POSTER NO: 198

Distinct molecular basis of malignancy and metastatic potential in human colorectal carcinoma

¹Kikuya Kato, ¹Shizuko Muro, ²Ichiro Takemasa, ¹Ryo Matoba, ²Morito Monden
¹Taisho Laboratory of Functional Genomics, Nara Institute of Science and Technology, 8916-5 Takayama, Ikoma, Nara, Japan, ²Clinical Oncology and Surgery, Osaka University Medical School, 2-2 Yamadaoka, Suita, Osaka, Japan

Colorectal carcinoma is one of the most prevalent cancers, and thus has been most intensively studied. Gene expression profiling is now being applied to human cancer studies. In particular, it is expected to be useful for elucidating the molecular features underlying variations among individual cancers.

We analyzed RNA purified from human colorectal carcinomas using a novel PCR array technique, which is a high throughput quantitative PCR system based on adapter-tagged competitive PCR (ATAC-PCR). With the PCR array, we can detect subtle changes in gene expression within cancer cells through the analysis of an RNA mixture derived from various cell types.

We first surveyed the expression of genes in colorectal cancer tissues through EST sequencing. We produced a 3' end-directed cDNA library from a mixture of RNAs, purified from eight colorectal cancer tissues. A total of 5465 EST clones were sequenced. Among them, we selected 1536 genes which were also deposited in the RefSeq database. We designed PCR primers for all 1344 genes, and 192 other genes known to be involved in colon cancer or identified as tumor-specific by previous microarray experiments. This gene set includes only those genes that are genuinely expressed in colorectal cancers, an advantage over the more universal sets, such as UniGene, including genes not detected in colorectal cancer. We analyzed sample RNAs derived from 104 cancer and 11 normal tissues. Consequently, this analysis yielded a data matrix consisting of 1536 genes x 115 tissue samples.

After data normalization and log conversion, we grouped genes by means of expression patterns by hierarchical cluster analysis. Genes are grouped into 88 clusters statistically significant. The sizes of clusters varied greatly, from one to 154 genes. There were eight clusters having more than 50 genes: nearly two thirds of genes belonged to one of them. We then examined these large clusters for correlation with clinical parameters such as prognosis, remote metastasis, and histological types. Hierarchical cluster analysis revealed that a cluster with 126 genes (designated as M-cluster) clearly discriminated tumor and normal tissues, and exhibited correlation with remote metastasis. The metastatic potential was correlated with low gene expression. Principal component analysis (PCA) revealed that cancer and normal tissues were clearly separated by the first two components. Samples with metastasis were on the opposite side of normal samples in the plot. M-cluster, however, showed no correlation with prognosis.

Another cluster with 136 genes, designated as P-cluster, was correlated with both prognosis and metastasis, but, did not classified tumors from normal tissues. Variation in gene expression essential for the correlation was extracted as the first component of PCA.

Because genes correlated with prognosis and metastasis were not identical, malignancy and metastatic potential are different properties, and are likely to be based on independent molecular mechanisms.