A methodology for normalizing fluid biomarker concentrations across reagent lots

Abstract

AbstractBackgroundLongitudinal studies of cerebrospinal fluid (CSF) and plasma biomarkers typically require the use of different assay reagent lots over time. Measured biomarker concentrations may drift from lot to lot, adding variability that may bias estimates and decrease the statistical power of biomarker analyses. Including lot as a covariate in models may not adequately adjust for lot effects, particularly if cohort characteristics change over time.MethodCSF and plasma samples were collected from participants enrolled in studies of memory and aging. CSF Aβ40, Aβ42, total tau (t‐tau), and phosphotau‐181 (p‐tau181) were measured using a fully automated assay platform (LUMIPULSE, Fujirebio). CSF and plasma NfL were measured using a plate‐based ELISA (UMAN Diagnostics) and Simoa‐HDX assays (Quanterix), respectively. “Bridging” samples were measured using an initial lot of reagents and were subsequently re‐analyzed using a different lot of reagents. Biomarker concentrations were transformed with the natural logarithm and compared via linear regression. If the slope (tested against 1) or intercept (tested against 0) were significantly different, the equation for the initial values as a function of the subsequent values was used to normalize the subsequent values, putting all values on one common scale.ResultMeasured concentrations for all analytes were highly correlated between lots. Slopes and/or intercepts for a linear regression of the log(initial value) as a function of the log(subsequent value) were significant for CSF Aβ40 (intercept: p<0.0001), t‐tau (slope: p<0.0001), p‐tau181 (intercept: p<0.0001), and plasma NfL (intercept: p=0.001), but were not significant for CSF Aβ42 or NfL.ConclusionThere is significant systematic variation in measured biomarker concentrations by reagent lot, even when using fully‐automated assay platforms. Running samples in larger batches with fewer lots is preferable to reduce lot effects. Using biomarker data unadjusted for lot effects or adjusted using simple correction factors may be problematic. Including two or three certified reference standards may not adequately control for these differences; measuring more samples across a wide range of values is likely necessary. Normalization of measured values between reagent lots may decrease variability in longitudinal data collected over several years, increase statistical power and improve the validity of biomarker analyses.