Background

Lithium is the most effective treatment for bipolar disorder, resulting in strong suicide prevention effects. The therapeutic range of lithium, however, is narrow and treatment initiation requires individual titration to address inter-individual variability. We aimed to improve lithium dose prediction using clinical and genomic data.

Methods

We performed a population pharmacokinetic study followed by a genome-wide association study (GWAS), including two clinical Swedish cohorts. Participants in cohort 1 were from specialised outpatient clinics at Huddinge Hospital, in Stockholm, Sweden, and participants in cohort 2 were identified using the Swedish National Quality Registry for Bipolar disorder (BipoläR). Patients who received a lithium dose corresponding to at least one tablet of lithium sulphate (6 mmol) per day and had clinically relevant plasma concentrations of lithium were included in the study. Data on age, sex, bodyweight, height, creatinine concentration, estimated glomerular filtration rate (eGFR), lithium preparation, number of tablets of lithium per day, serum lithium concentration, and medications affecting kidney function (C09 antihypertensives, C03 [except C03D] sodium-retaining diuretics, and non-steroidal anti-inflammatory drugs) were obtained retrospectively for several timepoints when possible from electronic health records, BipoläR, and the Swedish prescription registry. The median time between timepoints was 1·07 years for cohort 1 and 1·09 years for cohort 2. The primary outcome of interest was the natural logarithm of total body clearance for lithium (CL_Li) associated with the clinical variables. The residual effects after accounting for age and sex, representing the individual-level effects (CL_Li,age/sex), were used as the dependent variable in a GWAS.

Findings

2357 patients who were administered lithium (1423 women [60·4%] and 934 men [39·6%]; mean age 53·6 years [range 17–89], mainly of European descent) were included and 5627 data points were obtained. Age (variance explained [R²]: R²_cohort1=0·41 and R²_cohort2=0·31; both p<0·0001), sex (R²_cohort1=0·0063 [p=0·045] and R²_cohort2=0·026 [p<0·0001]), eGFR (R²_cohort1=0·38 and R²_cohort2=0·20; both p<0·0001), comedication with diuretics (R²_cohort1=0·0058 [p=0·014] and R²_cohort2=0·0026 [p<0·0001]), and agents acting on the renin–aldosterone–angiotensin system (R²_cohort1=0·028 and R²_cohort2=0·015; both p<0·0001) were clinical predictors of CL_Li. Notably, an association between CL_Li and serum lithium was observed, with a lower CL_Li being associated with higher serum lithium (R²_cohort1=0·13 and R²_cohort2=0·15; both p<0·0001). In a GWAS of CL_Li,age/sex, one locus was associated with a change in CL_Li (rs583503; β=–0·053 [95% CI –0·071 to –0·034]; p<0·00000005). We also found enrichment of the associations with genes expressed in the medulla (p=0·0014, corrected FDR=0·04) and cortex of the kidney (p=0·0015, corrected FDR=0·04), as well as associations with polygenic risk scores for eGFR (p value threshold: 0·05, p=0·01), body-mass index (p value threshold: 0·05, p=0·00025), and blood urea nitrogen (p value threshold: 0·001, p=0·00043). The model based on six clinical predictors explained 61·4% of the variance in CL_Li in cohort 1 and 49·8% in cohort 2. Adding genetic markers did not lead to major improvement of the models: within the subsample of genotyped individuals, the variance explained only increased from 59·32% to 59·36% in cohort 1 and from 49·21% to 50·03% in cohort 2 when including rs583503 and the four first principal components.

Interpretation

Our model predictors could be used clinically to better guide lithium dosage, shortening the time to reach therapeutic concentrations, thus improving care. Identification of the first genomic locus and PRS to be associated with CL_Li introduces the opportunity of individualised medicine in lithium treatment.

Funding

Stanley Medical Research Institute, Swedish Research Council, Swedish Foundation for Strategic Research, Swedish Brain Foundation, Swedish Research Council, Söderström-Königska Foundation, Bror Gadelius Minnesfond, Swedish Mental Health Fund, Karolinska Institutet and Hospital.

中文翻译：

---

使用群体药代动力学和药物基因组学改进锂剂量预测：瑞典的队列全基因组关联研究

背景

锂是双相情感障碍最有效的治疗方法，具有很强的自杀预防效果。然而，锂的治疗范围很窄，治疗开始需要个体滴定以解决个体间的差异。我们旨在利用临床和基因组数据改进锂剂量预测。

方法

我们进行了一项人群药代动力学研究，然后进行了一项全基因组关联研究 (GWAS)，其中包括两个临床瑞典队列。队列 1 的参与者来自瑞典斯德哥尔摩 Huddinge 医院的专科门诊，队列 2 的参与者是使用瑞典国家躁郁症质量登记处 (BipoläR) 确定的。接受锂剂量对应于每天至少一片硫酸锂 (6 mmol) 并具有临床相关血浆锂浓度的患者被纳入研究。年龄、性别、体重、身高、肌酐浓度、估计肾小球滤过率 (eGFR)、锂制剂、每日锂片数、血清锂浓度和影响肾功能的药物（C09 抗高血压药、C03 [除了 C03D] 钠潴留利尿剂和非甾体抗炎药）在可能的情况下从电子健康记录、BipoläR 和瑞典处方登记处回顾性地获得了几个时间点。时间点之间的中位时间对于队列 1 为 1·07 年，对于队列 2 为 1·09 年。感兴趣的主要结果是锂的全身清除率的自然对数（CL_Li ) 与临床变量相关。考虑到年龄和性别后的残余效应，代表个体水平效应（CL _Li,age/sex），被用作 GWAS 中的因变量。

发现

纳入了 2357 名服用锂的患者（1423 名女性 [60·4%] 和 934 名男性 [39·6%]；平均年龄 53·6 岁 [范围 17-89]，主要是欧洲血统），共有 5627 个数据点获得。年龄（方差解释 [ R ² ]：R ²_队列1 =0·41 和R ²_队列2 =0·31；均 p<0·0001），性别（R ²_队列1 =0·0063 [p=0·045] 和R ²_队列2 =0·026 [p<0·0001]），eGFR（R ²_队列1 =0·38 和R ²_队列2 =0·20；均p<0·0001），利尿剂联合用药（R ²_队列1=0·0058 [p=0·014] 和R ²_队列2 =0·0026 [p<0·0001]），以及作用于肾素-醛固酮-血管紧张素系统的药物（R ²_队列1 =0·028 和R ²_{队列 2} =0·015；p<0·0001) 都是 CL _Li的临床预测因子。值得注意的是，观察到 CL _Li与血清锂之间存在关联，较低的 CL _Li与较高的血清锂相关（R ²_队列1 =0·13 和R ²_队列2 =0·15；均 p<0·0001）。在 CL _Li,age/sex的 GWAS 中，一个基因座与 CL _{Li的变化有关}（rs583503；β=–0·053 [95% CI –0·071 至 –0·034]；p<0·00000005）。我们还发现与髓质（p=0·0014，校正的 FDR=0·04）和肾脏皮质（p=0·0015，校正的 FDR=0·04）以及与 eGFR（p 值阈值：0·05，p=0·01）、体重指数（p 值阈值：0·05，p=0·00025）和血尿素氮（p 值）的多基因风险评分相关阈值：0·001，p=0·00043）。基于六个临床预测因子的模型解释了 CL _{Li中 61·4% 的方差}在队列 1 和队列 2 中为 49·8%。添加遗传标记并没有导致模型的重大改进：在基因分型个体的子样本中，解释的方差仅在队列 1 中从 59·32% 增加到 59·36%当包括 rs583503 和四个第一主成分时，队列 2 中从 49·21% 到 50·03%。

解释

我们的模型预测因子可用于临床更好地指导锂剂量，缩短达到治疗浓度的时间，从而改善护理。_{与 CL Li}相关的第一个基因组位点和 PRS 的鉴定为锂治疗中的个体化医学带来了机会。

资金

斯坦利医学研究所、瑞典研究委员会、瑞典战略研究基金会、瑞典大脑基金会、瑞典研究委员会、Söderström-Königska 基金会、Bror Gadelius Minnesfond、瑞典心理健康基金、卡罗林斯卡学院和医院。