www.elsevier.com/locate/jbiosc

Journal of Bioscience and BioengineeringVOL. 116 No. 3, 371e379, 2013

Investigation of the physiological response to oxygen limited process conditionsof Pichia pastoris Mutþ strain using a two-compartment scale-down system

Bettina Lorantfy,1,2 Mohammadhadi Jazini,1 and Christoph Herwig1,*

Vienna University of Technology, Institute of Chemical Engineering, Research Division Biochemical Engineering, Gumpendorfer Strasse 1A 166/4, A-1060 Vienna,Austria1 and Research Center Pharmaceutical Engineering, A-8042 Graz, Austria2

Received 6 March 2013; accepted 27 March 2013Available online 3 May 2013

* CorrespondE-mail add

1389-1723/$http://dx.doi

Inhomogeneities in production-scale bioreactors influence microbial growth and product quality due to insufficientmixing and mass transfer. For this reason, lots of efforts are being made to investigate the effects of gradients thatimpose stress in large-scale reactors in laboratory scale. We have implemented a scale-down model which allowsseparating a homogeneous part, a stirred tank reactor (STR), and a plug flow reactor (PFR) which mimics the inho-mogeneous regimes of the large-scale fermenters. This scale-down model shows solutions to trigger oxygen limitedconditions in the PFR part of the scale-down setup for physiological analysis. The goal of the study was to investigate thescale-up relevant physiological responses of Pichia pastoris strain to oxygen limited process conditions in the abovementioned two-compartment bioreactor setup. Experimental results with non-induced cultures show that the specificgrowth rate significantly decreased with increasing the exposure time to oxygen limitation. In parallel more by-productswere produced. Examining physiological scalable key parameters, multivariate data analyses solely using on-line datarevealed that different exposures to the oxygen limitation significantly affected the culture performance. This work withthe small scale-downs setup reflects new approaches for a valuable process development tool for accelerating straincharacterization or for verifying CFD simulations of large-scale bioreactors. As a novel methodological achievement, thecombination of the two-compartment scale-down systemwith the proposed multivariate techniques of solely using on-line data is a valuable tool for recognition of stress effects on the culture performance for physiological bioprocess scale-up issues.

� 2013, The Society for Biotechnology, Japan. All rights reserved.

[Key words: Scale-down; Two-compartment reactor; Pichia pastoris; Stress response; Oxygen limitation]

The quality of recombinant pharmaceutical products is strictlyregulated by directives and standards of international andgovernmental bodies. Predictable yields for cells and product aswell as consistent product quality call for robust industrial-scalebioprocesses. The desired product quantity and quality are stronglyaffected by the heterogeneities in the concentration profiles whichoccur in production-scale bioreactors due to insufficient mixing (1).Regarding aerobic bioprocesses, the limits of mass transfer capacityand fluid dynamics are the most crucial that affect industrial-scaleperformance, resulting in concentration gradients for oxygen (2,3)and for the limiting substrate (4). For example, in industrial-scalefed-batch cultivations, the concentration of the limiting substratefluctuates due to the imperfection of mixing. This is largelyaffecting cell physiology as the feed solution is usually applied inconcentrated form to minimize dilution effects (5,6).

Therefore, the bioprocess performance differ between small-scale (laboratory scale) and the large-scale. While the hydraulicscale-up is mastered since decades, studying physiological effectsfocussing on scale-up relevant information still remained some-what unexploited (7,8). In order to investigate the impact of stress

ing author. Tel.: þ43 1 58801 166 400; fax: þ43 1 58801 166 980.ress: christoph.herwig@tuwien.ac.at (C. Herwig).

e see front matter � 2013, The Society for Biotechnology, Japan..org/10.1016/j.jbiosc.2013.03.021

effects of large-scale reactors, scale-downmodel systems should bedesigned (9,10). For example, scale-down models can focus onmimicking the desired inhomogeneities, such as pH gradients (11),mixing time (12) or the limiting substrate gradient (13). On onehand, the gradients in large-scale reactors are modelled andinvestigated by carrying out oscillatory experiments in laboratoryscale. For that, a single compartment well-mixed and aeratedfermenter is used which is considered as an ideal small-sizebioreactor. Regarding the two most critical compounds in aerobicbioprocesses, the effects of oscillating dissolved oxygen tension (14)and the oscillatory feed profile of the limiting substrate (15) onphysiology and productivity were reported. On the other hand, theinhomogeneities can be mimicked in laboratory scale with two-compartment systems, where one part represents the homoge-neous part, e.g., a continuous stirred tank reactor (STR) and theother part e another STR or a plug flow reactor (PFR) mimics theinhomogeneities in the large-scale reactors (7,16,17). The latterconfiguration allows perturbation of a part of the reaction volumeand therefore follows a scale-down approach of partly not well-mixed zones in large-scale.

The effects of gradients of limiting substrate, oxygen as well aspH on cell physiology and productivity were recently investigatedin the case of Escherichia coli (5e8,15), Bacillus subtilis (11,18) and

All rights reserved.

372 LORANTFY ET AL. J. BIOSCI. BIOENG.,

Saccharomyces cerevisiae (13,16). To our knowledge, such studieshave not been reported on Pichia pastoris yet. This strain has alsobecome a prosperous host organism due to its beneficial properties,e.g., easy cultivation to high cell densities, several available tools formolecular manipulations, high level of production of recombinantproteins, moreover, production of complex proteins with eukary-otic post-translational modifications (19e21).

In the era of the emerging recombinant techniques, the questionof faster screening and characterizing different recombinant strainsare also getting more highlighted in laboratory scale bioreactors.Regarding the acceleration of bioprocess scale-up to large-scalesfrom physiological point of view, the speed of process developmentalso requires reliable and quick tools and methodologies. Thesetechniques should not be necessarily targeted full physiologicalquantitation but only scale-up relevant physiological studies.

We have implemented a small-size two-compartment setupbased on the relevant two-compartment scale-down setups andaccording to the available modelling and simulation tools and re-sults (10,18). The goal of this study was to investigate the scale-uprelevant physiological responses of a P. pastoris strain to oxygenlimited process conditions with this small two-compartment scale-down setup. Repetitive batches were carried out with two differentsetup configurations and different residence times in the PFR. As anew approach for strain characterization, as demonstrated, therepetitive batches in a scale-down system offer faster alternativesfor acquiring bioprocess scale-up relevant physiological informa-tion. Hysteresis effect can be however hypothesized among therepetitive experiment items, the dependency and experiment cor-relation was checked and discussed. First, the perturbation impacton the maximum specific growth rate, rates and yields wereanalyzed. Subsequently, multivariate data analysis was performedwith the collected data. The dependence of the repetitive batchexperiments were also pre-checked with correlation test and foundto be sufficiently low for multivariate analysis. These mentionedsimple statistical and multivariate approaches can provide a basefor scale-up methodologies from the physiological point of view.

MATERIALS AND METHODS

Strain and media The experiments were carried out with P. pastoris Mutþ

SMD 1168 h (22). Preculture: Yeast nitrogen base medium (YNBM), per litre:potassium phosphate buffer (pH 6.0), 0.1 M; YNB w/o amino acids and ammoniasulphate (Difco TM), 3.4 g; (NH4)2SO4, 10 g; biotin, 400 mg; glucose, 20 g. Batchmedium: Basal salt medium (BSM), per litre: 85% phosphoric acid, 26.7 mL;CaSO4$2H2O, 1.17 g; K2SO4, 18.2 g; MgSO4$7H2O, 14.9 g; KOH, 4.13 g; glycerol asC-source, 10 g, Antifoam Struktol J650, 0.2 mL; PTM1, 4.35 mL; NH4OH asN-source. Trace element solution (PTM1), per litre: CuSO4$5H2O, 6.0 g; NaI 0.08 g;MnSO4$H2O, 3.0 g; Na2MoO4$2H2O, 0.2 g; H3BO3, 0.02 g; CoCl2, 0.5 g; ZnCl2,20.0 g; FeSO4$7H2O, 65.0 g; biotin, 0.2 g; H2SO4, 5 mL (23). As for the elementalcomposition of the biomass, the elemental analysis and the ash determinationprocedures were accomplished with previously washed and then lyophilizedbiomass samples according to the standard procedures (24).

Two-compartment scale-down setup The following small-size two-compartment scale-down systemwas implemented and characterized (Fig. 1A). Therepetitive batch experiments were carried out in a fully instrumented 3.6 Lautoclavable laboratory Labfors bioreactor (Infors, Switzerland) with 1 L batchworking volume which represents the homogeneous part of the setup (STR). Theon-line data monitoring and process control were executed with a ProcessInformation Management System (Lucullus, Biospectra AG, Switzerland). The pHand temperature were controlled by PID controllers and the pH value in thereactor was maintained at the physiological optimum value of 5 � 0.1 with theaddition of NH4OH with a certain known concentration. The reactor volume andthe base consumption were continuously monitored with the utilization oflaboratory scales with 0.1 g resolution (Sartorius, Germany and Mettler Toledo,Switzerland). The cultivation parameters in the bioreactor were 30�C and1200 rpm of agitation speed and 1.5 vvm air inlet flow. In order to triggerdifferent conditions from the homogeneous part of the setup, a plug flow reactor(PFR) with a defined working volume ratio was connected to the bioreactor. Thebroth was continuously circulated in silicone tubings (Roth, Germany and Tygon,Saint-Gobain, France) with the average inner diameter of 0.64 cm, outer diameterof 1 cm and 3.3 m length at a defined residence times between the two parts ofthe scale-down setup with a peristaltic pump (Ismatec, Switzerland). There are no

special materials required and no extra tubings needed for connecting the stirredtank reactor with the plug flow part of the setup, as the PFR itself consists ofsilicone tubing. To observe the conditions of the culture broth in the PFR beforereturning to the STR, the PFR was equipped with a sampling port and a stainlesssteel Flow cell (Hamilton Bonaduz, Switzerland) with the connection of thecombined pH as well as temperature sensor and pO2 sensor (Hamilton Bonaduz,Switzerland). The hydrodynamic characterization of the PFR is described in alatter section. As for the previously described two-compartment scale-downsetups, working volumes of 2.5e15 L were reported with focussing on mimickingonly restricted number of compound gradients affecting production (7,16,25,26).Although this setup with 1 L working volume operates in the same workingvolume range as reported, its great variability features wide variety ofperturbations like pH, substrate, temperature as well as oxygen related stresses atthe same time.

Setup design: different PFR configurations The study focused onmimicking regimes with low dissolved oxygen concentrations which are lower thanthe critical dissolved oxygen concentration limit for observing only purely oxidativemetabolism, executed with the small-size two-compartment scale-down setup. Thecritical dissolved oxygen concentration was reported between 30% and 40%,depending on the cultivation conditions and goals (27,28). The gradients in thedissolved oxygen concentration can be reasoned mainly by the insufficient mixingand secondarily by the unsatisfactory aeration. Therefore the low dissolvedoxygen concentration was set in the PFR via the following two differentconfigurations (Fig. 1B and C): Configuration A: Circulation of the broth withbubbles as dispersed system between the STR and the PFR with differentresidence times e with the aim of modelling the occurrence of low dissolvedoxygen concentration regimes due to insufficient mixing in large-scalebioreactors. In this case, however, oxygen is still present, the mass transferbetween the gas and liquid phase can only be carried out mainly via diffusion dueto the lack of satisfactory mixing. Configuration B: Circulation of the broth as onlyliquid without bubbles in the PFR with different residence times e with the aimof modelling regimes with both insufficient aeration and mixing in production-scale reactors. This scale-down setup configuration models those regimes in thelarge-scale bioreactors where the oxygen is not provided at all.

The setup has been characterized before implementation, with respect toperistaltic pump calibrations for the residence times and bubble elimination.Furthermore, investigations of the response times of the sensors in the flow cell,hydrodynamic as well as sedimentation calculations were also executed. The cir-culation flow in the PFR has been proved to be laminar (from 130 to 1010,Re < 2.4$104) with any of the flow rates of chosen residence times (from 7.6$10�3 L/min to 0.2 L/min). The Re number calculationwas executed with the usage of tubingdiameter of the PFR, broth density as well as viscosity, and the linear velocity in thePFR.

Though the flow in the PFR is in the laminar flow regime, in order to have nearlyideal plug flow in the recycle loop, Kenics static mixers were built in the recyclingloop in order to enhance turbulence of the liquid flow. This prevents any axialdispersion and at the same time, provides nearly ideal plug flow. With imple-menting these static mixers, there is negligible residence time distribution in PFR.Therefore, the average residence time was taken into account as the residence timeof PFR for the discussion of the experimental results.

As the broth circulates in the PFR as a liquidesolid system, comparing the linearvelocity in the loop for any residence times (10�2e10�1 m/s range) to the occurrenceof sedimentation (approximately 10�4 m/s range), the total amount of sedimentedbiomass in the PFR can be considered as negligible. The dry-weight measurementsin the PFR also showed that less than 1% of the total produced biomass is sed-imenting in the PFR zone during one batch experiment. Hence, it can be concludedthat the broth which enters to the PFR zone represented the well-mixed broth fromthe STR part of the setup.

Experimental design: repetitive batches With the goal of investigating theimpact of low dissolved oxygen concentration regimes on the overall performance,repetitive batch experiments were carried out with different setup configurationsand different residence times in the PFR. In order to avoid the impact of the inoc-ulum variations, the first experiment of the repetitive batch workflowwas repeatedin both setup configurations (Fig. 1B and C). As the working volume was always 1 L,100 mL from the previous batch remained as inoculum in the bioreactor, 900 mL offresh medium was added under sterile conditions.

Off-line analytics Off-line sampling was done regularly from the bioreactorand from the PFR. The following components were measured: biomass dry weight,substrate concentration (glycerol) andmetabolite concentrations. Acetatewas foundto be the only significant metabolite. Other metabolites (e.g., ethanol) were onlyfound in trace amounts which could not be quantitatively determined with ouranalytical methods (below 1 mg/L). All procedures were performed in duplicates.

To quantify the dry weight, cellular mass was measured gravimetrically in thereactor and in the PFR. Five and two mL of culture broth (from the bioreactor andfrom the PFR, respectively) were added to pre-weighted and dried glass tubes, fol-lowed by the immediate centrifugation at 5000 rpm for 10 min at 4�C (SigmaCentrifuge, Germany). The supernatant was then removed from the top and used forHPLC and enzymatic analysis of the substrate and the metabolite. The pellets werere-suspended twice in 5 mL of ice-cold distilled water using a vortex to wash them,

M

O2

CO2

GasAnalyzer

pH

pO2

Medium for repetitiveBatches

Base

Off-linteReactor

On-line

In-line

pO2

pH&T

Flowcell

M

Off-linePFR

Staticmixer

Flowcell

STR

M

Off-linePFR

Flowcell

STR

M

Bubbleelimination

A

B C

Air

SC

SC

SC

FIG. 1. (A) Simplified PFD diagram of the fully instrumented two-compartment scale-down setup. The PFR part of the setup is indicated with bold lines. The sensors in the flow cellrepresent the state of the broth at the end of the PFR, hence the distance between the end of the flow cell and the re-enter to the bioreactor was minimized. The PFR part was placedinto a waterbath at 30�C (Proline P18, Lauda, Germany). The facility of off-line sampling in the PFR and the more detailed structures of the PFR in the case of both configurations areshown in panels B and C. (B) In the case of configuration A, the broth is circulated in the PFR as a dispersed (gaseliquidesolid) system. The static mixer (Tube mixer, Kenics, UK)helps the broth to maintain its dispersed state in case of laminar flow and helps with decreasing the residence time dispersion in the PFR. The bypass line in the flow cell is requiredto split the flow, as most of the gas is lead in the bypass to minimize disturbance on the sensors in the flow cell (16). (C) In the case of configuration B, the broth is circulated in thePFR as only liquid with solid parts. As the broth enters the PFR, the bubbles of the broth were eliminated with the help of a secondary peristaltic pump (Ismatec, Switzerland). Thepump was operated according to previous calibrations with the flow rate difference between the aerated broth and the simple liquid. Therefore, this configuration was shown to beable to eliminate the bubbles from the broth as soon as it enters to the PFR region.

VOL. 116, 2013 PHYSIOLOGICAL RESPONSES TO OXYGEN LIMITATION 373

and then they were centrifuged again with the same procedure. The tubes with thepellets were dried for 72 h at 105�C oven (Heraeus, Germany). The tubes were keptin an exsiccator before the dry weight was determined.

The substrate glycerol was quantified by HPLC measurements (Agilent 1100Series, Agilent, USA) with SUPELCOGEL C-610H column (9 mm particle size,300� 7.8mm, Sigma Aldrich, USA) at 30�C, 0.1% H3PO4 in dist. water (traces of NaN3)eluent with 0.5 mL/min flow and with RI detector. Acetate was measured by thesame HPLC procedure and also via enzymatic analysis with photometric principleusing an enzymatic robot system (CuBiAn XC, Innovatis, Germany).

Multivariate data analysis The multivariate data analysis of the obtainedrepetitive batch data was carried out using DataLab software (provided by Prof. H.Lohninger, Vienna University of Technology). The data pre-treatments for multi-variate analysis were carried out according to the usual methods with datasetstandardization (calculation of the elements of the covariance matrix) (29). Datacorrelation checks were executed with the calculation of Pearsons’ correlationcoefficients between the applied variables. This correlation coefficient rangesfrom �1 to 1 and a value of 0 implies that there is no linear correlation betweenthe variables. Regarding the chosen variables for PCA, the Pearsons’ correlationcoefficients were always in the range of �0.3 in all cases which enables the use ofmultivariate techniques. For the data clustering, hierarchical cluster analysis wasutilized with Ward’s method and Euclidian distance measurement (30).

RESULTS

Repetitive batches data evaluation The execution of repet-itive batches offers a faster alternative for performing separatebatch experiments with different residence times in the PFR, withthe advantage of utilizing the same initial conditions and cultiva-tion parameters at the same time. This enabled better experimentaldata evaluation and also the recognition of possible patterns on theon-line signals. As to our knowledge, no similar work with repeti-tive batch experiments has been recently reported in the literature.

As the STR and the PFR part of the setup were also equippedwith pH and dissolved oxygen sensors, the difference betweenthese process parameters were monitored on-line. The dissolvedoxygen concentration in the STR during a certain batch experimentshowed the characteristic of exponential growth andwas above thecritical dissolved oxygen concentration. At the same time, in thePFR, the dissolved oxygen values were under the critical dissolved

374 LORANTFY ET AL. J. BIOSCI. BIOENG.,

oxygen concentration limits. The exposure of oxygen limitation inthe PFR was a dependent of the residence time in the PFR regardingsetup configuration A and configuration B (Fig. 2A and B). The pHvalues in the STR and the PFR were also monitored; however, sig-nificant differences could not be observed (data not shown).

According to the standard data evaluation methods (31), volu-metric rates and yields were calculated for the following compo-nents: substrate, biomass, acetate, oxygen, carbon dioxide. With

FIG. 2. (A) Dissolved oxygen concentration in the STR and in the PFR part of the scale-down setup configuration A for six repetitive batch experiments. The following sixbatches had the residence times: 8; 5.5; 3; 1.5; 1; 8 (repeating the first batch) min,respectively. The pO2 in the STR showed the expected behaviour of the dissolved ox-ygen concentration during batch cultivations and was always higher then the criticaldissolved oxygen concentration limit. The pO2 in the PFR indicated that the dissolvedoxygen concentration in the PFR falls below the critical dissolved oxygen concentrationlimit; moreover, the extension of oxygen limitation was dependent on the residencetimes in the PFR. (B) Dissolved oxygen concentration in the STR and in the PFR part ofthe scale-down setup configuration B for six repetitive batch experiments with thefollowing residence times: 7; 5; 3; 2; 1.5; 7 (repeating the first batch) min, respectively.As in the previous case, the pO2 in the STR showed the expected behaviour of dissolvedoxygen concentration during batch cultivations. The pO2 in the PFR indicated that thedissolved oxygen concentration in the PFR falls below the critical dissolved oxygenconcentration limit and the extension of oxygen limitation is strongly dependent onthe residence time in the PFR.

using the dataset of each repetitive batch, elemental and degree ofreduction balances could be closed within the uncertainty of5e10%. As the acetate production was one order of magnitudesmaller in comparison to the biomass production, data reconcilia-tion for pure oxidative growth was performed according to alreadydescribed methods (32). This statistically confirmed that the pro-duction of metabolites was not significant.

Maximum specific growth rate analysis When calculatingspecific growth rate from off-line dry weight measurements, wefound that a volumetric rate of biomass can hardly be quantifieddue to the frequency as well as the noisiness of off-line sampling(data not shown). This was even more pronounced for the specificgrowth rate, of course. Hence, physiological information from off-line data can hardly be extracted. As a requirement, we recentlypublished an approach for estimation of the extraction of infor-mation: the signal on biological activity of a specific compound hasto be significantly higher than the measurement noise (33).Therefore, we investigated the possibilities for extraction ofphysiological information using only on-line data.

To our knowledge, there are no studies published on usingon-line data for the characterization of the scale-up relevant phys-iological response of P. pastoris to critical low dissolved oxygenconcentrations. Therefore, we decided to analyze the influence onmaximum specific growth rate derived from on-line data underoxygen limited process conditions. Initially the impact of residencetimes with respective to different setup configurations (A or B) onthe maximum specific growth rates was analyzed. The maximumspecific growth rates were determined by exponential curve fittingto the carbon dioxide evolution rates (CER, mol/L/h) in the expo-nential growth phases of the repetitive batch experiments. The endpointof eachbatchwas consideredas zeropoint and theexponentialfitting was done for the least 2.5 h time period of each batchexperiment according to a previously published method (31). Theexponential curve fittingwas accomplished for data of an ideal one-compartment batch experiment as well. The coefficients of deter-mination (R2) of thefitswerehigher than0.98 in all cases. Fig. 3A andB demonstrates the exponential fittings in the case of the repetitivebatcheswith configuration A andwith configuration B, respectively.

The results showed that the maximum specific growth rate wasinfluenced differently in the two different setup configurations. Inconfiguration A (residence times 8e1.5 min), there were no sig-nificant differences in the maximum specific growth rates withregarding the different residence times in the PFR and the mmaxvalue of the ideal case. This suggests, however, exposed to the zonewithout oxygen supply; still remaining oxygen was available sup-plied by the bubbles. The bubbles were ideally suspended in theculture broth in the STR homogeneous part and as the circulatedflow is laminar in the PFR and were enhanced by static mixers, noaxial mixing took place.

In the case of configuration B, the exposure to low dissolvedoxygen concentrations with longer residence times had signifi-cantly higher influence on the decrease of the maximal specificgrowth rate. The maximum specific growth rate with the shortestresidence time (1 min) was significantly higher than the maximalspecific growth rate with the longest residence time (7 min) whichsuggests a classification of the performed repetitive batch experi-ments, which was analyzed below using a chemometric approach.

The observations of the influence of the oxygen limitations andresidence times on the maximum specific growth rate were sta-tistically supported by the use of ANOVA on the maximum specificgrowth rates for the two setup configurations. For configuration A,there is no significant difference among the maximum specificgrowth rates with different residence times, as p-value were higher(0.61) and the F-value were lower (1.04) than the critical valueswith applying 10% as the level of significance (0.1 and 55.83,

FIG. 3. (A) Analysis of the influence of oxygen limited process conditions on mmax inthe case of configuration A with different residence times in the PFR in comparisonwith an ideal one-compartment laboratory scale result. (B) Analysis of the influence onmmax in the case of configuration B with different residence times in the PFR incomparison with an ideal batch. In the case of longer residence times in the PFR themaximum specific growth rate has decreased.

FIG. 4. (A) Acetate production in the STR and in the PFR with setup configuration Awith different residence times. (B) Acetate production in the STR and in the PFR withsetup configuration B with different residence times.

VOL. 116, 2013 PHYSIOLOGICAL RESPONSES TO OXYGEN LIMITATION 375

respectively). For configuration B, with considering the same 10% asthe level of significance, the obtained F-value (117.2) was higherand the p-value was lower (0.07). This means that in contrast to theprevious setup case, the null hypothesis, namely, that the means ofthe maximum specific growth rates with this setup configurationwere equal, had to be rejected.

In order to prove that the differences among the maximumspecific growth rates are only reasoned by the different residencetimes in the PFR, the respiratory yields of each batch experimentswere also compared and statistically analyzed. The statistical re-sults (data not shown) also supported the observations which weremade with the maximum specific growth rates.

Acetate production: physiological response to low dissolvedoxygen concentrations Our off-line measurements with HPLCand with enzymatic robot analyzer showed that acetate was theonly metabolite which could be quantitatively detected. Further-more, the measured acetate values using two independent off-lineanalytic methods were in good agreement. Fig. 4A and B describesthe total acetate production per batch in the STR and in the PFRrelated to the amount of used substrate (mg acetate/g substrate)

as the mean value of two different analytic methods in the caseof both used setup configurations. In the case of the ideal one-compartment batch experiment, no acetate production wasdetected. The figures show that acetate was even produced withthe smallest residence time in the PFR, as the metabolism can bechanged in a number of s with the adaptation to new oxygenlimited environments. In addition, the choice of the residencetime ranges in the PFR proves that even 1e1.5 min exposure tocritical low dissolved oxygen concentration conditions wasenough to trigger a metabolic change and get the metabolismadapted to insufficient oxygen supply. Fig. 4A and B indicates thatacetate production was more significant in the PFR part of thesetup. Therefore, the metabolite acetate can be a markermetabolite for the exposure of critical low dissolved oxygenprocess conditions.

The acetate production was found to show no significant ten-dencies with the change of the residence times in the case of setupconfiguration A (Fig. 4A), where bubbles were also present in thePFR. In this case, significant difference in the acetate productionamong the cases with longer and the shorter residence times in thePFR could not be detected.

376 LORANTFY ET AL. J. BIOSCI. BIOENG.,

In contrast, the specific acetate production was found to bedependent on the residence time in the PFR in configuration B. Thespecific production of acetate was significantly higher in the case oflonger residence times than with shorter residence times. Hence,acetate production also supports the observations on themaximumspecific growth rate analysis.

Multivariate data analysis: PCA and clustering Initially,multivariate data analysis concerning hierarchical clustering andprincipal component analysis (PCA) was performed using off-linedata of each batch experiment. As input variables, residence timesin the PFR, the volumetric rates of substrate, biomass and acetatewere used. The results of the multivariate data analyses on off-linedata could not show any differentiation among the different casesof configurations or residence times (data not shown). Subse-quently, multivariate data analyses were performedwith using onlyon-line data. As variables, the following parameters were used: theresidence times in the PFR, the maximum specific growth ratecalculated from the CER data, the volumetric rates of oxygen andcarbon dioxide in the off-gas and the total amount of nitrogenwhich is consumed and can be calculated from the base con-sumption in each of the batch experiments. The datasets werestandardized and their covariance matrices were checked for datacorrelation.

Fig. 5AeD shows the results of the PCAwhich was donewith theabove mentioned on-line dataset of the repetitive batches. The firstthree principal components covered 78% of the total variance andshow how the data can be separated into groups: Configuration Ashorter residence times (3, 1.5 and 1 min); Configuration A longerresidence times (8; 5 and repeated 8 min); Configuration B shorterresidence times (1.5, 2, 3 min); Configuration B longer residencetimes (7, 5 and repeated 7 min). The separation of these groups canbe observed on the three dimensional graph of the first threeprincipal components (Fig. 5D) or on the two dimensional repre-sentation of principal component 1 and principal component 3(Fig. 5B). Principal component 1 and 3 cover 33.7% and 20% of thetotal variance, respectively. The first three principal componentscover only 78% of the total variance; higher numbers of principalcomponents cannot be graphically interpreted easily. However, thefirst three principal components have been shown to be sufficientfor clustering the experimental data. Fig. 5E shows the loadings ofthe used variables by the first three principal components. It can bealso observed that some relevant and also statistically investigatedvariables e the maximum specific growth rate and the respiratorydata e have higher loadings in principal component 1 and 3 whichshow the classification of the experimental data (Fig. 5B).

The identified clusters show that the two scale-down configu-rations, which were designed to investigate different kinds of ex-posures to oxygen limited process conditions, affected the cultureperformance differently. Furthermore, the clusters found with PCAusing only on-line variables were in good agreement with ourpreviously described hypothesis in the section analyzing physio-logical key parameters. Hence, the residence time in the PFR hadsignificant impact on the overall culture performance. However, theanalyses with the off-line data or with the maximum specificgrowth rate could only support this hypothesis in the case of setupconfiguration B. Therefore, this classification can be assigned as themarker of two different physiological states of the culture by themeans of longer or shorter residence times which reflect twodifferent process conditions with oxygen limitations.

The dendrogram shows the hierarchical clustering of experi-mental dataset (data not shown). The closest relation was betweenthe batches which were repeatedly done which supports the factthat memory effect during the repetitive batches can be neglected.Subsequently, the second closest relation was due to the similarresidence time ranges between the two different configurations.

The third clustering shows how the data can be divided into twogroups of data with shorter and longer residence times in the PFR.Finally, the obtained batch dataset can be divided into two groups,one with shorter and one with longer residence times which was inaccordancewith the PCA results and also with the hypothesis aboutthe effect of the residence time in the PFR to the physiological stateof the culture.

PCA with on-line data: a tool for identification limitationsand physiological excursions As information extraction fromthe off-line data was poor, multivariate data analysis on on-linedatasets offers a methodology for drawing conclusions on physio-logical regulations on the obtained experimental data. In addition,the executed PCA showed that the ideal batch behaviour is sepa-rated from the previously described four groups of repetitivebatcheswith the two configurations of the two-compartment scale-down setup (data not shown). As the hierarchical clustering resultpointed out, the ideal batch is obviously in closer relation with therepetitive batches with shorter residence times but still can beseparated. This observation also helps supporting the hypothesisthat longer residence time in the PFR results in greater influence onthe culture performance. As the ideal one-compartment data wasshown to be different from the experimental datasets with oxygenlimited conditions, it can be therefore stated that this multivariatetechnique offers a valuable process development tool for the followup and recognition of stress effects on the culture performancewithsolely using on-line derivable data. To our knowledge, no similarstudies have been reported with exploiting the potential of moni-toring physiological excursions with using on-line data from a two-compartment scale-down setup.

DISCUSSION

Hydraulic scale-up strategies have been settled since decades.However, the scale-up relevant effects on physiology have beenrarely assessed. This study describes a scale-down investigation onthe scale-up relevant physiological response of one P. pastoris strainto oxygen limited process conditions.

The scale-down setup was implemented for executing differentresidence times in the PFR. Based on CFD simulation of production-scale bioreactors (34), the range of 5e15% of the total workingvolume represents the inhomogeneous regime in the relevantscale-down setups (6,13). In the course of this study, the volumeratio of the PFR was fixed at 10% of the total working volume.

With the choice of the range of residence times in the PFR,different investigations canbeaimedaswell. Asmostof the reportedstudies were studying E. coli systems, the residence times in the PFRpartswere chosen in the rangeof 0e200 s (7,35). Itwas found that anexposure time to oxygen limitations as short as a couple of s wasenough for triggering metabolic switch in the case of E. coli cultures(36,37). To our knowledge, no similar studies have been reported forP. pastoris cultures. Hence, a wide range of 1e8 min residence timewas applied in the PFR for physiological studies under oxygenlimited process conditions. Referring to the hydraulic calculationsregarding the flow properties in the PFR, the residence time distri-bution in the PFR were neglected and the discussions were madewith using the average residence time in the PFR.

With the aim of creating a fast and useful process developmenttool, repetitivebatch experimentswere carriedout to characterize thehoststrain itself fromthescale-uprelevantphysiologicalpointof view.The feasibility of oxygen limited conditions in two different configu-rations in thePFRwasdemonstratedbyshowing theon-linedissolvedoxygen raw data in the STR and in the PFR for each experiment.

It is however documented that there is evolution in the course ofrepetitive batch experiments (38), the dependency of the resultswere on one hand checked with a statistical test (Pearsons’

FIG. 5. (AeD) Two and three dimensional representations of the result of PCA with using on-line data of the performed repetitive batches. The first three principal componentsenable clustering the obtained experimental data into four groups. (AeC) The two dimensional representations of the PCA results with principal component 1 (PC1) versus principalcomponent 2 (PC2), principal component 1 (PC1) vs. principal component 3 (PC3) and principal component 2 (PC2) vs. principal component 3 (PC3), respectively. (D) Threedimensional representation of the PCA results for the better observation of the classification of the experimental data. (E) The loadings of the variables by the principal components.It can be observed that the statistically investigated maximum specific growth rate as well as respiratory rates are highly pronounced in principal component 1 and principalcomponent 3.

VOL. 116, 2013 PHYSIOLOGICAL RESPONSES TO OXYGEN LIMITATION 377

correlation check) before applying multivariate data analysis. Thedata correlation was found negligible, therefore the dependency ofthe repetitive experiments could be also neglected. Moreover, thiswork has been carried out with non-induced cultures of a recom-binant P. pastoris strain that obviously show similar behaviour tothe wild type strain under the applied process conditions. On theother hand, as the documented presence of the adaptation to theenvironmental perturbation can impose changes (39), the memoryeffect of the repetitive batch experiments has been tested withrepeating a particular experiment in a repetitive section. Therewere no significant effects found on the relevant experimental re-sults (rates, yields, mmax). It is however to be considered that thiswork does not aim to provide a full quantitative physiologicalassessment of the culture under stress conditions, it aims intro-ducing a tool for faster strain characterization and bioprocessdevelopment.

The effect of low dissolved oxygen concentration regimes on theoverall batch performance could be primarily interpreted by theimpact on the maximum specific growth rate. It was demonstratedin the literature that exposure to oxygen limited process conditionsaffects cell physiology, namely lowers biomass yield and triggersby-product formation in the case of E. coli and P. pastoris cultiva-tions (37,40). Although the reduction of the specific growth rate inscale-down reactor is not novel, to the recent knowledge, howeverthere are no references available on statistical study on thedecrease of the specific growth rate on any microorganisms underoxygen limited process conditions. The exposure to low dissolvedoxygen concentrations with longer residence times in the PFR hadsignificantly greater influence on maximum specific growth rate.

The significant decrease in the maximum specific growth rate inparallel with the increase of the residence time in the PFR could beinterpreted by the fact that the more the culture was exposed tooxygen limited conditions, the more the pure oxidative metabolicpathways were affected. Under oxygen limitation, the syntheticpathways of ATP; moreover, the regeneration of intracellularNADH/NADþ levels might be blocked (41). Hence, the necessaryenergy for synthesizing new building blocks could not be suffi-ciently covered in cultures exposed to critical low dissolved oxygenconcentration.

The metabolite acetate was the main metabolite and producedin the PFR due to the exposure to insufficient oxygen supply. Inthe absence of oxygen, E. coli carries out a mixed-acid fermen-tation (42). By-product formation such as formate and acetatewere also reported under dynamic dissolved oxygen concentra-tion changes (35). In the case of yeasts, shift from oxidative tooxido-fermentative cultivation conditions trigger the formation ofacetaldehyde which based on the intracellular NADH/NADþ level,can be further transferred into acetate or ethanol (43). Forexample, under hypoxic fed-batch cultivation conditions withP. pastoris on the substrate glucose, the detection of ethanol as by-product was published (40). The residence time ranges showedthe proof that even exposures of 1e1.5 min to critical low dis-solved oxygen concentration regimes are sufficient to provokemetabolic perturbation from the purely oxidative metabolism.The metabolism gets adapted to insufficient oxygen supply in thePFR by producing acetate which fact is demonstrated by showingthat the production of acetate was more indicative in the PFRthan in the STR.

378 LORANTFY ET AL. J. BIOSCI. BIOENG.,

Based on the studies on the acetate formation and on maximumspecific growth rate, it can be concluded that the longer the resi-dence time in the PFR, themore the culture performance is affected.The standard off-line data evaluation methods were accomplishedand the elemental and degree of reduction balances could be closedwithin the uncertainty of 5e10%. However, the use of yields andspecific rates derived from off-line datasets could not support anyconclusions on the physiological effects on the culture perfor-mance. Therefore, the exploitation of on-line data and multivariatedata analysis techniques were required. Chemometric tools helpwith the extraction of relevant information in biotechnologicalapplications, considering the large number of inputs as well asoutputs and also their complex relations (44). Multivariate dataanalysis techniques are often applied for the treatment of spec-troscopy data collected during fermentation processes (45,46).Furthermore, principal component analysis has been demonstratedto be a tool for the detection of abnormal process conditions duringthe development of an industrial fed-batch cell culture process (47).Therefore, with the intention of extracting more information on theeffects of the different configurations and residence times on theoverall culture performance, multivariate techniques were utilized.

PCA and clustering of the performed repetitive batchesdemonstrated that the two different setup configurations showeddifferent behaviours and were also separable in the space of prin-cipal components. This was done using on-line data only. The effectof the residence time could be also identified using multivariatetechniques via creating different clusters of batches with longerand shorter residence times in the PFR, thereby representing thedifferent impact on the physiological state of the culture. The in-fluence of the different residence times in the PFR on the culturephysiology was described using three new latent variables as aresult of principal component analysis. In addition, the resultingfour separate groups represent the two significantly different setupconfigurations and two significantly different physiological states.The groups with the shorter residence times are in accordance withthe previously investigated approximate residence time range of0e200 s (7,35). With the chosen shorter residence times, themixing time range of large-scale bioreactors are covered, as mixingtimes are also previously reported in this time range. Hence, thephysiological changes in the culture can therefore be explained bythe formation of gradients due to insufficient mixing. Regarding theother groups with longer residence times, the impact of theadaption to longer exposure to low dissolved oxygen concentra-tions can be observed.

According to the classification based on the different setupconfigurations, it can also be concluded that the two different setupconfigurations represent different types of oxygen limitations. Inthe case of configuration A, oxygen limitation is only reasoned byimperfection in mixing, while oxygen is still present in thedispersed gas phase of the broth. The oxygen mass transfer can stilltake place between the gas and the liquid phase via diffusion.Therefore, the simple statistical analysis of the maximum specificgrowth rate, respiratory yields and acetate production did not showsignificant difference between the residence times, as the deter-mining time parameter for the oxygen limitation is not primarilythe mixing time but also the diffusion coefficient of oxygen in thebroth (approximately 3$10�5 m2/s at 30�C in water). In configura-tion B, where the oxygen limitation is affected by the absence ofaeration and mixing, the significantly different maximum specificgrowth rates with the residence times prove that in this case thelimiting time parameter is related to the mixing time, which cor-responds to the time ranges of the residence time in the PFR.

The results of this work clearly indicate that the combination ofthis scale-down setup and multivariate data evaluation of on-linedata could be a faster methodology for the investigation of stresseffects on the performance of the culture. However, without

extracting pure quantitative physiological information, this methodpresents a process development tool for obtaining only scale-uprelevant physiological knowledge on oxygen limitations. With thepossibility of speeding up the core entities for gaining scale-upknowledge, such as strain characterization, CFD simulation verifi-cation, this presented combined methodology is ready to be usedfor the characterization of other host strains as well.

ACKNOWLEDGMENTS

The authors would like to thank Cosima Koch for the usefulremarks on multivariate data analyses and Oliver Spadiut forproviding the Pichia pastoris Mutþ strain.

References

1. Lara, A. R., Galindo, E., Ramirez, O. T., and Palomares, L. A.: Living withheterogeneities in bioreactors. Understanding the effects of environmentalgradients on cells, Mol. Biotechnol., 34, 355e381 (2006).

2. Oosterhuis, N. M. G. and Kossen, N. W. F.: Dissolved oxygen concentration pro-files in a production-scale bioreactor, Biotechnol. Bioeng., 26, 546e550 (1984).

3. Larsson, G. and Enfors, S. O.: Influence of oxygen starvation on the respiratory ca-pacity of Penicillium chrysogenum, Appl.Microbiol. Biotechnol., 21, 228e233 (1985).

4. Larsson, G., Toernkvist, M., Wernersson, E. S., Traegaerdh, C., Noorman, H.,and Enfors, S. O.: Substrate gradients in bioreactors: origin and consequences,Bioprocess Eng., 14, 281e289 (1996).

5. Bylund, F., Collet, E., Enfors, S. O., and Larsson, G.: Substrate gradient for-mation in the large-scale bioreactor lowers cell yield and increases byproductformation, Bioprocess Eng., 18, 171e180 (1998).

6. Enfors, S. O., Jahic, M., Rozkov, A., Xu, B., Hecker, M., Jurgen, B., Kruger, E.,Schweder, T., Hamer, G., O’Beirne, D., and other authors: Physiological re-sponses to mixing in large scale bioreactors, J. Biotechnol., 85, 175e185 (2001).

7. Onyeaka, H., Nienow, A. W., and Hewitt, C. J.: Further studies related to thescale-up of high cell density Escherichia coli fed-batch fermentations: theadditional effect of a changing microenvironment when using aqueousammonia to control pH, Biotechnol. Bioeng., 84, 474e484 (2003).

8. Bylund, F., Castan, A., Mikkola, R., Veide, A., and Larsson, G.: Influence ofscale-up on the quality of recombinant human growth hormone, Biotechnol.Bioeng., 69, 119e128 (2000).

9. Noorman, H.: An industrial perspective on bioreactor scale-down: what wecan learn from combined large-scale bioprocess and model fluid studies, Bio-technol. J., 6, 934e943 (2011).

10. Neubauer, P. and Junne, S.: Scale-down simulators for metabolic analysis oflarge-scale bioprocesses, Curr. Opin. Biotechnol., 21, 114e121 (2010).

11. Amanullah, A., McFarlane, C. M., Emery, A. N., and Nienow, A. W.: Scale-down model to simulate spatial pH variations in large-scale bioreactors, Bio-technol. Bioeng., 73, 390e399 (2001).

12. Schilling, B. M., Pfefferle, W., Bachmann, B., Leuchtenberger, W., andDeckwer, W. D.: A special reactor design for investigations of mixing timeeffects in a scaled-down industrial L-lysine fed-batch fermentation process,Biotechnol. Bioeng., 64, 599e606 (1999).

13. George, S., Larsson, G., Olsson, K., and Enfors, S. O.: Comparison of the Baker’syeast process performance in laboratory and production scale, Bioprocess Eng.,18, 135e142 (1998).

14. Cortes, G., Trujillo-Roldan, M. A., Ramirez, O. T., and Galindo, E.: Productionof Beta-galactosidase by Kluyveromyces marxianus under oscillating dissolvedoxygen tension, Process Biochem., 40, 773e778 (2005).

15. Jazini, M. and Herwig, C.: Effect of post-induction substrate oscillation onrecombinant alkaline phosphatase production expressed in Escherichia coli,J. Biosci. Bioeng., 112, 606e610 (2011).

16. George, S., Larsson, G., and Enfors, S. O.: A scale-down two-compartmentreactor with controlled substrate oscillations: metabolic response of Saccha-romyces cerevisiae, Bioprocess Eng., 9, 249e257 (1993).

17. Bylund, F., Guillard, F., Enfors, S. O., Tragardh, C., and Larsson, G.: Scale downof recombinant protein production: a comparative study of scaling perfor-mance, Bioprocess Eng., 20, 377e389 (1999).

18. Junne, S., Klingner, A., Kabisch, J., Schweder, T., and Neubauer, P.: A two-compartment bioreactor system made of commercial parts for bioprocessscale-down studies: impact of oscillations on Bacillus subtilis fed-batch culti-vations, Biotechnol. J., 6, 1009e1017 (2011).

19. Cereghino, J. L. and Cregg, J. M.: Heterologous protein expression in themethylotrophic yeast Pichia pastoris, FEMS Microbiol. Rev., 24, 45e66 (2000).

20. Cregg, J. M., Lin, C. J., Shi, J., and Higgins, D. R.: Recombinant proteinexpression in Pichia pastoris, Mol. Biotechnol., 16, 23e52 (2000).

21. Lin, C. G. P., Lin, C. J., Ilgen, C., and Cregg, J. M.: Production of recombinantproteins in fermenter cultures of the yeast Pichia pastoris, Curr. Opin. Bio-technol., 13, 329e332 (2002).

VOL. 116, 2013 PHYSIOLOGICAL RESPONSES TO OXYGEN LIMITATION 379

22. Spadiut, O., Olsson, L., and Brumer, H., III: A comparative summary ofexpression systems for the recombinant production of galactose oxidase,Microb. Cell Fact., 9, 68 (2010).

23. Dietzsch, C., Spadiut, O., and Herwig, C.: A dynamic method based on thespecific substrate uptake rate to set up a feeding strategy for Pichia pastoris,Microb. Cell Fact., 10, 14 (2011).

24. Gurakan, T., Marison, I. W., Von, S. U., Gustafsson, L., and Gnaiger, E.: Pro-posals for a standardized sample handling procedure for the determination ofelemental composition and enthalpy of combustion of biological material,Thermochim. Acta, 172, 251e266 (1990).

25. Caspeta, L., Flores, N., Perez, N. O., Bolivar, F., and Ramirez, O. T.: The effect ofheating rate on Escherichia colimetabolism, physiological stress, transcriptionalresponse, and production of temperature-induced recombinant protein: ascale-down study, Biotechnol. Bioeng., 102, 468e482 (2009).

26. Hewitt, C. J., Onyeaka, H., Lewis, G., Taylor, I. W., and Nienow, A. W.:A comparison of high cell density fed-batch fermentations involving bothinduced and non-induced recombinant Escherichia coli under well-mixedsmall-scale and simulated poorly mixed large-scale conditions, Biotechnol.Bioeng., 96, 495e505 (2007).

27. Lee, C. Y., Lee, S. J., Jung, K. H., Katoh, S., and Lee, E. K.: High dissolved oxygentension enhances heterologous protein expression by recombinant Pichiapastoris, Process. Biochem., 38, 1147e1154 (2003).

28. Cunha, A. E., Clemente, J. J., Gomes, R., Pinto, F., Thomaz, M., Miranda, S.,Pinto, R., Moosmayer, D., Donner, P., and Carrondo, M. J. T.: Methanol in-duction optimization for scFv antibody fragment production in Pichia pastoris,Biotechnol. Bioeng., 86, 458e467 (2004).

29. Kourti, T.: Analysis, monitoring, control, and optimization of batch processes:multivariate dynamic data modeling, Biotechnol. Bioprocess, 33, 43e76 (2012).

30. Romesburg, H. C.: Cluster analysis for researchers. Lulu Press, North Carolina(2004).

31. Wechselberger, P., Seifert, A., and Herwig, C.: PAT method to gather bio-process parameters in real-time using simple input variables and first principlerelationships, Chem. Eng. Sci., 65, 5734e5746 (2010).

32. Jobe, A. M., Herwig, C., Surzyn, M., Walker, B., Marison, I., and vonStockar, U.: Generally applicable fed-batch culture concept based on thedetection of metabolic state by on-line balancing, Biotechnol. Bioeng., 82,627e639 (2003).

33. Wechselberger, P. and Herwig, C.:Model-based analysis on the relationship ofsignal quality to real-time extraction of information in bioprocesses, Bio-technol. Prog., 28, 265e275 (2012).

34. Noorman, H., Morud, K., Hjertager, B. H., Traegaardh, C., Larsson, G., andEnfors, S. O.: CFD modeling and verification of flow and conversion in a 1 m3bioreactor, BHR Group Conf. Ser. Publ., 5, 241e258 (1993).

35. Lara, A. R., Leal, L., Flores, N., Gosset, G., Bolivar, F., and Ramirez, O. T.:Transcriptional and metabolic response of recombinant Escherichia coli tospatial dissolved oxygen tension gradients simulated in a scale-down system,Biotechnol. Bioeng., 93, 372e385 (2006).

36. Schweder, T., Kruger, E., Xu, B., Jurgen, B., Blomsten, G., Enfors, S.-O., andHecker, M.: Monitoring of genes that respond to process-related stress inlarge-scale bioprocesses, Biotechnol. Bioeng., 65, 151e159 (1999).

37. Lara, A. R., Taymaz-Nikerel, H., Mashego, M. R., van, G. W. M., Heijnen, J. J.,Ramirez, O. T., and van, W. W. A.: Fast dynamic response of the fermentativemetabolism of Escherichia coli to aerobic and anaerobic glucose pulses, Bio-technol. Bioeng., 104, 1153e1161 (2009).

38. Koppram, R., Albers, E., and Olsson, L.: Evolutionary engineering strategies toenhance tolerance of xylose utilizing recombinant yeast to inhibitors derivedfrom spruce biomass, Biotechnol. Biofuels, 5, 32 (2012).

39. Zitomer, R. S., Carrico, P., and Deckert, J.: Regulation of hypoxic geneexpression in yeast, Kidney Int., 51, 507e513 (1997).

40. Baumann, K., Maurer, M., Dragosits, M., Cos, O., Ferrer, P., andMattanovich, D.: Hypoxic fed-batch cultivation of Pichia pastoris increasesspecific and volumetric productivity of recombinant proteins, Biotechnol.Bioeng., 100, 177e183 (2008).

41. Baumann, K., Carnicer, M., Dragosits, M., Graf, A. B., Stadlmann, J.,Jouhten, P., Maaheimo, H., Gasser, B., Albiol, J., Mattanovich, D., and otherauthors: A multi-level study of recombinant Pichia pastoris in different oxygenconditions, BMC Syst. Biol., 4, 141 (2010).

42. Smith, M. W. and Neidhardt, F. C.: Proteins induced by anaerobiosis inEscherichia coli, J. Bacteriol., 154, 336e343 (1983).

43. Rodrigues, F., Ludovico, P., and Leão, C.: Sugar metabolism in yeasts: anoverview of aerobic and anaerobic glucose catabolism, pp. 101e121, in:Rosa, C. A. and Szabo, P. G. (Eds.), Biodiversity and ecophysiology of yeasts; theyeast handbook. Springer-Verlag Berlin, Berlin (2006).

44. Rathore, A. S., Bhushan, N., and Hadpe, S.: Chemometrics applications inbiotech processes: a review, Biotechnol. Prog., 27, 307e315 (2011).

45. Kuligowski, J., Quintas, G., Herwig, C., and Lendl, B.: A rapid method for thedifferentiation of yeast cells grown under carbon and nitrogen-limited condi-tions by means of partial least squares discriminant analysis employinginfrared micro-spectroscopic data of entire yeast cells, Talanta, 99, 566e573(2012).

46. Dahlbacka, J., Weegar, J., Weymarn, N., and Fagervik, K.: On-line measure-ment of the substrate concentrations in Pichia pastoris fermentations using FT-IR/ATR, Biotechnol. Lett., 34, 1009e1017 (2012).

47. Gunther, J. C., Conner, J. S., and Seborg, D. E.: Fault detection and diagnosis inan industrial fed-batch cell culture process, Biotechnol. Prog., 23, 851e857(2007).