Molecular signaling distinguishes early from late recurrences in ER positive breast cancer

Unlike many other cancers, estrogen receptor-alpha (ER+) breast cancers are associated with cumulative risks of recurrence and death that persist for decades. We show that molecular differences between breast cancers that recur at distant sites early (² 3 years) or late (³ 5 years) support a robust molecular predictor of recurrence with Tamoxifen therapy and provide novel insights into the signaling features that differ between these recurrent cancers. We applied a support vector machine with recursive feature elimination to gene expression microarray data using a training-internal crossvalidation workflow that minimizes the gene selection bias problem. The resulting predictor was validated in an independent data set. Performance of the predictor suggests that it is possible to identify patients at increased risk of experiencing an early recurrence who require other treatments to prevent early metastasis. We implemented a Metropolis Sampling algorithm as a random walk to identify the protein-protein interactions (PPI) most closely associated with ER in 8 PPI databases. We then walked the gene expression data for the top PPIs to discover a signaling network driving early and late recurrent breast cancers. Consensus features between the training and validation datasets define a complex and highly connected network with major interactions among nodes including AR, CALD1, CALM(1,2,3), CDK1, EGFR, ESR1, ESR2, MAPK1, and SRC. The complexity illustrates the challenges in directing single agent or simple combination therapies to improve overall survival in ER+ breast cancers that will recur but also suggests potentially novel interventions to address this challenge. 49 independent samples (breast tumors) were arrayed

与绝大多数其他癌症不同，雌激素受体α阳性（estrogen receptor-alpha, ER+）乳腺癌的复发与死亡累积风险会持续数十年之久。本研究证实，早期（≤3年）或晚期（≥5年）发生远处转移复发的乳腺癌之间存在分子差异，此类差异既可为他莫昔芬（Tamoxifen）治疗后的复发预测提供可靠的分子标志物，也为两类复发癌症的信号通路特征差异提供了全新的研究视角。我们采用带递归特征消除的支持向量机（support vector machine, SVM），结合训练集内部交叉验证流程处理基因表达微阵列数据，以最大程度降低基因选择偏倚问题。所得预测模型在独立数据集上完成了有效性验证。该预测模型的性能表现表明，我们能够识别出早期复发风险升高的患者群体，这类患者需采用其他治疗手段以预防早期远处转移。我们借助以随机游走形式实现的梅特罗波利斯采样（Metropolis Sampling）算法，在8个蛋白质相互作用（protein-protein interaction, PPI）数据库中筛选出与ER密切相关的蛋白质相互作用事件。随后，我们针对筛选得到的Top级蛋白质相互作用的基因表达数据进行分析，以挖掘驱动早期与晚期复发乳腺癌的信号调控网络。训练集与验证集之间的共识特征共同构建了一个复杂且高度互联的信号网络，其节点间存在大量关键相互作用，涉及AR、CALD1、CALM(1,2,3)、CDK1、EGFR、ESR1、ESR2、MAPK1及SRC等分子。该网络的复杂性一方面凸显了针对复发性ER+乳腺癌，采用单一药物或简单联合疗法以改善总生存期所面临的重重挑战，另一方面也为破解这一难题提供了潜在的新型干预思路。本研究共对49份独立乳腺肿瘤样本完成了芯片阵列检测。