HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Full Text (PDF) All Versions of this Article: 16/2/246 01-0618fjev1    most recent Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by VAN RUISSEN, F. Articles by SCHALKWIJK, J. Search for Related Content PubMed PubMed Citation Articles by VAN RUISSEN, F. Articles by SCHALKWIJK, J.

Differential gene expression in premalignant human epidermis revealed by cluster analysis of serial analysis of gene expression (SAGE) libraries ¹

Department of Dermatology, University Medical Center, 6500 HB Nijmegen, The Netherlands

³Correspondence: Department of Dermatology, University Medical Center Nijmegen, PO Box 9101, 6500 HB Nijmegen, The Netherlands. E-mail: F.vanRuissen{at}derma.azn.nl.

SPECIFIC AIMS

We studied gene expression patterns of normal human epidermis, premalignant skin lesions (actinic keratosis), and cultured keratinocytes using serial analysis of gene expression (SAGE). To examine the large amount of data, we performed a data reduction, followed by two-way clustering analysis on transcripts and tissues, which separated the in vivo samples from the in vitro samples and visualized clusters of many presumably coregulated genes. Two gene clusters that were up-regulated in the premalignant skin lesion were studied in detail, and we could confirm these findings by Northern blot analysis and immunohistochemistry. On the basis of their presumed functions, these genes may be relevant in the onset and progression of carcinogenesis.

PRINCIPAL FINDINGS

1. Construction of SAGE libraries
Pure epidermal samples without admixture of components from the underlying tissue were obtained by dispase treatment of skin biopsies. Histological examination revealed a clear separation of epidermis and dermis. Using this separation in combination with MicroSAGE, we constructed SAGE libraries from normal epidermis of a healthy individual, epidermis from an actinic keratosis lesion of a kidney transplant patient, and normal epidermis from the same patient. Two other libraries were obtained from cultured human keratinocytes.

2. General results and statistics
Using SAGE, we generated five libraries containing >60,000 tags in total. A large proportion of these tags was found only once and therefore not useful for quantitative purposes. To obtain a true transcriptome as performed for yeast, far more tags need to be sequenced. Combining the two libraries of normal epidermis, we found that >12,500 putative unique genes were expressed. With the recent estimates of 30,000–40,000 genes in the human genome, this shows that keratinocytes have the potential to express at least 35% of the genes that reside within the genome.

3. Cluster analysis
To use the full potential of the data, we used cluster analysis to identify genes that behave similarly across the different libraries. Frequency distributions of the individual libraries were organized in an expression level matrix, where each row corresponds to a single transcript and each column represents the individual library. We compressed the data set to focus on a small subset of transcripts; most transcripts are expressed at low levels and may mask the effect of smaller subsets of relevant genes. After transformation of the final subset of most informative transcripts, cluster analysis was performed on both libraries and transcripts. The results are visualized in Fig. 1 . Clustering of the individual SAGE libraries separates the two in vitro libraries from the in vivo ones (data not shown). Cluster analysis shows several compact clusters of coregulated genes within the different libraries. Visual representation of the cluster analysis also reveals two clearly discernible clusters (A and B) that differentiate gene expression in actinic keratosis lesions from other samples. The genes contained in these two clusters can be of biological or diagnostic/prognostic significance for this particular type of tumor.

View larger version (13K): [in this window] [in a new window]   Figure 1. Visualization of the expression level matrix consisting of the 300 most informative transcripts. From top to bottom, the five different libraries are represented as follows: normal human epidermis (HEP), epidermis from actinic keratosis tumor material (AKP), normal epidermis from the actinic keratosis patient (AKN), resting cultured keratinocytes and cultured keratinocytes treated with tumor necrosis factor (TNF-). Clusters of coregulated genes clearly distinguishable are numbered from 1 to 5. Two clusters that differentiate gene expression in actinic keratosis lesions from the other samples are marked A and B. Gene expression levels are indicated using a color index; ranging from high (red) to low (dark blue).

4. Verification
To investigate whether the observed differences were not artificial and establish whether the findings derived from the pooled tumors obtained from one patient could be extrapolated more generally to this type of tumor. Expression of three genes identified in Fig. 1 (clusters A, B) showing an increased expression in the tumor were examined in more detail. Probes and antibodies for psoriasis-associated fatty acid binding protein (PA-FABP), also known as E-FABP, migration inhibitory factor-related protein 8 (MRP-8), and psoriasin were used to verify the data on Northern analysis and histological sections. Using Northern blot analysis (Fig. 2 A), it is shown that the difference in expression levels of these genes (as found in SAGE analysis of one patient) can also be demonstrated in tumors from four patients by using an independent technique. We extended these control experiments at the protein level by immunohistochemical examination of MRP-8 and PA-FABP (Fig. 2B ), which also confirmed the observed differences.

View larger version (115K): [in this window] [in a new window]   Figure 2. A) Confirmation of SAGE data by Northern blot analysis. Expression of three genes identified in clusters A and B from Fig. 1 was examined: PA-FABP, MRP-8, and psoriasin. Biopsy material from 4 different patients (N=normal; T=tumor; lanes 6–13), including the patient from whom the SAGE library was constructed (lanes 6 and 7), was compared against normal (n) epidermis (lanes 3–5). Lanes 1 and 2: TNF--stimulated cultured keratinocytes (+) and resting keratinocytes (-). To verify that changes in expression levels were not due to unequal loading of the samples, we included the 18S band as a control. In all tumors examined, the up-regulation of these genes could be confirmed. B) Confirmation of SAGE data by immunohistochemistry. Protein expression levels of MRP-8 (A, B) and E-FABP (C, D) were examined in normal human skin (A, C) and actinic keratosis lesions (B, D). MRP-8 is not expressed in normal epidermis but is found in the suprabasal compartment of the tumor. Staining of the basal cells is due to melanin. E-FABP is normally restricted to the stratum granulosum, but is expressed in the tumor throughout the suprabasal compartment.

5. Pathway and profile analysis by automated literature search
To look for known expression patterns and putative functions of these molecules, the set of reliable genes (clear identification in GenBank) found to be overexpressed in the tumor was subjected to conventional PubMed searches. In addition, we used the PubGene tool that was recently developed for automated literature-to-gene searches. When this set of genes, with their corresponding gene symbol according to the HUGO Nomenclature Committee, was subjected to the PubGene Subset network tool, a literature network of eight of the up-regulated genes based on concurrence was obtained. Within this cluster, four of the identified genes are located in a region on chromosome 1q21 that contains a known differentiation cluster.

CONCLUSIONS AND SIGNIFICANCE

Cellular changes within the transition of normal skin to a (pre)malignant state may involve only a small subset of genes. Relevant signals associated with this process may be overwhelmed by the noise of irrelevant information. To examine the data generated in the SAGE libraries, we first reduced the number of genes to be analyzed by compressing the data set and selecting the 300 most informative tags for further study. These tags were used to cluster the samples and the genes. Due to the complexity and amount of effort associated with the generation of SAGE libraries, only small numbers of samples (libraries) can be analyzed. Therefore, whenever genes are detected that are differentially expressed, independent methods for confirmation on larger sample numbers are mandatory in order to ascertain whether the data can be extrapolated (Fig. 3 ).

View larger version (61K): [in this window] [in a new window]   Figure 3. Schematic diagram of SAGE analysis on skin biopsies using data reduction and cluster analysis.

Figure 1 shows a number of tag clusters that can be easily recognized by visual inspection. The two clusters (A and B) we examined more closely were selected because they contain transcripts that were strongly up-regulated in the tumor vs. normal human epidermis and epidermis from the same patient. These genes appear to be tumor-specific in that they are not found at appreciable levels in the normal tissue. As such, they can be either causally involved in the process of carcinogenesis or a part of an alternative program of differentiation that is switched on by acquired alterations in genes that act as molecular switches. Identification of these genes would allow us to study initiating events in carcinogenesis. The two clusters of tags up-regulated in the tumor contain transcripts that cannot be unequivocally assigned to a Unigene cluster or a defined mRNA in the databases (GenBank). However, several tags that could be identified with certainty can be associated with the process of cancer development on the basis of their presumed function or expression pattern reported in the literature, such as MRP8, MRP14, PA-FABP, psoriasin, 14–3-3-sigma, PPP1R3, polymerase theta, and Bcl-X. At least five of these genes are known to be strongly up-regulated in psoriasis, a benign hyperproliferative skin condition characterized by grossly abnormal differentiation. However, in psoriasis the expression of this phenotype is reversible as the disease is presumably triggered by environmental stimuli, whereas in actinic keratosis this phenotype is permanently switched on. We would hypothesize that genes that act as a (reversible) molecular switch in psoriasis could be irreversibly activated by genetic alterations in premalignant epidermis leading to actinic keratosis and eventually to squamous cell carcinoma (SCC). In combination with mutations in tumor suppressor genes or DNA repair genes, the ‘on’ switch of a hyperproliferative phenotype would effectively favor tumor promotion and further acquisition of genetic alterations, leading to tumor progression and malignancy.

Remarkably, the expression profile from the ‘normal’ epidermis of the patient was more similar to the tumor epidermis than to normal epidermis from a healthy volunteer. The apparent difference between the normal skin of the healthy individual and the patient could be that the skin of the patient was from a sun-exposed area, whereas the skin from the healthy individual was not. Alternatively, the immunosuppressive therapy that kidney transplant patients undergo could affect epidermal gene expression, resulting in deviation from the skin of non-immunosuppressed individuals. Our findings clearly warrant further investigation in a larger series of patients.

To exploit the full potential of large-scale expression data, not only appropriate statistical tools but also ones that allow biological interpretation of the data are required. Sets of coregulated genes identified in SAGE or microarray experiments should be analyzed in the available biomedical literature or databases for common properties with respect to tissue/cell type expression patterns and regulatory mechanisms such as similarities in promoter regions, signaling cascades, response to drugs, etc. Automated literature and database search tools would be extremely useful. A recently developed web-based interface such as PubGene is a step in this direction. Although limited by inconsistencies of gene symbols, synonyms, and abbreviations used in the biomedical literature, it provides a powerful tool to identify networks of genes that may be related by function or expression patterns. Chromosomal localization of these genes confirms the idea that they are likely to be coregulated, because four (S100A7, S100A8, S100A9, SPRR2A) are located in a gene cluster on chromosome 1q21.

How does this kind of analysis applied to SAGE data compare with analysis of microarray data? The throughput of microarray technology is higher than that of SAGE, and it is anticipated that microarrays will ultimately be the preferred methodology. However, we and others have now shown that it is possible to obtain medium-sized SAGE libraries from very small tissue samples (10⁵-10⁶ cells) yielding 100 ng of mRNA or less, which is still below the sensitivity of microarray hybridization unless sample amplification methods are used. Another advantage is that SAGE does not select for known genes but potentially can discover new genes, whereas with cDNA microarrays only the genes present on the array are addressed. Finally, data from SAGE libraries are now publicly available and could be used for the type of analyses described here. Use and further development of cluster analysis algorithms and automated literature search tools will generate effective tools for large-scale expression data analysis.

FOOTNOTES

¹ To read the full text of this article, go to http://www.fasebj.org/cgi/doi/10.1096/fj.01-0618fje; to cite this article, use FASEB J. (December 28, 2001) 10.1096/fj.01-0618fje

² Both authors have equally contributed to the manuscript.

This article has been cited by other articles:

				E. D. Emberley, Y. Niu, L. Curtis, S. Troup, S. K. Mandal, J. N. Myers, S. B. Gibson, L. C. Murphy, and P. H. Watson The S100A7-c-Jun Activation Domain Binding Protein 1 Pathway Enhances Prosurvival Pathways in Breast Cancer Cancer Res., July 1, 2005; 65(13): 5696 - 5702. [Abstract] [Full Text] [PDF]

				R. W. Georgantas III, V. Tanadve, M. Malehorn, S. Heimfeld, C. Chen, L. Carr, F. Martinez-Murillo, G. Riggins, J. Kowalski, and C. I. Civin Microarray and Serial Analysis of Gene Expression Analyses Identify Known and Novel Transcripts Overexpressed in Hematopoietic Stem Cells Cancer Res., July 1, 2004; 64(13): 4434 - 4441. [Abstract] [Full Text] [PDF]

				B. Norman, J. Davis, and J. Piatigorsky Postnatal Gene Expression in the Normal Mouse Cornea by SAGE Invest. Ophthalmol. Vis. Sci., February 1, 2004; 45(2): 429 - 440. [Abstract] [Full Text] [PDF]

				D. Porter, J. Lahti-Domenici, A. Keshaviah, Y. K. Bae, P. Argani, J. Marks, A. Richardson, A. Cooper, R. Strausberg, G. J. Riggins, et al. Molecular Markers in Ductal Carcinoma in Situ of the Breast Mol. Cancer Res., March 1, 2003; 1(5): 362 - 375. [Abstract] [Full Text] [PDF]

This Article Full Text (PDF) All Versions of this Article: 16/2/246 01-0618fjev1    most recent Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by VAN RUISSEN, F. Articles by SCHALKWIJK, J. Search for Related Content PubMed PubMed Citation Articles by VAN RUISSEN, F. Articles by SCHALKWIJK, J.