Large scale gene expression profiling of metabolic shift of mammalian cells in culture

Journal of Biotechnology 107 (2004) 1–17

Large scale gene expression profiling of metabolic shift ofmammalian cells in culture

Rashmi Korkea, Marcela de Leon Gattia, Ally Lei Yin Lau b, Justin Wee Eng Limb,Teck Keong Seowb,1, Maxey Ching Ming Chungb,c, Wei-Shou Hua,∗

a Department of Chemical Engineering and Materials Science, University of Minnesota, 421 Washington Avenue SE, Minneapolis, MN55455-0132, USA

b Bioprocessing Technology Centre, National University of Singapore, Singapore, Singaporec Department of Biochemistry, National University of Singapore, Singapore, Singapore

Received 5 May 2003; received in revised form 22 August 2003; accepted 2 September 2003

Abstract

The metabolic state of hybridoma cells in continuous culture varies with the cultivation condition from which the cultureis initiated. At a metabolically shifted state, cells have markedly reduced glucose and other nutrient consumption and lactateproduction as compared to cells in batch culture or in continuous culture without a metabolic shift. Taking a combined genomicsand proteomics approach, we investigated the molecular mechanism of metabolic shift. Cells from continuous cultures at twodifferent steady states with a glucose consumption to lactate production molar ratio (�L/�G) of 0.08 and 1.4 were studied.Affymetrix GeneChips as well as cDNA microarrays were employed to identify differentially expressed mRNA transcripts, andthe differentially expressed proteins were identified using the 2D gel electrophoresis-mass spectrometry approach. The decreasein glucose metabolism upon metabolic shift is accompanied by a decrease in gene expression of a number of genes involved inits metabolism. However, the number of genes differentially expressed and the extent of differential expression upon metabolicshift are relatively moderate. The change in the expression of metabolic genes at the transcriptional level was confirmed by realtime PCR. The results suggest that metabolic shift is a combined effect of both biochemical events at reaction level and geneexpression at transcription and translation level. This approach of integrating transcriptional profiling, proteomic techniques andbiochemical analysis provides a more global view of the metabolism of mammalian cells in culture.© 2003 Elsevier B.V. All rights reserved.

Keywords:Cell culture; Metabolism; Genomics; DNA microarrays; Affymetrix GeneChips; Proteomics; Two-dimensional gel electrophoresis;Mass spectrometry; Metabolic shift

∗ Corresponding author. Tel.:+1-612-626-0587;fax: +1-612-626-7246.

E-mail address:[email protected] (W.-S. Hu).1 Present address: Department of Biological Sciences, National

University of Singapore, Singapore, Singapore.

1. Introduction

Cultured mammalian cells typically convert morethan half of the total carbon they consume into lac-tate, and more than one-third of the nitrogen theyuptake into ammonia (Zeng et al., 1998a,b). Theproduction of lactate and ammonia by hybridoma

0168-1656/$ – see front matter © 2003 Elsevier B.V. All rights reserved.doi:10.1016/j.jbiotec.2003.09.007

2 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

cells is undesirable because these two metaboliteshave adverse effects on cell growth (Hu et al., 1987;Omasa et al., 1992) and product formation (Doyleand Butler, 1990; Glacken et al., 1986; Kurano et al.,1990; Ozturk et al., 1992). By maintaining nutrientsat low concentrations in a fed-batch culture with dy-namic feeding in response to the estimated metabolicdemand, cellular metabolism has been altered suchthat lactate and ammonia accumulation is reducedwhereas cell and product concentrations are increased(Zhou et al., 1997). The shift in metabolism is re-flected in the molar ratio of lactate produced to glu-cose consumed (�L/�G). Subsequently, it has beendemonstrated that this shift in metabolism can besustained after the culture is switched to a continuousmode of operation and allowed to reach steady-state(low �L/�G state) (Europa et al., 2000). With areduction in conversion of substrates to lactic acid,a much higher cell and monoclonal antibody con-centration is achieved than a control culture that isinitiated from a batch culture without altering cel-lular metabolism (high�L/�G state). Comparablephysiological changes like metabolic shift have beeninduced in mammalian cells like BHK cells (Cruzet al., 1999), hybridoma cells (Follstad et al., 1999),and HEK 293 cells (Siegwart et al., 1999). Physi-ological changes like metabolic shift are probablyaccompanied by gene expression reprogramming inresponse to the environment at the transcription, tran-script processing, translation, or post-translationallevel. A better understanding of gene expression pro-file of cells undergoing the physiological change andthe identification of genes that are directly or indi-rectly involved in the genetic reprogramming willhelp in further exploitation of this phenomenon. Thegene expression profile related to metabolic shift andother phenomena of interest in cell culture processingcan be studied at the transcript level with microarraysand at the protein level with 2D gel electrophoresis(2-DE), as reviewed previously (Korke et al., 2002).By combining gene expression investigation at bothmRNA and protein levels, and biochemical studies atmetabolic reactions level, one can better understandthe regulation of metabolic shift. We have previouslyreported proteomic investigation on metabolic shift(Seow et al., 2001). We have also employed suppres-sion subtraction hybridization (SSH) to isolate genesequences which are differentially expressed upon

metabolic shift (Korke, 2002). SSH is a powerfulmethod for isolating rare sequences that are differ-entially expressed. However, the probabilistic natureof SSH and the laborious procedures limit its valuefor repeated applications. DNA microarrays, on theother hand, with a large number of gene sequencesas probes allow for a large-scale survey of gene ex-pression. In this investigation of metabolic shift, weemployed both cDNA and oligo DNA microarrays aswell as a more enhanced proteomic technique thanwe previously used. In this report we compare theresults obtained with these three methods.

2. Materials and methods

2.1. Cell culture

The mouse–mouse hybridoma cell line, MAK,which produces IgG monoclonal antibody and themedium used have been described previously (Zhouand Hu, 1994, 1995). Cells are maintained in a de-fined, serum-free medium consisting of a 1:1 mixtureof Dulbecco’s modified Eagle’s medium (DMEM)and Ham’s F12 medium supplemented with trans-ferrin, 2-mercaptoethanol, ethanolamine,l-ascorbicacid, sodium selenite and putrescine as describedpreviously (Europa et al., 2000). The concentrationsof glucose and glutamine used were 17.5 and 4 mM,respectively.

The continuous cultures were carried out in com-puter controlled stirred tank bioreactors with on-lineturbidity, oxygen consumption rate measurement andfeeding rate and effluent flow rate control. Two dif-ferent types of continuous cultures were performed.The continuous culture without metabolic shift (i.e.with a high�L/�G) was initiated as a batch culture,whereas the metabolically shifted culture was initiatedin a fed-batch mode as described previously to firstallow for decrease in�L/�G before it was switchedto a continuous mode. The bioreactor, setup and op-eration of fed-batch and continuous culture have beendescribed previously (Europa et al., 2000). The com-position of feed medium for the continuous cultureswas the same as that used for cell maintenance exceptfor glucose, glutamine, and other amino acids. Thefeed glucose and glutamine concentrations were 6.11and 4 mM, respectively while other amino acids were

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 3

1.9-fold that in cell maintenance media. The dilutionrate of the continuous culture was 0.03 h−1. Both cul-tures were allowed to reach steady-state as judged byless than 5% variation in the concentrations of cells,glucose and lactate.

2.2. RNA isolation

The cells in high and low�L/�G states were har-vested by centrifugation and stored in liquid nitrogenat about 40 million cells per cryovial using minimumamount of media. Total RNA was extracted from cellsamples twice with TriZol® (Invitrogen, Rockville,MD, USA) reagent according to the manufacturer’sprotocol. The resultant RNA pellet was resuspendedin 100�l water and stored at−80◦C until further pro-cessing. Purity and integrity of RNA was verified byrunning 5�g in a 1.4% agarose gel in 10 mM sodiumphosphate buffer, pH 6.8 at 100 V (Pelle and Murphy,1993).

Messenger RNA was isolated from total RNA usingOligotex® Spin Column isolation kit (Qiagen, SantaClara, CA, USA) according to the manufacturer’s pro-tocol. The concentration of the mRNA was estimatedspectrophotometrically by measuring the absorbanceat 260 nm and stored in 2�g aliquots at−80◦C untiluse.

2.3. cDNA microarrays

2.3.1. Microarray constructionThe cDNA microarray was constructed as described

in the University of Minnesota Mouse MicroarrayGroup website (http://www.bipl.ahc.umn.edu/MouseMicroArray/MainPage.htm). Briefly, poly-l-lysinecoated slides were obtained by dipping microslides (Fisher Scientific, Springfield, NJ, USA) in apoly-l-lysine (Sigma–Aldrich, St. Louis, MO, USA)solution in phosphate buffer saline (PBS) for 30 minfollowed by rinsing in water and drying completely.PCR products were amplified from the Brain Molec-ular Anatomy Project (BMAP) (http://www.trans.nih.gov/bmap/resources/resources.htm) library using M13Forward (5′-GTTTTCCCAGTCACGACGTTG-3′)and M13 Reverse (5′-TGAGCGGATAACAATTTCA-CACAG-3′) primers. PCR products were purifiedwith Multiscreen PCR purification filters (MilliporeCorporation). They were transferred to 384 well

plates and spotted on poly-l-lysine coated slides induplicate using the Total Array System (BioRobotics,Woburn, MA, USA) robotic arrayer. Following print-ing of slides, the spots were rehydrated in a humid-ified chamber (Sigma–Aldrich) in the presence of1× SSC at 35◦C for 10 min and snap-dried on ahot-plate for 3–5 s. To prevent non-specific bindingon the surface, the hydrated slides were incubated inblocking solution consisting of 0.16 M succinic an-hydride (Sigma–Aldrich) in 1-methyl-2-pyrrolidinone(Sigma–Aldrich) and containing 42.85 mM sodiumborate (pH 8.0) for 15 min. Excess blocking solutionwas drained and the slides were dipped in boilingwater for 1.5 min to denature the DNA on the spots.Finally, the slides were washed in 95% ethanol, driedby centrifugation at 200× g for 2 min at room tem-perature, stored at room temperature and used within30 days.

2.3.2. Labeling and hybridizationThe microarray labeling and hybridization were car-

ried out as described previously (May et al., 2001,http: //www.bipl .ahc.umn.edu/MouseMicroArray/MainPage.htm). In short, 2�g of mRNA from eachsample were used for first strand cDNA synthesisby reverse transcription using Superscript II® Re-verse Transcriptase (Invitrogen) in the presence ofamino allyl-dUTP (Sigma–Aldrich) according to themanufacturer’s protocol. The RNA was extracted andaliquoted at one time for all the labeling reactions toavoid variation between the RNA populations usedfor labeling in replicate experiments. The spottedampicillin resistance gene, which has minimal ho-mology to any spotted sequence, was used as a checkfor normalization. Five nanograms of mRNA gen-erated by in vitro transcription using MAXIscript®

SP6 kit (Ambion Inc., Austin, TX, USA) was spikedinto each reverse transcription reaction. The reac-tions were hydrolyzed by adding 10�l of 1 N NaOHand 10�l 0.5 M EDTA and incubated at 65◦C for15 min and then neutralized with addition of 25�l of1 M Tris–HCl pH 7.4. The cDNA was cleaned withthe Microcon 30 filter system (Millipore, Billerica,MA, USA) according to the manufacturer’s protocolfor subsequent coupling to the fluorolink monofunc-tional dyes (Amersham Biosciences, Piscataway, NJ,USA). Aliquots of Cy3 and Cy5 dye were added toeach of the pair of cDNA samples and incubated in

4 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

9�l of 1 M sodium borate buffer (pH 9.0) for 1 hat room temperature in the dark. The labeling dyeswere switched in the dye swap experiment whilekeeping all the other steps unchanged. The reactionswere then quenched by adding 4.5�l of 4 M hydrox-ylamine (Sigma–Aldrich) and incubated in the darkfor 15 min at room temperature. Unincorporated dyewas removed by cleaning up the cDNA using theQiaQuick PCR purification kit (Qiagen) accordingto the manufacturer’s protocol. The volume of thelabeled cDNAs was reduced to 15�l in a centrifu-gal evaporator (Jouan, Winchester, VA, USA) before3�l of 20× SSC and 1.5�l of 1 mg/ml poly A wereadded (Amersham Biosciences) to these samples.Then 0.45�l of 10% sodium dodecyl sulfate (SDS)was added before denaturing the labeled cDNAsby boiling for 2 min and letting cool down at roomtemperature for 5–10 min. The samples were thentransferred to the slide, placed in the hybridizationchamber and carefully covered with a coverslip, andhybridized at 63◦C for 5–6 h. The slides were thenwashed in wash solution I (2.85% SDS in 0.57× SSC)followed by wash solution II (0.057× SSC) and driedby centrifugation for 5 min at room temperature.

2.3.3. Image and data analysisThe hybridized microarrays were scanned using

ScanArray® Express (Packard BioScience Company,Billerica, MA, USA) Microarray Analysis System.The intensity of each spot was obtained for each pairof sample using the QuantArray® Analysis software(Packard BioScience Company) by the adaptive im-age quantification method. The spots for which thebackground-subtracted intensity of either Cy3- orCy5-labeled cDNA was less than twice of the back-ground standard deviation were excluded from furtheranalysis. The fluorescence intensity of the spots wasnormalized to compensate for different dye labelingeffects using cDNA Microarray Analysis Program(Tseng et al., 2001). Using this program, the microar-ray data was visually inspected through M–A plots toverify the need of non-linear normalization (Dudoitet al., 2002). Invariant genes were selected through theprogram’s built-in iterative process and used to obtainthe normalized log10-ratios for all sequences. None ofthe sequences classified invariant was later found dif-ferentially expressed. The ampicillin resistance genewas used to check for normalization efficiency.

The mean and standard deviation of the log2-trans-formed ratio for each spot were calculated from thereplicated microarray experiments. The data fromreplicate spots whose log2-transformed ratio wasmore than 1.5-fold the standard deviation from theaverage of those replicates was discarded and the av-erage log2-transformed ratio for the remaining spotswas calculated. In addition, a paired differencet-testwas performed for all sequences using this averagelog2-transformed ratio, corresponding standard devi-ation, and number of replicates. Among these spots,those whose absolute average log2-transformed ratiowas equal to or greater than 0.485 (i.e. arithmeticfold change≥1.4) were considered candidate dif-ferentially expressed sequences. Lastly, only thosesequences with higher than 90% confidence accord-ing to our P-values were classified as differentiallyexpressed.

2.4. Affymetrix GeneChips

2.4.1. GeneChip hybridization and stainingAffymetrix Murine Genome U74A v.2 GeneChips

(Affymetrix, Santa Clara, CA, USA) were used. The12,488 sequences include approximately 6,000 se-quences from the Mouse UNIGENE database whichhave been functionally characterized, while the restrepresent EST clusters.

Sample preparation and hybridization were per-formed following the recommendations given in theAffymetrix Technical Manual. In short, 2�g aliquotsof the same mRNA pool used in the cDNA mi-croarray experiments were reverse transcribed usingSSII-RT (Invitrogen) and T7-(dT)24 primer (ProligoLLC, Boulder, CO, USA) for the first strand synthesisfollowed by second strand synthesis using DNA poly-merase I (Invitrogen). The double-stranded cDNAwas cleaned with Phase Lock Gel (Eppendorf, West-bury, NY, USA) and stored at−80◦C. All aliquotswere in vitro transcribed and biotin-labeled with EnzoBioArray High Yield RNA Transcript Labeling Kit(Enzo Diagnostics, Farmingdale, NY, USA) beforehybridization. One microgram of mRNA typicallyyielded between 30 and 60�g of cRNA. After cleanupwith RNeasy columns (Qiagen), 20�g of cRNAwere fragmented into pieces from 35 to 200 bases,as confirmed by gel electrophoresis analysis. The20�g fragmented cRNA samples and all necessary

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 5

controls were hybridized overnight followed by probewashing and staining with streptavidin phycoerythrin(SAPE).

2.4.2. Image and data analysisThe GeneChips were scanned at the University

of Minnesota Affymetrix Lab. GeneChip ExpressionAnalysis Suite v.4.0 (Affymetrix) was used to cal-culate average difference values, and to scale totalchip fluorescence so that the mean chip intensityis equal for all samples (1000). Appropriateness oflinear normalization was verified by evaluating thelinearity between sample intensities. Firstly, geneswere classified as absent or present according to thesoftware algorithm and its default values. Amongstthe “present” sequences, those with a fold-changehigher than 1.6 were classified as candidate differ-entially expressed sequences. The difference callprovided by the Analysis Suite software analysis wasalso taken into account in this classification. If morethan 50% of the replicate comparisons for a givensequence were assigned a No Change (NC) call bythe software, then the sequence was not consideredfor differential expression. Lastly, theP-value for allsequences was calculated with the GeneData Expres-sionist suite v.4.0. Among the remaining candidates,only genes with higher than 90% confidence weretaken as differentially expressed. Genes were an-notated with the aid of the Analysis Center in theNetAffx website (http://www.netaffx.com/index2.jsp)as well as directly from the UNIGENE website (http://www.ncbi.nlm.nih.gov/UniGene/).

2.5. Real-time comparative quantitative PCR

2.5.1. Reaction preparationReverse transcription of mRNA was performed ac-

cording to the SuperscriptTM II RNase H− ReverseTranscriptase (Invitrogen) protocol for cDNA synthe-sis using 2�g of mRNA and 5�g of oligo-dT primers.The RT product was aliquoted and stored at−80◦C.The genes used for real-time PCR were phospho-fructokinase (PFK), triosephosphate isomerase (TPI),phosphoglycerate mutase (PGM), enolase (ENO3),pyruvate kinase (PK), and lactate dehydrogenase(LDH1). Two endogenous genes (�-tubulin (TUBA2)and ubiquitin B (UBB)) that showed high evidence ofnot being differentially expressed (ratio∼1 andP∼1)

according to microarray analysis were selected for nor-malization. Forward and reverse primers shown in theAppendix were designed to generate intron-spanningproducts of 100–150 bp using Array Designer 2 soft-ware (PREMIER Biosoft International, Palo Alto,CA, USA). Amplicon size and reaction specificitywere confirmed by agarose gel electrophoresis. Each12�l PCR mix contained 6�l SYBR® Green PCRMaster Mix (Applied Biosystems), 200 nM forwardand reverse primers and 10 ng template. Cycle condi-tions for the ABI 7700 (Applied Biosystems, FosterCity, CA, USA) were 94◦C for 15 min and 40 cyclesof 94◦C for 15 s, 57◦C for 30 s and 72◦C for 30 s. Athreshold of 10-fold the average standard deviation ofthe fluorescent level of the baseline cycles was usedto determineCt.

2.5.2. Data analysisThe calculated threshold cycle (Ct,normgenes) for

each gene amplification was normalized to an averageCt of the two endogenous control genes amplifiedfrom the corresponding sample. These values cameto be identical for both samples (14.5). The normal-ized Ct averaged from triplicate runs for each geneand each metabolic state led to the calculation of thefold change in mRNA level from low to high�L/�Gstates according the following formula (Talaat et al.,2002):

fold change= −2��Ct , where��Ct for genej

=(Ct,j − Ct,norm genes

)low �L/�G

−(Ct,j − Ct,norm genes

)high�L/�G

Unpairedt-test on the normalizedCt differences wasused to inquire about the statistical significance ofthe differential expression between the two metabolicstates.

2.6. Proteomic analysis

2.6.1. Sample preparation and 2D gel electrophoresisThe samples were prepared and separated by

2-DE essentially as described previously (Seowet al., 2001), with modifications in sample applicationfor the cup loading strip holders (Amersham Bio-sciences). The 18-cm pH 3-10 non-linear IPG stripswere rehydrated overnight in 340�l of 7 M urea

6 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

(Bio-Rad Laboratories, Hercules, CA, USA), 2 Mthiourea (Fluka Chemie AG, Buchs, Switzerland), 4%3-[(3-cholamidopropyl)dimethylammonio]-1-propane-sulfonate (USB, Amersham Biosciences), 20 mMdithiothreitol (Bio-Rad), and 0.5% IPG buffer (Amer-sham Biosciences). After rehydration, 10�l of samplewas applied onto the anodic end of the IPG strip us-ing a loading cup. The amount of protein loaded was∼120�g.

2.6.2. Gel staining, image analysis and massspectrometry

Visualization of the 2-DE gels was performed usingsilver staining, and the differentially displayed spotson the gels were detected using visual inspection andthe PDQuest software (Bio-Rad) as described previ-ously (Seow et al., 2001). Spots of interest were thenexcised and further prepared for mass spectrometryanalysis in 96-well microtiter plates. After reduction,alkylation, and trypsin digestion, the peptides wereextracted and identified using an Applied BiosystemsVoyager-DE STR matrix-assisted laser desorption/ionization time-of-flight mass spectrometer (MALDI-TOF MS) and/or tandem mass spectrometry (MS/MS)analyses on an Applied Biosystems QSTAR hybridquadrupole TOF MS/MS system equipped with anano-electrospray ionization (nESI) source as previ-ously described (Seow et al., 2001).

3. Results

3.1. Transcriptional analysis usingcDNA microarray

Hybridoma MAK cells were cultivated in con-tinuous cultures with different metabolism untilsteady-state was achieved. The metabolically shiftedculture was initiated in fed-batch mode with glucosecontrolled at low levels (9 mM). During the fed-batchperiod lactate production decreased. After switchingto continuous mode, the culture reached a steady-statewith a �L/�G of 0.08. In the case that the culturewas initiated in a batch mode the steady-state�L/�Gwas 1.4. Although the feed medium and the dilutionrate for both cultures were identical, the metabolismof the two cultures was very different. These ex-periments have been described previously (Europa

et al., 2000). The RNA was obtained from cells inthose two steady-states. The high density cDNA mi-croarray used to compare transcript expression ina high-throughput fashion includes 4972 genes and5203 ESTs. To account for the different efficienciesof Cy3 and Cy5 incorporation as well as the dye bi-ases caused by experimental variability in couplingreaction, dye swapping was carried out in duplicateby exchanging the dye used in labeling each pair ofsamples. Four cDNA microarrays were used to com-pare the two states. At least two replicates of eachsequence were present on each cDNA microarray.Hence, at least 8 replicate spots were used for eachgene to obtain the gene expression ratio. A total of123 sequences were identified as differentially ex-pressed (fold change >1.4,P < 0.1): 19 annotatedgenes up-regulated at high�L/�G state (Table 1) and24 up-regulated at the low�L/�G state (Table 2),in addition to 80 ESTs. The level of differential ex-pression is relatively small, from 1.4 to 3.3. Amongthose identified as differentially expressed a few areinvolved in central metabolism, whereas the role ofother differentially expressed genes is less clear.

3.2. GeneChip analysis

The same two mRNA samples used in the cDNAmicroarray experiments were hybridized in dupli-cate (4 arrays total) to Affymetrix GeneChip MurineGenome U74A v.2 with a total of 12,488 probes onthe chip. Four comparisons were possible with theduplicate experiments. A total of 62 sequences weredifferentially expressed between high and low�L/�Gstates. 57.2% of the genes were classified as absent.Tables 3 and 4list the 29 and 12 sequences thatare up-regulated in the high and low�L/�G states,respectively. In addition to those annotated genes,there are 12 and 9 ESTs up-regulated in high andlow �L/�G states, respectively. Most ESTs have lowintensity levels, indicating a low degree of abundanceor homology so the data analysis on these sequencesis less accurate. The level of differential expressionof those annotated genes, all withP-values smallerthan or equal to 0.1, range from 1.6- to 9.9-fold. Themajor classes of genes affected are involved in cen-tral metabolism. Overall, the number of genes whoseexpression level changed between the two metabolicstates is small.

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 7

Table 1Sequences up-regulated in high�L/�G state in cDNA microarray analysis

Sequence annotation Fold change P-values

Central metabolismTriosephosphate isomerase −2.8 0.093× 10−2

Enolase 1 −2.2 0.009× 10−2

Enolase 3, beta muscle −1.6 0.68× 10−2

Pyruvate kinase 3 −2.0 0.043× 10−2

Lactate dehydrogenase 1, a chain −1.9 0.011× 10−2

TranscriptionQuaking −1.5 0.022× 10−2

Activated RNA polymerase II transcriptional coactivator P15. A general coactivator, functionscooperatively with TAFS and mediates functional interactions between upstream activatorsand general transcriptional machinery as it binds ssDNA

−1.5 10× 10−2

Hairy and enhancer of split related-1 (Hesr-1). A transcriptional repressor of genes that requirea BHLH protein for their transcription and may act as a negative regulator of myogenesis byinhibiting the functions of MYOD and ASH1

−1.4 3.4× 10−2

Nucleotide metabolismNucleoside diphosphate kinase A. Metastatic process-associated, major role in synthesis of

nucleoside triphosphates (not ATP)−1.4 0.41× 10−2

EST, highly similar to cytosolic purine 5′-nucleotidase (H. sapiens). May have a critical role inthe maintenance of a constant composition of intracellular purine/pyrimidine nucleotides incooperation with other nucleotidases

−1.4 0.45× 10−2

Amino acid metabolismAromatic-l-amino-acid decarboxylase −1.4 0.23× 10−2

EST, highly similar to s-adenosylmethionine synthetase gamma form (H. sapiens). Catalyzesthe formation of s-adenosylmethionine from methionine and ATP

−1.4 0.52× 10−2

Lymphocyte specificLymphocyte antigen 6 complex, locus E. A group of small cysteine-rich cell surface proteins

that are anchored in the membrane by a glycosyl-phosphatidylinositol moieties and involvedin T-cell development

−3.3 0.007× 10−2

Small inducible cytokine A27. A chemotactic factor that attracts skin-associated memoryT-lymphocytes, may play a role in mediating homing of lymphocytes to cutaneous sites orin cell migration during embryogenesis

−1.4 0.38× 10−2

MiscellaneousTissue inhibitor of metalloproteinase. Complexes with metalloproteinases (collagenases) and

can irreversibly inactivate them−1.6 1.4× 10−2

Fatty acid binding protein 5, epidermal. High specificity for fatty acids with highest affinity forC18 chain length

−1.5 0.094× 10−2

EST, weakly similar to link protein (M. musculus). Stabilizes aggregates of proteoglycan withhyaluronic acid in extracellular cartilage matrix

−1.5 1.4× 10−2

p190 family of GTPases contain N-terminal GTPase, C-terminal Rho GAP domain −1.6 0.073× 10−2

Synuclein. Belongs to a novel class of substrates for G protein-coupled receptor kinases −1.4 3.6× 10−2

3.3. Real-time PCR analysis

The differential expression identified by microarrayanalysis was validated by real-time PCR on repre-sentative genes using the same mRNA samples. Sixglycolysis pathway genes were used for this valida-tion: phosphofructokinase (NT039621), triosephos-phate isomerase (NT039356), phosphoglycerate

mutase (NT039692), enolase (NT039520), pyru-vate kinase (NT039474), and lactate dehydrogenase(NT 039420). Table 5 shows the results for thesegenes normalized to two endogenous genes (� tubulinand ubiquitin B).P-values for all 6 genes tested fallbelow 0.1; in fact, only lactate dehydrogenase has aP-value higher than 0.05 (0.088). A comparison tothe microarray results (Tables 1 and 3) shows good

8 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

Table 2Sequences up-regulated in low�L/�G state in cDNA microarray analysis

Sequence annotation Fold change P-value

Protein processingEST, highly similar to ribosomal protein S21 (R. norvegicus) 2.1 0.0057× 10−2

Proteasome (prosome, macropain) subunit, beta 3. A multicatalytic proteinase complexcharacterized by its ability to cleave peptides and involved in an ATP/ubiquitin-dependentnon-lysosomal proteolytic pathway

1.8 0.015× 10−2

Lymphoctye Met-ase 1. 30-kDa serine protease that cleaves peptide substrates after methionineand leucine

1.8 0.026× 10−2

Hepatocyte growth factor activator inhibitor type 2. A new Kunitz-type serine protease inhibitor 1.5 0.070× 10−2

EST, highly similar to peptidyl-prolylcis–trans isomerase (A. thaliana). Accelerate folding ofproteins

1.5 2.3× 10−2

FK506-binding protein 12.6. FK506-binding protein 12.6 belongs to the peptidyl-prolylcis–trans isomerase family

1.4 0.41× 10−2

Ankyrin 3, epithelial. Membrane protein, plays an important role in polarized distribution ofmany integral membrane proteins

1.9 0.30× 10−2

Beta-spectrin 3. Associated with Golgi and cytoplasmic vesicles where spectrin is an importantstructural component of the plasma membrane skeleton

1.5 4.5× 10−2

EST, moderately similar to type II membrane protein (H. sapiens). A novel protein 1.4 3.58× 10−2

MitochondrialMitochondrial import inner membrane translocase subunit TIM13A. Strong resemblance to a

set of zinc-binding yeast proteins (Tim8p, Tim9p, Tim10p, Tim12p, and Tim13p) that areimplicated in the import of a class of transmembrane carrier proteins from the cytoplasm tothe mitochondrial inner membrane

1.5 0.14× 10−2

Mitochondrial import inner membrane translocase subunit TIM9 B 1.4 1.7× 10−2

Acyl-CoA-binding protein. Binds medium- and long-chain acyl-CoA esters with very highaffinity and may function as an intracellular carrier of acyl-CoA

1.4 0.92× 10−2

Neuron specificActivity-dependent neuroprotective protein. A novel neuronal protein 1.7 1.3× 10−2

Calsyntenin 1. A proteolytically processed postsynaptic membrane protein with a cytoplasmiccalcium-binding domain

1.5 1.8× 10−2

MiscellaneousH3 histone, family 3B. A central role in nucleosome formation 1.7 0.16× 10−2

Myeloid differentiation primary response gene 118. A novel gene induced by multiplecytokines, involved in the regulation of growth and apoptosis

1.6 0.030× 10−2

EST, highly similar to lens epithelium-derived growth factor (H. sapiens). A regulatory factor,may play important role for growth and survival of wide range of cell types

1.4 0.55× 10−2

EST, highly similar to orphan G protein-coupled receptor FEX (M. musculus). May beimportant receptor for signals controlling growth, differentiation of specific embryonictissues as it codes for a novel putative G protein-coupled receptor expressed in follicles

1.4 0.54× 10−2

EST, weakly similar to tubulin alpha-1 chain (M. musculus). The major constituent ofmicrotubules

1.5 0.072× 10−2

Potassium voltage gated channel, shaker related subfamily, member 5. Mediates thevoltage-dependent potassium ion permeability of excitable membranes assuming opened orclosed conformations in response to the voltage difference across the membrane and formsa potassium-selective channel through which K+ ions may pass in accordance with theirelectrochemical gradient

1.5 1.7× 10−2

snRNP core Sm protein homolog (Sm-X5) gene, two alternative splice forms.Doughnut-shaped heteromer in humans, binds specifically to the 3′-terminal U-tract of U6snRNA thereby facilitating U4/U6 duplex formation in vitro

1.5 0.18× 10−2

HLA-B associated transcript 4. A MHC class III specific gene 1.5 2.1× 10−2

RAN guanine nucleotide release factor. Guanine nucleotide release factor for Ran is aRas-related GTPase that is essential for the transport of protein and RNA between thenucleus and the cytoplasm

1.4 0.34× 10−2

Retinol dehydrogenase type 5. Catalyzes final step in the biosynthesis of 11-cis-retinaldehyde,the universal chromophore of visual pigments and active in the presence of NAD ascofactor but not in the presence of NADP

1.4 0.91× 10−2

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 9

Table 3Sequences up-regulated in high�L/�G state in AffymetrixGeneChip analysis

Sequence annotation Foldchange

P-value

Central metabolismPhosphofructokinase liver B-type −3.3 0.80× 10−2

Aldolase 1 A isoform −3.1 0.14× 10−2

Triosephosphate isomerase −3.3 1.8× 10−2

Phosphoglycerate kinase 1 −2.1 2.0× 10−2

Phosphoglycerate mutase 1 −1.9 7.3× 10−2

Enolase 3 beta muscle −1.8 0.36× 10−2

Pyruvate kinase 3 −1.9 4.5× 10−2

RNA bindingDead aspartate-glutamate-alanine-

aspartate box polypeptide 5.RNA-dependent helicase;DEAD-box proteins and theirrelatives are required for mostfundamental processesinvolving RNA

−2.6 3.4× 10−2

Lipid/fatty acidsNAD(P) dependent steroid

dehydrogenase like. Involvedin cholesterol biosynthesis

−1.8 0.59× 10−2

Fatty acid-binding protein 4,adipocyte

−3.7 0.21× 10−2

Cytoskeleton related proteinsLamin B1. Lamins are

components of the nuclearlamina, a fibrous layer on thenucleoplasmic side of the innernuclear membrane, which isthought to provide aframework for the nuclearenvelope and may also interactwith chromatin

−1.7 0.94× 10−2

Signal transductioncAMP-dependent protein kinase

inhibitor, gamma form.Extremely potent competitiveinhibitor of cAMP-dependentprotein kinase activity

−1.8 5.5× 10−2

Chemokine (c-x-c) receptor 3 −3.8 0.13× 10−2

TOB1 protein (transducer oferbb-2 1). Anti-proliferativeprotein that interacts with theerbb-2 receptor tyrosine kinase

−2.1 9.7× 10−2

Death-associated kinase 2 −2.1 0.049× 10−2

Muscle and microspikes RAS −2.2 1.1× 10−2

Lymphocyte specificLymphocyte specific 1 −4.1 10× 10−2

IA-associated invariant chain.Plays a critical role in MHCclass ii antigen processing

−2.0 1.2× 10−2

Table 3 (Continued)

Sequence annotation Foldchange

P-value

Lymphocyte antigen 6 complex,locus E. Involved in T-celldevelopment

−4.0 1.46× 10−2

Lymphocyte antigen 6 complex,locus D

−10.0 0.40× 10−2

CD52 antigen −2.6 0.72× 10−2

MiscellaneousAcetyl-coenzyme A synthetase 2 −2.3 2.9× 10−2

ATPase, Ca2+ transporting −1.7 7.9× 10−2

Histidine rich calcium bindingprotein

−1.7 0.91× 10−2

Membrane-spanning 4-domains,subfamily A, member 2

−2.5 2.0× 10−2

Tissue specific transplantationantigen p198-7

−1.8 3.8× 10−2

Insulin-like growth factor-bindingprotein 5

−1.8 2.1× 10−2

Potassium intermediate/smallconductance calcium-activatedchannel

−1.7 1.5× 10−2

Procollagen-proline,2-oxogulatarate 4-dioxygenase,alpha 1

−2.6 0.072× 10−2

agreement. Phosphofructokinase and phosphoglycer-ate mutase were not called differentially expressedin the cDNA microarray analysis although their foldchanges were−1.3 and−1.7, respectively, becausetheir statistical significance was not high (P > 0.1).Lactate dehydrogenase was not classified as differen-tially expressed in the Affymetrix GeneChip analysisalthough its fold change was−1.6 for the samereason.

3.4. Proteomic analysis

Cellular proteins were isolated and separated on2-DE gels within a pH range of 3–10. The cells usedwere from the same two cultures as used for transcrip-tion analysis. Nine and eight gel replicates were car-ried out for low and high�L/�G states, respectively;four of the most reproducible gels for each state wereused for differential analysis. A total of 43 spots dif-ferentially expressed between the high (Table 6) andlow (Table 7) �L/�G states were identified.

As in transcriptional analysis, the proteins identifiedas differentially expressed included those involved

10 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

Table 4Sequences up-regulated in low�L/�G state in AffymetrixGeneChip analysis

Sequence annotation Fold changeP-value

TranscriptionActivating transcription factor 5 1.9 10× 10−2

ApoptosisCaspase 7. Involved in the

activation cascade of caspasesresponsible for apoptosisexecution. Overexpressionpromotes programmed celldeath

3.3 4.6× 10−2

Cell cycleCyclin D2. Participates in cell

cycle control2.5 2.0× 10−2

Nucletotide metabolismGuanylate nucleotide binding

protein 32.4 7.2× 10−2

Lymphocyte specificT-cell death associated gene 2.1 2.5× 10−2

Signal transductionCytokine inducible

SH2-containing protein 22.7 1.6× 10−2

MiscellaneousTransmembrane 4 superfamily

member 72.4 2.3× 10−2

Uterine mRNA 1.7 0.21× 10−2

Demethyl-Q 7 2.1 0.37× 10−2

Calcium channel,voltage-dependent, gammasubunit 6

2.3 10× 10−2

Claudin 4 2.1 1.0× 10−2

T complex testis-specific protein 2.3 0.38× 10−2

Table 5Real-time PCR fold changes andP-values corresponding to lowover high�L/�G states normalized to endogenous genes (� tubu-lin and ubiquitin B)

Gene Fold change P-value

Phosphofructokinase −2.0 0.6× 10−2

Triosephosphate isomerase −7.3 1.3× 10−2

Phosphoglycerate mutase −2.4 2.6× 10−2

Enolase 3 −2.3 1.5× 10−2

Pyruvate kinase −2.9 0.2× 10−2

Lactate dehydrogenase 1 −2.4 8.8× 10−2

Table 6Proteins up-regulated in high�L/�G state in proteomic analysis

Sequence annotation Accessionnumber

Central metabolismGlucose-6-phosphate isomerase (M. musculus) P06745Triosephosphate isomerase (M. musculus) P17751Glyceraldehyde 3-phosphate dehydrogenase (M.

musculus)P16858

Phosphoglycerate mutase (R. norvegicus, M.musculus, H. sapiens)

112128/P18669

Phosphoglycerate mutase 2 (M. musculus) 12844989Phosphoglycerate mutase, muscle form (M.

musculus)O70250

Alpha enolase (M. musculus) P17182l-Lactate dehydrogenase A chain (M. musculus) P06151

MitochondrialATP synthase alpha chain, mitochondrial

precursor (M. musculus)Q03265

Short chain 3-hydroxyacyl-CoA dehydrogenase,mitochondrial precursor (M. musculus)

Q61425

Cytoskeletal proteinsActin, cytoplasmic 1 (H. sapiens) P02570/1

Protein/DNA processingHeterogeneous nuclear ribonucleoproteins

A2/B1 (M. musculus)O88569

ARP2/3 complex 34 kDa subunit (H. sapiens) O15144Platelet-activating factor acetylhydrolase IB beta

subunit (M. musculus)Q61506

Platelet-activating factor acetylhydroxylase 1Bgamma subunit (M. musculus)

Q61205

SET protein (M. musculus) 3953617Proteasome (prosome, macropain) 28 subunit, 3

(M. musculus)6755214

Miscellaneous6-Phosphogluconolactonase (M. musculus) 13384778Similar to dendritic cell protein (M. musculus) 13542790von Hippel-Lindau binding protein 1—putative

(M. musculus)12832152

GlutathioneS-transferase P 2 (M. musculus) P19157Homolog to electron transfer flavoprotein

beta-subunit (Beta-ETF)—putative (M.musculus)

12832367

Peroxiredoxin 1 (M. musculus) P35700GTP-binding nuclear protein RAN (H. sapiens) P28746

in energy metabolism and others with no apparentrelation to metabolic shift. A number of proteins,including triosephosphate isomerase (Fig. 1, labeled“P17751” in black), lactate dehydrogenase A chain(Fig. 1, labeled “P06151” in black), and phospho-glycerate mutase (Fig. 1, labeled 112128/P18669,

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 11

Table 7Proteins up-regulated in low�L/�G state in proteomic analysis

Sequence annotation Accessionnumber

Cytoskeletal proteinsActin, cytoplasmic 1, beta actin (H. sapiens)

fragmentP02570

Actin, cytoplasmic 2, gamma actin (H. sapiens)fragment

P02571

Nucleotide processingNucleotide binding protein 2 (M. musculus) Q92061

Protein processingProteasome activator complex subunit 1 (M.

musculus)P97371

Heat shock protein, 70 kDa 4 (M. musculus) 12805195

Signal transductionGuanine nucleotide binding protein (G protein),

beta polypeptide 2-like 1 (H. sapiens)5174447

Dual specificity mitogen-activated proteinkinase 1 (M. musculus)

P31938

MiscellaneousNuclear protein Hcc-1 (M. musculus) Q9D1J3Immunoglobulin-binding protein 1 (M. musculus) Q61249T-complex protein 1, epsilon subunit (M.

musculus)P80316

Voltage-dependent anion-selective channelprotein 2 (M. musculus)

Q60930

Homolog to inorganic pyrophosphatase (M.musculus)

12842843

High mobility group protein 1 (M. musculus) P07155Platelet-activating factor acetylhydrolase IB beta

subunit (M. musculus)Q61206

Platelet-activating factor acetylhydrolase IBgamma subunit (M. musculus)

Q61205

Peroxiredoxin 1 (M. musculus) P35700

12844989, and O70250 in black) were identified inmore than one spot. Those multiple spots that wereannotated as the same protein have similar molecularmasses but different isoelectric points. The differentspots representing the same protein were similarlydifferentially expressed. There were also four spotsthat yielded more than one protein. In two cases, itwas due to insufficient separation of the differentproteins (Fig. 1, labeled “Q61249/P02570/P02571”in gray and “P06151/P28746” in black). In the othercases, more than one protein was identified becausethey are closely related proteins, such as beta-actinand gamma-actin (labeled “P02570/P02571” in black)and phosphoglycerate mutase from rat and human

(labeled “112128/P18669” in black). It is interestingto note that both beta-actin and gamma-actin show upas up-regulated in both high�L/�G and low�L/�Gstates. The spot up-regulated at the low�L/�G stateand labeled “Q61249/P02570/P02571” in gray has amolecular mass and an isoelectric point that closelycorrespond to the respective theoretical values of bothforms of actin, indicating that this spot contains thefull-length beta-actin and/or gamma-actin. The otherspot (labeled “P02570/P02571” in black) has a molec-ular mass and an isoelectric point that are lower thanthe respective theoretical values, suggesting that theactins in the spot are probably truncated fragments. Apoint to note is that a similarly identified spot with thesame molecular mass and isoelectric point was alsodetected in the previous report (Seow et al., 2001) asdown-regulated in the low�L/�G state. This typeof actin fragments have been observed previously inmouse liver protein 2-DE gels (Sanchez et al., 2001).

3.5. Comparison of differentially expressedgenes identified

The genes identified as differentially expressed bycDNA array, GeneChip and proteomic analyses werecompared. The analysis of cDNA array and GeneChipidentify only those that have corresponding probe se-quence on the array. Since only a subset of the probesequences are present on both arrays, some sequencesidentified by each method as differentially expressedmay be present in only one while others on both. Weexamined whether those sequences, which are presentin both cDNA array and GeneChip, gave consistentresults from both methods. While the level of differ-ential expression for a particular gene may be givendifferent values by cDNA analysis and by GeneChip,the results obtained by both methods are generallyconsistent. None was called as up-regulated by oneand down-regulated by the other. However, 32 se-quences called as differentially expressed at low lev-els in GeneChip assay were classified as unchangedwith cDNA BMAP array assay. Those sequences weresubsequently considered variable and not presented infurther analysis.

The proteomic analysis yielded a larger numberof enzymes involved in central metabolism that weredifferentially expressed than the two mRNA basedmethods. Those identified as differentially expressed

12 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

Fig. 1. 2-DE gel of low�L/�G state with the differentially expressed protein spots labeled. Proteins down-regulated with metabolic shiftfrom high �L/�G state to low�L/�G state are labeled in black, while the proteins up-regulated with metabolic shift from high�L/�Gstate to low�L/�G state are labeled in gray.

at both mRNA and protein levels show a consistenttrend.

3.6. Genes whose expression are not calleddifferentially expressed

Among all the genes present in both cDNA arrayand GeneChip analyses, only a small fraction aredifferentially expressed at their mRNA level uponmetabolic shift. Both the BMAP cDNA array and

GeneChip array used included a few thousand an-notated sequences. Many biochemical pathways ofinterest do not have the gene sequence of all reactionsinvolved present as a probe sequence. Thus even ifany of the genes involved in a particular pathway havenot been identified as differentially expressed, it doesnot necessarily imply that the pathway as a wholeor in part has not altered its expression level uponmetabolic shift. It is possible that relevant genes arenot present as probes, or the results are too variable

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 13

Table 8Expression profile of glycosylation enzymes upon metabolic shift

Sequence annotation cDNAmicroarray

AffymetrixGeneChip

EST, highly similar toalpha-mannosidase (R.norvegicus)

−1.2 NA

EST, highly similar tooligosaccharyl transferase stt3subunit homolog (C. elegans)

NA −1.3

Acetylglucosaminyltransferase-likeprotein

1.2 A

Glucosaminyltransferase,I-branching enzyme

−1.1 A

EST, weakly similar toacetylglucosaminyltransferase-likeprotein

1.0 NA

EST, weakly similar toalpha-1,3-mannosyl-glycoproteinbeta-1,2-n-acetylglucos-aminyltransferase

NA −1.2

EST, weakly similar to polypeptideGalNAc transferase-T1

NA 1.0

Alpha(1,3)fucosyltransferaseFuc-TVII gene

1.0 A

UDP-Gal:betaGlcNAc beta1,3-galactosyltransferase-II

1.0 A

EST, weakly similar to beta1,4-galactosyltransferase II

NA −1.1

Alpha 2,8-sialyltransferase 1.2 AAlpha 2,3-sialyltransferase −1.3 AAlpha 2,6-sialyltransferase 1.0 −1.3EST, highly similar to

alpha-l-fucosidase precursor(R. norvegicus)

1.1 1.2

A: absence call. NA: sequence not available as a probe on thearray.

to be called as either differentially expressed or notdifferentially expressed. We examined the expres-sion profiles of all genes present in these arraysthat encode enzymes involved in the glycosylationof proteins. The results are shown inTable 8. Thetable lists genes and ESTs annotated as both highlyand weakly similar to enzymes involved in glyco-sylation. Some are present in both arrays as probe,others are absent in one of the two (denoted asNA, not available on the array). None of the gly-cosylation genes present on cDNA array showed asignificant change in mRNA level (i.e. having an ex-pression ratio higher than 1.4 andP-value lower than0.1). Similarly, those on GeneChip array either did

not change significantly in expression level or wereabsent.

4. Discussion

4.1. Metabolic shift and gene expression

Rapidly growing cells in their exponential phasecan be cultivated in both a high�L/�G state and ametabolically shifted state (low�L/�G) (Zhou et al.,1995). The cell and product concentrations attainedwith the metabolic shift are substantially higher thanat the high�L/�G state due to a drastic reduction ofthe inhibitory metabolite product concentrations. Thischange in metabolism is unlikely to incur alterationsat the genetic level as the phenomenon is completelyreversible. In this study we investigated the changesin mRNA and protein expression profiles associatedwith metabolic shift. To minimize the number of vari-ables involved we employed cells from continuousculture at steady-state at the same dilution rate. Theviability of both cultures was over 95%, thus the cellswere also growing at the same specific growth rate.These experiments have been discussed in great detailpreviously (Europa et al., 2000). The two metabolicstates investigated have very distinctive�L/�G ratiosof 0.08 and 1.4.

Previously, we have employed suppression sub-tractive hybridization (SSH) to isolate genes whichare differentially expressed, albeit at slightly dif-ferent metabolic states of�L/�G of 0.5 and 1.6(Korke, 2002). SSH is a powerful technique of iso-lating rare genes (genes which are expressed at lowlevels) that are differentially expressed. However, thepresence of large number of endogenous retroviralsequences (Intracisternal A particles) rendered theisolation of only a small number of true positive se-quences. As the mouse DNA arrays became available,we adopted microarray as a complementary methodto SSH.

We have also previously used 2-DE to identifyproteins which are differentially expressed uponmetabolic shift from a high to low�L/�G state(Seow et al., 2001). In this investigation the pH rangewas extended from 4–7 to 3–10. Furthermore, thecellular protein and mRNA samples used were iso-lated from the same continuous cultures and the same

14 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

steady-states. Since the alteration at mRNA level ofa particular gene is not necessarily reflected at theprotein level, using mRNA and protein preparationsfrom the same cell samples allow for a direct compar-ison between the transcriptional and the translationalprofiles related to metabolic shift. By using the cuploading method of sample introduction for the first di-mensional IEF, better resolution was obtained on the2-DE gels. The extension of the pH range to the alka-line region allowed many more spots to be distinctlyseparated.

4.2. Assessment of differential gene expression

The results of the two microarray (cDNA andGeneChip) and proteomic analyses corroborateeach other. To further confirm the results, real-timePCR was used to determine the degree of differen-tial expression of six genes: phosphofructokinase,triosephosphate isomerase, phosphoglycerate mutase,enolase, pyruvate kinase, and lactate dehydroge-nase. Among these genes, triosephosphate isomeraseand enolase were identified by all three methodsas differentially expressed. The two sequences notidentified by cDNA microarray were phosphofruc-tokinase which had a fold change (−1.3) lower thanthe cut off value of−1.4, and phosphoglycerate mu-tase (fold change= −1.7) which had a highP-value(>0.1). Affymetrix GeneChip analysis identified fiveof these six as differentially expressed, but lactatedehydrogenase (fold change= −1.6) had a highP-value. The extent of differential expression is rel-atively small, ranging from 1.6- to 2.8-fold changein cDNA microarray and 1.6- to 3.3-fold change inGeneChip array. Nevertheless, the differential ex-pression at mRNA level for all of those genes wasvalidated by real-time PCR, which in general showeda higher degree of differential expression, espe-cially for triosephosphate isomerase. Furthermore,the differential expression of four out of the six en-zymes was confirmed at protein level by proteomicanalysis.

4.3. Central metabolism genes

Taken together, the results of microarray, real-timePCR, and proteomic analysis suggest that the decreaseof metabolic flux in the energy metabolism pathway

upon metabolic shift is accompanied by a decreasedgene expression at both mRNA and protein levelsin the glycolysis pathway. The change at the fluxlevel, 3–5-fold reduction in glycolytic fluxes (Europaet al., 2000; Gambhir et al., 2003), is larger than the1.4–3.0-fold changes seen at the mRNA level. Di-rect quantitative comparison at protein level based on2-D gel image is difficult. Nevertheless, the resultsare consistent among metabolic fluxes, mRNA andprotein levels, all down regulated at the low�L/�Gstate.

It is interesting to note that except for phospho-fructokinase and pyruvate kinase, all the other gly-colytic enzymes identified as differentially expressedcatalyse reversible reactions. It thus appears that uponmetabolic shift not only the enzymes which play a keyrole in regulating the flux, such as phosphofructoki-nase, changes their expression level, but also othersnot directly involved in regulation. Taking a simplisticview, the reduction of glycolytic flux upon metabolicshift could be accomplished by altering the level ofphosphofructokinase alone without changes in the lev-els of the other enzymes. In that case, the intermediateinvolved in those reactions would settle to a new andlower level for the reduced flux.

However, a majority of those enzymes whose ex-pression level changed catalyze reactions whose reac-tants or products are also involved in other pathways.Sustaining a relatively constant flux in those pathwaysis necessary to sustain cell growth at the same specificgrowth rate. Triosephosphate isomerase catalyzesthe reversible reaction between dihydroxyacetonephosphate (DHAP) and glyceraldehyde-3-phosphate(GAP). Phosphoglycerate mutase catalyzes the re-versible reaction between glycerate-3-phosphate(G3P) and glycerate-2-phosphate (G2P). Enolasecatalyzes the reversible reaction between G2P andphosphoenolpyruvate (PEP). DHAP is involved inthe glycerolipid metabolism pathway where it formsglycerol-3-phosphate for glycerolipid synthesis. GAPis one of the intermediates in glycolysis that is acommon substrate for pentose phosphate pathwayvia the enzyme transketolase. G3P is a substrate forserine biosynthesis while PEP is used in shikimatesynthesis. By reducing the expression level of TPI,PGM and enolase, the intracellular concentration ofthe intermediates involved in the reactions catalysedby those enzymes are probably maintained at a higher

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 15

level than if those enzymes were maintained at con-stant levels. The higher level of those intermediatesallows the fluxes to those connected pathways to besustained without substantially increasing the enzymelevels in those pathways.

Pyruvate is the important link between glycoly-sis and tricarboxylic acid (TCA) cycle and it is alsothe substrate for lactate and alanine formation. Uponmetabolic shift lactate production is drastically re-duced. This reduction of lactate flux is therefore notmerely accomplished by a reduction of its substrate(pyruvate) concentration, but also the enzyme (LDH)level. Again, this suggests that upon metabolic shift,the pyruvate level is not reduced at the same proportionas the flux of glycolysis. This probably helps sustainthe flux of pyruvate into TCA cycle without elevatingpyruvate dehydrogenase, the entry point to TCA cycle.

4.4. Other differentially expressed genes

In addition to the genes involved in energy meta-bolism, others involved in transcription, translation,protein-processing and transport were also identifiedas differentially expressed. Those genes do not seem tobe directly related to each other or to the physiologicalphenomenon of metabolic shift. Many of those wereannotated on the basis of similarity of their sequence toone in the UNIGENE clusters. The “identity” of thosesequences is yet uncertain. It is likely that as the anno-tation gradually improves in the next few years, newinsight based on their “identity” will be revealed. Anumber of genes are annotated as being tissue specific,e.g. neuronal, for which the basis for annotation maybe sequence similarity due to a common motif. Somesequences have no functional annotation (ESTs).

The change in the transcripts of those genes couldbe collateral changes, e.g. being influenced by its rel-ative position in the genome to a differentially ex-pressed gene. It is also possible that the expression ofthose genes is truly related to metabolic shift, and yetbecause of the unavailability of their identity a mech-anistic understanding of their involvement is still elu-sive.

4.5. Concluding remarks

Metabolic shift of mammalian cells in cultureentails a reduction of fluxes through glycolysis by

3–5-fold, and through TCA cycle by a smaller de-gree (Europa et al., 2000; Gambhir et al., 2003). Thechanges in the metabolic fluxes may entail changinglevels of the reaction substrates and intermediatesand/or alterations at the transcription or translationlevels of enzymes involved or both. Furthermore,the alteration may entail only gene products directlyinvolved in cellular metabolism, or it may includeothers ostensibly unrelated to cellular metabolism butwhich are affected or affect metabolic shift.

The genomic and proteomic techniques utilized inour study consistently indicate that the metabolic shiftdoes involve alterations at the gene expression level.However, this seemly large change in the cellularmetabolism is accompanied by a moderate change ingene expression in terms of the number of genes andthe extent of their differential expression. The resultsseem to indicate that the metabolic shift is a relativelysmall perturbation on the global scale of mammalianphysiology. Nevertheless, our findings suggest thatmetabolic shift evolves from both changes at en-zyme levels and modulation of reaction rates throughnutrient and metabolic intermediate concentrations.

Acknowledgements

This work was supported in part by a grant toWSH from the National Science Foundation, USA(BES-97272) and by the Academic Health Centerat the University of Minnesota. cDNA libraries andclones derived from brain regions of adult mouse(strain C57BL6) were obtained from Research Genet-ics, Inc. These clones and libraries were constructedby M. Bento Soares, Ph.D. at the University of Iowaunder contract NO1 MH80014, which was awardedby the National Institute of Mental Health (NIMH)and the National Institute of Neurological Disordersand Stroke (NINDS) as part of the Brain MolecularAnatomy Project (BMAP). BMAP is an NIH initiativeto support molecular carography of the mammaliannervous system through the localization of genesand the analysis of their expression patterns. Wethank Pam Skinner, the Biomedical Genomics Center(BMGC) and the UMN mouse group for microarraypreparation. The generous gifts from Merck ResearchLaboratories and Sankyo Co. are also gratefully ac-knowledged.

16 R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17

Appendix A.

Primer sequences for real-time PCR

Gene Primer sequence Sequence ID

PFK Forward TGCCGCTGTTCGCTCTAC Sequence NT039621Reverse AACCACCTTGACCAGTCCAG Transcript NM021514.1

TPI Forward GCACAGGAAGTACACGAGAAG Sequence NT039356Reverse TGCTCCAGTCACAGAACCTC Transcript NM009415.1

PGM Forward TCTCGGCGATCCTCAGTTG Sequence NT039692Reverse AAGCGGTTCTCCAGGTTCC Transcript NM023418.1

ENO3 Forward AGCACTGGAACTCCGAGATG Sequence NT039520Reverse CAGCAGAGCAGGACCTAGAG Transcript NM007933.1

PK Forward TGCCGCCTGGACATTGAC Sequence NT039474Reverse CAGCCGAGCCACATTCATTC Transcript NM011099.1

LDH1 Forward TGGCAGACTTGGCTGAGAG Sequence NT039420Reverse ACCTTCACAACATCCGAGATTC Transcript XM193039.2

TUBA2 Forward TCTATGAAGGAGGTGGACGAAC Sequence NT039205Reverse GCGGTGCTGTTGCCAATG Transcript NM146116.1

UBB Forward CCTGGTCCTTCGCCTGAG Sequence NT039520Reverse CGTTCTCGATGGTGTCACTG Transcript NM011664.1

References

Cruz, H.J., Moreira, J.L., Carrondo, M.J.T., 1999. Metabolic shiftsby nutrient manipulation in continuous cultures of bhk cells.Biotechnol. Bioeng. 66, 104–113.

Doyle, C., Butler, M., 1990. The effect of pH on the toxicity ofammonia to a murine hybridoma. J. Biotechnol. 15, 91–100.

Dudoit, S., Hwa Yang, Y., Callow, M., Speed, T., 2002. Statisticalmethods for identifying differentially expressed genes inreplicated cDNA microarray experiments. Statistica Sinica 12,111–139.

Europa, A.F., Gambhir, A., Fu, P., Hu, W.-S., 2000. Multiplesteady-states with distinct cellular metabolism in continuousculture of mammalian cells. Biotechnol. Bioeng. 67, 25–34.

Follstad, B.D., Balcarcel, R.R., Stephanopoulos, G., Wang, D.I.C.,1999. Metabolic flux analysis of hybridoma continuous culturesteady-state multiplicity. Biotechnol. Bioeng. 63, 675–683.

Gambhir, A., Korke, R., Lee, J., Fu, P.-C., Hu, W.-S., 2003.Analysis of cellular metabolism of hybridoma cells at distinctphysiological states. J. Biosci. Bioeng. 95, 317–327.

Glacken, M.W., Fleischaker, R.J., Sinskey, A.J., 1986. Reduction ofwaste product excretion via nutrient control: Possible strategiesfor maximizing product and cell yields on serum in culturesof mammalian cells. Biotechnol. Bioeng. 28, 1376–1389.

Hu, W.S., Dodge, T.C., Frame, K.K., Himes, V.B., 1987. Effectof glucose on the cultivation of mammalian cells. Dev. Biol.Stand. 66, 279–290.

Korke, R., Rink, A., Seow, T.K., Chung, M., Beattie, C.W.,Hu, W.-S., 2002. Genomic and proteomic perspectives in cellculture engineering. J. Biotechnol. 94, 73–92.

Korke, R., 2002. Analysis of gene expression patterns ofmammalian cells at different metabolic states. PhD Thesis,Department of Chemical Engineering and Materials Science,University of Minnesota, Minneapolis, MN, USA.

Kurano, N., Leist, C., Messi, F., Kurano, S., Fiechter, A., 1990.Growth behaviour of Chinese hamster ovary cells in a compactloop bioreactor. Part 2. Effects of medium components andwaste products. J. Biotechnol. 15, 113–128.

May, B.J., Zhang, Q., Li, L.L., Paustian, M.L., Whittam, T.S.,Kapur, V., 2001. Complete genomic sequence of pasteurellamultocida, pm70. Proc. Natl. Acad. Sci. U.S.A. 98, 3460–3465.

Omasa, T., Kobayashi, M., Shioya, S., Suga, K.I., 1992. Effects oflactate concentration on hybridoma culture in lactate-controlledfed-batch operation. Biotechnol. Bioeng. 39, 556–564.

Ozturk, S.S., Riley, M.R., Palsson, B.O., 1992. Effects of ammoniaand lactate on hybridoma growth, metabolism, and antibodyproduction. Biotechnol. Bioeng. 39, 418–431.

Pelle, R., Murphy, N.B., 1993. Northern hybridization: Rapidand simple electrophoretic conditions. Nucleic Acids Res. 21,2783–2784.

Sanchez, J.C., Chiappe, D., Converset, V., Hoogland, C., Binz,P.A., Paesano, S., Appel, R.D., Wang, S., Sennitt, M., Nolan,A., Cawthorne, M.A., Hochstrasser, D.F., 2001. The mouse

R. Korke et al. / Journal of Biotechnology 107 (2004) 1–17 17

swiss-2d page database: A tool for proteomics study of diabetesand obesity. Proteomics 1, 136–163.

Seow, T.K., Korke, R., Liang, R., Ong, S.-E., Ou, K., Wong,K., Hu, W.S., Chung, M., 2001. Proteomic investigation ofmetabolic shift in mammalian cell culture. Biotechnol. Prog.17, 1137–1144.

Siegwart, P., Cote, J., Male, K., Luong, J.H.T., Perrier, M., Kamen,A., 1999. Adaptive control at low glucose concentration ofhek-293 cell serum-free cultures. Biotechnol. Prog. 15, 608–616.

Talaat, A.M., Howard, S.T., Hale, W.T., Lyons, R., Garner, H.,Johnston, S.A., 2002. Genomic DNA standards for geneexpression profiling in mycobacterium tuberculosis. NucleicAcids Res. 30, e104.

Tseng, G.C., Oh, M.K., Rohlin, L., Liao, J.C., Wong, W.H., 2001.Issues in cDNA microarray analysis: quality filtering, channelnormalization, models of variations and assessment of geneeffects. Nucleic Acids Res. 29, 2549–2557.

Zeng, A.P., Deckwer, W.D., Hu, W.S., 1998a. Determinants andrate laws of growth and death of hybridoma cells in continuousculture. Biotechnol. Bioeng. 57, 642–654.

Zeng, A.P., Hu, W.S., Deckwer, W.D., 1998b. Variation ofstoichiometric ratios and their correlation for monitoring andcontrol of animal cell cultures. Biotechnol. Prog. 14, 434–441.

Zhou, W., Hu, W.-S., 1994. On-line characterization of a hybridomacell culture process. Biotechnol. Bioeng. 44, 170–177.

Zhou, W., Hu, W.-S., 1995. Effect of insulin on a serum-freehybridoma culture. Biotechnol. Bioeng. 47, 181–185.

Zhou, W.C., Rehm, J., Hu, W.S., 1995. High viable cellconcentration fed-batch cultures of hybridoma cells throughon-line nutrient feeding. Biotechnol. Bioeng. 46, 579–587.

Zhou, W.-C., Rehm, J., Europa, A., Hu, W.-S., 1997. Alterationof mammalian cell metabolism by dynamic nutrient feeding.Cytotechnology 24, 99–108.