Gla

DC

a

ARRAA

KCCSGS

1

t2gaht

EEigm

9

h0

Journal of Biotechnology 189 (2014) 58–69

Contents lists available at ScienceDirect

Journal of Biotechnology

j ourna l ho me pa ge: www.elsev ier .com/ locate / jb io tec

ene expression measurements normalized to cell number revealarge scale differences due to cell size changes, transcriptionalmplification and transcriptional repression in CHO cells

ina Fomina-Yadlin, Zhimei Du, Jeffrey T. McGrew ∗

ell Sciences & Technology, Amgen Inc., Seattle, WA 98119, United States

r t i c l e i n f o

rticle history:eceived 9 July 2014eceived in revised form 18 August 2014ccepted 25 August 2014vailable online 4 September 2014

eywords:hinese Hamster Ovaryellular RNA contentynthetic RNA spike-in controlsene expressionpecific productivity

a b s t r a c t

Conventional approaches to differential gene expression comparisons assume equal cellular RNA con-tent among experimental conditions. We demonstrate that this assumption should not be universallyapplied because total RNA yield from a set number of cells varies among experimental treatments of thesame Chinese Hamster Ovary (CHO) cell line and among different CHO cell lines expressing recombinantproteins. Conventional normalization strategies mask these differences in cellular RNA content and, con-sequently, skew biological interpretation of differential expression results. On the contrary, normalizationto synthetic spike-in RNA standards added proportional to cell numbers reveals these differences andallows detection of global transcriptional amplification/repression. We apply this normalization methodto assess differential gene expression in cell lines of different sizes, as well as cells treated with a cellcycle inhibitor (CCI), an mTOR inhibitor (mTORI), or subjected to high osmolarity conditions. CCI treat-ment of CHO cells results in a cellular volume increase and global transcriptional amplification, whilemTORI treatment causes global transcriptional repression without affecting cellular volume. Similarly to

CCI treatment, high osmolarity increases cell size, total RNA content and antibody expression. Further-more, we show the importance of spike-in normalization for studies involving multiple CHO cell lines andadvocate normalization to spike-in controls prior to correlating gene expression to specific productivity(qP). Overall, our data support the need for cell number specific spike-in controls for all gene expressionstudies where cellular RNA content differs among experimental conditions.

© 2014 Elsevier B.V. All rights reserved.

. Introduction

A fundamental assumption of the entire field of compara-ive transcriptomics has recently been challenged (Loven et al.,012). Conventional approaches to both global and individualene expression measurements assume equal cellular RNA content

mong experimental conditions being compared. This assumptionas recently been questioned by the discovery of “global transcrip-ional amplification” of all actively-transcribed genes that results

Abbreviations: CCI, cell cycle inhibitor; CHO, Chinese Hamster Ovary; ERCC,xternal RNA Controls Consortium; HC, heavy chain; IRES, Internal Ribosomentry Site; LC, light chain; mTOR, mammalian target of rapamycin; mTORI, mTORnhibitor; qP , specific productivity; qPCR, quantitative real-time PCR; RNA-Seq, next-eneration RNA sequencing; RPKM, Reads Per Kilobase of exon model per Millionapped reads; VCD, viable cell density.∗ Corresponding author at: 1201 Amgen Court West, MS AW2/D2042, Seattle, WA8119, United States. Tel.: +1 206 265 7871.

E-mail address: mcgrewjt@amgen.com (J.T. McGrew).

ttp://dx.doi.org/10.1016/j.jbiotec.2014.08.037168-1656/© 2014 Elsevier B.V. All rights reserved.

from c-Myc overexpression in B cells (Lin et al., 2012; Nie et al.,2012). Induction of c-Myc increased cell size and both the mRNAand the rRNA cellular content resulting from augmentation ofthe entire cellular gene expression program, but these phenotypicobservations were masked by traditional global gene expressionanalysis (Lin et al., 2012; Nie et al., 2012). In order to allow detectionof global transcriptional amplification by transcriptomic analysis,Loven et al. (2012) applied the concept of adding spike-in controlsproportional to cell number. The use of synthetic spike-in RNA stan-dards for global gene expression analysis using microarrays waspreviously recommended to allow data comparison across differ-ent platforms and protocols used for data collection and analysis(Yang, 2006). Loven et al. (2012) used synthetic RNA spike-in stan-dards (ERCCs) for normalization to cell number, thus, revealingdifferences in cellular RNA content and enabling cross-platform

detection of transcriptional amplification by c-Myc. In addition,spike-in normalization to cell number was equally successful atdetecting “global transcriptional repression” in Rett syndromemodels driven by MECP2 loss-of-function (Li et al., 2013). Mutant

l of Bi

nvtstMys

dtmCSRoedD2ls22sRtnsoeGpnk(sKp2tmofop

2

2

eEmilgsEtIwCf

D. Fomina-Yadlin et al. / Journa

eurons displayed reduced size and cellular RNA content, but con-entional global gene expression analysis revealed no change inhe majority of genes (Li et al., 2013). In contrast, addition ofpike-in standards proportional to cell numbers revealed globalranscriptional and translational repression as a consequence of

ECP2 deletion (Li et al., 2013), advocating application of this anal-sis technique to all comparisons involving differences in cellularize.

The onset of the systems biology era in biotechnology has beenefined by a wide variety of ‘omics-based approaches to charac-erize the biological basis of desired phenotypic parameters and

anipulate them to enhance heterologous protein expression inhinese Hamster Ovary (CHO) cell lines (Kildegaard et al., 2013).o far, only conventional techniques that assume invariant totalNA yield per cell have been used to evaluate gene expressionf CHO cells. For example, many studies have been performed tolucidate the biological basis of specific productivity (qP) and pre-ict qP from gene expression profiles (Clarke et al., 2011a,b, 2012;oolan et al., 2012; Kang et al., 2014; Nissom et al., 2006; Yee et al.,009), but none took into account potential differences in cellu-

ar size or total RNA content. Other studies examining effects ofmall molecule treatments (e.g. sodium butyrate, Kantardjieff et al.,010) or bioprocess conditions (e.g. temperature, Kantardjieff et al.,010 and culture osmolarity, Shen et al., 2010) on gene expres-ion in CHO cells also did not take into account changes in cellularNA content. For this study, we selected two small molecules, oneargeting cell cycle progression and the other targeting mTOR sig-aling, which were hypothesized to have opposite effects on cellize in our biological system. Cell cycle arrest, achieved by eitherverexpression of an endogenous cell cycle inhibitor (p21CIP1) (Bit al., 2004) or addition of small molecule inhibitors that cause1 or G1/S arrest (Du et al., 2014; Fingar et al., 2002), has beenreviously shown to increase mammalian cell size. mTOR sig-aling through its downstream effectors, ribosomal protein S6inase (S6K1) and eukaryotic initiation factor 4E-binding protein4EBP1), has also been demonstrated to regulate mammalian cellize (Fingar et al., 2002). In fact, mTOR overexpression in CHO-1 cells stimulated cell cycle progression by promoting G1-to-Shase transition and increased cell size (Dreesen and Fussenegger,011). We showed that the assumption of constant cellular size andotal RNA yield was inaccurate for several small molecule treat-

ents, multi-cell line comparison studies and changes in culturesmolarity. Therefore, we assessed the use of spike-in standardsor evaluation of differential gene expression in CHO cells, focusingn case studies relevant to cell line development and productionrocesses.

. Materials and methods

.1. Cell culture and experimental treatments

Six CHO-derived cell lines (Rasmussen et al., 1998), eachxpressing a different monoclonal antibody (cell lines A, B, C, D,

and F), were cultured in a proprietary chemically-defined growthedia in vented shake-flasks at 36 ◦C, 5% CO2, 70% relative humid-

ty and shaken at 150–160 rpm in Kuhner incubators. Each celline was generated using a proprietary expression system. A sin-le mRNA encoded both the antibody light chain (LC) and the LCelectable marker as they were linked by an Internal Ribosomentry Site (IRES) sequence. Similarly, single mRNA encoded bothhe HC and the HC selectable marker as they were also linked by an

RES sequence. Viable cell density (VCD), viability and cell diameter

ere measured with a ViCell automated cell counter (Beckman-oulter, Inc., Brea, CA). Cellular volume was calculated using the

ormula for the volume of a sphere: V = 4/3�r3.

otechnology 189 (2014) 58–69 59

For the CCI-mTORI treatment study, cells were seeded fromday 4 growth cultures at 10 × 106 viable cells/mL into proprietarychemically-defined production medium in vented 24 deep-wellplates (3 mL volume per well) and shaken at 220 rpm in Kuhnerincubators. Cultures were treated on day 0 with either CCI (Du et al.,2014) (Amgen proprietary, PCT/US2013/074366, 10 �M final con-centration), mTORI (Amgen proprietary, WO/2010/132598, 0.5 �Mfinal concentration) or vehicle control (DMSO, 0.1% final concen-tration) and daily medium exchanges were performed using freshmedia containing an appropriate small molecule (CCI, mTORI orDMSO). Percent DMSO was kept constant among all experimentalconditions. Spent medium was used for daily titer measure-ments and daily specific productivity (qP) calculations. Titermeasurements were performed by affinity High Performance Liq-uid Chromatography (HPLC) using POROS A/20 Protein A column.Daily qP (pg/cell/day) was calculated according to the simplified for-mula: qP = daily titer/daily VCD. On day 3, 3 × 106 viable cells werecollected per condition for gene expression analysis (biological trip-licates), snap-frozen and stored at −70 ◦C for further processing.

For the multi cell line study, ten-day production assays withbolus feeds on days 3, 6 and 8 were performed in chemically-defined production medium in vented shake-flasks as previouslydescribed (Fomina-Yadlin et al., 2014). Titer samples were collectedon days 3, 6, 8 and 10. For each interval between days [m, n], qP wascalculated according to the formula: qP = titern/

∫ n

mVCDdt/(tn −

tm), where titern is the measured cumulative titer at tn and time (t)is expressed in days. On day 6, 3 × 106 viable cells were collectedper condition for gene expression analysis (biological triplicates),snap-frozen and stored at −70 ◦C.

For osmolarity level study, cells were seeded from day 3 growthcultures by 1:5 split at ∼0.5 × 106 viable cells/mL into proprietarychemically-defined growth media in 24 deep-well plates (3 mL vol-ume per well). Osmolarity was adjusted at 0-h time-point with 5 MNaCl solution in growth medium, as previously described (Shenet al., 2010). Daily VCD, viability and cell diameter measurementswere performed. On day 2, 1 × 106 viable cells were collected percondition for gene expression analysis (biological triplicates), snap-frozen and stored at −70 ◦C.

2.2. ERCC spike-in and RNA extraction

Addition of External RNA Controls Consortium (ERCC) controlswas done as previously described (Loven et al., 2012). Specifically,1 �L of 1:10 diluted ERCC RNA Spike-In Mix 1 (Ambion®, Life Tech-nologies, Grand Island, NY) was added per 1 × 106 cells. ERCC wasadded to frozen cell pellets with RLT lysis buffer, and total RNA wasisolated with the RNeasy Mini kit (Qiagen, Valencia, CA) accord-ing to the manufacturer’s protocol, including optional on-columnDNAse I digestion, and using 100 �L elution volume. RNA concen-tration was measured on the Nanodrop 2000 (Thermo Scientific,Wilmington, DE), and RNA quality was assessed using the 2100Bioanalyzer (Agilent, Santa Clara, CA) with the RNA 6000 Nano Kit(Agilent, Santa Clara, CA) to ensure all samples used for RNA-Seqanalysis had RNA Integrity Number (RIN) >9.

2.3. RNA-Seq sample processing and analysis

RNA library preparations, sequencing reactions, and initialbioinformatics analysis were conducted at GENEWIZ, Inc. (SouthPlainfield, NJ). Illumina TruSeq RNA library preparation, clustering,and sequencing reagents were used throughout the process follow-ing the manufacturer’s recommendations (Illumina, San Diego, CA).

Briefly, 1 �g of total RNA was used as starting material for librarypreparation with the Illumina Truseq RNA preparation Kit V2. Poly-T oligo-attached magnetic beads were used to purify mRNA, whichwas then fragmented for 8 min. at 94 ◦C. First strand and second

6 l of Bi

sTAePAbBltmsaSamao

tgba(pRfwmfCR

2

rtLPrI3GPisnt

i2

3

cidemgmhB

0 D. Fomina-Yadlin et al. / Journa

trand DNA were subsequently synthesized. Invitrogen Reverseranscriptase II was used with First Strand synthesis master mix.dapters were ligated after adenylation of the 3′ ends followed bynrichment and barcode addition for multiplexing by limited cycleCR. DNA libraries were validated using a DNA 1000 Chip on thegilent 2100 Bioanalyzer (Agilent, Santa Clara, CA), and quantifiedy using the Qubit 2.0 Fluorometer and by real time PCR (Appliediosystems, Carlsbad, CA, USA). The samples were clustered on six

anes of a flow cell, using the cBOT from Illumina. After clustering,he samples were loaded on the Illumina HiSeq 2500 instru-

ent according to manufacturer’s instructions. The samples wereequenced using a 1 × 50 Single Read (SR) configuration. Imagenalysis and base calling were conducted by the HiSeq Controloftware (HCS) on the HiSeq 2500 instrument. Sequence data wasligned to the Cricetulus griseus reference genome along with itsitochondrion, downloaded from NCBI (cgr ref CriGri 1.0 chrUn,

nd NC 007936.1). 92 ERCC spike-in sequences were added as partf the reference (Ambion®, Life Technologies, Grand Island, NY).

CLC Genomics Workbench 6.5.1 was used to align the reads tohe reference genome, allowing maximum of 2 mismatches. Theene hit count was performed also within the CLC Genomics Work-ench 6.5.1. The expression value was measured in RPKM, defineds Reads Per Kilobase of exon model per Million mapped readsMortazavi et al., 2008). Spike-in normalization of RPKM values waserformed as previously described (Loven et al., 2012). Specifically,PKM values were normalized to ERCC-spike-in standards with the

unction normalize.loess within the affy R package. Loess regressionas performed on the ERCC subset and used to re-normalize theatrix of all RPKM values. Differential expression analysis was per-

ormed with 2-way ANOVA in Array Studio (OmicSoft Corporation,ary, NC) on either the raw RPKM values or the ERCC-normalizedPKM values.

.4. Quantitative real-time PCR

For quantitative real-time RT-PCR (qPCR), 1 �g of total RNA waseverse-transcribed into cDNA using the SuperScript® VILOTM Mas-er Mix, according to the manufacturer’s protocol (InvitrogenTM,ife Technologies, Grand Island, NY). qPCR was performed with theower SYBR® Green PCR Master Mix following the manufacturer’secommendations (Applied Biosystems®, Life Technologies, Grandsland, NY) on the 7900HT Fast Real-Time PCR System with the84-Well Block Module (Applied Biosystems®, Life Technologies,rand Island, NY). Primer pairs for ERCC1-4 were designed withrimer3Plus and listed in Supplementary Table S1. For housekeep-ng CHO genes (ActB, B2M and TBP), previously published primerequences were utilized (Fomina-Yadlin et al., 2014). Average cycleumber (Ct) for either ERCC1-4 or the housekeeping genes was usedo normalize mAb expression.

Supplementary material related to this article can be found,n the online version, at http://dx.doi.org/10.1016/j.jbiotec.014.08.037.

. Results

Low-producing cell line (cell line A) cultures and high-producingell line (cell line B) cultures were treated with either a cell cyclenhibitor (CCI), an mTOR inhibitor (mTORI) or vehicle control onay 0 and maintained for 3 days with complete daily mediumxchanges. Daily spent medium was used for daily titer measure-ents and specific productivity (qP) calculations. CCI suppressed

rowth of both cell lines without affecting viability (Fig. 1A and B).TORI was less effective than the CCI at suppressing growth of the

igh-producing cell line B without affecting viability (Fig. 1A and). In addition to growth suppression, mTORI treatment decreased

otechnology 189 (2014) 58–69

viability of the low-producing cell line A (Fig. 1A and B). CCIincreased cell diameter, cell volume (calculated as a volume of asphere) and qP in both cell lines (Fig. 1C, D and E, respectively).However, most of the qP increase correlated with the increase incell volume (Fig. 1F). mTORI was hypothesized to decrease cellsize, but no experimental decreases in cell size were observed overthe course of the 3-day treatment for the two cell lines examined(Fig. 1C and D). Furthermore, mTORI treatment did not change qP

of either cell line (Fig. 1E).An underlying assumption of all conventional gene expression

measurements is no change in total RNA yield per cell. Since 3-dayCCI treatment increased cellular volume by 73% and 74% for celllines A and B, respectively, (Fig. 1D) we sought to evaluate the “nochange in RNA yield” assumption for this study. Indeed, CCI treat-ment increased the total RNA content per cell for both cell lines,while the RNA content per unit volume did not change significantly,indicating no change in “total RNA density” (Fig. 2A). Surprisingly,3-day mTORI treatment of both cell lines, which did not affect cellvolume, decreased the total RNA content per cell (Fig. 2A). There-fore mTORI decreased the “total RNA density,” suggesting globaltranscriptional repression.

Since the total RNA yield per cell changed with each experi-mental treatment, we used normalization to ERCC spike-in controlsto study global gene expression changes, as previously described(Loven et al., 2012). Three million cells were collected per con-dition on day 3, and ERCC controls were added to cell pelletsprior to total RNA extraction proportionally to the cell number col-lected per sample. Next generation sequencing (RNA-Seq) analysiswas performed on all day 3 samples, with the same RNA amountsequenced per sample. Statistical comparisons among experimen-tal and control conditions performed using 2-way ANOVA onun-normalized RPKM values revealed symmetrical volcano plotswith similar numbers of significantly up- and down-regulatedgenes in compound-treated cell lines A and B (Fig. 2B and C(top panels), respectively). However, performing 2-way ANOVA onERCC-normalized RPKM values resulted in skewed volcano plots,with mostly up-regulated genes for the CCI and down-regulatedgenes for the mTORI treatment of both cell lines A and B (Fig. 2Band C (bottom panels), respectively).

Using un-normalized RPKM values for examination of differ-ential expression yielded nearly symmetrical plots of averagetranscript abundance versus log 2-fold change between treatmentand control conditions (Fig. 3). In contrast, normalization to spike-in controls proportional to cell number revealed global up- anddown-regulation of gene expression corresponding to differencesin total RNA yield per cell following either the CCI or the mTORItreatments (Fig. 3A and B, respectively). Under equal total RNAassumption characteristic of conventional analysis strategies, 18.5%and 23.7% of total detected transcripts were significantly up- anddown-regulated by the CCI treatment of low-producing cell lineA, respectively (Fig. 3A (top panel)). In contrast, under equal cellnumber assumption facilitated by normalization to spike-in con-trols, 63.0% and 0.6% of total detected transcripts were significantlyup- and down-regulated by the CCI treatment, respectively (Fig. 3A(top panel)). These observations indicate that the CCI treatmentresults in global transcriptional amplification that is correlatedto the cell volume increase. mTORI treatment had the oppositeeffect on global gene expression. Under equal total RNA assump-tion, 27.9% and 30.7% of total detected transcripts were significantlyup- and down-regulated by the mTORI treatment of cell line A,respectively (Fig. 3B (bottom panel)). In contrast, under equal cellnumber assumption facilitated by normalization to spike-in con-

trols, 2.6% and 55.2% of total detected transcripts were significantlyup- and down-regulated by the mTORI treatment, respectively(Fig. 3B (bottom panel)), suggesting global transcriptional repres-sion. Analogous effects of normalization strategy on determination

D. Fomina-Yadlin et al. / Journal of Biotechnology 189 (2014) 58–69 61

Fig. 1. Cellular responses of two antibody-producing CHO cell lines to treatment with either the cell cycle inhibitor (CCI) or the mTOR inhibitor (MTORI). (A) Viable celldensity (VCD), (B) culture viability, (C) cell diameter, (D) cell volume, (E) specific productivity (qP ) and (F) specific productivity adjusted by volume factor are shown for the3-day time-course of compound treatment. Volume factor represents fold change with respect to the average cellular volume of day 0 controls for each cell line. Cell line Ai SD o

ohs

i2

gnp

s a low-producing cell line and cell line B is high-producing. Data represent mean ±

f differential expression between conditions were observed inigh-producing cell line B (Fig. S1), suggesting robustness of thepike-in normalization approach.

Supplementary material related to this article can be found,n the online version, at http://dx.doi.org/10.1016/j.jbiotec.014.08.037.

The ERCC normalization did not significantly alter the list ofenes identified as most highly regulated. In fact, the list of sig-ificantly up-regulated genes by the CCI treatment obtained byerforming ANOVA on un-normalized RPKM values was a subset

f three biological replicates.

of a larger list obtained by performing ERCC normalization prior todifferential expression analysis in both cell lines (Fig. S2). Similarly,the list of significantly down-regulated genes by mTORI treatmentgenerated by ANOVA on un-normalized RPKM values was a subsetof a larger list obtained by performing ANOVA on ERCC-normalizedRPKM values in both cell lines (Fig. S2). However, ERCC normal-

ization shifted the fold change up for those samples treated withthe CCI and shifted it down for those treated with the mTORI. Thedetailed analysis of the impact of the CCI and the mTORI on specificgenes will be the subject of a separate manuscript.

62 D. Fomina-Yadlin et al. / Journal of Biotechnology 189 (2014) 58–69

Fig. 2. Global gene-expression comparison of control samples to samples treated with either the cell cycle inhibitor (CCI) or the mTOR inhibitor (MTORI). (A) Total RNAextracted from 3 million cells per condition on day 3 of the treatment time-course with cell lines A and B. Three panels represent comparisons of total RNA, cell volume andthe ratio of total RNA to cell volume by cell line and treatment. Data represent mean ± SD of three biological replicates plotted as fold of control condition. **P < 0.01 and***P < 0.001 represent statistically significant differences between untreated controls and either CCI or MTORI treated conditions. Volcano plots depicting results of 2-wayA cell liE e p-vaa

i2

tcAad“ommmmcot

NOVA analyses comparing treatment and control conditions in cell line A (B) andach panel is constructed by plotting the log 2-fold change against the – log 10 of thnd −log 10(p-value) = 2.

Supplementary material related to this article can be found,n the online version, at http://dx.doi.org/10.1016/j.jbiotec.014.08.037.

In addition to the application of the ERCC spike-in normaliza-ion to measuring global gene expression, ERCC spike-in controlsould also be used for analysis of individual transcripts by qPCR.pplication of ERCC normalization to measurement of monoclonalntibody (mAb) expression in cell lines A (Fig. 4A and B) by qPCRemonstrated advantages over conventional normalization tohousekeeping” genes. In order to assess mAb expression, setsf primers were designed to measure the levels of the selectablearkers associated with the LC and the HC expression. In addition,olecule-specific LC and HC primers were used to further evaluateAb expression in the low- and the high-producing cell lines.

Ab expression in each cell line was assessed utilizing either the

onventional qPCR normalization via comparison to the averagef the housekeeping gene expression (Fig. 4A and B, top panel) orhe normalization to the expression average of the spiked-in ERCC

ne B (C) using either un-normalized or ERCC-normalized RPKM values are shown.lue. Red lines designate fold-change and significance cutoffs: |log 2-fold change| = 1

controls (Fig. 4A and B, bottom panel). Sets of primers to detect thetop 4 most abundant ERCC spike-in standards were designed toenable spike-in normalization strategy for qPCR. Normalization tospike-in controls, but not to housekeeping genes, revealed statisti-cally significant differences in mAb expression between untreatedcontrols and either the CCI or the mTORI treated conditions in bothcell line A and cell line B (Fig. 4A and B, respectively).

The CCI-mTORI study established the utility of spike-in nor-malization for the experimental treatments that change cell sizeand/or cause global transcriptional amplification or repression. Wesought to examine other types of experiments that would benefitfrom ERCC normalization for gene expression comparisons. Multi-cell line studies represent a common type of omic experimentsfrequently performed in the biopharmaceutical industry setting

to examine the biological basis of desired phenotypic parameters(Kang et al., 2014). Six CHO-derived mAb-expressing cell lines (A–F)with different productivity levels were run in a 10-day fed-batchproduction process with bolus feeds on days 3, 6 and 8 (Fig. 5), as

D. Fomina-Yadlin et al. / Journal of Biotechnology 189 (2014) 58–69 63

Fig. 3. Effects of ERCC normalization on detection of significant changes in gene expression. Significantly up-regulated (red) and down-regulated (blue) transcripts detectedby RNA-seq following 3-day treatment of cell line A with either (A) cell cycle inhibitor (CCI) or (B) mTOR inhibitor (MTORI) for either un-normalized (upper panel) onERCC-normalized (lower panel) RPKM data. For each panel, average expression value (intensity) is plotted against log 2-fold change, and statistical significance of differentiale of theo

pesTsaIro(bIdwctcScpu

xpression is determined by the p-value cutoff (p-value ≤ 0.01). (For interpretationf this article.)

reviously described (Fomina-Yadlin et al., 2014). The six cell linesxhibited different growth profiles (Fig. 5A), but all maintained rea-onably high viability throughout the 10-day fed-batch (Fig. 5B).he two largest cell lines, cell lines E and F, achieved lower cell den-ities in the fed-batch production cultures, a characteristic we havelso observed for other large cell lines (unpublished observations).n addition to growth differences, the cell lines covered a wideange of qP values during the production process, and qP s increasedver the course of the fed-batch for all but one cell line (cell line F)Fig. 5C). These six cell lines were originally selected for this studyecause of their diverse cell sizes during growth and propagation.n production, cellular volume fluctuated slightly among differentaily measurements, but remained fairly consistent for 5 cell lines,hereas the volume of cell line F increased dramatically over the

ourse of the fed-batch production process (Fig. 5D). Cell lines inhis study can be ranked by productivity (Fig. 5C), but the rankinghanges if cellular volume is factored into the calculation (Fig. 5E).

pecific productivity is defined independent of cell size. However,ertain cell lines that have an apparent sought-after productivityrofile become less desirable when you factor in their cellular vol-me (cell line F), while others remain top-ranked (cell lines B and

references to color in this figure legend, the reader is referred to the web version

D). The advantages of the cell lines B and D can be further visualizedby plotting specific productivity against cell line volume (Fig. 5F).

Total RNA extracted from the same number of cells on day 6 ofthe fed-batch production process varied significantly across the sixcell lines in this small multi-cell line study (Fig. 6A), with larger celllines having a greater total RNA yield (Fig. 6B). However, normaliza-tion of total RNA to cell volume eliminated most of the significantdifferences (Fig. 6C). Conventional normalization of mAb expres-sion to housekeeping genes was able to distinguish the cell linesfrom the opposite ends of the productivity spectrum (cell lines Aand B) (Fig. 6D). However, only ERCC normalization revealed thatlarger cells (cell lines E and F) had more product expression, whichwas masked by conventional normalization to housekeeping genes(Fig. 6D and E). The ERCC-normalized mAb signal to cellular volumeidentified cell line B as the most productive in terms of cell size(Fig. 6E).

In order to further expand the application of the spike-in nor-

malization methodology, we sought to examine other cultureconditions relevant to industrial bioprocess that can affect cellsize. Specifically, a sudden increase in culture osmolarity has beenshown to induce rapid cellular shrinking followed by hyperosmotic

64 D. Fomina-Yadlin et al. / Journal of Biotechnology 189 (2014) 58–69

Fig. 4. Gene expression measurements by qPCR for cell line A (A) and cell line B (B) following 3-day treatment with either the cell cycle inhibitor (CCI) or the mTOR inhibitor(MTORI). Gene expression measurements are classified into three categories: ERCC spike-in controls (ERCC1, 2, 3, 4), mAb expression, and housekeeping genes (ActB, B2M,TBP). Two panels correspond to two normalization strategies: conventional qPCR normalization to the average of housekeeping gene expression (top panel) and normalizationto the expression average of spiked-in ERCC controls (bottom panel). Data represent mean ± SD of three biological replicates. *P < 0.05, **P < 0.01 and ***P < 0.001 represents nd eit

rtawcHvSttonHtutaaalconmS

tatistically significant differences in mAb expression between untreated controls a

egulatory volume increase (Schliess et al., 2007). In a third study,wo osmolarity levels were chosen to model normal (300 mOsm/L)nd high (450 mOsm/L) culture osmolarity. Three different cell linesere chosen from the panel of cell lines used for the multi-cell line

omparison to examine osmolarity effects: cell lines B, D and F.igh osmolarity arrested cell growth, while maintaining cultureiability over the course of 48-h treatment in all 3 cell lines (Fig.3A). Furthermore, high culture osmolarity caused cellular volumeo increase in all cell lines examined (Fig. S3A). Consistent withhe volume increases, total RNA extracted from the same numberf cells at the 48-h time-point differed significantly between theormal and the high osmolarity levels in each cell line (Fig. S3B).owever, observed differences in the total RNA yield resulted from

he cell size increases, as evident through normalization to cell vol-me (Fig. S3C). Normalization to spike-in controls correspondingo cell numbers revealed significant differences between normalnd high osmolarity conditions for each cell line (Fig. S3D). Furtherdjustment by volume established that observed volume increasesccounted for transcriptional amplification induced by high osmo-arity in cell lines B and F (Fig. S3D). However, mAb expression inell line D was exceptionally transcriptionally responsive to the

smolarity changes, because the cellular volume increase couldot account for the entire magnitude of the up-regulation of itsAb-specific gene expression in high osmolarity condition (Fig.

3D).

her CCI or MTORI treated conditions.

Supplementary material related to this article can be found,in the online version, at http://dx.doi.org/10.1016/j.jbiotec.2014.08.037.

4. Discussion

Experimental treatments can alter either transcriptome com-position, RNA amount per cell, or both (Fig. 7A). When cellularRNA amount does not change, conventional normalization is wellsuited for differential gene expression comparisons among exper-imental conditions. However, in all cases of changing cellular RNAamount, conventional normalization strategies mask those changesand can, therefore, only be used for comparisons of RNA compo-sition. In contrast, spike-in normalization to cell number allowscomparisons of RNA amount. The three studies described in thismanuscript unequivocally establish the utility of spike-in normal-ization for experimental comparisons that involve differences incell size and RNA yield per cell.

These three case studies challenge traditional normalizationstrategies used for RNA-Seq and qPCR readouts and expand theapplication of previously-described spike-in normalization (Loven

et al., 2012) to several representative experimental comparisonsfrequently performed during gene expression studies by thebiopharmaceutical industry. Conventional gene expression mea-surement techniques, which assume no change in total RNA yield

D. Fomina-Yadlin et al. / Journal of Biotechnology 189 (2014) 58–69 65

Fig. 5. Phenotypic comparisons of 6 antibody-expressing CHO cell lines over the course of a fed-batch production process. (A) Viable cell density (VCD), (B) culture viability,(C) specific productivity (qP ), (D) cell volume and (E) specific productivity adjusted by volume factor are shown during 10-day fed-batch productions for 6 cell lines withdiverse expression levels (cell lines A, B, C, D, E and F). Volume factor represents fold change with respect to the average cellular volume of cell line C on day 0. Data representm specific

flBmcdia

asebomt

ean ± SD of three biological replicates. (F) Relationship between cell volume andircled.

rom a set number of cells, inherently mask any differences in cel-ular RNA amount among experimental conditions (Fig. 7A and). We demonstrate that cell size, which varies with certain smallolecule treatments, among cell lines and during production pro-

esses, should be factored into normalization and determination ofifferential expression. Synthetic spike-in standards allow normal-

zation to cell number and reveal differences in cellular RNA contentmong experimental conditions with unequal cell size (Fig. 7C).

Experimental conditions that cause global transcriptionalmplification/repression also require alternative normalizationtrategies because conventional strategies distort differential genexpression comparisons. Only spike-in normalization to cell num-

er was able to detect the drastic differences in the directionalityf global gene expression changes between CCI and mTORI treat-ents in the two phenotypically different cell lines. Utilizing this

echnique, we were able to demonstrate that inhibition of cell cycle

c productivity on day 6 of a fed-batch production process. Desirable cell lines are

progression leads to global transcriptional amplification and inhi-bition of mTOR signaling leads to global transcriptional repression.A previous study with a genetic cell cycle inhibitor showed thatcell cycle progression stopped, yet the cell size, RNA content, pro-tein content, qP, and mitochondrial content continued to increase(Bi et al., 2004). To account for these effects, the authors pro-posed that cell cycle inhibition decouples cell growth from cellcycle progression, i.e. it blocks cell division, but cell growth con-tinues, resulting in larger cells (Bi et al., 2004). Consistent withthat interpretation, RNA content per cell and qP increased duringtreatment of our two cell lines with the CCI (Fig. 1A). There-fore, global transcriptional amplification can account for much

of the qP increase observed for cells treated with the CCI. How-ever, the volume adjusted titer was higher in CCI treated cellsindicating the CCI may have effects independent of the volumeincrease (Fig. 1F). Furthermore, the mTORI result suggests that

66 D. Fomina-Yadlin et al. / Journal of Biotechnology 189 (2014) 58–69

Fig. 6. Effects of normalization on gene expression measurements for the 6 antibody-expressing CHO cell lines. (A) Total RNA extracted from 3 million cells per cell lineon day 6 of the fed-batch production process, ranked by the RNA yield (left-to-right). (B) Scatter plot of the total RNA yield from 3 million cells and cell volume. (C) TotalRNA yield for each cell line adjusted by volume factor. Volume factor represents fold change with respect to the average total RNA yield of cell line C. (D) Gene expressionmeasurements of two spike-in controls (ERCC1 and ERCC2), selectable markers of the antibody light chain (LC) and heavy chain (HC) and two housekeeping genes (ActB andB2M) by qPCR normalized to the housekeeping gene expression average and expressed as fold cell line C. (E) Gene expression measurements by qPCR normalized to theERCC expression average (top panel) and normalized to ERCC average and adjusted by volume (bottom panel). All data represent mean ± SD of three biological replicates.* gene e

tdbRcfHiiSma

P < 0.05, **P < 0.01 and ***P < 0.001 represent statistically significant differences in

ranscriptional amplification/repression can be independent ofetectable cell size changes. Thus, spike-in normalization shoulde required for all experimental comparisons with unequal totalNA yield extracted from the same number of cells. Hypotheti-ally, an alternate strategy could be to normalize to total RNA yieldrom equal cell numbers rather than to synthetic spike-in controls.owever, this approach requires that the ratio of mRNA-to-rRNA

s consistent across cell lines and treatments, but previous studies

ndicate that the ratio can vary significantly (Johnson et al., 1977;olanas et al., 2001), arguing against this strategy. In addition, nor-alization to spike-in controls added before RNA extraction should

ccount for all the variability introduced by sample processing and

xpression between cell line C and 5 other cell lines.

sequencing, further arguing that spike-in normalization should bethe method of choice.

One puzzling result was that cells treated with the mTORinhibitor had similar antibody qP despite lower levels of total RNAas well as lower LC- and HC-associated RNAs. However, others haveshown that mTOR inhibitors can increase expression of heterolo-gous proteins due to a delay in both apoptosis induction and cellviability drop (Lee and Lee, 2012). These authors also found that

mTOR inhibitors had minimal impact on specific productivity (Leeand Lee, 2012). In addition, previous studies have demonstratedthat mTOR inhibitors reduce translation of specific sets of mRNAsmore than others (Hsieh et al., 2012). We speculate that in our

D. Fomina-Yadlin et al. / Journal of Biotechnology 189 (2014) 58–69 67

Fig. 7. Graphical comparison of normalization strategies for analysis of global gene expression. (A) Graphical representation of the four possible impacts of experimentaltreatment on gene expression. Experimental treatment outcomes are classified based on changes in RNA composition and RNA amount. Transcript abundance of genes A(black), B (blue) and C (red) is color coded to visualize changes. (B) Conventional normalization strategy assuming equal cellular RNA content across experimental conditionsis compared to (C) the spike-in normalization strategy that allows normalization by cell number. Two outcomes for the volcano plots shown in (C) depend on the direction ofdifferential expression comparison utilizing the spike-in normalized data, i.e. whether the black cell is the control condition and the blue cell is the experimental condition(left volcano plot), or vice versa (right volcano plot). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

6 l of Bi

espe

pwlsscocsitsdKpctTrbtIm2t

tgaTw“kcrinEcs

gcssstcsate

A

MtfS

8 D. Fomina-Yadlin et al. / Journa

xperimental system, the mTOR inhibitor decreases translation ofome cellular proteins, but has a limited impact on heterologousrotein translation, so the cells maintain similar levels of antibodyxpression.

In addition to application of spike-in normalization for com-arison of experimental treatments that alter cellular RNA contentithin a cell line (e.g. small molecule treatments and culture osmo-

arity), we have demonstrated the utility of this normalizationtrategy for multi-cell line studies. Specifically, we suggest that thepike-in normalization to cell number should be performed beforeorrelating gene expression with desirable phenotypic parametersf CHO cell lines. Furthermore, these data indicate that qP cal-ulations could give misleading information concerning cell lineuitability for certain types of manufacturing. Since large cells typ-cally have more RNA (Figs. 2A and 6B), and presumably protein,he qP calculation is inherently biased toward larger cells. In facteveral studies have demonstrated cell size to be the major pro-uctivity determinant in CHO cell lines (Dinnis and James, 2005;im et al., 2001; Lloyd et al., 2000). In our multi-cell line study,lotting qP versus cell volume can help rank cell lines (Fig. 5F), andlearly distinguishes cell lines B and D that have lower qP, comparedo cell line F, but higher qP adjusted for cellular volume (Fig. 5E).herefore, a less biased qP would be based on a “per unit volume”ather than a “per cell” calculation. This type of calculation woulde more appropriate for cell line development and final clone selec-ion, which aim to develop cell lines with small size and high qP.t would be particularly valuable for perfusion cultures where cell

ass can account for >40% of the reactor volume (Schirmer et al.,010), and smaller cell lines can achieve the higher cell densitieshan larger cells for the same amount of cell mass.

The three studies described in this manuscript demonstratehat in addition to application of the ERCC normalization to globalene expression measurements, this normalization strategy can bepplied to measurements of individual gene expression by qPCR.he problem of “housekeeping” gene selection that plagues thehole approach to qPCR stems from the variability in the baseline

housekeeping” gene expression and distinct responses of “house-eeping” genes to various experimental treatments. Spike-in ERCControls eliminate the need for housekeeping gene selection andeveal cell size dependent differences in gene expression by allow-ng normalization to cell number. Application of the spike-inormalization for qPCR would not require addition of the entireRCC Master Mix containing 92 synthetic RNAs to the sample, butould rely on spiking-in either individual or a few synthetic RNAtandards proportional to the number of collected cells.

Data normalization strategies affect interpretation of bothlobal transcriptional analysis and analysis of individual trans-ripts. Our results argue that spike-in normalization to cell numberhould become a widespread practice for evaluation of gene expres-ion, and that addition of spike-in controls to reflect cell numberhould at least be used in all gene expression experiments wherehe “no change in RNA yield per cell” assumption is not valid acrossonditions. Furthermore, similar spike-in normalization strategieshould be developed and applied to other ‘omic techniques, suchs proteomics and metabolomics, where the cellular content ofhe analyte of interest (e.g. protein or metabolite) varies amongxperimental conditions.

cknowledgements

We thank the Small Molecule Team for CHO Cell Growth and

etabolism for sharing data and experimental conditions. We also

hank Scott Freeman, Kelsey Anderson and Natalie Jones for per-orming antibody titer analysis. We also thank Rajnita Charan,umana Dey and Louiza Dudin for media preparation. We thank

otechnology 189 (2014) 58–69

GENEWIZ, Inc. for performing RNA library preparations, sequenc-ing reactions, and initial bioinformatics analysis. Finally, we thankBrian Follstad for useful discussions on data analysis and presenta-tion.

References

Bi, J.X., Shuttleworth, J., Al-Rubeai, M., 2004. Uncoupling of cell growth and prolif-eration results in enhancement of productivity in p21CIP1-arrested CHO cells.Biotechnol. Bioeng. 85, 741–749.

Clarke, C., Doolan, P., Barron, N., Meleady, P., Madden, S.F., DiNino, D., Leonard,M., Clynes, M., 2012. CGCDB: a web-based resource for the investiga-tion of gene coexpression in CHO cell culture. Biotechnol. Bioeng. 109,1368–1370.

Clarke, C., Doolan, P., Barron, N., Meleady, P., O’Sullivan, F., Gammell, P., Melville, M.,Leonard, M., Clynes, M., 2011a. Large scale microarray profiling and coexpressionnetwork analysis of CHO cells identifies transcriptional modules associated withgrowth and productivity. J. Biotechnol. 155, 350–359.

Clarke, C., Doolan, P., Barron, N., Meleady, P., O’Sullivan, F., Gammell, P., Melville, M.,Leonard, M., Clynes, M., 2011b. Predicting cell-specific productivity from CHOgene expression. J. Biotechnol. 151, 159–165.

Dinnis, D.M., James, D.C., 2005. Engineering mammalian cell factories for improvedrecombinant monoclonal antibody production: lessons from nature? Bio-technol. Bioeng. 91, 180–189.

Doolan, P., Barron, N., Kinsella, P., Clarke, C., Meleady, P., O’Sullivan, F., Melville, M.,Leonard, M., Clynes, M., 2012. Microarray expression profiling identifies genesregulating sustained cell specific productivity (S-Qp) in CHO K1 production celllines. Biotechnol. J. 7, 516–526.

Dreesen, I.A., Fussenegger, M., 2011. Ectopic expression of human mTOR increasesviability, robustness, cell size, proliferation, and antibody production of chinesehamster ovary cells. Biotechnol. Bioeng. 108, 853–866.

Du, Z., Treiber, D., McCarter, J., Fomina Yadlin, D., Saleem, R., McCoy, R., Zhang,Y., Tharmalingam, T., Leith, M., Follstad, B., Dell, B., Grisim, B., Zupke, C.,Heath, C., Morris, A., Reddy, P., 2014. Use of a small molecule cell cycleinhibitor to control cell growth and improve specific productivity and prod-uct quality of recombinant proteins in CHO cell cultures. Biotechnol. Bioeng.,http://dx.doi.org/10.1002/bit.25332.

Fingar, D.C., Salama, S., Tsou, C., Harlow, E., Blenis, J., 2002. Mammalian cell size iscontrolled by mTOR and its downstream targets S6K1 and 4EBP1/eIF4E. GenesDev. 16, 1472–1487.

Fomina-Yadlin, D., Gosink, J.J., McCoy, R., Follstad, B., Morris, A., Russell, C.B.,McGrew, J.T., 2014. Cellular responses to individual amino-acid depletion inantibody-expressing and parental CHO cell lines. Biotechnol. Bioeng. 111,965–979.

Hsieh, A.C., Liu, Y., Edlind, M.P., Ingolia, N.T., Janes, M.R., Sher, A., Shi, E.Y., Stumpf, C.R.,Christensen, C., Bonham, M.J., Wang, S., Ren, P., Martin, M., Jessen, K., Feldman,M.E., Weissman, J.S., Shokat, K.M., Rommel, C., Ruggero, D., 2012. The trans-lational landscape of mTOR signalling steers cancer initiation and metastasis.Nature 485, 55–61.

Johnson, L.F., Abelson, H.T., Penman, S., Green, H., 1977. The relative amounts of thecytoplasmic RNA species in normal, transformed and senescent cultured celllines. J. Cell. Physiol. 90, 465–470.

Kang, S., Ren, D., Xiao, G., Daris, K., Buck, L., Enyenihi, A.A., Zubarev, R., Bondarenko,P.V., Deshpande, R., 2014. Cell line profiling to improve monoclonal antibodyproduction. Biotechnol. Bioeng. 111, 748–760.

Kantardjieff, A., Jacob, N.M., Yee, J.C., Epstein, E., Kok, Y.J., Philp, R., Betenbaugh,M., Hu, W.S., 2010. Transcriptome and proteome analysis of Chinese hamsterovary cells under low temperature and butyrate treatment. J. Biotechnol. 145,143–159.

Kildegaard, H.F., Baycin-Hizal, D., Lewis, N.E., Betenbaugh, M.J., 2013. The emergingCHO systems biology era: harnessing the ‘omics revolution for biotechnology.Curr. Opin. Biotechnol. 24, 1102–1107.

Kim, T.K., Chung, J.Y., Sung, Y.H., Lee, G.M., 2001. Relationship between cell sizeand specific thrombopoietin productivity in chinese hamster ovary cells duringdihydrofolate reductase-mediated gene amplification biotechnol. Biomol. Eng6, 332–336.

Lee, J.S., Lee, G.M., 2012. Rapamycin treatment inhibits CHO cell death in aserum-free suspension culture by autophagy induction. Biotechnol. Bioeng. 109,3093–3102.

Li, Y., Wang, H., Muffat, J., Cheng, A.W., Orlando, D.A., Loven, J., Kwok, S.M., Feldman,D.A., Bateup, H.S., Gao, Q., Hockemeyer, D., Mitalipova, M., Lewis, C.A., VanderHeiden, M.G., Sur, M., Young, R.A., Jaenisch, R., 2013. Global transcriptional andtranslational repression in human-embryonic-stem-cell-derived Rett syndromeneurons. Cell Stem Cell 13, 446–458.

Lin, C.Y., Loven, J., Rahl, P.B., Paranal, R.M., Burge, C.B., Bradner, J.E., Lee, T.I., Young,R.A., 2012. Transcriptional amplification in tumor cells with elevated c-Myc. Cell151, 56–67.

Lloyd, D.R., Holmes, P., Jackson, L.P., Emery, A.N., Al-Rubeai, M., 2000. Relation-

ship between cell size, cell cycle and specific recombinant protein productivity.Cytotechnology 34, 59–70.

Loven, J., Orlando, D.A., Sigova, A.A., Lin, C.Y., Rahl, P.B., Burge, C.B., Levens, D.L.,Lee, T.I., Young, R.A., 2012. Revisiting global gene expression analysis. Cell 151,476–482.

l of Bi

M

N

N

R

S

D. Fomina-Yadlin et al. / Journa

ortazavi, A., Williams, B.A., McCue, K., Schaeffer, L., Wold, B., 2008. Mapping andquantifying mammalian transcriptomes by RNA-Seq. Nat. Methods 5, 621–628.

ie, Z., Hu, G., Wei, G., Cui, K., Yamane, A., Resch, W., Wang, R., Green, D.R., Tes-sarollo, L., Casellas, R., Zhao, K., Levens, D., 2012. c-Myc is a universal amplifierof expressed genes in lymphocytes and embryonic stem cells. Cell 151, 68–79.

issom, P.M., Sanny, A., Kok, Y.J., Hiang, Y.T., Chuah, S.H., Shing, T.K., Lee, Y.Y., Wong,K.T., Hu, W.S., Sim, M.Y., Philp, R., 2006. Transcriptome and proteome profilingto understanding the biology of high productivity CHO cells. Mol. Biotechnol.34, 125–140.

asmussen, B., Davis, R., Thomas, J., Reddy, P., 1998. Isolation, characterization and

recombinant protein expression in Veggie-CHO: a serum-free CHO host cell line.Cytotechnology 28, 31–42.

chirmer, E.B., Kuczewski, M., Golden, K., Lain, B., Bragg, C., Chon, J., Cacciuttolo,M., Zarbis-Papastoitsis, G., 2010. Primary clarification of very high-density cellculture harvests by enhanced cell settling. BioProcess Intl. 8, 32–39.

otechnology 189 (2014) 58–69 69

Schliess, F., Reinehr, R., Haussinger, D., 2007. Osmosensing and signaling in theregulation of mammalian cell function. FEBS J. 274, 5799–5803.

Shen, D., Kiehl, T.R., Khattak, S.F., Li, Z.J., He, A., Kayne, P.S., Patel, V., Neuhaus, I.M.,Sharfstein, S.T., 2010. Transcriptomic responses to sodium chloride-inducedosmotic stress: a study of industrial fed-batch CHO cell cultures. Biotechnol.Prog. 26, 1104–1115.

Solanas, M., Moral, R., Escrich, E., 2001. Unsuitability of using ribosomal RNA asloading control for Northern blot analyses related to the imbalance betweenmessenger and ribosomal RNA content in rat mammary tumors. Anal. Biochem.288, 99–102.

Yang, I.V., 2006. Use of external controls in microarray experiments. Methods Enzy-mol. 411, 50–63.

Yee, J.C., Gerdtzen, Z.P., Hu, W.S., 2009. Comparative transcriptome analysis tounveil genes affecting recombinant protein productivity in mammalian cells.Biotechnol. Bioeng. 102, 246–263.