Using LongSAGE to Detect Biomarkers of Cervical Cancer Potentially Amenable to Optical Contrast Agent Labelling.

Sixteen longSAGE libraries from four different clinical stages of cervical intraepithelial neoplasia have enabled us to identify novel cell-surface biomarkers indicative of CIN stage. By comparing gene expression profiles of cervical tissue at early and advanced stages of CIN, several genes are identified to be novel genetic markers. We present fifty-six cell-surface gene products differentially expressed during progression of CIN. These cell surface proteins are being examined to establish their capacity for optical contrast agent binding. Contrast agent visualization will allow real-time assessment of the physiological state of the disease process bringing vast benefit to cancer care. The data discussed in this publication have been submitted to NCBIs Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) and are accessible through GEO Series accession number GSE6252.

Introduction

Clinical diagnosis of most cancers and their precursors is predominantly based on phenotypic markers such as appearance of cell nuclei. Classification and staging of disease is determined by evaluation of gross structural features, such as extent of local tumour invasion and presence of disease in other organs. It is now established that cancer arises as a result of successive genetic changes altering cellular processes including growth, angiogenesis, senescence, and apoptosis (Hanahan and Weinburg, 2000). Additionally, many cancers appear to have active inflammation and wound healing mechanisms (Chang et al. 2004). Proteins taking part in these cellular mechanisms are often strong candidates for biomarkers and molecular targets.

Cervical cancer is usually the result of a human papillomavirus (HPV) infection which initiates neoplastic progression mainly through viral oncoproteins E6 and E7 within the cervical transformation zone at the squamous/columnar junction. The role of HPV to the pathogenesis of cervical cancer has been addressed in recent reviews (zur Hausen, 2002; Woodman et al. 2007). Many HPV types produce only productive lesions following infection and are not associated with human cancers. In such lesions, the expression of viral gene products is carefully regulated, with viral proteins being produced at defined times and at regulated levels as the infected cell migrates towards the epithelial surface. The events that lead to viral synthesis in the upper epithelial layers appear common to both the low- and high-risk HPV types. Virus-induced cancers most often arise at sites where productive infection cannot be suitably supported. Productive infection can be divided into distinct phases, with different viral proteins playing specific roles (Doorbar, 2006). Upon infection, normal cells gradually advance through stages of cervical intraepithelial neoplasia (CIN). Mild dysplasia (CINI) presents as only a subset of the low third of the epithelium appearing dysplastic, moderate dysplasia (CINII) occurs where the dysplastic cells involve about one-half of the thickness of the epithelium of the cervix, and severe dysplasia (CINIII), or carcinoma-in-situ, is described as the condition where the entire thickness of the epithelium is disordered but the abnormal cells have not yet spread below the surface. If carcinoma-in-situ is not treated, it will often grow into an invasive cervical cancer. High grade dysplasia is considered the most advanced dysplasia with atypical changes in many of the cells and a very abnormal growth pattern of the glands; some of the glands are branching or budding. More than 50% of the cells have large, spotted nuclei and are frequently dividing while the cellular cytoplasm is reduced and looks abnormal. Cancer of the cervix was one of the most common causes of cancer death for American women, but between 1955 and 1992 the number of cervical cancer deaths in the United States dropped by 74% due to the introduction of the Pap test (Papanicolaou and Traut, 1943). Death rates from cervical cancer continue to decline by nearly 4% per year. Even so, the American Cancer Society reports that in 2006, about 3,700 of the 9,710 women diagnosed with cervical cancer in the United States have died from this disease.

HPV infection causes changes in expression levels of a wide variety of genes (Yim and Park, 2006). These differences in gene expression between pre-invasive neoplastic and non-neoplastic tissue give clues to the molecular basis of cancer. Early detection of cervical cancer based on molecular characterization would be clinically advantageous; risk of neoplastic lesion progression could be predicted and response to therapy could be monitored in real time at a molecular level. To monitor molecular characterisation of cancer it follows that the ability to optically image in real-time the molecular features of cancer in vivo is critical (González et al. 1999; Rajadhyaksha et al. 1999; White et al. 1999; Huzaira et al. 2001; Langley et al. 2001; Selkin et al. 2001; Collier et al. 2002) and requires safe, molecular-specific contrast agents whose images can be monitored rapidly and non-invasively during their uptake and distribution. The analysis presented here evaluates serial analysis of gene expression (SAGE) libraries to identify novel, cell-surface gene products. Upon mapping of highly differentially expressed SAGE tags to their corresponding genes, the gene products are candidates for antibody testing and optical contrast agent development.

Contrast Agents and Optical Imaging

Short of prevention, improved early stage cancer diagnosis would provide the greatest benefit for cancer patients. Because proteins may regulate gene expression, ligand-binding properties, molecular structure and dynamics on a temporal basis, protein biomarkers have a significant impact in cancer detection and therapy as therapies are becoming targeted to specific signal transduction and metabolic pathways. For example, breast cancers respond to HERCEPTIN (trastuzumab) if the tumor over-expresses Her-2/neu (Baselga et al. 2004; Ross et al. 2004). In the same way, GLEEVEC (imatinib) is most effective against cancers carrying the bcr-Abl translocation (Druker, 2004) and targeted molecular cancer therapy is already used successfully for the eradication of acute leukaemia (Frater et al. 2003; Yee and Keating, 2003). These examples imply that it will be important to produce biomarkers for all stages of cancer. Reliable diagnostics such as DNA screening and immunocytochemical analysis of known cervical neoplasia biomarkers p16INK4A and minichromosome maintenance (MCM) proteins are not implemented in vivo. Real-time biomarkers of the physiological state of the disease process or markers representative of treatment efficacy will bring immeasurable benefit to cancer care in terms of individualized agent selection and dosing. Furthermore, series of agents could be tested to determine empirically the localization of cancer and/or the most effective therapy.

Routine clinical cancer detection employs non-specific contrast agents such as acetic acid which enhance the nuclear backscattering but are limited by small signal magnitude. The field of molecular imaging is rapidly developing imaging agents with high affinity and specificity for targeted biomarkers. These new agents allow for the possibility of disease detection earlier than is currently feasible (Weissleder, 2001; Jaffer and Weissleder, 2005). For example, cancer metastases missed by conventional anatomically based imaging methods may be detected in patients by molecular imaging (Harisinghani et al. 2003). Optical imaging of tissue can be carried out non-invasively in real time, giving high spatial resolution (<1 μm lateral resolution). A number of optical techniques have been established including confocal microscopy (White et al. 1999; Collier et al. 2005), multispectral fluorescence imaging (Andersson-Engels et al. 1997; Ferris et al. 2001), reflectance spectroscopy with polarised and unpolarised light (Sokolov et al. 1999, 2002; Utzinger et al. 2001), multispectral reflectance imaging with polarised and unpolarised light (Ferris et al. 2001; Gurjar et al. 2001), and fluorescence spectroscopy (Gillenwater et al. 1998; Wagnières et al. 1998; Ramanujam, 2000; Sokolov et al. 2002). Together with emerging molecular tools (e.g. DNA screening, tissue proteomic and serum markers), biomarker imaging may soon be used for real-time screening, diagnosis, and detection of disease recurrence and progression (Rudin and Weissleder, 2003).

Contrast agents consist of a biomarker specific probe molecule, such as an antibody, conjugated to an optically suitable label. By topically applying molecular specific contrast agents to tissues, the scope of molecular changes that can be probed using optical imaging is significantly enhanced. Presently, contrast agents based on metal nanoparticles, organic fluorescent dyes, and quantum dots coupled to monoclonal antibodies against cancer specific biomarkers are being developed (Sokolov et al. 2003; Rahman et al. 2005).

SAGE Libraries and Tag Mapping

The SAGE technique is capable of producing a molecular representation of cervical tissue based on expressed genes. SAGE is not dependent on pre-existing databases of expressed genes and so provides an independent view of gene expression profiles within the mRNA populations (Velculescu et al. 1997). SAGE library construction is well documented in the literature (Velculescu et al. 1995 and 1997; Madden et al. 2000; Saha et al. 2002; Pleasance et al. 2003; Sander et al. 2005). Several recent gene expression profiles of in vitro HPV-infected cultured keratinocytes and from cervical carcinoma clinical samples have proposed changes in gene expression induced by HPV and in early cervical carcinomas (Thomas et al. 2001; Ruutu et al. 2002; Duffy et al. 2003; Pérez-Plasencia et al. 2005). Some studies have compared normal versus tumor-induced gene expression in cervical samples with the aim of identifying potential tumor markers of clinical value (Shim et al. 1998; Chen et al. 2003).

To identify genes expressed at dissimilar levels in preinvasive neoplastic and non-neoplastic untyped cervical tissue, we analysed sixteen long-SAGE libraries; 4 from normal cervical tissue samples, 3 of a mild dysplasia (CINI), 3 of moderate dysplasia (CINII), and 6 of severe dysplasia (CINIII), or carcinoma-in-situ. The CIN tissues are positive for MUC16. Raw numbers of longSAGE tags generated and library names are given in Tables 1 and 2. DiscoverySpace (Robertson et al. 2007), an in-house graphical software application backed by a relational database system designed to support SAGE gene expression analysis, was used to query data from over 25 publicly available data sources, as well as internal experimental results. Using DiscoverySpace, selected SAGE tag sequences were mapped to counterpart RefSeq (Pruitt et al. 2000, 2005) genes and confirmed using SAGE tag co-ordinates to establish gene identity through Ensembl (Hubbard et al. 2007; homo_sapiens_ core_41_36c). Genes were manually curated (EntrezGene) to ascertain gene identity and gene product localisation. These cervical longSAGE libraries were created from the epithelium of cervical biopsy samples collected just prior to LEEP (Loop Electrosurgical Excision Procedure). Tissue samples were placed into RNAIater and frozen at −80 °C within 10 minutes of being excised from the patient. These longSAGE libraries (Shadeo et al. 2007) have been submitted to the NCBI Gene Expression Omnibus (GEO) repository.

Table 1

Highly expressed genes with membrane-bound gene products up-regulated in cervical dysplasia stages CINI and CINIII. Up-regulated gene expression, from normal, ≥2-fold.

Tag sequence	Number of tags in disease libraries*	Number of tags in normal libraries	Gene	Fold increase from normal	RefSeq accession number
CINIII
GACCCAAGATAAAAGAA	704	16	PIGR	22.4	NM_002644
ATCCCCCTGGGCATCGG	52	2	SLC39A3	13.2	NM_144564
ACTCAGACCAGGTCCCA	94	4	STRA6	11.4	NM_022369
ACACAGTATTCGCTCTT	44	2	ITR	11.2	NM_180989
TTACTTCCCCACCCCTA	84	4	FADS2	10.7	NM_004265
GGAACTGTGAAGAGGCA	230	12	TSPAN1	9.7	NM_005727
GCACCTGTCGCCCAGTG	36	2	ANPEP	9.2	NM_001150
CCTGATCTGCGGTGTCC	308	18	MUC16	8.7	NM_024690
CGTTTTCTGATAACTCA	68	4	PTP4A2	8.6	XM_001132367
CAAATAAATTATGCGAT	98	6	TMPRSS2	8.3	NM_005656
TGCTCCTACCCTGCTCT	2022	190	FCGBP	5.4	XM_001131379
GCAGTGCCACTCAAGAA	726	68	SRD5A2L	5.4	NM_024592
CCTGGGAAGTGTTGTGG	730	98	MUC1	3.8	NM_001044391
AATATTTATATTGTATG	1880	400	CEACAM5	2.4	NM_004363
GTTCACATTAGAATAAA	3424	762	CD74	2.3	NM_001025158
GGGCATCTCTTGTGTAC	1602	350	HLA-DRA	2.3	NM_019111
TGCTGCCTGTTGTTATG	826	192	BST2	2.2	NM_004335
CTGACCTGTGTTTCCTC	596	144	HLA-B	2.1	NM_005514
GCAGGGCCTCATCTCAC	1950	484	FXYD3	2.0	NM_005971
CINI
GCACCTGTCGCCCAGTG	42	2	ANPEP	36.5	NM_001150
ATGTTAATAAAATAGGC	66	4	GPC4	28.7	NM_001448
ATAAATGATTAGACTAC	32	2	CLIC6	27.8	NM_053277
GGCCAAGAACTTTCACT	28	2	C14orf101	24.3	NM_017799
GACCCAAGATAAAAGAA	262	16	PIGR	23.0	NM_002644
TAGGTCAGGACCTTGGC	26	2	PTP4A3	22.6	NM_032611
GTATATAACTCTTAAAG	24	2	LOC644410**	20.8	XM_001133198
CCAGCTGCCTGGAGGAG	22	2	MGC45438	19.1	NM_152459
ATAAAAATTAGGGGGAT	22	2	SLC30A1	19.1	NM_021194
TGAGCTACCCCAGAGTC	20	2	FER1L4	17.4	NR_001442
TACATCAGTAAAGAGTT	374	42	GAS1	12.5	NM_002048
CATCACGGATCAATAGA	330	64	IL1R1	7.2	NM_000877
CCTGGGAAGTGTTGTGG	334	98	MUC1	4.1	NM_001044391
TGCAGATTGCAGTTCTG	366	132	PROM1	3.9	NM_006017
CACTTCAAGGGCAGCCT	238	102	LY6E	3.3	NM_002346
GTTCACATTAGAATAAA	1660	762	CD74	3.1	NM_001025158
GGGCATCTCTTGTGTAC	734	350	HLA-DRA	3.0	NM_019111
TGCTCCTACCCTGCTCT	402	190	FCGBP	3.0	XM_001131379
CTGACCTGTGTTTCCTC	256	144	HLA-B	2.5	NM_005514

GEO Series Accession Numbers GSE6252. 1,101,702 total longSAGE tags in four normal libraries, GEO Alias N3, N1, N2, N4; 2,165,777 total longSAGE tags in six CINIII libraries, GEO Alias C1, C3, C2, C4, C5, C6; 785,642 total longSAGE tags in three CINI libraries, GEO Alias M1, M2, M3.

Mapping unable to be confirmed through ENSEMBL or BLAT.

Table 2

Highly expressed genes with membrane-bound gene products up-regulated in normal cervical tissue. Up-regulated gene expression, from CINI and/or CINIII, ≥2-fold.

Tag sequence	Number of tags in disease libraries*	Number of tags in normal libraries	Gene	Fold increase from disease state	RefSeq accession number
CINIII
AGACTTGGCATACACAC	2	30	ACTR3	29.5	NM_005721
TAACTGGCCTTACGATG	6	54	DSG1	17.7	NM_001942
AGAGGGATGAGGCAACC	2	18	GJB2	17.7	NM_004004
AGTCCTGGAGGAGGAGA	2	18	PTPNS1	17.7	NM_001040022
GCTGACAGTCCCAAGTC	2	16	NIP	15.7	NM_144565
ACAGTCACCACGAGGAG	2	16	ODZ2	15.7	XM_047995
AGAGTTCTGTCACGGTC	8	62	UPK1A	15.2	NM_007000
CAGTATTTTAAATAATT	2	12	FLJ32028	11.8	NM_152680
ATAATCACCCTTCATCA	2	12	CACNA2D3	11.8	NM_018398
TGGTGGCTGCTTTTCCC	2	10	TMEM54	9.8	NM_033504
CCAGCGCCAACCAGTCA	272	482	LYPD3	3.5	NM_014400
GTTTCCAAAAAATGGTA	350	438	GJB2	2.5	NM_004004
GTGGAAGACGATAACCC	5664	7334	MAL	2.5	NM_002371
GCCCAGCATTCTCCACC	856	832	PSCA	1.9	NM_005672
CTTGATTCCCACGCTAC	236	230	QSCN6	1.9	NM_002826
GTTATTGAGGGCAAGAA	196	194	C3orf28	1.9	NM_014367
CINI
CAATCTTGCAGTGAAGA	6	226	TMPRSS11B	26.9	NM_182502
TCAGAATGATTCTGGTG	22	346		11.2
AGAGTTCTGTCACGGTC	2	62	UPK1A	22.1	NM_007000
GCATAACAACTCCCCAA	4	60	EMP1	10.7	NM_001423
TAATTTGCATTACTCTG	366	2036		4.0
GATCCAGATGCCTGAGG	4	74	LYNX1	13.2	NM_023946
CCAGCGCCAACCAGTCA	28	482	LYPD3	12.3	NM_014400
GATGAATCCGGGGTATG	4	68	TMEM45B	12.1	NM_138788
TATATTTTTAAATGTTC	2	30	TGFA	10.7	NM_003236
TAACCCCTCTCACCTGC	2	28	FGFR2	10.0	NM_022970
GGGACTAAAACTCACCC	2	28	MAL	10.0	NM_002371
GGAGGGAGCTGAGGGGT	2	28	EPHA1	10.0	NM_005232
GCCCAGCATTCTCCACC	82	832	PSCA	7.2	NM_005672
GTTATTGAGGGCAAGAA	30	194	C3orf28	4.6	NM_014367
GTTTCCAAAAAATGGTA	70	438	GJB2	4.5	NM_004004
GCTTCCTCGGTACCCTT	252	1590	RHCG	4.5	NM_016321
CCACAGGAGAATTCGGG	952	5624	PERP	4.2	NM_022121
GATATGTAAAAACTGTC	38	194	CLCA4	3.6	NM_012128
GCCTACCCGAGGAGAAG	132	616	TACSTD2	3.3	NM_002353

GEO Series Accession Number GSE6252. 1,101,702 total longSAGE tags in four normal libraries, GEO Alias N3, N1, N2, N4; 2,165,777 total longSAGE tags in six CINIII libraries, GEO Alias C1, C3, C2, C4, C5, C6; 785,642 total longSAGE tags in three CINI libraries, GEO Alias M1, M2, M3.

Any protein differentially expressed in cancer tissue, compared to normal tissue, or any protein known to be involved in cancer development, has potential as a candidate cancer biomarker. Genes presenting properties which identify them as likely targets for cancer diagnosis or prognosis must be separated from thousands of other genes which also may also possess clinical potential. Hundreds of potential candidates must be set aside in favour of gene products which offer the most promising characteristics. We focus on genes encoding membrane associated proteins because membrane-bound proteins are most likely to be accessible to topical application of contrast agents and have a rapid time frame for contrast agent visualization. Genes expressing the greatest number of tags combined with high levels of differential expression between dysplastic and normal tissue are the most likely to be observed by contrast agents in vivo.

For optical imaging of tumors by topical application in vivo of contrast agents to be of practical use, a large number of contrast agent receptors are required. One of the standard methods to detect candidate biomarkers is to identify genes with amplified expression in cancer and/or normal tissues. We compared transcription profiles and retained the most highly expressed membrane-bound gene products whose differential expression level is greater than two-fold. Many cell-surface proteins can potentially be developed as targets for optical contrast agents. Using longSAGE, we have also identified the most highly differentially expressed transcripts between disease and normal tissue. Table 1 specifies those genes (with cell-surface gene products) up-regulated in the CINI and CINIII stages of dysplasia and Table 2 lists genes up-regulated in normal tissue. Short descriptions of these protein biomarkers are given in the appendix and annotations of protein structural information, if available, are included. Given that CINII is difficult to determine clinically, it was not included in these comparisons. Contrast agent visualization of the epidermal growth factor receptor (EGFR) using an anti-EGFR monoclonal antibody has already been successful (Rahman et al. 2005). More of these markers should prove amenable to contrast agent development and topical formulations consisting of a range of contrast agents could help adjust to individual patient differences in gene expression.

Cervical Intraepithelial Neoplasia Stage Biomarkers

It is possible to evaluate a marker for presence or absence, but to correlate a marker or array of markers to changes in cellular localization relative to other markers is probably the most interesting and beneficial in terms of dysplasia progression, environment, therapy selection, and follow-up. The known function of these genes grants some insight into the biology of cervical neoplasia. For instance, several of these cell surface markers are involved in transport and/or signaling. MUCX and CD74, upregulated in CINI and CINIII, have signaling gene products known to be associated with carcinomas. CD74 is also known to be a high affinity binding protein for macrophage migration-inhibitory factor (MIF) which is implicated in tumor cell growth and angiogenesis. TSPAN1, upregulated in CINIII almost 10-fold, also plays a role in cell motility and growth. See Appendix for gene-specific references.

Our analysis of cervical cancer longSAGE expression profiles direct attention to some genes with relatively equal distribution in CINI and CINIII, such as PIGR. Another marker, ANPEP, is present at significantly different expression levels in CINI and CINIII. This knowledge expands the possibilities for rapid visualization between normal and stages of dysplasia in vivo. As discussed earlier, these cell surface targets were found by identifying differentially expressed genes. More often than not, a highly expressed tag is not localised to the cell surface and, for these purposes, does not warrant further attention. However, a highly differentially expressed gene whose gene product is not membrane-bound is sometimes found to be part of a mechanism which affects the cell surface and thereby the gene product becomes of potential use. TFF3 (trefoil factor 3), for instance, is up-regulated 13-fold in CINIII and 27-fold in CINI. Members of the trefoil family are characterised by having at least one copy of the trefoil motif, a 40-amino acid domain that contains three conserved disulphides. They are stable, secretory proteins whose functions are not defined but may protect the mucosa from insults, stabilize the mucus layer and affect healing of the epithelium. VANGL1 (Van Gogh-like protein 1) is an integral membrane protein which is serine/threonine phosphorylated and translocated to cytoplasmic vesicles in response to TFF3 stimulation (Kalabis et al. 2006). VANGL1 protein acts as a downstream effector of TFF3 signalling and regulates wound healing of intestinal epithelium. TFF3 is commonly expressed in hepatocellular carcinoma and its expression correlates with tumor grade (Khoury et al. 2005). TFF3 overexpression may be a critical process in mouse and human hepatocellular carcinogenesis (Okada et al. 2005). The group of trefoil factor peptides (TFF1-3) are part of the protective mechanism operating in the intestinal mucosa and play a fundamental role in epithelial protection, repair, and restitution (Vieten et al. 2005). TFF3 and the essential tumor angiogenesis regulator VEGF exert potent pro-invasive activity through STAT3 signalling in human colorectal cancer cells (Rivat et al. 2005). That VANGL1 returned to cell membranes within 45 minutes of TFF3 stimulation (Kalabis et al. 2006) could explain the low VANGL1 tag counts, 1–3 tags per library, observed in the longSAGE libraries.

Conclusions

Molecular specific contrast agents may provide the ability to directly image the cancer process; but biomarker discovery can be a lengthy process as candidate markers suitable for the task-at-hand must be identified from among thousands of proteins. SAGE-identified biomarkers hold promise for recognition of the stages of neoplasia by proteomic patterns. Optical contrast agents bound to these membrane-bound protein biomarkers will serve as a complement to histopathology, thus allowing more effective determination of tumor borders and non-invasive observation of response to treatment at a molecular level. We present fifty-six cell-surface gene products differentially expressed during progression of cervical intraepithelial neoplasia. Differential gene expression of these biomarkers will allow individualized selection of therapeutic combinations that best target the entire disease-protein system and advance understanding of carcinogenesis.