OBJECTIVE: The objectives of this study were to determine whether there are distinct gene expression profiles for endometrial and ovarian endometrioid carcinomas and to create a statistical model using these profiles to predict their organ of origin. METHODS: Expression profiles of seven stage I, grades 1 and 2, endometrioid endometrial carcinomas and seven stage I ovarian endometrioid carcinomas were analyzed as the training set. The test set included seven advanced endometrial carcinomas and nine dual primary endometrial and ovarian primary tumors of endometrioid histology. Unsupervised hierarchical clustering, multidimensional scaling (MDS) and predictive analysis of microarrays (PAM) were used to analyze the data. RESULTS: We identified 163 differentially expressed (DE) genes between endometrial and ovarian tumors. Both unsupervised hierarchical clustering and multidimensional scaling showed clear separation of the two groups. Pathway analysis of the 163 DE genes revealed significant biological differences between the two groups, which included metabolic and biosynthetic pathways. Further classification analysis on the training data using predictive analysis of microarray generated a 119-gene predictive model with 100% cross-validation accuracy. When the 119-gene model was applied to the test set of 16 samples, we found concordance in 11/16 samples and discordance in 5/16 samples with the pathologic classification. CONCLUSION: We conclude that, although similar in histologic appearance, endometrioid carcinomas of the ovary and endometrium have distinct gene expression patterns. A larger test set is needed to prospectively evaluate this prediction model further.
OBJECTIVE: The objectives of this study were to determine whether there are distinct gene expression profiles for endometrial and ovarian endometrioid carcinomas and to create a statistical model using these profiles to predict their organ of origin. METHODS: Expression profiles of seven stage I, grades 1 and 2, endometrioid endometrial carcinomas and seven stage I ovarian endometrioid carcinomas were analyzed as the training set. The test set included seven advanced endometrial carcinomas and nine dual primary endometrial and ovarian primary tumors of endometrioid histology. Unsupervised hierarchical clustering, multidimensional scaling (MDS) and predictive analysis of microarrays (PAM) were used to analyze the data. RESULTS: We identified 163 differentially expressed (DE) genes between endometrial and ovarian tumors. Both unsupervised hierarchical clustering and multidimensional scaling showed clear separation of the two groups. Pathway analysis of the 163 DE genes revealed significant biological differences between the two groups, which included metabolic and biosynthetic pathways. Further classification analysis on the training data using predictive analysis of microarray generated a 119-gene predictive model with 100% cross-validation accuracy. When the 119-gene model was applied to the test set of 16 samples, we found concordance in 11/16 samples and discordance in 5/16 samples with the pathologic classification. CONCLUSION: We conclude that, although similar in histologic appearance, endometrioid carcinomas of the ovary and endometrium have distinct gene expression patterns. A larger test set is needed to prospectively evaluate this prediction model further.
Authors: Diana L Kolbe; Julie A DeLoia; Patricia Porter-Gill; Mary Strange; Hanna M Petrykowska; Alfred Guirguis; Thomas C Krivak; Lawrence C Brody; Laura Elnitski Journal: PLoS One Date: 2012-03-05 Impact factor: 3.240