Lenburg Marc E, Liou Louis S, Gerry Norman P, Frampton Garrett M, Cohen Herbert T, Christman Michael F
Department of Genetics & Genomics, Boston University School of Medicine 715 Albany Street, E613 Boston, Massachusetts 02118, USA.
BMC Cancer. 2003 Nov 27;3:31. doi: 10.1186/1471-2407-3-31.
Renal cell carcinoma is a common malignancy that often presents as a metastatic-disease for which there are no effective treatments. To gain insights into the mechanism of renal cell carcinogenesis, a number of genome-wide expression profiling studies have been performed. Surprisingly, there is very poor agreement among these studies as to which genes are differentially regulated. To better understand this lack of agreement we profiled renal cell tumor gene expression using genome-wide microarrays (45,000 probe sets) and compare our analysis to previous microarray studies.
We hybridized total RNA isolated from renal cell tumors and adjacent normal tissue to Affymetrix U133A and U133B arrays. We removed samples with technical defects and removed probesets that failed to exhibit sequence-specific hybridization in any of the samples. We detected differential gene expression in the resulting dataset with parametric methods and identified keywords that are overrepresented in the differentially expressed genes with the Fisher-exact test.
We identify 1,234 genes that are more than three-fold changed in renal tumors by t-test, 800 of which have not been previously reported to be altered in renal cell tumors. Of the only 37 genes that have been identified as being differentially expressed in three or more of five previous microarray studies of renal tumor gene expression, our analysis finds 33 of these genes (89%). A key to the sensitivity and power of our analysis is filtering out defective samples and genes that are not reliably detected.
The widespread use of sample-wise voting schemes for detecting differential expression that do not control for false positives likely account for the poor overlap among previous studies. Among the many genes we identified using parametric methods that were not previously reported as being differentially expressed in renal cell tumors are several oncogenes and tumor suppressor genes that likely play important roles in renal cell carcinogenesis. This highlights the need for rigorous statistical approaches in microarray studies.
肾细胞癌是一种常见的恶性肿瘤，常表现为转移性疾病，目前尚无有效的治疗方法。为深入了解肾细胞癌的发生机制，已开展了多项全基因组表达谱研究。令人惊讶的是，这些研究在哪些基因存在差异调节方面的一致性很差。为了更好地理解这种不一致性，我们使用全基因组微阵列（45,000个探针集）对肾细胞肿瘤基因表达进行了分析，并将我们的分析结果与之前的微阵列研究进行了比较。
我们将从肾细胞肿瘤和相邻正常组织中分离出的总RNA与Affymetrix U133A和U133B阵列进行杂交。我们去除了存在技术缺陷的样本，并去除了在任何样本中均未表现出序列特异性杂交的探针集。我们使用参数方法在所得数据集中检测差异基因表达，并通过Fisher精确检验确定在差异表达基因中过度表达的关键词。
通过t检验，我们在肾肿瘤中鉴定出1234个基因的变化超过三倍，其中800个基因此前未被报道在肾细胞肿瘤中发生改变。在之前五项肾肿瘤基因表达微阵列研究中，只有37个基因被鉴定为在三项或更多研究中差异表达，我们的分析发现其中33个基因（89%）。我们分析的敏感性和效能的关键在于滤除有缺陷的样本和未可靠检测到的基因。
广泛使用的用于检测差异表达的样本投票方案未对假阳性进行控制，这可能是之前研究之间重叠性差的原因。在我们使用参数方法鉴定出的众多此前未被报道在肾细胞肿瘤中差异表达的基因中，有几个癌基因和肿瘤抑制基因可能在肾细胞癌发生中起重要作用。这凸显了在微阵列研究中采用严格统计方法的必要性。
