This is the accepted version of the journal article: Tomàs Gamisans, Màrius; Ødum, Anders Sebastian Rosenkrans; Workman, Mhairi; [et al.]. «Glycerol metabolism of Pichia pastoris (Komagataella spp.) characterised by 13 C-based metabolic flux analysis». New Biotechnology, Vol. 50 (May 2019), p. 52-59. DOI 10.1016/j.nbt.2019.01.005 This version is available at https://ddd.uab.cat/record/281356 under the terms of the license 1 Glycerol metabolism of Pichia pastoris 2 (Komagataella spp.) characterised by 13 3 C-based metabolic flux analysis 4 5 Màrius Tomàs-Gamisans1, Anders Sebastian Rosenkrans Ødum2, Mhairi Workman4, 6 Pau Ferrer1,3, Joan Albiol1* 7 1Department of Chemical, Biological and Environmental Engineering, Universitat 8 Autònoma de Barcelona, Bellaterra (Cerdanyola del Vallès), Catalonia, Spain 9 2 Department of Biotechnology and Biomedicine, Technical University of Denmark, 10 Lyngby, Denmark 11 3 Current affiliation: Luxembourg Institute of Science and Technology, Luxembourg 12 4 Novo Nordisk, Denmark. 13 *Corresponding author 14 15 Running title: 13C-MFA of P. pastoris growing on glycerol 1 16 Abstract 17 Metabolic flux analysis based on 13C-derived constraints has proved to be a powerful 18 tool for quantitative physiological characterisation of one of the most extensively used 19 microbial cell factory platforms, Pichia pastoris (syn. Komagataella spp.). Nonetheless, 20 the reduced number of carbon atoms and the symmetry of the glycerol molecule has 21 hampered the comprehensive determination of metabolic fluxes when used as 22 labelled C-source. Moreover, metabolic models typically used for 13C-based flux 23 balance analysis may be incomplete or misrepresent the actual metabolic network. To 24 circumvent these limitations, we reduced the genome-scale metabolic model 25 iMT1026-v3.0 into a core model and used it for the iterative fitting of metabolic fluxes 26 to the measured mass isotope distribution of proteinogenic amino acids obtained after 27 fractional 13C labelling of cells with [1,3-13C]-glycerol. This workflow allows obtaining 28 reliable estimates of in vivo fluxes in P. pastoris cells growing on glycerol as sole carbon 29 source, as well as revising previous assumptions concerning its metabolic operation 30 such as alternative metabolic branches, calculation of energetic parameters and 31 proposed specific cofactor utilisation. 32 33 Keywords: 13C-based metabolic flux analysis, Pichia pastoris, Komagataella spp, 34 glycerol, genome-scale metabolic model 35 Introduction 36 Glycerol is a side stream in conventional biodiesel production processes and, 37 therefore, its valorisation is a highly interesting option in the development of a 38 glycerol-based integrated biorefinery concept [1,2]. Glycerol is an attractive feedstock 39 to produce high value added compounds using microbial fermentation processes [3– 2 40 5]. Furthermore, the reduction degree of glycerol (4.67) is higher from that of glucose 41 (4.0), and therefore higher yields of certain metabolites can be obtained from this 42 compound [6]. However, crude glycerol typically contains several impurities such as 43 methanol [7], which is toxic for most microbes with the exception of methylotrophic 44 microorganisms. Notably, P. pastoris is able to efficiently use glycerol and/or methanol 45 as energy and carbon sources [8–10]. In fact, the conventional promoters used for 46 heterologous gene expression in this yeast (namely, PGAP, constitutive, and PAOX, 47 inducible) have been isolated from genes related to glycerol and methanol metabolism 48 [11,12]. Indeed, P. pastoris has been proven to grow on media containing crude 49 glycerol [13,14], and does not need to be genetically engineered for improved glycerol 50 utilisation, as done in other species like S. cerevisiae, [15–17], due to a more efficient 51 glycerol transport system [16].Taken together, these attributes make P. pastoris a cell 52 factory of high potential for the development of glycerol biorefineries. 53 However, there are very few systematic studies characterising growth of P. pastoris on 54 glycerol as a sole carbon source [18,19]. Early 13C-labelling experiments (CLE) 55 performed with P. pastoris using glucose or glycerol as sole carbon sources were based 56 on biosynthetically directed fractional (BDF) 13C-labeling of proteinogenic amino acids 57 with 2D-NMR, enabling the determination of metabolic flux ratios (METAFoR) [20]. 58 This methodology relies on the identification of conserved C-C bounds in proteinogenic 59 amino acids after feeding cells with a mixture of unlabelled and 13C-uniformly labelled 60 glucose as substrate [21,22]. However, the information derived using this technique 61 when using labelled substrates with a low number of carbons such as glycerol, limits its 62 application [18,20]. In addition, 13C-based metabolic flux studies reported so far 63 typically rely on metabolic models based on the pre-existing knowledge on 3 64 biochemical pathways of the central carbon metabolism, including the amino acid 65 biosynthesis pathways [18]. Such models have been extensively used, with just minor 66 modifications based on direct experimental observations, such as that glyoxylate 67 pathway or malic enzyme are not operative under the studied/similar growth 68 conditions [8], to improve metabolic flux determination using the available 69 experimental 13C datasets [9,23–25]. 70 Nevertheless, building such models by just combining well-known classical pathways 71 may easily result in a derived model with missing steps or pathways relevant for the 72 experimental conditions under study. As the 13C based method for metabolic flux 73 determination does not include the cofactor balances (typically NAD(H)/NADP(H)), 74 verification of those balances after the metabolic fluxes have been determined may 75 reveal those inconsistencies in the underlying model. One way to overcome such 76 inconvenience could be the use of a genome-scale metabolic model (GSMM). 77 A genome-scale description of the P. pastoris metabolism has been developed over 78 different versions such as the iMT1026 GSMM [26], which was subsequently adapted 79 for growth on glycerol and methanol [19]. A priori, this GSMM would be an alternative 80 of choice for 13C-MFA, due to its inclusion of a more complete number of pathways. 81 Nevertheless, large-scale 13C-MFA has significant limitations such as the requirement 82 for an accurate and complete atom transition mapping. Although there are databases 83 including the reaction atom mapping of biological pathways, GSMM-specific reactions 84 would require the additional effort to accurately annotate atom transitions. Another 85 important hindrance for genome-scale 13C-MFA is the resulting huge number of 86 variables that significantly impact computational complexity and performance, as well 87 as the difficulty or even impossibility in resolving all fluxes due to the robustness and 4 88 redundancy of metabolic pathways, which include parallel and alternative pathways 89 and cell compartmentalization [27]. Due to this, nowadays GSMMs are still not a 90 practical alternative to core metabolic models. Indeed, Gopalakrishnan and Maranas 91 [28] have concluded that reducing GSMMs down to a size similar to the currently used 92 core metabolism models used in 13C-MFA would be a feasible alternative to the use of 93 full GSMMs. In this regard, several algorithms for reducing GSMMs to core models 94 have been developed. One method that appears particularly suited to our purpose is 95 NetworkReducer [29], which allows to take into account the information derived from 96 previous 13C studies, such as the network topology -pathways proven to be active- in 97 the tested experimental conditions. Briefly, this is achieved by protecting relevant 98 reactions and applying phenotypic constrains while the algorithm successively 99 eliminates or combines reactions until a minimal model that fulfils all the phenotypic 100 constrains is obtained. This method has been recently employed in E. coli to produce a 101 new core model (EColiCore2) producing flux distributions equivalent to those 102 generated by the original GSMM [30]. 103 In this study, the genome-scale metabolic model iMT1026 v3.0 [19] is reduced to a 104 glycerol-specific core model. This reduced model is further used for 13C-MFA of P. 105 pastoris growing on glycerol as carbon source at different growth rates. In order to 106 circumvent the limitation of using a 3C-substrate, a 1- and 3-positionally labelled 107 glycerol is used, instead of a uniformly labelled substrate, together with the 108 measurement of proteinogenic amino acids content and subsequent iterative fitting of 109 metabolic fluxes to the measured mass isotope distribution (MIDs). Although this 110 method does not allow the level of resolution achieved in non-stationary CLE, we show 111 that such approach allowed us to improve the accuracy of the resolved fluxes in 5 112 comparison previous metabolic flux profiling studies based on METAFoR analysis 113 datasets. 114 115 Materials and Methods 116 Strain and cultivation conditions 117 The Pichia pastoris X-33 strain (Invitrogen-Thermofisher, Carlsbad, CA, USA) was used 118 throughout this study. Duplicate carbon-limited chemostat cultivations were 119 performed using a Sartorius 0.5-L bioreactor (Sartorius AG, Göttingen, Germany) at 120 dilution rates (D) of 0.05, 0.10 and 0.16 h-1 with a working volume of 0.3 L maintained 121 by a gravimetrically controlled peristaltic pump. Chemostat cultivations were 122 performed for at least 5 residence times (τ) prior to labelling. 123 Batch and chemostat media were adapted from Baumann et al. [31] by reducing 124 carbon and nitrogen sources concentrations for yielding an approximate final biomass 125 6 g/L when the steady state is reached. Thus, briefly, batch medium contained: 9.98 126 g/L glycerol, 0.46 g/L citric acid, 3.15 g/L (NH4)2HPO4, 0.006 g/L CaCl2·2H2O, 0.225 g/L 127 KCl, 0.125 g/L MgSO4·7H2O, 0.5 mL Biotin (0.2 g/L), 1.15 mL PTM1 trace salts stock 128 solution (prepared as described in [31]). pH was adjusted to 5.0 with 25% HCl. 129 Chemostat medium contained: 10 g/L glycerol, 0.818 g/L citric acid, 4.35 g/L 130 (NH4)2HPO4, 0.01 g/L CaCl2·2H2O, 1.7 g/L KCl, 0.65 g/L MgSO4·7H2O, 1.0 mL Biotin (0.2 131 g/L), 1.6 mL PTM1 trace salts stock solution and 0.2 mL/L of antifoam glanapon 2000 132 (Konc, Bussetti, Vienna, Austria). 133 An inoculum was cultivated overnight at 30C, 150 rpm in a 0.5-L shake flask 134 containing 75 mL of basal medium with glycerol and supplemented with biotin (1% 135 yeast nitrogen base, 4 · 10-5 % biotin, 1% glycerol). Bioreactor was inoculated an initial 6 136 OD600 of 0.3 – 0.5. Once glycerol was exhausted, continuous cultivations were started 137 at the corresponding flow rate. The aeration rate was 1 vvm and the off-gas O2 and 138 CO2 concentrations were measured using a Prima Pro Process Mass Spectrometer 139 (Thermo Fischer Scientific). Temperature was maintained at 25C. Stirring rate was 500 140 rpm and a pH 5.0 was controlled by automatic addition of 15% ammonia. 141 Labelling experiment and biomass harvest 142 After a minimum of 5τ of continuous cultivations with non-labelled glycerol, the feed 143 was switched to the labelled medium. Labelled feed medium composition was the 144 same as the unlabeled feed medium composition, replacing glycerol for 20% [1,3-13C]- 145 glycerol (CortecNet) and 80% unlabeled glycerol. Labelled medium was feed for at 146 least 2τ. Culture samples (50 – 100 mL) were collected, centrifuged (15 min, 16000g), 147 the supernatant discarded and pellets frozen with liquid N 2 and stored at -80C for 148 further extraction and analysis of proteinogenic amino acids. 149 Biomass and exometabolite analysis 150 Cell density and dry cell weight 151 Cell density was monitored by optical density at 600 nm. Dry cell weight (DCW) was 152 measured in duplicate by gravimetric methods. Briefly, a known volume of sample (5 153 to 10 mL) was filtered throughout a dried pre-weighted 0.45 μm polyether sulfone 154 filters (Frisenette, Knebel, Denmark) and washed with distilled water. Filters were 155 dried in a microwave oven at 150 W for 20 min and cooled down in a desiccator for at 156 least 2 h and finally weighted. 157 Exometabolite analysis 7 158 Samples taken for external metabolite analysis, were filtered through a 0.22 μm 159 syringe filters and stored at -20C until subsequent analysis. Glycerol, was the only 160 peak detected in HPLC analyses performed as described in [32]. 161 Proteinogenic amino acid MID determination 162 Amino acid extraction, derivatization and GC-MS analysis 163 Isotope distribution of the proteinogenic amino acid (MID) was determined as 164 described by Knudsen [33]. Briefly, 5 mg of biomass pellets were hydrolyzed with 6 M 165 hydrochloric acid at 105C for 16 h. Once at room temperature, samples were dried for 166 3h under a stream of nitrogen. Samples were redissolved in water and filtered through 167 Strata SCX (100 mg, 1 cc, Phenomenex, Torrance, CA, USA) columns and washed with 168 50% ethanol to remove all the impurities. Samples were eluted with 1 N NaOH and 169 additionally with the elution solution [33]. Two types of derivatives were prepared for 170 GC-MS analysis: N-ethoxycarbonyl-amino ethyl-esters (ECF) and N-dimethyl-amino- 171 methylene-methyl-esters (DMFDMA) following the accurate protocol described by 172 Knudsen [33]. Derivatized amino acid samples were analyzed in a GC-MS Agilent 6890 173 gas chromatograph coupled to an Agilent 5973 quadruple MS accordingly to the 174 specified settings [33]. GC/MS Translator (Agilent) was used to convert the resulting 175 row data files into readable for Agilent MassHunter Qualitative Analysis software. 176 MID correction for natural isotopes and washout kinetics 177 MID are uncorrected for naturally labelled atoms other than carbon backbone [34,35]. 178 OpenFlux [36] was used for correcting the MID of each amino acid according to the 179 expected fragmentation ions obtained in GC-MS analysis [37]. In addition, biomass was 180 harvested at 2 τ after the onset of labelling and thus, the fraction of labelled biomass 8 181 (Xlabelled) at the steady state was calculated according to a first-order wash-out kinetics 182 [18]: Xlabelled = 1 – e-t/, where t is the labelling time and  the residence time of the 183 chemostat. Corrected MIDs for each experimental replicate and corresponding 184 analysed peaks from each corresponding derivatisation method can be found in 185 additional file S2, Table S2. 186 Statistical analysis 187 Chemostat cultivation data was checked for consistency using elemental mass 188 balances and common reconciliation procedures [38]. The biomass molecular formula 189 used was selected according to the specific biomass composition of P. pastoris growing 190 on glycerol [19]. In all the cultivation sets, statistical consistency test was passed with a 191 confidence level of 95%. Consequently, there was no evidence of gross measurement 192 errors. 193 Core model generation 194 In order to obtain a representative network of the central carbon metabolism of 195 P. pastoris, including all the relevant reactions a core model (PpaCore) was derived 196 from iMT1026 v3.0 model using NetworkReducer [29] with CellNetAnalyser 2016.1 197 [39] under Matlab 2011. A detailed procedure and commands for model reduction can 198 be found in additional File S1. Default flux constrains for glycerol growth as the only 199 possible carbon source were set. Maximal growth rate in the reduced network is 200 constrained to be 99.9% of the maximal growth rate for the iMT1026 v3.0 (here 201 PpaGS, meaning ‘P. pastoris Genome-Scale’) corresponding to the glycerol chemostat 202 cultivations in our previous work [19]. In a first step, 46 reactions of the central carbon 203 metabolism were protected (additional File S1) and PpaGS was reduced to a pruned 9 204 model (PpaPruned). Subsequently, PpaPruned was further reduced to PpaCore by a 205 compressing procedure with a new set of 56 protected reactions (additional File S1). 206 Due to numerical reasons the ‘cof’ metabolite was removed from biomass equation. 207 The resulting PpaCore was tested for growth in glycerol and the same maximal growth 208 rates obtained in PpaGS were achieved for a constrained uptake of glycerol. 13 209 C-Metabolic flux analysis (13C-MFA) 210 Flux calculations were performed with OpenFLUX [36] under Matlab 2011 using 211 FMINCON from Matlab’s optimization toolbox. Previously to perform MFA, PpaCore 212 was adapted to OpenFLUX requirements in three steps: (1) each reversible reaction 213 was replaced by two paired irreversible reactions; (2) reactions mapping label 214 distribution in the measured proteinogenic amino acids were added according to the 215 appropriate compartmentalization [20]; (3) carbon atom transition equations were 216 added according to previous P. pastoris models [9] and databases [40]. In order to 217 avoid biased solutions, O2, reduction equivalents, energetic cofactors and additional 218 non-carbon-balanceable metabolites were defined as excluded metabolites. Moreover, 219 those reactions in PpaCore that uniquely contained excluded metabolites were also 220 removed from the final model for 13C-metabolic flux calculations. The resulting model 221 (additional File S2 Table S1) was used for 13C-MFA. Experimental MIDs showed an 222 average deviation below 5% and the model was fitted to the experimental data by the 223 least squares method detailed in [41] using the measured glycerol uptake rate and 224 specific biomass generation rates as constrains. The parameter estimation procedure 225 was repeated 100 times. Subsequently, the solution cluster with lower residual error 226 was used for sensitivity analysis using the non-linear approach of Antoniewicz [42]. 227 Sensitivity analysis was performed in order to find the lower and upper confidence 10 228 interval boundaries of calculated fluxes at a 95% confidence level [36]. Default 229 configuration settings were used for the sensitivity analysis. In those cases, where it 230 was impossible to determine individually the forward and reverse fluxes, only the net 231 fluxes were calculated and subjected to sensitivity analysis. Flux fitting to 13C MID was 232 performed individually for each experimental replicate and subsequently averaged. 233 Main results were depicted using Omix graphic software [43]. 234 Calculation of redox cofactor regeneration rates and energy requirements 235 Rates for redox cofactor regeneration and ATP synthesis were derived after calculation 236 of the other fluxes. Once the solution of the metabolic system was found, estimated 237 fluxes were further used for calculating the remaining reaction fluxes that contained 238 excluded metabolites. Redox cofactor balance was checked and any surplus of 239 reduction equivalents (NADH both cytosolic and mitochondrial) was considered a 240 source of electrons transferred to the electron transport chain (ECT). Therefore, 241 assuming the complete electron transfer from the surplus reduction equivalents to ETC 242 and taking into account the oxygen requirements for biomass synthesis (included in 243 the full biomass equation), the oxygen uptake rate could be calculated and compared 244 to the experimental values. In addition, since the model includes the complete ETC 245 with proton translocation to the mitochondrial intermembrane space and the 246 corresponding reaction for ATP synthesis [26], a theoretical maximal ATP generation 247 rate can be estimated. The total ATP generation was taken later into account for 248 calculating the energetic parameters. Essentially, two parameters were determined: 249 growth and non-growth associated maintenance energy (GAME and NGAME, 250 respectively) according to Pirt’s equation [44] and applying a linear regression between 251 ATP generation and growth rates [45]. The y-intercept would correspond to the 11 252 NGAME ( = 0.0 h-1) and the slope to GAME (Fig. S1). Those values can be later 253 compared to those obtained using the complete PpaGS model. 254 Results and discussion 255 Macroscopic growth parameter characterisation 256 Despite the increasing interest on using glycerol as feedstock for microbial 257 biorefineries, there is limited information on the physiology of P. pastoris growing on 258 such carbon source. In a recent study [19], we performed a series of cultivations at 259 different growth rates using glycerol as sole carbon source. These data allowed to 260 analyse the effects of growth rate on biomass composition in a comprehensive way. 261 In this study, a new series of carbon-limited chemostat cultivations using glycerol as 262 sole carbon source has been performed at three different dilution rates (corresponding 263 to a low, mid and high ranges) that had been already previously tested [19,20], 264 namely: 0.05, 0.10 and 0.16 h-1. In order to obtain equivalent datasets, the culture 265 medium composition was the same as in our recent study, except for glycerol 266 concentration in the feed medium, which was reduced to obtain a lower steady state 267 biomass concentration. 268 In agreement with our previous experiments, P. pastoris cells growing under these 269 conditions showed no by-product secretion and residual glycerol in the broth samples 270 was below the detection limit of the analytical method. All the measured external 271 macroscopic fluxes showed a clear correlation with the growth rate (R2 > 0.99), 272 consistent with our previous study [19], where a linear range for the cell’s macroscopic 273 variables when growing on glycerol was established between 0.05 and 0.16 h-1 growth 274 rate. Biomass yields (YXS) ranged between 0.70 – 0.72 gX · gS-1 (Table 1). These values 275 are in the upper range of YXS previously reported for glycerol as C-source [19] and are 12 276 also in agreement with biomass yields on glycerol reported for other yeast species 277 [46]. Similarly, the average experimental RQ (0.58  0.02) was close to 0.63  0.03, the 278 average RQ experimentally determined in our previous study for glycerol-grown cells. 279 Therefore, despite reducing the carbon source concentration in the feed medium, cells 280 showed comparable macroscopic profiles and therefore reflected comparable 281 cultivations conditions. 282 Reduction of P. pastoris genome-scale model 283 The iMT1026 v3.0 model was used for deriving central carbon metabolism model 284 in two reduction steps, generating PpaPruned and PpaCore. The main characteristics of 285 the reduced model are summarized in Table 2. In order to avoid infeasibilities due to 286 numerical tolerance constrains, ‘cof’ metabolite corresponding to cofactors in biomass 287 equation was removed. ‘COF’ reaction has stoichiometric coefficients of the order of 288 1x10-6 that can be lower than the minimal calculation tolerance. By removing COF, no 289 significant impact on the flux distribution or the predictability capacity was observed. 290 In the first reduction step, the 46 reactions were protected. These reactions included 291 the equivalent reactions to the previous P. pastoris central carbon metabolism model 292 [24] as well as relevant transport reactions across cell compartments. Protected 293 phenotypes were stablished in order to ensure the accuracy of the predictions, i.e. the 294 99.9% of the maximal predicted growth rate in the original model for a given glycerol 295 uptake rate. As reported previously [19], no by-product generation was detected when 296 growing in glycerol-limited chemostats. None of the by-product formation reactions 297 were protected and consequently none of them were present in PpaPruned in 298 agreement with experimental data previously reported [19]. Initially, the genome scale 299 metabolic model contained 2237 reactions and 1881 metabolites (175 external). As a 13 300 result of the network reduction the number of reactions and metabolites was 301 considerably reduced (495 reactions and 513 metabolites). The degrees of freedom 302 (dof) were also strongly reduced (from 485 to 4). This reduction could be due to the 303 fact that the pruned model only considers growth on glycerol with no by-product 304 generation. In the subsequent step, PpaPruned was further reduced by applying a loss- 305 free network compression step [29], where mainly consecutive, parallel and transport 306 reactions were lumped. As a result, PpaCore was generated including 77 reactions with 307 102 metabolites with no further reduction in degrees of freedom. The maximal 308 predicted growth rate was identical as the one obtained with the original model. 13 309 C-Metabolic flux analysis of glycerol-grown cells 310 The generated model was implemented in OpenFLUX code for 13C-MFA determination 311 as described above (Table S1). Flux values are presented in Error! No s'ha trobat 312 l'origen de la referència. as the 95% confidence interval (CI) in order to provide a more 313 informative description of the results taking into account the uncertainty of estimated 314 fluxes [42]. Although flux values expressed as mean of the optimal value ± SEM would 315 provide an overview of changes in the estimated fluxes, this would not indicate the 316 real uncertainty of the measurement due to the asymmetry of the interval. Then, in 317 this study metabolic fluxes are expressed as confidence interval boundaries. 318 To our knowledge, there is only one previous study analysing the metabolic flux 319 distribution of glycerol-grown yeasts based on 13C-labelling experiments [20] using 320 metabolic flux ratio analysis (METAFoR) and thus absolute metabolic flux values were 321 not provided. Moreover, such analysis was limited to the pyruvate node and 322 tricarboxylic acid cycle (TCA) due to the limitations caused by the use of a 3-carbon 323 source such as glycerol. Notably, the flux ratios derived from the metabolic flux 14 324 calculations performed in the present study are consistent with those obtained using 325 METAFoR analysis. In particular, Solà et al. [20] reported a very low activity through 326 glyoxylate cycle. This observation was based on the absence of carbon labelling 327 patterns compatible with the activity of this pathway, also supported by previous 328 studies in S. cerevisiae and P. stipitis [47]. In addition, Solà et al. [20] measured the 329 activity of isocitrate lyase (ICL), showing basal levels in both glucose- and glycerol- 330 grown P. pastoris chemostats. Since glucose is known to repress the glyoxylate 331 pathway in S. cerevisiae [48] and measured activities of ICL in P. pastoris were similar 332 for glucose and glycerol, it was concluded that the flux through glyoxylate cycle was 333 negligible. This consideration has also been subsequently assumed in later studies 334 [24,49–52] by omitting the glyoxylate pathway from the metabolic network for 13C- 335 MFA. As shown in Error! No s'ha trobat l'origen de la referència., the activity through 336 glyoxylate cycle found in the present study is very low, almost negligible, and thus 337 confirms the previous assumptions by Solà et al. [20]. A second observation by Solà 338 and co-workers was that the fraction of mitochondrial pyruvate derived from malate 339 was also very low or negligible, thus indicating that the malic enzyme is likewise almost 340 inactive in cells growing on glycerol-limited chemostats. Our calculations are in 341 agreement with this observation. However, for the lowest dilution rate tested (0.05 h - 342 1), the relative flux through the malic enzyme reaction appears to be higher than at the 343 other dilution rates. Nevertheless, this flux is less than 20% of the carbon flux 344 contribution to the mitochondrial pyruvate pool. 345 The flux split ratio between gluconeogenesis and pentose phosphate pathway cannot 346 be assessed when using biosynthetically directed fractional 13C labelling of 347 proteinogenic amino acids based on uniformly labelled glycerol or glycerol/methanol 15 348 mixtures [8,20]. Nevertheless, when applying the global fitting approach, fluxes 349 through the oxidative branch of the pentose phosphate pathway (PPP) have been 350 estimated for cells growing under such conditions [9]. Coherently with these previous 351 studies [9], our results indicate that the flux through the oxidative branch of PPP is 352 almost negligible (Error! No s'ha trobat l'origen de la referència.). Thus, the majority 353 of NADPH generated in cytosol would be produced in other reactions, such as the 354 glycerol oxidation. Flux directionality in non-oxidative branch of PPP could only be 355 determined within the 95% CI for cells growing at D = 0.05 h-1. In the other two 356 conditions tested, estimated CI includes both reaction directions as feasible Error! No 357 s'ha trobat l'origen de la referència.. This uncertainty on estimated PPP fluxes has 358 been previously described and attributed to the operation of PPP reactions close to 359 the thermodynamic equilibrium, i.e. bidirectionally feasible [53]. 360 The resulting core model included several mitochondrial transporters that also act as 361 redox shuttles [54]. Nevertheless, the malate/aspartate shuttle appears to be the 362 major redox shuttle and TCA cycle intermediate metabolite transporter Error! No s'ha 363 trobat l'origen de la referència., while the flux through other transporters was 364 estimated to be very low, almost negligible. High exchange rates of 365 mitochondrial/cytosolic oxaloacetate were observed in previous studies on glycerol 366 [20]. Authors suggested the existence of a highly active mitochondrial shuttle. 367 Moreover, in cultivations on mixtures of glycerol and methanol, a high transcription of 368 genes involved in the malate/aspartate shuttle was reported [25]. Therefore, those 369 previous studies support our results that point at the malate/aspartate shuttle as the 370 major redox shuttle and TCA cycle intermediate metabolites transporter. 16 371 Impact of dilution rate on metabolic flux distribution 372 Although the flux through the oxidative branch of PPP is very low, and could therefore 373 be considered as negligible, the upper bound of the CI of the relative flux through the 374 oxidative part of PPP appears to increase with the growth rate while the uncertainty in 375 the measured data is similar. Hence, while cultivations at 0.05 h-1 are estimated to 376 have a relative flux between 0 and 0.2, those at 0.10 and 0.16 h-1 showed a CI upper 377 bound of 6.7 and 13.9 respectively. Thus, it would suggest that at higher growth rates, 378 the average relative flux through the oxidative branch of PPP is in fact higher than at 379 lower growth rates. Consequently, the upper bounds for the non-oxidative part of the 380 PPP are also increased at higher growth rates. As recently detailed for P. pastoris [19], 381 the growth rate hypothesis (GRH) describes a positive correlation between the growth 382 rate and the RNA and protein content [55–57]. Moreover, the analysis performed of 383 the RNA content of cells growing on glycerol at different growth rates also depicts this 384 positive correlation [19]. Considering that major precursors for nucleic acids 385 biosynthesis are generated in the pentose phosphate pathway [45], an additional 386 demand of RNA would require an increase in PPP activity. Thus, the estimated increase 387 in relative flux through the oxidative branch of PPP at high growth rates is consistent 388 with the additional demand for RNA precursors, in agreement with the GRH recently 389 described for P. pastoris [19]. Concomitantly, the split ratio between gluconeogenesis- 390 PPP and lower glycolysis at the glyceraldehyde 3-phosphate node is also altered with 391 the growth rate. As a result of the increase in the relative flux though gluconeogenesis 392 and PPP at higher growth rates, there is a reduction of relative flux through the lower 393 glycolysis part and consequently to the TCA cycle (Fig. 1). Similar results were reported 394 by Jordà et al. [9] in chemostat cultivations using mixtures of glycerol and methanol as 17 395 carbon sources at low and high dilution rates (0.05 and 0.16 h -1). Among the different 396 mixtures of glycerol and methanol tested, a qualitative comparison of our results with 397 the reported condition at lower methanol:glycerol ratio (20:80, w/w) can be 398 performed. In this study, invariant absolute fluxes through the TCA cycle (citrate 399 synthase reaction) were observed when comparing low and high growth rates, while 400 the substrate uptake rate was much higher at the highest growth rate. Therefore, the 401 flux through citrate synthase reaction relative to the substrate uptake rate was 402 significantly reduced at the high growth rate, in agreement with the present results. 403 A correlation with the relative flux through the malate/-ketoglutarate transporter 404 and the growth rate is also observed. The upper bound of the CI increases with the 405 dilution rate, thus the relative flux for D=0.16 h-1 would be the highest possible with 406 respect to the other conditions. These results are also in agreement with those 407 described by Solà et al. [20]. There, authors described that at higher growth rates the 408 cytosolic-mitochondrial exchange flux of oxaloacetate was largely unidirectional from 409 the cytosol to the mitochondria. A flux increase in malate/-ketoglutarate exchange 410 reaction is coherent with the increased unidirectional transport of oxaloacetate into 411 the mitochondria, previously observed under similar growth conditions [20], as malate 412 is subsequently oxidised to oxaloacetate in this organelle. That is, a flux increase in the 413 malate/-ketoglutarate shuttle could in fact reflect an increase of the oxaloacetate 414 transport net flux into the mitochondria. 415 Redox cofactor regeneration and energy metabolism 416 Redox cofactors were excluded from the metabolic flux calculation step using carbon 417 labelling data in order to avoid biased solutions. Consequently, metabolic fluxes are 18 418 derived exclusively from the adjustment of experimental MIDs to the calculated 419 metabolic isotope distribution and measured input/output fluxes with no interference 420 of other additional information. Once the metabolic fluxes were estimated, a 421 calculation of both oxygen and cofactor regeneration was performed in order to check 422 whether the estimated solution implicitly satisfies the electron balance. Initially, the 423 estimated flux distribution solution predicted an excess of cytosolic NADH that could 424 not be oxidised. Cells have redox shuttles that are able to transport NADH indirectly 425 from cytosol to mitochondria [54]. One of these redox shuttles is the malate-aspartate 426 shuttle. In this particular case, despite being present in PpaCore, the exact specific 427 activity of the mitochondrial shuttle cannot be calculated as no carbon rearrangement 428 takes place. Consequently, only the net flux in the shuttle system can be calculated 429 (i.e. difference between the cytosolic and mitochondrial flux). Therefore, scaled fluxes 430 in both compartments would result in identical flux relative distributions, but with 431 additional amounts of NADH being translocated to the mitochondria. Regarding the 432 inability to predict scaled up fluxes in redox shuttles employing the currently used 433 constrains, an additional flux fitting was performed by adding the cytosolic NAD(H) 434 balance to the stoichiometric matrix for flux estimation. As a result, identical flux 435 distributions were obtained for reactions other than the mitochondrial redox shuttle. 436 Moreover, the increase in flux of mitochondrial redox shuttle in the new calculation 437 corresponds exactly to the calculated excess of cytosolic NADH. Correspondingly, the 438 theoretical oxygen consumption rates were calculated assuming that all the NADH 439 excess is consumed in the ETC. Oxygen requirements included in the biomass equation 440 were also taken into account and constrained in the ETC calculations. Thus, for each 441 growth rate, the theoretically calculated oxygen requirements account for over 97% of 19 442 the experimentally determined oxygen consumption rates. Therefore, metabolic fluxes 443 and calculated variables are highly consistent with the experimental data. 444 Using the metabolic flux distributions and a global electron balance, maximal ATP 445 generation rates were also calculated. Thus, growth associated and non-associated 446 maintenance energy was calculated as described in the Materials and Methods 447 section. The y-intersection in the regression of maximal ATP generation rates with the 448 growth rate is 1.22 ± 0.48 mmol ATP · gDCW-1 · h-1 and corresponds to the NGAME (See 449 Fig. S1 for the graphical representation). Despite the difference, this value is 450 comparable with the 2.51 reported in our previous study [19] for cultivations on 451 glycerol. In addition, GAME was estimated to be 88.8 ± 4.1 mmol ATP · gDCW-1, that is 452 higher but comparable to the 70.7 mmol ATP · gDCW-1 calculated in the previous study, 453 given that only 3 different growth rates are available. 454 It is worth mentioning that the oxidative branch of PPP is usually considered the 455 major source of cytosolic NADPH [49,58]. Nevertheless, the predicted flux through this 456 pathway in this case is very low, almost negligible, in accordance with previous 13C- 457 MFA estimations on mixtures of glycerol and methanol [9]. In fact, those studies 458 already suggested that alternative reactions must supply the required cytosolic 459 NADPH, otherwise a NADPH imbalance was observed. Our GSMM contains all the 460 relevant reactions producing NADPH in cytosol. As a result of automatic model 461 reduction, PpaCore contained the NADP+-dependent glycerol oxidation, although 462 previous core models proposed a NAD+-dependent glycerol oxidation. Therefore, it 463 seems plausible that the NADP+-dependent oxidation of glycerol pathway is used in 464 glycerol-grown cells. In this way, it would be the major source for cytosolic NADPH 465 formation. 20 466 Conclusions 467 The use of 13C-positionally labelled glycerol in the steady state CLEs combined with the 468 analysis of the labelling patterns of proteinogenic amino acids allowed, for the first 469 time, a more reliable estimation of metabolic fluxes through the central carbon 470 pathways of P. pastoris cells growing on glycerol as sole C-source within an acceptable 471 confidence range. A new protocol was used to determine the metabolic flux 472 distributions of P. pastoris growing in glycerol at different dilution rates. The results 473 allowed to verify previous hypothesis, calculate energetic parameters and propose 474 alternative cofactor utilization. 475 Furthermore, we have developed and tested a more robust steady state workflow for 476 13C-MFA of yeast growing on glycerol as sole carbon source, enabling efficient support 477 for glycerol-based metabolic and bioprocess engineering applications. 478 479 Acknowledgements 480 This work was supported by the project CTQ2013- 42391-R and CTQ2016-74959-R 481 (AEI/FEDER, UE) of the Spanish Ministry of Economy, Industry and Competitiveness, 482 2014-SGR-452 of the Reference Network in Biotechnology (XRB) (Generalitat de 483 Catalunya) and the grant FPU FPU12/06185 (M.T.) of the Spanish Ministry of 484 Education, Culture and Sport and the Short-Term Fellowship (M.T.) of the European 485 Molecular Biology Organisation. In addition, the authors would like to acknowledge 486 partial support from the ERA- IB IPCRES project. 487 488 489 21 490 References 491 [1] Kiss AA, Grievink J, Rito-Palomares M. A systems engineering perspective on process integration 492 in industrial biotechnology. J Chem Technol Biotechnol 2015;90:349–55. doi:10.1002/jctb.4584. 493 [2] Chen Z, Liu D. Toward glycerol biorefinery: metabolic engineering for the production of biofuels 494 and chemicals from glycerol. Biotechnol Biofuels 2016;9:205. doi:10.1186/s13068-016-0625-8. 495 [3] Valerio O, Horvath T, Pond C, Misra M, Mohanty A. Improved utilization of crude glycerol from 496 biodiesel industries: Synthesis and characterization of sustainable biobased polyesters. Ind Crops 497 Prod 2015;78:141–7. doi:10.1016/j.indcrop.2015.10.019. 498 [4] Johnson DT, Taconi KA. The glycerin glut: Options for the value-added conversion of crude 499 glycerol resulting from biodiesel production. Environ Prog 2007;26:338–48. 500 doi:10.1002/ep.10225. 501 [5] Yang F, Hanna M a, Sun R. Value-added uses for crude glycerol--a byproduct of biodiesel 502 production. Biotechnol Biofuels 2012;5:1–10. doi:10.1186/1754-6834-5-13. 503 [6] da Silva GP, Mack M, Contiero J. Glycerol: A promising and abundant carbon source for industrial 504 microbiology. Biotechnol Adv 2009;27:30–9. doi:10.1016/j.biotechadv.2008.07.006. 505 [7] Posada J a, Rincón LE, Cardona C a. Design and analysis of biorefineries based on raw glycerol: 506 addressing the glycerol problem. Bioresour Technol 2012;111:282–93. 507 doi:10.1016/j.biortech.2012.01.151. 508 [8] Solà A, Jouhten P, Maaheimo H, Sánchez-Ferrando F, Szyperski T, Ferrer P. Metabolic flux 509 profiling of Pichia pastoris grown on glycerol/methanol mixtures in chemostat cultures at low 510 and high dilution rates. Microbiology 2007;153:281–90. doi:10.1099/mic.0.29263-0. 511 [9] Jordà J, De Jesus SS, Peltier S, Ferrer P, Albiol J. Metabolic flux analysis of recombinant Pichia 512 pastoris growing on different glycerol/methanol mixtures by iterative fitting of NMR-derived 513 13C-labelling data from proteinogenic amino acids. N Biotechnol 2014;31:120–32. 514 doi:10.1016/j.nbt.2013.06.007. 515 [10] Çelik E, Ozbay N, Oktar N, Çalık P. Use of Biodiesel Byproduct Crude Glycerol as the Carbon 516 Source for Fermentation Processes by Recombinant Pichia pastoris. Ind Eng Chem Res 517 2008;47:2985–90. doi:10.1021/ie071613o. 518 [11] Cos O, Ramón R, Montesinos JL, Valero F. Operational strategies, monitoring and control of 519 heterologous protein production in the methylotrophic yeast Pichia pastoris under different 520 promoters: a review. Microb Cell Fact 2006;5:17. doi:10.1186/1475-2859-5-17. 521 [12] Gasser B, Prielhofer R, Marx H, Maurer M, Nocon J, Steiger M, et al. Pichia pastoris: protein 522 production host and model organism for biomedical research. Future Microbiol 2013;8:191–208. 523 doi:10.2217/fmb.12.133. 524 [13] Anastácio GS, Santos KO, Suarez PAZ, Torres FAG, De Marco JL, Parachin NS. Utilization of 525 glycerin byproduct derived from soybean oil biodiesel as a carbon source for heterologous 526 protein production in Pichia pastoris. Bioresour Technol 2014;152:505–10. 527 doi:10.1016/j.biortech.2013.11.042. 528 [14] Lopes M, Belo I, Mota M. Batch and fed-batch growth of Pichia pastoris under increased air 529 pressure. Bioprocess Biosyst Eng 2013;36:1267–75. doi:10.1007/s00449-012-0871-5. 530 [15] Ho P-W, Swinnen S, Duitama J, Nevoigt E. The sole introduction of two single-point mutations 531 establishes glycerol utilization in Saccharomyces cerevisiae CEN.PK derivatives. Biotechnol 532 Biofuels 2017;10:10. doi:10.1186/s13068-016-0696-6. 533 [16] Klein M, Swinnen S, Thevelein JM, Nevoigt E. Glycerol metabolism and transport in yeast and 534 fungi: established knowledge and ambiguities. Environ Microbiol 2017;19:878–93. 535 doi:10.1111/1462-2920.13617. 536 [17] Klein M, Carrillo M, Xiberras J, Islam Z, Swinnen S, Nevoigt E. Towards the exploitation of 537 glycerol’s high reducing power in Saccharomyces cerevisiae -based bioprocesses. Metab Eng 538 2016;38:464–72. doi:10.1016/j.ymben.2016.10.008. 539 [18] Ferrer P, Albiol J. 13C-Based Metabolic Flux Analysis in Yeast: The Pichia pastoris Case. In: 540 Mapelli V, editor. Methods Mol. Biol., vol. 1152, New York, NY: Springer New York; 2014, p. 209– 541 32. doi:10.1007/978-1-4939-0563-8_13. 542 [19] Tomàs-Gamisans M, Ferrer P, Albiol J. Fine-tuning the P. pastoris iMT1026 genome-scale 543 metabolic model for improved prediction of growth on methanol or glycerol as sole carbon 544 sources. Microb Biotechnol 2018;11:224–37. doi:10.1111/1751-7915.12871. 22 545 [20] Solà A, Maaheimo H, Ylönen K, Ferrer P, Szyperski T. Amino acid biosynthesis and metabolic flux 546 profiling of Pichia pastoris. Eur J Biochem 2004;271:2462–70. doi:10.1111/j.1432- 547 1033.2004.04176.x. 548 [21] Maaheimo H, Fiaux J, Petek Çakar Z, Bailey JE, Sauer U, Szyperski T. Central carbon metabolism 549 of Saccharomyces cerevisiae explored by biosynthetic fractional 13C labeling of common amino 550 acids. Eur J Biochem 2001;268:2464–79. doi:10.1046/j.1432-1327.2001.02126.x. 551 [22] Szyperski T, Glaser RW, Hochuli M, Fiaux J, Sauer U, Bailey JE, et al. Bioreaction Network 552 Topology and Metabolic Flux Ratio Analysis by Biosynthetic Fractional 13C Labeling and Two- 553 Dimensional NMR Spectroscopy. Metab Eng 1999;1:189–97. doi:10.1006/mben.1999.0116. 554 [23] Jordà J, Jouhten P, Cámara E, Maaheimo H, Albiol J, Ferrer P. Metabolic flux profiling of 555 recombinant protein secreting Pichia pastoris growing on glucose:methanol mixtures. Microb 556 Cell Fact 2012;11:57. doi:10.1186/1475-2859-11-57. 557 [24] Baumann K, Carnicer M, Dragosits M, Graf AB, Stadlmann J, Jouhten P, et al. A multi-level study 558 of recombinant Pichia pastoris in different oxygen conditions. BMC Syst Biol 2010;4:141. 559 doi:10.1186/1752-0509-4-141. 560 [25] Rußmayer H, Buchetics M, Gruber C, Valli M, Grillitsch K, Modarres G, et al. Systems-level 561 organization of yeast methylotrophic lifestyle. BMC Biol 2015;13:80. doi:10.1186/s12915-015- 562 0186-5. 563 [26] Tomàs-Gamisans M, Ferrer P, Albiol J. Integration and Validation of the Genome-Scale Metabolic 564 Models of Pichia pastoris: A Comprehensive Update of Protein Glycosylation Pathways, Lipid and 565 Energy Metabolism. PLoS One 2016;11(1):e0148031. doi:10.1371/journal.pone.0148031. 566 [27] Gopalakrishnan S, Maranas C. Achieving Metabolic Flux Analysis for S. cerevisiae at a Genome- 567 Scale: Challenges, Requirements, and Considerations. Metabolites 2015;5:521–35. 568 doi:10.3390/metabo5030521. 569 [28] Gopalakrishnan S, Maranas CD. 13C metabolic flux analysis at a genome-scale. Metab Eng 570 2015;32:12–22. doi:10.1016/j.ymben.2015.08.006. 571 [29] Erdrich P, Steuer R, Klamt S. An algorithm for the reduction of genome-scale metabolic network 572 models to meaningful core models. BMC Syst Biol 2015;9:48. doi:10.1186/s12918-015-0191-x. 573 [30] Hädicke O, Klamt S. EColiCore2: a reference network model of the central metabolism of 574 Escherichia coli and relationships to its genome-scale parent model. Sci Rep 2017;7:39647. 575 doi:10.1038/srep39647. 576 [31] Baumann K, Maurer M, Dragosits M, Cos O, Ferrer P, Mattanovich D. Hypoxic fed-batch 577 cultivation of Pichia pastoris increases specific and volumetric productivity of recombinant 578 proteins. Biotechnol Bioeng 2008;100:177–83. doi:10.1002/bit.21763. 579 [32] Liu X, Mortensen UH, Workman M. Expression and functional studies of genes involved in 580 transport and metabolism of glycerol in Pachysolen tannophilus. Microb Cell Fact 2013;12:27. 581 doi:10.1186/1475-2859-12-27. 582 [33] Knudsen PB, Workman M, Nielsen KF, Thykaer J. Development of scalable high throughput 583 fermentation approaches for physiological characterisation of yeast and filamentous fungi. Ph.D. 584 Thesis. Kgs. Lyngby. Technical University of Denmark, 2015. 585 [34] Fernandez CA, Rosiers C Des, Previs SF, David F, Brunengraber H. Correction of 13C Mass 586 Isotopomer Distributions for Natural Stable Isotope Abundance. J Mass Spectrom 1996;31:255– 587 62. doi:10.1002/(SICI)1096-9888(199603)31:3<255::AID-JMS290>3.0.CO;2-3. 588 [35] Van Winden WA, Wittmann C, Heinzle E, Heijnen JJ. Correcting mass isotopomer distributions 589 for naturally occurring isotopes. Biotechnol Bioeng 2002;80:477–9. doi:10.1002/bit.10393. 590 [36] Quek L-E, Wittmann C, Nielsen LK, Krömer JO. OpenFLUX: efficient modelling software for 13C- 591 based metabolic flux analysis. Microb Cell Fact 2009;8:25. doi:10.1186/1475-2859-8-25. 592 [37] Christensen B, Nielsen J. Isotopomer Analysis Using GC-MS. Metab Eng 1999;1:282–90. 593 doi:10.1006/mben.1999.0117. 594 [38] Noorman HJ, Romein B, Luyben KC a M, Heijnen JJ. Classification, error detection, and 595 reconciliation of process information in complex biochemical systems. Biotechnol Bioeng 596 2000;49:364–76. doi:10.1002/(SICI)1097-0290(19960220)49:4<364::AID-BIT2>3.0.CO;2-N. 597 [39] Klamt S, Saez-Rodriguez J, Gilles ED. Structural and functional analysis of cellular networks with 598 CellNetAnalyzer. BMC Syst Biol 2007;1:2. doi:10.1186/1752-0509-1-2. 599 [40] Caspi R, Altman T, Billington R, Dreher K, Foerster H, Fulcher C a., et al. The MetaCyc database of 600 metabolic pathways and enzymes and the BioCyc collection of Pathway/Genome Databases. 601 Nucleic Acids Res 2014;42:459–71. doi:10.1093/nar/gkt1103. 23 602 [41] Antoniewicz MR, Kelleher JK, Stephanopoulos G. Elementary metabolite units (EMU): A novel 603 framework for modeling isotopic distributions. Metab Eng 2007;9:68–86. 604 doi:10.1016/j.ymben.2006.09.001. 605 [42] Antoniewicz MR, Kelleher JK, Stephanopoulos G. Determination of confidence intervals of 606 metabolic fluxes estimated from stable isotope measurements. Metab Eng 2006;8:324–37. 607 doi:10.1016/j.ymben.2006.01.004. 608 [43] Droste P, Miebach S, Niedenführ S, Wiechert W, Nöh K. Visualizing multi-omics data in metabolic 609 networks with the software Omix—A case study. Biosystems 2011;105:154–61. 610 doi:10.1016/j.biosystems.2011.04.003. 611 [44] Pirt SJ. Maintenance energy: a general model for energy-limited and energy-sufficient growth. 612 Arch Microbiol 1982;133:300–2. doi:10.1007/BF00521294. 613 [45] Stephanopoulos GN, Aristidou AA, Nielsen J. Review of Cellular Metabolism. Metab. Eng., San 614 Diego: Academic Press; 1998, p. 21–79. doi:10.1016/B978-012666260-3/50003-0. 615 [46] Taccari M, Canonico L, Comitini F, Mannazzu I, Ciani M. Screening of yeasts for growth on crude 616 glycerol and optimization of biomass production. Bioresour Technol 2012;110:488–95. 617 doi:10.1016/j.biortech.2012.01.109. 618 [47] Fiaux J, Cakar ZP, Sonderegger M, Wuthrich K, Szyperski T, Sauer U. Metabolic-Flux Profiling of 619 the Yeasts Saccharomyces cerevisiae and Pichia stipitis. Eukaryot Cell 2003;2:170–80. 620 doi:10.1128/EC.2.1.170-180.2003. 621 [48] Gancedo JM. Yeast carbon catabolite repression. Microbiol Mol Biol Rev 1998;62:334–61. 622 [49] Jordà J, Rojas H, Carnicer M, Wahl A, Ferrer P, Albiol J. Quantitative Metabolomics and 623 Instationary 13C-Metabolic Flux Analysis Reveals Impact of Recombinant Protein Production on 624 Trehalose and Energy Metabolism in Pichia pastoris. Metabolites 2014;4:281–99. 625 doi:10.3390/metabo4020281. 626 [50] Nocon J, Steiger MG, Pfeffer M, Sohn SB, Kim TY, Maurer M, et al. Model based engineering of 627 Pichia pastoris central metabolism enhances recombinant protein production. Metab Eng 628 2014;24:129–38. doi:10.1016/j.ymben.2014.05.011. 629 [51] Heyland J, Fu J, Blank LM, Schmid A. Quantitative physiology of Pichia pastoris during glucose- 630 limited high-cell density fed-batch cultivation for recombinant protein production. Biotechnol 631 Bioeng 2010;107:357–68. doi:10.1002/bit.22836. 632 [52] Celik E, Calik P, Oliver SG. Metabolic flux analysis for recombinant protein production by Pichia 633 pastoris using dual carbon sources: Effects of methanol feeding rate. Biotechnol Bioeng 634 2010;105:317–29. doi:10.1002/bit.22543. 635 [53] Jordà J, Suarez C, Carnicer M, ten Pierick A, Heijnen JJ, van Gulik W, et al. Glucose-methanol co- 636 utilization in Pichia pastoris studied by metabolomics and instationary 13C flux analysis. BMC 637 Syst Biol 2013;7:17. doi:10.1186/1752-0509-7-17. 638 [54] Bakker BM, Overkamp KM, van Maris AJ, Kötter P, Luttik MAH, van Dijken JP, et al. Stoichiometry 639 and compartmentation of NADH metabolism in Saccharomyces cerevisiae. FEMS Microbiol Rev 640 2001;25:15–37. doi:10.1111/j.1574-6976.2001.tb00570.x. 641 [55] Kyle M, Acharya K, Weider LJ, Looper K, Elser JJ. Coupling of growth rate and body stoichiometry 642 in Daphnia: a role for maintenance processes? Freshw Biol 2006;51:2087–95. 643 doi:10.1111/j.1365-2427.2006.01639.x. 644 [56] Elser JJ, Acharya K, Kyle M, Cotner J, Makino W, Markow T, et al. Growth rate-stoichiometry 645 couplings in diverse biota. Ecol Lett 2003;6:936–43. doi:10.1046/j.1461-0248.2003.00518.x. 646 [57] Franklin O, Hall EK, Kaiser C, Battin TJ, Richter A. Optimization of biomass composition explains 647 microbial growth-stoichiometry relationships. Am Nat 2011;177:E29--42. doi:10.1086/657684. 648 [58] Blank LM, Lehmbeck F, Sauer U. Metabolic-flux and network analysis in fourteen 649 hemiascomycetous yeasts. FEMS Yeast Res 2005;5:545–58. doi:10.1016/j.femsyr.2004.09.008. 650 651 Figure Captions 652 Table 1. Reconciled macroscopic growth parameters for glycerol cultivations at 653 different growth rates. DSP (h-1) corresponds to the set point dilution rate and Dexp (h-1) to the 24 654 experimentally measured D. qS, qO2 and qCO2 are expressed in mmol · gDCW-1· h-1. Units for qX are Cmmol · gDCW-1· h- 655 1.Y XS represents biomass yield (gX · gS-1). RQ is the respiratory quotient. 656 Table 2. Main properties of P. pastoris models used and generated in this study. 657 Fig. 1. Metabolic flux distribution estimated for P. pastoris growing on glycerol at 658 different dilution rates. 0.05 h-1 (top box), 0.10 h-1 (middle box) and 0.16 h-1 (bottom box). 659 Results are expressed as the 95% confidence interval of the estimated fluxes relative to the 660 glycerol uptake rate in mmol glycerol · gDCW-1 · h-1. Lower and upper bounds of CI correspond to 661 the maximum and minimum CI bounds between the replicates. Flux directionality assumption 662 was represented by arrows; therefore, negative fluxes describe opposite direction. 663 25 664 Table 1 DSP 0.05 0.10 0.16 Dexp 0.046 ± 0.006 0.098 ± 0.009 0.166 ± 0.006 qS -0.72 ± 0.11 -1.45 ± 0.13 -2.52 ± 0.14 qX 1.62 ± 0.21 3.43 ± 0.31 5.89 ± 0.23 qCO2 0.55 ± 0.12 0.93 ± 0.07 1.65 ± 0.18 qO2 -0.90 ± 0.18 -1.66 ± 0.14 -2.89 ± 0.25 YXS 0.70 ± 0.02 0.73 ± 0.01 0.72 ± 0.01 RQ 0.60 ± 0.02 0.56 ± 0.01 0.57 ± 0.01 665 666 26 667 Table 2 668 iMT1026 v3.0 PpaPruned PpaCore Openflux PpaCore 669 # reactions 2237 495 77 120670 # internal metabolites 1706 513 102 100671 # external metabolites 175 9 9 9 672 Maximal biomass generation rate 0.0940 0.0939 0.0939 - 673 (g·gDCW-1·h-1) 674 27 675 Fig. 1 676 677 678 28 679 Additional File S1. Contains the original (PpaGS), reduced (PpaPruned) and 680 compressed model (PpaCore) in Cell Net Analyzer format. In addition, the instructions 681 followed for generating the reduced model are also included and described. PpaGS 682 model is derived from the original iMT1026 v3.0, but due to numerical tolerance 683 constrains in model reduction, the ‘cof_c’ metabolite was removed from the specific 684 biomass equation describing growth on glycerol. 685 686 Additional File S2. Contains: Table S1: The Reduced stoichiometric model used for 687 13C-MFA. Table S2: Corrected experimental and simulated MIDs for each experimental 688 replicate and corresponding analysed peaks from each corresponding derivatisation 689 method. Table S3: Metabolic flux distribution estimated for P. pastoris growing on 690 glycerol at different dilution rates. Fig. S1: Correlation among the calculated ATP 691 turnover and the corresponding growth rate. 692 693 694 29