A nano ultra-performance liquid chromatography-high resolution mass spectrometry approach for global metabolomic profiling and case study on drug-resistant multiple myeloma

Abstract

Global metabolomics relies on highly reproducible and sensitive detection of a wide range of metabolites in biological samples. Here we report the optimization of metabolome analysis by nanoflow ultraperformance liquid chromatography coupled to high-resolution orbitrap mass spectrometry. Reliable peak features were extracted from the LC-MS runs based on mandatory detection in duplicates and additional noise filtering according to blank injections. The run-to-run variation in peak area showed a median of 14%, and the false discovery rate during a mock comparison was evaluated. To maximize the number of peak features identified, we systematically characterized the effect of sample loading amount, gradient length, and MS resolution. The number of features initially rose and later reached a plateau as a function of sample amount, fitting a hyperbolic curve. Longer gradients improved unique feature detection in part by time-resolving isobaric species. Increasing the MS resolution up to 120000 also aided in the differentiation of near isobaric metabolites, but higher MS resolution reduced the data acquisition rate and conferred no benefits, as predicted from a theoretical simulation of possible metabolites. Moreover, a biphasic LC gradient allowed even distribution of peak features across the elution, yielding markedly more peak features than the linear gradient. Using this robust nUPLC-HRMS platform, we were able to consistently analyze ~6500 metabolite features in a single 60 min gradient from 2 mg of yeast, equivalent to ~50 million cells. We applied this optimized method in a case study of drug (bortezomib) resistant and drug-sensitive multiple myeloma cells. Overall, 18% of metabolite features were matched to KEGG identifiers, enabling pathway enrichment analysis. Principal component analysis and heat map data correctly clustered isogenic phenotypes, highlighting the potential for hundreds of small molecule biomarkers of cancer drug resistance.

Figures

Figure 1.

(A) nUPLC column set-up. (B) Base-peak chromatograms demonstrating interday reproducibility (across 2-weeks). (C) Scatter plot and (D) histogram showing the reproducibility of replicate analyses (n = 4) of yeast metabolite extracts. (E) Scatter plot and (F) histogram estimating the false discovery rate of pairwise metabolomics discovery experiments in a null experiment.

Figure 2.

(A) Sample workflow. (B) The level of protein in metabolite extracts was assessed by BCA assay in independent triplicates. For the samples extracted by 80% acetonitrile, the minimal detection amount of the method was used to represent an upper limit of protein in the sample. (C) Yeast metabolite extracts were analyzed in duplicate to determine how lysis conditions impacted feature detection.

Figure 3.

(A) Hyperbolic fit (R2 = 0.99) showing the saturation of detectable metabolite features with respect to amount of sample loaded with semi-log inset plot. (B) Selected 25 features showing the linear response of feature peak area with respect to amount of sample loaded. (C) Extracted ion chromatogram of the 325.246 m/z ion demonstrating peak symmetry at low (6e5 cells) and high (6e7 cells) levels of sample loading; scientific notation indicates peak area as detected using the Genesis algorithm.

Figure 4.

(A) Distribution of feature mass-to-charge ratios in yeast metabolite extracts. (B) Overlay of MS1 monoisotopic peak (403.279 m/z) at various MS resolutions. (C) Theoretical impact of MS resolution on feature discrimination. (D) Empirical impact of MS resolution on feature detection vs. predicted.

Figure 5.

(A) Base-peak chromatograms showing the impact of gradient length on chromatographic resolution. (B) Extracted ion currents (EICs) of 440.273 m/z under different gradient lengths showing increasing temporal resolution of isobaric metabolite features with longer gradient methods. (C) Duplicate analyses exploring feature detection as a function of gradient length and sample loading.

Figure 6.

(A) Gradient composition profiles for the linear (blue) and biphasic (red) methods. (B) Peak density plot through time for the linear and biphasic gradients. (C) Impact of gradient design on metabolite feature detection in yeast extracts.

Figure 7.

(A) Workflow for pairwise analysis of drug-sensitive and drug-resistant multiple myeloma cell lines. (B) Metabolite features detected and identified in myeloma cells. (C) Distribution of metabolite feature ratios between drug resistant and sensitive myeloma cells. Up and down-regulated metabolites are defined as being in the lower or top 5% as indicated by the dashed red lines. (D) Heat map of unfiltered metabolite features showing distinct clustering of BS and BR sample groups. (E) Multi-dimensional scaling plot showing distinct grouping between BR (red) and BS (blue) samples groups based on metabolite features.