Self-organizing-map-based molecular signature representing the development of hepatocellular carcinoma

Using high-density oligonucleotide array, we comprehensively analyzed expression levels of 12600 genes in 50 hepatocellular carcinoma (HCC) samples with positive hepatitis C virus (HCV) serology (well (G1), moderately (G2), and poorly (G3) differentiated tumors) and 11 non-tumorous livers (L1 and L0) with and without HCV infection. We searched for discriminatory genes of transition (L0 vs. L1, L1 vs. G1, G1 vs. G2, G2 vs. G3) with a supervised learning method, and then arranged the samples by self-organizing map (SOM) with the discriminatory gene sets. The SOM arranged the five clusters on a unique sigmoidal curve in the order L0, L1, G1, G2, and G3. The sample arrangement reproduced development-related features of HCC such as p53 abnormality. Strikingly, G2 tumors without venous invasion were located closer to the G1 cluster, and most G2 tumors with venous invasion were located closer to the G3 cluster (P=0.001 by Fisher's exact test). Our present profiling data will serve as a framework to understand the relation between the development and dedifferentiation of HCC.