Feature selection of DNA methylation biosignatures for neuropathy with comorbid drug abuse in the setting of HIV infection

点击翻译按钮获取中文摘要

Distal sensory polyneuropathy (DSPN) remains the most common neurological complication in HIV-infected population. Opiates and cocaine are often misused to manage neuropathic pain. These drugs, in turn, increase the risk of DSPN and exacerbate DSPN severity. The interplay of DSPN and substance use disorder (SUD) is associated with morbidity and mortality in HIV-infected individuals. Although neurotoxic effects of HIV-1 gp120 and antiretroviral medications contribute to DSPN through dysregulation of pro-inflammation genes, little is known about the mechanisms of DSPN and DSPN with comorbid SUD. A major barrier to advancing our understanding of DSPN is the inaccessibility of nerve tissues in living individuals and the absence of reliable biomarkers to inform the diagnosis of DSPN in the setting of SUD and HIV infection. Our overarching goal is to discover the DNA methylation signatures for DSPN, SUD, and their comorbidity. We focus on white blood cells (WBCs) because of their role in orchestrating and effecting immune responses and because they are highly accessible tissues. We hypothesize that DSPN emerges in the context of HIV infection as the result of HIV-1 enhanced expression of proinflammatory genes in many cell types, including WBCs. We also hypothesize that DSPN emerges as the result of SUD-enhanced expression of proinflammatory genes in WBCs. Thus, DNA methylation in WBCs is associated with DSPN that is influenced by SUD and contributes to HIV outcomes. To test the hypotheses, we will select methylation features in WBCs for DSPN, SUD, and their comorbidity using a combination of epigenome-wide association study (EWAS) and ensemble-based machine learning approaches. Leveraging two well-established independent cohorts, we will first identify methylation sites in WBCs for DSPN and SUD in four groups (DSPN+/SUD+, DSPN+/SUD-, DSPN-/SUD+, DSPN-/SUD-) in 2,000 HIV- infected samples using a 2-stage EWAS followed by a meta-EWAS. We will then select methylation features using machine learning methods and test the sensitivity and specificity to differentiate DSPN, SUD, and their comorbidity. We will apply our in-house developed bioinformatic package, smartFeatureSelection. The selected features will test associations with immune resilience (i.e. CD4+/CD8+) and HIV outcomes (i.e. frailty, mortality). Finally, we will explore the biological functions of the identified methylation sites in postmortem human brain and blood (N = 80) by RNA-seq and correlate methylation-regulated gene expression between WBCs and neural cells. We expect to discover a set of biologically meaningful methylation features as a marker for HIV- infected DSPN and SUD that can predict resilience and outcomes. Employing a rigorous design and a powerful computational approach, this application proposes the first epigenome-based prediction for DSPN, SUD, and their interaction. The identified features can serve as a biomarker for this complex condition and have potential clinical use. The results will enhance the knowledge of epigenetic mechanisms in blood and in brain for HIV-infected DSPN and SUD.