ORIGINAL ARTICLE

Comparative expression analysis of isolated humanadipocytes and the human adipose cell lines LiSa-2and PAZ6

EA van Beek1, AH Bakker2, PM Kruyt3, C Vink1, WH Saris2, NLW Franssen-van Hal1 and J Keijer1

1RIKILTFInstitute of Food Safety, Wageningen UR, Wageningen, The Netherlands; 2Maastricht University, Maastricht,The Netherlands and 3Gelderse Vallei Hospital, Ede, The Netherlands

Objective: To obtain insight in the extent to which the human cell lines LiSa-2 and PAZ6 resemble isolated primary humanadipocytes.Design: A combination of cDNA subtraction (representative difference analysis; RDA) and cDNA microarray analysis was usedto select adipose specific genes to compare isolated (pre-)adipocytes with (un)differentiated LiSa-2 and PAZ6 cells.Measurements: RDA was performed on adipose tissue against lung tissue. A total of 1400 isolated genes were sequenced andcDNA microarray technology was used for further adipose related gene selection. 30 genes that were found to be enrichedin adipose tissue were used to compare isolated human adipocytes and LiSa-2 and PAZ6 cells in the differentiated andundifferentiated states.Results: RDA and microarray analysis resulted in the identification of adipose enriched genes, but not in adipose specific genes.Of the 30 most differentially expressed genes, as expected, most were related to lipid metabolism. The second categoryconsisted of methyltransferases, DNMT1, DNMT3a, RNMT and SHMT2, of which the expression was differentiation dependentand higher in differentiated adipocytes. Using the 30 adipose expressed genes, it was found that isolated adipocytes on onehand, and PAZ6 and LiSa-2 adipocytes on the other, differ primarily in lipid metabolism. Furthermore, LiSa-2 cells seem to bemore similar to isolated adipocytes than PAZ6 cells.Conclusion: The LiSa-2 cell line is a good model for differentiated adipocytes, although one should keep in mind that the lipidmetabolism in these cells deviates from the in vivo situation Furthermore, our results imply that methylation may havean important function in terminal adipocyte differentiation.

International Journal of Obesity (2008) 32, 912–921; doi:10.1038/ijo.2008.10; published online 19 February 2008

Keywords: adipose tissue; adipocyte; cDNA microarray; stromal vascular cells; subtractive hybridization; RDA

Introduction

Adipose tissue is a dispersed tissue that has many functions

in the human body. Its primary function is the storage of

lipids. Its capacity to expand is essential to maintain healthy

levels of circulating lipids in times of abundant food supply

and provides a source of energy in times of need. Adipose

tissue was shown to be essential for glucose and lipid

homeostasis (reviewed in Waki and Tontonoz1), since too

little2,3 as well as too much adipose tissue4 results in insulin

resistance and dyslipidemia. Therefore adipose tissue and

adipocytes are subjected to studies to understand the

mechanisms of the development of obesity-associated

diseases.

A widely used approach to study adipocyte physiology is

to use cell lines. The most widely used cell lines are 3T3-L1

and its close relatives. 3T3-L1 is a mouse fibroblast cell line

that can be differentiated in vitro. Common adipogenic

markers like FABP4 and adiponectin were first identified in

3T3F42a5,6 and the importance of these molecules has been

shown in mice and human. Although the use of 3T3-L1 has

been of great importance for the understanding of adipose

physiology, not all mechanisms identified in 3T3-L1 and its

relatives, or for that matter in mice are the same in humans.

Therefore for improved extrapolation to humans it would be

worthwhile to study human cell lines. However, only few

human adipose cell lines exist. One cell line is PAZ6.7 This is

a human brown fat derived adipose cell line that can beReceived 5 September 2007; revised 30 November 2007; accepted 10 January

2008; published online 19 February 2008

Correspondence: Dr J Keijer, Food Bioactives, RIKILTFInstitute of Food Safety,

PO Box 230, Wageningen 6700AE, The Netherlands.

E-mail: jaap.keijer@wur.nl

International Journal of Obesity (2008) 32, 912–921& 2008 Nature Publishing Group All rights reserved 0307-0565/08 $30.00

www.nature.com/ijo

differentiated from pre-adipocyte to mature adipocytes.

Previously the PAZ6 cell line was used as a model for human

white adipocytes8–10 and brown adipocytes11,12 to study the

effect of hormones and other compounds on adipocyte

physiology, function and gene expression. Another cell line

is the liposarcoma cell line LiSa-213, that can also be

differentiated into mature adipocytes. Only a few studies

were performed using LiSa-2. These studies were focused on

the effect of HIV-protease inhibitors14 and on catalase

function.15 Both cell lines are poorly characterized and it

has not been established to what extent the undifferentiated

cells resemble stromal vascular cells (SVC) or differentiated

cells resemble isolated human adipocytes.

In this study, we aimed to gain insight in the extent

to which the human cell lines LiSa-2 and PAZ6 resemble

human adipocytes in the in vivo situation, using adipose tissue

properties. Therefore, we studied the expression levels of 30

adipose specific genes in isolated cells and in the undiffer-

entiated and differentiated cell lines. Isolation of adipose

tissue specific genes was attempted by using a combination of

subtractive hybridization and microarray analysis.

Subtractive hybridization, also named representation

difference analysis (RDA),16 is a method to enrich a target

mRNA population, in this case mRNA from human sub-

cutaneous adipose tissue, with low abundant and specific

sequences, by removing those sequences that are also

present in a second population, in this case lung mRNA.

The resulting cDNA library was sequenced. The success of the

subtraction procedure was assessed by spotting of the cDNA

inserts on a microarray and by hybridizing this cDNA

microarray with mRNA from different tissues, including

adipose tissue and lung. By hybridization with RNA from

adipose tissue and colon and lung tissue, 30 adipose

enriched genes were selected. These were used for compara-

tive analysis of isolated (pre-)adipocytes and (un)differen-

tiated cell lines using cDNA microarray technology.

Materials and methods

Tissue and cells

Subcutaneous adipose tissue was taken from eight morbid

obese women (age 36±4; BMI 44.6±4.7) who underwent

gastro plastic surgery. This was used for the FAT cDNA library

construction. Colon biopsies, subcutaneous and omental

adipose biopsies were taken from otherwise healthy subjects

diagnosed with colon cancer undergoing abdominal surgery

(colectomy). (Patient 1: male 70, BMI 32; pat 2: female 83,

BMI 34; pat 3: female 74, BMI 27.) The samples were either

homogenized in liquid N2 prior to RNA isolation, or were

used for (pre-)adipocyte isolation and subsequent RNA

isolation using TRIzol reagent (Invitrogen, Breda, The

Netherlands) as described previously.17 Isolation of (pre-)

adipocytes was performed according to Rodbell18 with minor

modifications as previously described.19 Fresh adipose

tissue samples (10 g) were cut into small pieces and treated

with 1 mg ml�1 collagenase type II (Sigma, Zwijndrecht, The

Netherlands) in Dulbecco’s modified Eagle’s medium

(DMEM)/HAM’s F12 (Invitrogen) with 4% BSA for 2 h

at 37 1C, rotating slowly.

The brown pre-adipocyte cell line PAZ6, kindly obtained

from Professor Dr D Strossberg7 and the white pre-adipocyte

cell line LiSa-2, kindly obtained from Professor Dr P Moller.13

were cultured as described.7 The cells were grown in DMEM/

Ham’s F12 (Invitrogen) with 2% fetal calf serum (FCS;

Bodinco, Alkmaar, The Netherlands), 100 U ml�1 penicillin

and 0.1 mg ml�1 streptomycin. To differentiate the cells,

differentiation medium was added at full confluency. Differen-

tiation medium is DMEM/Ham’s F12 supplemented with

15 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid

(HEPES), 1mM dexamethasone, 850nM insulin, 33mM biotin,

17mM pantothenate, 1 nM triiodothyronine, 10mM carbacyclin,

100U ml�1 penicillin, 0.1mg ml�1 streptomycin and 2% FCS.

During differentiation the medium was changed every other

day. Cells were cultured at 37 1C in a 95% air/5% CO2

atmosphere. Differentiation was assessed by Oil red-O staining

and differentiated cells were harvested after 9 days of initiation

of differentiation. RNA isolation of the PAZ6 and LiSa-2 cells

was performed as described previously.17

cDNA library and array construction

An adipocyte specific cDNA library (FAT library) was

obtained by subtractive hybridization, in which lung RNA

(Stratagene, Amsterdam, Netherlands; driver) was subtracted

from DNase treated RNA isolated from pooled subcutaneous

adipose tissue obtained from eight obese women (tester).

Lung was chosen as nonadipose tissue. cDNA was made

using the SMART PCR cDNA Synthesis kit (Clontech, Palo

Alto, CA, USA). Subtraction was done using the PCR-Select

cDNA Subtraction kit (Clontech) according the protocol of

the supplier. Prior to cloning, AmpliTaq (PerkinElmer,

Groningen, The Netherlands) was used to create an

A-overhang on both ends of the PCR products. The obtained

PCR products were ligated into pGEM-T easy (Promega,

Leiden, The Netherlands) and cloned in E. coli XL2-Blue

ultra competent cells (Stratagene). Cells were plated on

LB-agar plates containing 100 mg ml�1 ampicillin, 80 mg ml�1

X-gal and 0.5 mM IPTG and grown o/n at 37 1C. White

colonies were selected by a colony picker (Flexys), grown in

200 ml ‘freezing LB medium’ consisting of 200 ml LB-medium

with 36 mM K2HPO4, 13.2 mM KH2PO4, 1.7 mM natriumci-

trate, 0.4 mM MgSO4 �7H2O, 6.8 mM ammoniumsulphate,

4.4% (v/v) glycerol and 100mg ml�1 ampicillin in 96-well

plates and stored at �80 1C. For microarray construction,

cDNA inserts were amplified by PCR on glycerolstocks as

described by Franssen-van Hal et al.20 To determine the length

of the PCR products 1 ml of each reaction was run on 1%

agarose/TBE gels. Amplified products were purified using

L-50 fine Sephadex according to the manual. After precipita-

tion, the pellet was taken up in 12 ml 5� SSC of which 10 ml

Comparative expression analysis of adipocytesEA van Beek et al

913

International Journal of Obesity

was used for printing on silylated slides (CELAssociates,

Houston, TX, USA) using a PixSys 7500 arrayer (Cartesian

Technologies, Durham, NC, USA), as described.21 The array

consisted of a total of 3136 spots consisting of 2433

sequences originated from the adipose tissue enriched FAT

cDNA library, 280 selected named cDNAs in duplicate, 47

control luciferase clones (30, middle, 50 and full length)

to check cDNA synthesis efficiency, 2 Salmonella clones

(negative control), and 94 spots without DNA. This array was

used to characterize the FAT cDNA library sequences.

RNA labeling and microarray hybridization

RNA labeling for microarray hybridization was performed as

described previously.22 20 mg total RNA was labeled by

incorporation of Cy5 labeled dCTP in the cDNA. A standard

reference sample, consisting of a pool of total RNA from all

samples was used and labeled with Cy3 monofunctional dye.

Prior to hybridization, the Cy3- and Cy5-labelled samples

were mixed 1:1 (v/v). Prehybridization of the arrays was

performed as previously described.23 A gene frame, 75 ml

(Westburg, Leusden, The Netherlands) was used during an

overnight hybridization at 42 1C in a humid hybridization

chamber. All samples were hybridized in duplicate. After

hybridization the arrays were washed and dried as described

before.23 Scans with a pixel resolution of 10mm, a laser power

of 90%, PMT voltage of 60% for Cy3 scans and 65% for Cy5

scans, were obtained using a laser scanner (ScanArray

ExpressHT, PerkinElmer).

Sequencing and selection of uncharacterized cDNAs

To get insight in the composition of the adipose enriched

FAT cDNA library 1411 clones were selected for sequencing.

288 Randomly selected cDNAs and all genes with a

hybridization signal higher than 3� the background when

hybridized with RNA from PAZ6 cells (1123), were selected

for sequencing. Sequence analyses were performed by

Greenomics, Plant Research International, Wageningen,

The Netherlands. The obtained sequences were blasted

against the GenBank database. Redundant cDNAs of the

FAT library were removed to prevent possible influences of

over-represented sequences on data normalization. This was

done using similarity analysis in the BLASTCLUST program

in BLAST package, which clusters the input sequences based

on the pair wise sequence similarity. The sequences with

50% or higher similarity were considered in a same cluster,

and the longest sequence in each cluster was used to

represent the cluster. Clones from which the sequence

analysis failed and clones with an insert length less than

200 base pairs were removed. To obtain a set of genes suitable

for further analysis of hybridization experiments, a max-

imum of two clones representing the same gene were

selected from different clusters. In this case these cDNAs

represent different parts of the same gene. Representative

clones were checked to have hybridization signals represen-

tative for the whole cluster. All genes identified during RDA,

were grouped according to gene ontology molecular func-

tion using eGON.24

The 387 resulting representative clones were printed on a

new human cDNA array, also containing cDNAs that were

selected in a similar manner from cDNA libraries of intestinal

origin (human colon cancer cell lines CaCo-2 and HT-29 and

human small intestine, all constructed with HeLa as driver).

The array was constructed as described above and consisted

of a total of 3153 spots consisting of 387 representative

sequences originated from the adipose tissue enriched FAT

cDNA library (FAT), 2456 representative clones originating

from the intestinal libraries (INT), 298 selected named

cDNAs (CO). 93 of the named genes were related to obesity.

Furthermore, two control luciferase clones to check cDNA

synthesis efficiency, two spots containing the gene coding

for granulated starch and eight spots only SSC. This array was

used in all experiments described, except sequence charac-

terization.

Data normalization and analysis

The software package ArrayVision (GE Healthcare, Diegem,

Belgium) was used to collect average spot intensities and

background values of the arrays and data were further

processed in Microsoft Excel. All spots with a cy3 of cy5

signal lower than twice the average background were

excluded from the dataset before data analysis. The data

normalization was performed exactly as described pre-

viously;22 briefly: in this experimental design a reference

sample is used for each slide, thereby making it possible to

use multiple slides in a single experiment. The data normal-

ization consists of two corrections; first, the hybridization

signal is corrected for fluctuations in the amount of DNA

spotted and for variations in the hybridization conditions

within a slide. Therefore the cy5 signal on spot a on slide A is

multiplied by the median of the cy3 levels on spot a of all

slides divided by the corresponding cy3 signal on spot a on

slide A. Second, the hybridization levels are also corrected for

differences in labeling efficiency between samples and for

differences in amount of sample RNA used in the labeling

reaction. This correction 2 is performed by multiplying the

hybridization signals obtained after correction 1 by the

median of the median cy5 signals of all hybridized slides

after correction 1 divided by the median of the cy5 signals of

all spots on slide A after correction 1.

Dot plots of duplicate arrays were made in Microsoft Excel

and outliers due to staining were excluded from further

analysis. Discriminant analyses were performed using Gene-

maths XT 1.02 (Applied Maths, St Martens Lathem, Belgium).

A Student’s t-test was performed using Microsoft Excel.

Results

A combination of subtractive hybridization and microarray

analysis was used to obtain adipose specific genes.

Comparative expression analysis of adipocytesEA van Beek et al

914

International Journal of Obesity

Subtractive hybridization, or RDA, should result in selection

of genes present in one source over the other (enrichment),

with all genes being present in the same frequency (normal-

ization). A cDNA library, the FAT library, of mRNAs selected

for transcription in adipose tissue and not in lung tissue was

obtained. The effectiveness of the procedure was investigated

by printing the cDNAs on a microarray slide, hybridizing this

to adipocyte RNA and sequencing of the cDNAs correspond-

ing to transcripts which give an average signal of three times

above the background (1123 clones), as well as 288

additional, randomly selected clones. In total 1248 cDNAs

were successfully sequenced and their frequency was deter-

mined to assess the extent of normalization (Figure 1). The

majority of genes, 88%, were present once or twice on the

array, while 4% of the genes were present in multiple copies,

that is, occurred 20 till 250 times. The genes that are present

most frequently are listed in Table 1. Several genes with an

important role in adipose physiology, such as perilipin,

FABP4, CD36, SCD, LPL and vimentin, are present between

3 and 20 times, while this category seems to be lacking

among the more abundant genes.

To obtain functional insight in the cloned genes obtained

by subtraction, first a selection was made in which each gene

was presented once. Of 295 genes, 211 could be associated to

a unigene cluster and 152 are genes with GO annotations for

molecular function (Figure 2). Most genes were categorized

to binding (115), protein binding (62), ion binding (39),

nucleic acid binding (30), nucleotide binding (22) and lipid

binding (8). The other represented categories are catalytic

activity (56), structural molecule activity (25), transporter

activity (14), enzyme regulator activity (9), signal transducer

activity (16), transcription regulator activity (6), translation

regulator activity (6), antioxidant activity (3) and motor

activity (2).

To establish to which extent the FAT library is specific for

fat, a new microarray was constructed that was subsequently

used for hybridization to RNA from adipose and nonadipose

tissues. A total of 387 sequenced FAT cDNA clones representing

all unique sequences were printed on this microarrray, next

to cDNA clones obtained form subtractions of human small

intestine and human colon cell lines and named cDNAs,

corresponding to proteins with a known function. This

human microarray contained 3141 sequences (3153 includ-

ing empty spots, luciferase and starch), of which 12%

originated from the FAT library (Figure 3a), 10% were genes

with a possible regulatory function in adipose tissue or colon

tissue and 78% originated from the intestinal libraries. This

microarray was then hybridized with RNAs from subcuta-

neous adipose tissue biopsies, omental adipose tissue

biopsies, isolated omental adipocytes or colon biopsies as

well as the lung RNA that was used as a driver in the

subtractive hybridization. Genes with an average expression

value of higher than two times the background were used for

further analysis. To assess the extent of enrichment and to

select possible adipose specific genes, the 30 most discrimi-

nating genes were selected from the following comparisons

using supervised principal component analysis: discriminating

between adipose tissue and nonadipose tissue (lung and

colon) with higher mRNA levels in adipose tissue; discrimi-

nating between adipose tissue and lung with higher mRNA

levels in lung tissue. The origins of these clones were then

determined (Figure 3). The genes highly expressed in adipose

tissue and discriminating between adipose tissue and lung

have a distribution (Figure 3b) that is very different from the

distribution of all genes spotted on the array (Figure 3a). 56%

originates from the FAT cDNA library, while this selection

represents 12% of the array. Also the named genes selected

for their possible function in adipose tissue increases from

3% (distribution of all genes spotted on the array) to 10%

(genes higher expressed in adipose tissue). If the genes

discriminating between adipose and lung with higher mRNA

levels in lung were selected, only one gene (3%) remains that

originates from the FAT cDNA library (Figure 3c). When the

most discriminating genes between colon and lung were

selected for comparison, the distribution resembles the

distribution of all genes spotted on the array, albeit with a

slight increase in the relative amount of genes originating

from the FAT cDNA library (Figure 3d).

The 30 most discriminating adipose genes (Figure 3b) were

grouped according to function (Table 2). Most have a

function in lipid metabolism (13 out of 30). Four genes have

methyltransferase activity. The remaining genes function in

the regulation of transcription (4 out of 30), cell structure (2

out of 30) or have other functions (7 out of 30). The ratio of

expression in adipose tissue versus nonadipose tissue varies

from 26.3 fold for fatty acid binding protein 4 (FABP4) to 1.5

fold for acyl-CoA synthetase long-chain family member 1

(ACSL1; Table 2). The four genes with methyltransferase

activity have a 1.8 fold higher expression in adipose tissue.

All genes were individually tested for statistical significance

and all P-values were found to be smaller than 0.05 and most

smaller than 0.001 (Table 2).

To assess the true adipocyte origin of these genes, their

expression in adipocytes and pre-adipocytes was determined.

All 30 genes are indeed expressed above two times theFigure 1 Frequency distribution of genes identified in the FAT library.

The FAT library is described in the Materials and methods section.

Comparative expression analysis of adipocytesEA van Beek et al

915

International Journal of Obesity

background in isolated human primary subcutaneous

adipocytes. Next, we investigated whether these 30 genes

are also expressed in the cell lines PAZ6 and LiSa-2. In Table 3

the expression levels (%) are given relative to their expres-

sion in isolated adipocytes (100%).

Of the 30 genes, 9 have at least a 50% lower expression in

mature differentiated LiSa-2 and 12 in PAZ6, as compared to

isolated adipocytes. Remarkably, most of these genes func-

tionally belong to lipid metabolism. In fact, of the genes

belonging to lipid metabolism, only three (FABP4, SCD and

ALOX5) have a high expression in both differentiated cell

lines and three genes (LPL, CD36 and ACSL1) have high

expression in differentiated LiSa-2 cells, but a low expression

in PAZ6 cells. SCD and CD36 are particularly highly

expressed in differentiated LiSa-2 cells (2- and 1.5-fold

higher than in adipocytes).

The methyltransferases are all expressed to the same extent

in both differentiated cell lines as compared to adipocytes.

This is also true for all genes in the regulation of transcrip-

tion category, except for early B-cell factor (EBF) whose

expression is twofold less in differentiated PAZ6 cells. Of the

other genes, three, VIM, TGFBR3 and SLC19A2, show

a higher expression in differentiated LiSa-2 cells, compared

to isolated adipocytes.

Table 1 Genes overabundant represented in the FAT library

Unigene gene Accession number Unigene title Times identified

MALAT1 NR_002819 Metastasis associated lung adenocarcinoma transcript 1 233

CALM2 NM_001743 Calmodulin 2 (phosphorylase kinase, delta) 78

SAA1 NM_000331 Homo sapiens serum amyloid A1 57

NC_001807 Mitochondrion, complete genome 55

FTL NM_000146 Ferritin L chain mRNA, complete cds 54

TRBVOR9@ NG_001337.1 T-cell receptor beta variable orphans on chromosome 9 32

EEF1A1 NM_001402 Eukaryotic translation elongation factor 1 alpha 1 31

AF254896 AF254896 Cytochrome b gene 24

DCN NM_133507 Homo sapiens decorin 24

PLIN NM_002666 Perilipin 19

FABP4 NM_001442 Fatty acid binding protein 4, adipocyte 18

FHL1 NM_001449 Four and a half LIM domains 1 17

HBB NM_000518 Hemoglobin, beta 16

MYL6 NM_021019 Myosin, light polypeptide 6, alkali, smooth muscle and

nonmuscle

12

M10098 18S rRNA gene 10

B2M NM_004048 Beta-2-microglobulin 10

CD36 NM_000072 CD36 antigen (collagen type I receptor, thrombospondin

receptor)

8

SCD NM_005063 Stearoyl-CoA desaturase (delta-9-desaturase) 8

ANXA1 NM_000700 Annexin A1 7

ANXA2 NM_004039 Annexin A2 7

LPL NM_000237 Lipoprotein lipase 7

RPL6 NM_000970 Ribosomal protein L6 transcript variant 2 7

SPARC NM_003118 Secreted protein, acidic, cysteine-rich (osteonectin) 7

ACTB NM_001101 Actin, beta 5

RPS6 NM_001010 Ribosomal protein S6 4

AK000624 cDNA FLJ20617 fis, clone KAT05223 3

ADH1B NM_000668 Alcohol dehydrogenase IB (class I), beta polypeptide 3

SEPP1 NM_005410 Selenoprotein P, plasma, 1 3

TXNIP NM_006472 Thioredoxin interacting protein 3

VIM NM_003380 Vimentin 3

YWHAZ NM_003406 Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase

activation protein, zeta polypeptide

3

Figure 2 Molecular function of unique genes identified in the FAT library.

Unique is used to indicate single representation. Activity is abbreviated as act.

Comparative expression analysis of adipocytesEA van Beek et al

916

International Journal of Obesity

To see whether the expression of these 30 genes is

related to adipogenesis and to further characterize the cell

lines, we also analyzed the expression of these genes in

undifferentiated LiSa-2 and PAZ6 cells and in the SVC

fraction of adipose tissue. Using these data we can also

observe similarities and differences between SVC and LiSa-1

and PAZ6 pre-adipocytes and shed light on some differences

between in vivo and in vitro differentiation. SCD, FABP4, the

methyltransferases DNMT3A, DNMT1 and RNMT, RAR�,

and MMP1 display differentiation related expression; they

are well expressed in isolated adipocytes and the differen-

tiated cell lines, while these genes are markedly lower

expressed in SVC. Three other genes, DDIT3, VIM and

SLC19A2 behave the same in primary cells and the both cell

lines and display similar expression in differentiated and

undifferentiated conditions. Lisa-2 seems more similar to the

primary cells than PAZ6 as indicated by similar expression of

LPL, CD36, SHMT2, EBF, SUI1 and TIGA1, while GPX1 is

similarly expressed in the primary cells and PAZ6. Some

genes (SAA1, PLIN, APOD, KIAA1560, ALDH2 and

C10orf116) are regulated during differentiation in vivo,

however they are not regulated during adipogenesis in both

cell lines.

Discussion

Isolated primary adipocytes as well as cell lines are widely

used for mechanistic studies on adipose physiology.

Although specific effects observed in cell lines have been

confirmed in primary adipocytes by targeted analysis,12,25,26

no profiling techniques have been used to examine the

extent to which adipocyte cell lines resemble primary

adipocytes. This is particularly true for the human situation.

In this study, we isolated adipose tissue genes by RDA and

studied the expression of the 30 most differential adipose

expressed genes in isolated adipocytes and two human

adipocyte cell lines LiSa-2 and PAZ6 using cDNA microarray

technology. We observed that differentiated LiSa-2 cells

resemble isolated adipocytes the most.

Differences in gene expression between isolated adipocytes

and the differentiated cell lines were observed particularly

in genes involved in fatty acid metabolism, compared to the

other categories (Table 3). This may be explained by several

reasons. Adipocytes differentiated in vitro are unable to

develop fully into unilocular adipocytes. While mature

white adipocytes have a single large lipid droplets, both

differentiated LiSa-2 cells and differentiated PAZ6 cells

display multiple small lipid droplets (unpublished observa-

tions). In part this may be due to increased buoyancy

accompanying lipid accumulation, resulting in detachment

of the cells from the bottom of the culture flask and

consequential loss of maturing cells (unpublished observa-

tions). A second reason may be in the genetic makeup of

LiSa-2 cells and PAZ6 cells. Both cell lines are immortal. LiSa-

2 is a cancer cell line and PAZ6 is immortalized using SV40

genes.7 The immortality may be the cause of the continuous

cell growth that is observed during differentiation (unpub-

lished observations). A third reason for a difference in lipid

metabolism related genes between cell lines and isolated

adipocytes is that the in vitro culture medium may lack

essential nutrients, signaling molecules or other factors that

are present in vivo.

Figure 3 Distribution of the origin of genes on the human cDNA array. (a) All genes on the human cDNA array. (b) Thirty most discriminating genes between lung

and adipose tissue, higher expressed in adipose tissue. (c) Thirty most discriminating genes between lung and adipose tissue, higher expressed in lung tissue.

(d) Thirty most discriminating genes between lung and colon tissue. FAT is adipose derived, INT is intestine derived and CO are named genes associated with obesity

(CO-FAT) or others processes (CO-g).

Comparative expression analysis of adipocytesEA van Beek et al

917

International Journal of Obesity

The 30 most discriminating genes between adipose tissue

and nonadipose tissue (lung and colon) were categorized

to function. As expected the majority has a function in fatty

acid metabolism. In our study the expression of the leptin

gene was at background levels.

The second category, comprised of four genes with

methyltransferase activity, was less anticipated. The role of

these methyltransferases in adipose tissue is not known.

Although the differences in expression levels between adipose

tissue versus nonadipose tissue are not high (between 1.8 and

3.3), the occurrence of four methyltransferases in the

selection and their differentiation related expression pattern

(Table 3) suggests a functional role and further examination

may be worthwhile. DNMT1 and DNMT3a are involved in

de novo DNA methylation. DNMT1 is associated with main-

tenance of methylation patterns during replication, while

DNMT3a is associated with developmental methylation

of, in particular, single copy genes.27 Interestingly, DNMT1

interacts with the retinoblastoma protein,28,29 which is an

important adipose developmental switch30 and has a role in

the regulation of energy expenditure.31 De novo DNA

methylation occurs during embryogenesis, but can also occur

in adult cells to alter the developmental program of the cell.32

It is hypothesized that DNA methylation is set up in a tissue-

specific manner, thereby silencing the genes opposing the

proper development and leaving essential genes unmethy-

lated and thereby helping to direct terminal differentia-

tion.33,34 Indeed, prevention of DNA methylation during

differentiation of mammary epithelial cells inhibits the

completion of acinar morphogenesis.35 Both DNMT136 and

DNMT3a32 have been implicated in terminal differentiation.

Therefore, we speculate that the high expression of methyl-

transferases in differentiated adipocytes may be involved in

(terminal) differentiation of adipocytes.

Table 2 Thirty most discriminating genes between adipose (AD) and nonadipose (non-AD) tissues, higher expressed in adipose tissue

Unigene gene Accession number Unigene title AD/non-AD P-value

Lipid metabolism

FABP4 NM_001442 Fatty acid binding protein 4, adipocyte 26.3 0.001

SAA1 NM_199161 Serum amyloid A1 12.7 0.04

PLIN NM_002666 Perilipin 6.3 o0.001

SCD NM_005063 Stearoyl-CoA desaturase (delta-9-desaturase) 6.0 0.016

ADH1B NM_000668 Alcohol dehydrogenase IB (class I), beta polypeptide 5.9 0.003

APOD NM_001647 Apolipoprotein D 4.5 0.004

LPL NM_000237 Lipoprotein lipase 4.0 0.001

KIAA1560 NM_020918 Glycerol 3-phosphate acyltransferase, mitochondrial 3.8 0.04

ANXA1 NM_000700 Annexin A1 3.3 0.003

ALDH2 NM_000690 Aldehyde dehydrogenase 2 family 2.9 o0.001

CD36 NM_000072 CD36 antigen (collagen type I receptor, thrombospondin receptor) 2.7 0.002

ALOX5 NM_000698 Arachidonate 5-lipoxygenase 2.2 0.002

ACSL1 NM_001995 Acyl-CoA synthetase long-chain family member 1 1.5 0.001

Methyltransferase activity

DNMT3A NM_175630 DNA (cytosine-5-)-methyltransferase 3 alpha 3.3 o0.001

DNMT1 NM_001379 DNA (cytosine-5-)-methyltransferase 1 2.9 o0.001

RNMT NM_003799 RNA (guanine-7-) methyltransferase 2.5 o0.001

SHMT2 NM_005412 Serine hydroxymethyltransferase 2 (mitochondrial) 1.8 o0.001

Regulation of transcription

RARB NM_016152 Retinoic acid receptor, beta 7.2 0.002

EBF NM_024007 Early B-cell factor 2.7 o0.001

SUI1 NM_005801 Putative translation initiation factor 2.0 o0.001

DDIT3 NM_004083 DNA-damage-inducible transcript 3 1.7 0.004

Cell structure

MMP1 NM_002421 Matrix metalloproteinase 1 (interstitial collagenase) 2.6 o0.001

VIM NM_003380 Vimentin 2.4 o0.001

Other

C10orf116 NM_006829 Homo sapiens chromosome 10 open reading frame 116 3.9 o0.001

TGFBR3 NM_003243 Transforming growth factor receptor beta 3 2.1 o0.001

GCK NM_033508 Glucokinase (hexokinase 4, maturity onset diabetes of the young 2),

transcript variant 3

2.0 0.001

RTN3 NM_201430 Reticulon 3 1.6 o0.001

GPX1 NM_201397 Glutathione peroxidase 1 1.6 o0.001

TIGA1 NM_053000 TIGA1 1.5 o0.001

SLC19A2 NM_006996 Solute carrier family 19 (thiamine transporter), member 2 1.3 0.01

Categorized by function. Relative expression in adipose tissue over non-adipose tissue (AD/non AD) and P-value (Student’s t-test) is given.

Comparative expression analysis of adipocytesEA van Beek et al

918

International Journal of Obesity

RNMT is required for the formation of the 50-terminal

mGpppN cap together with RNA 50-triphosphatase and

guanylyltransferase37 and capping is important for gene

expression, because it functions in mRNA processing,

stability and translation.38 This implies that differentiated

adipocytes are still metabolically active, as is widely accepted

nowadays.

Mitochondrial serine hydroxymethyltransferase, SHMT2,

functions in one carbon metabolism and in the interconver-

sion of serine and glycine. It is thought to function in the

regulation of cell growth and proliferation39 and was shown

to be under control of c-Myc, which has an important role

in mitochondrial biogenesis and the expression of nuclear

encoded mitochondrial genes.40 It is likely that higher

expression of SHMT2 in adipocytes is related to increased

mitochondrial density.

The relative small differences agree with previous stu-

dies41,42 but may in part also be due to the selection of genes

used, which was directed at white adipose characteristics and

not to identify differences between white and brown adipose

tissues.

LiSa-2 and PAZ6 are both used to study adipocyte

function,8–10 but have a different physiological background.

LiSa-2 is a white adipose cell line of liposarcoma origin,

while the PAZ6 cell line is an immortalized brown pre-

adipocyte cell line. White adipose tissue is specialized in

energy storage and brown adipose tissue has an important

function in the generation of heat and for this it expresses

Table 3 Thirty most adipose discriminating genes, their expression levels in isolated (pre-)adipocytes, LiSa-2 and PAZ6 cells and expression ratios between

adipocytes and pre-adipocytes

Unigene gene Adipocytes Pre-adipocytes Ratio

Isolated adipocytes LiSa-2 PAZ6 SVC LiSa-2 PAZ6 Isolated cells LiSa-2 PAZ6

Lipid metabolism

FABP4 100 80 97 43 7 7 2.3 11.4 13.9

SAA1 100 12 14 15 10 16 6.7 1.2 �1.1

PLIN 100 28 20 17 21 26 5.9 1.3 �1.3

SCD 100 215 101 15 34 33 6.7 6.3 3.1

ADH1B 100 7 7 99 6 6 1.0 1.2 1.2

APOD 100 24 24 242 24 22 �2.5 1.0 1.1

LPL 100 93 29 37 32 42 2.7 2.9 �1.4

KIAA1560 100 40 31 33 31 39 3.0 1.3 �1.3

ANXA1 100 48 74 166 85 43 �1.7 �1.7 1.7

ALDH2 100 19 20 45 23 26 2.2 �1.3 �1.3

CD36 100 165 57 88 89 50 1.1 1.9 1.1

ALOX5 100 89 87 84 61 64 1.2 1.5 1.4

ACSL1 100 84 59 79 58 75 1.3 1.4 �1.3

Methyltransferase activity

DNMT3A 100 81 71 56 32 40 1.8 2.5 1.8

DNMT1 100 93 92 55 36 48 1.8 2.6 1.9

RNMT 100 97 83 64 45 51 1.6 2.2 1.6

SHMT2 100 101 103 65 59 80 1.5 1.7 1.3

Regulation of transcription

RARB 100 113 90 41 31 24 2.4 3.6 3.8

EBF 100 91 46 92 81 40 1.1 1.1 1.2

SUI1 100 119 90 145 133 48 �1.4 �1.1 1.9

DDIT3 100 102 92 79 83 73 1.3 1.2 1.3

Cell structure

MMP1 100 90 91 59 53 50 1.7 1.7 1.8

VIM 100 152 84 143 195 77 �1.4 �1.3 1.1

Other

C10orf116 100 12 12 48 13 12 2.1 �1.1 1.0

TGFBR3 100 130 74 90 NA 52 1.1 NA 1.4

GCK 100 82 74 78 49 68 1.3 1.7 1.1

RTN3 100 87 78 79 69 83 1.3 1.3 �1.1

GPX1 100 61 88 84 80 98 1.2 �1.3 �1.1

TIGA1 100 82 64 165 101 91 �1.7 �1.3 �1.4

SLC19A2 100 147 85 92 81 90 1.1 1.8 �1.1

All expression levels (%) were set relative to that of isolated adipocytes (100%). Expression levels higher than 70% are shaded and ratios lower than �1.5 and higher

than 1.5 are shaded.

Comparative expression analysis of adipocytesEA van Beek et al

919

International Journal of Obesity

of the mitochondrial uncoupling protein UCP1.43 In view

of their origin it is not unexpected that LiSa-2 more

closely resembles primary adipocytes than PAZ6 (Table 3).

In line with their distinct origins, differences in gene

expression between the two cell lines can be observed,

but are not pronounced (Table 3). In particular SHMT2 is

higher expressed in undifferentiated PAZ6 cells and the

expression of SUI-1, a general eukaryotic transcription factor

involved in transcription initiation, differs (Table 3). The

relative small differences agree with previous studies23,44 but

may in part also be due to the selection of genes used, which

was directed at white adipose characteristics and not to

identify differences between white and brown adipose

tissues.

Nowadays, whole genome arrays are available and have

successfully been used for comparative analysis, for example,

of white and brown adipose tissue and muscle.44 We used a

more focused approach and applied RDA to select adipose

tissue expressed genes for further analysis. Theoretically the

RDA procedure both selects for specific genes and normalizes

their frequency. When the resulting FAT cDNA library was

analyzed, we found that most genes were present only once

or twice, but 31 genes were overrepresented. The goal of

normalization, resulting in a pool of cDNAs with each gene

being equally represented, was therefore not fully obtained.

This is in agreement with other studies,45,46 where normal-

ization was also not fully complete. Among the over-

represented genes, two categories could be distinguished,

the moderately overrepresented genes, predominantly genes

with a function in adipose tissue biology, and the highly

overrepresented genes, which are expressed to a similar

extent in lung and adipose tissue. The latter include

MALAT1, calmodulin2, ferritin, serum amyloid A and

mitochondrial genome hits. This shows that RDA is

particularly sensitive to sequences that are expressed in both

tissues that are used in the procedure. The results also imply

that random isolation of RDA-derived clones would have

resulted in a large number of false-positives, while a more

detailed analysis of unique sequences clearly shows that the

FAT cDNA library is enriched for adipose sequences (Figure 3).

The 30 most discriminant genes selected for further analysis

are not adipose specific genes as they are also expressed in

lung tissue,47 although to a lesser extent. In our hands RDA

was not effective in the isolation of tissue specific genes but,

in combination with microarray selection, was effective in

the isolation of adipose enriched sequences. Our results may

also indicate that genes exclusively expressed in adipocytes

do not exist.

However, as we only used 30 genes for expression analysis

from an in-house constructed microarray, further studies

using full genome microarrays should be used before we can

conclude that adipocyte specific genes do not exist. Further-

more, the use of full genome arrays may result in a more

distinct picture regarding the similarity and the use of the

LiSa-2 cell line as an alternative for the widely used murine

cell line 3T3-L1.

In this study, we used a selection of 30 adipose genes on

cDNA microarrays for comparative analysis of human

adipose cell lines to support their use for studies on adipose

physiology. Our analysis indicates that the cell lines are good

models for studying adipocytes physiology provided that

caution is taken with regard to lipid metabolism related

genes. This disadvantage may be overcome by adapting

culture conditions that more closely resemble the in vivo

situation.

Acknowledgements

We thank Professor Dr D Strossberg for providing the PAZ6

cell line and Professor Dr P Moller for providing the LiSa-2

cell line. The research reported in this paper was supported

by the ZonMW program Diet and Chronic diseases (VCZ

980-10-012) and by the Ministry of Agriculture, Nature

Management and Food Quality (803-71-54701)

References

1 Waki H, Tontonoz P. Endocrine functions of adipose tissue.doi:10.1146/annurev.pathol.2.010506.091859. Annu Rev Pathol2007; 2: 31–56.

2 Moitra J, Mason MM, Olive M, Krylov D, Gavrilova O, Marcus-Samuels B et al. Life without white fat: a transgenic mouse. GenesDev 1998; 12: 3168–3181.

3 Reitman ML, Arioglu E, Gavrilova O, Taylor SI. Lipoatrophyrevisited. Trends Endocrinol Metab 2000; 11: 410–416.

4 Rosen ED, Spiegelman BM. Adipocytes as regulators of energybalance and glucose homeostasis. Nature 2006; 444: 847–853.

5 Hunt CR, Ro JH, Dobson DE, Min HY, Spiegelman BM. AdipocyteP2 gene: developmental expression and homology of 50-flankingsequences among fat cell-specific genes. Proc Natl Acad Sci USA1986; 83: 3786–3790.

6 Scherer PE, Williams S, Fogliano M, Baldini G, Lodish HF. A novelserum protein similar to C1q, produced exclusively in adipocytes.J Biol Chem 1995; 270: 26746–26749.

7 Zilberfarb V, Pietri-Rouxel F, Jockers R, Krief S, Delouis C, Issad Tet al. Human immortalized brown adipocytes express functionalbeta3-adrenoceptor coupled to lipolysis. J Cell Sci 1997; 110:801–807.

8 Zilberfarb V, Siquier K, Strosberg AD, Issad T. Effect of dexa-methasone on adipocyte differentiation markers and tumournecrosis factor-alpha expression in human PAZ6 cells. Diabetologia2001; 44: 377–386.

9 Strobel A, Siquier K, Zilberfarb V, Strosberg AD, Issad T. Effect ofthiazolidinediones on expression of UCP2 and adipocyte markersin human PAZ6 adipocytes. Diabetologia 1999; 42: 527–533.

10 Grosfeld A, Zilberfarb V, Turban S, Andre J, Guerre-Millo M, IssadT. Hypoxia increases leptin expression in human PAZ6 adiposecells. Diabetologia 2002; 45: 527–530.

11 Jockers R, Issad T, Zilberfarb V, de Coppet P, Marullo S, StrosbergAD. Desensitization of the beta-adrenergic response in humanbrown adipocytes. Endocrinology 1998; 139: 2676–2684.

12 Brydon L, Petit L, Delagrange P, Strosberg AD, Jockers R.Functional expression of MT2 (Mel1b) melatonin receptors inhuman PAZ6 adipocytes. Endocrinology 2001; 142: 4264–4271.

13 Wabitsch M, Bruderlein S, Melzner I, Braun M, MechtersheimerG, Moller P. LiSa-2, a novel human liposarcoma cell line with

Comparative expression analysis of adipocytesEA van Beek et al

920

International Journal of Obesity

a high capacity for terminal adipose differentiation. Int J Cancer2000; 88: 889–894.

14 Chow WA, Guo S, Valdes-Albini F. Nelfinavir induces liposarcomaapoptosis and cell cycle arrest by upregulating sterol regulatoryelement binding protein-1. Anticancer Drugs 2006; 17: 891–903.

15 Yano S, Arroyo N, Yano N. Catalase binds Grb2 in tumor cellswhen stimulated with serum or ligands for integrin receptors. FreeRadic Biol Med 2004; 36: 1542–1554.

16 Lisitsyn N, Wigler M. Cloning the differences between twocomplex genomes. Science 1993; 259: 946–951.

17 van Beek EA, Bakker AH, Kruyt PM, Hofker MH, Saris WH, KeijerJ. Intra- and interindividual variation in gene expression inhuman adipose tissue. Pflugers Arch 2007; 453: 851–861.

18 Rodbell M. Metabolism of isolated fat cells. I. Effect of hormoneson glucose metabolism and lipolysis. J Biol Chem 1964; 239:375–380.

19 Bakker AH, Van Dielen FM, Greve JW, Adam JA, Buurman WA.Preadipocyte number in omental and subcutaneous adiposetissue of obese individuals. Obes Res 2004; 12: 488–498.

20 Franssen-van Hal NL, Vorst O, Kramer E, Hall RD, Keijer J. Factorsinfluencing cDNA microarray hybridization on silylated glassslides. Anal Biochem 2002; 308: 5–17.

21 Schena M, Shalon D, Heller R, Chai A, Brown PO, Davis RW.Parallel human genome analysis: microarray-based expressionmonitoring of 1000 genes. Proc Natl Acad Sci USA 1996; 93:10614–10619.

22 Pellis L, Franssen-Van Hal NL, Burema J, Keijer J. The intraclasscorrelation coefficient applied for evaluation of data correction,labeling methods and rectal biopsy sampling in DNA microarrayexperiments. Physiol Genomics 2003; 21: 21.

23 Boeuf S, Klingenspor M, Van Hal NLW, Schneider T, Keijer J,KLaus S. Differential gene expression in white and brownpreadipocytes. Physiol Genomics 2001; 7: 15–25.

24 Beisvag V, Junge FK, Bergum H, Jolsum L, Lydersen S, Gunther CCet al. GeneToolsFapplication for functional annotation andstatistical hypothesis testing. BMC Bioinformatics 2006; 7: 470.

25 Mineo H, Oda C, Chiji H, Kawada T, Shimizu K, Taira T.Thiazolidinediones exhibit different effects on preadipocytesisolated from rat mesenteric fat tissue and cell line 3T3-L1 cellsderived from mice. Cell Biol Int 2007; 31: 703–710.

26 Payne VA, Au WS, Gray SL, Nora ED, Rahman SM, Sanders R et al.Sequential regulation of DGAT2 expression by C/EBPbeta andC/EBPalpha during adipogenesis. J Biol Chem 2007; 282:21005–21014.

27 Hermann A, Gowher H, Jeltsch A. Biochemistry and biology ofmammalian DNA methyltransferases. Cell Mol Life Sci 2004;61: 2571–2587.

28 Robertson KD, Ait-Si-Ali S, Yokochi T, Wade PA, Jones PL, WolffeAP. DNMT1 forms a complex with Rb, E2F1 and HDAC1 andrepresses transcription from E2F-responsive promoters. Nat Genet2000; 25: 338–342.

29 Pradhan S, Kim GD. The retinoblastoma gene product interactswith maintenance human DNA (cytosine-5) methyltransferaseand modulates its activity. EMBO J 2002; 21: 779–788.

30 Hansen JB, Jorgensen C, Petersen RK, Hallenborg P, De Matteis R,Boye HA et al. Retinoblastoma protein functions as a molecularswitch determining white versus brown adipocyte differentiation.Proc Natl Acad Sci USA 2004; 101: 4112–4117.

31 Dali-Youcef N, Mataki C, Coste A, Messaddeq N, Giroud S, Blanc Set al. Adipose tissue-specific inactivation of the retinoblastomaprotein protects against diabesity because of increased energyexpenditure. Proc Natl Acad Sci USA 2007; 104: 10703–10708.

32 Wang Y, Zhang P, Li W, Hou L, Wang J, Liang Y et al. Mousefollicular and marginal zone B cells show differential expressionof Dnmt3a and sensitivity to 50-azacytidine. Immunol Lett 2006;105: 174–179.

33 Warnecke PM, Clark SJ. DNA methylation profile of the mouseskeletal alpha-actin promoter during development and differ-entiation. Mol Cell Biol 1999; 19: 164–172.

34 Walsh CP, Bestor TH. Cytosine methylation and mammaliandevelopment. Genes Dev 1999; 13: 26–34.

35 Plachot C, Lelievre SA. DNA methylation control of tissuepolarity and cellular differentiation in the mammary epithelium.Exp Cell Res 2004; 298: 122–132.

36 Rai K, Nadauld LD, Chidester S, Manos EJ, James SR, Karpf ARet al. Zebra fish Dnmt1 and Suv39h1 regulate organ-specificterminal differentiation during development. Mol Cell Biol 2006;26: 7077–7085.

37 Shuman S. Capping enzyme in eukaryotic mRNA synthesis.Prog Nucleic Acid Res Mol Biol 1995; 50: 101–129.

38 Furuichi Y, Shatkin AJ. Viral and cellular mRNA capping: past andprospects. Adv Virus Res 2000; 55: 135–184.

39 Yoshida E, Atkinson TG, Chakravarthy B. Neuroprotective geneexpression profiles in ischemic cortical cultures preconditionedwith IGF-1 or bFGF. Brain Res Mol Brain Res 2004; 131: 33–50.

40 Li F, Wang Y, Zeller KI, Potter JJ, Wonsey DR, O’Donnell KA et al.Myc stimulates nuclearly encoded mitochondrial genes andmitochondrial biogenesis. Mol Cell Biol 2005; 25: 6225–6234.

41 Sadowski HB, Wheeler TT, Young DA. Gene expression during3T3-L1 adipocyte differentiation. Characterization of initialresponses to the inducing agents and changes during commit-ment to differentiation. J Biol Chem 1992; 267: 4722–4731.

42 Urs S, Smith C, Campbell B, Saxton AM, Taylor J, Zhang B et al.Gene expression profiling in human preadipocytes and adipo-cytes by microarray analysis. J Nutr 2004; 134: 762–770.

43 Nedergaard J, Golozoubova V, Matthias A, Asadi A, Jacobsson A,Cannon B. UCP1: the only protein able to mediate adaptive non-shivering thermogenesis and metabolic inefficiency. BiochimBiophys Acta 2001; 1504: 82–106.

44 Timmons JA, Wennmalm K, Larsson O, Walden TB, Lassmann T,Petrovic N et al. Myogenic gene expression signature establishesthat brown and white adipocytes originate from distinct celllineages. Proc Natl Acad Sci USA 2007; 104: 4401–4406.

45 Kok EJ, Franssen-van Hal NL, Winnubst LN, Kramer EH, DijksmaWT, Kuiper HA et al. Assessment of representational differenceanalysis (RDA) to construct informative cDNA microarrays forgene expression analysis of species with limited transcriptomeinformation, using red and green tomatoes as a model. J PlantPhysiol 2007; 164: 337–349.

46 Zervos EE, Tanner SM, Osborne DA, Bloomston M, Rosemurgy AS,Ellison EC et al. Differential gene expression in patientsgenetically predisposed to pancreatic cancer. J Surg Res 2006;135: 317–322.

47 Wheeler DL, Church DM, Federhen S, Lash AE, Madden TL,Pontius JU et al. Database resources of the National Center forBiotechnology. Nucleic Acids Res 2003; 31: 28–33.

Comparative expression analysis of adipocytesEA van Beek et al

921

International Journal of Obesity