Evaluation and correction of injection order effects in LC-MS/MS based targeted metabolomics

Abstract

For large-scale and long-term metabolomics studies that involve a large batch or multiple batches of analyses, batch effects cause nonbiological systematic biases that may lead to false positive or false negative findings. Quantitative monitoring and correction of batch effects is critical to the development of reproducible and robust metabolomics platforms either for untargeted or targeted analyses. To achieve sufficient retention and separation of a broad range of metabolites with diverse chemical structures and physicochemical properties, LC-MS/MS based targeted metabolomics often involves 3 complemented chromatographic separation methods, including reversed-phase liquid chromatography (RP-LC), hydrophilic interaction liquid chromatography (HILIC), and ion-pair liquid chromatography (IP-LC). The purpose of this study is to quantitatively evaluate intra-batch variations or injection order effects of the RP-LC, HILIC, and IP-LC methods for targeted metabolomics analyses, and develop strategies to minimize intra-batch variations and correct injection order effects for problematic metabolites. Both RP-LC and HILIC methods exhibit robust intra-batch reproducibility in 0.2 µM standard mix QC, with ~96 % of the measured metabolites showing acceptable intra-batch variations (<20 %); whereas, the intra-batch reproducibility for some metabolites in cell matrix QC may be compromised due to stability issue, suboptimal chromatographic retention, and/or matrix effects causing ionization suppression and/or retention instability. The IP-LC method exhibits significant injection order effects, which could be effectively corrected by the developed exponential models of signal drift trends as a function of injection order for individual targeted metabolites.

1. Introduction

Metabolomics, together with genomics, transcriptomics, and proteomics, are invaluable tools for better understanding system biology [1–4]. Metabolomics focuses on comprehensive, quantitative analyses of small molecule metabolites in cells, tissues or organisms [3–5]. The metabolites have diverse chemical structures and physicochemical properties, present at a wide concentration range, and change more rapidly than any genetic materials [4]. Several analytical technologies including nuclear magnetic resonance (NMR), gas chromatography coupled with mass spectrometry (GC–MS), liquid chromatography coupled with mass spectrometry (LC-MS) are commonly used for metabolomic analyses [3,6,7]. In particular, given the broad metabolite coverage, large dynamic range, high sensitivity and specificity, high-performance liquid chromatography coupled with high-resolution mass spectrometry (LC-HRMS) or tandem mass spectrometry (LC-MS/MS) has been increasingly applied for metabolomics, either untargeted or targeted analyses [6–9].

LC-MS based metabolomics, however, has its technical limitation, notably batch effects that LC-MS responses may vary between batches or within a batch of analyses [10–12]. Inter-batch variability can occur due to variations in laboratory conditions, instrument conditions, reagent lots, or LC column lots [13–15]. Intra-batch variability or injection order effects are often caused by instrumental fluctuations that could be related to chromatographic variation (e.g., retention time shift and peak shape changes) and/or mass spectrometry signal drift [15,16]. Injection order effects are frequently observed in LC-HRMS data due to the gradual contamination of the LC column and/or MS detector in continuous analyses of a large number of samples in an analytical run without instrument maintenance (which is defined as a batch) [17,18]. Analyses of samples in batches with intermittent cleaning and conditioning of the instrument between batches could mitigate the contamination problem to a degree. However, this may introduce other technical variations (such as difference in the instrument operating conditions) leading to inter-batch variability [18]. For large-scale and long-term metabolomics studies that involve a large batch or multiple batches of analyses, batch effects cause nonbiological systematic biases that may lead to false positive or false negative findings.

Quantitative monitoring and correction of batch effects is critical to the development of reproducible and robust metabolomics platforms either for untargeted or targeted analyses. Most studies on batch effects have been focused on LC-HRMS based untargeted metabolomics [11,17]; whereas, the issue is largely overlooked in LC-MS/MS based targeted metabolomics analyses. In contrast to the untargeted metabolomics that involves a large-scale detection of all detectable metabolite features including chemical unknowns, targeted metabolomics focuses on analysis of a predefined group of metabolites with known chemical structures [7,8,19]. Since reference standards are available in targeted metabolomics, calibration standard curves for individual determined metabolites are prepared in each batch of analyses thus efficiently correcting for inter-batch variability. However, intra-batch variability or injection order effects remain the main cause of non-biological systematic bias in LC-MS/MS based targeted metabolomics.

To achieve sufficient retention and separation of a broad range of metabolites with diverse chemical structures and physicochemical properties, LC-MS/MS based targeted metabolomics often involves 3 complemented chromatographic separation methods, including reversed-phase liquid chromatography (RP-LC), hydrophilic interaction liquid chromatography (HILIC), and ion-pair liquid chromatography (IP-LC) [8,19]. Despite their wide applications, no studies have been published, to our best knowledge, to systematically evaluate injection order effects of these methods in targeted metabolomics. In the present study, we quantitatively evaluated injection order effects of the RP-LC, HILIC, and IP-LC methods for targeted analyses of ~250 chemically diverse endogenous metabolites. In addition, we proposed strategies to minimize intra-batch variations and correct injection order effects for problematic metabolites, which could be implemented as standard operation procedures for targeted metabolomics studies.

2. Materials and methods

2.1. Chemicals and reagents

Metabolite reference standards were purchased from Sigma Aldrich (St. Louis, MO) and Fisher Scientific (Minneapolis, MN) with the minimal 95 % purity or the highest available purity. Hexafluora-2-isopropanol (HFLP) and triethylamine (TEA) were purchased from CovaChem, LLC (Loves Park, IL) and Alfa Aesar (Ward Hill, MA), respectively. Formic acid, ammonium acetate, acetonitrile, methanol, and isopropanol were LC-MS grade from Fisher Scientific (Minneapolis, MN). Water was filtered and deionized with a US Filter PureLab Plus UV/UF system (Siemens, Detroit, MI, USA) and used throughout in all aqueous solutions.

2.2. Stock solution and quality control (QC) samples

Standard stock solution for each individual metabolite was prepared, depending on its solubility, in DMSO, 10–50 % acetonitrile, acetonitrile, 5–10 % methanol, or methanol at the final concentration of 10 mM, and stored in a brown glass vial at − 80 °C. Standard mix of targeted metabolites at the concentration of 100 µM was prepared in acetonitrile/water (50/50, v/v), aliquoted, and stored at −80 °C. Standard mix quality control samples at the individual metabolite concentration of 0.2 µM were prepared freshly on each batch of analysis by serial dilutions of the standard mix stock in water (for RP-LC and IP-LC method) or acetonitrile/water (90/10, v/v) (for HILIC method).

To prepare cell matrix QC samples, human breast cancer cell lines, MDA-MB-231 and MCF-7 (obtained from Division of Cancer Treatment and Diagnosis of the National Cancer Institute, Frederick, MD) were cultured in RPMI-1640 medium, supplemented with 10 % fetal bovine serum (FBS), 100 U/mL penicillin, and 100 μg/mL streptomycin. Cells were cultured in 10-cm dish until ~80 % confluence, and cell pellet was collected and snap-frozen in liquid nitrogen. Metabolites in frozen cell pellet were extracted twice with 80 % methanol (pre-cooled at −80 °C) as described previously with modifications [19]. The supernatants from both extractions were combined, aliquoted (1.2 mL each) in Eppendorf tubes, and dried in a CentriVap® Refrigerated Centrifugal Concentrator (Kansas City, MO) at 6 °C. The dried extract aliquots were stored at −80 °C until analysis. Prior to the LC-MS/MS analysis, one aliquot of the dried cell extract was reconstituted in 50 µL water (for RP-LC and IP-LC methods) or acetonitrile/water (90/10, v/v) (for HILIC method), followed by vortex-mixing and centrifugation. The supernatant was collected and used as the original cell matrix QC.

Some metabolites in the original cell matrix QC may be unquantifiable (i.e., below the lower limit of quantitation). To ensure all targeted metabolites were quantifiable, spiked cell matrix QC was prepared by spiking 5 µM standard mix (for RP-LC and HILIC methods) or 10 µM standard mix (for IP-LC method) into the original cell matrix QC sample.

2.3. Instrumentation

All LC-MS/MS analyses were performed on an AB SCIEX (Foster City, CA) QTRAP 6500 system, which consists of a SHIMADZU (Kyoto, Japan) Nexera ultra high-performance liquid chromatography system coupled with a hybrid triple quadrupole and linear ion trap mass spectrometer. Analyst® 1.6 software was used for system control and data acquisition, and MultiQuant 3.0 software was used for data processing and quantitation.

Routine instrument maintenance between batches was performed by purging the pump heads, solvent lines, and autosampler with methanol/water (1:1, v/v), acetonitrile/methanol/isopropanol/water (1:1:1:1, v/v), and corresponding mobile phases for 10 min each. HPLC column was washed with water and acetonitrile at a flow rate of 0.25 mL/min (RP and HILIC methods) or 0.5 mL/min (IP method) for 90 min. The MS source (sample cone and baffle components) was cleaned by sequentially rinsing with 0.1 % formic acid in water, methanol/water (1:1, v/v), and methanol. During the sample analyses, autosampler was washed following each injection with two alternative purging methods of external purging with methanol/water (1:1, v/v) and acetonitrile/methanol/isopropanol/water (1:1:1:1, v/v), as well as internal purging with corresponding mobile phase.

2.4. LC-MS/MS based targeted metabolomics

A targeted metabolomics platform, which involved 3 different chromatographic separation methods (including RP-LC, HILIC, and IP-LC) coupled with tandem mass spectrometry, was developed to quantitatively determine a total of ~250 endogenous metabolites with diverse physicochemical properties, which are involved in major human metabolic pathways including glycolysis, pentose phosphate pathway, tricyclic acid (TCA) cycle, amino acid metabolism, nucleotide and nucleoside metabolism, and fatty acid metabolism. Supplementary Tables S1, S2, and S3 summarize the list of metabolites in each method, as well as the MS transitions, retention time, linear dynamic ranges, and pathways involved for individual metabolites.

Chromatographic separation of small organic acids, amino acids metabolites, nucleobases, nucleosides, monophosphate nucleosides, sugar derivatives, and fatty acid derivatives was achieved based on RP-LC on a Synergi Polar-RP column (80 Å, 2.0 mm × 150 mm, 4 µm). The gradient elution consisted of mobile phase A (0.03 % formic acid in water) and mobile phase B (0.03 % formic acid in acetonitrile), at a flow rate of 0.25 mL/min and column temperature of 40 °C. The gradient program was as follows: 0–0.3 min, 0 % B; 0.3–25 min, 0–95 % B; 25–25.1 min, 95–0 %; 25.1–30 min, 0 % B. Separation of highly polar metabolites that were not well retained or did not show good dynamic ranges on the RP-LC (e.g., some amino acids and acyl carnitines) was achieved based on HILIC mechanism on an Atlantis HILIC Silica column (2.1 mm × 150 mm, 3 µm). The gradient elution consisted of mobile phase A (10 mM ammonium formate, pH 3.0) and mobile phase B (0.1 % formic acid in acetonitrile), at a flow rate of 0.25 mL/min and column temperature of 40 °C. The gradient program was as follows: 0–0.5 min, 95 % B; 0.5–10.5 min, 95–40 % B; 10.5–15 min, 40 % B; 15–15.5 min, 40–95 % B;15.5–16 min, 95–40 % B; 16–16.5 min, 40–95 % B; 16.5–17 min, 95–40 % B;17–17.5 min, 40–95 % B; 17.5–18 min, 95–40 % B; 18–18.5 min, 40–95 % B; 18.5–24 min, 95 % B. Of note, “W-wash” was incorporated into the mobile phase gradient program (16–18 min) to effectively diminish carryover effects of some metabolites (e.g., histidine, agmatine, and methionine sulfoxide) in the HILIC method. Instead of a single gradient slope, “W-wash” involves three repeats of the gradient change from 40 % to 95 % and back to 40 % mobile phase B (acetonitrile) thereby forming a W-shape gradient. Chromatographic separation of coenzyme A and derivatives, mono-, di-, or tri-phosphate nucleotides, and sugar phosphates was achieved based on IP-LC mechanism on an Atlantis T3 column (2.1 mm × 100 mm, 3.0 µm). The gradient elution consisted of mobile phase A (100 mM hexofluoro-2-propanol and 8.6 mM triethylamine in water, final pH, 8.3) and mobile phase B (10 % acetonitrile in mobile phase A), at a flow rate of 0.5 mL/min and column temperature of 40 °C. The gradient program was as follows: 0–0.3 min, 0 % B; 0.3–5 min, 0–10 % B; 5.0–20 min, 10–100 % B; 20–21 min, 100 % B; 21–22.1 min, 100–0 % B; 22–30 min, 0 % B.

Column eluents were monitored under both positive and negative ionization modes using the multiple reaction monitoring (MRM) on the QTRAP 6500 mass spectrometer. Mass spectrometric parameters (including ionization polarity, product ion, collision energy, declustering potential, and cell exit potential) were optimized to obtain the most sensitive and specific mass transitions for individual metabolites by direct infusion of the standard solutions into the ion source with a syringe pump. Other optimized mass spectrometric parameters were as follows: ion Spray (IS) potential was 5500 V for positive ionization mode and 4500 V for negative ionization mode; nebulizer gas (GS1) and bath gas (GS2) were 50 psi; curtain gas (CUR) was 30 psi; collision gas (CAD) was set medium level; and source temperature (TEM) was 475 °C. The dwell time for each MRM transition was optimized in all methods, which ranged from 3 to 50 ms for individual targeted metabolites in the RP-LC and HILIC methods, and 10 to 50 ms in the IP-LC method.

2.5. Evaluation of injection order effects

As the standard operation procedure for metabolomic sample analyses in the KCI Pharmacology and Metabolomics Core Facility, the number of injections in one continuous batch analysis is limited within 100 (which takes maximum 2–3 days continuous instrument run); for studies with more than 100 samples, the samples are analyzed in batches with intermittent routine instrument cleaning and maintenance (as described on Page 7). This procedure provides the confidence on the stability of metabolites in post-extracted samples in the autosampler and optimal instrument condition for each batch analysis. As consistent with our standard operation procedure in real metabolomics sample analysis, the intra-batch variability (injection order effects) was evaluated by analyzing 5 types of QCs (including 0.2 µM standard mix, original MCF-7 cell matrix, original MDA-MB-231 cell matrix, spiked MCF-7 cell matrix, and spiked MDA-MB-231 cell matrix) using the RP-LC, HILIC, and IP-LC methods in a continuous batch analysis (without instrument maintenance) over 2–3 days. The injection sequences of QC and blank samples in the RP-LC, HILIC, and IP-LC methods are presented in Supplementary Table S4.

2.6. Data analysis

The signal intensities of individual targeted metabolites in QC samples were acquired as peak areas without any data transformation, normalization, or scaling. Principal component analysis (PCA) was performed using MetaboAnalyst (https://www.metaboanalyst.ca/) [20], and used for visual inspection of clustering of QC samples, whereby highly reproducible data would cluster together in a PCA scores plot. The intra-batch variation for each targeted metabolite was assessed as the mean relative standard deviation (RSD) of peak areas for the same type of QC samples that were analyzed in a continuous run. RSD is calculated as: $SD=\frac{\sigma }{Mean}$, where mean is the mean peak area and $\sigma$ is the standard deviation of peak areas of individual metabolites from the same type of QC samples.

The trend of signal drift was expressed as the percentage of peak area of each individual metabolite relative to the peak area in the first injection of the same type of QCs as the function of injection order. Exponential, logarithmic, and polynomial functions were tested, and the exponential function (Eq. (1)) was found to best fit the signal drift trends of the targeted metabolites.

$$Y_i=\alpha \cdot e^{\beta X_i}+\gamma$$ (1)

where ${\mathrm{X}}_{\mathrm{i}}$ is the injection order $\mathrm{i}$; ${\mathrm{Y}}_{\mathrm{i}}$ is the percentage of peak area of a particular metabolite in injection $\mathrm{i}$ relative to its peak area in the first injection; $\mathrm{\alpha }$, $\mathrm{\beta }$, and $\mathrm{\gamma }$ are the model parameters describing the signal drift trend for this metabolite. The developed exponential model (Eq. (1)) provided a correction factor for correcting injection order effects of the metabolite in an experimental sample, as demonstrated by Eq. (2).

$$C_i^{\prime }=\frac{C_i}{\alpha \cdot e^{\beta X_i}+\gamma }$$ (2)

Where ${\mathrm{C}}_{\mathrm{i}}$ and $C_i^{\prime }$ is the observed (measured) and corrected concentration of a metabolite in a sample with the injection order $\mathrm{i}$.

3. Results and discussion

3.1. Overview of intra-batch variations of RP-LC, IP-LC, and HILIC methods in targeted metabolomics

A commonly implemented approach for monitoring intra-batch variations (injection order effects) in untargeted metabolomics is the inclusion of quality control (QC) samples in the analysis [11]. The QC samples can be either a mixture of known laboratory grade standard references, a pooled sample from the experimental samples, or biologically similar samples (if experimental samples are not available for pooled samples). A set of identical QC samples is interspersed with the experimental samples in the same batch of data acquisition, which provides a reference for not only tracking instrumental fluctuations but also post-acquisition correction for variations if appropriate mathematical models are applied [11,18]. In the present study, we evaluated the intra-batch variations of the RP-LC, IP-LC, and HILIC methods by interspersing 5 types of QC samples (including 0.2 µM standard mix in water, original MCF-7 cell matrix, original MDA-MB-231 cell matrix, spiked MCF-7 cell matrix, and spiked MDA-MB-231 cell matrix) in a continuous analytical run over 2–3 days of analyses (Supplementary Table S4). The signal intensities of individual targeted metabolites in QC samples were acquired as peak areas.

Principal component analysis (PCA) was used for visual inspection of clustering of QC data, whereby highly reproducible data would cluster together in a PCA scores plot. The 3D PCA scores of plots (PC1 vs PC2 vs PC3) showed that each set of QC data was closely clustered together in the RP-LC and HILIC methods; whereas, there was an apparent signal drift pattern (indicating significant injection order effects) in the IP-LC method (Fig. 1). Intra-batch variations in each method were further quantitatively assessed by the mean relative standard deviation (RSD) of individual metabolite peak areas from each set of QC samples. Based on the United States Food and Drug Administration Guidance on Bio-analytical Method Validation (https://www.fda.gov/regulatory-information/search-fda-guidance-documents/bioanalytical-method-validation-guidance-industry), a commonly accepted criterion for the reproducibility of single-analyte tests is RSD ≤ 15 %, except for the lower limit of quantitation (LLOQ) at which RSD ≤ 20 % is acceptable. For analysis of a large number of metabolites, an acceptance criterion of RSD ≤ 20 % is generally accepted [10], and thus was used in the present study. The intra-batch RSD for individual metabolites in each method are presented in Supplementary Table S1, S2, and S3. The distributions of RSD for 5 sets of QC samples analyzed by the 3 methods are summarized in Fig. 2. Collectively, the RP-LC and HILIC methods exhibited excellent intra-batch reproducibility, while the IP-LC exhibited significant injection order effects.

Fig. 1.

Principal component analysis based on the peak areas of targeted metabolites in 5 types of QC samples analyzed by (A) RP-LC method, (B) HILIC method, and (C) IP-LC method. The 3D scores plots of the first, second, and third principal component are shown. Each set of QC data is closely clustered in the RP-LC and HILIC methods, indicating good intra-batch reproducibility; whereas, there is an apparent signal drift pattern (indicating injection order effects) in the IP-LC method.

Fig. 2.

The distributions of intra-batch variations of 5 types of quality control (QC) samples analyzed by (A) RP-LC method, (B) HILIC method, and (C) IP-LC method. Intra-batch variation is assessed as the mean relative standard deviation (RSD) of peak areas of individual targeted metabolites in the QC samples that were analyzed in a continuous analytical batch. BLQ, below the lower limit of quantitation.

3.2. Intra-batch reproducibility of the RP-LC method

RP-LC is suitable for the analysis of non-polar compounds because the retention mainly relies on hydrophobic interactions between analytes and stationary phases. In our targeted metabolomics, 122 metabolites such as small organic acids (e.g., succinic acid, fumaric acid, kynurenic acid), amino acids (e.g., isoleucine, leucine, tyrosine, glutamic acid), and nucleoside substitutes (e.g., guanosine, adenosine, deoxyinosine) etc. were quantitated by the RP-LC method (Supplementary Table S1). In the 0.2 µM standard mix solution QC and spiked cell matrix QC samples, ~ 96 % of 122 quantifiable metabolites showed RSD ≤ 20 %, while only ~4 % showed RSD > 20 % (Fig. 2A). In the original (unspiked) MCF-7 and MDA-MB-231 cell matrix QC samples, about 25 % of targeted metabolites were unquantifiable (below the LLOQ); ~ 6 % showed RSD > 20 %; and the rest showed RSD ≤ 20 % (Fig. 2A). In the spiked MCF-7 and MDA-MB-231 cell matrix QCs, the overall intra-batch reproducibility was similar to that in the standard mix QC (Fig. 2A). Notably, the metabolites with RSD > 20 % in the 0.2 µM standard mix QC (i.e., glutathione, l-cysteine, nicotinic acid adenine dinucleotide, spermidine, and spermine) had low sensitivity, poor peak shape, and/or stability issue. The metabolites with RSD > 20 % in the original MCF-7 and MDA-MB-231 cell matrix QCs mostly appeared to be “low-abundant” or less sensitive metabolites with peak areas of 1000–30,000, for which matrix effects may be significant to interfere chromatographic separation and/or ionization efficiency. The metabolites with RSD > 20 % in the spiked MCF-7 and MDA-MB-231 cell matrix QCs (i.e., adenosine, deoxyadenosine, cytidine, inosine, homocysteine, and spermine) exhibited apparent matrix interference and/or stability issue. Collectively, the RP-LC method exhibits robust reproducible chromatographic performance for targeted metabolomics in general. The reproducibility may be compromised for some metabolites with stability issue (e.g., l-cysteine and nicotinic acid adenine dinucleotide), suboptimal chromatographic retention (e.g., spermidine, and spermine), and/or matrix interference/effect (e.g., adenosine, deoxyadenosine, and cytidine). If these problematic metabolites are of particular interest, alternative sample preparation and/or chromatographic methods would be needed for further validation.

3.3. Intra-batch reproducibility of the HILIC method

HILIC method is commonly used for the analysis of hydrophilic and polar metabolites that fail to retain on standard reversed-phase stationary phases [21,22]. While multimodal retention mechanisms are involved, the HILIC mode of separation mainly relies on liquid–liquid partitioning of the analyte between mobile phase and water-enriched solvent layer partially immobilized onto the surface of the stationary phase [21,22]. Given the fact that the more polar compounds have a stronger interaction with the stationary aqueous layer than the less polar compounds, the HILIC separation is mainly dependent on the difference in polarity of the compound and degree of solvation [21]. By using a more volatile mobile phase, the HILIC method coupled with electrospray ionization mass spectrometry often offers better sensitivity for analysis of polar compounds than RP-LC method. However, the reproducibility of HILIC may be less optimal than RP-LC because subtle differences in temperature, pH, or solvent additive concentrations could alter the peak shape and retention time in HILIC analysis [23].

The HILIC method was used for quantitation of 71 highly polar metabolites (which were not well retained, showed poorer sensitivity or narrow dynamic range in the RP-LC), including bile acid metabolites, acylcarnitines, nucleobase, and small organic acids etc (Supplementary Table S2). For the 0.2 µM standard mix solution QC, the intra-batch reproducibility of the HILIC method was equally excellent as that of the RP-LC method, whereby ~96 % of 71 quantifiable metabolites showed RSD ≤ 20 %, while only ~4 % showed RSD of 20–50 % (Fig. 2B). For those with RSD of 20–50 % (including glutamine, glycine, and cystine), the ionization efficiency appeared to be low (with peak area < 30,000 in the 0.2 µM standard mix solution). In the original (unspiked) MCF-7 and MDA-MB-231 cell matrix QC samples, about 50 % of 71 metabolites were below the LLOQ, which could be attributable to intrinsic low abundances of these metabolites in cells or significant matrix effects to suppress ionization signals (Fig. 2B). In the spiked MCF-7 and MDA-MB-231 cell matrix QCs, ~ 89 % of 71 quantifiable metabolites showed RSD ≤ 20 % and ~11 % had RSD > 20 % (Fig. 2B). Notably, those metabolites with RSD > 20 % showed either low ionization efficiency (e.g., aconitic acid, 2-oxo-4-methylthiobutanoic acid, citicoline, taurodeoxycholic acid, and taurochenodesoxycholic acid) or retention instability (e.g., S-adenosylhomocysteine, cystathionine, maleic acid, malonyl-l-carnitine). Taken together, the HILIC method exhibited reproducible chromatographic performance for analysis of the 0.2 µM standard mix solution QC; whereas, the intra-batch reproducibility for some metabolites in biological samples may be compromised due to significant matrix effects resulting in ionization suppression or retention instability.

While the inclusion of stable isotope-labeled internal standards can efficiently correct for matrix effects, it is not feasible to have isotope-labeled internal standards for individual metabolites in metabolomics analysis. Alternatively, a practical approach to diminish matrix effects is to dilute biological samples [24–26]. 20-fold dilution of most biological samples was found to effectively diminish matrix effects in our routine analyses and published studies by us and others [24–26]. Thus, as the standard operation procedure in our metabolomics analyses, both original and 20-fold diluted post-extracted samples are analyzed in a same analytical batch, and the metabolite concentrations determined from the diluted samples are reported if the concentrations are within the linear calibration curve ranges.

3.4. Injection order effects in the IP-LC method

IP-LC is a variant of RP-LC that achieves the retention and separation of charged or ionic compounds on traditional reversed-phase hydrophobic stationary phases by the addition of ion pair reagents into the mobile phase [27]. Through formation of ion pairs with opposite charged ion pair reagents, the retention of charged analytes can be significantly improved on reversed-phase stationary phases [27]. However, IP-LC has its own disadvantages when coupling with mass spectrometry detection, most notably system contamination and ion-suppression.

The IP-LC method was developed for the determination of 62 metabolites (Supplementary Table S3). Significant injection order effects were observed, whereby the peak areas of most metabolites decreased as the function of injection order (Fig. 1). Notably, < 10 % of the targeted metabolites showed acceptable intra-batch reproducibility (RSD < 20 %) in all 5 types of QC samples (Fig. 2C). Injection order effects frequently observed in the IP-LC method are largely attributable to a build-up of ion-pairing reagents in the LC-MS/MS system during a continuous batch analysis, which contaminates the LC column and/or MS detector thus causing a gradual decrease of MS signals [28,29]. For either targeted or untargeted metabolomics analyses involving IP-LC method, injection order effects can cause significantly non-biological systematic bias, and thus require appropriate correction.

While several algorithms have been developed and applied for calibration or correction of injection order effects in LC-HRMS based untargeted metabolomics, few has been implemented in targeted metabolomics analyses. In this study, we developed a simple correction algorithm specifically for targeted analyses, which was based on the trend of signal drifts of individual targeted metabolites as the function of injection order. The signal drift trends for all quantifiable metabolites in 0.2 µM standard QC were well fitted by an exponential model (Eq. (1)), with the correlation coefficient R² > 0.90. Similar fittings were made for spiked MCF-7 cell matrix QC and spiked MDA-MB-231 cell matrix QC, except for 3 metabolites (i.e., phosphoenolpyruvic acid, 3-hydroxy-3-methylglutaryl-CoA, and thiamine monophosphate) that could not be fitted by the exponential model. These 3 metabolites showed a sharp signal decline in the later injected QCs, probably due to their instability in cell matrix. Exponential model fittings for the signal drift trends of representative metabolites are graphically illustrated in Fig. 3.

Fig. 3.

Exponential model fittings of the IP-LC signal drift trends as a function of injection order for representative metabolites in (A) 0.2 µM standard mix QC, (B) spiked MCF-7 cell matrix QC, and (C) spiked MDA-MB-231 cell matrix QC. The trend of signal drift, expressed as the percentage of peak area of each individual metabolite relative to the peak area in the first injection of the same type of QCs as the function of injection order, is fitted by the exponential function (Eq. (1)).

A correction factor for each individual metabolite, which was calculated based on the respective exponential model of the signal drift trend in the QC, was used to estimate the “true” signal (peak area or concentration) of a metabolite in the samples (Eq. (2)). The performance of the correction models was evaluated by examining the PCA scores plots of the corrected data and their RSD distributions for 5 types of QC samples. As shown in Fig. 4, after correction by the exponential models developed from either 0.2 µM standard mix QCs or spiked MDA-MB-231 cell matrix QCs, the signals from each type of QC samples was closely clustered in the PCA scores plots. In addition, the percentage of metabolites with acceptable intra-batch reproducibility (RSD < 20 %) was increased from 0 % (for pre-corrected data) to 75–100 % for all 5 types of QCs (Fig. 5). Collectively, these results suggest that the injection order effects in the IP-LC method were effectively corrected by the developed exponential models of signal drift trends for individual targeted metabolites using either standard mix QC or spiked cell matrix QC.

Fig. 4.

Principal component analysis based on the peak areas corrected for injection order effects in 5 types of QC samples analyzed by the IP-LC method. The peak areas of individual metabolites are corrected by the respective exponential equations of signal drift trends developed from (A) 0.2 µM standard mix QC, and (B) spiked MDA-MB-231 cell matrix QC. The 3D scores plots of the first, second, and third principal component are shown. Each set of QC data is closely clustered together, suggesting injection order effects are effectively corrected.

Fig. 5.

The distributions of intra-batch variations of the IP-LC method for 5 type of QC samples after the peak areas of individual metabolites are corrected for injection order effects using the respective exponential equations of signal drift trends developed from (A) 0.2 µM standard mix QC, and (B) spiked MDA-MB-231 cell matrix QC. Intra-batch variation is assessed as the mean relative standard deviation (RSD) of the corrected peak areas of individual metabolites in the QC samples analyzed in a continuous analytical batch.

4. Conclusion

Three complemented chromatographic separation methods (i.e., RP-LC, HILIC, and IP-LC) coupled with tandem mass spectrometry are applied in targeted metabolomics analyses of ~250 metabolites to achieve effective separation and retaining of all metabolites. Both RP-LC and HILIC methods exhibit robust intra-batch reproducibility for 0.2 µM standard mix QC, with ~96 % of the measured metabolites showing RSD < 20 %; whereas, the intra-batch reproducibility for some metabolites in cell matrix QCs may be compromised due to stability issue, suboptimal chromatographic retention, and/or matrix effects causing ionization suppression and/or retention instability. Given the fact that 20-fold dilution of biological samples could effectively diminish matrix effects [24–26], a standard operation procedure is implemented to analyze both original and 20-fold diluted post-extracted samples in an analytical batch and report the metabolite concentrations determined from the diluted samples if the concentrations are within the linear calibration curve ranges. The IP-LC method exhibits significant injection order effects due to a build-up of ion-pairing reagents in the LC-MS/MS system causing a gradual decrease of MS signals. With interspersing either standard mix QC or spiked cell matrix QC samples in a continuous batch analysis, exponential models of signal drift trends for individual metabolites can be developed to correct injection order effects of the IP-LC method.

Supplementary Material

Appendix A. Supplementary material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jchromb.2022.123513.

Data availability

Data will be made available on request.