Department of Medicine, Duke University Medical Center, DUMC Box 2629, Durham, NC, 27710, USA.
Department of Biostatistics and Bioinformatics, Duke University Medical Center, Durham, NC, USA.
J Neurovirol. 2023 Feb;29(1):78-93. doi: 10.1007/s13365-022-01102-2. Epub 2022 Nov 8.
This study sought to identify neuroimaging and immunological factors associated with substance use and that contribute to neurocognitive impairment (NCI) in people with HIV (PWH). We performed cross-sectional immunological phenotyping, neuroimaging, and neurocognitive testing on virally suppressed PWH in four substance groups: cocaine only users (COC), marijuana only users (MJ), dual users (Dual), and Non-users. Participants completed substance use assessments, multimodal MRI brain scan, neuropsychological testing, and blood and CSF sampling. We employed a two-stage analysis of 305 possible biomarkers of cognitive function associated with substance use. Feature reduction (Kruskal Wallis p-value < 0.05) identified 53 biomarkers associated with substance use (22 MRI and 31 immunological) for model inclusion along with clinical and demographic variables. We employed eXtreme Gradient Boosting (XGBoost) with these markers to predict cognitive function (global T-score). SHapley Additive exPlanations (SHAP) values were calculated to rank features for impact on model output and NCI. Participants were 110 PWH with sustained HIV viral suppression (33 MJ, 12 COC, 22 Dual, and 43 Non-users). The ten highest ranking biomarkers for predicting global T-score were 4 neuroimaging biomarkers including functional connectivity, gray matter volume, and white matter integrity; 5 soluble biomarkers (plasma glycine, alanine, lyso-phosphatidylcholine (lysoPC) aC17.0, hydroxy-sphingomyelin (SM.OH) C14.1, and phosphatidylcholinediacyl (PC aa) C28.1); and 1 clinical variable (nadir CD4 count). The results of our machine learning model suggest that substance use may indirectly contribute to NCI in PWH through both metabolomic and neuropathological mechanisms.
这项研究旨在确定与物质使用相关的神经影像学和免疫学因素，以及导致 HIV 感染者（PWH）神经认知障碍（NCI）的因素。我们对 4 个物质使用组（可卡因使用者 COC、大麻使用者 MJ、双重使用者 Dual、非使用者 Non-users）的病毒抑制 PWH 进行了横断面免疫表型、神经影像学和神经认知测试。参与者完成了物质使用评估、多模态 MRI 脑扫描、神经心理学测试以及血液和 CSF 采样。我们采用了两阶段分析，对与物质使用相关的 305 种可能的认知功能生物标志物进行了分析。特征减少（Kruskal Wallis p 值<0.05）确定了与物质使用相关的 53 种生物标志物（22 种 MRI 和 31 种免疫），以及临床和人口统计学变量，用于模型纳入。我们使用极端梯度提升（XGBoost）和这些标记物来预测认知功能（总体 T 评分）。计算了 SHapley Additive exPlanations（SHAP）值，以对影响模型输出和 NCI 的特征进行排名。参与者是 110 名 HIV 病毒持续抑制的 PWH（33 名 MJ、12 名 COC、22 名 Dual 和 43 名 Non-users）。预测全球 T 评分的十个最高排名生物标志物包括 4 个神经影像学生物标志物，包括功能连接、灰质体积和白质完整性；5 个可溶性生物标志物（血浆甘氨酸、丙氨酸、溶血磷脂酰胆碱（lysoPC）aC17.0、羟基神经鞘氨醇（SM.OH）C14.1 和磷脂酰胆碱二酰（PC aa）C28.1）；以及 1 个临床变量（最低 CD4 计数）。我们的机器学习模型结果表明，物质使用可能通过代谢组学和神经病理学机制间接导致 PWH 的 NCI。
