Predicting Drug Resistance in Cancer Genomes by DMA Methylation Profiling

卵巢癌是所有妇科肿瘤中最致命的。虽然卵巢肿瘤对化疗药物的耐药性是一个常见的问题，这种多因素现象的潜在机制仍然知之甚少。虽然CpG岛（CpG i）甲基化可能在癌症耐药的复杂性中起着重要作用，但这在卵巢癌中尚未得到广泛解决，顺铂诱导的获得性DNA甲基化的新现象也没有得到广泛解决。我们先前在耐药卵巢癌中对CpGi基因座进行了全基因组询问，并确定了一个CpGi亚组，该亚组在耐药细胞系和复发肿瘤中差异性高甲基化，并且与较短的生存期密切相关。此外，在这些基因座中，我们确定了保守的DNA序列，甲基化倾向的CpGi的特征。我们假设CpGi甲基化与顺铂耐药相关。我们将通过使用卵巢癌耐药模型和患者肿瘤样本来验证这一假设。我们将进行计算建模，使用耐药的高甲基化CpGi基因座作为我们的机器训练集。一个无监督的学习方案，具体地说，一个序列聚类算法，将被用来分类CpGi序列成组的相似序列。CpGi序列中的模式或顺序将用于进一步选择与卵巢癌中药物诱导的DNA甲基化相关的甲基化倾向CpGi序列。甲基化特异性寡核苷酸（MSO）微阵列方法将用于确定顺铂选择的子卵巢癌亚系、药物敏感的亲代系、化疗前的原发性卵巢肿瘤和复发时获得的肿瘤中CpGi基因座的详细甲基化模式。此外，序列预测和MSO结果将使用未用药卵巢癌细胞系和多次化疗处理的相同细胞系之间的时程甲基化比较来直接评估。 顺铂的时间。我们设计了这个时间过程的实验系统，以确定CpGi基因位点易受药物诱导的从头甲基化，并直接关联CpGi与卵巢癌顺铂耐药的发展。为了进一步确认特定的、相关的基因座的作用，我们将使用小干扰RNA技术敲除药物敏感细胞系中的这些基因，并检查随后对药物敏感性的影响。生物学实验的结果将使用监督或无监督学习方法进行分析。对实验数据的计算分析将指导假设的重新表述。整个过程可以迭代，以生成和完善数学模型，预测药物诱导的CpGi甲基化，并可能识别表观遗传复发生物标志物， 卵巢癌

Ovarian cancer is the most lethal of all gynecological neoplasms. Although ovarian tumor resistance to chemotherapeutic drugs is a common problem, the underlying mechanisms of this multifactorial phenomenon remain poorly understood. While CpG island (CpGi) methylation likely plays a prominent role in the complexity of drug resistance in cancer, this has not been widely addressed in ovarian cancer, nor has the emerging phenomenon of acquired DNA methylation induced by cisplatin. We previously performed a genome-wide interrogation of CpGi loci in drug-resistant ovarian cancer and identified a subset of CpGi that were differentially hypermethylated in drug-resistant cell lines and relapse tumors and also were strongly correlated with shorter survival. Furthermore, within these loci, we identified conserved DNA sequences, characteristic of methylation-prone CpGi. We hypothesize that CpGi methylation is associated with cisplatin resistance. We will test this hypothesis by using models of ovarian cancer drug resistance and patient tumor samples. We will conduct computational modeling, using the drug-resistant, hypermethylated CpGi loci as our machine training set. An unsupervised learning scheme, specifically, a sequence clustering algorithm, will be used to classify CpGi sequences into groups of similar sequences. Patterns or subsequences in CpGi sequences will be used to further select methylation-prone CpGi sequences associated with drug-induced DNA methylation in ovarian cancer. The methylation-specific oligonucleotide (MSO) microarray method will be used to determine detailed methylation patterns of the CpGi loci in cisplatin-selected daughter ovarian cancer sublines, the drug-sensitive parental line, primary ovarian tumors taken before chemotherapy, and tumors obtained at relapse. Furthermore, the sequence prediction and MSO results will be evaluated directly, using a time-course methylation comparison between a drug-naive ovarian cancer cell line and the same cell line treated multiple times with cisplatin. We have designed this time-course experimental system to determine CpGi loci susceptible to drug-induced ate novo methylation and also to directly associate CpGi with the development of cisplatin resistance in ovarian cancer. To further confirm the roles of specific, implicated loci, we will use small interfering RNA technology to knock down such genes in drug sensitive cell lines and examine the subsequent effects on drug sensitivity. The results from the biological experiments will be analyzed using supervised or unsupervised learning methods. The computational analysis of experimental data will guide the reformulation of hypotheses. The entire procedure can be iterated to generate and refine mathematical models that predict drug-induced CpGi methylation and perhaps identify epigenetic relapse biomarkers in ovarian cancer.