Buckhaults Phillip, Zhang Zhen, Chen Yu-Chi, Wang Tian-Li, St Croix Brad, Saha Saurabh, Bardelli Alberto, Morin Patrice J, Polyak Kornelia, Hruban Ralph H, Velculescu Victor E, Shih Ie-Ming
The Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins, Johns Hopkins Medical Institutions, Baltimore, Maryland 21231, USA.
Cancer Res. 2003 Jul 15;63(14):4144-9.
Identifying the primary site in cases of metastatic carcinoma of unknown origin has profound clinical importance in managing cancer patients. Although transcriptional profiling promises molecular solutions to this clinical challenge, simpler and more reliable methods for this purpose are needed. A training set of 11 serial analysis of gene expression (SAGE) libraries was analyzed using a combination of supervised and unsupervised computational methods to select a small group of candidate genes with maximal power to discriminate carcinomas of different tissue origins. Quantitative real-time PCR was used to measure their expression levels in an independent validation set of 62 samples of ovarian, breast, colon, and pancreatic adenocarcinomas and normal ovarian surface epithelial controls. The diagnostic power of this set of genes was evaluated using unsupervised cluster analysis methods. From the training set of 21,321 unique SAGE transcript tags derived from 11 libraries, five genes were identified with expression patterns that distinguished four types of adenocarcinomas. Quantitative real-time PCR expression data obtained from the validation set clustered tumor samples in an unsupervised manner, generating a self-organized map with distinctive tumor site-specific domains. Eighty-one percent (50 of 62) of the carcinomas were correctly allocated in their corresponding diagnostic regions. Metastases clustered tightly with their corresponding primary tumors. A classification map diagnostic of tumor types was generated based on expression patterns of five genes selected from the SAGE database. This expression map analysis may provide a reliable and practical approach to determine tumor type in cases of metastatic carcinoma of clinically unknown origin.
确定原发部位对于不明原发灶转移性癌患者的治疗具有深远的临床意义。尽管转录谱分析有望为这一临床挑战提供分子解决方案，但仍需要更简单、更可靠的方法来实现这一目的。我们使用监督式和非监督式计算方法相结合，对一个包含11个基因表达序列分析（SAGE）文库的训练集进行分析，以筛选出一小部分具有最大鉴别不同组织来源癌能力的候选基因。采用定量实时PCR检测这些基因在一个独立验证集中的表达水平，该验证集包含62份卵巢、乳腺、结肠和胰腺腺癌样本以及正常卵巢表面上皮对照样本。使用非监督式聚类分析方法评估这组基因的诊断能力。从11个文库中获得的21,321个独特SAGE转录标签的训练集中，鉴定出5个基因，其表达模式可区分4种腺癌类型。从验证集获得的定量实时PCR表达数据以非监督方式对肿瘤样本进行聚类，生成一个具有独特肿瘤部位特异性区域的自组织图谱。81%（62个样本中的50个）的癌被正确分配到其相应的诊断区域。转移灶与其相应的原发肿瘤紧密聚类。基于从SAGE数据库中选择的5个基因的表达模式生成了一个诊断肿瘤类型的分类图谱。这种表达图谱分析可能为确定临床来源不明的转移性癌的肿瘤类型提供一种可靠且实用的方法。
