A molecular correlate to the Gleason grading system for prostate adenocarcinoma

Abstract

Adenocarcinomas of the prostate can be categorized into tumor grades based on the extent to which the cancers histologically resemble normal prostate glands. Because grades are surrogates of intrinsic tumor behavior, characterizing the molecular phenotype of grade is of potential clinical importance. To identify molecular alterations underlying prostate cancer grades, we used microdissection to obtain specific cohorts of cancer cells corresponding to the most common Gleason patterns (patterns 3, 4, and 5) from 29 radical prostatectomy samples. We paired each cancer sample with matched benign lumenal prostate epithelial cells and profiled transcript abundance levels by microarray analysis. We identified an 86-gene model capable of distinguishing low-grade (pattern 3) from high-grade (patterns 4 and 5) cancers. This model performed with 76% accuracy when applied to an independent set of 30 primary prostate carcinomas. Using tissue microarrays comprising >800 prostate samples, we confirmed a significant association between high levels of monoamine oxidase A expression and poorly differentiated cancers by immunohistochemistry. We also confirmed grade-associated levels of defender against death (DAD1) protein and HSD17 beta4 transcripts by immunohistochemistry and quantitative RT-PCR, respectively. The altered expression of these genes provides functional insights into grade-associated features of therapy resistance and tissue invasion. Furthermore, in identifying a profile of 86 genes that distinguish high- from low-grade carcinomas, we have generated a set of potential targets for modulating the development and progression of the lethal prostate cancer phenotype.

Fig. 1.

Gene-expression changes associated with prostate cancer grade. (A) Genes differentially expressed between microdissected benign and neoplastic prostate epithelium were identified in a supervised analysis of transcript profiles of 32 matched samples from 29 individuals. A Statistical Analysis Of Microarrays (SAM) one-sample t test comparing cancerous epithelium to matched benign epithelium across all Gleason patterns identified 736 differentially expressed genes (false discovery rate ≤0.05%). Cancer-associated transcript alterations for several genes with previously described alterations in prostate carcinoma are shown. (B) Identification of gene-expression alterations predictive of Gleason pattern. PAM analysis of gene-expression profiles generated from microdissected Gleason pattern 3, 4, and 5 cancers. Circles represent the training error. Triangles represent the leave-one-out cross-validation error. The x axis shows the number of array clones (and corresponding unique genes) used for classification. (C) Principal-components analysis of grade-specific prostate cancers using the 40-gene (51-clone) Gleason pattern classifier. For each sample, the score on the first (x axis) and second (y axis) principal component is plotted. Gleason pattern 3 (circle) samples are generally distinct from Gleason pattern 4 and 5 sample space (triangle and diamond, respectively). Pattern 4 and 5 samples were intermixed, indicating a high degree of molecular similarity. Three misclassified samples (under cross-validation of the 40-gene model) between pattern 3 and pattern 4/5 cancers are denoted in blue.

Fig. 2.

Genes associated with specific Gleason pattern prostate cancers. (A) Shown are the 86 genes exhibiting the greatest discriminatory power to partition cancer grades using an independent set of prostate cancer tissues. Genes are ordered according to t test score. The 40 genes used in the original PAM model are denoted by asterisks. (B) Application of the 86-gene Gleason classification model to an independent test set of 23 prostate cancers of singular pattern (e.g., 3 + 3) or exclusively high grade (4 + 4 or 4 + 5) resulted in an overall classification accuracy of 74%. The predicted pattern of 7 additional cancers of mixed low grade and high grade (3 + 4 and 4 + 3) varied between pattern 3 and pattern 4/5. The functional properties of grade-classifying genes are annotated in Table 1, which is published as supporting information on the PNAS web site.

Fig. 3.

IHC analysis of Gleason grade-associated gene expression. Representative immunohistochemical staining of Gleason pattern 3 (a and e), 4 (b and f), and 5 (c and g) prostate adenocarcinoma for MAOA (a–c) and DAD1 (e–g) expression. Magnifications are ×400. Summary of MAOA (1,358 scoreable samples) (d) and DAD1 (437 scoreable samples) (d and h) expression quantitation in tissue cores comprised of benign secretory epithelium and specific Gleason pattern adenocarcinomas.