Transcriptomic score: A biomarker index of Alzheimer’s risk and resilience

Abstract

AbstractBackgroundOne definition of resilience is the absence of disease despite high levels of risk. And in most cases, risk itself is best quantified by a multivariate profile rather than a single risk factor. Thus, multivariate risk profiles may be most appropriate for defining risk and resilience. RNA expression may serve as a risk biomarker, but no method has been proposed to build a whole‐transcriptome risk index of Alzheimer’s disease (AD) that could identify resilient individuals at the highest transcriptomic risk. Here, we derived a novel method for transcriptome‐wide gene expression data called “transcriptomic risk score” (TRS). We also employed the algorithm to build a “transcriptomic resilience score” (TReS) that offsets that risk.MethodAnalogous to a polygenic risk score, TRS is a weighted rank‐sum statistic that represents a subject’s relative expression level across the transcriptome, where the standardized beta coefficients are used as weights. We leveraged the TRS approach on peripheral blood gene expression profiles of 364 normal controls and 188 AD cases from ADNI and AddNeuroMed batch 1. Using TRSas a means of risk quantification, we then identified resilient normal controls as those with the highest percentiles of risk. High‐risk AD cases with matched TRSs range were retained for comparison. Expression levels of individual transcripts not included in the best‐fit TRS were contrasted between the high‐risk normal controls and high‐risk AD cases using mixed‐effect linear models. A TReS was derived using the residual transcriptomic variation that buffers against the transcriptomic risk. The association between TReS and case status was evaluated using a logistic regression model.ResultThe derived TReS differentiated resilient subjects (high‐risk normal controls) and high‐risk AD cases in independent replication samples (216 normal controls and 223 AD cases), with significant differentiation seen at all evaluated p‐value bins and a maximum of 3% variance accounted for at a p‐value threshold of 0.1.ConclusionThis analysis indicated that TRS and TReS analysis of blood has the potential of facilitating the understanding of phenotypic variability and improving clinically useful biomarker profiles for AD.