Division of Infectious Diseases and Geographic Medicine, Stanford University School of Medicine, Stanford, California.
Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco, San Francisco, California.
Am J Respir Crit Care Med. 2021 Dec 1;204(11):1317-1326. doi: 10.1164/rccm.202103-0564OC.
Standardized dosing of antitubercular drugs contributes to a substantial incidence of toxicities, inadequate treatment response, and relapse, in part due to variable drug concentrations achieved. SNPs in the (-acetyltransferase-2) gene explain the majority of interindividual pharmacokinetic variability of isoniazid (INH). However, an obstacle to implementing pharmacogenomic-guided dosing is the lack of a point-of-care assay. To develop and test a classification algorithm, validate its performance in predicting isoniazid clearance, and develop a prototype pharmacogenomic assay. We trained random forest models to predict acetylation genotype from unphased SNP data using a global collection of 8,561 phased genomes. We enrolled 48 patients with pulmonary tuberculosis, performed sparse pharmacokinetic sampling, and tested the acetylator prediction algorithm accuracy against estimated INH clearance. We then developed a cartridge-based multiplex quantitative PCR assay on the GeneXpert platform and assessed its analytical sensitivity on whole blood samples from healthy individuals. With a 5-SNP model trained on two-thirds of the data ( = 5,738), out-of-sample acetylation genotype prediction accuracy on the remaining third ( = 2,823) was 100%. Among the 48 patients with tuberculosis, predicted acetylator types were 27 (56.2%) slow, 16 (33.3%) intermediate, and 5 (10.4%) rapid. INH clearance rates were lowest in predicted slow acetylators (median 14.5 L/h), moderate in intermediate acetylators (median 40.3 L/h), and highest in fast acetylators (median 53.0 L/h). The cartridge-based assay accurately detected all allele patterns directly from 25 μl of whole blood. An automated pharmacogenomic assay on a platform widely used globally for tuberculosis diagnosis could enable personalized dosing of INH.
抗结核药物的标准化给药方案导致毒性、治疗反应不足和复发的发生率相当高，部分原因是达到的药物浓度不同。乙酰转移酶-2 基因中的 SNP 解释了异烟肼（INH）个体间药代动力学变异性的大部分。然而，实施基于药物基因组学的剂量指导的一个障碍是缺乏即时检测（point-of-care assay）。为了开发和测试分类算法，验证其预测异烟肼清除率的性能，并开发原型药物基因组学检测。我们使用全球范围内的 8561 个定相基因组数据集，训练随机森林模型，从未定相 SNP 数据中预测乙酰化基因型。我们招募了 48 例肺结核患者，进行了稀疏的药代动力学采样，并根据估计的 INH 清除率测试了乙酰化预测算法的准确性。然后，我们在 GeneXpert 平台上开发了基于盒式的多重定量 PCR 检测，并评估了其对健康个体全血样本的分析灵敏度。使用三分之二数据（= 5738）训练的 5-SNP 模型，对剩余三分之一数据（= 2823）的外样本乙酰化基因型预测准确性为 100%。在 48 例肺结核患者中，预测的乙酰化类型为 27 例（56.2%）为慢代谢型，16 例（33.3%）为中代谢型，5 例（10.4%）为快代谢型。预测的慢代谢型 INH 清除率最低（中位数 14.5 L/h），中间代谢型（中位数 40.3 L/h）和快代谢型（中位数 53.0 L/h）中等。基于盒式的检测能够从 25 μl 全血中直接准确地检测到所有等位基因模式。一种在全球广泛用于结核病诊断的平台上的自动化药物基因组学检测，可以实现 INH 的个体化给药。
