Guidelines for Selection of Internal Standard-Based Normalization Strategies in Untargeted Lipidomic Profiling by LC-HR-MS/MS

Abstract

Due to variation in instrument response caused by various sources of errors throughout an analytical assay, data normalization plays an indispensable role in untargeted LC-MS profiling, yet limited accepted guidelines on this topic exist. In this work, a systematic comparison of several normalization techniques, mainly focusing on internal standard-based approaches, has been performed to derive some general recommendations. For generation of untargeted lipidomic data, a comprehensive ultra-high performance liquid chromatography (UHPLC)-electrospray ionization (ESI)-quadrupole time of flight (QTOF)-MS/MS method was utilized. To monitor instrument stability and evaluate normalization performance, quality control (QC) samples, prepared from aliquots of all experimental samples, were embedded in the sequence. Stable isotope labeled standards, representing differing lipid classes, were spiked to each sample as internal standards for postacquisition normalization. Various metrics were used to compare distinct normalization strategies, with reduction of variation in QC samples being the critical requirement for acceptance of successful normalization. The comparison of intragroup coefficients of variation (CVs), median absolute deviations (MADs), and variance enables simple selection of the best performance of normalization with improved and coherent results. Furthermore, the importance for normalization in critical data sets, showing only minor effects between groups with high variation and outliers, is pointed out. Apart from normalization, also, influences of used raw data types are demonstrated. In addition, effects of various factors throughout the processing workflow were investigated and optimized. Eventually, implementation of quality control samples, even if not required for normalization, provided a useful basis for assessing data quality. Due to lack of consensus for selecting optimum normalization, suggestions for validating data integrity are given.