysis. Fold change rates of genes with an FDR below 94 Results

5% in both strains were analyzed using BioCyc’s omics viewer [358]. For identified interesting pathways the fold changes were also considered, even if the FDR was above 5%. However, these results were evaluated cautiously and are not shown in the figures but descried in the text. Since the FDR rates are partially very high, some trends are indicated, but these data cannot be taken for a fact. All of the here indicated FDR values are listed in Supplementary table 1.

Results and Discussion MCC-IMS and microarray analysis of two yeast strains Carbon-limited fed-batch fermentations at pilot-scale (7.5-12 L) were run with a laboratory strain using Verduyn minimal salt medium and an industrial strain using molasses medium. The Crabtree effect was induced in both experiments by gradually increasing the feed rate. In each experiment, an ethanol sensor and an MCC-IMS were used for online monitoring of aqueous ethanol and volatile metabolites, respectively. Additionally, samples for microarray analyses were taken once per minute. Because of the high cost of the microarray analyses, only 40 samples were analyzed. 19 samples each were measured from the experiments of the laboratory strain on minimal medium and the industrial strain on the molasses medium, and two samples from an experiment with the laboratory strain on industrial molasses medium. From Figure 17, it is observable that in both experiments the MCC-IMS detected changes in the ethanol concentration minutes before the commercially available ethanol sensor. The MCC-IMS is not only capable of detecting ethanol but can also be used to detect other signals in the same run. Here, besides ethanol, isoamyl alcohol was detected. This has the advantage that detection of the Crabtree effect becomes more robust since it is not only relying on one single substance but two. Because of the early detection of the signals in the MCC-IMS, the microarray samples were divided in three groups. The first group (early) spans samples taken prior to any observable changes of the IMS signals when compared to the period before the feed was enhanced. This group also contained samples taken before the feed rate was enhanced. The early group encompassed 7 samples of the lab strain and 3 samples of the industrial strain. The second group (mid) included samples taken in the time period, when MCC-IMS signals changed, but the ethanol detector signal stayed constant. The mid group contained 8 samples in the lab strain and 11 samples in the industrial strain. Finally, the third group (late) consisted of samples taken after increased ethanol levels were detected with the ethanol sensor. The late group contained 4 samples of the lab strain and 5 samples of the industrial strain.

95 Chapter 2

Figure 17: Volatile signals of Saccharomyces cerevisiae CEN.PK 113-7D (A) and Saccharomyces cerevisiae DHW (B) in a fed-batch that was slowly overfed to trigger the Crabtree effect. The grey background represents the feed rate. The commercially available Alcoline sensor form Biotechnology Kempe was used for ethanol quantification in the fermentation broth along with the MCC-IMS Breath Discovery from B&S Analytik for the monitoring of volatile metabolites in the fermentation head space. The black dotted line indicates the start of the overfeeding process, the red dotted lines indicate the division into early (left) mid (middle) and late (right) phase. D and M behind the substance names describe the measured dimer and monomer respectively. 96 Results

The data was investigated for differences between the laboratory and the industrial strain. To identify changes that might be based on the medium rather than on the strain, transcriptional changes of the laboratory strain on molasses medium were evaluated as well. So far, this comparison has not been performed in any other study that we know of. The strongest differences in single pathways were detected in respiration, the TCA cycle and the metabolism of leucine and isoleucine. Some changes could be observed between the early and the mid samples, however the strongest changes were observed between the early and late samples. In the laboratory strain, the aerobic respiratory pathway was downregulated alongside with the tricarboxylic acid pathway (Figure 18). This is conform with the Crabtree effect induced changes described in the yeast literature [289, 359]. However, the industrial strain showed opposing result: In this strain, the genes of the TCA cycle and the respiratory chain, indicated in Figure 18 A and B by an orange color, were upregulated. Such a behavior has up to now not been described. In the aerobic respiratory pathway, the indicated genes that did not show a strong transcriptional change (SDH3, SDH2, QCR10, QCR9, QCR8, QCR7, QCR6, RIP1, COX11, COX19, COX16, COX20 and COX5B) had a FDR above 5%. However, the trends followed the indicated ones for all genes but COX19 in the laboratory conditions and COX20 and COX5B in the industrial conditions. This means that these genes were downregulated in the laboratory strain and upregulated in the industrial strain, while the exceptions showed the opposite behavior. It also should be noted that the FDR values for COX20 and COX5B were above 50%, indicating these values, as all values not indicated in Figure 18, only display tendencies and not hard facts. In summary, respiration was downregulated in the laboratory strain, while it was upregulated in the industrial yeast strain. The TCA cycle genes showed a similar gene expression pattern as those of the respiratory chain, i.e., downregulation of all genes of the lab strain and 4 genes of the industrial strain and upregulation of zero genes of the lab strain and 3 genes of the industrial strain. In addition to the expression changes indicated in Figure 18 B, the genes PYC1, PYC2, CIT1, LPD1, KGD1, KGD2, MDH1 also showed a similar regulation as the in Figure 18 indicated ones for each strain, i.e., downregulation in the lab strain and upregulation in the industrial strain, they were however not plotted, since their FDR was above 5%. ACO2, IDH1 and IDH2 showed a different expression profile. ACO2 was upregulated in the laboratory strain and downregulated in the industrial strain and therefore opposing the regulation pattern of the other genes in the TCA cycle, IDH1 and IDH2 were upregulated in the lab strain. However, the expression data of the IDH genes showed a 30% FDR value and might therefore not be as reliable as the other trends indicated in Figure 18. In experiments with the laboratory strain on industrial molasses medium, it was observed that the laboratory strain had a much lower critical growth rate (about 0.13 h-1) in comparison to the industrial strain (about 0.19 h-1) in 1 L fed-batch experiments (data not shown). These results in combination with the microarray data evaluation (Figure 18 A and B) indicate that the industrial strain might not slow down its respiration as the laboratory strain does, since the industrial strain did not seem to be affected by the glucose repression as the laboratory strain was. Thus, in the industrial strain, the biomass yield is theoretically higher, since more of the glucose could be channeled through the respiratory pathway instead of the fermentative pathway compared to the laboratory strain. This makes the industrial strain theoretically much more suitable for biomass production compared the laboratory strains. However, as ethanol was still produced, it has to be shown if this negatively affects the baking ability. This result also gives hints about the importance of the Crabtree effect in industrial yeast production. Since the Crabtree effect seems not to be as distinctive in the industrial conditions 97 Chapter 2 as anticipated from the laboratory results [294], its impact on the industrial yeast production might not be as big as could be anticipated from experiments with laboratory strains. However, higher yields could be achieved, if the Crabtree effect is avoided, since fermentative growth yields less biomass compared to respiratory growth. Since the transcriptome could also vary from the proteome and fluxome, further experiments would be necessary to proof this theory. The leucine biosynthesis was upregulated in the laboratory strain but was downregulated in the industrial strain. This is an interesting finding, since in both strains the production of the leucine degradation product 3-methyl-1-butanol increases during aerobic fermentation. The downregulation in the industrial strain however, might be attributed to the medium because the laboratory strain on industrial molasses medium indicated also a downregulation of the leucine metabolism (data not shown). Not depicted in Figure 18 C are the values for the genes ILV5, LEU9, LEU2 and BAT1 because their calculated false discovery rate was in at least one of the data sets above the threshold of 5%. However, the data still indicates a vague change of expression. All genes not indicated in Figure 18 C are upregulated in the laboratory strain and downregulated in the industrial strain. The only exception is LEU5, which was upregulated in both strains, however, the FDR of the industrial strain data was very high (55%) and might give a wrong indication. LEU4 codes for the major isoenzyme of α-isopropylmalate synthase as described by Casalone et al. in 2000 [360]. Therefore, it would be expected that the major gene coding for the enzyme catalyzing 3-methyl-2-oxobutanoate to 2-isopropylmalate would show a similar expression pattern compared to the other genes in the pathway, but this was not the case here. One possible explanation is that in the here tested laboratory strain, LEU9 codes for the major isoenzyme, another possibility is that based on the growth conditions either LEU9 or LEU4 is expressed as known for BAT1 and BAT2 [361]. BAT2 codes for the major branched-chain amino acid aminotransferase isoenzyme in the stationary phase, while the BAT1 encoded enzyme dominates in the exponential growth phase [361]. Here in both strains BAT2 was downregulated, while BAT1 was upregulated in the laboratory strain. This reflects the activity of the rest of the pathway, where genes were upregulated in the laboratory strain. The transaminases encoded by BAT1 and BAT2 are both, involved in the leucine biosynthesis and degradation. This and an indication based on the microarray data of higher THI3 (ketoisocaproate decarboxylase) expression, both enzymes involved in the degradation of leucine, can explain the higher levels of 3-methyl-1-butanol in the samples during the Crabtree effect at least for the laboratory strain.

98 Results

Figure 18: Transcriptional alterations of the most differing pathways based on expressional changes between Saccharomyces cerevisiae CEN.PK 113-7D (left) and Saccharomyces cerevisiae DHW (right). The two strains were run in fed-batch fermentations and overfeed gradually. The respiration pathway (A), the TCA-cycle (B) and the branched-chain amino acid production (C) are depicted. The genes without color coding are not displayed, since it had a false discovery rate of more than 5% in at least one of the tested strains.

99 Chapter 2

Conclusion and Outlook In this chapter, two different yeast strains, a laboratory and an industrial yeast strain were compared. Since in industrial yeast production, undefined molasses medium, and in laboratory applications defined medium is used, we replicated these conditions here in pilot scale. Here off-gas analysis and microarrays have been used to compare the two strains and media. The laboratory and the industrial strain showed no differences in the production of ethanol and isoamylalcohol in response of the Crabtree effect. Therefore, it can be assumed that all tests described in chapter 2.3 can be transferred to the here tested industrial strain under industrial conditions. While the metabolites detected in the off-gas showed similar behavior in both strains, differences in the expression of the genes of the leucine, valine biosynthesis pathway and the genes in the respiratory pathway including the TCA cycle could be observed in the microarray experiments. Due to upregulation of the respiratory pathway, it could be concluded that the Crabtree effect in the industrial strain is not as distinct as in the laboratory strain. However, since transcriptome data can only indicate hints and are not necessarily connected to the proteome and fluxome, especially in such short time intervals, further experiments are necessary to verify this assumption. In further experiments the physiology could be investigated. However, this might be challenging especially in industrial molasses medium. This diverse medium might make it difficult to use standard analytics (e.g., HPLC, OD-measurements). Therefore, special methods have to be developed to investigate the physiology and prove if the here seen changes in the transcriptome can be extrapolated to the yeasts physiology.

100

Chapter 3

General discussion

General discussion and outlook

General discussion Possible impact of this thesis The insights into yeast metabolism gained in this thesis could be used in many applications. The here introduced methods could have a major influence on future studies and the production of volatile compounds. These methods could be used as a blueprint for future studies in other organisms. Instead of implementing completely new pathways, the here generated knowledge could be used to simplify the production of volatiles. It would also be possible to use online metabolic flux analysis with the methods developed here. Lastly, the differentiation of metabolic states through the headspace could be used in process control along with product quality analysis and the possibility to detect possible contaminations.

Impact on future studies This thesis could be used as a blueprint for the investigation of volatile compounds in other organisms. For example, the investigation of soil bacteria could be interesting, since some of them use VOCs to communicate with each other in sparsely populated terrains [92]. The here used methods are not specific for S. cerevisiae but can also be applied for other microorganisms including prokaryotic organisms. Through minor adaptations to the methods used in this thesis, the VOCs even of multi- cellular organisms can be analyzed. For example volatile measurements of Begonia semperflorens have been conducted by Barrios-Collado et al. [297]. The identification of the measured volatiles is a challenge and can be pursuit using fragmentation of the metabolites in a mass spectrometer and subsequent comparison of the fragments with a database like NIST [362]. As many VOCs are rather small, isomers not easily identified by differences in mass occur, and hence it is always a good idea to implement a second source of verification. For that purpose, it is possible to run a standard of the expected metabolite. The yeasts metabolic network has been investigated in great detail, however, prior to this thesis little was known about the volatile metabolism [64, 194, 227, 229, 230]. The lack of information about the biochemistry of VOCs can be compared to the novel annotation of not well investigated organisms: basic information, like the genome sequence, can be compared by computational tools to existing databases. Here, with the aim to investigate possible pathways connecting new metabolites to the already known biochemical network. As described in this thesis, in a next step, the enzymatic capabilities can be investigated using databases, to make sure that the organism is capable of enzymatically catalyzing the reaction. However, if no enzyme with the required activity is identified or the enzymatic capabilities of an organism are not well investigated, there is still a chance of the presence of undiscovered enzymes or enzymes with moonlighting activity. Moonlighting describes enzymes that have the capability of catalyzing two independent reactions [276]. A good starting point for connecting new metabolites to an already existing metabolic network is the use of biological databases (e.g., KEGG) and tools that search for the connection (here fmm, pathpred and metaroute have been used) [207-209, 231, 232]. For the production of compounds that have not been described yet, the tool BNICE can be used, this program predicts new metabolic pathways based on a set of enzymatic reactions [234]. Using these tools is very convenient, since not all pathways have to be investigated manually. However, it has to be noted that the metabolic pathways need to be verified,

105 Chapter 3 before they can be assumed correct. FBA and TFBA can be used to increase the plausibility of added reactions as described in chapter 2.1 [219, 235]. It was also attempted to calculate which metabolites are volatile at room temperature based on their molecular weight. However, structural complexity and other factors like functional groups have a big influence on volatility, making a priori estimation of volatility difficult [363-365], although highly interesting for basic research and applications.

Strain engineering & possible use of volatile products With all the new knowledge gained in this thesis, where can it be applied? The here gained information could be used to knock out genes that would weaken the Crabtree effect and therefore, a nearly Crabtree negative strain that still holds the advantages like taste and texture of the industrial strain could be produced. The MCC-IMS could be used to detect changes in the 3D-topographic plot of the wildtype and knockout strain, to ensure that the flavor of the yeast strain is not affected by these changes. This could be tested in a laboratory strain and later the same deletions could be introduced into the different industrial strains that are used in the yeast production plants. This is possible because of the great similarities that were revealed in the microarray experiments between the laboratory and the industrial strain and conditions. Since the industrial strains are mostly not haploid, the CRISPR-Cas9 system is predestined to be used in these strains. It was shown before that polyploid strains can be manipulated easily [366-368]. Also new yeast strains could first be developed in the dry lab through application of the newly generated metabolic model and in the next step the created strains could be constructed in the wet lab. For example, by overexpressing pathways investigated in this thesis, yeasts could be designed that would produce a higher amount of selected VOCs and therefore have a special flavor. There are several compounds in yeasts that can create pleasant flavors if concentrated correctly. For example, acetate esters have a fruity aroma, while 2-phenylethanol smells like roses, also creamy or caramel like flavors can be found in yeasts repertoire [100]. If treated correctly, the designed flavor would not only enhance the yeasts flavor, but still appear in products created with this yeast. This way it would be possible to create special beers with strong taste of banana or a wine with a rose like bouquet. Since the initial research on volatile compounds comes mostly from the investigation of foodstuffs and beverages like wine, previous studies could be used to create new products with just the right amount of volatile compounds [9, 291, 369-371]. The balance of different volatiles is important because too high concentrations are also recognized as unpleasant [75]. These metabolites are not restricted to foodstuffs and beverages but could also be used in cosmetics, fragrances and perfumes. For some of the metabolites investigated in this thesis biotechnological processes are already in development [1, 372-375]. Another application would be the production of valuable metabolites that can be used for the production of bioplastics or biofuels. Furans and terpenes might be the biofuels of the future. Also, mycodiesels are close in composition to conventional diesel fuel and could take their place in the coming years [376, 377]. Terpenes are not only interesting, regarding biofuels but also for the production of drugs. Based on isoprene, medical relevant drugs could be produced directly in the yeasts [378].

106 General discussion and outlook

The use of different metabolic conditions An advantage of yeast is the possibility to easily control its metabolic shift from aerobe respiration to fermentation also known as the Crabtree effect [289]. Using this effect, it would be possible to let the strain accumulate biomass in a low sugar and highly aerated fermenter. Once enough biomass is present, the fermentation could be overfed to induce the production of special volatiles. Therefore, the strain needs to be engineered to only produce the product in question while fermenting. Through this thesis this engineering should be a lot easier than before because of the vast additions to the yeast metabolic model and the microarray experiments performed in both, a laboratory and an industrial strain.

Online flux analysis The here introduced methods are highly sensitive and online measurements are possible. When the detected peaks and masses can be assigned to metabolites, together with the enhanced yeast model it would be possible to take flux balance analysis to the next level and calculate live fluxes. Online flux analysis has been performed before, however through the noninvasive measurement of analytes and the large range, this method could still be enhanced [379]. This would increase the yield of industrial products, since the process parameters could be adjusted according to the microbial current metabolic state. If correctly applied, the yield of the desired product could be increased, especially in difficult production processes, e.g., where overfeeding might be problematic.

Conclusion In this thesis, the volatile space of S. cerevisiae has been investigated. Therefore, computational methods, novel analysis techniques and finally transcriptome analyses have been performed. Since volatile metabolites are of high value and are important in industrial processes (aroma, taste), not only laboratory, but also near industrial conditions have been examined in this thesis. Finally, we were able to enhance the yeasts metabolic network, to increase the number of tools available for the investigation of volatile compounds in microorganisms and to use the newly gained knowledge to compare laboratory with industrial conditions.

The consensus yeast metabolic model The consensus yeast metabolic network model (Yeast 7.5), is considerably lacking the volatile space [230, 380]. In this work, these shortcomings were addressed by investigating the yeast literature for known volatile metabolites and expanding the consensus model. A list of known volatile compounds has been created and possible pathways have been investigated. Therefore, the theoretical known enzymatic repertoire of yeast has been used, as well as computational tools and new pathways have been created. The newly created pathways have been verified through a thermodynamic flux analysis. In addition, 13 pathways have been verified by investigating 8 ADH knockout mutants that contain only one enzyme with ADH activity each and one mutant that contains no enzyme with any ADH activity. In total 219 reactions and 225 metabolites have been added to the metabolic model, while it was not practically possible to experimentally verify all of the reactions added; this might be addressed in future studies. It is possible that not all enzymes that have been added to the model are required for its function because some proteins might possess moonlighting activity [273, 276].

107 Chapter 3

Moonlighting describes enzymes with more than one catalytic activity, making it possible to have a smaller number of enzymes carrying out more reactions than anticipated up to this day [273, 276]. In conclusion, the metabolic yeast model covers now volatile compounds, and includes reactions that connect the volatiles with the know biochemistry. Novel experimental evidence can be easily used to curate the model in the future.

Sesi-Orbitrap To examine the volatile space of S. cerevisiae, new analytical tools have been used. The SESI- Orbitrap combines high time resolution with high mass resolving power and high mass accuracy, making online investigation of volatile dynamics in near real time possible. The Crabtree effect was examined as distinct metabolic shift with industrial importance in food production. Using the SESI- Orbitrap it was possible to investigate the Crabtree effect in a detail that it has never been investigated in before. It was possible to investigate sub-minute changes of metabolism for the first time and to observe metabolic network operation in near real time. Due to the new technology and the high redundancy of the measurements, the system delivers robust data. Up to now there are no comparable measurements with such high time resolution and mass accuracy. However, due to the high cost and high maintenance of an Orbitrap, its use outside of the lab is limited. For industrial applications another device, the MCC-IMS, was considered for the evaluation of quick changes in the volatile yeast space.

MCC-IMS

The MCC-IMS was used for measurements of yeast fermentations, since it can operate non-invasive, online, is a low-cost device and requires a relatively low maintenance [109, 381, 382]. Because of these properties, the IMS is already used in industrial applications, such as air quality control, food quality analysis, medical diagnostics, scanning for drugs, chemical warfare agents and explosives in the military and for airport safety controls [109, 115, 117, 383-388]. Here it was shown that the two metabolic conditions of aerobic respiration and aerobic fermentation could be distinguished in yeast using an MCC-IMS. In the beginning, there were problems with too high ethanol concentrations in the MCC-IMS, these problems could be solved by introducing a shorter sample loop and changing the MCC in cooperation with the B&S Analytik (Dortmund Germany). In its final state, the MCC- IMS could successfully distinguish different metabolic conditions in yeast. This has been reported before, but never with an MCC-IMS [61, 62, 185]. Through its high sensitivity the device was even able to detect metabolic changes prior to an industrial ethanol sensor. Additionally, to its fast detection, the measurement is also more robust since two reporter metabolites that are produced early during the metabolic change, can be used to predict the alteration in metabolism. As shown by Vautz and Baumbach in 2008, the IMS is also suitable for process control [389].

Laboratory vs. industrial conditions Laboratory and industrial conditions are hardly comparable, yet studies are mostly performed under laboratory conditions. Here, both were investigated and compared. The metabolic shift could be detected in both conditions. While the laboratory strain S. cerevisiae CEN.PK 113-7D was tested in 108 General discussion and outlook minimal salt medium as well as in industrial molasses medium, the industrial strain S. cerevisiae DHW was solely tested under industrial conditions. All tests were carried out in pilot scale with a volume for the fed batch of 8-11 L. Distinguishing between the metabolic states aerobic respiration and aerobic fermentation was possible in all conditions. This is based on the robust measurements that did not only yield one, but two distinct metabolic reporters at an early stage. The advantage of two reporters is that the system becomes more stable, if for example the drift gas flask is dirty and one reporter cannot be detected anymore, it is likely that the other reporter will still show up. In addition to the volatile space, the transcriptome of the cells was also investigated to compare the two conditions. Here it turned out that the gene regulation of the two strains was essentially similar with exceptions in leucine and isoleucine metabolism, the TCA cycle and the respiratory pathway. This is especially interesting for industrial yeast production, since it indicates that the Crabtree effect might not be as significant in industrial yeast fermentations as it is in the laboratory.

109

Chapter 4

Outlook

The results generated in this thesis might contribute to future studies. First of all, the expansion of the metabolic model will be helpful for studies concerning the metabolic network of yeast. The yeast community will benefit from these new pathways especially if solutions need to be found to produce large amounts of volatile organic compounds. However, for a rich aroma it is important to take into consideration that not only one compound is responsible. Therefore, further investigations need to be performed to find out what combination of metabolites will produce a pleasant aroma. Once the aroma is constructed, metabolically engineered yeasts could be used to produce the flavor and positively alter the aroma of a finished product. The final strains will have to be adjusted repeatedly, to ensure that the intended flavor is achieved in the final product. The data generated using the SESI-Orbitrap is very rich. In this thesis, the data has been evaluated for effects during the Crabtree effect, however due to the high time resolution and excellent mass resolution of the single time points, it is possible to gain even more insight without the need of further experiments. For example, it should be possible to detect enzymatic activity based on mass transfers and the accompanying increase respectively decrease of masses involved in the enzymatic reaction. In the best case, it is even possible to determine reaction rates based on this dataset. The here identified “reporter” volatile compounds could be interesting in industrial processes that need to monitor yeast fermentations, since the reporters can indicate the change of the metabolic state. The system could be checked online for example using the MCC-IMS described in chapter 2.3 and based on the desired product the feed could be adjusted to gain biomass or the production of ethanol. However, the use of the MCC-IMS is not only limited to yeast fermentations, using this thesis as blueprint, it would also be possible to adjust the systems to other industrial processes. In addition, MCC-IMS data would enable yeast producers to detect contaminations in very low concentrations and early in an industrial process and contribute to avoid unnecessary costs.

113

Appendix

Matlab Scripts for the evaluation of SESI-Orbitrap data

Driver:

%% Script for loading the data and running the analysis automatically for the file that equals n % Order as indicated in Exps %Creates two clustergrams, the first shows all masses that are changing between the glucose pulse %and the rise of the ethanol signal, the second shows all masses that rise after the increase of the %ethanol signal %% cd('C:\Measurements'); Data = load('Filename'); Exps = {'id20160513_Mini6TestR1'; 'id20160513_Mini6TestR4'; 'id20160513_Mini6TestR3'; 'id20160517_WachstumCENR1Puls1'; 'id20160518_WachstumCCENR2Puls110muL'; 'id20160518_WachstumVHR4Puls1_100muL'; 'id20160519_PulsVHR3mit100mgGlc_160518172415'; 'id20160519_PulsCENR1mit100mgGlc_160518172415'; 'id20160520_PulsVHR4_100mgGLc_160518172415'; 'id20160520_PulsCENR2_100mgGLc_160518172415'}; PulsTimes = [23.4; 13.5; 20.1; 21.1; 50.4; 58; 40.5; 50; 44.5; 45]/60; % Conversion of minutes to hours for each timepoint after that the glucose pulse was applied ExpNames = fieldnames(Data.mz);n=4 [ClustergramMaxInInterval, ClustergramMaxPastInterval,... resi, mz, tt, tPuls, tEtOHSig] = Analyse_Glc_Puls(Data,... Exps(n), PulsTimes(n)); PosEtOH = find(roundn(Data.mz.(Exps[96]), -3) == 93.0910); PosPuls = min(find(roundn(Data.tiempo.(Exps[96]), -3) > roundn(PulsTimes(n),-3))); time = transpose(Data.tiempo.(Exps{n}))*60;

Analysis: function [ClustergramMaxInInterval, ClustergramMaxPastInterval, resi, mz, tt, tPuls, tEtOHSig] = Analyse_Glc_Puls(Data, Exps, PulsTimes) %% Script for data analysis of high resolution SESI-Orbitrap data % we are searching for a timeseries the slope of which rises or falls sooner over the threshold than that of EtOH % rawdata time in h %% Res = struct(); for e = 1: length(Exps) PosEtOH = find(roundn(Data.mz.(Exps[79]), -3) == 93.0910); tPuls = min(find(roundn(Data.tiempo.(Exps{e}), -3) > roundn(PulsTimes(e),-3))); % deltas between 2 data points RawData = Data.tt.(Exps{e}); %% smoothing of the data using “moving average” and “sgolay” MyNormedData = medfilt1(RawData(:,:),27,[],2); DeltaSig = diff(MyNormedData,1,2); tEtOHSig = min(find(DeltaSig(PosEtOH,:) > max(DeltaSig(PosEtOH,1:tPuls)))); minEtOHSig = MyNormedData(PosEtOH, tEtOHSig+1); % detections threshold MyNormedData(abs(MyNormedData) < minEtOHSig) = 0; %% finds signals that have larger changes during the measurements than the minimal EtOH % signal MinMaxSignalPrePuls = iqr(MyNormedData(:,1:tPuls),2); MinMaxSignalPostPuls = iqr(MyNormedData(:,PosEtOH:end),2); RelevantSignals = abs(MinMaxSignalPrePuls - MinMaxSignalPostPuls) > minEtOHSig; %create variable mz for plotting mz=Data.mz.(Exps{e}); tt=Data.tt.(Exps{e}); res = zeros(size(MyNormedData)); for i = 1: size(res,1) res(i,:) = sgolayfilt(MyNormedData(i,:),1,35); end MyNormedData=res; resi = zeros(size(MyNormedData)); for i = 1: size(MyNormedData,1) xmin = min(MyNormedData(i,:),[],2); xmax = max(MyNormedData(i,:),[],2); xdiff2 = xmax - xmin; for o= 1: size(MyNormedData,2) xi = MyNormedData(i,o); xdiff1 = xi-xmin; 117 Appendix

xdiff = xdiff1 / xdiff2; resi(i,o) = xdiff ; end end %% find relevant signals that max between tpulse and t EtOH MaxInInerval = max(resi(:,tPuls:tEtOHSig),[],2) > max(resi(:,tEtOHSig:end),[],2); MinInInerval = min(resi(:,tPuls:tEtOHSig),[],2) < min(resi(:,tEtOHSig:end),[],2); x = find(RelevantSignals & MaxInInerval); lables = cellstr(num2str(Data.mz.(Exps{e}))); lables = cellstr(num2str([1:length(Data.mz.(Exps{e}))]')); ClustergramMaxInInterval.(Exps{e}) = clustergram(resi(x, tPuls:tEtOHSig),'RowLabels',lables(x),'Cluster', 'Column','LogTrans','false'); %% Find signal with max after t EtOH signal but have a signal before t EtOH MaxPostEtOH = max(resi(:,tEtOHSig:end),[],2) > max(resi(:,tPuls:tEtOHSig),[],2); x = find(RelevantSignals & MaxPostEtOH); lables = cellstr(num2str(Data.mz.(Exps{e}))); lables = cellstr(num2str([1:length(Data.mz.(Exps{e}))]')); ClustergramMaxPastInterval.(Exps{e}) = clustergram(resi(x, tPuls:tEtOHSig),'RowLabels',lables(x),'Cluster', 'Column','LogTrans','false'); end end

Plot:

%x = vector of masses that will be plotted against ethanol X=[57.0700 58.0735 69.0701 70.0732 71.0857 72.0887 73.0649 74.0680 118.0652 119.0885 127.1117 128.1152 166.9860 177.0763 196.1123]'; for i = 1: size(X,1) figure;plot((time),resi(PosEtOH,:),'k','LineWidth', l) hold on %PlotX a = find(roundn(mz,-4) == X(i,:)); tPulsi=time(tPuls) %transforms time value from column number to value tEtOHSigi=time(tEtOHSig) %transforms time value from column number to value b=mz(a,:); plot((time),resi(a,:),'LineWidth', l) plot([tPulsi;tPulsi],[0,1],'k','LineWidth', l) plot([tEtOHSigi;tEtOHSigi],[0,1],'k','LineWidth', l) legend('Ethanol',num2str(b)) xlabel('time [min]','fontsize', 15) ylabel('normalized data','fontsize', 15) set(gca,'fontsize',15) %xlim([40 90]); %sets x axis to the wanted time figure;plot(tt(a,:),'LineWidth', l)

Correlation:

%% a and b are the m/z fragments that are to compare a = find(roundn(Data.mz.(Exps{n}),-4) == 89.0962); b = find(roundn(Data.mz.(Exps{n}),-4) == 71.0857); %normalization of the data xmina = min(Data.tt.(Exps{n})(a,:),[],2); xmaxa = max(Data.tt.(Exps{n})(a,:),[],2); xdiff2 = xmaxa - xmina; for o= 1: size(Data.tt.(Exps{n}),2) xi = Data.tt.(Exps{n})(a,o); xdiff1 = xi-xmina; xdiffa = xdiff1 / xdiff2; resia(1,o) = xdiffa ; end xminb = min(Data.tt.(Exps{n})(b,:),[],2); xmaxb = max(Data.tt.(Exps{n})(b,:),[],2); xdiff2 = xmaxb - xminb; for o= 1: size(Data.tt.(Exps{n}),2) xi = Data.tt.(Exps{n})(b,o); xdiff1 = xi-xminb; xdiffb = xdiff1 / xdiff2; resib(1,o) = xdiffb ; end 118 Appendix

%plot the data figure;plot(resi(PosEtOH,:),'k') hold on plot (resia, 'g') plot (resib, 'r') plot([tPuls;tPuls],[0,1],'k') plot([tEtOHSig;tEtOHSig],[0,1],'k') cx = corr2(resi(a,:),resi(b,:)) cxy = num2str(cx) %axis labeling and legend c=mz(a,:); d=mz(b,:); legend('Ethanol',num2str(c), num2str(d),'corr coeff: ', cxy) xlabel('time [min]') ylabel('normalized data')

119 Appendix

Supplementary data

Supplementary table 1: FDR Values of transcript level changes between early and late group. FDR- LS describes the FDR values in the laboratory strain, while FDR-IS describes the FDR value in the industrial strain. Name FDR-LS [%] FDR-IS [%] SDH4 0,12 0,04 SDH3 0,08 10,40 SDH2 3,65 34,23 SDH1 0,36 0,61 YJL045W 0,02 1,49 NDE2 0,02 0,10 NDI1 10,80 58,28 NDE1 93,21 10,13 QCR10 3,73 15,58 QCR9 0,70 10,40 QCR8 1,84 24,29 QCR7 6,75 28,01 QCR6 5,36 15,56 QCR2 0,92 0,93 COR1 0,08 1,79 CYT1 0,34 1,15 RIP1 0,20 5,43 COB 100,00 88,52 COX11 10,45 12,38 COX19 7,51 37,47 COX16 0,01 60,03 COX23 0,50 3,09 COX18 83,57 45,83 COX20 0,02 76,23 COX5B 0,09 53,37 PYC1 0,06 9,93 PYC2 11,27 15,10 CIT3 0,01 0,03 CIT1 0,05 48,23 ACO1 1,02 3,11 ACO2 0,09 45,97 IDH1 29,01 5,04 IDH2 30,80 12,36 LPD1 2,83 60,74 KGD1 0,02 14,72 KGD2 0,09 35,94 LSC1 0,03 1,42 LSC2 0,00 11,41 MDH1 13,22 2,45 ILV2 0,51 2,26 120 Appendix

ILV6 10,92 3,71 ILV5 17,92 55,02 ILV3 0,32 0,56 LEU9 0,01 87,15 LEU4 0,01 0,20 LEU1 0,00 1,71 LEU2 16,98 24,35 BAT1 0,05 19,27 BAT2 0,18 0,51 CHA1 89,04 65,61 ILV1 0,00 51,96

121

References

1. Ebert, B.E., C. Halbfeld, and L.M. Blank, Exploration and exploitation of the yeast volatilome. Current Metabolomics, 2016. 4: p. 1-17. 2. Hazelwood, L.A., et al., The Ehrlich pathway for fusel alcohol production: a century of research on Saccharomyces cerevisiae metabolism. Applied and Environmental Microbiology, 2008. 74(8): p. 2259-2266. 3. Pires, E.J., et al., Yeast: the soul of beer's aroma - a review of flavour-active esters and higher alcohols produced by the brewing yeast. Applied Microbiology and Biotechnology, 2014. 98(5): p. 1937-49. 4. Carrau, F.M., et al., De novo synthesis of monoterpenes by Saccharomyces cerevisiae wine yeasts. FEMS Microbiology Letters, 2005. 243(1): p. 107-15. 5. Baumbach, J.I. and G.A. Eiceman, Ion mobility spectrometry: Arriving on site and moving beyond a low profile. Applied Spectroscopy, 1999. 53(9): p. 338a-355a. 6. Halbfeld, C., B. Ebert, and L. Blank, Multi-capillary column-ion mobility spectrometry of volatile metabolites emitted by Saccharomyces cerevisiae. Metabolites, 2014. 4(3): p. 751-774. 7. Schmidt, R., et al., Volatile affairs in microbial interactions. The ISME Journal, 2015. 9(11): p. 2329-2335. 8. Dulac, C. and A.T. Torello, Molecular detection of pheromone signals in mammals: from genes to behaviour. Nature Reviews: Neuroscience, 2003. 4(7): p. 551-62. 9. Cho, I.H. and D.G. Peterson, Chemistry of bread aroma: a review. Food Science and Biotechnology, 2010. 19(3): p. 575-582. 10. Baardseth, P., et al., The effects of bread making process and wheat quality on french baguettes. Journal of Cereal Science, 2000. 32(1): p. 73-87. 11. MartinezAnaya, M.A., Enzymes and bread flavor. Journal of Agricultural and Food Chemistry, 1996. 44(9): p. 2469-2480. 12. Frasse, P., et al., The influence of fermentation on volatile compounds in french bread dough. Food Science and Technology-Lebensmittel-Wissenschaft & Technologie, 1993. 26(2): p. 126- 132. 13. Bertram, G.L., Studies on crust color .1. The importance of the browning reaction in determining the crust color of bread. Cereal Chemistry, 1953. 30(2): p. 127-139. 14. Barnes, H.M. and C.W. Kaufman, Industrial aspects of browning reaction. Industrial and Engineering Chemistry, 1947. 39(9): p. 1167-1170. 15. Zentner, H., Influence of glycine and lysine on dough fermentation. Journal of the Science of Food and Agriculture, 1961. 12(12): p. 812. 16. Rooney, L.W., A. Salem, and J.A. Johnson, Studies of carbonyl compounds produced by sugar- amino acid reactions .I. Model systems. Cereal Chemistry, 1967. 44(5): p. 539. 17. Salem, A., L.W. Rooney, and J.A. Johnson, Studies of carbonyl compounds produced by sugar- amino acid reactions .2. In bread systems. Cereal Chemistry, 1967. 44(6): p. 576. 18. Chang, C.Y., L.M. Seitz, and E. Chambers, Volatile flavor components of breads made from hard red winter-wheat and hard white winter-wheat. Cereal Chemistry, 1995. 72(3): p. 237-242. 19. Kirchhoff, E. and P. Schieberle, Determination of key aroma compounds in the crumb of a three- stage sourdough rye bread by stable isotope dilution assays and sensory studies. Journal of Agricultural and Food Chemistry, 2001. 49(9): p. 4304-4311. 20. Schieberle, P. and W. Grosch, Changes in the concentrations of potent crust odourants during storage of white bread. Flavour and Fragrance Journal, 1992. 7(4): p. 213-218. 21. Zehentbauer, G. and W. Grosch, Crust aroma of baguettes - I. Key odorants of baguettes prepared in two different ways. Journal of Cereal Science, 1998. 28(1): p. 81-92. 22. Zehentbauer, G. and W. Grosch, Crust aroma of baguettes II. Dependence of the concentrations of key odorants on yeast level and dough processing. Journal of Cereal Science, 1998. 28(1): p. 93-96. 23. Heenan, S.P., et al., Characterisation of fresh bread flavour: Relationships between sensory characteristics and volatile composition. Food Chemistry, 2009. 116(1): p. 249-257. 125 References

24. Welke, J.E., et al., Characterization of the volatile profile of Brazilian Merlot wines through comprehensive two dimensional gas chromatography time-of-flight mass spectrometric detection. Journal of Chromatography A, 2012. 1226: p. 124-139. 25. Swiegers, J.H., et al., Yeast and bacterial modulation of wine aroma and flavour. Australian Journal of Grape and Wine Research, 2005. 11(2): p. 139-173. 26. Patrignani, F., et al., Production of volatile and sulfur compounds by 10 Saccharomyces cerevisiae strains inoculated in Trebbiano must. Frontiers in Microbiology, 2016. 7: p. 243. 27. Mateo, J.J., et al., Yeast starter cultures affecting wine fermentation and volatiles. Food Research International, 2001. 34(4): p. 307-314. 28. Schreier, P., Flavor composition of wines: a review. CRC Critical Reviews in Food Science and Nutrition, 1979. 12(1): p. 59-111. 29. Boulton, R.B., Principles and practices of winemaking. The Chapman & Hall enology library. 1996, New York: Chapman & Hall. xiv, 604 p. 30. Rapp, A., Volatile flavour of wine: correlation between instrumental analysis and sensory perception. Nahrung, 1998. 42(6): p. 351-63. 31. Rapp, A. and H. Mandery, Wine aroma. Experientia, 1986. 42(8): p. 873-884. 32. Picard, M., et al., Identification of piperitone as an aroma compound contributing to the positive mint nuances perceived in aged red Bordeaux wines. Journal of Agricultural and Food Chemistry, 2016. 64(2): p. 451-460. 33. Capone, D.L., et al., Evolution and occurrence of 1,8-cineole (eucalyptol) in australian wine. Journal of Agricultural and Food Chemistry, 2011. 59(3): p. 953-959. 34. Rapp, A., et al., Fremdkomponenten im Aroma von Trauben und Weinen interspezifischer Rebsorten. Vitis, 1980. 19: p. 13-23. 35. Rapp, A., Aromastoffe des Weines. Chemie in unserer Zeit, 1992. 6: p. 273-284. 36. Rapp, A., P. Pretorius, and D. Kugler, Foreign and undesirable flavours in wine, in Developments in Food Science, C. George, Editor. 1992, Elsevier. p. 485-522. 37. Romano, F., et al., Approccio psicofisico allanalisi sensoriale dei vini, in The Aromatic Substances in Grapes and Wines, A. Scienza, G. Versini, and M. all´Adige, Editors. 1989. p. 427-440. 38. Augustyn, O.P.H., A. Rapp, and C.J. van Wyk, Some volatile aroma components of Vitis vinifera L. cv. Sauvignon blanc. South African Journal of Enology and Viticulture, 1982(3): p. 53-59. 39. Bayonove, C., C. Cordonnier, and P. Dubois, etude d´une fraction caractéristique de l'arôme du raisin de la variete cabernet sauvignon mise en evidence de la 2-methoxy-3-isobutylpyrazine. Comptes Rendus de l'Académie des Sciences, 1975(281): p. 75-78. 40. Snowdon, E.M., et al., Mousy off-flavor: a review. Journal of Agricultural and Food Chemistry, 2006. 54(18): p. 6465-6474. 41. Strauss, C.R., 2-Acetyltetrahydropyridines a cause of the “mousy” taint in wine. Chemistry and Industry, 1984: p. 109-110. 42. Heresztyn, T., Formation of substituted tetrahydropyridines by species of Brettanomyces and Lactobacillus isolated from mousy wines . American Journal of Enology and Viticulture, 1986. 37(2): p. 127-132. 43. Herderich, M., The occurrence of 2-acetyl-1-pyrroline in mousy wines. Natural Product Letters, 1995. 7: p. 129-132. 44. Grbin, P.R., P.J. Costello, and M. Herderich, Developments in the sensory, chemical and microbiological basis of mousy taint in wine. Sonoma County Wine Library. 1995: Proceedings of the Ninth Australian Wine Industry Technical Conference. 45. Grbin, P.R. and P.A. Henschke, Mousy off-flavour production in grape juice and wine by Dekkera and Brettanomyces yeasts. Australian Journal of Grape and Wine Research, 2000. 6(3): p. 255-262. 46. Tucknott, O.G., Taints in fermented juice products: mousy taint in cider, in Annual Report 1977. 1978: University of Bristol, 1978. 47. Rapp, A. and G. Versini, Flüchtige phenolische Verbindungen in Wein. Deutsche Lebensmittel- Rundschau, 1996. 92(2): p. 42-48. 126 References

48. Sefton, M.A. and R.F. Simpson, Compounds causing cork taint and the factors affecting their transfer from natural cork closures to wine - a review. Australian Journal of Grape and Wine Research, 2005. 11(2): p. 226-240. 49. The chemistry of wine: stabilization and treatments. Handbook of Enology, ed. P. Ribereau- Gayon, et al. Vol. 2. 2006: Wiley. 50. Riu, M., et al., Quantification of chloroanisoles in cork using headspace solid-phase microextraction and gas chromatography with electron capture detection. Journal of Chromatography A, 2006. 1107(1-2): p. 240-7. 51. Vieira, P., S.M. Rocha, and A.J.D. Silvestre, Simultaneous headspace solid phase microextraction analysis of off-flavour compounds from Quercus suber L. cork. Journal of the Science of Food and Agriculture, 2007. 87(4): p. 632-640. 52. Vlachos, P., et al., Matrix effect during the application of a rapid method using HS-SPME followed by GC-ECD for the analysis of 2,4,6-TCA in wine and cork soaks. Food Chemistry, 2007. 105(2): p. 681-690. 53. Fischer, U., Wine aroma, in Flavours and Fragrances, R.G. Berger, Editor. 2007, Springer-Verlag: Berlin Heidelberg. p. 241-267. 54. Christoph, N. and C. Bauer-Christoph, Falvour and spirit drinks: raw materials, fermentation, destillation, and ageing, in Flavours and Fragrances, R.G. Berger, Editor. 2007, Springer: Berlin- Heidelberg. p. 219-239. 55. Morath, S.U., R. Hung, and J.W. Bennett, Fungal volatile organic compounds: a review with emphasis on their biotechnological potential. Fungal Biology Reviews, 2012. 26(2-3): p. 73-83. 56. Kanchiswamy, C.N., M. Mainoy, and M.E. Maffei, Chemical diversity of microbial volatiles and their potential for plant growth and productivity. Frontiers in Plant Science, 2015. 6. 57. Alpha, C.J., et al., Mycofumigation by the volatile organic compound-producing fungus Muscodor albus induces bacterial cell death through DNA damage. Applied and Environmental Microbiology, 2015. 81(3): p. 1147-1156. 58. Bitas, V., et al., Sniffing on microbes: diverse roles of microbial volatile organic compounds in plant health. Molecular Plant-Microbe Interactions, 2013. 26(8): p. 835-43. 59. Hertel, M., et al., Detection of signature volatiles for cariogenic microorganisms. European Journal of Clinical Microbiology & Infectious Diseases, 2016. 35(2): p. 235-44. 60. Piechulla, B. and J. Degenhardt, The emerging importance of microbial volatile organic compounds. Plant, Cell & Environment, 2014. 37(4): p. 811-2. 61. Bruce, A., et al., Production of volatile organic compounds by Trichoderma in media containing different amino acids and their effect on selected wood decay fungi. Holzforschung, 2000. 54(5): p. 481-486. 62. Blom, D., et al., Production of plant growth modulating volatiles is widespread among rhizosphere bacteria and strongly depends on culture conditions. Environmental Microbiology, 2011. 13(11): p. 3047-3058. 63. Maffei, M.E., J. Gertsch, and G. Appendino, Plant volatiles: production, function and pharmacology. Natural Product Reports, 2011. 28(8): p. 1359-1380. 64. Arn, H. and T.E. Acree, Flavornet: a database of aroma compounds based on odor potency in natural products, in Developments in Food Science, C.T.H.C.J.M.T.H.P.F.S. E.T. Contis and A.M. Spanier, Editors. 1998, Elsevier: Amsterdam. p. 27. 65. de Lacy Costello, B., et al., A review of the volatiles from the healthy human body. Journal of Breath Research, 2014. 8(1): p. 014001. 66. Lemfack, M.C., et al., mVOC: a database of microbial volatiles. Nucleic Acids Research, 2014. 42(D1): p. D744-D748. 67. Simó, C., A. Cifuentes, and V. García-Cañas, Fundamentals of advanced omics technologies: from genes to metabolites. 1st ed. Comprehensive analytical chemistry. 2014, Amsterdam; Boston: Elsevier. 467 pages. 68. Lim, F.Y., et al., Toward awakening cryptic secondary metabolite gene clusters in filamentous fungi. Methods in Enzymology, 2012. 517: p. 303-324. 127 References

69. Breitling, R., et al., Metabolomics for secondary metabolite research. Metabolites, 2013. 3(4): p. 1076-1083. 70. Spraker, J. and N. Keller, Waking sleeping pathways in filamentous fungi, in Natural Products. 2014, John Wiley & Sons, Inc. p. 277-292. 71. Claeson, A.S., M. Sandstrom, and A.L. Sunesson, Volatile organic compounds (VOCs) emitted from materials collected from buildings affected by microorganisms. Journal of Environmental Monitoring, 2007. 9(3): p. 240-245. 72. Korpi, A., J. Jarnberg, and A.L. Pasanen, Microbial volatile organic compounds. Critical Reviews in Toxicology, 2009. 39(2): p. 139-193. 73. Herwig, C. and U. von Stockar, A small metabolic flux model to identify transient metabolic regulations in Saccharomyces cerevisiae. Bioprocess and Biosystems Engineering, 2002. 24(6): p. 395- 403. 74. Ferreira, B.S., et al., Recombinant Saccharomyces cerevisiae strain triggers acetate production to fuel biosynthetic pathways. Journal of Biotechnology, 2004. 109(1-2): p. 159-167. 75. Miyake, T. and T. Shibamoto, Quantitative analysis of acetaldehyde in foods and beverages. Journal of Agricultural and Food Chemistry, 1993. 41(11): p. 1968-1970. 76. Margalith, P., Flavor microbiology. 1st ed. 1981, Springfield, Ill.: Thomas. xiii, 309 p. 77. Cheraiti, N., F.X. Sauvage, and J.M. Salmon, Acetaldehyde addition throughout the growth phase alleviates the phenotypic effect of zinc deficiency in Saccharomyces cerevisiae. Applied Microbiology and Biotechnology, 2008. 77(5): p. 1093-1109. 78. Becher, P.G., et al., Yeast, not fruit volatiles mediate Drosophila melanogaster attraction, oviposition and development. Functional Ecology, 2012. 26(4): p. 822-828. 79. Ehrlich, F., Über die bedingungen der Fuselölbildung und über ihren Zusammenhang mit dem Eiweißaufbau der Hefe. Berichte der deutschen chemischen Gesellschaft, 1907. 40(1): p. 1027- 1047. 80. Styger, G., B. Prior, and F.F. Bauer, Wine flavor and aroma. Journal of Industrial Microbiology & Biotechnology, 2011. 38(9): p. 1145-1159. 81. Styger, G., et al., Genetic analysis of the metabolic pathways responsible for aroma metabolite production by Saccharomyces cerevisiae. Applied Microbiology and Biotechnology, 2013. 97(10): p. 4429-4442. 82. Van Dijken, J.P. and W.A. Scheffers, Redox balances in the metabolism of sugars by yeasts. FEMS Microbiology Letters, 1986. 32(3-4): p. 199-224. 83. Park, S.-H., S. Kim, and J.-S. Hahn, Metabolic engineering of Saccharomyces cerevisiae for the production of isobutanol and 3-methyl-1-butanol. Applied Microbiology and Biotechnology, 2014. 98(21): p. 9139-9147. 84. Chen, E.C.H., The relative contribution of Ehrlich and biosynthetic pathways to the formation of fusel alcohols. Journal of the American Society of Brewing Chemists, 1978. 36(1): p. 39-41. 85. Saerens, S.M., et al., Production and biological function of volatile esters in Saccharomyces cerevisiae. Microbial Biotechnology, 2010. 3(2): p. 165-177. 86. Malcorps, P. and J.P. Dufour, Short-chain and medium-chain aliphatic-ester synthesis in Saccharomyces cerevisiae. European Journal of Biochemistry, 1992. 210(3): p. 1015-1022. 87. Vadali, R.V., G.N. Bennett, and K.Y. San, Applicability of CoA/acetyl-CoA manipulation system to enhance isoamyl acetate production in Escherichia coli. Metabolic Engineering, 2004. 6(4): p. 294-299. 88. Cordente, A.G., et al., Modulating aroma compounds during wine fermentation by manipulating carnitine acetyltransferases in Saccharomyces cerevisiae. FEMS Microbiology Letters, 2007. 267(2): p. 159-166. 89. Dillemans, M., et al., The amplification effect of the ILV5 gene on the production of vicinal diketones in Saccharomyces cerevisiae. Journal of the American Society of Brewing Chemists, 1987. 45: p. 81.

128 References

90. Krogerus, K. and B.R. Gibson, Influence of valine and other amino acids on total diacetyl and 2,3-pentanedione levels during fermentation of brewer's wort. Applied Microbiology and Biotechnology, 2013. 97(15): p. 6919-30. 91. Alves, Z., et al., Exploring the Saccharomyces cerevisiae volatile metabolome: indigenous versus commercial strains. PloS One, 2015. 10(11): p. e0143641. 92. Penuelas, J., et al., Biogenic volatile emissions from the soil. Plant, Cell & Environment, 2014. 37(8): p. 1866-1891. 93. Lambrechts, M.G. and I.S. Pretorius, Yeast and its importance for wine aroma - a review. South African Journal for Enology and Viticulture, 2000. 21: p. 97-129. 94. Tahara, S., K. Fujiwara, and J. Mizutani, Metabolites of Sporobolomyces odorus. 2. Neutral constituents of volatiles in cultured broth of Sporobolomyces odorus. Agricultural and Biological Chemistry, 1973. 37(12): p. 2855-2861. 95. Wache, Y., et al., Role of beta-oxidation enzymes in gamma-decalactone production by the yeast Yarrowia lipolytica. Applied and Environmental Microbiology, 2001. 67(12): p. 5700-5704. 96. Nordström, K., Formation of esters from acids by Brewer‘s yeast IV. Effect of higher fatty acids and toxicity of lower fatty acids. Journal of the Institute of Brewing, 1964. 70(3): p. 233-242. 97. Schewe, H., et al., Biotechnological production of terpenoid flavour and fragrance compounds in tailored bioprocesses, in Developments in Food Science, L.P.B. Wender and P. Mikael Agerlin, Editors. 2006, Elsevier: Amsterdam. p. 45-48. 98. Rottava, I., et al., Microbial oxidation of (-)-alpha-pinene to verbenol production by newly isolated strains. Applied Biochemistry and Biotechnology, 2010. 162(8): p. 2221-2231. 99. King, A. and J. Richard Dickinson, Biotransformation of monoterpene alcohols by Saccharomyces cerevisiae, Torulaspora delbrueckii and Kluyveromyces lactis. Yeast, 2000. 16(6): p. 499-506. 100. Carrau, F., E. Boido, and E. Dellacassa, Yeast diversity and flavor compounds, in Fungal Metabolites, J.-M. Mérillon and G.K. Ramawat, Editors. 2016, Springer International Publishing: Cham. p. 1-29. 101. Dzubak, P., et al., Pharmacological activities of natural triterpenoids and their therapeutic implications. Natural Products Reports, 2006. 23(3): p. 394-411. 102. Rodriguez, J.M., et al., Fungal metabolic model for 3-methylcrotonyl-CoA carboxylase deficiency. Journal of Biological Chemistry, 2004. 279(6): p. 4578-4587. 103. Moss, J. and M.D. Lane, The biotin-dependent enzymes. Advances in Enzymology and Related Areas of Molecular Biology, 1971. 35: p. 321-442. 104. Anderson, M.D., et al., 3-Methylcrotonyl-coenzyme A carboxylase is a component of the mitochondrial leucine catabolic pathway in plants. Plant Physiology, 1998. 118(4): p. 1127-1138. 105. Oswald, M., et al., Monoterpenoid biosynthesis in Saccharomyces cerevisiae. FEMS Yeast Research, 2007. 7(3): p. 413-421. 106. von Reuss, S.H., et al., Octamethylbicyclo[3.2.1]octadienes from the rhizobacterium Serratia odorifera. Angewandte Chemie, International Edition, 2010. 49(11): p. 2009-2010. 107. Weise, T., et al., VOC emission of various Serratia species and isolates and genome analysis of Serratia plymuthica 4Rx13. FEMS Microbiology Letters, 2014. 352(1): p. 45-53. 108. Cumeras, R., et al., Review on Ion Mobility Spectrometry. Part 2: hyphenated methods and effects of experimental parameters. Analyst, 2015. 140(5): p. 1391-1410. 109. Mäkinen, M.A., O.A. Anttalainen, and M.E.T. Sillanpää, Ion Mobility Spectrometry and Its Applications in Detection of Chemical Warfare Agents. Analytical Chemistry, 2010. 82(23): p. 9594-9600. 110. Cable, A. Some aspects of the use of intelligent systems engineering in the design of airport security programmes. in Intelligent Systems Engineering, 1992., First International Conference on (Conf. Publ. No. 360). 1992. IET. 111. Eiceman, G.A., et al., Separation of ions from explosives in differential mobility spectrometry by vapor-modified drift gas. Analytical Chemistry, 2004. 76(17): p. 4937-4944.

129 References

112. Westhoff, M., et al., Ion mobility spectrometry: a new method for the detection of lung cancer and airway infection in exhaled air? First results of a pilot study. CHEST Journal, 2005. 128(4_MeetingAbstracts): p. 155S-a-155S. 113. Westhoff, M., et al., Ion mobility spectrometry for the detection of volatile organic compounds in exhaled breath of patients with lung cancer: results of a pilot study. Thorax, 2009. 64(9): p. 744-8. 114. Ruzsanyi, V. and J. Baumbach, Analysis of human breath using IMS. International Journal for Ion Mobility Spectrometry, 2005. 8: p. 5-7. 115. Baumbach, J.I., Process analysis using ion mobility spectrometry. Analytical and Bioanalytical Chemistry, 2006. 384(5): p. 1059-70. 116. Collins, D. and M. Lee, Developments in ion mobility spectrometry–mass spectrometry. Analytical and Bioanalytical Chemistry, 2002. 372(1): p. 66-73. 117. Cumeras, R., et al., Review on ion mobility spectrometry. Part 1: current instrumentation. Analyst, 2015. 140(5): p. 1376-90. 118. Hill, H.H. and G. Simpson, Capabilities and limitations of ion mobility spectrometry for field screening applications. Field Analytical Chemistry and Technology, 1997. 1(3): p. 119-134. 119. Tabrizchi, M., T. Khayamian, and N. Taj, Design and optimization of a corona discharge ionization source for ion mobility spectrometry. Review of Scientific Instruments, 2000. 71(6): p. 2321-2328. 120. Leasure, C.S., et al., Photoionization in air with ion obility spectrometry using a hydrogen discharge lamp. Analytical Chemistry, 1986. 58(11): p. 2142-2147. 121. Sielemann, S., et al., Detection of alcohols using UV-ion mobility spetrometers. Analytica Chimica Acta, 2001. 431(2): p. 293-301. 122. Gillig, K.J., et al., Coupling high-pressure MALDI with ion mobility/orthogonal time-of flight mass spectrometry. Analytical Chemistry, 2000. 72(17): p. 3965-3971. 123. Shumate, C.B. and H.H. Hill, Coronaspray nebulization and ionization of liquid samples for ion mobility spectrometry. Analytical Chemistry, 1989. 61(6): p. 601-606. 124. Bradbury, N.E. and R.A. Nielsen, Absolute values of the electron mobility in hydrogen. Physical Review, 1936. 49(5): p. 388. 125. Tyndall, A.M. and C.F. Powell, The mobility of ions in pure gases. Proceedings of the Royal Society of London Series A, 1930. 129(809): p. 162-180. 126. Revercomb, H. and E.A. Mason, Theory of plasma chromatography/gaseous electrophoresis. Review. Analytical Chemistry, 1975. 47(7): p. 970-983. 127. Hill, H.H., W.F. Siems, and R.H. St. Louis, Ion mobility spectrometry. Analytical Chemistry, 1990. 62(23): p. 1201A-1209A. 128. Stach, J. and J.I. Baumbach, Ion mobility spectrometry-basic elements and applications. International Journal for Ion Mobility Spectrometry, 2002. 5: p. 1-21. 129. Dzidic, I., et al., Comparison of positive ions formed in nickel-63 and corona discharge ion sources using nitrogen, argon, isobutane, ammonia and nitric oxide as reagents in atmospheric pressure ionization mass spectrometry. Analytical Chemistry, 1976. 48(12): p. 1763-1768. 130. Eiceman, G.A., et al., Positive reactant ion chemistry for analytical, high temperature ion mobility spectrometry (IMS): effects of electric field of the drift tube and moisture, temperature, and flow of the drift gas. International Journal for Ion Mobility Spectrometry, 1998. 1: p. 28-37. 131. Sielemann, S., et al., Determination of MTBE next to benzene, toluene and xylene within 90 s using GC/IMS with multi-capillary column. Journal of Integrative Bioinformatics, 2001. 4(2): p. 69-73. 132. Vidal-de-Miguel, G., et al., Low-sample flow secondary electrospray ionization: improving vapor ionization efficiency. Analytical Chemistry, 2012. 84(20): p. 8475-8479. 133. Gaugg, M.T., et al., Expanding metabolite coverage of real-time breath analysis by coupling a universal secondary electrospray ionization source and high resolution mass spectrometry a pilot study on tobacco smokers. Journal of Breath Research, 2016. 10(1): p. 016010.

130 References

134. Barrios-Collado, C.s., et al., Capturing in vivo plant metabolism by real-time analysis of low to high molecular weight volatiles. Analytical Chemistry, 2016. 88(4): p. 2406-2412. 135. Perry, R.H., R.G. Cooks, and R.J. Noll, Orbitrap mass spectrometry: instrumentation, ion motion and applications. Mass Spectrometry Reviews, 2008. 27(6): p. 661-699. 136. Barrios-Collado, C., G. Vidal-de-Miguel, and P.M.-L. Sinues, Numerical modeling and experimental validation of a universal secondary electrospray ionization source for mass spectrometric gas analysis in real-time. Sensors and Actuators B: Chemical, 2016. 223: p. 217-225. 137. Jordan, A., et al., A high resolution and high sensitivity proton-transfer-reaction time-of-flight mass spectrometer (PTR-TOF-MS). International Journal of Mass Spectrometry, 2009. 286(2): p. 122-128. 138. Bunge, M., et al., On-line monitoring of microbial volatile metabolites by proton transfer reaction- mass spectrometry. Applied and Environmental Microbiology, 2008. 74(7): p. 2179-2186. 139. Romano, A., et al., Proton transfer reaction–mass spectrometry: online and rapid determination of volatile organic compounds of microbial origin. Applied Microbiology and Biotechnology, 2015. 99(9): p. 3787-3795. 140. Weise, T., et al., Volatile organic compounds produced by the phytopathogenic bacterium Xanthomonas campestris pv. vesicatoria 85-10. Beilstein Journal of Organic Chemistry, 2012. 8(1): p. 579-596. 141. Critchley, A., et al., The proton transfer reaction mass spectrometer and its use in medical science: applications to drug assays and the monitoring of bacteria. International Journal of Mass Spectrometry, 2004. 239(2): p. 235-241. 142. O'Hara, M. and C.A. Mayhew, A preliminary comparison of volatile organic compounds in the headspace of cultures of Staphylococcus aureus grown in nutrient, dextrose and brain heart bovine broths measured using a proton transfer reaction mass spectrometer. Journal of Breath Research, 2009. 3(2): p. 027001. 143. Zehm, S., et al., Detection of Candida albicans by mass spectrometric fingerprinting. Current Microbiology, 2012. 64(3): p. 271-275. 144. Jaksch, D., et al., The effect of ozone treatment on the microbial contamination of pork meat measured by detecting the emissions using PTR-MS and by enumeration of microorganisms. International Journal of Mass Spectrometry, 2004. 239(2): p. 209-214. 145. Mayr, D., et al., Detection of the spoiling of meat using PTR–MS. International Journal of Mass Spectrometry, 2003. 223: p. 229-235. 146. Silcock, P., et al., Microbially induced changes in the volatile constituents of fresh chilled pasteurised milk during storage. Food Packaging and Shelf Life, 2014. 2(2): p. 81-90. 147. Soukoulis, C., et al., Proton transfer reaction time‐of‐flight mass spectrometry monitoring of the evolution of volatile compounds during lactic acid fermentation of milk. Rapid Communications in Mass Spectrometry, 2010. 24(14): p. 2127-2134. 148. Makhoul, S., et al., Proton‐transfer‐reaction mass spectrometry for the study of the production of volatile compounds by bakery yeast starters. Journal of Mass Spectrometry, 2014. 49(9): p. 850- 859. 149. Makhoul, S., et al., Volatile compound production during the bread-making process: effect of flour, yeast and their interaction. Food and Bioprocess Technology, 2015. 8(9): p. 1925-1937. 150. Capozzi, V., et al., PTR-MS characterization of VOCs associated with commercial aromatic bakery yeasts of wine and beer origin. Molecules, 2016. 21(4): p. 483. 151. Grob, R.L., Theory of gas chromatography, in Modern Practice of Gas Chromatography. 2004, John Wiley & Sons, Inc.: Hoboken, New Jersey. p. 23-63. 152. Colón, L.A. and L.J. Baird, Detectors in modern gas chromatography, in Modern Practice of Gas Chromatography. 2004, John Wiley & Sons, Inc. p. 275-337. 153. Barry, E.F., Columns: packed and capillary; column selection in gas chromatography, in Modern Practice of Gas Chromatography. 2004, John Wiley & Sons, Inc. p. 65-191.

131 References

154. Tranchida, P.Q., et al., Heart-cutting multidimensional gas chromatography: a review of recent evolution, applications, and future prospects. Analytica Chimica Acta, 2012. 716: p. 66-75. 155. Marriott, P. and R. Shellie, Principles and applications of comprehensive two-dimensional gas chromatography. TRAC Trends in Analytical Chemistry, 2002. 21(9–10): p. 573-583. 156. Seeley, J.V. and S.K. Seeley, Multidimensional gas chromatography: fundamental advances and new applications. Analytical Chemistry, 2012. 85(2): p. 557-578. 157. Jagerdeo, E., et al., Analysis of ethyl carbamate in wines using solid-phase extraction and multidimensional gas chromatography/mass spectrometry. Journal of Agricultural and Food Chemistry, 2002. 50(21): p. 5797-5802. 158. van Ruth, S.M., Methods for gas chromatography-olfactometry: a review. Biomolecular Engineering, 2001. 17(4–5): p. 121-128. 159. Zhang, Y., et al., Identification of potent odorants in a novel nonalcoholic beverage produced by fermentation of wort with shiitake (Lentinula edodes). Journal of Agricultural and Food Chemistry, 2014. 62(18): p. 4195-4203. 160. Schmidt, R. and W.S. Cain, Making scents: dynamic olfactometry for threshold measurement. Chemical Senses, 2010. 35(2): p. 109-20. 161. Campo, E., et al., Prediction of the wine sensory properties related to grape variety from dynamic- headspace gas chromatography−olfactometry data. Journal of Agricultural and Food Chemistry, 2005. 53(14): p. 5682-5690. 162. Eiceman, G.A., H.H. Hill, and J. Gardea-Torresdey, Gas chromatography. Analytical Chemistry, 2000. 72(12): p. 137-144. 163. Franc, C., F. David, and G. de Revel, Multi-residue off-flavour profiling in wine using stir bar sorptive extraction–thermal desorption–gas chromatography–mass spectrometry. Journal of Chromatography A, 2009. 1216(15): p. 3318-3327. 164. Perl, T., et al., Detection of characteristic metabolites of Aspergillus fumigatus and Candida species using ion mobility spectrometry–metabolic profiling by volatile organic compounds. Mycoses, 2011. 54(6): p. e828-e837. 165. Aznar, M. and T. Arroyo, Analysis of wine volatile profile by purge-and-trap–gas chromatography–mass spectrometry: Application to the analysis of red and white wines from different Spanish regions. Journal of Chromatography A, 2007. 1165(1–2): p. 151-157. 166. Verstrepen, K.J., et al., Expression levels of the yeast alcohol acetyltransferase genes ATF1, Lg- ATF1, and ATF2 control the formation of a broad range of volatile esters. Applied and Environmental Microbiology, 2003. 69(9): p. 5228-5237. 167. Rowan, D.D., Volatile metabolites. Metabolites, 2011. 1(1): p. 41-63. 168. Schmidt, K. and I. Podmore, Solid phase microextraction (SPME) method development in analysis of volatile organic compounds (VOCS) as potential biomarkers of cancer. Journal of Molecular Biomarkers & Diagnosis, 2015. 2015. 169. Xu, L., C. Basheer, and H.K. Lee, Developments in single-drop microextraction. Journal of Chromatography A, 2007. 1152(1): p. 184-192. 170. Liu, H. and P.K. Dasgupta, Analytical chemistry in a drop. solvent extraction in a microdrop. Analytical Chemistry, 1996. 68(11): p. 1817-1821. 171. Jeannot, M.A. and F.F. Cantwell, Solvent microextraction into a single drop. Analytical Chemistry, 1996. 68(13): p. 2236-2240. 172. Ahmadi, F., et al., Determination of organophosphorus pesticides in water samples by single drop microextraction and gas chromatography-flame photometric detector. Journal of Chromatography A, 2006. 1101(1): p. 307-312. 173. Liu, W. and H.K. Lee, Continuous-flow microextraction exceeding1000-fold concentration of dilute analytes. Analytical Chemistry, 2000. 72(18): p. 4462-4467. 174. Theis, A.L., et al., Headspace solvent microextraction. Analytical Chemistry, 2001. 73(23): p. 5651- 5654. 175. Ouyang, G., W. Zhao, and J. Pawliszyn, Kinetic calibration for automated headspace liquid-phase microextraction. Analytical Chemistry, 2005. 77(24): p. 8122-8128. 132 References

176. Lapthorn, C., F. Pullen, and B.Z. Chowdhry, Ion mobility spectrometry-mass spectrometry (IMS- MS) of small molecules: separating and assigning structures to ions. Mass Spectrometry Reviews, 2013. 32(1): p. 43-71. 177. Schneider, T., et al., An integrative clinical database and diagnostics platform for biomarker identification and analysis in ion mobility spectra of human exhaled air. Journal of Integrative Bioinformatics, 2013. 10: p. 218. 178. Eiceman, G.A., et al., Fragmentation of butyl acetate isomers in the drift region of an ion mobility spectrometer. International Journal of Mass Spectrometry and Ion Processes, 1988. 85(3): p. 265- 275. 179. Becker, S., et al., Surfaced-enhanced laser desorption/ionization time-of-flight (SELDI-TOF) differentiation of serum protein profiles of BRCA-1 and sporadic breast cancer. Annals of Surgical Oncology, 2004. 11(10): p. 907-914. 180. Makarov, A., Electrostatic axially harmonic orbital trapping: a high-performance technique of mass analysis. Analytical Chemistry, 2000. 72(6): p. 1156-1162. 181. Wu, C., et al., Construction and characterization of a high-flow, high-resolution ion mobility spectrometer for detection of explosives after personnel portal sampling. Talanta, 2002. 57(1): p. 123-134. 182. Graus, M., M. Müller, and A. Hansel, High resolution PTR-TOF: quantification and formula confirmation of VOC in real time. Journal of the American Society for Mass Spectrometry, 2010. 21(6): p. 1037-1044. 183. Fiehn, O., et al., Identification of uncommon plant metabolites based on calculation of elemental compositions using gas chromatography and quadrupole mass spectrometry. Analytical Chemistry, 2000. 72(15): p. 3573-3580. 184. Pagnotti, V.S., N.D. Chubatyi, and C.N. McEwen, Solvent assisted inlet ionization: an ultrasensitive new liquid introduction ionization method for mass spectrometry. Analytical Chemistry, 2011. 83(11): p. 3981-5. 185. Halbfeld, C., B.E. Ebert, and L.M. Blank, Multi-capillary column-ion mobility spectrometry of volatile metabolites emitted by Saccharomyces cerevisiae. Metabolites, 2014. 4(3): p. 751-74. 186. Hamm, S., et al., A chemical investigation by headspace SPME and GC–MS of volatile and semi- volatile terpenes in various olibanum samples. Phytochemistry, 2005. 66(12): p. 1499-1514. 187. Pragst, F., et al., Analysis of fatty acid ethyl esters in hair as possible markers of chronically elevated alcohol consumption by headspace solid-phase microextraction (HS-SPME) and gas chromatography-mass spectrometry (GC-MS). Forensic Science International, 2001. 121(1–2): p. 76-88. 188. Ochiai, N., et al., Optimization of a multi‐residue screening method for the determination of 85 pesticides in selected food matrices by stir bar sorptive extraction and thermal desorption GC‐ MS. Journal of Separation Science, 2005. 28(9‐10): p. 1083-1092. 189. Aprea, E., et al., Application of PTR-TOF-MS to investigate metabolites in exhaled breath of patients affected by coeliac disease under gluten free diet. Journal of Chromatography. B: Analytical Technologies in the Biomedical and Life Sciences, 2014. 966: p. 208-13. 190. Nguyen, N.N., et al., Megafiller: A retrofitted protein function predictor for filling gaps in metabolic networks. Journal of Proteomics and Bioinformatics, 2014. S9: p. 003. 191. Engel, S.R., et al., Saccharomyces Genome Database provides mutant phenotype data. Nucleic Acids Research, 2010. 38(Database issue): p. D433-6. 192. Karp, P.D., et al., Expansion of the BioCyc collection of pathway/genome databases to 160 genomes. Nucleic Acids Research, 2005. 33(19): p. 6083-6089. 193. Jewison, T., et al., YMDB: the yeast metabolome database. Nucleic Acids Research, 2012. 40(Database issue): p. D815-20. 194. Heavner, B.D., et al., Version 6 of the consensus yeast metabolic network refines biochemical coverage and improves model performance. Database: The Journal of Biological Databases and Curation, 2013. 2013: p. bat059.

133 References

195. Forster, J., et al., Genome-scale reconstruction of the Saccharomyces cerevisiae metabolic network. Genome Research, 2003. 13(2): p. 244-253. 196. Kuepfer, L., U. Sauer, and L.M. Blank, Metabolic functions of duplicate genes in Saccharomyces cerevisiae. Genome Research, 2005. 15. 197. Christian, N., et al., An integrative approach towards completing genome-scale metabolic networks. Molecular Biosystems, 2009. 5(12): p. 1889-1903. 198. Cakir, T. and M.J. Khatibipour, Metabolic network discovery by top-down and bottom-up approaches and paths for reconciliation. Frontiers in Bioengineering and Biotechnology, 2014. 2: p. 62. 199. Shahzad, K. and J.J. Loor, Application of top-down and bottom-up systems approaches in ruminant physiology and metabolism. Current Genomics, 2012. 13(5): p. 379-394. 200. Simeonidis, E., et al., Why does yeast ferment? A flux balance analysis study. Biochemical Society Transactions, 2010. 38(5): p. 1225-1229. 201. Molenaar, D., et al., Shifts in growth strategies reflect tradeoffs in cellular economics. Molecular Systems Biology, 2009. 5: p. 323. 202. Machado, D. and M.J. Herrgård, Co-evolution of strain design methods based on flux balance and elementary mode analysis. Metabolic Engineering Communications, 2015. 2: p. 85-92. 203. Lewis, N.E., H. Nagarajan, and B.O. Palsson, Constraining the metabolic genotype-phenotype relationship using a phylogeny of in silico methods. Nature Reviews: Microbiology, 2012. 204. Kanehisa, M., et al., The KEGG databases at GenomeNet. Nucleic Acids Research, 2002. 30(1): p. 42-46. 205. Caspi, R., et al., MetaCyc: a multiorganism database of metabolic pathways and enzymes. Nucleic Acids Research, 2006. 34(Database issue): p. D511-6. 206. Schellenberger, J., et al., BiGG: a biochemical genetic and genomic knowledgebase of large scale metabolic reconstructions. BMC Bioinformatics, 2010. 11: p. 213. 207. Chou, C.H., et al., FMM: a web server for metabolic pathway reconstruction and comparative analysis. Nucleic Acids Research, 2009. 37(Web Server issue): p. W129-34. 208. Blum, T. and O. Kohlbacher, MetaRoute: fast search for relevant metabolic routes for interactive network navigation and visualization. Bioinformatics, 2008. 24(18): p. 2108-2109. 209. Moriya, Y., et al., PathPred: an enzyme-catalyzed metabolic pathway prediction server. Nucleic Acids Research, 2010. 38: p. W138-W143. 210. Hatzimanikatis, V., et al., Exploring the diversity of complex metabolic networks. Bioinformatics, 2005. 21(8): p. 1603-1609. 211. Hadadi, N., et al., A computational framework for integration of lipidomics data into metabolic pathways. Metabolic Engineering, 2014. 23C: p. 1-8. 212. Brunk, E., et al., Characterizing strain variation in engineered E. coli using a multi-omics-based workflow. Cell Systems, 2016. 2(5): p. 335-346. 213. Zur, H., E. Ruppin, and T. Shlomi, iMAT: an integrative metabolic analysis tool. Bioinformatics, 2010. 26(24): p. 3140-3142. 214. Jensen, P.A. and J.A. Papin, Functional integration of a metabolic network model and expression data without arbitrary thresholding. Bioinformatics, 2011. 27(4): p. 541-7. 215. van Berlo, R.J.P., et al., Predicting metabolic fluxes using gene expression differences as constraints. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2011. 8(1): p. 206-216. 216. Faria, J.P., et al., Genome-scale bacterial transcriptional regulatory networks: reconstruction and integrated analysis with metabolic models. Briefings in Bioinformatics, 2014. 15(4): p. 592-611. 217. Machado, D. and M. Herrgard, Systematic evaluation of methods for integration of transcriptomic data into constraint-based models of metabolism. PLoS Computational Biology, 2014. 10(4): p. e1003580. 218. Orth, J.D., I. Thiele, and B.O. Palsson, What is flux balance analysis? Nature Biotechnology, 2010. 28(3): p. 245-8.

134 References

219. Henry, C.S., L.J. Broadbelt, and V. Hatzimanikatis, Thermodynamics-based metabolic flux analysis. Biophysical Journal, 2007. 92(5): p. 1792-805. 220. Crown, S.B., W.S. Ahn, and M.R. Antoniewicz, Rational design of 13C-labeling experiments for metabolic flux analysis in mammalian cells. BMC Systems Biology, 2012. 6. 221. Yang, H., D.E. Mandy, and I.G. Libourel, Optimal design of isotope labeling experiments. Methods in Molecular Biology, 2014. 1083: p. 133-47. 222. Noh, K., A. Wahl, and W. Wiechert, Computational tools for isotopically instationary C-13 labeling experiments under metabolic steady state conditions. Metabolic Engineering, 2006. 8(6): p. 554-577. 223. Mollney, M., et al., Bidirectional reaction steps in metabolic networks: IV. optimal design of isotopomer labeling experiments. Biotechnology and Bioengineering, 1999. 66(2): p. 86-103. 224. Seijo, M., K.C. Soh, and V. Hatzimanikatis, A novel approach to find the missing links in genome- scale metabolic models: The BridgIT method, in FOSBE 2012. 2012: Tsuruoka, Japan. 225. Tervo, C.J. and J.L. Reed, FOCAL: an experimental design tool for systematizing metabolic discoveries and model development. Genome Biology, 2012. 13(12). 226. Mo, M.L., B.Ø. Palsson, and M.J. Herrgård, Connecting extracellular metabolomic measurements to intracellular flux states in yeast. BMC Systems Biology, 2009. 3. 227. Herrgard, M.J., et al., A consensus yeast metabolic network reconstruction obtained from a community approach to systems biology. Nature Biotechnology, 2008. 26(10): p. 1155-60. 228. Dobson, P.D., et al., Further developments towards a genome-scale metabolic model of yeast. BMC Systems Biology, 2010. 4(1): p. 145. 229. Heavner, B.D., et al., Yeast 5 – an expanded reconstruction of the Saccharomyces cerevisiae metabolic network. BMC Systems Biology, 2012. 6(1): p. 55. 230. Aung, H.W., S.A. Henry, and L.P. Walker, Revising the representation of fatty acid, glycerolipid, and glycerophospholipid metabolism in the consensus model of yeast metabolism. Industrial Biotechnology, 2013. 9(4): p. 215-228. 231. Kanehisa, M., et al., KEGG: new perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Research, 2017. 45(D1): p. D353-D361. 232. Kanehisa, M. and S. Goto, KEGG: Kyoto encyclopedia of genes and genomes. Nucleic Acids Research, 2000. 28(1): p. 27-30. 233. Kanehisa, M., et al., From genomics to chemical genomics: new developments in KEGG. Nucleic Acids Research, 2006. 34. 234. Soh, K.C. and V. Hatzimanikatis, DREAMS of metabolism. Trends in Biotechnology, 2010. 28(10): p. 501-8. 235. Jankowski, M.D., et al., Group contribution method for thermodynamic analysis of complex metabolic networks. Biophysical Journal, 2008. 95(3): p. 1487-99. 236. Hyduke, D., et al., COBRA toolbox 2.0. 2011. 237. Verduyn, C., et al., Effect of benzoic acid on metabolic fluxes in yeasts: a continuous-culture study on the regulation of respiration and alcoholic fermentation. Yeast, 1992. 8(7): p. 501-517. 238. Jakociunas, T., et al., Multiplex metabolic pathway engineering using CRISPR/Cas9 in Saccharomyces cerevisiae. Metabolic Engineering, 2015. 28: p. 213-22. 239. Electrocompetent E. coli. 2017 [cited 2017 10.10.2017]; Available from: http://barricklab.org/twiki/bin/view/Lab/ProtocolsElectrocompetentCells. 240. Jessop-Fabre, M.M., et al., EasyClone-MarkerFree: A vector toolkit for marker-less integration of genes into Saccharomyces cerevisiae via CRISPR-Cas9. Biotechnology Journal, 2016. 11(8): p. 1110-7. 241. Joska, T.M., et al., A universal cloning method based on yeast homologous recombination that is simple, efficient, and versatile. Journal of Microbiological Methods, 2014. 100: p. 46-51. 242. Bitinaite, J., et al., USER friendly DNA engineering and cloning method by uracil excision. Nucleic Acids Research, 2007. 35(6): p. 1992-2002. 243. Bitinaite, J. and N.M. Nichols, DNA cloning and engineering by uracil excision, in Current Protocols in Molecular Biology. 2001, John Wiley & Sons, Inc.

135 References

244. Vaisvila, R. and J. Bitinaite, Gene synthesis by assembly of deoxyuridine-containing oligonucleotides, in Enzyme Engineering: Methods and Protocols, J.C. Samuelson, Editor. 2013, Humana Press: Totowa, NJ. p. 165-171. 245. Famili, I., et al., Saccharomyces cerevisiae phenotypes can be predicted by using constraint-based analysis of a genome-scale reconstructed metabolic network. Proceedings of the National Academy of Sciences, 2003. 100(23): p. 13134-13139. 246. Hazelwood, L.A., et al., The Ehrlich pathway for fusel alcohol production: a century of research on Saccharomyces cerevisiae metabolism. Applied and Environmental Microbiology, 2008. 74(8): p. 2259-66. 247. Bruce, A., et al., Identification of volatile organic compounds (VOCs) from bacteria and yeast causing growth inhibition of sapstain fungi. Holzforschung, 2004. 58(2): p. 193-198. 248. Hazelwood, L. CORNUCOPIA - Non-conventional yeasts for food fermentation. in 25th VH Yeast Conference. 2012. Research Institute for Baker's Yeast, Berlin (self-published). 249. Callejon, R.M., et al., Volatile and sensory profile of organic red wines produced by different selected autochthonous and commercial Saccharomyces cerevisiae strains. Analytica Chimica Acta, 2010. 660(1-2): p. 68-75. 250. Perez-Coello, M.S., et al., Characteristics of wines fermented with different Saccharomyces cerevisiae strains isolated from the La Mancha region. Food Microbiology, 1999. 16(6): p. 563-573. 251. Patel, S. and T. Shibamoto, Effect of 20 different yeast strains on the production of volatile components in Symphony wine. Journal of Food Composition and Analysis, 2003. 16(4): p. 469- 476. 252. Torija, M.J., et al., Effects of fermentation temperature and Saccharomyces species on the cell fatty acid composition and presence of volatile compounds in wine. International Journal of Food Microbiology, 2003. 85(1-2): p. 127-136. 253. Mateos, J.A.R., F. Perez-Nevado, and M.R. Fernandez, Influence of Saccharomyces cerevisiae yeast strain on the major volatile compounds of wine. Enzyme and Microbial Technology, 2006. 40(1): p. 151-157. 254. Ames, J.M. and G.M. Leod, Volatile components of a yeast extract composition. Journal of Food Science, 1985. 50(1): p. 125. 255. Kanehisa, M., et al., KEGG as a reference resource for gene and protein annotation. Nucleic Acids Research, 2016. 44(D1): p. 457-462. 256. Mavrovouniotis, M.L., Group contributions for estimating standard gibbs energies of formation of biochemical compounds in aqueous solution. Biotechnology and Bioengineering, 1990. 36(10): p. 1070-1082. 257. Atsumi, S., T. Hanai, and J.C. Liao, Non-fermentative pathways for synthesis of branched-chain higher alcohols as biofuels. Nature, 2008. 451(7174): p. 86-9. 258. Dickinson, J.R., The catabolism of amino acids to long chain and complex alcohols in Saccharomyces cerevisiae. Journal of Biological Chemistry, 2002. 278(10): p. 8028-8034. 259. Fujii, T., et al., Nucleotide sequences of alcohol acetyltransferase genes from lager brewing yeast, Saccharomyces carlsbergensis. Yeast, 1996. 12(6): p. 593-8. 260. Nagasawa, N., et al., Cloning and nucleotide sequence of the alcohol acetyltransferase II gene (ATF2) from Saccharomyces cerevisiae Kyokai No. 7. Bioscience, Biotechnology, and Biochemistry, 1998. 62(10): p. 1852-1857. 261. Vasserot, Y., V. Steinmetz, and P. Jeandet, Study of thiol consumption by yeast lees. Antonie Van Leeuwenhoek, 2003. 83(3): p. 201-207. 262. Knoll, L.J., D.R. Johnson, and J.I. Gordon, Biochemical studies of three Saccharomyces cerevisiae acyl-CoA synthetases, Faa1p, Faa2p, and Faa3p. Journal of Biological Chemistry, 1994. 269(23): p. 16348-56. 263. Lutstorf, U. and R. Megnet, Multiple forms of alcohol dehydrogenase in Saccharomyces cerevisiae. Archives of Biochemistry and Biophysics, 1968. 126(3): p. 933-944.

136 References

264. Drewke, C., J. Thielen, and M. Ciriacy, Ethanol formation in adh0 mutants reveals the existence of a novel acetaldehyde-reducing activity in Saccharomyces cerevisiae. Journal of Bacteriology, 1990. 172(7): p. 3909-3917. 265. Smith, M.G., S.G. Des Etages, and M. Snyder, Microbial synergy via an ethanol-triggered pathway. Molecular and Cellular Biology, 2004. 24(9): p. 3874-3884. 266. de Smidt, O., J.C. du Preez, and J. Albertyn, The alcohol dehydrogenases of Saccharomyces cerevisiae: a comprehensive review. FEMS Yeast Research, 2008. 8(7): p. 967-78. 267. Wenger, J.I. and C. Bernofsky, Mitochondrial alcohol dehydrogenase from Saccharomyces cerevisiae. Biochimica et Biophysica Acta (BBA) - Enzymology, 1971. 227(3): p. 479-490. 268. Wills, C., P. Kratofil, and T. Martin, Functional mutants of yeast alcohol dehydrogenase, in Genetic Engineering of Microorganisms for Chemicals, A. Hollaender, et al., Editors. 1982, Springer US: Boston, MA. p. 305-329. 269. Weimer, E.P., E. Rao, and M. Brendel, Molecular structure and genetic regulation of SFA, a gene responsible for resistance to formaldehyde in Saccharomyces cerevisiae, and characterization of its protein product. Molecular and General Genetics MGG, 1993. 237(3): p. 351-358. 270. Larroy, C., et al., Properties and functional significance of Saccharomyces cerevisiae ADHVI. Chemico-Biological Interactions, 2003. 143: p. 229-238. 271. Larroy, C., et al., Characterization of the Saccharomyces cerevisiae YMR318C (ADH6) gene product as a broad specificity NADPH-dependent alcohol dehydrogenase: relevance in aldehyde reduction. Biochemical Journal, 2002. 361(1): p. 163-172. 272. Hauser, M., et al., A transcriptome analysis of isoamyl alcohol-induced filamentation in yeast reveals a novel role for Gre2p as isovaleraldehyde reductase. FEMS Yeast Research, 2007. 7(1): p. 84-92. 273. Gancedo, C., C.L. Flores, and J.M. Gancedo, The expanding landscape of moonlighting proteins in yeasts. Microbiology and Molecular Biology Reviews, 2016. 80(3): p. 765-77. 274. Hult, K. and P. Berglund, Enzyme promiscuity: mechanism and applications. Trends in Biotechnology, 2007. 25. 275. Messenguy, F. and J.M. Wiame, The control of ornithinetranscarbamylase activity by arginase in Saccharomyces cerevisiae. FEBS Letters, 1969. 3(1): p. 47-49. 276. Gancedo, C. and C.-L. Flores, Moonlighting proteins in yeasts. Microbiology and Molecular Biology Reviews, 2008. 72(1): p. 197-210. 277. Doerks, T., et al., Systematic identification of novel protein domain families associated with nuclear functions. Genome Research, 2002. 12(1): p. 47-56. 278. Fischer, E., N. Zamboni, and U. Sauer, High-throughput metabolic flux analysis based on gas chromatography–mass spectrometry derived 13C constraints. Analytical Biochemistry, 2004. 325(2): p. 308-316. 279. Rühl, M., et al., Collisional fragmentation of central carbon metabolites in LC-MS/MS increases precision of 13C metabolic flux analysis. Biotechnology and Bioengineering, 2012. 109(3): p. 763- 771. 280. Jin, C., et al., Progress in the production and application of n-butanol as a biofuel. Renewable and Sustainable Energy Reviews, 2011. 15(8): p. 4080-4106. 281. Adkins, J., et al., Engineering microbial chemical factories to produce renewable “biomonomers”. Frontiers in Microbiology, 2012. 3: p. 313. 282. McGovern, P.E., et al., Fermented beverages of pre- and proto-historic China. Proceedings of the National Academy of Sciences of the United States of America, 2004. 101(51): p. 17593-8. 283. Sicard, D. and J.-L. Legras, Bread, beer and wine: Yeast domestication in the Saccharomyces sensu stricto complex. Comptes Rendus Biologies, 2011. 334(3): p. 229-236. 284. McGovern, P.E., et al., Neolithic resinated wine. Nature, 1996. 381(6582): p. 480-481. 285. Mortimer, R.K., Evolution and variation of the yeast (Saccharomyces) genome. Genome Research, 2000. 10(4): p. 403-9. 286. McGovern, P.E., A. Mirzoian, and G.R. Hall, Ancient Egyptian herbal wines. Proceedings of the National Academy of Sciences of the United States of America, 2009. 106(18): p. 7361-6. 137 References

287. Goffeau, A., et al., Life with 6000 genes. Science, 1996. 274(5287): p. 546-567. 288. Ro, D.K., et al., Production of the antimalarial drug precursor artemisinic acid in engineered yeast. Nature, 2006. 440(7086): p. 940-3. 289. de Deken, R.H., The Crabtree effect: a regulatory system in yeast. Microbiology, 1966. 44(2): p. 149-156. 290. Torrens, J., et al., Volatile compounds of red and white wines by headspace--solid-phase microextraction using different fibers. Journal of Chromatographic Science, 2004. 42(6): p. 310- 6. 291. Polaskova, P., J. Herszage, and S.E. Ebeler, Wine flavor: chemistry in a glass. Chemical Society Reviews, 2008. 37(11): p. 2478-89. 292. Stevens, R., Beer flavour: I. Volatile products of dermentation: a review. Journal of the Institute of Brewing, 1960. 66(6): p. 453-471. 293. Kobayashi, M., H. Shimizu, and S. Shioya, Beer volatile compounds and their application to low- malt beer fermentation. Journal of Bioscience and Bioengineering, 2008. 106(4): p. 317-323. 294. Frick, O. and C. Wittmann, Characterization of the metabolic shift between oxidative and fermentative growth in Saccharomyces cerevisiae by comparative 13C flux analysis. Microbial Cell Factories, 2005. 4: p. 30. 295. Tejero Rioseras, A., et al., Comprehensive real-time analysis of the yeast volatilome. Scientific Reports, 2017. 7(1): p. 14236. 296. Wu, C., W.F. Siems, and H.H. Hill, Secondary electrospray ionization ion mobility spectrometry/mass spectrometry of illicit drugs. Analytical Chemistry (Washington), 2000. 72(2): p. 396-403. 297. Barrios-Collado, C., et al., Capturing in vivo plant metabolism by real-time analysis of low to high molecular weight volatiles. Analytical Chemistry, 2016. 88(4): p. 2406-12. 298. Cole, R.B., Some tenets pertaining to electrospray ionization mass spectrometry. Journal of Mass Spectrometry, 2000. 35(7): p. 763-772. 299. Barrios-Collado, C., G. Vidal-de-Miguel, and P. Martinez-Lozano Sinues, Numerical modeling and experimental validation of a universal secondary electrospray ionization source for mass spectrometric gas analysis in real-time. Sensors and Actuators B: Chemical, 2016. 223: p. 217-225. 300. Vidal-de-Miguel, G., et al., Low-sample flow secondary electrospray ionization: improving vapor ionization efficiency. Analytical Chemistry, 2012. 84(20): p. 8475-9. 301. Bateman, K.P., et al., Quantitative-qualitative data acquisition using a benchtop Orbitrap mass spectrometer. Journal of the American Society for Mass Spectrometry, 2009. 20(8): p. 1441-50. 302. Martinez-Lozano Sinues, P., et al., Circadian variation of the human metabolome captured by real-time breath analysis. PloS One, 2014. 9(12): p. 114422. 303. Sinues, P.M., M. Kohler, and R. Zenobi, Monitoring diurnal changes in exhaled human breath. Analytical Chemistry, 2013. 85(1): p. 369-73. 304. Kessner, D., et al., ProteoWizard: open source software for rapid proteomics tools development. Bioinformatics, 2008. 24(21): p. 2534-6. 305. Hong, S.Y., A.S. Zurbriggen, and A. Melis, Isoprene hydrocarbons production upon heterologous transformation of Saccharomyces cerevisiae. Journal of Applied Microbiology, 2012. 113(1): p. 52-65. 306. Anderson, M.S., et al., Farnesyl diphosphate synthetase. Molecular cloning, sequence, and expression of an essential gene from Saccharomyces cerevisiae. Journal of Biological Chemistry, 1989. 264(32): p. 19176-19184. 307. Morales, M.L., et al., Monitoring volatile compounds production throughout fermentation by Saccharomyces and non-Saccharomyces strains using headspace sorptive extraction. Journal of Food Science and Technology, 2017. 54(2): p. 538-557. 308. Fan, J., et al., Quantitative flux analysis reveals folate-dependent NADPH production. Nature, 2014. 510(7504): p. 298-302. 309. Link, H., et al., Real-time metabolome profiling of the metabolic switch between starvation and growth. Nature Methods, 2015. 12(11): p. 1091-7. 138 References

310. Mortimer, R.K., Evolution and variation of the yeast (Saccharomyces) genome. Genome Research, 2000. 10(6): p. 891-891. 311. McGovern, P.E., et al., Fermented beverages of pre- and proto-historic China. Proceedings of the National Academy of Sciences of the United States of America, 2004. 101(51): p. 17593-17598. 312. Sicard, D. and J.L. Legras, Bread, beer and wine: Yeast domestication in the Saccharomyces sensu stricto complex. Comptes rendus: Biologies, 2011. 334(3): p. 229-236. 313. Paddon, C.J., et al., High-level semi-synthetic production of the potent antimalarial artemisinin. Nature, 2013. 496(7446): p. 528-32. 314. Kjeldsen, T., Yeast secretory expression of insulin precursors. Applied Microbiology and Biotechnology, 2000. 54(3): p. 277-86. 315. Thim, L., et al., Secretion and processing of insulin precursors in yeast. Proceedings of the National Academy of Sciences of the United States of America, 1986. 83(18): p. 6766-70. 316. Basso, L.C., et al., Yeast selection for fuel ethanol production in Brazil. FEMS Yeast Research, 2008. 8(7): p. 1155-63. 317. Quantz, M., Versuchsanstalt der Hefeindustrie e.V., Berlin, Germany. 2014. 318. Paczia, N., et al., Extensive exometabolome analysis reveals extended overflow metabolism in various microorganisms. Microbial Cell Factories, 2012. 11. 319. Suomalainen, H. and L. Nykänen, The aroma components produced by yeast in nitrogen-free sugar solution. Journal of the Institute of Brewing, 1966. 72(5): p. 469-474. 320. Maddula, S., et al., Detection of volatile metabolites of Escherichia coli by multi capillary column coupled ion mobility spectrometry. Analytical and Bioanalytical Chemistry, 2009. 394(3): p. 791- 800. 321. Kunze, N., et al., Detection and validation of volatile metabolic patterns over different strains of two human pathogenic bacteria during their growth in a complex medium using multi-capillary column-ion mobility spectrometry (MCC-IMS). Applied Microbiology and Biotechnology, 2013. 97(8): p. 3665-76. 322. Kotiaho, T., et al., Membrane inlet ion mobility spectrometry for online measurement of ethanol in beer and in yeast fermentation. Analytica Chimica Acta, 1995. 309(1-3): p. 317-325. 323. Kolehmainen, M., P. Ronkko, and A. Raatikainen, Monitoring of yeast fermentation by ion mobility spectrometry measurement and data visualisation with self-organizing maps. Analytica Chimica Acta, 2003. 484(1): p. 93-100. 324. Vautz, W., J.I. Baumbach, and J. Jung, Beer fermentation control using ion mobility spectrometry - Results of a pilot study. Journal of the Institute of Brewing, 2006. 112(2): p. 157-164. 325. van Dijken, J.P., et al., An interlaboratory comparison of physiological and genetic properties of four Saccharomyces cerevisiae strains. Enzyme and Microbial Technology, 2000. 26(9-10): p. 706-714. 326. Blank, L.M., L. Kuepfer, and U. Sauer, Large-scale 13C-flux analysis reveals mechanistic principles of metabolic network robustness to null mutations in yeast. Genome Biology, 2005. 6(6): p. R49. 327. Eiceman, G.H., Z. Karpas, and H.H. Hill, Ion mobility spectrometry. Third Edition. ed. xvi, 428 pages. 328. Bödeker, B., W. Vautz, and J. Baumbach, Peak finding and referencing in MCC/IMS data. International Journal for Ion Mobility Spectrometry, 2008. 11(1-4): p. 83-87. 329. Ruzsanyi, V., J.I. Baumbach, and G.A. Eiceman, Detection of the mold markers using ion mobility spectrometry. International Journal for Ion Mobility Spectrometry, 2003. 6: p. 53-57. 330. Crabtree, H.G., Observations on the carbohydrate metabolism of tumours. Biochemical Journal, 1929. 23(3): p. 536-45. 331. Randez-Gil, F., I. Corcoles-Saez, and J.A. Prieto, Genetic and phenotypic characteristics of baker's yeast: relevance to baking. Annual Review of Food Science and Technology, 2013. 4: p. 191-214. 332. Ejiofor, A.O., N. Okafor, and E.N. Ugwueze, Development of baking yeast from Nigerian palm- wine yeasts. World Journal of Microbiology & Biotechnology, 1994. 10(2): p. 199-202. 333. Fischer, K. Straightforward prognostication of durability of baker’s yeast. in 18th VH Yeast Conference. 2005. Research Institute for Baker's Yeast, Berlin (self-published). 139 References

334. Porro, D., et al., Recombinant protein production in yeasts. Molecular Biotechnology, 2005. 31(3): p. 245-59. 335. Jouhten, P., et al., Oxygen dependence of metabolic fluxes and energy generation of Saccharomyces cerevisiae CEN.PK113-1A. BMC Systems Biology, 2008. 2. 336. Chen, G.C. and F. Jordan, Brewers-yeast pyruvate decarboxylase produces acetoin from acetaldehyde - a novel tool to study the mechanism of steps subsequent to carbon-dioxide loss. Biochemistry, 1984. 23(16): p. 3576-3582. 337. Heidlas, J. and R. Tressl, Purification and properties of two oxidoreductases catalyzing the enantioselective reduction of diacetyl and other diketones from baker's yeast. European Journal of Biochemistry, 1990. 188(1): p. 165-74. 338. Romano, P. and G. Suzzi, Origin and production of acetoin during wine yeast fermentation. Applied and Environmental Microbiology, 1996. 62(2): p. 309-15. 339. Bartowsky, E.J. and P.A. Henschke, The 'buttery' attribute of wine-diacetyl-desirability, spoilage and beyond. International Journal of Food Microbiology, 2004. 96(3): p. 235-252. 340. Gonzalez, E., et al., Characterization and functional role of Saccharomyces cerevisiae 2,3-butanediol dehydrogenase. Chemico-Biological Interactions, 2001. 130-132(1-3): p. 425-34. 341. Pronk, J.T., H.Y. Steensma, and J.P. vanDijken, Pyruvate metabolism in Saccharomyces cerevisiae. Yeast, 1996. 12(16): p. 1607-1633. 342. Ng, C.Y., et al., Production of 2,3-butanediol in Saccharomyces cerevisiae by in silico aided metabolic engineering. Microbial Cell Factories, 2012. 11: p. 68. 343. Walker, V. and G.A. Mills, 2-Pentanone production from hexanoic acid by Penicillium roqueforti from blue cheese: Is this the pathway used in humans? Scientific World Journal, 2014. 344. Dickschat, J.S., et al., Pyrazine biosynthesis in Corynebacterium glutamicum. European Journal of Organic Chemistry, 2010(14): p. 2687-2695. 345. Gancedo, J.M., Yeast carbon catabolite repression. Microbiology and Molecular Biology Reviews, 1998. 62(2): p. 334-361. 346. Kayikci, O. and J. Nielsen, Glucose repression in Saccharomyces cerevisiae. FEMS Yeast Research, 2015. 15(6). 347. DeRisi, J.L., V.R. Iyer, and P.O. Brown, Exploring the metabolic and genetic control of gene expression on a genomic scale. Science, 1997. 278(5338): p. 680-686. 348. Shalon, D., S.J. Smith, and P.O. Brown, A DNA microarray system for analyzing complex DNA samples using two-color fluorescent probe hybridization. Genome Research, 1996. 6(7): p. 639- 645. 349. Schena, M., et al., Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science, 1995. 270(5235): p. 467-470. 350. Bodenmiller, B., et al., PhosphoPep[mdash]a database of protein phosphorylation sites in model organisms. Nature Biotechnology, 2008. 26(12): p. 1339-1340. 351. Oliveira, A.P. and U. Sauer, The importance of post-translational modifications in regulating Saccharomyces cerevisiae metabolism. FEMS Yeast Research, 2012. 12(2): p. 104-17. 352. Kresnowati, M.T., et al., When transcriptome meets metabolome: fast cellular responses of yeast to sudden relief of glucose limitation. Molecular Systems Biology, 2006. 2: p. 49. 353. ter Linde, J.J.M., et al., Genome-Wide Transcriptional Analysis of Aerobic and Anaerobic Chemostat Cultures of Saccharomyces cerevisiae. Journal of Bacteriology, 1999. 181(24): p. 7409-7413. 354. Regenberg, B., et al., Growth-rate regulated genes have profound impact on interpretation of transcriptome profiling in Saccharomyces cerevisiae. Genome Biology, 2006. 7(11): p. 107-107. 355. Martinez, J.L., et al., Gcn4p and the Crabtree effect of yeast: drawing the causal model of the Crabtree effect in Saccharomyces cerevisiae and explaining evolutionary trade-offs of adaptation to galactose through systems biology. FEMS Yeast Research, 2014. 14(4): p. 654-62. 356. Bell, P.J.L., V.J. Higgins, and P.V. Attfield, Comparison of fermentative capacities of industrial baking and wild-type yeasts of the species Saccharomyces cerevisiae in different sugar media. Letters in Applied Microbiology, 2001. 32(4): p. 224-229.

140 References

357. Benjamini, Y. and Y. Hochberg, Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society. Series B, 1995. 57(1): p. 289- 300. 358. Paley, S.M. and P.D. Karp, The pathway tools cellular overview diagram and omics viewer. Nucleic Acids Research, 2006. 34(13): p. 3771-8. 359. Broach, J.R., Nutritional control of growth and development in yeast. Genetics, 2012. 192(1): p. 73-105. 360. Casalone, E., et al., Identification by functional analysis of the gene encoding α-isopropylmalate synthase II (LEU9) in Saccharomyces cerevisiae. Yeast, 2000. 16(6): p. 539-545. 361. Eden, A., G. Simchen, and N. Benvenisty, Two yeast homologs of ECA39, a target for c-Myc regulation, code for cytosolic and mitochondrial branched-chain amino acid aminotransferases. Journal of Biological Chemistry, 1996. 271(34): p. 20242-20245. 362. NIST chemistry webbook, NIST standard reference database number 69, ed. P.J. Linstrom and W.G. Mallard. 2017, National Institute of Standards and Technology, Gaithersburg MD. 363. Donahue, N.M., et al., A two-dimensional volatility basis set – Part 2: Diagnostics of organic- aerosol evolution. Atmospheric Chemistry and Physics (Print), 2012. 12(2): p. 615-634. 364. Donahue, N.M., et al., A two-dimensional volatility basis set: 1. organic-aerosol mixing thermodynamics. Atmospheric Chemistry and Physics (Print), 2011. 11(7): p. 3303-3318. 365. Gross, J. and G. Sadowski, Perturbed-chain SAFT: An equation of state based on a perturbation theory for chain molecules. Industrial & Engineering Chemistry Research, 2001. 40(4): p. 1244- 1260. 366. Stovicek, V., I. Borodina, and J. Forster, CRISPR–Cas system enables fast and simple genome editing of industrial Saccharomyces cerevisiae strains. Metabolic Engineering Communications, 2015. 2: p. 13-22. 367. Zhang, G.-C., et al., Construction of a quadruple auxotrophic mutant of an industrial polyploid Saccharomyces cerevisiae strain by using RNA-guided Cas9 nuclease. Applied and Environmental Microbiology, 2014. 80(24): p. 7694-7701. 368. Wang, Y., et al., Simultaneous editing of three homoeoalleles in hexaploid bread wheat confers heritable resistance to powdery mildew. Nature Biotechnology, 2014. 32(9): p. 947-51. 369. Ilc, T., D. Werck-Reichhart, and N. Navrot, Meta-analysis of the core aroma components of grape and wine aroma. Frontiers in Plant Science, 2016. 7: p. 1472. 370. González-Barreiro, C., et al., Wine aroma compounds in grapes: a critical review. Critical Reviews in Food Science and Nutrition, 2015. 55(2): p. 202-218. 371. He, Y., et al., Wort composition and its impact on the flavour-active higher alcohol and ester formation of beer – a review. Journal of the Institute of Brewing, 2014. 120(3): p. 157-163. 372. Hua, D. and P. Xu, Recent advances in biotechnological production of 2-phenylethanol. Biotechnology Advances, 2011. 29(6): p. 654-660. 373. Wang, H., et al., A continuous and adsorptive bioprocess for efficient production of the natural aroma chemical 2-phenylethanol with yeast. Enzyme and Microbial Technology, 2011. 48(4-5): p. 404-7. 374. Etschmann, M.M. and J. Schrader, An aqueous-organic two-phase bioprocess for efficient production of the natural aroma chemicals 2-phenylethanol and 2-phenylethylacetate with yeast. Applied Microbiology and Biotechnology, 2006. 71(4): p. 440-3. 375. Yilmaztekin, M., T. Cabaroglu, and H. Erten, Effects of fermentation temperature and aeration on production of natural isoamyl acetate by Williopsis saturnus var. saturnus. BioMed Research International, 2013. 2013: p. 870802. 376. Strobel, G.A., Bioprospecting--fuels from fungi. Biotechnology Letters, 2015. 37(5): p. 973-82. 377. Xue, J. and B.K. Ahring, Enhancing isoprene production by genetic modification of the 1-deoxy- d-xylulose-5-phosphate pathway in Bacillus subtilis. Applied and Environmental Microbiology, 2011. 77(7): p. 2399-405. 378. Jansen, D.J. and R.A. Shenvi, Synthesis of medicinally relevant terpenes: reducing the cost and time of drug discovery. Future Medicinal Chemistry, 2014. 6(10): p. 1127-48. 141 References

379. Dias, J., et al., On-line adaptive metabolic flux analysis: Application to PHB production by mixed microbial cultures. Biotechnology Progress, 2009. 25(2): p. 390-398. 380. Yeastnet. 2014 [cited 2014 02 September]; Available from: http://sourceforge.net/projects/yeast/files/. 381. Zimmermann, S., et al., Miniaturized low-cost ion mobility spectrometer for fast detection of chemical warfare agents. Analytical Chemistry, 2008. 80(17): p. 6671-6676. 382. Harris, G.A., M. Kwasnik, and F.M. Fernandez, Direct analysis in real time coupled to multiplexed drift tube ion mobility spectrometry for detecting toxic chemicals. Analytical Chemistry, 2011. 83(6): p. 1908-15. 383. Mäkinen, M., M. Nousiainen, and M. Sillanpää, Ion spectrometric detection technologies for ultra-traces of explosives: a review. Mass Spectrometry Reviews, 2011. 30(5): p. 940-973. 384. Palmer, P.T. and T.F. Limero, Mass spectrometry in the U.S. space program: past, present, and future. Journal of the American Society for Mass Spectrometry, 2001. 12(6): p. 656-675. 385. Eiceman, G.A., et al., Monitoring volatile organic compounds in ambient air inside and outside buildings with the use of a radio-frequency-based ion-mobility analyzer with a micromachined drift tube. Field Analytical Chemistry & Technology, 2000. 4(6): p. 297-308. 386. Ewing, R.G., et al., A critical review of ion mobility spectrometry for the detection of explosives and explosive related compounds. Talanta, 2001. 54(3): p. 515-529. 387. Bödeker, B., et al., Biomarker validation—room air variation during human breath investigations. International Journal for Ion Mobility Spectrometry, 2010. 13(3): p. 177-184. 388. Verkouteren, J.R. and J.L. Staymates, Reliability of ion mobility spectrometry for qualitative analysis of complex, multicomponent illicit drug samples. Forensic Science International, 2011. 206(1-3): p. 190-6. 389. Vautz, W. and J.I. Baumbach, Analysis of bio-processes using ion mobility spectrometry. Engineering in Life Sciences, 2008. 8(1): p. 19-25.

142

Curriculum vitae

Personal Data Name: Christoph Halbfeld Born: February 7th, 1988 in Moers, Germany Nationality: German Mobile: 0049 1520 4453659 Email: [email protected]

Education 2012-2017 Research assistant and doctoral candidate at the Institute of Applied Microbiology of the RWTH Aachen University, Aachen, Germany 2012 Braunwald course: Biosystems Engineering Summer School Braunwald, Switzerland 2010-2012 Master of Science in Biology at the RWTH Aachen University, Aachen, Germany 2008-2011 Bachelor of Science in Biology at the RWTH Aachen University, Aachen, Germany 2007 Abitur at the Anne-Frank-Gesamtschule Rheinkamp, Moers, Germany

Work Experience 2016 Research stay in the Zenobi Group at the Laboratory of Organic Chemistry at the ETH Zürich, Switzerland 2013-2017 Research assistant and doctoral candidate at the Institute of Applied Microbiology of the RWTH Aachen University, Aachen, Germany 2013 Research stay at B&S Analytik-GmbH Dortmund, Germany

Publications

Halbfeld, C.; Ebert, B.; Blank, L. Multi-capillary column-ion mobility spectrometry of volatile metabolites emitted by Saccharomyces cerevisiae. Metabolites 2014, 4, 751-774. Ebert, B.; Halbfeld, C.; Blank, L. Exploration and exploitation of the yeast volatilome. Current Metabolomics 2017, 5, 102-118.

Oral Presentations

Halbfeld, C.; Ebert, E.; Blank, L. (14.04.2015) YeastScent – Volatile metabolites as quantitative proxies for metabolic network operation of Saccharomyces cerevisiae. 28th VH-Yeast Conference. Berlin, Germany Halbfeld, C.; Ebert, E.; Blank, L. (30.07.2015) YeastScent – Volatile metabolites as quantitative proxies for metabolic network operation of Saccharomyces cerevisiae. 24th Annual ISIMS Conference. Cordoba, Spain Halbfeld, C.; Ebert, E.; Blank, L. (23.09.2015) YeastScent – Volatile metabolites as quantitative proxies for metabolic network operation of Saccharomyces cerevisiae. 6. Symposium Metaboliten in Prozessabluft und Ausatemluft. Reutlingen, Germany

147 Curriculum vitae

Halbfeld, C.; Sippel, A.K.; Pollmann, E.; Ebert, E.; Quantz, M.; Zierow, J.; Baumbach, J.I.; Blank, L. (25.04.2017) Investigation of the Crabtree effect using off-gas analysis. 30th VH-Yeast Conference. Berlin, Germany

Poster presentations

Halbfeld, C.; Ebert, E.; Blank, L. (22.02.2016) YeastScent – Volatile metabolite measurements for yeast fermentation control. Bioprocessing days 2016. Recklinghausen, Germany (Poster award) Halbfeld, C.; Ebert, B.; Martinez-Lozano Sinues, P.; Blank, L. (20.02.2017) Discovering dynamics of volatile metabolites in yeast fermentations using SESI-Orbitrap-MS. Bioprocessing days 2017. Recklinghausen, Germany

148