gene-expression profiling reveals distinct expression patterns for classic versus variant merkel...

Gene-expression profiling reveals distinct expression patterns for Classic

versus Variant Merkel cell phenotypes and new classifier genes todistinguish Merkel cell from small-cell lung carcinoma

Mireille Van Gele1, Glen M Boyle2, Anthony L Cook3,5, Jo Vandesompele1, Tom Boonefaes4,Pieter Rottiers4, Nadine Van Roy1, Anne De Paepe1, Peter G Parsons2, J Helen Leonard5 andFrank Speleman*,1

1Center for Medical Genetics, Ghent University Hospital, Ghent B-9000, Belgium; 2Melanoma Genomics Group, Queensland Instituteof Medical Research, Brisbane, Queensland 4006, Australia; 3Institute for Molecular Biosciences, University of Queensland, Brisbane,Queensland 4072, Australia; 4Department of Molecular Biomedical Research, Flanders Interuniversity Institute for Biotechnology andUniversity of Ghent, Ghent B-9000, Belgium; and 5Queensland Radium Institute Research Unit, Queensland Institute of MedicalResearch, Brisbane, Queensland 4006, Australia

Merkel cell carcinoma (MCC) is a rare aggressive skintumor which shares histopathological and genetic featureswith small-cell lung carcinoma (SCLC), both are ofneuroendocrine origin. Comparable to SCLC, MCC celllines are classified into two different biochemical sub-groups designated as ‘Classic’ and ‘Variant’. With the aimto identify typical gene-expression signatures associatedwith these phenotypically different MCC cell linessubgroups and to search for differentially expressed genesbetween MCC and SCLC, we used cDNA arrays toprofile 10 MCC cell lines and four SCLC cell lines. Usingsignificance analysis of microarrays, we defined a set of 76differentially expressed genes that allowed unequivocalidentification of Classic and Variant MCC subgroups. Weassume that the differential expression levels of some ofthese genes reflect, analogous to SCLC, the differentbiological and clinical properties of Classic and VariantMCC phenotypes. Therefore, they may serve as usefulprognostic markers and potential targets for the develop-ment of new therapeutic interventions specific for eachsubgroup. Moreover, our analysis identified 17 powerfulclassifier genes capable of discriminating MCC fromSCLC. Real-time quantitative RT–PCR analysis of thesegenes on 26 additional MCC and SCLC samplesconfirmed their diagnostic classification potential, openingopportunities for new investigations into these aggressivecancers.Oncogene (2004) 23, 2732–2742. doi:10.1038/sj.onc.1207421Published online 2 February 2004

Keywords: Merkel cell carcinoma; small-cell lung carci-noma; differential expression profiling; differentialdiagnosis; Classic/Variant

Introduction

Merkel cell carcinoma (MCC) is a rare aggressive skintumor and is assumed to arise from normal Merkel cells.Merkel cells are neuroendocrine in origin, expressingmarkers such as neuron-specific enolase and bombesin.These cells are located in the basal layer of theepidermis, where they often function as slow-actingmechanoreceptors (Halata et al., 2003). MCC mostlyaffects elderly people and occurs predominantly on thesun-exposed areas of the skin, suggesting UV exposurein its etiology (Miller and Rabkin, 1999; Van Gele et al.,2000; Popp et al., 2002).In a previous study, we have determined by compara-

tive genomic hybridization (CGH) the patterns ofgenomic imbalances which occur in MCC (Van Geleet al., 1998). Interestingly, the observed under- and over-representations of partial chromosomal regions werequite similar to those observed in small-cell lungcarcinoma (SCLC) (Ried et al., 1994; Levin et al.,1995; Petersen et al., 1997; Van Gele et al., 1998). Bothare neuroendocrine tumors with remarkable histopatho-logical similarities, that is, small round cells oftencontaining dense core granules and expressing severalidentical immunohistochemical markers (Ratner et al.,1993; Metz et al., 1998; Schmidt et al., 1998; Goesslinget al., 2002). Therefore, the possibility exists that theyarise from a common cell lineage. Further support forthis hypothesis comes from the fact that cell linesderived from MCC and SCLC resemble each other intheir biochemical, morphological and growth character-istics. Similar to SCLC, MCC cell lines are classified intotwo groups: ‘Classic’ and ‘Variant’ defined on theirbiochemical markers and neurosecretory granule status,which are further subdivided into four subtypes (I–IV)based on morphology, colony shape and aggregation(Carney et al., 1985; Gazdar et al., 1985; Leonard et al.,1993, 1995a, 2002; Leonard and Bell, 1997). At present,little is known about the genes associated with thesecharacteristics.

Received 26 August 2003; revised 17 November 2003; accepted 2December 2003

*Correspondence: F Speleman; E-mail: [email protected]

Oncogene (2004) 23, 2732–2742& 2004 Nature Publishing Group All rights reserved 0950-9232/04 $25.00

www.nature.com/onc

ON

CO

GE

NO

MIC

S

In contrast to MCC, numerous molecular geneticstudies have been performed on SCLC which havecontributed to the understanding of SCLC pathogenesis(Fong et al., 1999; Minna et al., 2002). In addition,histochemical markers and differentially expressed genesdistinguishing Classic from Variant SCLC cell lines havebeen identified and could lead to an improved under-standing of the underlying genetic basis responsible forthe biological and clinical heterogeneity among small-cell lung cancers (Broers et al., 1985; Zhang et al., 2000).In order to obtain further insights into the complex

and heterogeneous molecular pathogenesis of MCC, wedetermined the gene-expression profiles of 10 MCC celllines and four SCLC cell lines using Atlas cDNA arrayscontaining 1891 unique genes involved in many cellularfunctions. This study offers potential insights into thegenes and signalling pathways involved in MCC andSCLC, a prerequisite for the development of newrational therapeutic interventions, which could lead toan improved patient survival or even complete remis-sion. Furthermore, we identified genes not previouslyimplicated in these cancers, whose expression enableddiscrimination between MCC and SCLC and maytherefore aid in the differential diagnosis of cases whereexisting markers such as cytokeratin 20 are unable todifferentiate between these two neuroendocrine cancers.

Results

Validation of atlas cDNA array data

The expression level of 1891 genes was measured by theuse of Atlas Human and Human Cancer 1.2 arrays for10 MCC and four SCLC cell lines. After primary dataanalysis and normalization, the data of 1083 geneswhich had an expression value above background in atleast six of the analysed samples were used forhierarchical clustering. Of these 1083 genes, 206 geneswere present on both the Human and Human Cancer1.2 arrays and a high correlation was found between theexpression levels of these common genes for each sample(mean Spearman rank correlation coefficient¼ 79.3%).Reliability and reproducibility of the array gene-expres-sion data were further supported by (a) comparison ofthe array gene-expression levels and real-time RT–PCRexpression levels for 25 selected genes (mean Spearmanrank correlation coefficient¼ 76.1%) (see below), (b) ahighest degree of similarity evidenced by hierarchicalclustering for expression patterns of cell lines MCC14/1and MCC14/2, derived from the same tumor (seeFigure 1b) and (c) confirmation of the differential geneexpression for ASCL1 in SCLC versus MCC, asreported in the literature (see Discussion).

Hierarchical clustering analysis of MCC and SCLC celllines

Hierarchical clustering was used to identify similaritiesin gene-expression patterns between MCC and SCLCcell lines. Clustering of the 14 samples was based on the

gene-expression levels of the 1083 preselected genes.Figure 1a shows the complete cluster diagram. Thedendrogram (Figure 1b) summarizes the degree ofsimilarity in gene expression among the 14 analysed

Figure 1 (a) Scaled-down representation of the hierarchicalcluster diagram of 1083 selected genes and 10 MCC cell lines andfour SCLC cell lines. A row in the cluster indicates expression of aspecific gene across all the 14 samples. A column indicates thesample in which the gene is expressed. The color scale (ExpressionIndex) shown at the bottom (�3 to þ 3 in log base 2 units) indicatesthat the relative expression level of the gene is greater, less than orequal to the geometric mean expression across all 14 samples,respectively. (b) Dendrogram representing similarities in theexpression patterns between experimental samples from Figure 1a.The symbols below the sample names reflect the different(sub)groups of tumor cell lines (circle: Variant MCC; asterix:Classic MCC; square: SCLC). (c) Real-time-based hierarchicalcluster analysis of MCC cell lines and tumor samples for ninegenes, which can distinguish between the Classic and Variant MCCsubtypes (asterix: Classic MCC cell lines; square: MCC tumors;circle: Variant MCC cell lines. (d) Average linkage hierarchicalcluster analysis of MCC and SCLC cell lines and tumor samples for14 genes identified by SAM as able to differentiate between MCC(1, asterix) and SCLC (2, square) and quantified by real-time PCR.Same color scale as above for both figures, but expressed in log 10base units

Gene-expression profiling of MCC and SCLCMV Gele et al

2733

Oncogene

samples. Two major subgroups were observed. Exceptfor MCC cell line UISO, group 1 contained all VariantMCC cell lines (circle:MCC26, MCC13, MCC14/2 andMCC14/1 (see Table 1)), indicating that these samplesresemble each other in their gene-expression patterns.Group 2 included UISO, and all Classic MCC cell lines(asterix:MKL-2, MKL-1, T95-45, MCC6 and MCC5)which were mixed with all the four SCLC cell lines(square), of which GLC4 was the only Variant one (seeTable 1). From this analysis, we concluded thatclustering of MCC cell lines into different subgroupspredominantly coincided with the Classic or Variantphenotypes of the respective cell lines. However,hierarchical clustering of the 1083 preselected genescould not unequivocally separate MCC cell lines fromSCLC cell lines. This does, however, further support aputative ontogenetic relationship for both tumors.

Identification of differentially expressed genes in Classicversus Variant MCC cell lines

Unsupervised analysis of array data enables coherentpatterns of gene expression to be identified, but provideslittle information about the statistical significance.Therefore, we decided to exploit a supervised strategyin order to identify a specific set of genes whoseexpression pattern could discriminate Classic versusVariant MCC cell lines. To this purpose, Classic (n¼ 5)and Variant (n¼ 5) MCC cell lines were predefined asthe two sample groups. Subsequently, a two-class SAManalysis on the log transformed data matrix containing1365 genes (see Material and methods) was performed.A cutoff value delta, depending on an arbitrary false-positive rate, was chosen to identify significantlydifferentially expressed genes. For this analysis, a deltavalue of 0.90572 was used. This led to the identificationof a total of 239 differentially expressed genes with a

median ‘false discovery rate’ (FDR) of 2.5%, whichmeans that there are about six false positives on average.Highly differentially expressed significant genes werefurther selected if a differential expression pattern(4two-fold difference) was present in at least four ofthe five Classic MCC cell lines as compared to theVariant ones and vice versa. In all, 46 genes with arelative elevated expression in MCC Classic cell linesand 30 genes with an increased expression level in theVariant MCC cell lines were identified (see Table 2).Hierarchical cluster analysis of these 76 genes clearlyclassified the MCC cell lines in their respectivephenotypic groups (data not shown). Genes with kinaseactivities but also genes encoding for ligand and voltage-gated ion channels, neuromediators, GDP/GTP exchan-gers and signal-transduction receptors were seen athigher expression levels in Classic MCC cell linesrelative to Variant ones. Genes with a higher expressionin Variant cell lines compared to Classic ones includedgenes involved in cell cycle control and proliferation (seeTable 2).

Identification of differentially expressed genes in MCCversus SCLC

Hierarchical clustering analysis performed in this studyshowed a high degree of similarity between MCC andSCLC. We were interested, however, to search for geneswhich could distinguish between the two tumor types.Therefore, the same supervised strategy as outlinedabove was applied to identify genes differentiallyexpressed in MCC versus SCLC cell lines. Due to thesmall gene-expression level variances between MCC andSCLC, as a result of their similarity, we had to choose arather low delta (D¼ 0.33878) in order to include asmany genes as possible with significant but sometimessmall expression differences in MCC versus SCLC and

Table 1 Characteristics of the MCC and SCLC cell lines used for cDNA array hybridization

Cell line Morphological type Colony shape Colony aggregation Classificationa

MCC cell linesMCC5 I 3-db Tight ClassicMCC6 I 3-d Tight ClassicMCC13 IV Flat NAc VariantMCC14/1 IV Flat NA VariantMCC14/2 IV Flat NA VariantMCC26 IV Flat NA VariantUISO IV Flat NA VariantMKL-1 III 2-d Loose ClassicMKL-2 III 2-d Loose ClassicT95-45 II 3-d Loose Classic

SCLC cell linesNCI-H69 II 3-d Loose ClassicNCI-H146 II 3-d Loose ClassicCOR-L88 III 2-d Loose ClassicGLC4 III 3-d Loose Variant

aClassic MCC cell lines express neuroendocrine markers including neuron-specific enolase and Chromogranin A, and contain neurosecretorygranules (Leonard et al., 1993). Variant MCC cell lines have a selective loss of neuroendocrine markers including Chromogranin A, and do notcontain neurosecretory granules, as evidenced by electron microscopy (Leonard et al., 1995a). Classic SCLC cell lines express levels of L-dopadecarboxylase and bombesin while Variant ones have undetectable levels of L-dopa decarboxylase and bombesin (Carney et al., 1985; Gazdar et al.,1985). b3-d; three-dimensional clusters, 2-d; two-dimensional clusters. cNot applicable, adherent growing cell lines.


2734

Oncogene

Table 2 List of Classic and Variant MCC classifier genes

Spota GB Accb Symbol Gene/protein identity Fold chc Chrom locd Core P-value

Classic specific genesC_E09m D78345 IGHG3 Membrane-bound and secreted immunoglobulin

gamma heavy chain100 14q32.33 ND ND

C_E08k U78095 SPINT2 Kunitz-type serine protease inhibitor 2 50 19q13.1 83.5 3.80E-04C_D02h U96136 CTNND2 Delta catenin 38.46 5p15.2 ND NDH_D01d X53179 CHRNB2 Cholinergic receptor nicotinic beta polipeptide 2 25 1q21.3 ND NDH_D08e J05252 PCSK2 Neuroendocrine convertase 2 20 20p11.2 ND NDH_D07h L19761 SNAP25 25-kDa synaptosomal-associated protein 20 20p12–11.2 ND NDH_F06k X51405 CPE Carboxypeptidase H 11.11 4q32.3 ND NDH_B13d D10924 CXCR4 CXC chemokine receptor type 4 11.11 2q21 ND NDC_E09h AF003521 JAG2 Jagged homolog 2 11.11 14q32 ND NDC_A10n U72649 BTG2 NGF-inducible anti-proliferative protein PC3 10 1q32 93.4 1.03E�06H_A13n U07139 CACNB3 Dihydropyridine-sensitive L-type calcium

channel beta-3 subunit9.09 12q13 ND ND

H_B05f X74979 DDR1 Epithelial discoidin domain receptor 1 8.33 6p21.3 ND NDC_F09a U86759 NTN2L Netrin-2 8.33 16p13.3 ND NDH_D01g Y00757 SGNE1 Secretory granule endocrine protein I 8.33 15q13–14 ND NDC_B10h L42374 PPP2R5B Protein phosphatase 2A B56-beta 8.33 11q12 ND NDC_A04m X80343 CDK5R1 Cyclin-dependent kinase 5 activator 7.69 17q11.2 ND NDC_D09j AF011466 EDG4 G protein-coupled receptor EDG4 7.69 19p12 ND NDC_B13d L07597 RPS6KA1 Ribosomal protein S6 kinase II alpha 1 7.69 3 ND NDC_C12k U66197 FGF12 FHF-1 7.14 3q28 ND NDH_C05b L05500 ADCY1 Adenylate cyclase type I 7.14 7p13–12 ND NDH_B14a L14595 SLC1A4 Neutral amino-acid transporter A 6.67 2p15–13 ND NDH_C14e M23410 JUP Junction plakoglobin 6.67 17q21 94.5 1.12E�06C_B02f D50925 PASK KIAA0135 6.67 2q37.3 ND NDC_C02l Y11416 TP73 p73 6.67 1p36.33 ND NDC_F13b L11931 SHMT1 Cytosolic serine hydroxylmethyltransferase 6.67 17p11.2 ND NDC_B08i U27193 DUSP8 Dual-specificity protein phosphatase 8 6.25 11p15.5 ND NDC_C02f L07540 RFC5 Replication factor C 36-kDa subunit 6.25 12q24.2–24.3 ND NDH_C12k L09561 POLE DNA polymerase II subunit A 6.25 12q24.3 ND NDC_F13n U96876 INSIG1 Insulin-induced protein 1 5.89 7q36 ND NDC_D04l Y08110 LR11 Low-density lipoprotein receptor-related

protein LR115.56 11q23.2–24.2 ND ND

C_C03b U77352 MADD MAP kinase-activating death domain protein 5.27 11p11.2 ND NDH_D07g AF040255 DCX Doublecortin 5.26 Xq22.3–23 ND NDC_A11f AF006484 CDK2AP1 Putative oral tumor-suppressor protein 4.76 12q24.31 ND NDC_B13l D84064 HGS HRS 4.55 17q25 ND NDH_B03m X80907 PIK3R2 Phosphatidylinositol 3-kinase regulatory

beta subunit4.55 19q13.2–13.4 ND ND

H_E12f U04810 TROAP Trophinin-associated protein 4.55 12p11.1 ND NDH_A05k U40343 CDKN2D Cyclin-dependent kinase 4 inhibitor 2D 4.35 19p13 ND NDC_C08e U86529 GSTZ1 Glutathione transferase zeta 1 4 14q24.3 ND NDH_C13k D38073 MCM3 MCM3 DNA replication licensing factor 4 6p12 ND NDH_C08g M93426 PTPRZ1 Protein-tyrosine phosphatase zeta 4 7q31.3 ND NDC_D11e U24497 PKD1 Autosomal dominant polycystic kidney

disease protein 13.57 16p13.3 ND ND

H_C05e X70326 MLP MARCKS-like protein 3.33 1p34.3 75.8 2.67E�03C_E05h X61157 ADRBK1 Beta-adrenergic receptor kinase 1 3.33 11q13 ND NDH_B01a X91906 CLCN5 Chloride channel protein 5 2.86 Xp11.23–11.22 ND NDH_A07n U69883 KCNN1 Calcium-activated potassium channel HSK1 2.7 19p13.1 ND NDH_A08i X60188 MAPK3 Mitogen-activated protein kinase 3 2.38 16p12–11.2 ND ND

Variant specific genesC_F04g X56134 VIM Vimentin 42.71 10p13 90.8 7.30E�06H_A07d X16707 FOSL1 FOS-like antigen 1 25.05 11q13 80.8 8.39E�04H_A05e M76125 AXL axl oncogene 21.14 19q13.1 90.3 9.56E�06H_A03h X59798 CCND1 G1/S-specific cyclin D1 16.56 11q13 83.9 2.17E�04H_F05n X03124 TIMP1 Tissue inhibitor of metalloproteinase 1 16.09 Xp11.3–11.23 ND NDH_C06h U28014 CASP4 Caspase 4 10.57 11q22.2–22.3 ND NDH_A11g X57766 MMP11 Matrix metalloproteinase 11 9.15 22q11.23 ND NDH_C14h M23254 CAPN2 M-type calcium-activated neutral proteinase 9.13 1q41–42 ND NDH_E03f M24069 CSDA Cold shock domain protein A 8.31 12p13.1 ND NDC_F09g M96322 AKAP12 Gravin 7.85 6q24–25 ND NDH_D10j Z36715 ELK3 ets domain protein elk-3 7.59 12q23 ND NDH_E09g M73780 ITGB8 Integrin beta 8 7.32 7p21.3 ND NDH_F03b M34664 HSPD1 Heat shock 60-kDa protein 7.19 2q33.1 ND NDC_F14a J03040 SPARC Secreted protein acidic and rich in cysteine 7.04 5q31.3–32 ND NDH_E13k X66945 FGFR1 Fibroblast growth factor receptor 1 6.78 8p11.2–11.1 ND NDC_D09m U90313 GSTTLp28 Glutathione-S-transferase-like protein 6.67 10q24.33 ND ND


2735

Oncogene

vice versa. This led to the identification of a total of 121differentially expressed genes with a median FDR of27.9%, that is, 33.8 false-positive genes on average. Inaddition, the 121 SAM-identified genes were furtherselected as highly differentially expressed significantgenes in MCC versus SCLC if a differential expressionpattern (4two-fold difference) was present in at leastfour of the MCC cell lines as compared to SCLC or in atleast three of the four SCLCs as compared to MCC. Inthis way, the use of a low delta value merged with arelatively high FDR was justified. Subsequently, 12 andinitially seven genes showed higher and lower expressionlevels respectively in MCC versus SCLC (see Table 3 andfootnote f). Genes more highly expressed in MCCincluded, for example, brain-derived neurotrophic factor,heat shock-related 70-kDa protein 2, neurogranin andintergrin alpha 3. In three of the four SCLC cell lines,elevated levels of the neuronal differentiation markerASCL1 (achaete-scute homolog 1), the basic helix–loop–helix transcription factor ID2 (inhibitor of DNA-binding protein 2) and GPX2 (glutathione peroxidase-related protein 2) were observed relative to those seen inMCC cell lines. The lower expression level of ASCL1 incell line GLC4 was of interest, as this was the onlyVariant SCLC cell line examined. All others had aClassic phenotype (see Table 1) consistent with ASCL1being required for the neuroendocrine phenotype ofSCLC (Borges et al., 1997). A complete list of SAMgenes differentially expressed between MCC and SCLCis given in Table 3. Two genes, FLT1 and EGR1(Table 3, footnote f), were however excluded fromfurther analysis, as differences in array gene-expressionlevels were not confirmed by real-time quantitative RT–PCR analysis. The classification potential of theremaining 17 differentially expressed genes betweenMCC and SCLC was visualized after re-clustering. Eachcell line clearly clustered into either the MCC or SCLCsubgroup (data not shown). The low number of classifier

genes being identified in this analysis illustrated the highdegree of similarity at the RNA expression level betweenMCC and SCLC, and further supports a putativeontogenetic relationship between both tumor types.

Validation of array gene-expression levels andclassification of additional MCC and SCLC cell lines/tumor samples by real-time RT–PCR analysis

To verify the array gene-expression levels by anindependent and sensitive method, we performed real-time quantitative RT–PCR on the same RNA samplesof the 14 cell lines used for filter array analysis. In all, 25SAM selected genes were quantified. Of these, 16 genes(ASCL1, GPX2, ID2, TFAP4, FLT1, IGFBP2, PRKCA,ITGA3, BDNF, ILK, PRKR, CDC25B, EGR1, CHD2,MAP2K3 and HSPA2) were previously selected by theirability to distinguish between MCC and SCLC (seeTable 3). The other nine genes selected arbitrarily fromthe SAM list (SPINT2, MYC, AXL, CCND1, JUP,BTG2, FOSL1, VIM and MLP) had Classic versusVariant classification capability (see Table 2). In general,the quantitative RT-PCR data correlated very well withthe array hybridization data (see Tables 2 and 3).Figures 2a and b illustrate relative real-time RT–PCRdata and array hybridization data for the genes ASCL1and IGFBP2, with Spearman rank correlation coeffi-cients of 70.5% (P-value¼ 4.82E�03) and 92.1%(P-value¼ 2.98E�06), respectively.Average linkage hierarchical cluster analysis of 12

MCC cell lines and 10 MCC tumors for nine SAM geneswith phenotypic classification potential and quantifiedby real-time PCR resulted in separation of all (adherent)Variant MCC cell lines, all but one Classic MCC cellline (MCC5) and nine of the 10 MCC tumors(Figure 1c). These results illustrate that, even with alimited selected set of differential genes, a distinctionbetween Classic and Variant MCC cell lines can easily

Table 2 (continued )


H_B12g U48959 MYLK Smooth muscle and nonmuscle myosinlight chain kinase

6.18 3q21 ND ND

C_B10c L31951 MAPK9 Mitogen-activated protein kinase 9 6.16 5q35 ND NDH_F02g U10117 SCYE1 Endothelial-monocyte activating

polypeptide II6.12 4q24 ND ND

C_C10b U60520 CASP8 Caspase 8 5.74 2q33–34 ND NDC_E03i AF031385 CYR61 Cysteine-rich anigogenic inducer 61 4.97 1p31–22 ND NDH_B11e M59371 EPHA2 Ephrin type-A receptor 2 4.87 1p36 ND NDH_D08i M13667 PRNP Major prion protein 4.66 20pter-p12 ND NDH_F02l D00760 PSMA2 Proteasome subunit alpha type 2 4.23 7p13–12 ND NDH_A02d J04101 ETS1 ets1 proto-oncogene 3.93 11q23.3 ND NDC_A05i M25753 CCNB1 G2/mitotic-specific cyclin B1 3.83 5q12 ND NDH_A12c V00568 MYC myc proto-oncogene 3.67 8q24.12–24.13 94.7 1.00E�06C_E11l M34671 CD59 CD59 glycoprotein 3.64 11p13 ND NDH_E13g D13866 CTNNA1 Alpha1 catenin 3.53 5q31 ND NDH_F07k X87212 CTSC Cathepsin C 3.33 11q14.1–14.3 ND ND

aSpot location of each gene on either the Atlas Human (H) or Cancer (C) arrays. bGenBank accession number. cFold change was calculated bydividing the mean expression level of all Classic MCC cell lines to the mean expression level of all Variant MCC cell lines for Classic specific genes(and vice versa for Variant specific genes). dChromosomal location of each gene. eExpression of a number of genes was confirmed by real-timequantitative RT–PCR. The Spearman rank correlation coefficient (%) was calculated between the array gene-expression levels and the real-timeRT–PCR levels. ND, not done.


2736

Oncogene

be made. Originally, the grouping of Classic and Variantphenotypes for MCCs was cell line based (Leonard et al.,1993; Leonard et al., 1995a, 2002; Leonard and Bell,1997). However, we have observed concordant resultsbetween tumor samples and their respective cell lines inseveral genomic deletion analyses (Leonard and Hayard,1997; Leonard et al., 2000; Van Gele et al., 2000; Cooket al., 2001) and in immunohistochemical studies. Inparticular, the transcription factor HATH1, shown to beexpressed in normal Merkel cells and in Classic MCCcell lines was also expressed only in those biopsies fromwhich Classic MCC cell lines, were derived (Leonardet al., 2002). In the present study, almost all MCCtumors clustered in one group which showed a higherdegree of similarity with the Classic cell lines comparedto the Variant cell lines, although they shared somegene-expression features common to both cell linegroups. Given the concordance previously seen betweentumors and their respective cell lines, these groupingsare likely to have clinical significance and only furtherstudies on additional tumors would determine if thoseexamined could be typed as having a ‘Classic’phenotype.It should be mentioned that cell line MCC19 was

recently thought to be a Variant MCC suspension cellline based on the lack of HATH1 expression (Leonardet al., 2002). However, MCC19 still expresses, similar toClassic MCC suspension cell lines, the transcriptionfactor Brn-3c (Leonard et al., 2002) and ChromograninA, a neuroendocrine marker. Therefore, it is probablynot surprising that MCC19 is outlying the Variantcluster group, but shows instead a high degree ofsimilarity with Classic MCC suspension cell lines such as

MKL-1 (Figure 1c). Although PCR-based hierarchicalclustering of the nine selected SAM genes analysed inthis study could not classify MCC19 as a separate thirdbiological MCC subgroup, as suggested by HATH1reactivity (Leonard et al., 2002), quantitative PCRanalysis of a larger panel of differentially expressedgenes followed by clustering with an extended numberof Variant MCC suspension cell lines may enable this tooccur.In order to confirm the classification strength of the

SAM selected genes (MCC versus SCLC), real-time RT–PCR analysis was extended to two further MCC celllines, 10 MCC tumor samples, 12 additional SCLC celllines and two SCLC tumors for ASCL1, GPX2, ID2,TFAP4, IGFBP2, PRKCA, ITGA3, BDNF, ILK, PRKRCDC25B, CHD2, MAP2K3 and HSPA2. Averagelinkage hierarchical cluster analysis based on theSpearman rank correlation coefficient as a similaritymeasure showed two major clusters (Figure 1d). Exceptfor one SCLC cell line (NCI-N464), cluster 1 containedall MCC cell lines and MCC tumor samples. Cluster 2consisted of all the remaining SCLC cell lines and SCLCtumors. Real-time PCR-based gene-expression profilingtherefore resulted in an almost perfect classification ofthe different tumors or cell lines into their respectivetumor groups. In addition, the cell lines MCC14/1 andMCC14/2 established from the same tumor sampleremained clustered next to each other in a subgroupwith other Variant MCC cell lines. The cell lines MKL-1and MKL-1 clone 2, also derived from a same patient,were found in different subclusters. The MKL-1 clone 2has, however, been grown separately from MKL-1 for along time, and has clonally evolved, apart from

Table 3 List of MCC and SCLC classifier genes


MCC specific genesH_F09f M61176 BDNF Brain-derived neurotrophic factor 16.67 11p13 92.9 4.61E�06H_F04a L26336 HSPA2 Heat shock-related 70-kDa protein 2 11.11 14q24.1 95.1 2.04E�06H_C06c Y09689 NRGN Neurogranin 7.69 11q24 ND NDH_E06i M59911 ITGA3 Integrin alpha 3 7.14 17q21.32 85.9 8.20E�05H_B03i M22199 PRKCA Protein kinase C alpha polypeptide 6.25 17q22–23.2 75.9 1.67E�03H_C06j M35663 PRKR Interferon-inducible RNA-dependent protein kinase 4.17 2p22–21 89.9 1.24E�05H_B12h L36719 MAP2K3 Mitogen-activated protein kinase kinase 3 3.7 17q11.2 80.7 4.91E�04H_A01k D84212 STK6 Aurora-related kinase 1 3.57 20q13.2–13.3 ND NDH_E06h M34064 CDH2 Cadherin 2 2.94 18q11.2 83.3 2.17E�04H_A08j D88435 GAK Cyclin G-associated kinase 2.86 4p16 ND NDH_B03g U40282 ILK Integrin-linked kinase 2.5 11p15.5–15.4 60.9 2.09E�02C_A11l M81934 CDC25B Cell division cycle 25 homolog B 2.44 20p13 82.9 2.51E�04

SCLC specific genesf

H_D01h L08424 ASCL1 Achaete-scute homolog 1 70.72 12q22–23 70.5 4.82E�03H_F06a X53463 GPX2 Glutathione peroxidase-related protein 2 10.43 14q24.1 83.6 1.33E�03H_D11m M97796 ID2 Inhibitor of DNA binding 2 protein 5.06 2p25 87.7 3.83E�05C_A08f M35410 IGFBP2 Insulin-like growth factor-binding protein 2 3.33 2q33–34 92.1 2.98E�06H_E04d S73885 TFAP4 AP4 basic helix–loop–helix DNA-binding protein 2.68 16p13 78.9 7.95E�04

aSpot location of each gene on either the Atlas Human (H) or Cancer (C) arrays. bGenBank accession number. cFold change was calculated bydividing the mean expression level of all MCC cell lines to the mean expression level of all SCLC cell lines for MCC-specific genes (and vice versa forSCLC specific genes). dChromosomal location of each gene. eExpression of a number of genes was confirmed by real-time quantitative RT–PCR.The Spearman rank correlation coefficient (%) was calculated between the array gene-expression levels and the real-time RT–PCR levels. ND, notdone. fInitially identified SAM genes FLT1 and EGR1 were excluded for further analysis, as their array gene-expression levels were not confirmedby real-time RT–PCR analysis (Cor¼�31.0%; P-value¼ 2.81E�01 and Cor¼�14.2%; P-value¼ 6.26E�01, respectively).


2737

Oncogene

becoming karyotypically unrelated to MKL-1 (Rosenet al., 1987; Van Gele et al., 2002).The value of using cell lines for gene-expression

profiling and validation of differentially expressed genesfor tumor types such as SCLC where tumor material isvery difficult to obtain in sufficient amount or numbershas recently been demonstrated in a global geneexpression analysis by Pedersen et al. (2003). Therefore,we believe that, in this study, the use of a limited numberof available SCLC tumors combined with a largernumber of SCLC cells in culture for validation of SCLCclassifier genes is justified and reliable.

Discussion

In this study, expression profiling of 10 MCC and fourSCLC cell lines was performed through analysis of 1891unique genes. Hierarchical clustering was used in a firstgeneral attempt to assess the classification power of theobtained expression data set. Cluster analysis of 1083preselected genes allowed the MCC cell lines tosegregate into two different subgroups mainly associated

with their Classic or Variant phenotypes. On the otherhand, this analysis could not distinguish between MCCand SCLC, emphasizing their biological/genetic rela-tionship. Therefore, we adopted a superviseddata-mining strategy, that is, two-class SAM analysis,in order to identify (a) phenotypic classifier genes whichallow to separate Classic from Variant MCC cell linesand (b) diagnostic classifier genes which may aid in thedifferential diagnosis of MCC and SCLC.This led to the identification of 76 highly differentially

expressed significant genes, of which 46 showed higherexpression in the Classic cell lines and 30 were morehighly expressed in the Variant MCC cell lines. A subsetof genes with higher levels of expression in the Classiccell lines are involved in signal-transduction pathways,and could lead to increased cell growth when over-expressed. This is exemplified by genes such as MAPK3and MADD involved in the mitogen-activated protein(MAP) kinase pathway, and genes such as PI3-K p85beta in the phosphatidylinositol 3-kinase (PIK3) path-way. In addition, Classic cell lines showed higher levelsof expression of genes coding for neuromediators(SGNE1) and neurotransmittors (PCSK2), and proteinsinvolved in neuronal development such as doublecortin(DCX) and MARCKS-like protein (MLP). This is inkeeping with the observed neuroendocrine and moredifferentiated phenotypes associated with the ClassicMCC cell lines. Ligand and voltage-gated ion channelsand receptors essential for neurotransmission were alsoexpressed in the Classic cell lines. Some of these ionchannels are known to play a role during mechanicalstimulation of normal Merkel cell receptors (Baumannet al., 2000; Tazaki et al., 2000). Variant MCC cell linescould have lost expression of some of these ion channels.Their specific function in MCC tumor cells, however,has yet to be elucidated.Genes with higher expression in Variant cell lines were

involved in cell cycle control (CCND1 and CCNB1) andcell proliferation (HSP60, MMP11, MAPK9, FOSL1,AXL, MYC and ETS1). Some of these genes maycorrelate with the shorter doubling time and aggressivenature of the Variant MCC cell lines, as illustrated bytheir higher cloning efficiency in soft agar and theirreduced sensitivity to radiation (Leonard et al., 1995b).In addition, we observed high expression of vimentin, amesenchymal marker, together with FOSL1 (aliasFRA1, FOS-related antigen 1). A tight correlation ofvimentin and FOSL1 expression was also recently foundin highly invasive breast cancer cell lines, pointing to apossible role in tumor progression and enhanced cellmigration of these cancer cells (Zajchowski et al., 2001).These two genes could be significant prognostic markersfor the more aggressive MCC Variant types. Increasedexpression of vimentin was also observed by immuno-histochemical studies in Variant SCLC cell lines (Broerset al., 1985, 1986), and as a result of a suppressionsubtractive hybridization experiment comparing a Clas-sic to a Variant SCLC cell line (Zhang et al., 2000).These observations could point at a similar mechanismof tumor progression or metastatic properties betweenMCC and SCLC Variant phenotypes.

Figure 2 Histogram comparing relative real-time PCR expressionlevels (gray bars) and array hybridization levels (white bars) of (a)ASCL1 and (b) of IGFBP2 in MCC and SCLC cell lines (ordinate,log10 space). The normalized expression level of each gene wasdivided by its geometric mean across all 14 samples. The Spearmanrank correlation coefficient (Sp r) for both genes is indicated on thehistogram


2738

Oncogene

To extend and validate the information from theSAM analysis data set, nine genes able to differentiatebetween Classic and Variant cell lines were arbitrarilyselected from the list and quantified by real-time PCR in12 MCC cell lines and 10 other MCC tumors. Clusteranalysis of the PCR results confirmed the separation ofthe phenotypically different MCC cell lines, andillustrates that, even with a limited selected set ofdifferential genes, a distinction between Classic andVariant MCC cell lines can be made.In addition, comparison of our results found for

MCC with recently published gene-expression profiles ofSCLC cell lines by Pedersen et al. (2003) revealed anumber of identical genes expressed in the Classic celllines, but not in the Variant ones of both tumor types.These included the neuroendocrine gene SGNE1 andneuronal markers doublecortin and MACMARKS orMLP. These genes could be used as markers fordistinction between Classic and Variant phenotypes ofboth tumor types. Moreover, the nine selected SAMgenes able to distinguish between Classic and VariantMCC cell lines could also be used as classifiers forClassic and Variant SCLC cell lines (personal observa-tion). Two of these nine genes, vimentin (see also above)and SPINT2, were previously found to be differentiallyexpressed between a Classic and more aggressiveVariant SCLC cell line after a suppression subtractivehybridization experiment (Zhang et al., 2000). Interest-ingly, SCLC tumors derived from Variant cell lines aremore aggressive and patients have a worse prognosis(Gazdar et al., 1985). Comparable to SCLC, lack ofexpression of Classic marker genes described here oroverexpression of Variant MCC classifier genes couldindicate a subset of more aggressive MCCs, for whichmore intensive treatment and closer follow-up may bewarranted in a similar way to our recent results for Brn-3c/HATH1 expression (Leonard et al., 2002). Real-timePCR analysis of 10 MCC tumors did not separate theminto distinct subgroups. However, the number of Classicversus Variant classifier genes analysed by real-timePCR in this study was limited to nine genes, andextension of this panel and also increased numbers ofpatients, for which survival data and treatment proce-dures are also available, might be beneficial. In addition,future investigations of genes or disregulated pathwaysinvolved in Classic and Variant MCC and/or SCLCcell lines could lead to potential targets for thedevelopment of new therapeutic strategies specific foreach (sub)group.A second goal of the study was to identify genes which

were able to distinguish MCC from SCLC. This led tothe identification of 17 classifier genes whose gene-expression levels showed significant differential geneexpression between MCC and SCLC samples. In all, 12of these genes showed a higher expression in MCC ascompared to SCLC. Of particular interest was brain-derived neurotrophic factor (BDNF), which is known tostimulate the mechanotransducing properties of normalMerkel cells (Carroll et al., 1998). In addition, over-expression of BDNF in murine skin was shown to beassociated with an increase in Merkel cell number

(Botchkarev et al., 1999). Higher expression in MCC ascompared to SCLC could contribute, in patients, to anincrease in the numbers of Merkel cells typicallyobserved in MCC tumors (Moll et al., 1996).Consequently, if BDNF could be downregulated inMCC patients, this might have an antiproliferativeeffect. This hypothesis warrants further investigation.Expression of CDC25B was recently observed in SCLCcell lines (Pedersen et al., 2003), and we showed noweven higher levels of expression in MCC, suggesting thata possible upregulation of CDC25B in MCC may occur.None of the other 10 genes were previously shown to beimplicated in MCC and SCLC biology. One strikingfinding was the differential expression of the alphasubunit of PKC (PRKCA) in MCC compared to SCLCcell lines. Protein kinase C is a key protein involved inthe regulation of cell growth and activation of the MAPkinase pathway (Buchner, 2000), and this finding is inkeeping with the observed expression for MAP2K3.Both overexpression and downregulation of PRKCAhave been previously observed in different human tumortypes and correlated with malignant transformation andproliferative activity of PRKCA (Benzil et al., 1992;Scaglione-Sewell et al., 1998). PRKCA could thus beinvolved both in MCC and SCLC, albeit through adifferent mechanism in each of these tumor types.Further investigation of this gene and other MCC andSCLC classifier genes such as ITGA3, HSPA2, CDH2,NRGN, GAK, PRKR and STK6 should elucidate theirrole in MCC and/or SCLC biology.Five of the 17 SAM-identified MCC and SCLC

classifier genes had a higher expression level in SCLC ascompared to MCC. Our data confirmed the previouslyreported expression of the neuroendocrine differentia-tion marker ASCL1 in (Classic) SCLCs and its lack ofexpression in MCC cell lines (Bhattacharjee et al., 2001;Garber et al., 2001; Leonard et al., 2002; Pedersen et al.,2003). The ID2 basic helix–loop–helix transcriptionfactor demonstrated higher levels of transcripts inClassic SCLCs compared to MCCs. Gene-expressionprofile analysis of small-cell lung cancer cells byPedersen et al. (2003) also detected expression of ID2in SCLC. ID2 plays a role in cell proliferation anddifferentiation and is able to disrupt the antiproliferativeeffects of retinoblastoma family members (Iavaroneet al., 1994). It is possible that disruption of the RB1pathway through increased expression of ID2 could bean important mechanism in neuroendocrine SCLCswhich may not occur in MCC. For the three remaininggenes (GPX2, IGFBP2 and TFAP4) expressed in SCLCcell lines but not in MCC, no previous involvement inSCLC has been described. Further investigation of thesegenes should clarify their role in SCLC or MCC biology.Our gene-expression profiling and clustering with only

17 MCC and SCLC classifier genes identified throughSAM analysis was extended through real-time RT–PCRon the original cell lines, as well as additional cell linesand tumor samples. Our results showed that the selectedgenes were able to effectively cluster the samples,providing an additional and simple test to differentiatebetween MCC and SCLC.


2739

Oncogene

In conclusion, we generated a gene expression-basedclassification of at least two biological and possiblyclinically distinct subgroups of MCC. Interestingly,some of the differentially expressed genes are typicallyclassifying Classic and Variant phenotypes of MCC aswell as SCLC. Further investigation could result in amore selective therapeutic treatment applicable for bothtumor types and improvement of patient outcome. Inaddition, we demonstrated for the first time a highdegree of similarity at the gene-expression level betweenMCC and SCLC. Furthermore, we were able to identifya subset of genes by supervised analysis, which may behelpful in the differential diagnosis of MCC and SCLC.Our study also serves as the first step towards a furtherdetailed study of differentially expressed genes involvedin cell proliferation, signal transduction and neuro-transmission, finally leading to more insight into thecomplex and heterogeneous biology of MCC andSCLC.

Materials and methods

Cell lines

Merkel cell carcinoma cell lines MCC5, MCC6, MCC13,MCC14/1, MCC14/2 and MCC26 were established at theQueensland Radium Institute Research Unit, Queensland,Australia and have been described in detail previously byLeonard et al. (1993, 1995a). MCC cell line UISO wasdescribed by Ronan et al. (1993) and MKL-1 by Rosen et al.(1987). MKL-2 was established at the Robert H LurieComprehensive Cancer Center, IL, USA and reported byVan Gele et al. (2002). T95-45 was established at the Center forMedical Genetics, Ghent, Belgium. Most cell lines werepreviously analysed by CGH and/or multiplex fluorescencein situ hybridization (Van Gele et al., 1998, 2002). Small-celllung carcinoma cell lines NCI-H69 and NCI-H146 wereobtained from the American Type Culture Collection, COR-L88 was a gift from Dr P Twentyman, Cambridge, UK andGLC4 was a gift from Dr Marc Maliepaard, Amsterdam, TheNetherlands. The morphological characteristics, growth beha-vior and subgroup classification (Classic versus Variant) of thecell lines are summarized in Table 1. All cell lines were grownto 70% confluency in RPMI 1640 (Invitrogen) supplementedwith antibiotics, 10% fetal calf serum and 1% L-glutamine.Cells from 10 tissue culture flasks (75 cm2) were pelleted, quickfrozen in liquid N2 and stored at �801C.

cDNA array hybridization

Total RNA from cell lines was extracted using Proteinase Kand phenol/chloroform (Sigma), followed by a sodium acetateprecipitation in ethanol (MCC13, MCC14/1, MCC14/2 andMCC26) or the Atlas Pure Total RNA Labelling System(Clontech – BD Biosciences). The resuspended RNA wassubsequently DNase I (Roche) (2U/50 mg) treated. The qualityand integrity of the Dnase-treated RNA were checked byethidium bromide agarose gel electrophoresis. Expressionanalysis was performed using the Atlas Human 1.2 (7850-1)and Atlas Human Cancer 1.2 (7851-1) nylon arrays (Clontech– BD Biosciences). Both filters contained 1176 genes, of which461 were present on both arrays. For each sample, 12.5mg oftotal RNA was used in the cDNA probe synthesis with[a-32P]dATP (NEN Life Science Products) and performed

according to the Clontech Atlas cDNA Expression ArraysUser Manual. Purification of the probe, hybridization andwashes were performed by following the manufacturer’sinstructions. Each RNA sample was simultaneously hybridizedto both filters in the same hybridization bottle. After thewashes, membranes were exposed for one to three nights tophosphoimager plates and scanned with a PhosphoImagerSystem using ImageQuaNT (Molecular Dynamics –Amersham Biosciences).

Analysis of cDNA arrays

The scanned gel images were converted to a 16-bit taggedimage file format. Signal intensities were quantified using theVisualGrid software version 2.1 (GPC Biotech). The Arra-yAn2 software (T. Boonefaes, P. Rottiers, and J. Grooten.ArrayAn2: optimized algorithms for primary data analysis ofcDNA arrays, manuscript in preparation) was used for furtherprimary data analysis. In short, the spot intensities werecorrected for the local background signal intensity, followedby a spot quality-control step to exclude spots influenced byintense signals of adjacent spots. The detection limit forexpression values above background was calculated based onthe variation of the local background intensity. Constitutivegenes were selected (50% of the spots showing the lowestcoefficient of variation over all arrays) and used fornormalization.

Expression data analysis

Genes with an expression value above the background level inat least six of the analysed samples were selected for furtheranalysis. This resulted in a total of 1083 genes, of which 412were common genes (i.e. 206 genes were present on both arrays(Human and Human Cancer 1.2). Cluster and Treeviewsoftware were used for unsupervised hierarchical clusteringand visualization of the data (Eisen et al., 1998). Prior toclustering, genes were mean centered and the expression datamatrix was log transformed (base 2). Subsequently, completelinkage clustering using Spearman rank correlation coefficientas similarity metric was performed to the samples and genes.The complete expression data matrix is available as a tabdelimited file from the authors on request.We used the Significance Analysis of Microarrays or SAM

algorithm (Tusher et al., 2001), which allows supervisedidentification of significantly differentially expressed genesbetween predefined sample groups. In order to include lessrepresentative genes for the SAM analysis, the filter thresholdwas lowered by including genes expressed above backgroundin at least four of the analysed samples (1365 genes).

Real-time quantitative RT–PCR

In all, 25 SAM identified genes were quantified by real-timequantitative RT–PCR on the same 14 RNA samples as forarray hybridizations. In order to validate the array gene-expression data, the normalized array and real-time PCR datawere each mean centered for the genes. The Spearman rankcorrelation coefficient was then calculated between the arraygene-expression levels and real-time PCR expression levels foreach gene using the Statistical Package for the Social Sciences(SPSS) Version 11.0 software. Primer sequences for all 25genes were designed with Primer Express 1.0 software (AppliedBiosystems) using the default TaqMan parameters, withmodified minimum amplicon length requirements (75 bp).The primer sequences are submitted in a public database(RTPrimerDB) for real-time PCR primers and probes (Pattynet al., 2003) (gene: primer-ID; ASCL1: 373, GPX2: 346, ID2:


2740

Oncogene

102, IGFBP2: 349, FLT1: 348, TFAP4: 347, EGR1: 389,BDNF: 352, HSPA2: 390, ITGA3: 351, PRKCA: 350, PRKR:354, MAP2K3: 388, CDH2: 387, ILK: 353, CDC25B: 355,VIM: 356, FOSL1: 357, AXL: 360, CCND1: 87, MYC: 18,SPINT2: 358, BTG2: 361, JUP: 359, MLP: 362). In order toconfirm the classification potential of the above-mentionedgenes, real-time PCR analysis was extended to two furtherMCC cell lines, 10 MCC tumors, 12 additional SCLC cell linesand two SCLC tumors. Tumor samples (MCCT1, T2, T3,SCLCT1 and T2) were collected at the Department ofDermatology or Pathology, Ghent University Hospital,Ghent, Belgium, and tumor samples MCCT4, T5, T6, T7,T8, T9 and T10 were collected at the Department ofPathology, University Hospital, Leuven, Belgium. Tumorbiopsies were homogenized with an Ultra-Turrax T25 (IKA-Werke) in 2ml lysis buffer (Qiagen). Total RNA of biopsieswas extracted using the RNeasy Midi Kit (Qiagen), accordingto the manufacturer’s instructions. Total RNA from MCC cellline MCC19 (Type II, suspension cell line with three-dimensional loose colonies thought to be Variant (Leonardet al., 2002)) was isolated at the Queensland Radium InstituteResearch Unit using Total RNA Isolation Reagent (AppliedBiotechnologies) from cells in exponential growth. The RNAof SCLC cell lines NCI-H446, POVD and AFL was a gift fromDr G Sozzi (Milan, Italy), RNA of GLC1, GLC7, GLC28,GLC36 and GLC45 was kindly provided by Dr K Kok(Groningen, The Netherlands) and RNA of NCI-H60, NCI-H82, NCI-H250, NCI-N464 and MCC cell line MKL-1(subclone 2) (Type III, suspension cell line with two-dimensional loose colonies classified as Classic) was a giftfrom H Salwen (IL, USA). All RNAs were quantified using theRibogreen reagent (Molecular Probes) on a TD-360 fluorom-eter (Turner Designs). The relative gene-expression levels weredetermined using an optimized two-step SYBR green I RT–PCR assay, as described by Vandesompele et al. (2002a). The

standard curve method (serial dilutions of a cDNA mixturecontaining two SCLC and two MCC samples) or thecomparative Ct method was used for quantification. PCRreagents were obtained from Eurogentec as SYBR Green Imastermixes, and used according to the manufacturer’sinstructions. PCR reactions were run on an ABI Prism 5700Sequence Detection System (Applied Biosystems). To correctfor differences in RNA quantities and cDNA synthesisefficiency, relative gene-expression levels were normalizedusing the geometric mean of five housekeeping genes (UBC,HPRT1, GAPD, TBP and HMBS) according to Vandesom-pele et al. (2002b). In order to perform hierarchical clustering,a real-time-based expression matrix was created by dividingthe normalized gene-expression level of each gene by itsgeometric mean across all samples, and data were subsequentlylog transformed (base 10).

Acknowledgements

This work was supported by GOA Grant 12051397, FWOGrant G.0028.00 and the Queensland Cancer Fund and theQueensland Radium Institute Research Unit. Anthony LCook is supported by a University of Queensland Mid YearScholarship. Jo Vandesompele is sponsored by VEO-grant011V1302. Nadine Van Roy is a postdoctoral researcher of theFund for Scientific Research, Flanders. This paper presents theresearch results of the Belgian program of Interuniversity Polesof attraction initiated by the Belgian State, Prime Minister’sOffice, Science Policy Programming. The scientific responsi-bility is assumed by us. We would like to thank Drs MMHughes, VM Hinkley, RW Allison, W Cockborn, TJ Harrisand O Williams for their support in collecting the QueenslandMCC specimens, from which the MCC cell lines wereestablished, and H Salwen for providing MCC cell lineMKL-2.

References

Baumann KI, Senok SS, Chan E and Yung WH. (2000).Merkel Cells, Merkel Cell Carcinoma and Neurobiology ofthe Skin, Suzuki H and Ono T (eds) Elsevier Science B.V.:Tokyo, pp 73–81.

Benzil DL, Finkelstein SD, Epstein MH and Finch PW.(1992). Cancer Res., 52, 2951–2956.

Bhattacharjee A, Richards WG, Staunton J, Li C, Monti S,Vasa P, Ladd C, Beheshti J, Bueno R, Gillette M, Loda M,Weber G, Mark EJ, Lander ES, Wong W, Johnson BE,Golub TR, Sugarbaker DJ and Meyerson M. (2001). Proc.Natl. Acad. Sci. USA, 98, 13790–13795.

Borges M, Linnoila RI, van de Velde HJ, Chen H, Nelkin BD,Mabry M, Baylin SB and Ball DW. (1997). Nature, 386,852–855.

Botchkarev VA, Kief S, Paus R and Moll I. (1999). J. Invest.Dermatol., 113, 691–692.

Broers JL, Carney DN, de Ley L, Vooijs GP and RamaekersFC. (1985). Proc. Natl. Acad. Sci. USA, 82, 4409–4413.

Broers JL, Carney DN, Klein Rot M, Schaart G, Lane EB,Vooijs GP and Ramaekers FC. (1986). J. Cell Sci., 83,37–60.

Buchner K. (2000). J. Cancer Res. Clin. Oncol., 126, 1–11.Carney DN, Gazdar AF, Bepler G, Guccion JG, MarangosPJ, Moody TW, Zweig MH and Minna JD. (1985). CancerRes., 45, 2913–2923.

Carroll P, Lewin GR, Koltzenburg M, Toyka KV andThoenen H. (1998). Nat. Neurosci., 1, 42–46.

Cook AL, Pollock PM, Welch J, Walsh MD, Bowman RV,Baumann KC, Hayward NK and Leonard JH. (2001). Int. J.Cancer, 93, 361–367.

Eisen MB, Spellman PT, Brown PO and Botstein D. (1998).Proc. Natl. Acad. Sci. USA, 95, 14863–14868.

Fong KM, Sekido Y and Minna JD. (1999). J. Thorac.Cardiovasc. Surg., 118, 1136–1152.

Garber ME, Troyanskaya OG, Schluens K, Petersen S,Thaesler Z, Pacyna-Gengelbach M, van de Rijn M, RosenGD, Perou CM, Whyte RI, Altman RB, Brown PO, BotsteinD and Petersen I. (2001). Proc. Natl. Acad. Sci. USA, 98,13784–13789.

Gazdar AF, Carney DN, Nau MM and Minna JD. (1985).Cancer Res., 45, 2924–2930.

Goessling W, McKee PH and Mayer RJ. (2002). J. Clin.Oncol., 20, 588–598.

Halata Z, Grim M and Bauman KI. (2003). Anat. Rec., 271A,225–239.

Iavarone A, Garg P, Lasorella A, Hsu J and Israel MA. (1994).Genes Dev., 8, 1270–1284.

Leonard JH and Bell JR. (1997). Australas. J. Dermatol., 38,91–98.

Leonard JH, Bell JR and Kearsley JH. (1993). Int. J. Cancer,55, 803–810.

Leonard JH, Cook AL, Nancarrow D, Hayward N, Van GeleM, Van Roy N and Speleman F. (2000). Cancer Detect.Prev., 24, 620–627.


2741

Oncogene

Leonard JH, Cook AL, Van Gele M, Boyle GM, Inglis KJ,Speleman F and Sturm RA. (2002). Int. J. Cancer, 101, 103–110.

Leonard JH, Dash P, Holland P, Kearsley JH and Bell JR.(1995a). Int. J. Cancer, 60, 100–107.

Leonard JH and Hayard N. (1997). Genes ChromosomesCancer, 20, 93–97.

Leonard JH, Ramsay JR, Kearsley JH and Birrell GW.(1995b). Int. J. Radiat. Oncol. Biol. Phys., 32, 1401–1407.

Levin NA, Brzoska PM, Warnock ML, Gray JW andChristman MF. (1995). Genes Chromosomes Cancer, 13,175–185.

Metz KA, Jacob M, Schmidt U, Steuhl KP and Leder LD.(1998). Graefe’s Arch. Clin. Exp. Ophthalmol., 236,

561–566.Miller RW and Rabkin CS. (1999). Cancer Epidemiol.

Biomarkers Prev., 8, 153–158.Minna JD, Roth JA and Gazdar AF. (2002). Cancer Cell, 1,49–52.

Moll I, Gillardon F, Waltering S, Schmelz M and Moll R.(1996). J. Cutan. Pathol., 23, 109–117.

Pattyn F, Speleman F, De Paepe A and VandesompeleJ. (2003). Nucleic Acids Res., 31, 122–123.

Petersen I, Langreck H, Wolf G, Schwendel A, Psille R, VogtP, Reichel MB, Ried T and Dietel M. (1997). Br. J. Cancer,75, 79–86.

Pedersen N, Mortensen S, Sorensen SB, Pedersen MW,Rieneck K, Bovin LF and Poulsen HS. (2003). CancerRes., 63, 1943–1953.

Popp S, Waltering S, Herbst C, Moll I and Boukamp P.(2002). Int. J. Cancer, 99, 352–360.

Ratner D, Nelson BR, Brown MD and Johnson TM. (1993).J. Am. Acad. Dermatol., 29, 143–156.

Ried T, Petersen I, Holtgreve-Grez H, Speicher MR, SchrockE, du Manoir S and Cremer T. (1994). Cancer Res., 54,1801–1806.

Ronan SG, Green AD, Shilkaitis A, Huang TS and Das GuptaTK. (1993). J. Am. Acad. Dermatol., 29, 715–722.

Rosen ST, Gould VE, Salwen HR, Herst CV, Le Beau MM,Lee I, Bauer K, Marder RJ, Andersen R, Kies MS, Moll R,Franke WW and Radosevich JA. (1987). Lab. Invest., 56,302–312.

Scaglione-Sewell B, Abraham C, Bissonnette M, Skarosi SF,Hart J, Davidson NO, Wali RK, Davis BH, Sitrin M andBrasitus TA. (1998). Cancer Res., 58, 1074–1081.

Schmidt U, Muller U, Metz KA and Leder LD. (1998). Am. J.Dermatopathol., 20, 346–351.

Tazaki M, Tazaki Y, Inoue T and Shimono M. (2000). MerkelCells, Merkel Cell Carcinoma and Neurobiology of the Skin,Suzuki H and Ono T (eds) Elsevier Science B.V.: Tokyo, pp.63–71.

Tusher VG, Tibshirani R and Chu G. (2001). Proc. Natl. Acad.Sci. USA, 98, 5116–5121.

Vandesompele J, De Paepe A and Speleman F. (2002a). Anal.Biochem., 303, 95–98.

Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N,De Paepe A and Speleman F. (2002b). Genome Biol., 3,research0034.001–research0034.0011.

Van Gele M, Kaghad M, Leonard JH, Van Roy N, NaeyaertJM, Geerts ML, Van Belle S, Cocquyt V, Bridge J, Sciot R,De Wolf-Peeters C, De Paepe A, Caput D and Speleman F.(2000). Br. J. Cancer, 82, 823–826.

Van Gele M, Leonard JH, Van Limbergen H, Van Roy N,Van Belle S, Cocquyt V, De Paepe A, Salwen H andSpeleman F. (2002). Int. J. Cancer, 101, 137–145.

Van Gele M, Speleman F, Vandesompele J, Van Roy N andLeonard JH. (1998). Cancer Res., 58, 1503–1508.

Zajchowski DA, Bartholdi MF, Gong Y, Webster L, Liu HL,Munishkin A, Beauheim C, Harvey S, Ethier SP andJohnson PH. (2001). Cancer Res., 61, 5168–5178.

Zhang L, Cilley RE and Chinoy MR. (2000). J. Surg. Res., 93,108–119.


2742

Oncogene

gene-expression profiling reveals distinct expression patterns for classic versus variant merkel...

Documents