Zhang Yan-Yan, Liu Houfu, Summerfield Scott G, Luscombe Christopher N, Sahi Jasminder
Drug Metabolism and Pharmacokinetics, Platform Technology and Science China, GlaxoSmithKline R&D , Shanghai, China.
David Jack Centre for R&D, GlaxoSmithKline R&D , Park Road, Ware, Hertfordshire, SG12 0DP, U.K.
Mol Pharm. 2016 May 2;13(5):1540-50. doi: 10.1021/acs.molpharmaceut.6b00031. Epub 2016 Apr 4.
Estimation of uptake across the blood-brain barrier (BBB) is key to designing central nervous system (CNS) therapeutics. In silico approaches ranging from physicochemical rules to quantitative structure-activity relationship (QSAR) models are utilized to predict potential for CNS penetration of new chemical entities. However, there are still gaps in our knowledge of (1) the relationship between marketed human drug derived CNS-accessible chemical space and preclinical neuropharmacokinetic (neuroPK) data, (2) interpretability of the selected physicochemical descriptors, and (3) correlation of the in vitro human P-glycoprotein (P-gp) efflux ratio (ER) and in vivo rodent unbound brain-to-blood ratio (Kp,uu), as these are assays routinely used to predict clinical CNS exposure, during drug discovery. To close these gaps, we explored the CNS druglike property boundaries of 920 market oral drugs (315 CNS and 605 non-CNS) and 846 compounds (54 CNS drugs and 792 proprietary GlaxoSmithKline compounds) with available rat Kp,uu data. The exact permeability coefficient (Pexact) and P-gp ER were determined for 176 compounds from the rat Kp,uu data set. Receiver operating characteristic curves were performed to evaluate the predictive power of human P-gp ER for rat Kp,uu. Our data demonstrates that simple physicochemical rules (most acidic pKa ≥ 9.5 and TPSA < 100) in combination with P-gp ER < 1.5 provide mechanistic insights for filtering BBB permeable compounds. For comparison, six classification modeling methods were investigated using multiple sets of in silico molecular descriptors. We present a random forest model with excellent predictive power (∼0.75 overall accuracy) using the rat neuroPK data set. We also observed good concordance between the structural interpretation results and physicochemical descriptor importance from the Kp,uu classification QSAR model. In summary, we propose a novel, hybrid in silico/in vitro approach and an in silico screening model for the effective development of chemical series with the potential to achieve optimal CNS exposure.
评估药物通过血脑屏障（BBB）的摄取情况是设计中枢神经系统（CNS）治疗药物的关键。从物理化学规则到定量构效关系（QSAR）模型等计算机模拟方法被用于预测新化学实体穿透中枢神经系统的潜力。然而，在以下几个方面我们的认识仍存在差距：（1）已上市的人用药物衍生的中枢神经系统可及化学空间与临床前神经药代动力学（neuroPK）数据之间的关系；（2）所选物理化学描述符的可解释性；（3）体外人P-糖蛋白（P-gp）外排率（ER）与体内啮齿动物未结合脑血比（Kp,uu）之间的相关性，因为这些是药物研发过程中常规用于预测临床中枢神经系统暴露的检测方法。为了填补这些差距，我们研究了920种市售口服药物（315种中枢神经系统药物和605种非中枢神经系统药物）以及846种化合物（54种中枢神经系统药物和792种葛兰素史克 proprietary化合物）的中枢神经系统类药物性质边界，这些化合物具有可用的大鼠Kp,uu数据。根据大鼠Kp,uu数据集确定了176种化合物的精确渗透系数（Pexact）和P-gp ER。进行了受试者工作特征曲线分析，以评估人P-gp ER对大鼠Kp,uu的预测能力。我们的数据表明，简单的物理化学规则（最酸性pKa≥9.5且TPSA<100）与P-gp ER<1.5相结合，为筛选可穿透血脑屏障的化合物提供了机制性见解。为作比较，使用多组计算机模拟分子描述符研究了六种分类建模方法。我们提出了一个使用大鼠神经药代动力学数据集且具有出色预测能力（总体准确率约为0.75）的随机森林模型。我们还观察到Kp,uu分类QSAR模型的结构解释结果与物理化学描述符重要性之间具有良好的一致性。总之，我们提出了一种新颖的计算机模拟/体外混合方法以及一种计算机模拟筛选模型，用于有效开发有可能实现最佳中枢神经系统暴露的化学系列药物。
