A Software Tool for Rapid and Automated Preprocessing of Large-Scale Serum Metabolomic Data by Multisegment Injection-Capillary Electrophoresis-Mass Spectrometry

Abstract

Mass spectrometry (MS)-based metabolomics often rely on separation techniques when analyzing complex biological specimens to improve method resolution, metabolome coverage, quantitative performance, and/or unknown identification. However, low sample throughput and complicated data preprocessing procedures remain major barriers to affordable metabolomic studies that are scalable to large populations. Herein, we introduce PeakMeister as a new software tool in the R statistical environment to enable standardized processing of serum metabolomic data acquired by multisegment injection-capillary electrophoresis-mass spectrometry (MSI-CE-MS), a high-throughput separation platform (<4 min/sample) which takes advantage of a serial injection format of 13 samples within a single analytical run. We performed a rigorous validation of PeakMeister by analyzing 47 cationic metabolites consistently measured in 5,000 serum and 420 quality control samples from the Brazilian National Survey on Child Nutrition (ENANI-2019) comprising a total of 224,983 metabolite peaks acquired in 40 days across three batches over an eight-month period. A migration time index using a panel of 11 internal standards was introduced to correct for large variations in migration times, which allowed for reliable peak annotation, peak integration, and sample position assignment for serum metabolites having two flanking internal standards or a single comigrating stable-isotope internal standard. PeakMeister accelerated data preprocessing times by 30-fold compared to manual processing of MSI-CE-MS data by an experienced analyst using vendor software, while also achieving excellent peak annotation fidelity (median accuracy >99.9%), acceptable intermediate precision (median CV = 16.0%), consistent metabolite peak integration (mean bias = -2.1%), and good mutual agreement when quantifying 16 plasma metabolites from NIST SRM-1950 (mean bias = -1.3%). Reference ranges are also reported for 40 serum metabolites in a national nutritional survey of Brazilian children under 5 years of age from the ENANI-2019 study. MSI-CE-MS in conjunction with PeakMeister allows for rapid and automated processing of large-scale metabolomic studies that tolerate nonlinear migration time shifts without complicated dynamic time warping or effective mobility scale transformations.

Figure 1

PeakMeister is an automated data preprocessing tool for peak picking and sample position assignment in metabolomics when using a serial sample injection format in MSI-CE-MS. A targeted molecular feature list and adjustable parameters are user defined and conveniently stored in a xlsx file as an input. PeakMeister first performs a mass calibration to correct for mass drift and then extracts and smooths the user supplied panel of internal standards and serum metabolites to generate EIEs for peak detection. An algorithm is then applied to the internal standard EIEs to identify the most abundant peaks. This information is then used in conjunction with MTIs for MT prediction of each serum metabolite peak. The predicted MTs are then used to identify sample peaks in the corresponding EIEs, including the correct injection order of samples. After identifying and integrating all peaks, PeakMeister generates EIE plots and csv files with an output of integrated peak areas and MTs for all annotated serum metabolites.

Figure 2

(A) Plot of 47 serum metabolites (green circles) and 11 internal standards (yellow circles) measured consistently by MSI-CE-MS from the ENANI-2019 study. (B) MT variability in MSI-CE-MS is extensive after analysis of 5,433 serum samples (including 420 QCs) with MT shifts exceeding 20 min for certain serum metabolites. This leads to high mean absolute error (MAE) when comparing absolute MT changes over time (run 1 used as reference). (C) However, MT can be accurately predicted with PeakMeister when using a co-migrating stable-isotope internal standard, such as creatinine that had a MAE reduced from 95 s to only 0.30 s. Alternatively, use of MTIs in PeakMeister for serum metabolites having two flanking internal standards when stable-isotope internal standards were lacking, such as isoleucine, also resulted in a major reduction in MAE from 106 s to only 1.5 s. However, late migrating metabolites with only one internal standard, such as serum hypoxanthine, had a more modest decrease in MAE from 146 to 15.3 s. Overall, MSI-CE-MS is prone to greater long-term MT drift when using different capillaries (dotted gray lines) while acquiring data across 3 batches (solid black lines) over an 8 month period. (D) Overall, PeakMeister provided a 36-fold improvement in MT prediction accuracy when using MTIs or RMTs than MTs in MSI-CE-MS with a median MAE of only 2.9 s for all 47 serum metabolites.

Figure 3

Principal component analysis (PCA) comparing intermediate precision of a large-scale serum metabolome analysis by MSI-CE-MS with data preprocessing using (A) PeakMeister and (B) manual data integration following a QC-based batch correction. Both data processing methods had an analogous intermediate precision when integrating 47 serum metabolites from pooled QC samples with a median CV of 16% (n = 420) over 3 batches of analyses with similar biological variation (61% and 52%) in individual serum samples from the ENANI-2019 study (n = 5,000). (C) Overall precision was consistent between the two data preprocessing techniques except for three serum metabolites, including CysGly-CySS, unknown 7, and SDMA, which were low abundance metabolites that had more variable peak integration outcomes by PeakMeister. Control charts of F-Phe (40 μmol/L) added to all serum samples (n = 5,420) demonstrated excellent long-term precision for (D) PeakMeister with a mean CV of 7.8% that was comparable to (E) manual integration with a mean CV of 7.5% after batch correction.