translatome analysis of cho cells

7/23/2019 Translatome Analysis of Cho Cells

http://slidepdf.com/reader/full/translatome-analysis-of-cho-cells 1/10

Journal of Biotechnology 167 (2013) 215–224

Contents lists available at ScienceDirect

Journal of Biotechnology

journal homepage: www.elsevier .com/ locate / jb iotec

Translatome analysis of CHO cells to identify key growth genes

Franck C. Courtes a,b, Joyce Lin a, Hsueh Lee Lim a, Sze Wai Ng a, Niki S.C. Wong a,c,Geoffrey Koh a, Leah Vardy d, Miranda G.S. Yap a,b, Bernard Loo a,∗∗, Dong-Yup Lee a,b,∗

a Bioprocessing TechnologyInstitute,A*STAR (Agency for Science, Technologyand Research),20 BiopolisWay,#06-01 Centros, Singapore 138668, Singaporeb Department of Chemical and Biomolecular Engineering, National University of Singapore, 4 Engineering Drive 4, Singapore117576, Singaporec AbbVie Pte Ltd., 8 Biomedical Grove, #03-01, Neuros, Singapore 138665, Singapored Institute ofMedical Biology, A*STAR(Agency for Science, Technology andResearch), 8A Biomedical Grove,#06-06 Immunos, Singapore138648, Singapore

a r t i c l e i n f o

Article history:

Received 23 March 2013

Received in revised form 14 June 2013

Accepted 10 July 2013

Available online 19 July 2013

Keywords:

Translatome

CHO cells

Polysome profiling

Cellular growth

a b s t r a c t

We report the first investigation of translational efficiency ona global scale, also known as translatome, of

a Chinese hamster ovary (CHO) DG44 cell line producing monoclonal antibodies (mAb). The translatome

data was generated via combined use of high resolution and streamlined polysome profiling technology

and proprietary Nimblegen microarrays probing for more than 13K annotated CHO-specific genes. The

distribution of ribosome loading during the exponential growth phase revealed the translational activity

corresponding to the maximal growth rate, thus allowing us to identify stably and highly translated genes

encoding heterogeneous nuclear ribonucleoproteins (Hnrnpc andHnrnpa2b1), protein regulator of cytoki-

nesis 1 (Prc1), glucose-6-phosphate dehydrogenase (G6pdh), UTP6 small subunit processome (Utp6) and

RuvB-like protein 1 (Ruvbl1) as potential key players for cellular growth. Moreover, correlation analysis

between transcriptome and translatome data sets showed that transcript level and translation efficiency

were uncoupled for 95% of investigated genes, suggesting the implication of translational control mech-

anisms such as the mTOR pathway. Thus, the current translatome analysis platform offers new insights

into gene expression in CHO cell cultures by bridging the gap between transcriptome and proteome data,

which will enable researchers of the bioprocessing field to prioritize in high-potential candidate genes

and to devise optimal strategies for cell engineering toward improving culture performance.

© 2013 Published by Elsevier B.V.

1. Introduction

Chinese hamster ovary (CHO) cells are one of the most com-

monly used mammalian host cell lines for the production of

recombinant proteins of therapeutic interest ( Jayapal et al., 2007).

In order to meet the increasing demand for such recombinant pro-

teins, consistent efforts havebeen madefor significantlyenhancing

the production capacity of CHO cell cultures (Barnes and Dickson,

2006). Some of the major improvements were achieved by empiri-

cally designed strategies (Kim et al., 2012; Lim et al., 2010; Wurm,

2004), but without clear understanding of the underlying cellular

mechanisms. The lack of such crucial understanding has hindered

the development of knowledge-based strategies to fully optimize

and control cell culture processes. To address this limitation, sev-

eral “-omics” profiling technologies have been successfully utilized

(Kuystermans et al., 2007; O’Callaghan and James, 2008; Omasa

∗ Corresponding author at: Department of Chemical and Biomolecular Engineer-

ing, National University of Singapore, 4 Engineering Drive 4, Singapore 117576,

Singapore. Tel.: +65 6516 6907; fax: +65 6779 1936.∗∗ Corresponding author.

E-mailaddresses:[email protected] (B. Loo), [email protected] (D.-Y. Lee).

et al., 2010; Selvarasuet al., 2012), thereby gaining a more in-depth

insight into these cellular mechanisms. For example, a targeted

transcriptomic approach was used to track the transcriptional reg-

ulation accompanying the onset of apoptosis between batch and

fed-batchcultures (Wong et al., 2006). As a result, seven core apop-

totic genes were detected as differentially expressed, including

Faim, Agl2, Fadd and Requiem. Interestingly, the transcriptome of

CHO cells treated with sodium butyrate revealed that transcripts

pertaining to apoptosis, cellcycle and protein folding (DeLeon Gatti

et al., 2007; Yee et al., 2008) as well as carbohydrate metabolism

and signal transduction pathways (Klausing et al., 2011) were dif-

ferentially regulated. The transcriptome of CHO cells undergoing

a temperature shift at 33◦C suggested that transcripts enriched

in protein trafficking and cytoskeleton reorganization were up-

regulated (Yee et al., 2009).

Similarly, Lee et al. (1996) initiated the exploration of the

CHO proteome by identifying 3 proteins, CCND1, CCNE and

e2F-1, that correlated with mitogenic signals. Later, a prelim-

inary proteome map of CHO culture was established by 2-D

gel electrophoresis (Champion et al., 1999). Temperature shift

to 31 ◦C affected the quantity of 10 proteins in a CHO cell cul-

ture (Kaufmann et al., 1999) and an increase of osmotic pressure

from 300 to 450mOsmkg−1, significantly changed the level of

0168-1656/$ – seefrontmatter © 2013 Published by Elsevier B.V.

http://dx.doi.org/10.1016/j.jbiotec.2013.07.010

http://localhost/var/www/apps/conversion/tmp/scratch_2/dx.doi.org/10.1016/j.jbiotec.2013.07.010


http://www.sciencedirect.com/science/journal/01681656

http://www.elsevier.com/locate/jbiotec

mailto:[email protected]






http://crossmark.dyndns.org/dialog/?doi=10.1016/j.jbiotec.2013.07.010&domain=pdf

http://www.elsevier.com/locate/jbiotec

http://www.sciencedirect.com/science/journal/01681656




216 F.C.Courtes et al. / Journal of Biotechnology 167 (2013) 215–224

glyceraldehyde-3-phosphate dehydrogenase, pyruvate kinase and

tubulin (Lee et al., 2003). The proteome of CHO cells engineered

with the apoptosis inhibitor Bcl-XL to minimize cell death high-

lighted thirty-two differentially expressed proteins enriched in

protein metabolism, transcription and cytoskeleton functional

clusters (Carlage et al., 2009). In another study, twelve differentially

expressed proteins involved in glucose metabolism and protein

translation/folding were identified through comparison between

high and low mAb producer CHO cells (Meleady et al., 2011).

With such successful applications of each level of omics pro-

filing, clearly the study on correlation between transcriptome and

proteome hasbeen a topic of greatinterestto theresearchcommu-

nity. However, only a handful attempted to combine both levels of

information (Baik et al., 2006; Doolan et al., 2010; Nissom et al.,

2006; Yee et al., 2008) and commonly concluded that there was

a general lack of correlation between mRNA and protein expres-

sion levels in CHO cells. The same conclusion was also reported

in various human cell lines and yeast (de Nobel et al., 2001; Gygi

et al., 1999; Pradet-Balade et al., 2001), implying that a change of

a transcript level may not be necessarily accompanied by a simi-

lar change (direction and amplitude) at the protein level. In fact,

it has been estimated that the concentration of only 20–40% of

the total proteins is determined by the corresponding transcript

level in mammalian cells (Cox et al., 2005; Tian et al., 2004), sug-gesting that not all mRNA are translated with the same efficiency

according to the cellular requirements (Brockmann et al., 2007).

Thus, for a more inclusive interpretation of the “-omics” landscape

in CHO cells, deeper understanding of translational regulation is

highly required. Herein, the “-omics” for investigatingtranslational

regulation is commonly referred to as the translatome.

Translatome information can be obtained via a combination

of polysome profiling and microarray technologies (Maˇ sek et al.,

2011), allowing us to cluster mRNAs with respect to ribosome

loading, and hence provide a snapshot of their translational effi-

ciency (Aravaet al., 2005). Theseclustered mRNAs can thereafter be

quantified via microarrays (global)or qRT-PCR (targeted) technolo-

gies. Recently, such translatome approach has been successfully

employed on several organisms including bacteria (Picard et al.,2012), yeast (Arava et al., 2003; Preiss et al., 2003), stem cells

(Sampath et al., 2008) and human cell lines (Thomas and Johannes,

2007) but, it is still newto CHOcells, whichare industriallyrelevant

mammalian host cells. Thus, the objective of the current study is

to establish and apply a translatomics platform to CHO cell culture

in order to elucidate their translational characteristics. We showed

the relevance of such translatome analysis as a new approach to

identify potential cell engineering targets for enhancing cellular

growth. Moreover, we explored the relationship between trans-

criptome and translatome to further understand the implication of

translational control mechanism in gene expression of this mam-

malian expression host.

2. Materials and methods

2.1. Cell line and cell culture

A stable Chinese hamster ovary (CHO) DG44 cell line express-

ing a recombinant monoclonal immunoglobulin G (IgG) against

a human rhesus-D antigen was utilized in this study (Chusainow

et al., 2009). All cultures were performed in 3 L disposable Erlen-

meyer flasks. Initial working volumes were 1L of a proprietary

protein free and chemically defined medium, supplemented with

8 mM l-glutamine (Sigma–Aldrich, St. Louis, MO), 600g mL −1

G418 (Sigma–Aldrich, St. Louis, MO), 250n M methotrexate

(Sigma–Aldrich, St. Louis, MO) and 0.1% (v/v) Pluronic® F68 (Invi-

trogen, Carlsbad,CA). Thecultures were seededat 3×

105

cell mL −1

andgrown in suspension at 37◦C,under an8% CO2 atmosphere and

shaker platforms set at 110rpm in a humidifiedincubator (Kühner,

Germany).Cell densitiesand viabilities weredetermined by thetry-

pan blue exclusion method using a Cedex automated cell counter

(Innovatis, Roche, Basel, Switzerland).

2.2. Polysome profiling and fractionation

For polysome profiling, 20×106 cells were incubated with

100g mL −1 of cycloheximide (Sigma–Aldrich, St. Louis, MO) for

10min at 37 ◦C to arrest ribosome progression on mRNAs before

harvesting the cells. The harvested cells were centrifuged at

1500rpm for 5 min at 4 ◦C and resuspended in 240L of resuspen-

sion buffer (20mM Tris–HCl pH 7.4, 20mM NaCl, 30mM MgCl2,

RNasin, 100g mL −1 Heparin, and 5g mL −1 of cycloheximide).

Cells were then lysed for 10min on ice in 265L of lysis buffer

(1.2% (v/v) Triton X-100, 1.2% (v/v) deoxycholate). The nuclei and

cell debris were removed by centrifugation at 12,000rpm for

10min. Nucleic acid content was quantified by absorbance read-

ing at 260 nm on an aliquot of collected supernatant. Thereafter,

equal optical density (A260) unit per volume unit were loaded

onto 11.5mL linear 10–50% sucrose gradients and centrifuged at

34,000rpm for 2 h at8 ◦C in a SW41 bucket rotor (Beckman, Fuller-

ton, CA). Polysome profiles were monitored at 260nm with anEM-1 UV Monitor (BioRad, Hercules, CA). For fractionation of the

polysomes, 20 fractions (0.5mL each) were collected with a Bio-

Comp piston gradient fractionator linked to the EM-1 UV Monitor.

The fractions were incubated in 10% (v/v) SDS and 1.2% (v/v) pro-

teinase K at 42 ◦C for 30 min and stored at −80 ◦C before RNA

extraction.

2.3. RNA pooling, extraction and purification

Thirteen fractions starting from the 80S peak (Fig. 1) were

pooled in monosome (pool B) and polysome (pool A) enriched

pools according to an adaptation of the strategy previously estab-

lished by Hendrickson et al. (2009). RNAwas extractedfrom 900L

of each pool by phenol–chloroform extraction. Equal volumes of the aqueous phase containing the RNA were precipitated with 0.1

volume of 5M sodium acetate (Ambion, Austin, TX) and 1 vol-

ume of isopropanol (Sigma–Aldrich) with overnight incubation at

−80 ◦C. The RNA pellets were then washed with 75% (v/v) ethanol

(Sigma–Aldrich) and precipitated with lithium chloride (Ambion,

Austin, TX) and incubated overnight at −80 ◦C. RNA precipitates

were washed with 70% ethanol and subjected to a final round

of sodium acetate and ethanol precipitation. Finally, the precipi-

tated RNA pellets were washed in 75% ethanol and resuspended

in 30L of RNase-free water. RNA concentration was determined

using NanoDrop (Thermo Scientific, Bremen, Germany) and RNA

samples were stored at−80 ◦C until runs of microarrays.

2.4. Microarray hybridization, washing and scanning

Microarray data were generated from one biological sample per

day during the exponential growth phase, so that the data from

four time points could be obtained whilst using an affordable num-

ber of microarray chips. Importantly, the exponential phase can

represent a steady physiological state where cells are growing at

their maximum growth rate and the genes supporting growth are

likely to be steady in expression. Under this assumption, four time

points could be considered as biological replicates, offering the

advantage to provide information-richer data that cover the full

exponential growth phase. For transcriptome, total RNAs were iso-

lated from 5×106 cells using RNeasy Mini Kit (Qiagen, Valencia,

CA). For translatome samples, total RNAs were isolated and pooled

in pool A and B as described above. Fifty nanograms of total RNAs



F.C. Courtes et al. / Journal of Biotechnology 167 (2013) 215–224 217

Table 1

List of primers pairs utilized for normalization of themicroarrays data.

# Target gene Forward (5 > 3) Reverse (5 > 3) Efficiency

1 CLPP CAGAAGGAGACTCCCACAGC AGAGCAGCCTCAATCGACAT 0.998

2 HIST1H CCAAAAGTCCACCGAGCTAC CACAGGTTGGTGTCCTCAAA 1.013

3 NFAT5 GGGTCAAACGACGAGATTGT TCCAGCTTTTGAGTTGCCTT 0.954

4 MRG2 ACACTCACACTCCCACGTCA TGCTGCTGTCTCATGGTTTC 1.013

5 PRRX1 TAAAAACGCTTCCCTCCTCA AGAGTGGGCCATTCATTCAC 0.986

6 RBMX2 CTGGAAGCAGGGCAGTAAAG TCTTCCTCCCCTCTCTGGAT 0.940

7 MAP3 K TTAAGGCAGGTGAACAAGGG CTGCTGCTGCACATAAGCTC 1.0458 GRINA CGCCATACTCTGCATCTTCA TGTGTACAGGTTCAGAGCCG 0.930

9 CRIP1 ACTCGTGCAGGACCAAGTTC GTAGCAGGGATGGTTGCAGT 1.049

10 ITGB5 CATCCAGATGACACCACAGG CATCCTTCATGGAGAGGGAA 0.923

11 TTYH2 ACAGAACACACTGAAGCCCC GGCTCACAGTATTCACGGGT 0.909

12 SLC19A1 CTTCACAATCGAGCAGGTGA CAGGATCAGGATTGGCTTGT 0.954

13 ALG3 ATAAAGGTGGTGGTTGCTCG ACCAGGCTGGCCTTAAACTT 0.913

14 PSMC3IP ATTTTGCAGACCAGGACCAG TGCATCTCAGGAGTGGTCAG 0.966

15 DOCK8 CTCATGATGGCTGGGAATCT AATTGTCCCCTGGGTAAAGG 0.969

from each sample were amplified and reverse transcribed using

TransPlex Whole Transciptome Amplification Kit (Sigma–Aldrich,

St. Louis, MO). cDNA labeling, hybridization and washing were car-

ried out as described in NimbleGen Arrays User’s Guide – Gene

Expression Arrays protocol. Labeled cDNA were synthesized using

NimbleGen One-Color DNA Labeling Kit (NimbleGen, Roche, Basel,Switzerland) from one microgram of cDNA and then purified using

salt precipitation method. Unique Sample Tracking Controls (STCs)

from NimbleGen Sample Tracking Control Kit (NimbleGen, Madi-

son, WI) was added to each labeled cDNA sample. The STCs are

to track the sample locations, after hybridization, and to monitor

Fig. 1. Schematic overview of the translatomic platform. Ribosome bound RNAs

were separatedon 10–50% sucrose gradient andthe polysome profilewas fraction-

ated in thirteen differentfractions to allow high resolution of thetranslatomedata.

The fractions were reduced to pool A (enriched in polysome) and pool B (enriched

in monosome) according to the sigmoidal-trend repartition of volumes devised by

Hendrickson et al. (2009). Translatomedefinedas the measurement of translational

efficiency of every gene was calculated as the ratio of microarray intensity obtain

for poolA and poolB.

for any cross-contamination during loading and hybridization of

target to array. The targets were hybridized to the 12×135K

CHO NimbleGen arrays (proprietary) for 16h. The arrays were

washed and scanned using Axon GenePix 4000B Scanner. Data

was extracted using NimbleScan v2.6 software. The Sample Track-

ing Report indicated that there was no cross-contamination of thesamples.

2.5. Pre-processing of microarrays data

All the chips were background corrected using NimbleScan

V2.6 software. On each chip, there were a total of 135,883 probes

covering a total of 13,514 annotated CHO genes and other con-

trol genes. The intensities of probes corresponding to same gene

were averaged and quantile normalization was then performed on

transcriptome, pool A and pool B samples separately using the R

package AffyPLM (http://svitsrv25.epfl.ch/R-doc/library/affyPLM/

html/normalize.exprSet.html). The normalized microarray inten-

sity values were validated by quantitative RT-PCR performed on a

selectionof 15 genes (Table 1) that covered the range of microarraynormalized values for transcriptome and translatome (ratio pool

A over pool B). Each qRT-PCR was performed in technical dupli-

cates in a 96-well iQ real-time PCR plate (Biorad, Hercules, CA) for

40 cycles 95◦C for 5 s and 60 ◦C for 10s. Total volume of reaction

was 10L per well. For each technical duplicate, reaction mixtures

were prepared as master mix with 8L of cDNA solution, 10L of

SsoFastTM EvaGreen® supermix (Biorad, Hercules, CA) and 2L of

500M primer pair.

2.6. Microarrays data analysis

For determining translational efficiency of each gene, intensity

value measured in pool A was divided by its counterpart intensity

value measured in pool B. A 1.5 ratio fold change was consideredas a significant differential enrichment between pool A and pool B.

For functional clustering, the online program DAVID was utilized

(Huangetal.,2009). Thisprogramenriched genes in GOterms based

on the existing functional annotation between gene identifiers

and GO terms. GO terms were themselves clustered in annota-

tion clusters that grouped together GO terms of similar biological

meaning due to sharing similar enriched gene members. Input list

of selected genes were clustered under the GOTERM BP FAT gene

ontology system. Only GO terms with a P -value lower than 0.05

were accepted as significantly enriched. In brief, P -values were cal-

culated according to a modified Fisher exactP -value that testedthe

random chance for genes to be enriched in GO terms. Thereafter,

the P -values of accepted GO terms pertaining to one annotation

cluster were averaged.

http://svitsrv25.epfl.ch/R-doc/library/affyPLM/html/normalize.exprSet.html







Fig. 2. Framework of translatome analysis forthe identification of key growth genes.Left panel:Cellular growth performance of CHO cells in batch culture. The logarithmic

scale was usedto determinethe linearityrange of theexponential growth phasecharacterizedby maximum and constantgrowth rate. Rightpanel: Global translation activity

measured from polysome profiling. Only the polysome profile of day 1 is shown as an example. Monosomes (M ) and polysomes (P ) area under the curve were determined

with the trapezium rule and presented in the barchart. The ratio of P /M (blue bars×10 forpresentation purpose) was an indicator of global translation, which appeared to

be relatively stable over time. (For interpretation of thereferences to color in this figure legend, thereader is referred to theweb version of thearticle.)

3. Results and discussion

3.1. Generation of first translatome data in CHO cells

We developed a translatome platform for recombinant CHO

cells by combining high-resolution polysome profiling technol-

ogy and proprietary NimbleGenmicroarrays withstreamlined data

processing. To generate translatome data, total mRNA was sep-

arated with respect to ribosome loading density on a sucrose

gradient by centrifugal force. As such, higher density RNAs (more

ribosomes attached) were spun toward the bottom of centrifuge

tubes (Fig. 1). Thirteen different fractions from the sucrose gradi-

ent were collected ranging from the 80S monosome peak (fraction1) to the highest polysome degree (fraction 13), via polysome pro-

filing. Note thathigher number of fractions increases the resolution

of translatome data. For example, twelve fractions allow for a finer

separation of differentially translated mRNAs than four fractions

and thus a 3-fold increase in the level of information within one

sample. In this regard, thirteen fractions collected in the current

work can provide a high resolution of the translatome informa-

tion as compared to other previous translatome studies based on

two (Tebaldi et al., 2012), four (Sampath et al., 2008) and twelve

(Thomas and Johannes, 2007) fractions.

Across the fractions collected, monosomes were enriched in

mRNAs with low translation activity while polysomes represented

mRNAs that were being highly translated. As a result, each frac-

tion contained mRNA of different translation efficiencies, which

were then pooled together in two pools based on a sigmoidal trend

repartition of the fractions adapted from the pooling approach

developed by Hendrickson et al. (2009) f or human HEK293T cells.

The two pools were enriched in polysomes (pool A: highly trans-

lated mRNA) or monosomes (pool B: poorly translated mRNA).

Each pool was then probed on separate proprietary NimbleGen

CHO microarray chip covering 13,514 annotated CHO genes. For

eachgene,translatomedata wereprocessed as microarray intensity

ratios of polysome-enriched pool (pool A) to monosome-enriched

pool (pool B), noted as A/B. Interestingly, recent next-generation

sequencing technologies have enabled high-throughput quantifi-

cation of RNA samples, as an alternative to microarrays (Becker

et al., 2011). However, this approach still requires extensive bioin-formatic analysis to properly handle huge amount of sequence

reads ( Jacob et al., 2010; Wang et al., 2009). Thus, the more robust

and established microarrays technology available can be still used

in this study.

Processing translatome data with large number of fractions is

one of main challenges since it is not straightforward to weigh

the microarray intensities of mRNAs found in the different frac-

tions toward one translational efficiency value per gene. Preiss

et al. (2003) collected twelve fractions and pooled a selection of

them in two pools with no weightage of their respective contri-

bution to the overall translational efficiency. By doing so, they

could determine translational efficiency as the ratio of the two

pools, although the benefit of the twelve fractions resolution was

actually reduced to two fractions. The translational efficiency can




also be assessed by graphical visualization showing the microarray

intensity of mRNAs in each fraction (Thomas and Johannes, 2007)

but this representation remains semi-quantitative and tedious for

global scale analysis. Recently, Sampath et al. (2008) suggested a

mathematical approach that could process the microarray intensi-

ties from four fractions into one numerical value per gene, which

could then conveniently be used for further computational analy-

sis. However, this mathematical approach was designed only for

four fractions and it is not readily applicable to a higher num-

ber of fractions. Unlike other translatome data processing, the

sigmoidal pooling strategy utilized in this study allocated a differ-

ent volume percentage of each fraction (Fig. 1), thereby weighing

their contribution, and ensuring the maintenance of the high-

resolution information even after reduction to the two pools (pools

A and B).This strategy allowed us to achieve high-resolution trans-

latome data via streamlined data processing. It should be noted

that in future, we need to validate whether this proposed sigmoid-

approach can reliably extract the relative translation efficiency in

each fraction. It could be achieved by comparing the quantitative

value(ratio of pool A/poolB) against the corresponding trend of the

relative intensity-distribution within the 13 fractions, of each gene

or statistically representative subset of genes, serving as a semi-

quantitative reference. Now, resultant data can be further analyzed

to identify key growth genes as potential cell engineering targetsas well as to explore translational control mechanisms in CHO cell

cultures.

3.2. Global translation activity during exponential cell growth

Cellular growth rate is known to reach its maximum during

the exponential growth phase. Based on the principle that cellular

growth is essentially driven by protein accumulation via transla-

tion throughout the G1 phase of the cell cycle, global translation

activitywas assessed by polysomeprofilingon exponentiallygrow-

ing CHO cells. Fig. 2 (left panel) displays the growth performance

of the cells; as indicated by the logarithmic-scale plot, the linear-

ity range corresponding to the exponential growth phase occurred

from day 1 till day 4 and at a maximal constant growth rate max

of 0.64day−1. Samples for translatome were collected on days 1–4

of the batch culture as well as for the transcriptome samples in

order to investigate the correlation between transcriptome and

translatome as discussed in the next section. Global translation

activity supported the constant and maximal growth rate (max)

during the exponential growth phase as shown by the distribution

of monosomes and polysomes via polysome profiling (right panel

in Fig.2). Thebar chartdepicts theareaunder thecurves ingreenfor

polysome (P ) and in red for monosome (M ) which were utilized to

calculate global translation as the ratio of polysome to monosome

via polysome profile analysis (Maˇ sek et al., 2011). Global transla-

tion was relatively constant at an average of P /M =1.6±0.1 over

the four days and corresponded to the maximum global transla-

tion value that was observed over a full-length culture (data notshown). The fact that this ratio was maximum and rather constant

was in good agreement with the expected physiological state of

exponentially growing CHO cells, where ideally there was neither

limitation of nutrients, nor accumulation of toxic molecules which

could have affected maximal growth ratemax. In summary, global

translation supported maximum growth rate by producing high

level of cellular proteins. Thus, the identity and biological function

of mRNAs specifically enriched in polysomes (highly translated)

can be further analyzed via the translatome data.

3.3. Translatome: a new strategy to identify growth genes

Specific growth rate and global translation trends in theleft and

right panels of Fig. 2, respectively, were observed to be constant

Fig. 3. Workflow of data processing toward the identification of key growth genes

in CHO cells. Shortlisted genes in the translatome data (current study) and overlap

comparison with the genes previously identified by transcriptome and proteome

analysis are summarized.

during the exponential growth phase. Hence, only genes with

constant translational efficiency (relative standard deviation< 10%

over the four days) were regarded as potentially associated with

cellular growth, thereby identifying 4003 genes with annotated

functions (Fig. 3). These genes were, then, statistically analyzedbased on both average and standard deviation of their respec-

tive translational efficiencies from day 1 to day 4. The highest

and lowest translational efficiencies were A/B= 4.38±0.35 and

A/B =0.09±0.01 respectively; the overall average was 1.24. Such

distribution indicated that there were more enriched genes in

polysome fractions during the exponential growth phase and was

in goodagreement withthe observed global translation trend (right

panel in Fig. 2). Herein, we only selected highly translated genes,

which are positively correlated with the cellular growth. Amongst

the 4003 genes, 1079 (26.9%) genes were classified as highly trans-

lated based on a fold change of A/B> 1.5. Note that this moderately

stringent threshold is often used in high-throughput data analysis

(Dalman et al., 2012) and allowed for subsequent functional

enrichment analysis using the program DAVID. From the signifi-cantly enriched functional clusters (enrichmentP -value< 0.05),we

further shortlisted high-potential cellular growth candidates by

applying more stringent condition ( A/B> 2.5), as such resulting in a

total of 43 non-redundant genes listed in Table 2. Among them, 17

genes were enriched in two or more different functional clusters.

This first CHO translatome information was compared withpre-

viously reported growth-associated genes in CHO cells (Fig. 3).

It should be noted that 21 (Doolan et al., 2010) and 41 (Clarke

et al., 2012) key growth genes were identified by a comparison

of transcriptome and proteome data between fast and slow grow-

ing CHO cells while 13 by coexpression network analysis of 295

transcriptome profiles of CHO cells (Clarke et al., 2011). Inter-

estingly, among the 43 shortlisted genes by our translatomics

platform, five genes (Hnrnpc , Hnrnpa2b1, Mcm5, Pcna, Vcp) were




Fig. 4. Correlation between transcript level and translational efficiency. (A) Genes with both stable transcript level and translational efficiency over the exponential growth

phase. (B) Correlation plot of transcriptome vs translatome. Thresholds for high and low transcript levels were defined by the average TCav plus and minus the standard

deviation ( ), respectively while a 1.5-fold change was used for the translatome data. Green and yellow regions correspond to uncoupled transcript level and translational

efficiency. Blue region marks the genes for which there is positive correlation while genes in the red region are negatively correlated. (C) Validation of microarray data via

qRT-PCR using transcriptome and translatome (ratio A/B) from day 1. Linear regression between microarrays and qRT-PCR values displayedR2 coefficients greater than 0.7,

whichwas comparableto the correlationobtainedfor translatome data measured with Affymetrixarrays (Sampath et al., 2008). (For interpretationof the references to color

in this figure legend, thereader is referred to theweb version of thearticle.)




Table 2

Functional enrichment of genes with constant and high translational efficiency during the exponential growth phase.

Annotationcluster P -value Gene IDa Details A/B

RNA processing 1.0E−03

Utp6 UTP6, small subunit (SSU) processome component 3.36

Hnrnpc Heterogeneous nuclear ribonucleoprotein C 2.88

App Amyloid beta (A4) precursor protein Gene 2.69

Hnrnpa2b1 Heterogeneous nuclear ribonucleoprotein A2/B1 2.68

Fip1l1 FIP1 like 1 2.56

Syncrip Sy napt ot agmin bin ding, cyt oplasmicRNA int eract ing prot ein 2.54

Parn Poly(A)-specific ribonuclease ( deadenylation n uclease) 2.50

Cell cycle 1.8E−03

Prc1 Protein regulator of cytokinesis 1 Gene 3.18

Ruvbl1 RuvB-like protein 1 Gene 3.02

Vps4b Vacuolar protein sorting 4b (yeast) Gene 2.76

App Amyloid beta (A4) precursor protein Gene 2.69

Anxa11 Annexin A11 Gene 2.67

Pdcd6ip Programmed cell death 6 interacting protein Gene 2.67

Mcm5 Minichromosomemaintenance deficient5, cell divisioncycle 2.57

Chtf18 CTF18, chromosome transmission fidelity factor 18 2.54

tRNA aminoacylation 5.3E−03 Wars Tryptophanyl-tRNA synthetase 3.26

Nucleic transport of proteins 6.5E−03

Pcna Proliferating cell nuclear antigen 3.22

Snx14 Sorting nexin 14 3.09

Ipo8 Importin 8 2.81

Sec31a Sec31 homolog A 2.77

Vps4b Vacuolar protein sorting 4b 2.76

Exoc1 Exocyst complex component 1 2.74

App Amyloid beta (A4) precursor protein 2.69

Pdcd6ip Programmed cell death 6 interacting protein 2.67

Vcp Valosin containing protein 2.55

Catabolic processes 9.0E−03

Rnpep Arginyl aminopeptidase (aminopeptidase B) 3.29

Lgmn Legumain 3.18

Hltf Helicase-like transcription factor 2.81

Uba2 Ubiquitin-like modifier activating enzyme 2 2.67

Capn2 Calpain 2 2.60

Ube4a Ubiquitination factor E4A, UFD2 homolog 2.59

Vcp Valosin containing protein 2.55

Ctsl Cathepsin L 2.54

Fbxw8 F-box and WD-40 domain protein 8 2.50

Parn Poly(A)-specific ribonuclease ( deadenylation n uclease) 2.50

Chromosomes organization and modification 1.0E−02

Ruvbl1 RuvB-like protein 1 3.02

Actl6A Actin-like 6A 2.89

Hltf Helicase-like transcription factor 2.81

Hat1 Histone aminotransferase 1 Gene 2.77

Aspartate metabolic processes 1.2E−02 Thnsl1 Threonine synthase-like 1 3.25

Asns Asparagine synthetase 2.96

Copper ion homeostasis 1.6E−02 App Amyloid beta (A4) precursor protein 2.69

DNA metabolic processes/DNA repair 1.7E−02

Top2b Topoisomerase (DNA) II beta 3.55

Pcna Proliferating cell nuclear antigen 3.22

Ruvbl1 RuvB-like protein 1 3.02

Gtf2h4 General transcription factor II H, polypeptide 4 2.94

Ccdc47 Coiled-coil domain containing 47 2.77

Mcm5 Minich ro mo so me m aint enance deficie nt 5, celldivis ion cycle 2.57

Chtf18 CTF18, chromosome transmission fidelity factor 18 2.54

Macromolecular complex assembly 2.0E−02

Anxa5 Annexin A5 Gene 3.05

Atl3 Atlastin GTPase 3 2.91

Ipo8 Importin 8 2.81

Picalm Phosphatidylinositol b inding clathrin a ssembly p rotein 2.66

Cdh1 Cadherin-1E-Cad/CTF1E-Cad/CTF2E-Cad/CTF3 2.54

Carbohydrate metabolic processes 2.0E−02 G6pdh Glucose-6-phosphate dehydrogenase 3.44

Gbe1 Glucan (1,4-alpha-), branching enzyme 1 2.66

Cofactor metabolism 2.8E−02 G6pdh Glucose-6-phosphate dehydrogenase 3.44

Gstt2 Glutathione S-transferase, theta 2 2.86

RNA localization 3.3E−

02 Hnrnpa2b1 Heterogeneous nuclear ribonucleoprotein A2/B1 2.68Ribosome biogenesis 3.4E−02 Utp6 UTP6, s mall subunit ( SSU) process om e com pon en t, ho mo log 3.36

a Biological significance and cellular functionsfor theselectedgenes in bold areprovided in themain text.

conserved with transcriptomic and proteomic studies, despite

the differences in cell lines, culture conditions, microarrays tech-

nology and CHO annotation databases used. Of them, the geneVcp was previously reported to alter cellular growth after siRNA

functional engineering (Doolan et al., 2010) whereas two hetero-

geneous nuclear ribonucleoprotein genes (HnrnpcA/B= 2.88±0.22

andHnrnpa2b1A/B= 2.67±0.09)were identified from themostsig-

nificantly enriched annotation cluster, RNA processing, with the

lowest P -value (Table 2). Overexpressing these genes in CHO cells

could help maintain cellular homeostasis under stress conditions

(Hossain et al., 2007) and increase proliferation by binding to the

5UTRof the regulatorof gene expression c-myc and modulating its

translation(Kimet al., 2003). Inthesamecluster,thegeneUtp6 cod-

ingfor UTP6 small subunit processome component had the highest

translational efficiency ( A/B=3.36±0.25) and was also enriched in

the ribosome biogenesis annotation cluster. Overexpressing Utp6

in CHO cells would enhance cellular growth through an increase of

the number of ribosomes and an improvement of their efficiency

viaUtp6 role in the nucleolar processing of pre-18S ribosomal RNA

(Champion et al., 2008).




Fromsecond cluster of Table2, two genesPrc1 ( A/B= 3.17±0.23)

andRuvbl1 ( A/B=3.02±0.28) coding for protein regulatorof cytoki-

nesis 1 and RuvB-like protein 1, respectively, can be also suggested

for engineering the cell cycle of CHO cells and fostering the cel-

lular growth. Protein regulator of cytokinesis 1 is crucial for the

correct splitting of the chromosomes by binding to and regulat-

ing the microtubules, thereby ensuring that chromosome number

is maintained from one generation to the next during cytokine-

sis (Subramanian et al., 2010). The gene Ruvbl1 possesses ATPase

and helicase activities and has been recently implicated in sev-

eral cellular processes including cell cycle checkpoint activation,

DNA repair and replication, snoRNP assembly, telomere regulation,

centromere stability and chromosome segregation ( Jha and Dutta,

2009; Niewiarowski et al., 2010).

Lastly, we found the gene G6pdh from two functional clus-

ters, carbohydrate metabolic processes and cofactor metabolism,

as one of highly recommended targets for cell engineering strate-

gies aimed at enhancing the metabolic efficiency of cells, toward

higher cellular growth. G6pdh codes for the rate-limiting enzyme

involved in the pentose phosphate pathway, which generates a

major source of NADPH coenzyme (Tian et al., 1998). Thus, this

source of NADPH is likely essential for exponential cellular growth

by fueling anabolicreactions suchas lipidsynthesis, which is essen-

tial for cellular-membrane growth (Stanton, 2012).In summary, as the second and last biosynthesis step (after

transcription) along the gene expression process, translation has

consequently a decisive impact onthe formationof proteins by reg-

ulating translational efficiency. Hence, the current translatomics

platformcan guide us to better understanding of thesetranslational

control mechanisms, complementing translatomic and proteomic

strategy to identify potential cell engineering targets that support

growth on the basis of the assumption that the target genes are

highly and steadily translated to meet the cellular requirement

during the exponential growth phase.

3.4. Translational control mechanisms in CHO cells

The relationship between transcriptome and translatome datawasstudied in orderto explore translationalcontrol mechanisms in

exponentially growing CHO cells. Similarly to the previous statisti-

cal analysis of translatome data, the correlation between transcript

level and translation efficiency was investigated by selecting the

genes whose transcript levels showed a stable trend over the four

days of the exponential growth phase (relative standard devi-

ation < 10% over the four days). There were 6198 genes with a

constant level of transcript across the growth phase (Fig. 4A).

The overlap of these 6198 (transcriptome) and 4003 (translatome)

genes resulted in 1353 genes that displayed constant trends for

both their transcript level and translational efficiency. Each gene

was thencharacterized according to an average value of bothtrans-

criptome (absolute value) and translational efficiency (ratio A/B)

(Fig. 4B). In parallel, microarrays data were validated by qRT-PCR (Fig. 4C).

Arguably, there was no strong correlation between transcript

level and translation efficiency as indicated by the Pearson coeffi-

cient, which was−0.057. This correlation was further characterized

by classifying genes in nine different regions based on threshold of

high and low transcriptome and translatome (Fig. 4B). High and

low transcriptome were defined as the average TCav plus or minus

one standard deviation ( ) calculated for the 1353 genes (TCav +

and TCav − ) while high and low translatome were defined upon

a 1.5-fold change. Based on these criteria, genes were classified

in coupled or uncoupled groups (Fig. 4B). Transcript level and

translation efficiency were considered as coupled if both were sig-

nificantly high or low in positive correlation (blue; supplementary

Table 1). They were instead regarded as uncoupled if (a) both were

significantly high or low but in negative correlation (red; supple-

mentary Table 2), (b) only translational efficiency was significantly

high or low (green; supplementary Table 3) and (c) only transcript

level was significantly high or low (yellow; supplementary Table

4). The overall degree of uncoupling (95% – green, yellow and red

zones) between transcript level and translational efficiency was

higher than the degree of coupling (5% – blue zone).

Of interest, genes in the blue region were significantly enriched

in cellular ion (metal/calcium) homeostasis, signal transduction,

nucleotides biosynthesis and regulation of phosphorylation. On

the other hand, the 657 genes with uncoupled transcript level

and translation efficiency (green, yellow and red regions) were

mainly enriched in protein transport, localization and complex

assembly as well as in the generation of precursor metabolites and

energy. Although this analysis was conducted based on a subset

of genes, it was highly consistent with other studies (Halbeisen

and Gerber, 2009; Preiss et al., 2003; Tebaldi et al., 2012), presum-

ably representing the characteristics of the whole CHO cell gene

expression and indicating the complex regulation of gene expres-

sion through the involvement of translational control mechanisms

(Pradet-Balade et al., 2001).

Comparison of transcriptome and translatome data could also

be a valuablestrategyto devise adequate cellengineeringstrategies

in future, by identifying potential bottleneck along the geneexpres-sion pathway. For example, the previously highlighted key growth

genes, Hnrnpc , Utp6, Pcna, Vcp andMcm5 with a high and constant

translational efficiency ( A/B> 1.5) were present in the green region

(green marker with black line; Fig. 4B) where transcripts level and

translational efficiency were uncoupled. In this case, translational

efficiency was highand thereforeunlikely to representa bottleneck

along the gene expression process. Instead, moderate transcript

level could be limiting the flow of genetic information available

for translation. We support the idea that overexpression of such

genes leading to increased level of mRNA (shift toward the blue

region) could be an appropriate strategy for cell engineering in

which the increase in mRNA level would allow for the synthesis

of more proteins with the high translational efficiency.

Interestingly, the correlation analysis also implied the involve-ment of the mTOR pathway on translation control in CHO cells.

The yellowregion of high transcript level and average translational

efficiency (ratio A/B∼1) appeared to be enriched with 20 differ-

ent ribosomal proteins (yellow marker with black line; Fig. 4B).

Ribosomal proteins are known to be 5TOP mRNA translationally

regulated by the mTOR pathway (Meyuhas, 2000) which is a mas-

ter regulator of translational activity conserved across species (Ma

and Blenis, 2009). Recent studies of the mTOR pathway in CHO

cells demonstrated the strong potential of this pathway to alter

growth and productivity (Dreesen and Fussenegger, 2011; Lee and

Lee, 2012). Therefore further investigation of the mTOR pathway is

required to understand the impact of translational control in CHO

cells.

4. Conclusions

In this study, a translatomic platform has been applied to

CHO cells cultures as a novel strategy to identify and priori-

tize high potential candidate-genes for improving cellular growth.

Future work such as performing cell engineering validation on the

potential targets and adding proteomics datato complement trans-

criptome and translatome will further enhance our understanding

of mammalian cell culture. Moreover, we expect translatome data

to foster future work related to translational efficiency in CHO

cells. For example, preferential codon usage affecting translational

efficiency (Massaer et al., 2001; Navon and Pilpel, 2011; Wang

et al., 2006) will be ideally complemented and assessed by trans-

latome data. In addition, the specific translational efficiency of



http://refhub.elsevier.com/S0168-1656(13)00306-4/sbref0210








































































































































































































































































































































































































































































































































































































































































































































































































































































































Nissom, 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., others, 2006. Transcriptome and proteome profiling tounderstanding the biology of high productivity CHO cells. Molecular Biotech-nology 34 (2), 125–140.

O’Callaghan, P., James, D., 2008. Systems biotechnology of mammaliancell factories.Briefings in Functional Genomics and Proteomics 7 (2), 95–110.

Omasa, T., Onitsuka, M., Kim, W.D., 2010. Cell engineering and cultivation of chi-nese hamster ovary (CHO) cells. Current Pharmaceutical Biotechnology 11 (3),233–240.

Picard, F.,Milhem,H., Loubiere,P., Laurent, B.,Cocaign-Bousquet,M., Girbal, L.,2012.Bacterialtranslationalregulations:high diversitybetweenall mRNAsand major

role in gene expression. BMC Genomics 13, 528.Pradet-Balade, B., Boulmé, F., Beug, H., Müllner, E., Garcia-Sanz, J., 2001. Transla-

tion control: bridging the gap between genomics and proteomics? Trends inBiochemical Sciences 26 (4), 225–229.

Preiss, T., Baron-Benhamou, J., Ansorge, W., Hentze, M., 2003. Homodirectionalchanges in transcriptome composition and mRNA translation induced byrapamycin and heat shock. Nature Structural & Molecular Biology 10 (12),1039–1047.

Sampath, P., Pritchard, D.K., Pabon, L., Reinecke, H., Schwartz, S.M., Morris, D.R.,Murry, C.E., 2008. A hierarchical network controls protein translation duringmurine embryonic stem cell self-renewal and differentiation. Cell Stem Cell 2(5), 448–460.

Selvarasu, S., Ho,Y.S., Chong, W.P., Wong, N.S., Yusufi, F.N., Lee, Y.Y., Yap,M.G., Lee,D.Y., 2012. Combined in silico modeling and metabolomics analysis to charac-terize fed-batch CHO cell culture. Biotechnology and Bioengineering 109 (6),1415–1429.

Stanton,R.C., 2012. Glucose-6-phosphatedehydrogenase, NADPH, and cell survival.IUBMB Life 64 (5), 362–369.

Subramanian, R., Wilson-Kubalek,E.M., Arthur, C.P.,Bick, M.J.,Campbell, E.A.,Darst,

S.A., Milligan, R.A., Kapoor, T.M., 2010. Insights into antiparallel microtubule

crosslinking by PRC1, a conserved nonmotor microtubule binding protein. Cell142 (3), 433–443.

Tebaldi, T., Re, A., Viero, G., Pegoretti, I ., Passerini, A., Blanzieri, E., Quattrone,A., 2012. Widespread uncoupling between transcriptome and translatomevariations after a stimulus in mammalian cells. BMC Genomics 13,220.

Thomas, J.D.,Johannes, G.J., 2007. Identificationof mRNAsthat continueto associatewith polysomesduring hypoxia. RNA 13 (7), 1116–1131.

Tian, Q., Stepaniants, S.B., Mao, M., Weng, L., Feetham, M.C., Doyle, M.J., Yi, E.C., Dai,H.,Thorsson, V.,Eng, J.,others,2004. Integratedgenomic andproteomic analysesof gene expression in Mammalian cells.Molecular & Cellular Proteomics 3 (10),

960–969.Tian, W.N., Braunstein, L.D., Pang, J., Stuhlmeier, K.M., Xi, Q.C., Tian, X., Stanton,

R.C., 1998. Importance of glucose-6-phosphate dehydrogenase activity for cellgrowth. Journal of BiologicalChemistry 273 (17), 10609–10617.

Wang,Z., Gerstein,M., Snyder,M., 2009. RNA-Seq:a revolutionarytool fortranscrip-tomics. Nature reviews. Genetics 10 (1), 57–63.

Wang, Z.Y., Tian, S.F., Wang, X.P., Wang, S.F., Li, R.Q., Chu, Y.L., Ruan, L.,2006. Codonusage of Chinese hamster ovary cells.Zhonghua Shi YanHe LinChuang Bing DuXue ZaZhi20 (3),266–269.

Wong, D.C., Wong, K.T., Lee, Y.Y., Morin,P.N., Heng, C.K., Yap, M.G., 2006. Transcrip-tionalprofilingof apoptoticpathways in batch andfed-batch CHO cell cultures.Biotechnology and Bioengineering 94 (2), 373–382.

Wurm, F., 2004. Productionof recombinant protein therapeutics in cultivatedmam-malian cells. NatureBiotechnology 22 (11), 1393–1398.

Yee, J.C.,de Leon Gatti,M., Philp, R.J.,Yap, M., Hu,W.S., 2008.Genomicand proteomicexploration of CHO and hybridoma cells under sodium butyrate treatment.Biotechnology and Bioengineering 99 (5), 1186–1204.

Yee,J.C., Gerdtzen,Z.P.,Hu, W.S., 2009.Comparativetranscriptomeanalysisto unveilgenes affecting recombinant protein productivityin mammalian cells. Biotech-

nology and Bioengineering 102 (1), 246–263.







































































































































































































































































































































































































































































translatome analysis of cho cells

Documents