Integrated Genomic and Transcriptional Profiling Identifies Chromosomal Loci with Altered Gene Expression in Cervical Cancer

Abstract

For a better understanding of the consequences of recurrent chromosomal alterations in cervical carcinomas, we integrated genome-wide chromosomal and transcriptional profiles of 10 squamous cell carcinomas (SCCs), 5 adenocarcinomas (AdCAs) and 6 normal controls. Previous genomic profiling showed that gains at chromosome arms 1q, 3q, and 20q as well as losses at 8q, 10q, 11q, and 13q were common in cervical carcinomas. Altered regions spanned multiple megabases, and the extent to which expression of genes located there is affected remains unclear. Expression analysis of these previously chromosomally profiled carcinomas yielded 83 genes with significantly differential expression between carcinomas and normal epithelium. Application of differential gene locus mapping (DIGMAP) analysis and the array CGH expression integration tool (ACE-it) identified hotspots within large chromosomal alterations in which gene expression was altered as well. Chromosomal gains of the long arms of chromosome 1, 3, and 20 resulted in increased expression of genes located at 1q32.1-32.2, 3q13.32-23, 3q26.32-27.3, and 20q11.21-13.33, whereas a chromosomal loss of 11q22.3-25 was related to decreased expression of genes located in this region. Overexpression of DTX3L, PIK3R4, ATP2C1, and SLC25A36, all located at 3q21.1-23 and identified by DIGMAP, ACE-it or both, was confirmed in an independent validation sample set consisting of 12 SCCs and 13 normal ectocervical samples. In conclusion, integrated chromosomal and transcriptional profiling identified chromosomal hotspots at 1q, 3q, 11q, and 20q with altered gene expression within large commonly altered chromosomal regions in cervical cancer.

INTRODUCTION

Although the implementation of cervical screening programs has resulted in a decrease in the incidence of cervical cancer in developed countries, it still remains the second most common cancer in women worldwide. Histologically, cervical carcinomas can be classified into multiple types, including squamous cell carcinomas (SCCs) in 70-80% and adenocarcinomas (AdCAs) in 5-10% of cases (Fu and Reagan, 1989; Pisani et al., 2002).

Persistent infection with high-risk (i.e., oncogenic) human papillomavirus (hr-HPV) is causally involved in cervical carcinogenesis. Deregulated expression of the viral oncoproteins E6 and E7 interferes with cell cycle control due to their ability to induce degradation of the tumor suppressor proteins TP53 and RB1, respectively. This results in uncontrolled cell proliferation and accumulation of specific (epi) genetic changes in the host cell genome, driving progression to a malignant phenotype (zur Hausen, 2000; Steenbergen et al., 2005). Only a minority of all women infected with hr-HPV will ultimately develop cervical cancer, emphasizing the multistep nature of cervical carcinogenesis. More insight into the (epi) genetic alterations that occur during cervical carcinogenesis is essential for the discovery of genes that are biologically relevant in cervical cancer development and progression.

Microarray-based technology enables high-resolution genome-wide screening for altered chromosomal regions (array-based comparative genomic hybridization (array CGH)) as well as altered gene expression. Recurrent chromosomal alterations found by (array) CGH involved many losses (chromosome arms 2q, 3p, 4p, 5q, 6q, 8q, 10q, 11q, 13q, 18q) and gains (1, 3q, 5p, 8q, 20q, and Xq) (Heselmeyer et al., 1996, 1997; Dellas et al., 1999; Allen et al., 2000; Hidalgo et al., 2000,2005; Yang et al., 2001; Umayahara et al., 2002; Rao et al., 2004; Wilting et al., 2006). Even when using high-resolution array CGH, affected regions are still multiple megabases in size and contain substantial amounts of genes (Wilting et al., 2006). The extent to which these recurrent chromosomal alterations have functional relevance by resulting in abnormal expression of the involved genes is still largely unknown. Similarly, several studies have been conducted to determine transcriptional profiles of cervical carcinomas, yielding large numbers of genes with altered expression (Cheng et al., 2002; Chen et al., 2003; Sopov et al., 2004; Santin et al., 2005; Chao et al., 2006; Wong et al., 2006; Gius et al., 2007). Identification of tumor cell specific and functionally relevant genes from expression array analysis is hampered by the large number of secondary or responsive changes in gene expression not only in tumor cells, but in interspersed stromal and invading immune cells as well. Indeed, previous studies using microdissection or in situ hybridization approaches showed that a substantial subset of differentially expressed genes in cervical carcinomas was specific for stromal cells (Chen et al., 2003; Gius et al., 2007).

To identify genes with altered expression related to chromosomal alterations we integrated genome-wide mRNA expression profiles and previously generated high-resolution chromosomal signatures of cervical carcinomas, for which RNA and DNA were isolated from the same frozen sections (Wilting et al., 2006). For this purpose two innovative statistical approaches were used, namely differential gene locus mapping (DIGMAP) and the array CGH and expression integration tool (ACE-it). Whereas DIGMAP compares expression of windows of genes located next to each other between carcinomas and normal epithelium, ACE-it determines the association between chromosomal alterations and gene expression by comparing the expression of a particular gene between carcinomas that show an alteration at the locus of that gene and carcinomas without a chromosomal alteration (Yi et al., 2005; van Wieringen et al., 2006). Although only a limited number of samples were included in this study, the statistical analyses used do correct for this small sample size. Integration resulted in the identification of five chromosomal hotspots where common chromosomal alterations were associated with changes in gene expression. Differential expression of a subset of genes identified by this integrated approach was confirmed in an independent validation sample set using real-time RT-PCR.

MATERIALS AND METHODS

Tissue Specimens

We used frozen specimens of SCCs and AdCAs, all of which were collected during the course of routine clinical practice at the Department of Obstetrics and Gynaecology at the VU University Medical Center (Amsterdam) (Table 1). Normal epithelial control samples were obtained from histologically normal frozen biopsies or smears of non-cancer patients undergoing hysterectomy (Table 2). Tissue biopsies were quick-frozen and stored in liquid nitrogen immediately after completion of the surgical procedure. Cervical smears were collected directly after hysterectomy using brushes. Brushes were stored in TRIzol Reagent (Life Technologies, Inc. Breda, The Netherlands), which stabilizes RNA molecules, immediately after a cytological specimen was made for microscopical examination. For microarray analysis 10 SCCs, 5 AdCAs, and 6 samples of normal epithelium were used. Normal controls included five samples of normal squamous epithelium (NSE) and one sample of RNA from normal endocervical columnar epithelium (NCE), which are further specified in Table 2. RNA isolated from two different biopsies of the same SCC was hybridized as a biological replicate. One of the normal ectocervical epithelial control samples was hybridized twice as a technical replicate to determine technical variation. Real-time RT-PCR was performed on all carcinomas included in microarray analysis as well as an independent set of 12 SCCs and 13 histologically normal squamous epithelial samples of the ectocervix (Tables 1 and 2).

TABLE 1.

Summary of Clinical Data and HPV Typing of Carcinomas Analyzed

Sample	Differentiation grade	Tumour stage	HPV type	Age (years)	Technique
SCC 2	PD	IB	16	39	MA/qRT-PCR
SCC 3	Unknown	IIA	16	78	qRT-PCR
SCC 4	MD	IIA	67	62	MA/qRT-PCR
SCC 11	PD	IB	16	75	qRT-PCR
SCC 12	PD	IB	16	44	MA/qRT-PCR
SCC 15	PD	IB	16	47	MA/qRT-PCR
SCC 20	PD	IB	45	34	qRT-PCR
SCC 27	MD	IB2	69	49	MA/qRT-PCR
SCC 28	MD	IB2	35	48	MA/qRT-PCR
SCC 29	MD	IBI/IIB	16	48	qRT-PCR
SCC 32	MD	MA	16	37	MA/qRT-PCR
SCC 36	MD	MA	16	72	MA/qRT-PCR
SCC 38	MD	IBI/IIA	16	51	MA/qRT-PCR
SCC 39	PD	IBI	33	40	MA/qRT-PCR
SCC 40	PD	IBI	45	57	qRT-PCR
SCC 42	PD	IBI	16	79	qRT-PCR
SCC 44	PD	Unknown	16	51	qRT-PCR
SCC 45	PD	G3	45	52	qRT-PCR
SCC 46	PD	G2	31	65	qRT-PCR
SCC 49	PD	G2	16	32	qRT-PCR
SCC 51	PD	Unknown	16	25	qRT-PCR
SCC 54	PD	IBI	33	60	qRT-PCR
AdCA 1	WD	IB	16	34	MA/qRT-PCR
AdCA 2	WD	IB	16	35	qRT-PCR
AdCA 7	MD	IB	16	31	MA/qRT-PCR
AdCA 10	MD	IB2	16	39	MA/qRT-PCR
AdCA 11	MD	IB1	18	64	MA/qRT-PCR
AdCA 12	MD	IB2	18	41	MA/qRT-PCR
AdCA 14	Unknown	IIB	18	42	qRT-PCR
AdCA 15	PD	IB1	18	39	qRT-PCR

SCC, squamous cell carcinoma; AdCA, adenocarcinoma; WD, well differentiated; MD, moderately differentiated; PD, poorly differentiated; MA, micro-array; qRT-PCR, real-time RT-PCR.

TABLE 2.

Summary of Clinical Data and HPV Typing of Normal Samples Analyzed

Sample	Origin	HPV type	Age (yrs)	Technique	Type of material
NCE	(endo)cervix	-;-;-;-;58/70	43;75;59;43;28	MA	pool of 5 smears
NSE1A/B	(ecto)cervix	-;-;-;-	59;47;48;52	MA	pool of 4 smears
NSE2	(ecto)cervix	-;58;-	40;56;52	MA	pool of 3 frozen tissues
NSE3	uvula	-	43	MA	frozen tissue
NSE4	uvula	-	52	MA	frozen tissue
NSE5	uvula		48	MA	frozen tissue
NSE6	(ecto)cervix	35	36	qRT-PCR	frozen tissue
NSE7	(ecto)cervix	-	45	qRT-PCR	frozen tissue
NSE8	(ecto)cervix	-	36	qRT-PCR	frozen tissue
NSE9	(ecto)cervix	-	39	qRT-PCR	frozen tissue
NSE10	(ecto)cervix	16	30	qRT-PCR	frozen tissue
NSE11	(ecto)cervix	-	29	qRT-PCR	frozen tissue
NSE12	(ecto)cervix	16	33	qRT-PCR	frozen tissue
NSE13	(ecto)cervix	16	30	qRT-PCR	frozen tissue
NSE14	(ecto)cervix	-	34	qRT-PCR	frozen tissue
NSE15	(ecto)cervix	70	31	qRT-PCR	frozen tissue
NSE16	(ecto)cervix	16	31	qRT-PCR	frozen tissue
NSE17	(ecto)cervix	-	47	qRT-PCR	frozen tissue
NSE18	(ecto)cervix	42	33	qRT-PCR	frozen tissue

NCE, normal columnar epithelium; NSE, normal squamous epithelium; MA, microarray; qRT-PCR, real-time RT-PCR.

This study followed the ethical guidelines of the Institutional Review Board of the VU University Medical Center.

RNA Isolation

Total RNA was isolated from all samples using TRIzol Reagent according to the manufacturers' instructions.

Frozen tissue specimens were mounted using Tissue-Tek O.C.T. (Sakura Finetek Europe B.V., Zoeterwoude, The Netherlands) and were serially sectioned at -20°C according to the sandwich method in which the outer sections were H&E stained for histological assessment by an experienced pathologist and sequential series of in-between cryo-sections were used for RNA extraction. Only tumor specimens containing ≥70% tumor cells were included. If necessary (< 70% epithelial cells) frozen biopsies of normal samples were first enriched for epithelial cells by means of laser capture microdissection using a Leica ASLMD microscope (Leica, Heidelberg, Germany). Of these samples 10 μm thick cryosections were cut and mounted on PEN foil coated slides (Leica). Sections were stained with Mayers' hematoxylin for 1 min and completely dehydrated by rinsing the slides in 50, 70 and 100% ethanol followed by an 8 min incubation in xylene. Slides were stored in closed 50 ml tubes at -80°C to avoid rehydration of the tissue and subsequent RNA degradation until microdissection was performed. In a pilot experiment using this protocol RNA integrity was assessed using RNA Nano Chips on an Agilent 2100 Bioanalyzer (Agilent Technologies, Waldbronn, Germany). RNA quality of non-micro-dissected samples was checked by agarose gel electrophoresis.

HPV Typing

DNA was isolated after RNA extraction with TRIzol following the manufacturers' recommendations (Wilting et al., 2006). HPV typing was performed using the general primer GP5+/6+ polymerase chain reaction (PCR) followed by reverse line blot, as described previously (van den Brule et al., 2002). Results are shown in Table 1.

Expression Microarray Analysis

We used oligo arrays containing the 19K Human Release 1.0 oligonucleotide library, designed by Compugen (San Jose, CA) and obtained from Sigma-Genosys (Zwijndrecht, The Netherlands). In total 18,861 60-mer oligos representing 17,260 unique genes were spotted on the array. Arrays were produced at the VUmc Microarray facility (Braakhuis et al., 2006; Muris et al., 2007). In short, cDNA was prepared from 15 μg of total sample RNA and 15 μg of Universal Human Reference RNA (Stratagene, La Jolla, California, USA) using oligo-dT₂₀-VN primer (Invitrogen, Breda, The Netherlands) and coupled to Cy3 (test sample) or Cy5 (reference). Slides were pre-hybridized for 45 min at 37°C with a pre-hybridization solution containing 30 μg salmon sperm DNA (Gibco, Breda, The Netherlands), 12 μg poly(A) (Pharmacia), 60 μg yeast tRNA (Sigma), and 24 μg Cot-1 DNA (Invitrogen) dissolved in 127 μl hybridization mix (0.2% SDS, 8% glycerol, 50% formamide, and 0.1% dextrane sulfate in 2× SSC). Pre-hybridization was followed by probe hybridization for 14 h at 37°C. Both pre-hybridization and hybridization were performed in HybStation 12 (Perkin-Elmer Life Sciences). Hybridized arrays were then scanned using ScanArray Express (Perkin-Elmer Life Sciences, Zaventum, Belgium) and quantified in ImaGene 5.6.1 software (BioDiscovery Ltd, Marina del Rey, CA) using default settings. Flagged spots were excluded from further analysis.

The entire dataset described here (including previously described array CGH data (Wilting et al., 2006)) is available from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/projects/geo/) through series accession number GSE6473.

Microarray Data Analysis

Data pre-processing

Microarray data were normalized using Lowess regression. When both intensities were below 50, hybridization was assumed inefficient and the ratio value was considered `missing.' The cutoff “50” is based on technical reproducibility experiments on our platform. To obtain more stable ratios, intensity values below 50 were substituted by 50 when the other channel was above 50. Genes with missing ratio values in more than 20% of arrays were excluded from analysis. Remaining missings were imputed using K-nearest neighbor imputation.

Cluster analysis

Unsupervised hierarchical clustering was performed on all genes that met our quality criteria in Spotfire DecisionSite 7.3 (Spotfire Inc., Göteborg, Sweden) using the Euclidean distance measure and complete linkage.

LIMMA differential gene expression analysis

Differentially expressed genes were determined using the Linear Models for MicroArray data (LIMMA) statistical package from BioConductor (www.bioconductor.org) (Smyth, 2004). Genes with False Discovery Rate (FDR) below 0.05 were considered statistically significant. Since we included three normal squamous epithelial controls of non-cervical origin, we applied an additional conservative selection criterion in which only significant genes that showed a fold change (FC) of at least two to the normal cervical controls were considered truly differentially expressed.

DIGMAP analysis

Differential gene locus mapping (DIGMAP) analysis was carried out to investigate aberrant expression patterns of groups of genes based on their chromosomal location as described by Yi et al., (2005). Statistically significant loci were identified using the TTest, which generated confidence (T) scores by log-transformation of the reciprocal P value. An automated scanning method employing sliding window analysis was used to find so-called differential flag regions (DFRs). In this study, regions with a T score greater than 2 SDs from the mean of total T scores for all sliding windows and consisting of at least five consecutive significant sliding windows were considered a DFR if at least 70% of genes located in this region showed concordant differential expression (log₂(FC) > 0 indicates increased expression; log₂(FC) < 0 indicates decreased expression).

ACE-it

In the carcinomas mRNA expression levels of genes were directly linked to gene copy numbers as measured by array CGH using the Array CGH Expression integration tool (ACE-it) (van Wieringen et al., 2006). This tool identifies genes with differential expression between carcinomas that showed either a gain or a loss and carcinomas without a chromosomal alteration. The relation between gene expression and gene dosage is tested for all genes present on the oligo-array which fulfil the grouping criteria, using the one-sided Wilcoxon-rank test followed by Benjamini-Hochbergs' multiplicity correction for FDR control. We allowed for five contaminating samples within the groupings and considered gene dosage and gene expression linked for genes with an FDR below 0.2.

Real-Time RT-PCR

Intron-flanking primers were selected for nine genes (i.e., DEK, SEMP1, ITGAV, SYCP2, MAL, ATP2C1, SLC25A36, PIK3R4, and DTX3L) using Primer Express 2.0 (Applied Biosystems, Warrington, United Kingdom) or taken from the RT-primer database of the University of Gent in Belgium (http://medgen.ugent.be/rtprimerdb) (Table 3). Total RNA was reverse transcribed using oligo-d T₂₀ primer (Invitrogen) and the resulting cDNA was used for real-time PCR on the ABI/Prism 7700 Sequence Detector System (Taqman-PCR; Applied Biosystems). cDNA corresponding to 25 ng of total RNA was amplified in a total reaction volume of 25 μl containing 12.5 μl 2× Sybr Green master mix (Perkin-Elmer/Applied Biosystems) and 0.5 μM primers. Relative expression values for each gene were determined from a standard curve, using primary keratinocytes (EK) or the cervical cancer cell line SiHa. The slopes of all standard curves were between -3.1 and -3.9 with a correlation coefficient of at least 0.98, ensuring sufficient efficiency of all experiments. All PCR experiments were performed in duplicate (delta Ct ≤1.5 between replicates) and mean values were used for calculations.

TABLE 3.

Primer Sequences Used for Real-Time RT-PCR

Gene	Primers (5′-3′)	Size (bp)	Annealing (°C)
DEK	F: AGAGAGGTTGACAATGCAAGTCT	71	56
	R: TCTGCCCCTTTCCTTGTG
SEMP1	F: GATGAGGATGGCTGTCATTG	75	55
	R: TACCATGCTGTGGCAACTAAA
ITGAV	F: TTGTTGCTACTGGCTGTTTTG	89	60
	R: TCCCTTTCTTGTTCTTCTTGAG
SYCP2	F: ACAGAAAACTGAAGACTACCTTTGTTA	88	55
	R: TCATCAGCTCCATTCAAATTAAA
MAL	F: GCAAGACGGCTTCACCTACAG	74	59
	R: GCAGAGTGGCTATGTAGGAGAACA
ATP2C1	F: GGATGTTCAGCAGCTTTCACAA	93	59
	R: TCTGTAGCGACTTAATAATTTTCATCTTG
SLC25A36	F: CCAGTGTCAACCGAGTAGTGTCT	76	57
	R: AGGAACGAGGCCCTTCTTTT
PIK3R4	F: GACTGCTACAAAAACCCCATGTT	90	60
	R: CGGCACCATAACGTATCCATAA
DTX3L	F: CAGTGAAAGGGCAGCTAAGG	74	60
	R: GCACAGGTTTTTCGTCAACA

F, Forward primer; R, Reverse primer; ITGAV primer sequences were retrieved from Rogojina et al (Rogojina et al., 2003).

Statistical Analysis of Real-time RT-PCR Results

Linear (Pearson) correlation between microarray results and real-time RT-PCR values was determined. Expression levels as measured by real-time RT-PCR were compared between carcinomas and normal cervical epithelium, using the non-parametric Wilcoxon-rank test. Two-sided P values below 0.1 were considered statistically significant.

RESULTS

SCC Expression Profiles Differ from those of AdCAs

In this study, expression profiles of 10 SCCs and 5 AdCAs of the cervix were determined. To investigate the similarity between the overall expression profiles of these two histological subtypes of cervical cancer we performed unsupervised hierarchical clustering analysis with the expression values of all genes fulfilling the quality criteria (n = 12.831). This resulted in three clusters, one cluster containing only SCCs and two clusters containing AdCAs (Fig. 1). Within the SCC group, the biological replicates (SCC12A and SCC12B) clustered very closely together, indicating that the dataset is reliable. Interestingly, one of the AdCA groups (including AdCA1 and 12) clustered more closely to the SCC group than to the other AdCA group (including AdCA7, 10 and 11). This result suggests that expression profiles of a subset of the AdCAs included in this study are more similar to SCCs than to the other AdCAs. In concordance with this observation, the chromosomal profile of AdCA12 also showed alterations that are in general more specific for SCCs, including gains of 3q and 20q (Wilting et al., 2006). However, AdCAs still form a separate cluster, indicating that expression profiles differ between SCCs and AdCAs. Differences in both expression and chromosomal profiles between SCC and AdCA of the cervix have been described previously (Contag et al., 2004; Chao et al., 2006; Wilting et al., 2006).

Figure 1.

Dendrogram obtained by unsupervised hierarchical clustering of SCCs and AdCAs. SCC12A and B are biological replicates, respectively.

Identification of Differentially Expressed Genes

To determine which genes were differentially expressed in cervical carcinomas we compared the expression profiles of all tumors (n = 15) to those of normal epithelium (n = 6) using the LIMMA statistical package. In this way 39 differentially expressed genes were identified (24 upregulated genes and 15 downregulated genes). Since cluster analysis separated SCCs and AdCAs based on their overall expression profiles, we also conducted separate analyses for both tumor types. The comparison of SCCs (n = 10) with normal epithelium identified 55 overexpressed genes and 21 downregulated genes. These 76 genes included all but seven of the genes that were differentially expressed between all tumors and normal epithelium. When we compared only AdCAs (n = 5) to normal epithelium no significant differences in expression were found, which is possibly due to the small number of samples investigated. All genes showing differential expression between carcinomas and normal epithelium in the comparisons described above were combined into one list of 83 genes (58 upregulated and 25 downregulated genes) (Table 4). Significantly differential expression between SCCs and AdCAs was found for 16 genes, 8 of which were higher expressed in SCCs (Table 5).

TABLE 4.

Differentially Expressed Genes as Determined by LIMMA Statistical Analysis Between Carcinomas and Normal Epithelium

			All carcinomas		Only SCCs
GenBank Acc No	Gene symbol	Cytoband	FDR	FC (normal cervix)	FDR	FC (normal cervix)
Downregulated genes in cervical carcinomas compared to normal epithelial tissue
NM_014658	RAPIGAI	1p36.12	0.1665	2.3	0.0303	3.0
NM_004425	ECMI	1q21.2	0.0497	2.7	0.0590	2.2
NM_002371	MAL	2q11.1	0.0032	34.3	0.0039	33.1
J00129	FGB	4q31.3	0.3926	3.8	0.0486	6.6
AK001007	FLJ10145	5q23.2	0.0878	2.5	0.0363	3.0
NM_002770	PRSS2	7q34	0.0200	10.5	0.0214	10.6
NM_004063	CDH17	8q22.10.1	0.1067	2.6	0.0363	3.2
AL137555	C9orf88	9q34.11	0.0497	2.2	0.0624	2.0
AK022845	C9orf58	9q34.13	0.0122	2.6	0.0010	3.4
AL 122071	SLC16A9	10q21.2	0.1540	3.7	0.0363	5.4
NM_000543	SMPDI	11p15.4	0.1040	2.1	0.0454	2.4
M62402	IGFBP6	12q13.13	0.0243	5.8	0.0255	5.9
X07695	KRT4	12q13.13	0.0271	17.7	0.0363	15.3
NM_016039	CLE7	14q22.1	0.0497	2.4	0.0483	2.5
NM_002435	MPI	15q24.1	0.0123	4.5	0.0133	4.7
NM_001520	GTF3C1	16p12.1	0.0537	2.9	0.0448	3.1
AB045292	M83	16p13.3	0.1714	2.3	0.0043	3.2
NM_017839	AYTL1	16q12.2	0.0580	2.9	0.0363	3.2
NM_006373	VAT1	17q21.31	0.0271	2.4	0.0308	2.4
NM_002476	MYL4	17q21.32	0.0497	2.8	0.0483	2.8
NM_014921	LPHN1	19p13.12	0.0394	1.9	0.0311	2.0
NM_020428	CTL2	19p13.2	0.0521	2.1	0.0354	2.3
NM_001928	DF	19p13.3	0.0464	9.2	0.0363	9.9
NM_006087	TUBB4	19p13.3	0.0497	2.1	0.0624	2.0
NM_006307	SRPX	23p11.4	0.0497	4.8	0.0549	4.8
Upregulated genes in cervical carcinomas compared to normal epithelial tissue
AB050716	TINAGL1	1p35.2	0.0497	4.6	0.0043	6.3
NM_003579	RAD45L	1p34.1	0.0499	1.9	0.0150	2.2
NM_OO6417	IF144	1p31.1	0.0394	4.7	0.0212	5.6
NM_004428	EFNA1	1q22	0.0497	2.6	0.0549	2.4
AK024944	FLJ21291	1q32.1	0.0249	2.6	0.0067	2.9
AL137717	DKFZp434J1630	2p11.2	0.0499	2.7	0.0576	2.7
NM_004688	NMI	2q23.3	0.1365	3.3	0.0486	4.0
NM_002210	ITGAV	2q32.1	0.0090	2.4	0.0106	2.3
NM_007315	STAT1	2q32.2	0.0654	3.3	0.0043	4.5
NM_000090	COL3A1	2q32.2	0.0543	20.5	0.0354	27.8
AK000160	MYO1B	2q32.3	0.0497	2.6	0.0034	3.6
NM_000094	COL7A1	3p21.31	0.2214	2.1	0.0392	3.6
AK025135	DTX3L	3q21.1	0.03	3.1	0.0303	3.5
AL122079	CCDC14	3q21.1	0.031	2.5	0.0311	2.6
NM_004526	MCM2	3q21.3	0.0497	2.3	0.0039	3.0
Y08991	PIK3R4	3q21.3	0.2749	1.7	0.0450	2.1
NM_014382	ATP2C1	3q21.3	0.1186	2.0	0.0454	2.4
NM_018155	SLC25A36	3q23	0.1792	1.8	0.0483	2.4
NM_003071	SMARCA3	3q24	0.0592	1.9	0.0448	2.2
AF086432	GPR87	3q25.1	0.4196	8.6	0.0491	19.5
NM_003875	GMPS	3q25.31	0.0338	1.8	0.0150	2.0
NM_016625	RSRC1	3q25.32	0.0871	2.0	0.0354	2.4
NM_003722	TP73L	3q28	0.4562	7.9	0.0363	21.4
AF101051	SEMP1	3q28	0.7625	4.4	0.0491	13.7
AL049229	DKFZp564O1016	3q29	0.1792	2.0	0.0317	2.6
Z24724	ATP13A3	3q29	0.0090	3.0	0.0043	3.3
AK024639	ATP13A3	3q29	0.1081	2.1	0.0454	2.4
AL050097	DKFZp586B0319	3q29	0.1058	3.6	0.0486	4.4
NM_001553	IGFBP7	4q12	0.0627	2.8	0.0363	3.3
NM_000582	SPP1	4q22.1	0.1311	1.8	0.0483	3.0
NM_003118	SPARC	5q33.1	0.0959	2.4	0.0363	3.1
NM_002932	DKFZp779B1535	5q33.1	0.1242	2.9	0.0465	3.5
NM_002341	LTB	6p21.33	0.2267	2.0	0.0491	2.6
NM_002800	PSMB9	6p21.32	0.1248	3.3	0.0363	4.5
NM_003472	DEK	6p22.3	0.0497	2.4	0.0106	2.9
NM_018950	HLA-F	6p22.1	0.1186	10.1	0.0363	15.5
NM_000416	IFNGR1	6q23.3	0.1540	4.1	0.0311	5.8
NM_002835	PTPN12	7q11.23	0.0497	2.3	0.0635	2.1
NM_014791	MELK	9p13.2	0.0497	1.9	0.0324	2.1
NM_001786	CDC2	10q21.2	0.0497	2.5	0.0180	2.9
NM_005127	CLECSF2	12p13.31	0.3683	2.9	0.0255	5.6
NM_002583	PAWR	12q21.2	0.0960	1.9	0.0354	2.2
NM_002345	LUM	12q21.33	0.0243	3.1	0.0235	3.2
NM_001845	COL4A1	13q34	0.0394	2.2	0.0235	2.6
AF035787	RF-IP4	14q32.33	0.1792	5.8	0.0363	9.5
NM_004048	B2M	15q21.1	0.0537	4.7	0.0454	5.2
NM_004165	RRAD	16q22.1	0.0497	3.6	0.0067	4.6
NM_001793	CDH3	16q22.1	0.1949	3.0	0.0214	4.5
NM_003150	STAT3	17q21.2	0.0959	1.9	0.0483	2.1
NM_002592	PCNA	20p12.3	0.1067	2.4	0.0363	2.9
NM_014258	SYCP2	20q13.33	0.0497	4.8	0.0265	6.1
NM_006806	BTG3	21q21.1	0.0125	2.3	0.0039	2.7
AK023589	FLJ13527	21q22.3	0.0884	1.9	0.0278	2.2
NM_005940	MMP11	22q11.23	3.70.0	13.7	0.0043	15.7
AF019225	APOL1	22q12.3	0.1235	4.2	0.0486	6.0
NM_001953	ECGF1	22q13.33	0.2007	7.4	0.0246	14.0
M97168	XIST	23q13.2	0.0394	6.7	0.0149	9.4
X56196	XIST	23q13.2	0.0394	4.5	0.0150	6.5
X56197	XIST	23q13.2	0.0532	2.6	0.0280	3.4
AK001758	RLR1	23q25	0.0537	2.2	0.0432	2.4
NM_004961	GABRE	23q28	0.3781	2.2	0.0265	4.1

For genes printed in bold, elevated expression was related to increased copy numbers as determined by DIGMAP and/or ACE-it analysis. FDR, False Discovery Rate; FC, Fold change.

TABLE 5.

Differentially Expressed Genes as Determined by LIMMA Statistical Analysis Between SCCs and AdCAs

GenBank AccNo	gene symbol	cytoband	FDR
Downregulated genes in SCCs compared to AdCAs
X73079	PIGR	1q32.1	0.0143
AK022207	MLPH	2q37.3	0.0219
NM_017862	FHFR	4p16	0.0212
NM_017540	GALNT10	5q33.2	0.0219
NM_001306	CLDN3	7q 11.23	0.0212
X74956	MUC5B	11 p15.5	0.0197
AB045292	M83	16p13.3	0.0212
NM_003225	TFF1	21q22.3	0.0212
Upregulated genes in SCCs compared to AdCAs
AF101051	SEMP1	3q28	0.0012
NM_003722	TP73L	3q28	0.0197
NM_006718	PLAGL1	6q24.2	0.0335
NM_002855	PVRL1	11q23.3	0.0213
NM_005127	CLECSF2	12p13.31	0.0197
NM_001941	DSC3	18q12.1	0.0143
NM_004961	GABRE	23q28	0.0197
AK026383	FLJ22730	8 or 14	0.0213

FDR, False Discovery Rate.

Validation of Differentially Expressed Genes by Real-time RT-PCR

To confirm the results obtained by LIMMA differential gene expression analysis we determined expression values of five selected genes (ITGAV, MAL, SEMP1, DEK, and SYCP2) by real-time RTPCR in all carcinomas analyzed by microarray. ITGAV and MAL were the most significantly upand down-regulated gene in carcinomas compared to normal, respectively. SEMP1 showed increased expression in SCCs compared with AdCAs as well as compared to normal epithelium, whereas upregulation of DEK and SYCP2 was previously shown to be associated with hr-HPV infection (Wise-Draper et al., 2005; Slebos et al., 2006). Overall, real-time RT-PCR and microarray values showed a good correlation as was previously shown for the platform used (Fig. 2; Pearson correlation r = 0.77, < 0.0005) (Muris et al., 2007).

Figure 2.

The overall correlation between microarray and real-time RT-PCR values for DEK, ITGAV, MAL, SEMP1 and SYCP2.

In addition, we verified differential expression of these genes in a separate validation sample set consisting of 13 normal epithelial samples of the ectocervix and 12 SCCs. Expression values of DEK, ITGAV, SEMP1, and SYCP2 showed more variation within the group of SCCs than in the normal samples, indicating various levels of (over)expression within the carcinomas. Nevertheless, increased expression of DEK and ITGAV as well as decreased expression of MAL in SCCs compared to normal ectocervical epithelium was statistically significant, thereby verifying our microarray results (Figs. 3A-3C; P < 0.05). Expression of SYCP2 and SEMP1 was also increased in SCCs, albeit at a lower level of significance (Figs. 3D-3E; P < 0.1).

Figure 3.

Real-time RT-PCR results for a subset of genes identified by differential expression analysis. Box plots of the log-transformed expression levels of (A). DEK,(B). ITGAV,(C). MAL, (D). SEMP1 and (E). SYCP2 are shown in the independent validation set consisting of SCCs and normal ectocervical epithelial samples. The upper and lower boundaries of the boxes represent the 75th and 25th percentiles, respectively. The black line within the box represents the median, the whiskers represent the minimum and maximum values that lie within 1.5 inter quartile range from the end of the box. Values outside this range are represented by triangles. ** P < 0.05; * P < 0.10.

Altered Gene Expression Related to the Chromosomal Location

To identify genes with altered expression associated with recurrent genetic alterations, we next aimed for the integration of gene expression and gene copy numbers. Genome-wide chromosomal profiles of the carcinomas transcriptionally profiled in this study were previously generated and described (Wilting et al., 2006).

Using gene expression data of carcinomas and normal epithelium, differential gene locus mapping analysis (DIGMAP) was performed (Yi et al., 2005), in which all genes present on the array were grouped based on their chromosomal location. This yielded loci exhibiting differential gene expression between either SCCs or AdCAs and normal epithelium. A number of these loci overlapped with frequently altered chromosomal alterations found by array CGH (Table 6). Genes located at 1q32.1 showed an overall increase in expression in both SCCs and AdCAs compared with normal epithelium. On the long arm of chromosome 3 two loci were identified which showed differential gene expression between SCCs and normal epithelium but not between AdCAs and normal epithelium. In agreement with these findings, gains of chromosome 1q were frequently found in both SCCs and AdCAs, whereas gains of 3q were more specific for SCCs (Wilting et al., 2006). The fact that we also identified loci with downregulated expression at chromosome arms 7q, 9q 16p, and 19p which do not overlap with common chromosomal losses, suggests that other regulatory mechanisms of gene expression, such as epigenetic alterations, also play a role in cervical cancer. An overview of all genes located within the regions identified by DIGMAP is given in Supplementary Table 1.

TABLE 6.

Loci with Differential Gene Expression Determined by DIGMAP Analysis in Relation to Chromosomal Alterations

DIGMAP analysis		Array CGH analysis
Comparison	Locus	% gains/losses in SCC	% gains/losses in AdCA
SCC vs normal	1q32.1	67	57
	3q 13.32-q22.3	100	14
	3q26.32-q27.3	100	29
	7q22.1	11	14
	9q34.11-q34.2	0	0
	16p13.3	11	0
AdCA vs normal	1q32.1-q32.2	67	57
	9q34.3	0	0
	19p13.3	11	0

Using gene expression and CGH array data, the Array CGH and Expression integration tool (ACEit) was applied (van Wieringen et al., 2006), in which expression and CGH array data were aligned for all carcinomas. Expression of genes was compared between carcinomas that showed either a gain or a loss of the respective gene locus and carcinomas without such chromosomal alteration. As a consequence, this analysis could only be performed on those genes for which sufficient numbers of carcinomas with and without chromosomal alteration were present (n = 654). Genes located at 1p36.23 (1 gene), 3q13.12-13.33 (8 genes), 3q21.1-23 (22 genes), 20p12.2 (1 gene), and 20q11.21-13.33 (15 genes) showed increased expression in carcinomas with gains at these particular regions, whereas expression of 16 genes located at 11q22.3-25 and 1 gene located at 18q23 was decreased in carcinomas with a loss of the chromosomal region in which they reside (FDR < 0.2) (Supplementary Table 2).

Results from array CGH, DIGMAP and ACE-it analysis are combined in Fig. 4. Particularly for chromosomes 1 and 3, integration of expression and chromosomal profiles resulted in the identification of chromosomal hotspots with altered gene expression within larger chromosomally altered regions. Whereas CGH analysis showed frequent gains (>50%) of the complete long arm of both chromosome 1 and 3, altered gene expression was only found at 1q32.1-32.2, 3q13.32-23, and 3q26.32-27.3.

Figure 4.

An overview of frequencies of gains and losses on chromosomes (A). 1, (B). 3, (C). 11, and (D). 20 as determined by array CGH is depicted in gray. Loci identified by DIGMAP and ACE-it are shown in blue and yellow respectively. [Color figure can be viewed in the online issue, which is available at www. interscience.wiley.com.]

Validation of Integration Results by Real-time RT-PCR

To verify the results obtained by integrating transcriptional and chromosomal profiles we selected 4 genes located at chromosome 3q21.1-23, namely DTX3L, PIK3R4, ATP2C1 and SLC25A36. Of these genes DTX3L and ATP2C1 were identified by DIGMAP, SLC25A36 by ACE-it and PIK3R4 was found in both analyses. In addition, significantly differential expression of these four genes was found between carcinomas and normal epithelium by LIMMA analysis. Real-time RTPCR confirmed increased expression of all four genes in the independent validation sample set consisting of 12 SCCs and 13 normal ectocervical samples (Fig. 5, P < 0.05). The fact that overexpression of genes identified by different methods of integration could be validated in an independent data set underlines the validity of our approach and shows that the use of multiple statistical analyses has additional value.

Figure 5.

Real-time RT-PCR results for genes identified by integrated analysis. Box plots of the log-transformed expression levels of (A). ATP2C1,(B). SLC25A36, (C). PIK3R4 and (D). DTX3L are shown in the independent validation set consisting of SCCs and normal ectocervical epithelial samples. The upper and lower boundaries of the boxes represent the 75th and 25th percentiles, respectively. The black line within the box represents the median, the whiskers represent the minimum and maximum values that lie within 1.5 inter quartile range from the end of the box. Values outside this range are represented by triangles. ** P < 0.05.

DISCUSSION

In the present study, we integrated transcriptional profiles of cervical carcinomas with their chromosomal profiles to identify genes with altered expression associated with copy number alterations. Others and we previously showed that frequent chromosomal alterations in cervical carcinomas are usually quite large in size. Integration allowed for fine mapping of these large chromosomally altered regions resulting in the identification of loci in which gene expression is altered as well.

To the best of our knowledge, this is the first study on cervical carcinomas in which high-resolution transcriptional and chromosomal profiles were generated from the same frozen tissue sections, which prevents possible confounding of the results by tumor heterogeneity. Nevertheless, a recent study by Narayan et al. (2007), using array CGH and expression array data of two independent sets of cervical carcinomas, revealed concordant regions in which an association between chromosomal alterations and elevated gene expression was found (i.e., 3q27-29 and 20q13.1). Fitzpatrick et al., (2006) identified potential chromosomal alterations important for the development of cervical cancer by alignment of expression microarray data. Comparable with our results altered expression of genes located at 1q, 3q, and 11q was identified in either high-grade premalignant lesions or invasive carcinomas.

We chose to compare expression patterns in carcinomas to those of unmatched normal controls instead of tumor-surrounding normal epithelium of the patient to ensure absence of genetically changed cells that otherwise might be present as a possible consequence of field cancerization. A significant proportion of carcinomas are surrounded by a field of cells that are clonally related to the carcinoma and thus show cancer-related genotypic alterations (Braakhuis et al., 2003; Braakhuis et al., 2004). Chu et al., (1999) already showed a local field effect of genomic instability associated with (pre)malignant cervical lesions. On the other hand, the use of unmatched controls may result in increased variability due to inter-person differences. To reduce this potential inter-patient variability, we hybridized pools of RNA obtained from normal cervical epithelium of various individuals. As we aimed to identify genes that next to hrHPV contribute to cervical cancer development, both HPV positive and negative normal control samples were included to minimize identification of HPVinduced transcriptional changes.

Even though we only included a relatively small amount of samples, differential gene expression analysis without integration to chromosomal alterations still identified 83 significantly differentially expressed genes between carcinomas and normal epithelium. Altered expression of a subset of these genes was validated in an independent sample set. A number of genes found to be differentially expressed in this study have been previously reported as modified in cervical cancer, including MAL, KRT4, EFNA1, MCM2, SEMP1, SPP1, PSMB9, CDH3, MMP11, and CLECSF2 (Chen et al., 2003; Hatta et al., 2004; Wu et al., 2004; Santin et al., 2005; Vazquez-Ortiz et al., 2005; Chao et al., 2006; Wong et al., 2006). In addition, differential expression of a number of genes detected by LIMMA, including ITGAV, STAT1, MCM2, TP73L,SEMP1, SPP1, B2M, CDH3, STAT3, and ECGF1, could be confirmed at the protein level by searching a high throughput immunohistochemistry staining database constructed on a panel of cancers (http//:www.hpr.se).

Integration analysis using DIGMAP and ACE-it, showed that gene expression and copy numbers were related at 1q32.1-32.2, 3q13.12-23, 3q26.32-27.3, 11q22.3-25, and 20q11.21-13.33. Especially for chromosomes 1 and 3 integration analysis resulted in the fine mapping of large chromosomally altered regions. Whereas the most frequent area of gain on both chromosomes was about 40 Mb in size, loci identified by integration ranged from 4-20 Mb. DIGMAP and ACE-it use very different approaches for integration of expression and genomics, which explains why partly different loci were identified. Despite these differences three genes located at 3q13.33 and 21 genes located at 3q21.1-22.2 were common in both analyses. Both approaches have their advantages and disadvantages. Because DIGMAP analysis uses gene sets rather than individual genes, this method is likely to detect smaller differences in expression that are persistent within the gene set than LIMMA does for individual genes. DIGMAP analysis detects coordinated changes in expression of genes located next to each other, which only provides indirect evidence of the involvement of a chromosomal alteration in the regulation of gene expression. This is underlined by the fact that DIGMAP also identified a number of loci that do not show frequent chromosomal alterations. On the other hand, altered gene expression does not have to be caused by the same mechanism in all tumor samples. Genes showing altered expression as a consequence of different mechanisms may potentially be the most interesting ones. In addition, DIGMAP enables analysis of all genes present on the array of which the genomic location is known. In contrast to DIGMAP, ACE-it allows determination of a more direct correlation between a chromosomal alteration and changes in expression of the genes located there, provided that no other differences exist between carcinomas with and without an alteration. Therefore, this approach is quite sensitive for detection of subtle but consistent changes in expression related to chromosomal alterations. A disadvantage of ACE-it, especially when working with small sample sizes, is the fact that a sufficient number of carcinomas with and without a certain chromosomal alteration are needed to perform the analysis.

To examine the potential value of both approaches we subsequently validated differential expression of four genes located at 3q21.1-21.3, which were identified either by DIGMAP (DTX3L and ATP2C1), ACE-it (SLC25A36) or both (PIK3R4). Using an independent validation sample set, overexpression of all four genes could be confirmed. This underlines the suggestion that both analyses can yield valuable information and are complementary to each other. Of these genes DTX3L encodes a member of the deltex family of proteins, which function as E3 ligases and modify Notch signaling (Takeyama et al., 2003). The Notch signaling pathway has been implicated in cervical cancer before, although results are contradictory as to whether its role is oncogenic or tumor suppressive (Zagouras et al., 1995; Talora et al., 2002; Radtke and Raj, 2003). PIK3R4 encodes the regulatory subunit of the class III phosphatidylinositol-kinase (PI3K) signaling pathway, which was shown to be involved in amino acid-induced mTOR activation and may as such be essential for cell growth control (Nobukuni et al., 2005). Mutations in PIK3R4 have recently been described in breast cancer (Wood et al., 2007). Whereas relatively little is known about class III PI3K signaling in cervical cancer, deregulated class I PI3K signaling, of which the catalytic subunit PIK3CA is also located at 3q, has been described in cervical cancer (Bader et al., 2005; Bertelsen et al., 2006).ATP2C1, encoding a P-type cation transport ATPase, plays an essential role in the epidermis by keeping basal keratinocytes in their undifferentiated state (Yoshida et al., 2006). Specific cellular functions of the protein encoded by SLC25A36 are unknown at the moment, but Gene Ontology indicates that the protein is located in the mitochondrial membrane and possesses transporter activity.

A gain of chromosome 3q is the most frequently found chromosomal alteration in cervical carcinomas and has been described in a number of other solid tumors as well (Sugita et al., 2000). The fact that the frequency of this event gradually increases from low-grade lesions to high-grade lesions, reaching a maximum in invasive carcinomas indicates that a 3q gain is closely tied to malignant progression of cervical lesions (Heselmeyer et al., 1996; Kirchhoff et al., 1999, 2001; Umayahara et al., 2002; Alazawi et al., 2004). Therefore, markers based on this chromosomal alteration are likely to be specific for progressive premalignant lesions and carcinomas. Two promising candidate oncogenes are located within this region, namely hTR and PIK3CA (Ma et al., 2000; Sugita et al., 2000). Their potential value as marker for cervical cancer is supported by recent studies showing increased gene copy numbers of hTR and increased PIK3CA protein expression in scrapings of women with underlying high-grade premalignant disease compared with scrapings of women with mild or no underlying disease (Heselmeyer-Haddad et al., 2003, 2005; Goto et al., 2006). In this study expression values of hTR were not available. PIK3CA was included in the 3q26.32-27.3 region identified by DIGMAP, but did not show significantly differential expression independent of chromosomal location (LIMMA analysis). Other interesting genes identified by DIGMAP, but not LIMMA, located within this region include LAMP3, overexpression of which was associated with metastasis in cervical cancer (Kanao et al., 2005), ST6GAL1, enhanced expression of which was shown in cervical squamous cell carcinoma (Wang et al., 2003), and RCF4, for which an association between gene dosage and gene expression in cervical cancer has been previously described (Narayan et al., 2007). In summary, besides PIK3CA and hTR, other genes located within the 3q gain are overexpressed in cervical carcinomas and may be good candidate marker genes as well.

To enable reliable statistical testing we expanded our normal control group, of which we had limited amounts of high-quality RNA available, with normal squamous epithelium from another source. We are aware of the fact that the heterogeneity within our normal control group may have resulted in the identification of lower numbers of significantly differentially expressed genes. Nonetheless, this approach apparently did not result in false positive outcomes of identified differentially expressed genes since we were able to validate their differential expression in a separate set of samples, in which only normal cervical tissue specimens were used as controls. Another limitation of this study is the relatively small sample size. We would like to emphasize that identification of genes showing significant upregulation in tumors associated with increased copy number of the gene locus in such a small sample group indicates that altered structure and activity of these genes are common in cervical carcinomas. This is supported by confirmation of increased expression of a subset of these genes in an independent validation sample set. We do realize that a larger sample size is needed to identify genes that are less strongly affected.

In conclusion, this study shows that integrated genome-wide transcriptional and chromosomal profiling of cervical carcinomas appears a promising strategy for fine mapping of large chromosomally altered regions to identify loci with altered gene expression that might be biologically meaningful. Further studies of the genes identified in this study are needed to assess their biological relevance in (cervical) carcinogenesis as well as their potential value as biomarker for improved detection of highgrade cervical lesions and carcinomas.