Classification of alcohol abuse by plasma protein biomarkers

Abstract

Background: Biochemical diagnostics of ethanol intake would improve alcohol abuse treatment and have applications in clinical trial and public safety settings. Self-reporting of alcohol use has clinical utility but lacks the desired reliability. Previously, proposed single-analyte biochemical tests of alcohol intake suffer from low sensitivity and specificity or examine only acute drinking and have therefore seen limited clinical use.

Methods: To address this unmet need, plasma protein biomarker discovery and validation were performed with an alcohol self-administering nonhuman primate model system to develop a diagnostic that accurately classifies subjects into nondrinking, nonabusive drinking, and abusive drinking categories.

Results: A 17-plasma protein panel was determined that correctly classifies abusive drinking with 100% sensitivity and also differentiates any level of drinking from alcohol abstinence with 88% accuracy.

Conclusions: The biomarker panel reflects changes in multiple organ systems and suggests robust changes in the plasma proteome with drinking that might serve as a sensitive and specific diagnostic test. The specific plasma proteins altered with alcohol self-administration might represent indicators of alcohol-induced stress on a variety of organ systems.

Figure 1

(A) Time course of non-human primate alcohol self-administration and plasma sampling points. Two independent samples (A and B) were collected from each condition (Naïve, Induction, and Drinking) for biomarker discovery and classification analysis testing. (B) The cumulative self-administered intake of the monkeys at Induction and Drinking time points. (C) In the two weeks prior to plasma sampling time points, the average daily intake was the same for A and B samples in the Induction or Drinking states. Data are presented with mean lines and dots for each animal and time point (n=10 for each sample).

Figure 2

(A) Using a three protein panel and Support Vector Machine (SVM) classification algorithm, alcohol abuse samples (Drinking) were correctly classified from non-alcohol abusing samples (Naïve and Induction) with 92% accuracy and 100% sensitivity. (B) Using a fourteen protein panel and SVM classification algorithm, alcohol using samples (Induction and Drinking) were correctly classified from non-alcohol using samples (Naïve) with 88% accuracy and 95% sensitivity. (C) Heatmap representation of the seventeen protein biomarker panel and clustering of the sample groups. Mean expression levels, normalized to a mean naïve level of 1, for each drinking state and time point were clustered by condition. Increased abundance, compared to mean Naive, is presented in red and reduced abundance in green. Independent samples from each drinking state clustered together.