Metabolic Fluxes of Acetic Acid Bacteria Under Cocoa Pulp Fermentation

Home , Acetoin, Acetoin dehydrogenase

2 The key to acetate: Metabolic fluxes of acetic acid bacteria under cocoa pulp

3 fermentation simulating conditions

5 Philipp Adlera, Lasse Jannis Freya, Antje Bergera, Christoph Josef Boltenb, Carl Erik

6 Hansenb and Christoph Wittmanna,c*

9 a Institute of Biochemical Engineering, Technische Universität Braunschweig, 38106

10 Braunschweig, Germany

11 b Nestlé Research Center, Vers-Chez-Les-Blanc, 1000 Lausanne 26 Switzerland

12 c Institute of Systems Biotechnology, Saarland University, 66123 Saarbrücken,

13 Germany

17 *Corresponding address: Christoph Wittmann, Institute of Systems Biotechnology,

18 Saarland University, Campus A 1.5, 66123 Saarbrücken, Germany, phone 0049-681-

19 30271971, FAX: 0049-681-30272972, Email: [email protected]

22 Abstract

23 Acetic acid bacteria (AAB) play an important role during cocoa fermentation, as their

24 main product acetate is a major driver for development of desired cocoa flavors. Here,

25 we investigated the specialized metabolism of these bacteria under cocoa pulp

26 fermentation simulating conditions. A carefully designed combination of parallel 13C

27 isotope experiments allowed the elucidation of intracellular fluxes in the complex

28 environment of cocoa pulp, including lactate and ethanol as primary substrates,

29 among undefined ingredients. We demonstrate that AAB exhibit a functionally

30 separated metabolism during co-consumption of carbon-two and carbon-three

31 substrates. Acetate is almost exclusively derived from ethanol, while lactate serves for

32 formation of acetoin and biomass building blocks. Although suboptimal from cellular

33 energetics, this allows maximized growth and conversion rates. The functional

34 separation results from lack of phosphoenolpyruvate carboxykinase and malic

35 enzymes, typically present in bacteria to interconnect metabolism. In fact,

36 gluconeogenesis is driven by pyruvate phosphate dikinase. Consequently, a balanced

37 ratio of lactate and ethanol is important for optimum performance of AAB. As lactate

38 and ethanol are individually supplied by lactic acid bacteria and yeasts, during the

39 initial phase of cocoa fermentation, respectively, this underlines the importance of a

40 well-balanced microbial consortium for a successful fermentation process. Indeed,

41 AAB performed best and produced the largest amounts of acetate in mixed culture

42 experiments, when lactic acid bacteria and yeasts were both present.

45 Key words

46 Metabolic flux, 13C, fluxome, systems biology, acetate, cocoa

47 2

48 Introduction

49 Acetic acid bacteria (AAB) play an important role in cocoa fermentation (1). During

50 fermentation of pulp, surrounding the cocoa beans, they form acetate. Acetate then

51 diffuses into the beans (2–4), where it initiates a cascade of chemical and biochemical

52 reactions leading to precursor molecules for cocoa flavor (2, 5, 6). Potential

53 substrates for AAB are lactate and ethanol, which are individually produced by lactic

54 acid bacteria (LAB) (mainly Lactobacillus fermentum) and yeasts (diverse yeasts such

55 as Saccharomyces cerevisiae, Hanseniaspora oppuntiae, and Candida crusei) during

56 the fermentation process (6–12). Hereby, degradation of lactate by AAB is desired,

57 since remaining lactate may provide off-flavor in the final cocoa product (11, 13, 14).

58 In recent years, AAB were extensively analyzed for their contribution to cocoa

59 fermentation. Obviously, the most prevalent AAB species is Acetobacter pasteurianus

60 (13, 15–17). In addition, A. ghanensis and A. senegalensis are found during

61 spontaneous cocoa bean fermentation (9, 13, 17, 18). Further studies provided first

62 insights into basic microbiological properties of such strains and macroscopic

63 dynamics during cocoa pulp fermentation (12, 15, 18–20). At this point, it appears

64 relevant to resolve the metabolic contribution of AAB to a greater detail. Basic

65 knowledge that we share indicates a unique metabolism among members of the

66 genus Acetobacter, including a non-functional Embden-Meyerhof-Parnas (EMP)

67 pathway and the generation of energy from incomplete oxidation of ethanol into

68 acetate (21, 22). A few species are able to catabolize lactate (23–26). A pathway from

69 lactate to acetate as final product has been proposed, however the role of some of its

70 enzymes remains unclear (26). In general, little is known about the in vivo contribution

71 of these carbon core pathways to cellular metabolism of AAB. Particularly, their life

72 style in complex environments such as cocoa pulp is largely unknown. In the past

73 decade, 13C metabolic flux analysis has emerged as a routine approach to study 3

74 individual microbes grown on single-carbon substrates on the level of molecular

75 carbon fluxes (in vivo reaction rates) (27, 28). Without resolving details, advanced

76 experimental designs meanwhile provide fluxes for complex microbial systems,

77 including mixed populations (29) and nutrient mixtures (30).

79 Here, we applied 13C based metabolic flux analysis to study the metabolism of

80 Acetobacter during cocoa pulp fermentation. Acetobacter pasteurianus NCC 316 and

81 A. ghanensis DSM 18895 were selected as representative strains of two major

82 species of AAB, naturally occurring in cocoa pulp (23, 31). A validated set-up (12)

83 unraveled metabolic fluxes in AAB under cocoa pulp simulating conditions. As central

84 finding, concerted use of lactate and ethanol enables optimum fluxes with regard to

85 growth and energy metabolism. Considering recent insights into extracellular and

86 molecular fluxes of LAB during cocoa fermentation (12, 20, 23, 30) this suggests that

87 the efficiency of AAB depends on compositional traits provided by LAB and yeasts

88 during the initial fermentation process so that the overall success of cocoa

89 fermentation requires a well-balanced microbial consortium.

91 Material and Methods

92 Strains and maintenance. Acetobacter pasteurianus NCC 316, Lactobacillus

93 fermentum NCC 575 and Saccharomyces cerevisiae NYSC 2 were obtained from the

94 Nestlé Culture Collection (Lausanne, Switzerland). A. pasteurianus NCC 316 and L.

95 fermentum NCC 575 were previously isolated from cheese (1978) and from coffee

96 (1984), respectively. S. cerevisiae NYSC 2 is the commonly used laboratory strain

97 X2180-1B (MATα gal2 SUC2 mal CUPJ). Acetobacter ghanensis DSM 18895 was

98 obtained from the German Culture Collection (DSMZ, Braunschweig, Germany).

99 Escherichia coli K-12 was used as a reference strain for enzyme activity assays and 4

100 was supplied by the Coli Genetic Stock Center (CGSC, New Haven, CT, USA). A.

101 pasteurianus NCC 316 and A. ghanensis DSM 18895 were stored in mannitol – yeast

102 extract – peptone (MYP) medium (32). L. fermentum NCC 575 was maintained in

103 Lactobacilli MRS (deMan-Rogosa-Sharpe) broth (32). S. cerevisiae NYSC 2 was

104 stored in yeast malt (YM) medium (32). E. coli K-12 was maintained in LB medium

105 (synonymously called lysogeny broth or Luria Bertani broth) (33). All frozen stocks

106 were stored at -80°C and contained 15% (vol/vol) glycerol.

107

108 Cultivation. Cells of A. pasteurianus NCC 316 and A. ghanensis DSM 18895 were

109 re-activated on MYP agar at 30°C for 48 h before use. For pre-cultures and main

110 cultivation of Acetobacter species, the cocoa pulp simulation medium for acetic acid

111 bacteria was used (PSM-AAB) (12). It contained per liter: 10 g ethanol, 7.2 g sodium

112 lactate, 10 g yeast extract (Roth, Karlsruhe, Germany), 5 g soy peptone (Roth), and 1

113 ml Tween 80 (Sigma-Aldrich, Taufkirchen, Germany). The initial pH of the medium

114 was 4.5. In 13C tracer experiments, lactate and ethanol were replaced by equimolar

115 amounts of 98% [U-13C] sodium lactate (Cambridge Isotopes, Andover, MA, USA),

116 98% [3-13C] sodium lactate (Cambridge Isotopes), and 99% [U-13C] ethanol (Sigma-

117 Aldrich), respectively. Pre-cultures and main cultures of L. fermentum NCC 575 and

118 S. cerevisiae NYSC 2 were performed in cocoa pulp simulation medium for lactic acid

119 bacteria (PSM-LAB) (12), which was prepared as described previously (30). For all

120 experiments, cells from agar plates, previously obtained from frozen stocks

121 (Acetobacter), or frozen stocks (L. fermentum and S. cerevisiae) were used to

122 inoculate the first pre-culture (37°C, 12-24 h). Cells were transferred to the second

123 pre-culture, which was incubated at 37°C. Cells were then harvested in the mid-

124 exponential growth phase, washed (6000 × g, 5 min; 4°C) with a peptone mixture (10

125 g liter−1 yeast extract [Roth], 5 g liter−1 soy peptone [Roth]), and used to inoculate the 5

126 main culture to an initial optical density (OD600) of 0.01. All cultivations of A.

127 pasteurianus NCC 316 and A. ghanensis DSM 18895 were performed in disposable

128 baffled shake flasks (250 ml, PreSens Precision Sensing GmbH, Regensburg,

129 Germany) on a rotary shaker at 280 rpm (Infors Multitron II, Infors, Bottmingen,

130 Switzerland). Dissolved oxygen was monitored using a shake-flask reader (SFR,

131 PreSens Precision Sensing GmbH). To account for potential evaporation of volatiles,

132 their loss was monitored in control experiments without cells, incubated under the

133 fermentation conditions described above and evaporation rates were used to correct

134 formation and consumption rates. Cultivations with L. fermentum NCC 575 and S.

135 cerevisiae NYSC 2 were conducted in non-baffled 250-ml shake flaks at 37°C and a

136 rotation speed of 50 rpm.

137

138 In order to study the entire cocoa pulp fermentation, PSM-LAB was inoculated with

139 either L. fermentum, S. cerevisiae or a mixture of both microorganisms and then

140 incubated for 24 h. Subsequently, the fermentation broth was filtered (filter top 250,

141 0.22 µm, TPP Techno Plastic Products, Trasadingen, Switzerland), the pH was

142 adjusted to 4.5, and the medium was further incubated with A. pasteurianus NCC

143 316.

144

145 Quantification of substrates and products. Concentrations of organic acids

146 (lactate, acetate, pyruvate, citrate, and α-ketoglutarate), acetoin and ethanol in culture

147 supernatants were determined by HPLC (Elite-LaChrom, Hitachi, West Chester, PA,

148 USA), equipped with an Aminex HPX-87H column (300 × 7.8 mm, Bio-Rad, Hercules,

149 CA, USA) as stationary phase, using 12.5 mM H2SO4 as mobile phase at a flow rate

150 of 0.5 ml min−1 and a column temperature of 45°C. Quantification of acetoin was

151 carried out via UV detection at 290 nm. For detection of all other substances, a 6

152 refractive index detector was used. Samples were diluted 1:10 with deionized water

153 prior to analysis.

154

155 Cell growth was monitored as optical density (OD600) at 600 nm (Libra S11,

156 Biochrome, Cambridge, UK). Cell dry weight (CDW) was determined gravimetrically

157 after filtration of culture broth (cellulose acetate, 0.2 µm, Sartorius, Göttingen,

158 Germany) and subsequent drying of the filters at 105°C until weight constancy. The

159 correlation factor between OD600 and CDW (three biological replicates each) was

−1 160 CDW = 0.356 × OD600 (g liter ) for A. pasteurianus NCC 316, and CDW = 0.240 ×

−1 161 OD600 (g liter ) for A. ghanensis DSM 18995.

162

163 GC/MS analysis of 13C labeling pattern of amino acids and of secreted acetoin

164 and acetate. For analysis of 13C labeling patterns, samples were taken from the first

165 exponential growth phase of the AAB. Mass isotopomer distributions of proteinogenic

166 amino acids were analyzed by GC/MS in selective ion monitoring mode (34). The

167 labeling pattern of acetate in culture supernatants was determined by GC/MS as

168 described previously (30). To determine the labeling pattern of acetoin, a novel

169 method was adapted from a protocol for multifunctional carbonyl compounds (35).

170 Shortly, 500 µl of culture supernatant was mixed with 50 µl of a 0.5% (w/vol) aqueous

171 O-(2,3,4,5,6-pentafluorobenzyl)hydroxylamine hydrochloride (PFBHA) solution and

172 incubated for 30 min at 80°C. Subsequently, PFBHA derivatives were extracted with

173 methyl-t-butylether (MTBE). Excess PFBHA was removed by back-extraction with 0.1

174 M HCl (500 µl). To eliminate traces of water in the MTBE extracts, dry sodium sulfate

175 powder was added. Remaining water in the organic phase was trapped into clumps of

176 sodium sulfate. After removal of the particles by centrifugation for a few seconds

177 (Sprout mini-centrifuge, Heathrow Scientific LLC, Vernon Hills, Il, USA), samples were 7

178 derivatized with 50 µl of bis-(trimethylsilyl)trifluoroacetamide (BSTFA) at 80°C for 30

179 min. Subsequently, PFBHA-BSTFA derivatives were analyzed by GC/MS (7890A,

180 5975C quadrupole detector, Agilent Technologies, Santa Clara, CA, USA). The oven

181 program was as follows: 60°C for 2 min, ramp: 15°C min−1, final temperature: 325°C.

182 Samples were analyzed in SIM mode at m/z 340-344 (C1-C4) and m/z 312-314 (C3-

183 C4). For method validation, a 0.05% (w/v) solution of naturally labeled acetoin was

184 treated and analyzed as described above.

185

186 Enzyme activity assays. For enzyme assays, cells from the first exponential growth

187 phases were harvested by centrifugation (25800 x g, 4°C, 10 min), washed with 100

188 mM Tris-HCl (pH 7.5, 10 mM MgCl2, 0.75 mM dithiothreitol) and re-suspended in 10

189 ml of the same buffer. Disruption was performed mechanically (100 µm silica glass

190 beads, 2 x 20 s, 6.0 m s-1, FastPrep-24, MP Biomedicals, Santa Ana, CA, USA). Cell

191 debris was removed by centrifugation at 17,000 x g for 5 min at 4°C and protein

192 content in crude extract was determined by the Bradford method (36) using a reagent

193 solution (Rothi-Quant, Roth) with bovine serum albumin as standard. Enzyme assays

194 were performed at room temperature in a microplate reader (Varioskan Flash,

195 Thermo Scientific, Waltham, MA, USA). Phosphoenolpyruvate carboxykinase (EC

196 4.1.1.32) and phosphoenolpyruvate carboxylase (EC 4.1.1.31) assays were

197 performed as described previously (37). Malic enzyme activity (EC 1.1.1.40) was

198 determined as described previously (38). As a positive control, an isocitrate

199 dehydrogenase assay was used (39). Enzyme activities were calculated from change

-1 -1 200 in absorbance. Molar extinction coefficients were ε340 = 6.22 mM cm for NADH +

+ + -1 -1 201 H and NADPH + H , and ε300 = 0.6 mM cm for oxaloacetate, respectively

202

203 Metabolic network. 8

204 The metabolic pathway network of Acetobacter (Figure 1) was derived from the

205 genomic repertoire of A. pasteurianus IFO 3283 and A. pasteurianus 386B, recently

206 sequenced (21, 40). A. pasteurianus possesses membrane-bound, pyrroloquinoline

207 quinone (PQQ) dependent alcohol and aldehyde dehydrogenases for oxidation of

208 ethanol into acetate, as well as genes encoding intracellular alcohol dehydrogenase

209 and aldehyde dehydrogenase (Table S1, Supplementary Material) (21, 40–43).

210 Lactate is metabolized via lactate dehydrogenase and further metabolized into

211 acetate (26). It has been reported that A. pasteurianus strains accumulate acetoin,

212 when incubated with lactate (23–25). Indeed, two different pathways for acetoin

213 formation are annotated. Furthermore, there is genetic evidence for a modified TCA

214 cycle, in which succinyl-CoA synthetase and malate dehydrogenase are replaced by

215 succinyl-CoA:acetate CoA transferase (APA01_00320, APA01_00310,

216 APA386B_2589) and malate:quinone oxidoreductase (APA01_01110, APA01_11550,

217 APA386B_2675) (21, 40). This modified pathway has been described previously(44).

218 The glyoxylate pathway and the Entner-Doudoroff (ED) pathway are not encoded in

219 the genome of A. pasteurianus IFO 3283 and A. pasteurianus 386B, respectively (21,

220 40). There are predicted genes for phosphoenolpyruvate carboxylase in the genome

221 sequences of A. pasteurianus IFO 3283 and A. pasteurianus 386B (21, 40).

222 Additionally, activity of phosphoenolpyruvate carboxykinase is predicted putatively

223 (APA01_16650) (40), however, since there is no experimental evidence, it was not

224 considered in the network. Due to absence of phosphofructokinase, the EMP pathway

225 is non-functional (21).

226

227 Estimation of metabolic fluxes.

228 Metabolic flux distributions were estimated with OpenFLUX v 2.1 (45). Simulated

229 mass isotopomer data were corrected for natural occurrence of isotopes in 9

230 derivatization residues and non-carbon atoms (46), whereby the natural abundance of

231 isotopes reflected values from laboratory chemicals (47). For statistical evaluation of

232 the obtained fluxes, Monte-Carlo analysis was performed with 100 individual

233 parameter estimations, assuming Gaussian noise for the experimental data (48).

234

235 Results

236 Computer-based experimental design predicts optimum combinations of 13C

237 lactate and ethanol to resolve metabolic fluxes in Acetobacter. In the medium

238 that represented cocoa pulp, lactate and ethanol were the primary substrates, among

239 complex ingredients. In order to elucidate metabolic fluxes in AAB growing on this

240 complex nutrient mixture, a strategy of combined isotope experiments was developed

241 (Figure 2). Of central importance was the identification of suitable tracer substrates to

242 resolve the fluxes of interest. In a set of computer simulations with varied

243 combinations of 13C labelled lactate and ethanol, excellent resolution of metabolic

244 fluxes was obtained by a combination of [U-13C] lactate, [U-13C] ethanol, and [3-13C]

245 lactate. The relative contribution of de-novo synthesis and uptake from the medium to

246 amino acid supply could be quantified using a mixture of [U13C] lactate and [U13C]

247 ethanol, as deduced previously (49). Individual contribution of lactate and ethanol to

248 central metabolism was elucidated by applying mixtures of one universally labeled

249 and one naturally labeled substrate in two separate labeling experiments. For a more

250 detailed resolution of pathway fluxes, a mixture of [3-13C] lactate and naturally labeled

251 ethanol was found adequate. Taken together, metabolic flux analysis of the acetic

252 acid bacteria integrated labeling information from four parallel tracer studies (Figure

253 2).

254

255 Quantification of the labeling profile of acetoin. Previously developed protocols

256 allow precise labeling analysis of proteinogenic amino acids (50) and of secreted

257 acetate (30) from isotope experiments. In order to provide the full picture, a method

258 for labeling analysis of secreted acetoin, another typical product of acetic acid

259 bacteria, had to be developed. Tests with different reagents and incubation conditions

260 (data not shown) finally provided a stable PFBHA-BSTFA derivate of acetoin that

261 eluted as separate peak after 9.4 min, as shown in the GC/MS chromatogram (Figure

262 3 A). Accuracy of the measurement was validated by analyzing naturally and

263 universally 13C labeled acetoin, respectively. Resulting mass isotopomer distributions

264 perfectly matched with theoretical values, derived from the natural isotope abundance

265 (47) (Table 1). Potential isobaric overlay of acetoin ion clusters with sample matrix

266 was excluded by comparative analysis of acetoin from supernatant of a culture of A.

267 pasteurianus NCC 316 grown on naturally labelled substrates (Table 1). Accordingly,

268 the developed method was appropriate to quantify the labeling of acetoin from isotope

269 experiments in cocoa pulp simulation medium.

270

271 Growth physiology of A. ghanensis DSM 18895 and A. pasteurianus NCC 316.

272 A. pasteurianus NCC 316 co-consumed lactate and ethanol within 8 h of fermentation

273 (Figure 4). During the first growth phase, acetate was the main product with a total

274 amount of 140 mM. Additionally, acetoin (18 mM) and pyruvate were produced (13

275 mM). Biomass reached a concentration of 1.4 CDW g liter−1 and the pH remained

276 relatively stable at 4.3. The fluxes of substrate uptake and product formation during

277 this phase are listed in Table 3. Concentration of dissolved oxygen (DO) decreased

278 and reached a minimal level of 8% after 8 h of cultivation, which ensures fully aerobic

279 conditions during the process. With depletion of lactate and ethanol, the dissolved

280 oxygen level sharply increased. A. pasteurianus NCC 316 then revealed an 11

281 intermediate lag-phase (Figure S1, Supplementary Material). After 20 h, cells started

282 to metabolize acetate, indicated by the decrease of DO. Further on, biomass

283 increased to a maximum of 2.7 g liter−1 and the final pH after 54 h of cultivation was

284 8.0.

285

286 A. ghanensis DSM 18895 exhibited similar fermentation characteristics (Figure S2,

287 Supplementary Material). During the first phase of fermentation, lactate and ethanol

288 were co-utilized. The concentration of acetate at the end of this growth phase (9 h)

289 was 139 mM, while the DO reached a minimum of 28%, biomass concentration was

290 1.0 g liter−1 and the pH was 4.4. Additionally, 12 mM of acetoin, but no significant

291 amounts of pyruvate were formed. While ethanol had been exhausted completely

292 when growth stopped, considerable amounts of lactate (22 mM) remained in the

293 medium and were not metabolized further during the transient lag-phase. During a

294 second growth phase, lactate and acetate were then metabolized and biomass

295 concentration reached a maximum of 1.9 g liter−1. The final pH was 8.5 after 54 h of

296 fermentation.

297

298 Overall, both AAB species preferred a combination of lactate and ethanol for their

299 metabolism. However, in the absence of lactate and ethanol, cells could re-utilize

300 acetate.

301

302 Metabolic origin of proteinogenic amino acids. In a first set of isotope

303 experiments, cells were grown on fully 13C labelled lactate or ethanol, respectively,

304 together with non-labelled complex nutrients. According to the experimental design,

305 the 13C labeling pattern of amino acids from cell protein directly provided information

306 on their metabolic origin, i.e. relative contribution of either lactate or ethanol to de- 12

307 novo biosynthesis of amino acids during the first exponential growth phase. GC/MS

308 analysis of proteinogenic amino acids of A. ghanensis DSM 18895 and A.

309 pasteurianus NCC 316 revealed a clear picture (Table S5, Supplementary Material).

310 Aromatic amino acids and amino acids of the serine family were exclusively produced

311 from lactate. These amino acids were naturally labeled, when [U-13C] ethanol was

312 present in the fermentation medium, but were enriched in 13C, when [U-13C] lactate

313 was used. Likewise, also amino acids of the pyruvate family were almost exclusively

314 derived from lactate. This was also observed for lysine and isoleucine, which

315 predominantly originated from lactate. Ethanol contributed only to a small set of

316 biosynthetic pathways. Amino acids, stemming from the TCA cycle, including

317 compounds that belong to the aspartate and the glutamate family were synthesized

318 form both, lactate and ethanol.

319

320 Contribution of lactate and ethanol to the major fermentation products of A.

321 pasteurianus NCC 316 and A. ghanensis DSM 18895. The same isotope

322 experiments were next inspected for contribution of lactate and ethanol to the major

323 products: acetate and acetoin. For this purpose, labeling of the latter was quantified

324 from culture supernatants that were derived from cultures with cocoa pulp simulation

325 medium (first exponential growth phase), containing [U-13C] lactate or [U-13C] ethanol,

326 respectively, as trace (Tables S7 and S8, Supplementary Material). Taken together,

327 both microorganisms produced most of acetate (96-98%) from ethanol (Figure 5B).

328 Analogously, acetoin resulted from metabolism of lactate. Together with the picture

329 from amino acid biosynthesis, this indicated a surprising separation of metabolism.

330 Both strains exhibited two functional pathway modules, related to the utilization of

331 lactate and ethanol with only weak exchange fluxes.

332 13

333 Enzyme repertoire of A. pasteurianus NCC 316 and A. ghanensis DSM 18895.

334 The obviously poor interconnection of the metabolism of lactate and ethanol raised

335 questions on the enzymatic set-up around the PEP-pyruvate-acetyl CoA node, where

336 the two carbon sources are channeled into core metabolism (Figure 1). A.

337 pasteurianus NCC 316 and A. ghanensis DSM 18895 were tested positive for

338 isocitrate dehydrogenase, an enzyme of the oxidative TCA cycle (Table 2). In both

339 strains, PEP carboxylase was active as anaplerotic enzyme. PEP carboxykinase and

340 malic enzyme, enzymes of gluconeogenesis, however, were not expressed (Table 2).

341

342 Impact of the microbial community on metabolite profiles in simulated cocoa

343 pulp fermentations. A set of experiments investigated performance of AAB during an

344 entire two-step cocoa fermentation process. The study combined three different set-

345 ups for the first phase (only L. fermentum, only S. cerevisiae, L. fermentum plus S.

346 cerevisiae) with a second fermentation phase, inoculated with A. pasteurianus NCC

347 316. In all cases, A. pasteurianus NCC 316 was grown on a filtrate of pre-fermented

348 medium from one of the first-phase incubations. High amounts of lactate and acetate,

349 but no ethanol was obtained, when L. fermentum NCC 575 was used as a starter in

350 the first phase of simulated cocoa pulp fermentation (Table 4). Citrate, which was part

351 of the simulation medium, was largely metabolized by the lactic acid bacteria. The

352 second fermentation phase using A. pasteurianus NCC 316 as starter strain resulted

353 in a significant reduction of lactate, together with a consumption of acetate, whereas

354 low levels of citrate remained unchanged. A completely different picture was

355 observed, when the cocoa pulp simulating medium was incubated with S. cerevisiae

356 NYSC 2. Here, extremely large amounts of ethanol were formed, whereas only low

357 levels of lactate and acetate resulted. The concentration of citrate remained almost

358 unchanged (12). This mixture appeared not optimal for A. pasteurianus NCC 316. The 14

359 AAB only partially utilized ethanol and formed only small amount of acetate (20 mM).

360 The best performance of A. pasteurianus NCC 316, visualized by largest amounts of

361 acetate were observed, when L. fermentum and S. cerevisiae were co-cultivated

362 during the initial step of fermentation. After the second phase of fermentation a total

363 amount of 121 mM acetate was observed as sole product, while citrate, lactate and

364 ethanol had been consumed almost completely.

365

366

367 Discussion

368 Acetic acid bacteria exhibit a functionally separated metabolism during co-

369 consumption of carbon two and carbon three substrates. Both Acetobacter

370 species synthesized acetate almost exclusively from ethanol, while only 2-4%

371 originated from lactate (Figure 5 B). This was contrary to expectations, because

372 previously it has been suggested that ethanol as well as lactate are converted by A.

373 pasteurianus 386B into acetate (12), but is in line with recent indications from

374 metabolite balancing (23). Based on 13C labeling, our results clearly show that lactate

375 is poorly oxidized into acetate. Vice versa ethanol does not contribute to acetoin

376 formation. The latter is completely derived from lactate via pyruvate as a precursor

377 (Figure 1). Furthermore, our results suggest that the pyruvate decarboxylase pathway

378 is not involved in acetoin formation, since this pathway would use acetaldehyde,

379 produced during assimilation of ethanol (51). Instead, acetoin is synthesized solely via

380 α-acetolactate. Absence of pyruvate decarboxylase activity (low activity) seems likely

381 responsible for the low contribution of lactate to acetate formation.

382

383 From an energetic perspective, complete oxidation of lactate via the TCA cycle seems

384 optimal (4 moles ATP per mole of lactate) (Table 5). In contrast, conversion of lactate

385 into acetoin leads to low ATP formation. Further experiments showed that lactate-

386 grown A. pasteurianus NCC 316 indeed shifts from acetoin formation to complete

387 oxidation of lactate into CO2, when ethanol is omitted from the medium (data not

388 shown). We conclude that the TCA cycle is suppressed in the presence of ethanol

389 (52) and that acetoin is an overflow metabolite of pyruvate. Nevertheless, acetoin

390 formation may be advantageous for cellular homoeostasis, as this compound is less

391 toxic than the organic acid. In this regard, acetoin formation could be favorable over

392 acetate production. Analogously, oxidation of ethanol into acetate and formation of 16

393 acetoin from lactate seem to be inefficient from an energetic point of view (Table 5).

394 However, the cells metabolized both products in a second growth phase (Figure S1)

395 without significant loss of energy compared to the total oxidation of lactate and

396 ethanol (Table 5). Additionally, complete oxidation of lactate via acetate as an

397 overflow metabolite is less efficient (2 moles of ATP per mole of lactate) than

398 complete oxidation of lactate via acetoin (3 moles of ATP per mole of lactate). All this

399 together indicates that A. pasteurianus NCC 316 and A. ghanensis DSM 18895 have

400 adapted their metabolism to a strategy which aims at optimization of speed rather

401 than efficiency (53).

402

403 As a central finding, our data demonstrate a strict separation of carbon-two (ethanol,

404 acetaldehyde, acetate) and carbon-three (lactate, pyruvate, phosphoenolpyruvate)

405 metabolism in AAB. As example, this becomes obvious from the biosynthesis of

406 amino acids. These are synthesized from precursors of the TCA cycle, from the

407 pentose phosphate pathway and from gluconeogenesis (Figures 1 and 4A).

408 Interestingly, all amino acids stemming from intermediates upstream of pyruvate are

409 produced exclusively from lactate during the first growth phase. This confirms that

410 carbon from ethanol does not enter this upper part of metabolism. As exception of the

411 otherwise strongly separated functional modules of lactate and ethanol metabolism, a

412 minor fraction of lactate-derived carbon replenishes the TCA cycle. Such a separated

413 metabolism is rarely observed in prokaryotic cells and rather a property of eukaryotes,

414 which possess cellular compartments (54). One rare example of a bacterium with a

415 disconnected metabolism is L. fermentum (30, 55). Here, glucose is an energy

416 source, while fructose is entirely used as electron acceptor for NADH + H+. Thus,

417 separation of metabolism seems to be a property of bacteria with a rather specialized

418 metabolism. AAB are specialized to membrane-bound oxidating systems for acetate 17

419 production (41, 56, 57). These systems are directly coupled to respiratory chains,

420 located at the cytoplasmic membrane and allow oxidation of external substrates in a

421 rather simple manner (41). Since reaction takes place in the periplasm, no further

422 transport across the membrane into the cytoplasm is required. Thus, although this

423 incomplete oxidation is apparently inefficient (Table 5), it enables bacteria to generate

424 energy with a small set of enzymes. In contrast, channeling of ethanol or acetate into

425 central metabolism via acetyl-CoA synthetase costs two extra moles of ATP (22, 52,

426 53). Additionally, the glyoxylate shunt is missing in A. pasteurianus (21, 40) and A.

427 ghanensis (31). As a consequence, ethanol cannot serve as an efficient source for

428 precursors that drain into biomass such as oxaloacetate, phosphoenolpyruvate (PEP)

429 and pyruvate. Analogously, it was shown previously that A. ghanensis DSM 18895

430 and A. pasteurianus NCC 316 did not grow on ethanol with ammonium as the only

431 nitrogen source (31). However, due to direct drain of intermediates from TCA cycle to

432 biomass, amino acids are partly derived from ethanol (Figure 5A). Although, as

433 discussed above, a part of ethanol replenishes the TCA cycle, gluconeogenic

434 intermediates are not derived from ethanol.

435

436 Functional separation of carbon two and the carbon three metabolism in acetic

437 acid bacteria under pulp simulating conditions is created by a lack of

438 phosphoenolpyruvate carboxykinase and malic enzyme. The link between

439 metabolism of carbon-two compounds such as ethanol and of carbon-three

440 compounds through gluconeogenesis, respectively, are cataplerotic reactions of

441 phosphoenolpyruvate carboxykinase (PEPCK) and malic enzyme (ME) (Figure 1).

442 These reactions convert intermediates from TCA cycle (oxaloacetate, malate) into

443 PEP, which in turn enters the gluconeogenesis.

444 18

445 A deeper inspection of labeling patterns provides direct evidence on their in vivo

446 significance. In the case of a back flux from oxaloacetate to PEP, one would expect

13 447 equal C enrichment in carbons C1 and C2 of oxaloacetate and in corresponding

448 carbons of PEP. In contrast to significant enrichment in the oxaloacetate pool of A.

449 pasteurianus NCC 316, the PEP pool is completely non labeled (Figure 6 A). We

450 conclude that the flux through PEPCK is zero. In agreement, cells do not exhibit

451 activity for this enzyme (Table 2). These results suggest that, in fact, the genome of

452 A. pasteurianus NCC 316 is lacking PEPCK. Also, ME is not active, so that a potential

453 contribution of TCA cycle metabolites to gluconeogenetic back flux can be excluded.

454 In fact, the upper part of metabolism is directly supplied from lactate, e.g. for

455 biosynthesis of amino acids of the serine and the aromatic family, respectively (Table

456 S5, Supplementary Material). In this regard, Acetobacter species can make use of

457 pyruvate phosphate dikinase, which converts pyruvate directly into PEP (58, 59).

458 Notably, this reaction requires two moles of ATP. Figure 6B shows that the 13C

459 labeling patterns of pyruvate and PEP from isotope experiments are virtually identical.

460 Similarly, A. aceti lacks oxaloacetate decarboxylation during growth on carbon-three

461 substrates, while pyruvate phosphate dikinase is active (58, 60). Thus, pyruvate

462 phosphate dikinase seems to play an important role for gluconeogenesis of A.

463 pasteurianus NCC 316 and A. ghanensis DSM 18895. Analogously to A. aceti (61),

464 anaplerotic phosphoenolpyruvate carboxylase (PEPC) was present in A. pasteurianus

465 NCC 316 and A. ghanensis DSM 18895 (Table 2).

466

467 Different enzymes form the metabolic link between glycolysis, gluconeogenesis and

468 TCA cycle (Figure 7). In general, this node is flexible to adjust metabolism to anabolic

469 and energetic demands under a given condition (27). Additionally, in some bacteria,

470 carbon flux control through the PEP-pyruvate-oxaloacetate node is complex, e.g. PC 19

471 and ME are simultaneously active during growth on glucose (62, 63). In contrast, A.

472 pasteurianus NCC 316 and A. ghanensis DSM 18895 use a rather reduced enzymatic

473 set-up. For instance, to form PEP out of pyruvate, Corynebacterium glutamicum uses

474 pyruvate carboxylase (PC) and PEPCK (Figure 7). In this set of reactions, two moles

475 of ATP are consumed in order to produce one mole of PEP. Although genes for these

476 enzymes are annotated in the genomes of A. pasteurianus IFO 3283_01 and 386B

477 (Table S1), this mode is not used by A. pasteurianus NCC 316 and A. ghanensis

478 DSM 18895. This seems reasonable, because in the presence of an active pyruvate

479 phosphate dikinase lack of PEPCK prevents futile cycling. Due to this reduced

480 enzymatic set-up, however, there is no interconnection between carbon-three and

481 carbon-two substances in these microorganisms, at least not under the conditions

482 studied. Switching the regulation of malic enzyme expression could be the key event

483 to switch to different metabolic modes, probably needed in other environments.

484

485 The full picture - metabolic flux distribution of A. pasteurianus.

486 Information on the metabolic repertoire from genome annotation (Figure 1), enzyme

487 measurements (Table 2), together with experimental data on growth and product

488 formation (Table 3) and 13C patterns from the parallel isotope studies (Table S12) was

489 now integrated. Based on experimental findings, the metabolic network could be

490 condensed into a model (Table S11, Supplementary Material) that allowed high

491 resolution flux estimations (more detailed information is available in the Appendix).

492 The resulting metabolic flux distribution is shown in Figure 8. Ethanol is converted

493 almost completely into acetate. Because, corresponding dehydrogenases are directly

494 linked to the respiratory chain in A. pasteurianus (41, 56), one molecule of ethanol

495 oxidized into acetate account for two electron pairs. With a P/O ratio of 0.5 (53), this

496 corresponds to 62 mmol ATP per 100 mmol of total substrate influx and 53% of total 20

497 ATP formation. The TCA cycle contributed to 39% of total ATP formation, which is

498 composed of 17% ATP from ethanol metabolism and 23% ATP from lactate

499 metabolism. Lactate conversion into acetoin is responsible for 8% of the ATP pool.

500 Thus, ethanol is the major energy source of A. pasteurianus NCC 316 under

501 conditions of cocoa fermentation. In contrast, lactate is the main source for the

502 pyruvate pool, which in turn supplies the cell with precursors for biomass compounds

503 such as amino acids. However, the major part of pyruvate (64% of the total pyruvate /

504 PEP pool) is directed towards acetoin forming pathway(s). As a result, carbon flux

505 through the TCA cycle is only 15% of the total carbon uptake. This might be due to

506 suppression of the TCA cycle in the presence of ethanol (52). Considering the fully

507 aerobic growth conditions, this suggests that acetoin is formed as an overflow-

508 metabolite. Additionally, decreased activity of the TCA cycle prevents a catabolic

509 breakdown of acetate.

510

511 In addition to lactate and ethanol as main energy and carbon sources, respectively,

512 some amino acids played a superior role in central metabolism of A. pasteurianus

513 NCC 316 (Figure 8). A small portion of alanine is catabolized and supports the

514 pyruvate pool. Obviously, aspartate serves as a source of oxaloacetate and

515 replenishes the TCA cycle, replacing anaplerosis via PEP carboxylase (PEPC).

516 Glutamate is involved in biosynthetic reactions as a donor of amino groups (64),

517 therefore there is a steady interconversion between glutamate and its keto acid, α-

518 keto glutarate. Although alanine, aspartate and glutamate are metabolized,

519 biosynthesis of these amino acids is active, suggesting that the involved biosynthetic

520 reactions are constitutive in A. pasteurianus NCC 316.

521

522 Bacteria require NADPH + H+ for many biosynthetic processes. The cofactor is

523 typically formed through the pentose phosphate pathway (glucose 6-phosphate

524 dehydrogenase) and the TCA cycle (isocitrate dehydrogenase) (65–67). Since the

525 pentose phosphate pathway is poorly active in A. pasteurianus NCC 316, we

526 wondered, how A. pasteurianus covers its biosynthetic demand for NADPH + H+.

527 Considering cellular composition (68) and incorporation of amino acids from the

528 extracellular environment, the biosynthetic NADPH + H+ requirement equals

529 approximately 7 mmol/g CDW. Due to low biomass yield (Table 3), this corresponds

530 to only 5% of total substrate uptake and 7% of ethanol conversion. In contrast, E. coli

531 requires 16-17mmol NADPH + H+ per g CDW, which corresponds to 140% of the

532 glucose uptake flux (69). Through isocitrate dehydrogenase, the TCA cycle provides

533 A. pasteurianus NCC 316 with 22 mmol of NADPH + H+ per g CDW (Figure 8).

534 Additionally, A. pasteurianus is able to produce NAD(P)H + H+ via soluble alcohol and

535 aldehyde dehydrogenases (21, 51). Among species of Acetobacter, only a soluble

536 aldehyde dehydrogenase of Acetobacter rancens (later synonym of A. pasteurianus),

537 which requires NADP+, has been characterized (70). Furthermore, an NAD+-

538 dependent soluble alcohol dehydrogenase was identified in A. pasteurianus (51). This

539 indicates that additionally 15 mmol of NADPH + H+ per g CDW are formed from

540 intracellular metabolic breakdown of ethanol. Hence, the biosynthetic NADPH

541 demand of A. pasteurianus NCC 316 is fully covered, in fact there seems to be an

542 apparent excess of NADPH + H+. A proton-translocating nicotinamide nucleotide

543 transhydrogenase might be involved in NADPH oxidation (21).

544

545 Taken together, the metabolic flux distribution highlights specific roles of lactate and

546 ethanol for central metabolism of A. pasteurianus NCC 316 during cocoa pulp

547 fermentation. In this constellation, the main role of ethanol is to generate metabolic 22

548 energy via acetate production. While lactate is also an energy source, it additionally

549 supports growth by forming precursors, particularly via gluconeogenesis and pentose

550 phosphate pathway. Thereby, acetate production is intensified, which is one of the

551 main purposes of cocoa fermentation. Consequently, a balanced ratio of lactate and

552 ethanol is important for success of the fermentation process.

553

554 Species interactions determine the quality of cocoa fermentation. The formation

555 of adequate amounts of acetate is critical for well-fermented cocoa beans (1, 2),

556 Among the simulated triculture cocoa pulp fermentation, the largest amount of acetate

557 is obtained when L. fermentum NCC 575 and S. cerevisiae NYSC 2 are co-cultivated

558 in the first phase of fermentation. LAB produce lactic acid, which supports growth of

559 A. pasteurianus NCC316. Ethanol supplied by yeasts is required for synthesis of

560 acetate by the acetic acid bacteria and suppression of acetate-degrading enzymes.

561 Furthermore, LAB reduce the amount of citrate and AAB degrade lactate, which

562 positively affects the flavor of the cocoa beans (14). When the cultivation is performed

563 with only L. fermentum NCC 575 in the first phase of cultivation, A. pasteurianus

564 NCC316 even degrades acetate. For this reason, high levels of ethanol, produced

565 during the initial phase of cocoa pulp fermentation, seem to be important in order to

566 repress breakdown of acetate. Similarly, low amounts of acetate are produced, when

567 only yeast is present in the first phase of simulated cocoa fermentation, since AAB

568 form little biomass on ethanol (53, 71). These findings clearly show that the pathway

569 activities of LAB, AAB and yeast is important for a successful fermentation of cocoa

570 pulp (11, 23, 72).

571

572 Acknowledgements

573 We thank Stéphane Duboux and Christof Gysler for their support in providing strains

574 of the Nestlé Culture Collection. We further appreciate the technical assistance of

575 Elena Kempf, Sandra Hübner and Martin Wewers.

576

577

578

579 Appendix: Metabolic model for flux calculations. The metabolic model for flux

580 estimations was derived from the genomic repertoire (see Figure 1) and experimental

581 findings of this study. Extracellular fluxes are given in Table 3. Since PEPCK and ME

582 were found inactive, these reactions were not included. Furthermore, the

583 gluconeogenesis and pentose phosphate pathway were lumped, since these

584 pathways could not be distinguished in this complex environment. Additionally, the

585 gluconeogenesis, pentose phosphate pathway and TCA cycle were considered

586 irreversible. Because acetoin was produced exclusively from lactate, the acetoin

587 synthesis pathway was simplified. Pyruvate and PEP were pooled, due to potentially

588 parallel activity of PEPC and pyruvate carboxylase. Additionally, malate and

589 oxaloacetate as well as erythrose 4-phosphate and pentose 5-phosphate were

590 lumped, respectively. The drain of precursors to biomass was derived from the

591 bacterial cell composition (68). Additionally, the degrees of de novo synthesis of

592 amino acids were considered, as determined from the experiments. The degrees of

593 synthesis of some amino acids (arginine, cysteine, methionine, and proline) were not

594 determined directly but deduced from other amino acids of the same families. The

595 contribution of histidine to the biomass was negligible (68). The mass isotopomer

596 distributions of the [M-57] fragments of alanine, valine, glutamate and aspartate, the

597 [M-95] fragment of glutamate, the fragments at m/z = 302 of valine, aspartate, and

598 additionally the labeling of acetate were used to fit the model. For this reason, the

599 corresponding uptake fluxes of the amino acids were determined by the model fit.

600 Additionally, deamination of alanine, aspartate and glutamate was considered in the

601 metabolic model.

602

603 Identical flux distributions, obtained from diverse parameter optimization runs with

604 multiple initialization values, indicated that a global optimum was found. The mean

605 squared error between experimental and simulated mass isotopomer distributions

606 was 4.2·10-4, indicating an excellent fit. From Monte-Carlo simulations an averaged

607 error of 0.88 of the resulting metabolic fluxes was obtained.

608

609

610

611

612 Nomenclature

613 3PG 3-phosphoglycerate

614 AAB acetic acid bacteria

615 AcCoA acetyl coenzyme A

616 Ala alanine

617 aKG alpha-ketoglutarate

618 Asp aspartate

619 BSTFA bis-(trimethylsilyl)trifluoroacetamide

620 DO dissolved oxygen

621 E4P erythrose 4-phosphate

622 GC/MS gas chromatography / mass spectrometry

623 Glu glutamate

624 Gly glycine

625 Ile isoleucine

626 LAB lactic acid bacteria

627 Leu leucine

628 Lys lysine

629 MDH malate dehydrogenase

630 ME malic enzyme

631 MTBE methyl-t-butylether

632 OAA oxaloacetate

633 P5P pentose 5-phosphate

634 PC pyruvate carboxylase

635 PEP phosphoenolpyruvate

636 PEPC phosphoenolpyruvate carboxylase

637 PEPCK phosphoenolpyruvate carboxykinase 27

638 PFBHA O-(2,3,4,5,6-pentafluorobenzyl)hydroxylamine hydrochloride

639 Phe phenylalanine

640 PP pathway pentose phosphate pathway

641 Pro proline

642 PSM cocoa pulp simulation medium

643 Pyr pyruvate

644 Ser serine

645 SD standard deviation

646 Tyr tyrosine

647 TCA tricarboxylic acid

648 Valine valine

649

650

651

652

653

654

655

656

657

658 Table 1: Mass isotopomer distribution of acetoin determined by GC/MS. The data

659 comprise means and standard deviations (SD) from duplicate measurements of

660 standard and culture samples and and theoretical values inferred from natural isotopic

661 enrichment in non-labeled substances (47).

662

Fragment Mass Mass isotopomer distribution ± SD isotopomer a fraction Theoretical Unlabeled standard Unlabeled sample

M-15 m+0 0.794 0.792 ± 0.000 0.791 ± 0.002

m+1 0.157 0.156 ± 0.000 0.157 ± 0.000

m+2 0.044 0.044 ± 0.000 0.044 ± 0.000

m+3 0.005 0.006 ± 0.000 0.006 ± 0.000

m+4 0.000 0.003 ± 0.000 0.002 ± 0.002

M-43 m+0 0.811 0.806 ± 0.000 0.807 ± 0.000

m+1 0.142 0.150 ± 0.000 0.151 ± 0.000

m+2 0.041 0.043 ± 0.000 0.042 ± 0.000 663

664 asupernatant of A. pasteurianus NCC 316 grown in cocoa pulp simulation medium.

665

666

667

668

669

670

671

672

673

674

675

676 Table 2: Enzymatic repertoire of A. pasteurianus NCC 316 and A. ghanensis DSM 18895

677 during the first growth phase in cocoa pulp simulation medium.

678

Enzyme activity ± SD (U (g CDW)-1) Enzyme A. pasteurianus NCC 316 A. ghanensis DSM 18895

Isocitrate dehydrogenase 66.9 ± 1.3 51.9 ± 5.3 (positive control)

Malic enzyme - decarboxylation n.d.a n.d.a - carboxylation n.d.a n.d.a

PEP carboxylase 20.3 ± 4.3 65.2 ± 15.3

PEP carboxykinase - decarboxylation n.d.a n.d.a - carboxylation n.d.a n.d.a

679

680 aNot detected. Limit of detection < 1 U (g CDW)-1.

681

682

683

684

685

686

687

688

689 Table 3: Extracellular fluxes of substrates and products, and corresponding standard

690 deviations (SD) determined from biological duplicates. The data reflect average fluxes from

691 the first exponential growth phase of A. pasteurianus NCC 316.

692

693

694

Lactate Ethanol Acetate Acetoin Pyruvatea Extracellular flux ± 41.3 ± 1.5 106.0 ± 2.2 94.0 ± 3.0 14.3 ± 1.3 0 SD (mmol (g CDW)-1) 695

696 aPyruvate was not formed during the exponential growth phase.

697

698

699

700

701

702

703

704 Table 4: Formation of organic acids and ethanol during simulated two-phase cocoa

705 pulp fermentation. For the first phase, PSM-LAB medium was inoculated with L.

706 fermentum NCC 575 and/or S. cerevisiae NYSC 2. The cell-free supernatant of each

707 fermentation broth, obtained after incubation for 24 h, was inoculated with A.

708 pasteurianus NCC 316 and incubated for another 12 h (second phase).

Concentration ± SD (mM)

Citrate Lactate Acetate Ethanol

1st phasea 1.3 ± 0.2 91.1 ± 1.1 106.6 ± 2.0 n.d.d L. fermentum NCC 575 2nd phaseb 1.0 ± 0.2 21.4 ± 1.3 76.0 ± 2.2 n.d. d

1st phasea 47.0 ± 0.4 0.9 ± 0.1 7.5 ± 1.3 360.9 ± 14.5 S. cerevisiae NYSC 2 2nd phaseb 47.1 ± 1.1 0.7 ± 0.1 21.0 ± 5.7 239.2 ± 7.0

L. fermentum 1st phasea 0.2 ± 0.2 104.1 ± 0.2 117.2 ± 3.5 75.4 ± 6.9 NCC 575 + S. cerevisiae 2nd phaseb 0.1 ± 0.1 5.8 ± 6.6 121.5 ± 15.6 n.d. d NYSC 2c 709

710 aAmounts after 24 h of cultivation with L. fermentum NCC 575 and/or S. cerevisiae NYSC 2.

711 bAmounts after 12 h of cultivation with A. pasteurianus NCC 316.

712 cCo-cultivation of both microorganisms.

713 dNot detected. Limit of detection < 0.2 mM.

714

715

716

717 Table 5: Energetic efficiency of different metabolic modes for consumption of lactate,

718 ethanol, acetate and acetoin.

719

Mode of NADH + ATP ATP Pathways / key reactions a metabolism H+ (GTP) equivalents Ethanol : 2 0 1 Membrane bound dehydrogenase Acetate (56, 57)

Ethanol : CO2 6 -1 2 Acetyl-CoA synthetase, TCA (22)

Acetate : CO2 4 -1 1 Acetyl-CoA synthetase, TCA (22) Lactate : 2 0 1 Pyruvate decarboxylase (73) acetate

Lactate : CO2 6 1 4 Pyruvate dehydrogenase, TCA (22) Lactate : 1 0 0.5 Acetolactate synthase (21, 40) acetoin

b Acetoin : CO2 10 0 5 Acetoin dehydrogenase, Acetyl- CoA synthetase, TCA (21, 40) 720 aP/O ratio = 0.5 (53).

721 bEquivalent to two molecules of lactate

722

723

724 Figure Legends

725

726 Figure 1: Metabolic network of the central carbon metabolism of Acetobacter

727 pasteurianus. Abbreviations of metabolites are listed in the nomenclature section of

728 this manuscript. Enzymatic reactions and there corresponding genes are given in

729 Supplementary Material.

730

731 Figure 2: Labeling strategy to elucidate metabolic fluxes of Acetobacter pasteurianus

732 and Acetobacter ghanensis DSM 18895 in cocoa pulp simulation medium using

733 specifically enriched isotopic tracer substances and mass spectral labeling analysis of

734 proteinogenic amino acids and extracellular products. (A) Theoretical 13C mass

735 isotopomer distribution of alanine obtained by de novo synthesis from the labeled

736 substrates ([U-13C] lactate, [U-13C] ethanol) or amino acid uptake (12C). (B) and (C)

737 Contribution of lactate and ethanol to formation of amino acid precursors. (D)

738 Elucidation of specific metabolic routes of [3-13C] lactate utilization. Black circles

739 indicate 13C labeled carbon atoms, and white circles represent unlabeled carbon

740 (12C).

741

742 Figure 3: GC/MS analysis of the 13C pattern of acetoin in culture supernatant. (A)

743 Total ion current of the sample after derivatization with O-(2,3,4,5,6-

744 pentafluorobenzyl)hydroxylamine (PFBHA) and bis-(trimethylsilyl)trifluoroacetamide

745 (BSTFA), with acetoin eluting after 9.42 min, (B) corresponding mass spectrum of

746 PFBHA-BSTFA-acetoin, (C) chemical structure of PFBHA-BSTFA-acetoin and

747 corresponding fragments M-15 and M-43.

748

749 Figure 4: Profile of the first growth phase of A. pasteurianus NCC 316 grown in cocoa

750 pulp simulation medium. The data are means from biological duplicates. Black circles,

751 lactate; white circles, acetate; white diamonds, acetoin; black squares, ethanol; white

752 squares, CDW; white stars, pyruvate; black line, dissolved oxygen.

753

754 Figure 5: (A) Relative contribution of lactate and ethanol to biosynthesis of amino

755 acids. The squares indicate fractions originating from lactate and ethanol that were

756 found in the proteinogenic amino acids of A. pasteurianus NCC 316 (left) and A.

757 ghanensis DSM 18895 (right) during the first growth phase in cocoa pulp simulation

758 medium. The full data are given in Supplementary Material (Table S5). Abbreviations

759 of metabolites are listed in Nomenclature. (B) Relative contribution (%) of lactate and

760 ethanol to formation of acetate and acetoin in A. pasteurianus NCC 316 (upper

761 values) and A. ghanensis DSM 18895 (lower values). Violet circles represent nodes

762 between the modules of lactate and ethanol metabolism.

763

764 Figure 6: Biosynthesis of phosphoenolpyruvate from lactate or ethanol via

765 phosphoenolpyruvate carboxykinase (A) and pyruvate phosphate dikinase (B) by A.

766 pasteurianus NCC 316. (A) Carbon transition in the reaction of phosphoenolpyruvate

767 carboxykinase and experimental labeling profiles of oxaloacetate and

12 13 768 phosphoenolpyruvate resulting from growth on a mixture of [ C3] lactate and [ C2]

769 ethanol. Experimental data are derived from the fragments of aspartate and tyrosine

770 at m/z = 302 (Asp 302, Tyr 302), containing carbon atoms C1 and C2, as described in

771 Supplementary Material. (B) Carbon transition in the reaction of pyruvate phosphate

772 dikinase and experimental labeling profiles of pyruvate and phosphoenolpyruvate,

13 12 773 resulting from growth on a mixture of [ C3] lactate and [ C2] ethanol. Experimental

774 data are derived from the fragments of valine and tyrosine at m/z = 302 (Val 302, Tyr

775 302), containing carbon atoms C1 and C2, as described in Supplementary Material.

776

777 Figure 7: The phosphoenolpyruvate-pyruvate-oxaloacetate node in (A)

778 Corynebacterium glutamicum and Bacillus subtilis, (B) Escherichia coli, and (C) A.

779 pasteurianus. Abbreviations of metabolites are listed in Nomenclature, and enzyme

780 names are given in Supplementary Material.

781

782 Figure 8: Metabolic fluxes of A. pasteurianus NCC 316 during the first growth phase

783 in cocoa pulp simulation medium. The data are given as relative fluxes, normalized to

784 the cumulative uptake flux of lactate and ethanol (Tables S10 and S11,

785 Supplementary Material). The relative flux intensity is illustrated by the arrow

786 thickness and the individual contribution of lactate and ethanol to the fluxes is

787 indicated by the arrow color. The net direction of a reversible reaction is indicated by

788 a small arrow. The fluxes were derived via metabolite and isotopomer balancing

789 recruiting extracellular fluxes (Table S11, Supplementary Material), and 13C labeling

790 pattern of proteinogenic amino acids and acetate. Simulated and experimental mass

791 isotopomer distributions are presented in Figure S3 (Supplementary Material). In

792 addition to the net-fluxes, 90% confidence intervals from Monte-Carlo analysis are

793 shown. Abbreviations of metabolites are listed in Nomenclature, extracellular

794 substrates and products are indicated by [x].

795

796 References 797 798 1. Biehl B, Passern D, Sagemann W. 1982. Effect of acetic acid on subcellular 799 structures of cocoa bean cotyledons. J. Sci. Food Agric. 33:1101–1109. 800 2. de Brito ES, García NHP, Gallão MI, Cortelazzo AL, Fevereiro PS, Braga MR. 801 2001. Structural and chemical changes in cocoa (Theobroma cacao L) during 802 fermentation, drying and roasting. J. Sci. Food Agric. 81:281–288. 803 3. Quesnel VC. 1965. Agents inducing the death of cacao seeds during fermentation. 804 J. Sci. Food Agric. 16:441–447. 805 4. Forsyth WGC, Quesnel VC. 1963. The mechanism of cacao curing. Adv. 806 Enzymol. Relat. Areas. Mol. Biol. 25:457–492. 807 5. Camu N, de Winter T, Addo SK, Takrama JS, Bernaert H, de Vuyst L. 2008. 808 Fermentation of cocoa beans: influence of microbial activities and polyphenol 809 concentrations on the flavour of chocolate. J. Sci. Food Agric. 88:2288–2297. 810 6. Schwan RF, Wheals AE. 2004. The microbiology of cocoa fermentation and its 811 role in chocolate quality. Crit. Rev. Food. Sci. Nutr. 44:205–221. 812 7. Papalexandratou Z, de Vuyst L. 2011. Assessment of the yeast species 813 composition of cocoa bean fermentations in different cocoa-producing regions 814 using denaturing gradient gel electrophoresis. FEMS Yeast Res 11:564–574. 815 8. Daniel H, Vrancken G, Takrama JF, Camu N, de Vos P, de Vuyst L. 2009. Yeast 816 diversity of Ghanaian cocoa bean heap fermentations. FEMS Yeast Res. 9:774– 817 783. 818 9. Papalexandratou Z, Lefeber T, Bahrim B, Lee OS, Daniel H, de Vuyst L. 2013. 819 Hanseniaspora opuntiae, Saccharomyces cerevisiae, Lactobacillus fermentum, 820 and Acetobacter pasteurianus predominate during well-performed Malaysian cocoa 821 bean box fermentations, underlining the importance of these microbial species for 822 a successful cocoa bean fermentation process. Food Microbiol. 35:73–85. 823 10. Jespersen L, Nielsen D, Honholt S, Jakobsen M. 2005. Occurrence and 824 diversity of yeasts involved in fermentation of West African cocoa beans. FEMS 825 Yeast Res. 5:441–453. 826 11. Lefeber T, Papalexandratou Z, Gobert W, Camu N, de Vuyst L. 2012. On- 827 farm implementation of a starter culture for improved cocoa bean fermentation and 828 its influence on the flavour of chocolates produced thereof. Food Microbiology 829 30:379–392. 830 12. Lefeber T, Janssens M, Camu N, de Vuyst L. 2010. Kinetic analysis of 831 strains of lactic acid bacteria and acetic acid bacteria in cocoa pulp simulation 832 media toward development of a starter culture for cocoa bean fermentation. Appl. 833 Environ. Microbiol. 76:7708–7716. 834 13. Camu N, de Winter T, Verbrugghe K, Cleenwerck I, Vandamme P, 835 Takrama JS, Vancanneyt M, de Vuyst L. 2007. Dynamics and biodiversity of 836 populations of lactic acid bacteria and acetic acid bacteria involved in spontaneous 837 heap fermentation of cocoa beans in Ghana. Appl. Environ. Microbiol. 73:1809– 838 1824. 839 14. Holm CS, Aston JW, Douglas K. 1993. The effects of the organic acids in 840 cocoa on the flavour of chocolate. J. Sci. Food Agric. 61:65–71. 841 15. Papalexandratou Z, Falony G, Romanens E, Jimenez JC, Amores F, 842 Daniel H, de Vuyst L. 2011. Species diversity, community dynamics, and 843 metabolite kinetics of the microbiota associated with traditional Ecuadorian 844 spontaneous cocoa bean fermentations. Appl. Environ. Microbiol. 77:7698–7714. 845 16. Papalexandratou Z, Vrancken G, de Bruyne K, Vandamme P, de Vuyst L. 846 2011. Spontaneous organic cocoa bean box fermentations in Brazil are 37

847 characterized by a restricted species diversity of lactic acid bacteria and acetic acid 848 bacteria. Food Microbiol. 28:1326–1338. 849 17. Camu N, Gonzalez A, de Winter T, van Schoor A, de Bruyne K, Vandamme 850 P, Takrama JS, Addo SK, de Vuyst L. 2008. Influence of turning and 851 environmental contamination on the dynamics of populations of lactic acid and 852 acetic acid bacteria involved in spontaneous cocoa bean heap fermentation in 853 Ghana. Appl. Environ. Microbiol. 74:86–98. 854 18. Lefeber T, Gobert W, Vrancken G, Camu N, de Vuyst L. 2011. Dynamics 855 and species diversity of communities of lactic acid bacteria and acetic acid bacteria 856 during spontaneous cocoa bean fermentation in vessels. Food Microbiol. 28:457– 857 464. 858 19. Garcia-Armisen T, Papalexandratou Z, Hendryckx H, Camu N, Vrancken 859 G, Vuyst L, Cornelis P. 2010. Diversity of the total bacterial community associated 860 with Ghanaian and Brazilian cocoa bean fermentation samples as revealed by a 16 861 S rRNA gene clone library. Appl. Microbiol. Biotechnol. 87:2281–2292. 862 20. Lefeber T, Janssens M, Moens F, Gobert W, de Vuyst L. 2011. Interesting 863 starter culture Strains for controlled cocoa bean fermentation revealed by 864 simulated cocoa pulp fermentations of cocoa-specific lactic acid bacteria. Appl. 865 Environ. Microbiol. 77:6694–6698. 866 21. Illeghems K, de Vuyst L, Weckx S. 2013. Complete genome sequence and 867 comparative analysis of Acetobacter pasteurianus 386B, a strain well-adapted to 868 the cocoa bean fermentation ecosystem. BMC Genomics 14:526. 869 22. Sakurai K, Arai H, Ishii M, Igarashi Y. 2011. Transcriptome response to 870 different carbon sources in Acetobacter aceti. Microbiology 157:899–910. 871 23. Moens F, Lefeber T, de Vuyst L. 2014. Oxidation of metabolites highlights the 872 microbial interactions and role of Acetobacter pasteurianus during Cocoa Bean 873 Fermentation. Appl. Microbiol. Biotechnol. 80:1848–1857. 874 24. de Ley J, Schell J. 1962. Lactate and pyruvate catabolism in acetic acid 875 bacteria. J Gen. Microbiol. 29:589–601. 876 25. de Ley J. 1959. On the formation of acetoin by Acetobacter. J Gen. Microbiol. 877 21:352–365. 878 26. Chandra Raj K, Ingram L, Maupin-Furlow J. 2001. Pyruvate decarboxylase: 879 a key enzyme for the oxidative metabolism of lactic acid by Acetobacter 880 pasteurianus. Arch. Microbiol. 176:443–451. 881 27. Sauer U, Eikmanns BJ. 2005. The PEP–pyruvate–oxaloacetate node as the 882 switch point for carbon flux distribution in bacteria. FEMS Microbiol. Rev. 29:765– 883 794. 884 28. Kohlstedt M, Becker J, Wittmann C. 2010. Metabolic fluxes and beyond— 885 systems biology understanding and engineering of microbial metabolism. Appl. 886 Microbiol. Biotechnol. 88:1065-1075. 887 29. Sauer U, Lasko DR, Fiaux J, Hochuli M, Glaser R, Szyperski T, Wüthrich 888 K, Bailey JE. 1999. Metabolic flux ratio analysis of genetic and environmental 889 modulations of Escherichia coli central carbon metabolism. J. Bacteriol. 181:6679– 890 6688. 891 30. Adler P, Bolten CJ, Dohnt K, Hansen CE, Wittmann C. 2013. Core fluxome 892 and meta fluxome of lactic acid bacteria under cocoa pulp fermentation simulating 893 conditions. Appl. Microbiol. Biotechnol. 79:5670–5681. 894 31. Cleenwerck I, Camu N, Engelbeen K, de Winter T, Vandemeulebroecke K, 895 de Vos P, de Vuyst L. 2007. Acetobacter ghanensis sp. nov., a novel acetic acid 896 bacterium isolated from traditional heap fermentations of Ghanaian cocoa beans. 897 Intern. J. Sys. Evol. Microbiol. 57:1647–1652. 38

898 32. Atlas RM. 2006. Handbook of microbiological media for the examination of 899 food, 2nd ed. CRC Press, Boca Raton, FL. 900 33. Bertani G. 1951. Studies on lysogenesis I: The mode of phage liberation by 901 lysogenic Escherichia coli. J. Bacteriol. 62:293–300. 902 34. Wittmann C, Hans M, Heinzle E. 2002. In vivo analysis of intracellular amino 903 acid labelings by GC/MS. Anal. Biochem. 307:379–382. 904 35. Spaulding R, Charles M. 2002. Comparison of methods for extraction, 905 storage, and silylation of pentafluorobenzyl derivatives of carbonyl compounds and 906 multi-functional carbonyl compounds. Anal. Bioanal. Chem. 372:808-816. 907 36. Bradford MM. 1976. A rapid and sensitive method for the quantitation of 908 microgram quantities of protein utilizing the principle of protein-dye binding. Anal. 909 Biochem. 72:248–254. 910 37. Jetten MSM, Sinskey AJ. 1993. Characterization of phosphoenolpyruvate 911 carboxykinase from Corynebacterium glutamicum. FEMS Microbiol. Lett. 111:183– 912 188. 913 38. Gourdon P, Baucher M, Lindley ND, Guyonvarch A. 2000. Cloning of the 914 malic enzyme gene from Corynebacterium glutamicum and role of the enzyme in 915 lactate metabolism. Appl. Microbiol. Biotechnol. 66:2981–2987. 916 39. Becker J, Klopprogge C, Schröder H, Wittmann C. 2009. Metabolic 917 engineering of the tricarboxylic acid cycle for improved lysine production by 918 Corynebacterium glutamicum. Appl. Microbiol. Biotechnol. 75:7866–7869. 919 40. Azuma Y, Hosoyama A, Matsutani M, Furuya N, Horikawa H, Harada T, 920 Hirakawa H, Kuhara S, Matsushita K, Fujita N, Shirai M. 2009. Whole-genome 921 analyses reveal genetic instability of Acetobacter pasteurianus. Nucleic Acids Res. 922 37:5768–5783. 923 41. Matsushita K, Toyama H, Adachi O. 1994. Respiratory chains and 924 bioenergetics of acetic acid bacteria, p 248. In Rose A, Tempest D (ed), Advances 925 in microbial physiology. Academic Press, San Diego, CA. 926 42. Kondo K, Beppu T, Horinouchi S. 1995. Cloning, sequencing, and 927 characterization of the gene encoding the smallest subunit of the three-component 928 membrane-bound alcohol dehydrogenase from Acetobacter pasteurianus. J. 929 Bacteriol. 177:5048–5055. 930 43. Kanchanarach W, Theeragool G, Yakushi T, Toyama H, Adachi O, 931 Matsushita K. 2010. Characterization of thermotolerant Acetobacter pasteurianus 932 strains and their quinoprotein alcohol dehydrogenases. Appl. Microbiol. Biotechnol. 933 85:741-751. 934 44. Matsutani M, Hirakawa H, Yakushi T, Matsushita K. 2011. Genome-wide 935 phylogenetic analysis of Gluconobacter, Acetobacter, and Gluconacetobacter. 936 FEMS Microbiol. Lett. 315:122–128. 937 45. Quek L, Wittmann C, Nielsen LK, Krömer JO. 2009. OpenFLUX: efficient 938 modelling software for 13C-based metabolic flux analysis. Microb. Cell Fact. 8:25. 939 46. van Winden WA, Wittmann C, Heinzle E, Heijnen JJ. 2002. Correcting mass 940 isotopomer distributions for naturally occurring isotopes. Biotechnol. Bioeng. 941 80:477–479. 942 47. Rosman KJR, Taylor PDP. 1998. Isotopic compositions of the elements 1997. 943 Pure Appl. Chem. 70:217–235. 944 48. Wittmann C, Heinzle E. 2002. Genealogy profiling through strain improvement 945 by using metabolic network analysis: metabolic flux genealogy of several 946 generations of lysine-producing Corynebacteria. Appl. Environ. Microbiol. 68:5843– 947 5859.

948 49. Wittmann C, Heinzle E. 2005. Metabolic activity profiling by 13C tracer 949 experiments and mass spectrometry in Corynebacterium glutamicum. In Barredo 950 JL (ed), Methods in Biotechnology. Humana Press, Totowa, NJ. 951 50. Wittmann C, Krömer JO, Kiefer P, Binz T, Heinzle E. 2004. Impact of the 952 cold shock phenomenon on quantification of intracellular metabolites in bacteria. 953 Anal. Biochem. 327:135–139. 954 51. Chinnawirotpisan P, Theeragool G, Limtong S, Toyama H, Adachi OO, 955 Matsushita K. 2003. Quinoprotein alcohol dehydrogenase is involved in catabolic 956 acetate production, while NAD-dependent alcohol dehydrogenase in ethanol 957 assimilation in Acetobacter pasteurianus SKU1108. J Biosci Bioeng 96:564–571. 958 52. Saeki A, Matsushita K, Takeno S, Taniguchi M, Toyama H, Theeragool G, 959 Lotong N, Adachi O. 1999. Enzymes responsible for acetate oxidation by acetic 960 acid bacteria. Biosci. Biotechnol. Biochem. 63:2102–2109. 961 53. Luttik M, van Spanning R, Schipper D, van Dijken JP, Pronk JT. 1997. The 962 low biomass yields of the acetic acid bacterium Acetobacter pasteurianus are due 963 to a low stoichiometry of respiration-coupled proton translocation. Appl. Environ. 964 Microbiol. 63:3345–3351. 965 54. Bowsher CG, Tobin AK. 2001. Compartmentation of metabolism within 966 mitochondria and plastids. Journal of Experimental Botany 52:513–527. 967 55. Aarnikunnas J, von Weymarn N, Rönnholm K, Leisola M, Palva A. 2003. 968 Metabolic engineering of Lactobacillus fermentum for production of mannitol and 969 pure L-lactic acid or pyruvate. Biotechnol. Bioeng. 82:653–663. 970 56. Yakushi T, Matsushita K. 2010. Alcohol dehydrogenase of acetic acid 971 bacteria: structure, mode of action, and applications in biotechnology. Appl. 972 Microbiol. Biotechnol. 86:1257–1265. 973 57. Adachi O, Miyagawa E, Shinagawam E., Matsushita K, Ameyana M. 1978. 974 Purification and properties of particulate alcohol dehydrogenase from Acetobacter 975 aceti. Agric. Biol. Chem. 42:2331–2340. 976 58. Schwitzguebel J, Ettlinger L. 1979. Pyruvate, orthophosphate dikinase from 977 Acetobacter aceti. Arch. Microbiol. 122:103-108. 978 59. Benziman M, Eizen N. 1971. Pyruvate-phosphate dikinase and the control of 979 gluconeogenesis in Acetobacter xylinum. J. Biol. Chem. 246:57–61. 980 60. Schwitzguébel J, Ettlinger L. 1980. Oxaloacetate decarboxylase from 981 Acetobacter aceti. Arch. Microbiol. 124:63-68. 982 61. Schwitzguébel J, Ettlinger L. 1979. Phosphoenolpyruvate carboxylase from 983 Acetobacter aceti. Arch. Microbiol. 122:109-115. 984 62. Petersen S, Graaf AA de, Eggeling L, Möllney M, Wiechert W, Sahm H. 985 2000. In vivo quantification of parallel and bidirectional fluxes in the anaplerosis of 986 Corynebacterium glutamicum. J. Biol. Chem. 275:35932–35941. 987 63. Dauner M, Storni T, Sauer U. 2001. Bacillus subtilis metabolism and 988 energetics in carbon-limited and excess-carbon chemostat culture. J. Bacteriol. 989 183:7308–7317. 990 64. Stephanopoulos G, Aristidou ANJ. 1998. Metabolic engineering. Principles 991 and methodologies, 1st ed. Academic Press, San Diego. 992 65. Csonka LN, Fraenkel DG. 1977. Pathways of NADPH formation in 993 Escherichia coli. J. Biol. Chem. 252:3382–3391. 994 66. Rühl M, Le Coq D, Aymerich S, Sauer U. 2012. 13C-flux analysis reveals 995 NADPH-balancing transhydrogenation cycles in stationary phase of nitrogen- 996 starving Bacillus subtilis. J. Biol. Chem. 287:27959–27970. 997 67. Becker J, Klopprogge C, Zelder O, Heinzle E, Wittmann C. 2005. Amplified 998 expression of fructose 1,6-bisphosphatase in Corynebacterium glutamicum 40

999 increases in vivo flux through the pentose phosphate pathway and lysine 1000 production on different carbon sources. Appl. Microbiol. Biotechnol. 71:8587–8596. 1001 68. Ingraham JL, Maaloe O, Neidhardt FC. 1983. Growth of the bacterial cell. 1002 Sinauer Associates, Sunderland, MA. 1003 69. Sauer U, Canonaco F, Heri S, Perrenoud A, Fischer E. 2004. The soluble 1004 and membrane-bound transhydrogenases UdhA and PntAB have divergent 1005 functions in NADPH metabolism of Escherichia coli. J. Biol. Chem. 279:6613– 1006 6619. 1007 70. Hommel R, Kurth J, Kleber H. 1988. NADP+-dependent aldehyde 1008 dehydrogenase from Acetobacter rancens CCM 1774: purification and properties. 1009 J. Basic Microbiol. 28:25–33. 1010 71. Russell JB, Cook GM. 1995. Energetics of bacterial growth: balance of 1011 anabolic and catabolic reactions. Microbiol. Reviews 59:48–62. 1012 72. Ho VTT, Zhao J, Fleet G. 2014. Yeasts are essential for cocoa bean 1013 fermentation. Intern. J. Food Microbiol. 174:72–87. 1014 73. Chandra Raj K, Ingram L, Maupin-Furlow J. 2001. Pyruvate decarboxylase: 1015 a key enzyme for the oxidative metabolism of lactic acid by Acetobacter 1016 pasteurianus. Arch. Microbiol. 176:443–451. 1017