Treatment resistance is a major clinical challenge of diffuse large B-cell lymphoma (DLBCL) where approximately 40% of the patients have refractory disease or relapse. Since DLBCL is characterized by great clinical and molecular heterogeneity, the purpose of the present study was to investigate whether miRNAs associated to single drug components of R-CHOP can improve robustness of individual markers and serve as a prognostic classifier. Fifteen DLBCL cell lines were tested for sensitivity towards single drug compounds of the standard treatment R-CHOP: rituximab (R), cyclophosphamide (C), doxorubicin (H), and vincristine (O). For each drug, cell lines were ranked using the area under the dose-response curve and grouped as either sensitive, intermediate or resistant. Baseline miRNA expression data were obtained for each cell line in untreated condition, and differential miRNA expression analysis between sensitive and resistant cell lines identified 43 miRNAs associated to growth response after exposure towards single drugs of R-CHOP. Using the Affymetrix HG-U133 platform, expression levels of miRNA precursors were assessed in 701 diagnostic DLBCL biopsies, and miRNA-panel classifiers predicting disease progression were build using multiple Cox regression or random survival forest. Classifiers were validated and ranked by repeated cross-validation. Prognostic accuracies were assessed by Brier Scores and time-varying area under the ROC curves, which revealed better performance of multivariate Cox models compared to random survival forest models. The Cox model including miR-146a, miR-155, miR-21, miR-34a, and miR-23a~miR-27a~miR-24-2 cluster performed the best and successfully stratified GCB-DLBCL patients into high- and low-risk of disease progression. In addition, combination of the Cox miRNA-panel and IPI substantially increased prognostic performance in GCB classified patients. As a proof of concept, we found that expression data of drug associated miRNAs display prognostic utility and adding these to IPI improves prognostic stratification of GCB-DLBCL patients treated with R-CHOP.

中文翻译：

---

与R-CHOP单一药物成分相关的MicroRNAs可识别结局较差的弥漫性大B细胞淋巴瘤患者，并为国际预后指标增加预后价值

抗药性是弥漫性大B细胞淋巴瘤（DLBCL）的主要临床挑战，其中约40％的患者患有难治性疾病或复发。由于DLBCL具有很大的临床和分子异质性特征，因此本研究的目的是调查与R-CHOP单一药物成分相关的miRNA是否可以提高单个标记物的稳健性并作为预后分类器。测试了15种DLBCL细胞系对标准治疗R-CHOP的单一药物化合物的敏感性：利妥昔单抗（R），环磷酰胺（C），阿霉素（H）和长春新碱（O）。对于每种药物，使用剂量反应曲线下的面积对细胞系进行排名，并分为敏感，中度或耐药性。在未处理的条件下获得了每个细胞系的基线miRNA表达数据，敏感和耐药细胞系之间的差异miRNA表达分析确定了暴露于R-CHOP单一药物后与生长反应相关的43个miRNA。使用Affymetrix HG-U133平台，在701例诊断性DLBCL活检中评估了miRNA前体的表达水平，并使用多重Cox回归或随机生存森林构建了预测疾病进展的miRNA面板分类器。通过重复交叉验证对分类器进行验证和排名。通过在ROC曲线下的Brier分数和时变面积评估了预后准确性，与随机生存森林模型相比，多变量Cox模型的性能更好。Cox模型包括miR-146a，miR-155，miR-21，miR-34a，miR-23a〜miR-27a〜miR-24-2簇表现最好，成功地将GCB-DLBCL患者分为疾病进展的高危和低危。此外，Cox miRNA-panel和IPI的组合可显着提高GCB分类患者的预后性能。作为概念的证明，我们发现药物相关miRNA的表达数据显示出预后效用，并将这些数据添加到IPI可以改善用R-CHOP治疗的GCB-DLBCL患者的预后分层。