Interlaboratory comparison of standardised metabolomics and lipidomics analyses in human and rodent blood using the MxP® Quant 500 kit

Abstract

Metabolomics and lipidomics are pivotal in understanding phenotypic variations beyond genomics. However, quantification and comparability of mass spectrometry (MS)-derived data are challenging. Standardised assays can enhance data comparability, enabling applications in multi-center epidemiological and clinical studies. Here we evaluated the performance and reproducibility of the MxP^® Quant 500 kit across 14 laboratories. The kit allows quantification of 634 different metabolites from 26 compound classes using triple quadrupole MS. Each laboratory analysed twelve samples, including human plasma and serum, lipaemic plasma, NIST SRM 1950, and mouse and rat plasma, in triplicates. 505 out of the 634 metabolites were measurable above the limit of detection in all laboratories, while eight metabolites were undetectable in our study. Out of the 505 metabolites, 412 were observed in both human and rodent samples. Overall, the kit exhibited high reproducibility with a median coefficient of variation (CV) of 14.3 %. CVs in NIST SRM 1950 reference plasma were below 25 % and 10 % for 494 and 138 metabolites, respectively. To facilitate further inspection of reproducibility for any compound, we provide detailed results from the in-depth evaluation of reproducibility across concentration ranges using Deming regression. Interlaboratory reproducibility was similar across sample types, with some species-, matrix-, and phenotype-specific differences due to variations in concentration ranges. Comparisons with previous studies on the performance of MS-based kits (including the AbsoluteIDQ p180 and the Lipidyzer) revealed good concordance of reproducibility results and measured absolute concentrations in NIST SRM 1950 for most metabolites, making the MxP^® Quant 500 kit a relevant tool to apply metabolomics and lipidomics in multi-center studies.

Figure 1.

Graphical overview of participating laboratories, project samples, and plate design. Each participating laboratory is denoted by a unique letter and color code used throughout this manuscript. Analogously, project sample types are denoted by different symbols.

Figure 2:

Metabolites measured in project samples across participating laboratories. The top panel provides an overview of the metabolite coverage of the MxP^® Quant 500 kit. The central panel shows how many of these metabolites were detected in at least one measurement above LOD for the project samples within this study. The horizontal bars depict metabolite classes and numbers of metabolites that were detected in all 14 laboratories, in 10 to 13, in 4–9, or in 1–3 laboratories, respectively. Overall, 505 metabolites were measured in all 14 laboratories. The pie chart displays the proportion of these metabolites that were measured in all project samples (i.e., 0 % missingness), with less than 5 %, 25 %, and 50 %, or at least 50 % missingness. In the bottom panel, Venn diagrams depict comparisons of metabolite coverage between the different sample types for the 505 metabolites detectable in all laboratories.

Figure 3.

Laboratory-specific missingness. Top panel: Laboratories showed differences in the number of missing measurements. Bottom panel: Thereby, higher laboratory-specific LODs correlated with higher overall missingness. Color coding of laboratories as in Figure 1.

Figure 4:

Normalisation of metabolites using QC2. Top panel: Mean absolute percentage error (MAPE) distribution of measured QC2 concentrations of all metabolites detected in QC2 samples compared to their reference target values provided by the kit manufacturer is shown separated by the measurement method used: (from left to right) after LC separation: (i) metabolites quantified with multi-point calibration curves (7-point calibration), (ii) metabolites quantified using 1-point calibration detected in positive or (iii) negative mode, and (iv) metabolites detected by flow injection analysis (FIA). (Metabolite abbreviations: Asp – asparagine, PEA - phenylethylamine, HCys – homocysteine, AA – arachidonic acid, AbsAcid – abscisic acid.) Central panel: PCA score plot for PC1 and PC2 before (left) and after (right) QC2 normalisation (center). Bottom panel: Average interlaboratory CVs of all sample types for 561 metabolites before and after QC2 normalisation.

Figure 5:

Pairwise laboratory comparisons of measurements using Deming regression. Top panel: Deming regression was used to assess pairwise systematic constant and proportional differences of measurements. Results for each metabolite are provided as matrices indicating whether concentrations between two laboratories were not significantly different (“N”), or showed significant constant (“C”), proportional (“P”) or constant and proportional differences (“CP”). In addition, we also provide matrices containing and color-coding the observed relative differences for each metabolite in the minimum, median, and maximum ranges. Center panel: Circular heatmaps visualising the log-median relative differences of each laboratory for each metabolite for the median, minimum, and maximum concentration ranges using a color gradient from saturated green (LMRD ≤ 2) to saturated red (LMRD > 8); white color indicates missing information due to lack of measurements for the metabolite in this laboratory. Bottom panel: Inlets of the circular heatmaps show selected examples of observed patterns of reproducibility.

Figure 6:

CV values in different sample types. The top panel shows all CVs as calculated for each metabolite and sample type. Vertical dotted lines indicate the overlay of CV values for specific metabolites that have been selected as examples in the lower right panel. Sample NIST SRM 1950 is highlighted in orange. The bottom left panel illustrates the number of metabolites with CVs below 25 % (color) or 10 % (grey) within each sample type. The bottom right panel shows the distribution of CVs across sample types for acetylcarnitine (C2), HexCer(d16:1/22:0), PC ae C38:1, and TG(18:0_30:1).

Figure 7:

Comparison of metabolite concentrations in the NIST SRM 1950 sample between studies. Top panel: Mean absolute percentage error (% MAPE) of data from this project versus those (i) certified by NIST (or provided as reference) (28), (ii) measured by Ghorasaini et al. (18), and (iii) Liu et al. (27). Color coding for metabolites represents metabolite classes as in Figure 2. Bottom panel: A correlation plot of amino acid values measured in this study versus those reported for the SRM 1950 probe as certified or reference by NIST. For each amino acid, individual concentrations measured in the Laboratories A-N are shown. Color coding of the laboratories as in Figure 1. Abbreviations: EPA - eicosapentaenoic acid (FA(20:5)), DHA - docosahexaenoid acid (FA(22:6)), Cit - citrulline, C2 – acetylcarnitine, H1 – hexoses, CA – cholic acid.