Table 1_Machine learning–derived genetic risk scores identify IL21 as a predictor of response to omalizumab and dupilumab in asthma.docx

RationaleGenetic risk scores (GRS) of Th1/2/17-related loci may be associated with response to biologics. We leveraged previously published machine learning-derived GRSs associated with plasma proteins from the INTERVAL/UK-Biobank study.
MethodsWe assessed 42 Th1/2/17-related GRSs and SNPs for association with response (≥50% reduction in exacerbations) to biologics in 172 White patients with moderate-to-severe asthma in the Mass General Brigham Biobank (MGBB: 92 omalizumab, 38 mepolizumab, 42 dupilumab). Replication was sought in 243 individuals in the All of Us (AoU) cohort (111 omalizumab, 58 mepolizumab, 74 dupilumab). Models adjusted for age, sex, BMI, baseline exacerbations, and principal components 1–10. AUROC was used to evaluate top predictors; type I error was assessed using random GRS sets (target FDR ≤20%).
ResultsFemales comprised a large proportion; mean BMI was 28–35 kg/m2. IL21 GRS was associated with omalizumab response in MGBB (OR: 1.7, 95% CI: 1.03–2.87) with similar direction in AoU (1.5, 0.91–2.45). IL21 also predicted dupilumab response in MGBB (2.4, 1.05–5.44) but in the opposite direction in AoU (0.57, 0.31–1.06). IL21 replicated as a predictor of omalizumab [AUROC, 95% CI: MGBB 0.62 (0.50–0.74), AoU: 0.71 (0.61–0.81)] and dupilumab [AUROC, 95% CI, MGBB 0.76 (0.58–0.95), AoU: 0.75 (0.64–0.86)]. Adding IL5RA (omalizumab) or CCL17 (dupilumab) modestly improved AUROC but not significantly. No GRS predicted mepolizumab response.
ConclusionsUsing ML-based GRS applied to an independent cohort of asthma patients, we found that IL-21-related GRSs were predictors of response to omalizumab and dupilumab.

研究依据：与T辅助细胞1/2/17（Th1/2/17）相关位点的遗传风险评分（Genetic Risk Scores, GRS）可能与生物制剂应答存在关联。本研究利用此前已发表的、来自INTERVAL/英国生物银行（INTERVAL/UK-Biobank）研究中与血浆蛋白相关的机器学习衍生遗传风险评分。
方法：本研究在麻省总医院布里格姆生物银行（Mass General Brigham Biobank, MGBB）的172例白人中重度哮喘患者中，评估42个Th1/2/17相关GRS及单核苷酸多态性（Single Nucleotide Polymorphism, SNP）与生物制剂应答（急性加重次数较基线减少≥50%）的关联：其中92例接受奥马珠单抗（omalizumab）治疗、38例接受美泊利单抗（mepolizumab）治疗、42例接受度普利尤单抗（dupilumab）治疗。本研究在“我们所有人”（All of Us, AoU）队列的243例个体中开展验证：其中111例接受奥马珠单抗治疗、58例接受美泊利单抗治疗、74例接受度普利尤单抗治疗。所有模型均校正年龄、性别、体质量指数（BMI）、基线急性加重次数及主成分1至10。采用受试者工作特征曲线下面积（Area Under the Receiver Operating Characteristic curve, AUROC）评估最优预测因子；采用随机GRS集评估I类错误，设定目标错误发现率（False Discovery Rate, FDR）≤20%。
结果：本研究队列女性占比颇高，平均BMI为28~35 kg/m²。在MGBB队列中，白细胞介素21（IL-21）GRS与奥马珠单抗应答显著相关（比值比Odds Ratio, OR=1.7，95%置信区间Confidence Interval, CI：1.03~2.87），且在AoU队列中效应方向一致（OR=1.5，95%CI：0.91~2.45）。IL-21 GRS同样可预测MGBB队列中的度普利尤单抗应答（OR=2.4，95%CI：1.05~5.44），但在AoU队列中效应方向相反（OR=0.57，95%CI：0.31~1.06）。IL-21 GRS作为奥马珠单抗[AUROC, 95%CI：MGBB队列0.62（0.50~0.74），AoU队列0.71（0.61~0.81）]与度普利尤单抗[AUROC, 95%CI：MGBB队列0.76（0.58~0.95），AoU队列0.75（0.64~0.86）]的预测因子得到验证。加入IL5RA（奥马珠单抗）或CCL17（度普利尤单抗）可小幅提升AUROC值，但未达到统计学显著性。未发现可预测美泊利单抗应答的GRS。
结论：基于机器学习衍生的GRS，在独立哮喘患者队列中开展本研究后，我们发现IL-21相关GRS可预测奥马珠单抗与度普利尤单抗的治疗应答。