Comparison of High-Resolution Fourier Transform Mass Spectrometry Platforms for Putative Metabolite Annotation

Abstract

Fourier transform ion cyclotron resonance (FTICR) and Orbitrap mass spectrometry (MS) are among the highest-performing analytical platforms used in metabolomics. Non-targeted metabolomics experiments, however, yield extremely complex datasets that make metabolite annotation very challenging and sometimes impossible. The high-resolution accurate mass measurements of the leading MS platforms greatly facilitate this process by reducing mass errors and spectral overlaps. When high resolution is combined with relative isotopic abundance (RIA) measurements, heuristic rules, and constraints during searches, the number of candidate elemental formula(s) can be significantly reduced. Here, we evaluate the performance of Orbitrap ID-X and 12T solariX FT-ICR mass spectrometers in terms of mass accuracy and RIA measurements and how these factors affect the assignment of the correct elemental formulas in the metabolite annotation pipeline. Quality of the mass measurements was evaluated under various experimental conditions (resolution: 120, 240, 500 K; automatic gain control: 5 × 10⁴, 1 × 10⁵, 5 × 10⁵) for the Orbitrap MS platform. High average mass accuracy (<1 ppm for UPLC-Orbitrap MS and <0.2 ppm for direct infusion FT-ICR MS) was achieved and allowed the assignment of correct elemental formulas for over 90% (m/z 75–466) of the 104 investigated metabolites. ¹³C₁ and ¹⁸O₁ RIA measurements further improved annotation certainty by reducing the number of candidates. Overall, our study provides a systematic evaluation for two leading Fourier transform (FT)-based MS platforms utilized in metabolite annotation and provides the basis for applying these, individually or in combination, to metabolomics studies of biological systems.

Graphical Abstract

Non-targeted metabolomics studies yield extremely complex datasets due to their comprehensive nature. When mass spectrometry (MS) analytical platforms are used, hundreds to thousands of endogenous metabolites can be detected in biological samples owing to this platform’s high sensitivity.¹ As the data are collected without a predefined list of target metabolites, the confident annotation of selected species is a prerequisite step to a robust biological interpretation of the results. However, annotation is currently recognized as the major bottleneck in non-targeted metabolomics studies due to the wide structural diversity of metabolites² and because the majority of small molecule structures in biological systems are still unknown.

Unlike proteins and nucleic acids, which are derived from a well-defined set of building blocks (e.g., amino acid, nucleobases) and can therefore be sequenced, biogenic small molecules may incorporate building blocks from many metabolic pathways and can be further transformed through a number of routes. The existence of structural and stereoisomers adds additional complexity to the metabolome. Though a number of databases that contain large lists of endogenous metabolites in biological systems are available for searching, such as the Human Metabolome Database (HMDB),³ METLIN,⁴ mzCloud (https://www.mzcloud.org), LIPID MAPS,⁵ LipidBlast,⁶ National Institute of Standards and Technology (NIST) library (http://chemdata.nist.gov), etc., their coverage is still largely incomplete.²

The process of metabolite annotation is typically started by searching experimental mass measurements against databases with a certain mass tolerance window to obtain potential candidate elemental formulas. These putative metabolite annotations are then refined by matching orthogonal information (e.g., tandem mass spectrometry (MS/MS) fragmentation information,⁷ retention time, collision cross section (CCS),⁸ drift time,⁹ etc.). Using more than one type of supporting orthogonal information is preferable to increase the confidence in the annotations.¹⁰ Assignments are ultimately validated by matching pure reference chemical standards under identical experimental conditions, if available. The initial assignment of accurate elemental formulas is, therefore, of central importance for metabolite annotation. High-resolution mass measurements greatly facilitate elemental formula assignments by reducing mass errors and spectral overlaps. When applied together with heuristic rules, analysis of isotope abundance pattern, and constraints during searches (e.g., hydrogen/carbon ratios, ring double bond equivalents (RDBE) restrictions, elemental ratios of nitrogen, oxygen, phosphorous, and sulfur vs. carbon, etc.),^11,12 the number of candidate elemental formulas found can be significantly reduced, greatly simplifying the process.

Currently, Fourier transform ion cyclotron resonance (FTICR) MS¹³ and Orbitrap MS¹⁴ are among the highest-performing analytical MS platforms in metabolomics.^15,16 Both employ a Fourier transform to determine the frequency of ion motion in confining magnetic and/or electric fields followed by mass calibration and thus are referred to as Fourier transform-based mass spectrometry (FTMS).¹⁷ The inherent high fidelity of FTMS in the determination of ion motion frequency allows very high mass accuracy and resolving power, enabling a highly detailed investigation of biological metabolomes.^15,16 The latest generation of Orbitrap instruments is able to achieve mass resolution up to 1 M full width at half maximum (FWHM) at m/z 200 and mass accuracies better than 1 ppm,¹⁸ whereas FT-ICR mass spectrometers can achieve maximum resolving power >10 M and mass accuracy better than 0.2 ppm.¹⁹ However, high mass accuracy alone is not sufficient for determining the most likely elemental formula. Even with <1 ppm mass accuracy, the number of potential elemental formula candidates increases with increasing of m/z.²⁰ Therefore, additional information is needed to further reduce the number of possible assignments and improve annotation confidence. For example, Kind and Fiehn found that mass spectrometers capable of 3 ppm mass accuracy with 2% error in the isotopic abundance patterns outperformed mass spectrometers with <1 ppm mass accuracy alone in determining correct elemental formulas.²⁰ It is now generally accepted that relative isotopic abundance (RIA) of stable elemental isotopes such as ¹³C, ¹⁸O, ¹⁵N, and ³⁴S is highly informative for the correct assignment of elemental formulas, as it can be used to estimate the number of various atoms present in the composition of a given metabolite. Ultrahigh resolution is a desirable feature for accurate RIA measurements, as it helps to reduce spectral overlap, producing isotopic clusters free from spectral interferences. For example, FT-ICR MS has been shown to be capable of distinguishing CHONPS elemental isotopes and unambiguously determining molecular formulas.^21,22

Though FT-ICR MS has unparalleled high resolving power, its main disadvantage is the relatively slow acquisition rate and lack of automatic gain control (AGC), somewhat limiting its application to liquid chromatography-mass spectrometry (LC-MS)-based metabolomics.¹⁶ As a result, FT-ICR MS in direct infusion mode (DI-MS) is typically favored over LC FT-ICR MS approaches.^15,16,21 However, the recent implementation of absorption mode and ICR signal detection at the high order harmonic frequencies have enabled faster acquisition rates,^23,24 leading to even more successful applications in non-targeted metabolomics.^21,25–27

Together with time-of-flight instrumentation, Orbitrap MS coupled with LC is perhaps one of the most widely used metabolomics platforms due to its high mass resolution, mass accuracy, and sensitivity, with high-field Orbitrap analyzers having scan rates readily compatible with ultra performance liquid chromatography (UPLC).^16,28–30

To date, a number of studies have been conducted to evaluate the performance of FTMS in terms of metabolite annotation. For example, Knolhoff et al. compared the mass accuracy and RIA between Orbitrap and quadrupole-time-of-flight (Q-TOF) mass spectrometers.³¹ Xu et al. evaluated accurate mass measurements and RIA of linear ion trap quadrupole (LTQ)-Orbitrap MS,³² and Barbier Saint Hilaire et al. evaluated the performance of the Orbitrap Fusion mass spectrometer by accurately measuring masses and RIA for metabolite annotation purposes.³³ Mitchell et al. developed a method for determining elemental formulas using detected sets of isotopic peaks from spectra generated by FTMS platforms.³⁴ Thompson et al. developed a flow injection (FI) continuous accumulation of selected ions (CASI) FT-ICR MS workflow to reveal isotopic fine structures and better assign elemental formulas.³⁵ In addition, Weber et al. characterized the accuracy of RIA measurements of two FTMS platforms, LTQ FT-ICR MS and LTQ Orbitrap MS, for elemental formula assignment.³⁶ Those studies provided valuable insights into the performance of leading FTMS analytical platforms in the metabolomics field at the time of publication.

In this study, we evaluate the mass and ¹³C and ¹⁸O RIA measurement accuracies produced by high-field Orbitrap ID-X and 12T solariX FT-ICR platforms and investigate how these affect the assignment of the correct elemental formulas in metabolite annotation applications. For Orbitrap MS, we assessed the performance at different resolution and automatic gain control (AGC) settings. Mixtures containing 104 small molecular weight metabolites (m/z 75–620) in model solutions and complex Caenorhabditis elegans biological matrices were used for the experiments. DI-FT-ICR MS and UPLC-Orbitrap MS workflows were developed to evaluate the performance of the platforms under optimized conditions. To the best of our knowledge, this is the first report systematically evaluating the performance of two of the highest-performing FTMS platforms at the present time in a metabolomics context.

EXPERIMENTAL SECTION

Sample Preparation.

One hundred and four selected chemical standard compounds were purchased from Sigma-Aldrich (St. Louis, MO) and used to prepare stock solutions at a concentration of 0.001 M in LC-MS grade methanol (Fisher Scientific, Pittsburgh, PA). These standards were selected because they map to key metabolic pathways, some altered in cancer.³⁷ The list of standard compounds used is provided in the Supporting Information section (Table S1). If a chemical could not be completely dissolved in methanol, LC-MS grade water (Fisher Scientific, Pittsburgh, PA) was added to increase the solubility. A pooled sample of 104 standard compounds was diluted with methanol at a final concentration of 5 μM.

Three-2.0 mm zirconium oxide beads and ~75 μL of 0.5 mm glass beads were added to each lyophilized PD1074 C. elegans sample (~10 mg). Samples were placed in a TissueLyser II (QIAGEN, Hilden, Germany) and homogenized at 1800 rpm, at 80 °C for 3 min. One and a half milliliter of 80% methanol (in water) was added to each homogenized sample. Samples were then shaken using an Isotemp high-speed shaker (Fisher Scientific, Pittsburgh, PA) at 1500 rpm for 30 min, and centrifuged at 22 100g for 5 min. The supernatant was collected, dried, and stored at −80 °C. Prior to MS analysis, dried C. elegans matrices were resuspended with 1 mL LC-MS grade methanol containing the 104 standard compounds at a 5 μM concentration.

FTMS Analysis.

Samples were subjected to DI-MS analysis on a 12T solariX FT-ICR mass spectrometer equipped with a Paracell (Bruker, Bremen, Germany) and UPLC-MS analysis on an Orbitrap ID-X mass spectrometer (Thermo Fisher Scientific, Waltham, MA) in both positive and negative ion modes. For UPLC-Orbitrap MS analysis, liquid chromatography was performed with an Acquity UPLC BEH Amide, 2.1 × 150 mm², 1.7 μm-particle column on a Vanquish UPLC system (Thermo Fisher Scientific, Waltham, MA). Mobile phase A was water/acetonitrile (80:20 v/v), 10 mM ammonium formate, and 0.1% formic acid. Mobile phase B was acetonitrile with 0.1% formic acid. All solvents and additives used to prepare mobile phases were LC-MS grade and purchased from Fisher Chemical (Pittsburgh, PA). The gradient program used for chromatographic separation can be found in the Supporting Information section (Table S2). The column temperature was 40 °C, while samples were maintained at 4 °C in the autosampler. The injection volume was 2 μL. The spray voltages were 3.5 and −2.5 kV in positive and negative ion modes, respectively. In both modes, various combinations of resolution settings (120, 240, and 500 K) and AGC target values (5 × 10⁴, 1 × 10⁵, 5 × 10⁵) were used in data acquisition. The instrument was externally calibrated with a Pierce FlexMix calibration solution (Thermo Fisher Scientific, Waltham, MA), and the internal calibration feature (EASY-IC) was turned on.

For DI-FT-ICR MS analysis, samples were introduced to the ion source through a syringe pump at a 4.0 μL min⁻¹ flow rate (total sample volume consumed: ~32 μL per 8.35 min DI-MS analysis). MS acquisition parameters in both positive and negative ion modes can be found in the Supporting Information (Table S3). A spectral size of 8 M was used for data acquisition with an ion accumulation time of 0.025 s and a scan range of 73.71–1000.00 m/z. Before data acquisition, the instrument was externally calibrated with 0.1 mg mL⁻¹ sodium trifluoroacetate (NaTFA) in water/acetonitrile (50:50 v/v). Each sample was run in triplicate. CASI experiments were also attempted; however, no significant performance improvements were observed and were therefore not further pursued.

Data Processing.

Following data acquisition, MZmine 2.51^38,39 was used for preprocessing of profile data generated from UPLC-Orbitrap MS and DI-FT-ICR MS analyses. Raw data generated from the FT-ICR platform was converted to mzML files using msConvert⁴⁰ prior to importing into MZmine. The procedure included peak picking, chromatogram deconvolution, and peak area integration. The parameters for processing the raw data collected from Orbitrap MS and FT-ICR MS can be found in the Supporting Information. For all of the datasets, mass errors were calculated and used to assign elemental formulas (mass error tolerance: <0.001 m/z or 3.0 ppm) with [C_0–50H_0–100N_0–15O_0–20P_0–7S_0–8 ± H]^± limits, heuristic rules, and RDBE restrictions applied through the MZmine built-in functions (Supporting Information).¹¹ RIA errors and the number of elemental atoms were calculated using eqs 1 and 2, respectively

$${\text{RIA}}^{\text{error}}=\frac{{\text{RIA}}^{\text{exp}.}-{\text{RIA}}^{\text{theo}.}}{{\text{RIA}}^{\text{theo}.}}\times 100\%$$ (1)

$$\text{no}{.}^{\text{atoms}}=\frac{{\text{RIA}}^{\text{exp}.}}{\text{relative}\text{abu}.}$$ (2)

where exp. stands for experimental, theo. stands for theoretical, and abu. stands for abundance (¹³C: 1.1%, ¹⁸O: 0.2%).

A three-step filtering approach was used for assigning elemental formulas for every investigated m/z value: (1) Assign candidate elemental formulas solely by mass accuracy using the built-in formula prediction function in MZmine. (2) Calculate the number of carbon atoms based on ¹³C RIA and then use the no.C_exp ± 1 to filter elemental formula candidates. (3) Calculate the number of oxygen atoms based on ¹⁸O RIA and use the no.O_exp ± 1 to further filter elemental formula candidates.

Data generated in this work are available through the NIH Metabolomics Workbench (http://www.metabolomicsworkbench.org/) with project ID PR001131 and doi: 10.21228/M8HD7B (Orbitrap MS, study ID: ST001777 and FT-ICR MS, study ID: ST001778).

RESULTS AND DISCUSSION

Mass Accuracy Evaluation.

A pooled sample containing 104 metabolite standards in methanol, with m/z values ranging from 75 to 620, was analyzed using both DI-FT-ICR MS and UPLC-Orbitrap MS. Pairs of retention time and monoisotopic accurate masses measured for individual standards were used to identify the metabolites in the mixture. The impact of different resolution and AGC target settings on mass accuracy was investigated for the UPLC-Orbitrap MS platform and compared with the DI-FT-ICR MS platform. The number of metabolites species detected and their corresponding mass accuracy for different settings are shown in Table 1.

Table 1.

Average Mass Accuracy for Metabolite Monoisotopic Ions Detected in a Model Mixture under Various Experimental Conditions

	Average Mass Error (ppm)a				Number of Metabolites Detected (out of 104)d
	AGC╲Rb	120 K	240 K	500 K	AGC╲R	120 K	240 K	500 K
Pos. Mode	5e4	0.34	0.41	0.46	5e4	74	72	72
	1e5	0.29	0.33	0.37	1e5	73	72	70
	5e5	0.33	0.36	0.35	5e5	72	71	65
	FT-ICRc		0.10		FT-ICR		75
Neg. Mode	5e4	0.36	0.31	0.29	5e4	88	83	78
	1e5	0.37	0.32	0.36	1e5	87	82	75
	5e5	0.69	0.71	0.88	5e5	82	75	71
	FT-ICR		0.11		FT-ICR		93

^aThe mass error was calculated as the mean absolute error for all detected species.

^bR: mass resolution (defined as FWHM at m/z 200), AGC: automatic gain control target, both settings were for the Orbitrap platform.

^cFor the FT-ICR MS platform, a spectral size of 8 M (~700 K mass resolution at m/z 200) and 0.025 s ion accumulation time were used for data acquisition. Pos.: positive, neg.: negative.

^dFor the Orbitrap MS platform, 45–62 overlapping species were detected in both modes; the highest number (62) were detected at a resolution setting of 120 K with an AGC target value of 5 × 10⁴ and the lowest number of overlapping species (45) were detected at a resolution setting of 500 K with an AGC target value of 5 × 10⁵. For the FT-ICR MS platform, a total of 65 species were detected in both ion modes.

As expected, metabolite standards exhibited a wide range of ionization efficiencies in the different ionization modes tested. For the Orbitrap platform, 65–74 species were detected in positive ion mode and 71–88 species were detected in negative ion mode, with 45–62 species overlapping between both modes. A lower number of metabolites were detected when resolution and/or AGC target values were increased in Orbitrap experiments. The lowest number of metabolites—65 and 71 species in positive and negative modes, respectively—were detected at a resolution setting of 500 K with an AGC target value of 5 × 10⁵ (Table 1). These ion losses were attributed to the lower scan speeds at the highest resolving power setting and the increased ion filling at high AGC target values that results in lower abundance ion species becoming undetectable.⁴¹ In addition, the resolution and AGC target settings also affected the detected overall peak abundances. For example, the peak area for choline (m/z 104.10699 in positive mode) was 1.9 × 10⁹ at 500 K resolution with an AGC target of 5 × 10⁴. The peak area for the same species was 6.4 × 10⁸ at 500 K resolution with an AGC target of 5 × 10⁵, a 66% decrease. When the AGC target setting remained at 5 × 10⁵, reducing the resolution to 120 K increased the peak area to 1.0 × 10⁹, a 56% increase. These results suggest that the choice of high resolution and AGC target settings can result in significant signal loss, introducing changes in the relative quantitation of low abundance metabolites. For the FT-ICR platform, 75 and 93 metabolite species were detected in positive and negative ion modes, respectively. Sixty-five of these were detected in both ion modes. Compared with the Orbitrap dataset, the number of metabolites detected using DI-FT-ICR was comparable in positive mode and higher in negative mode, likely due to the lack of UPLC dilution effects in DI mode and the higher sample volume consumed by the FT-ICR platform. These results should be interpreted with caution, as the UPLC-MS platform is less sensitive to ionization suppression effects than the DI platform, so for more complex samples, the DI platform may detect a lower number of analytes. Detection differences could also be ascribed to the differences in solvents present during the ionization process (UPLC mobile phase vs methanol). A list of detected metabolites is provided in the Supporting Information (Table S1).

Under the various experimental Orbitrap conditions investigated, the absolute mass accuracy for metabolite species was, on average, below 0.5 ppm, except at the 5 × 10⁵ AGC target setting in negative ion mode. This very high mass accuracy is a result of the combined external and internal calibration used in the Orbitrap platform. In agreement with the findings by Barbier Saint Hilaire et al., resolution had little effect on mass accuracy, whereas the highest AGC target setting was observed to have a negative impact.³³ Though not observed in positive mode, increasing the AGC target to 5 × 10⁵ increased the mass error to ~0.7–0.9 ppm in negative ion mode. This reduction in mass accuracy was attributed to higher space charge effects with increased ion filling at the higher AGC target setting. Space charge effects are known to affect frequency shifts of ions with adjacent frequencies, resulting in a mass accuracy decrease.⁴² For FT-ICR, the observed mass accuracy was excellent, reaching ~0.1 ppm in both ion modes and only requiring external calibration. This ultrahigh mass accuracy is primarily a result of the high magnetic field in the ICR cell that minimizes ion cyclotron frequency shifts due to space charge effects.¹⁹

C. elegans extract samples with spiked-in metabolite species were analyzed to investigate the impact of a complex biological matrix on accurate mass measurements. The experimental conditions were identical to those used in the experiments described above. The performance of the Orbitrap platform for detecting spiked metabolites in the biological sample was similar to that in the model mixture. A number of metabolites not detected in the model mixture were detected in experiments with the spiked biological sample, likely due to higher ion abundances stemming from endogenous contributions. The Orbitrap mass accuracy was comparable with that for metabolites in model solutions. Mass errors were below 0.5 ppm, except at the 5 × 10⁵ AGC target setting in negative ion mode, where mass errors were in the 0.61–1.13 ppm range. This mass error increase was explained by the added space charge effects in the presence of ions from the more complex biological matrix at such a high AGC target setting. For FT-ICR MS experiments, the average mass error was 0.17 ppm in both ion modes with external calibration, illustrating the excellent performance in detecting metabolites in complex biological matrices. A summary of the mass accuracy results and the number of detected metabolite species in the C. elegans matrix are provided in Table S4.

Evaluation of Relative Isotope Abundance Errors.

As the vast majority of metabolites contain carbon and/or oxygen atoms, RIA measurement of ¹³C/¹²C and ¹⁸O/¹⁶O pairs can provide valuable information for assigning elemental formulas by constraining the number of elemental atoms.³² The ability to resolve isotopic fine structures—for example, distinguishing ¹⁸O₁ from ¹³C₂—is essential for accurate RIA measurements. In evaluating the accuracy of RIA measurements for the two FTMS platforms, we first identified ¹³C₁ and ¹⁸O₁ isotopes from the datasets collected under various experimental conditions. For the Orbitrap platform, 45–61 ¹³C₁ isotopic signals were detected for the targeted metabolites in positive ion mode under various resolutions and AGC target settings. Due to lower natural abundance of the ¹⁸O₁ isotope, only 12–29 ¹⁸O₁ isotopic signals were detected under the same conditions. As with monoisotopic ion species, increased AGC target and resolution settings resulted in a reduced number of ¹³C₁ and ¹⁸O₁ signals, especially for lower abundance ions. For FT-ICR, 55 ¹³C₁ and 19 ¹⁸O₁ isotopic signals were detected in positive ion mode. The lower number of detected isotopes with FT-ICR was attributed to the ion suppression effects caused by the DI electrospray sample introduction, whereas UPLC was used for Orbitrap experiments, reducing interferences through chemical separation.⁴³ A similar trend was observed in negative ion mode for the Orbitrap platform; 36–60 ¹³C₁ and 8–22 ¹⁸O₁ isotopes were detected, with a lower number of isotopes detected at higher AGC target and resolution settings. Contrary to experiments in positive ion mode, 80 ¹³C₁ and 45 ¹⁸O₁ isotopes were detected in negative mode with FT-ICR, more than the species detected with Orbitrap. This was attributed to the higher ionization efficiency of some of the investigated metabolites in negative ion mode that contributed to better detection.

RIA errors were calculated for the detected ¹³C₁ and ¹⁸O₁ isotopes using eq 2 defined in the experimental section. In agreement with previous findings using Orbitrap MS,^32,33 RIA errors for ¹³C₁ isotopes were affected by resolution and varied with peak abundance, whereas AGC target setting had little effect. In positive ion mode, the mean absolute RIA errors for ¹³C₁ isotopes were 6.2, 8.8, and 13.3% at resolution settings of 120, 240, and 500 K, respectively. At 500 K resolution, RIA errors were greatly increased for monoisotopic species with abundances below 1.0 × 10⁶, with the shift mainly toward negative RIA error values. This type of shift has been previously reported^32,33,36 and may be explained by the instability of small ion clouds in the Orbitrap.^33,44,45 The increased RIA errors could also be attributed to the limitations of AGC to maintain nonequivalent scans consistently during liquid chromatography and the flooring performed by the data acquisition software during Fourier transformation. Therefore, the more intense the monoisotopic ion signal, the more accurate the ¹³C₁ RIA measurement for determining elemental formulas. For FT-ICR, there was no observable correlation between RIA deviation and peak abundance. The mean absolute RIA error value of ¹³C₁ isotopes was 12.2%, which was comparable to the Orbitrap MS platform at 500 K mass resolution. In negative ion mode Orbitrap MS, the mean absolute RIA error values of ¹³C₁ isotopes were 5.8, 10.2, and 13.4% at resolution settings of 120, 240, and 500 K, respectively. FT-ICR MS data generated a mean absolute ¹³C₁ RIA error value of 10.1% in negative ion mode. Higher ¹³C₁ RIA errors were observed for ion species with abundance below 1.0 × 10⁶ at 240 and 500 K resolution with the Orbitrap, whereas FT-ICR data did not exhibit similar RIA deviation trends. RIA errors for ¹³C₁ isotopes under various experimental conditions are illustrated in Figure 1.

Figure 1.

Orbitrap RIA errors for ¹³C₁ isotopes at 120 K (a, d), 240 K (b, e), and 500 K (c, f) resolution settings; and at ~700 K (g, h) resolution setting using FT-ICR MS. RIA errors of the ¹³C₁ (M + 1) isotopic ions are expressed as a function of the abundance of the corresponding monoisotopic ion (M). For Orbitrap MS, data were collected in positive (a–c) and negative (d–f) ion modes. In all cases, data acquired at different AGC target settings (5 × 10⁴, 1 × 10⁵, and 5 × 10⁵) were combined. For FT-ICR MS, data were collected in positive (g) and negative (h) ion modes. In both ion modes, a transient size of 8 M (~700 K resolution) was used. Mass resolution was defined at FWHM at m/z 200.

RIA measurement errors for ¹⁸O₁ isotopes were generally higher than those for ¹³C₁ isotopes. Similar trends were reported in previous studies with older Orbitrap instruments.^32,33 For the ID-X Orbitrap platform, the mean absolute RIA error values for ¹⁸O₁ isotopes in positive ion mode across chromatographic peaks were 16.0, 24.5, and 43.8% at resolution settings of 120, 240, and 500 K, respectively. In negative ion mode, the mean absolute RIA errors were 12.5, 18.0, and 30.3% at 120, 240, and 500 K, respectively. Compared with the more abundant ¹³C₁ isotopes, the higher ¹⁸O₁ RIA errors were caused by the lower isotopic ion abundances. Overall, increasing the resolution setting was observed to lower the quality of RIA measurements. For FT-ICR MS, the mean absolute ¹⁸O₁ RIA errors across 200 scans were 16.8 and 13.3% in positive and negative ion modes, respectively. RIA errors for ¹⁸O₁ isotopes in the synthetic standard mixture are provided in the Supporting Information (Figure S1).

The impact of the biological matrix on RIA measurement errors for ¹³C₁ and ¹⁸O₁ isotopes was also evaluated. For Orbitrap MS, 48–61 and 16–36 ¹³C₁ and ¹⁸O₁ isotopes, respectively, were detected in positive ion mode, with a lower number of detected isotopic species at higher AGC target and resolution settings. The mean absolute ¹³C₁ RIA errors ranged from 4.7 to 8.8%, and ¹⁸O₁ RIA errors ranged from 12.2 to 30.5%, with increased RIA errors at higher resolution settings. In negative ion mode, 37–70 ¹³C₁ and 11–29 ¹⁸O₁ isotopes were detected, with mean absolute ¹³C₁ RIA errors ranging from 7.2 to 13.4%, and ¹⁸O₁ RIA errors ranging from 11.9 to 30.6%. Compared with standards in model mixtures, the improved number of detected isotope species and RIA errors were attributed to the added abundance of endogenous metabolites in the biological matrix. A list of endogenous metabolites detected in the biological matrix but not detected in model mixtures is given in Table S5. For FT-ICR MS, the results were similar to the standards in neat solvent. In positive ion mode, 57 ¹³C₁ and 22 ¹⁸O₁ isotopes were detected, with mean absolute RIA errors of 11.7 and 19.0% for ¹³C₁ and 22 ¹⁸O₁ isotopes, respectively. In negative ion mode, 79 ¹³C₁ and 48 ¹⁸O₁ isotopes were detected, with mean absolute RIA errors of 10.2 and 12.8% for ¹³C₁ and 22 ¹⁸O₁ isotopes, respectively. A summary of observed RIA errors for ¹³C₁ and ¹⁸O₁ isotopes in the C. elegans biological matrix under various experimental conditions is provided in the supporting information section (Figures S2 and S3).

Impact of Mass Accuracy and RIA Measurement on Elemental Formula Assignments.

In evaluating the impact of mass accuracy and RIA errors on elemental formula assignments, we first compared the performance of the two FTMS platforms at similar resolution settings (~500 K at m/z 200 for Orbitrap MS and m/z 400 for FT-ICR MS). We determined an AGC target setting of 1 × 10⁵ as optimal due to the higher AGC setting (5 × 10⁵), resulting in lower mass accuracy and the lower AGC setting (5 × 10⁴), decreasing the overall sensitivity.³⁵ C. elegans extracts with spiked-in metabolites were used for these experiments, as these samples better mimic complex biological samples from a typical metabolomics study. In assigning elemental formulas, a mass error tolerance <3 ppm was used, together with heuristic rules and RDBE restrictions as the filtering criteria. For Orbitrap MS, 94.3% (66 out of 70) and 90.1% (73 out of 81) of metabolites investigated were assigned the correct elemental formula as the top 1 candidate in positive and negative ion modes, respectively. Among those metabolites, 67.1% (47 out of 70) and 66.7% (54 out of 81) had a single correct elemental formula in positive and negative modes, respectively. For FT-ICR MS, 100.0% (77 out of 77) and 92.6% (87 out of 94) of metabolites were assigned the correct elemental formula in positive and negative ion modes, respectively. Of these, 68.8% (53 out of 77) and 67.0% (63 out of 94) only had a single candidate elemental formula in positive and negative ion modes, respectively.

Starting at m/z 225, more than one candidate elemental formulas can be generated within a 0.001 m/z or 3 ppm mass error threshold; from m/z 285, the number of candidate elemental formulas exceeds 10. Though accurate mass measurements allow the assignment of correct elemental formulas as the top candidate in most cases, further filtering criteria are needed to reduce such numbers and increase confidence in the assignments.

RIA for ¹³C₁ isotopes was then applied as the second filtering criterion to elemental formula elucidation. The number of carbon atoms (no.C_exp) was calculated based on ¹³C₁ RIA using eq 2 and the no.C_exp ± 1 was used to filter elemental formula candidates. Using ¹³C₁ RIA as a filter, the number of candidate elemental formulas was greatly reduced, and the success of elemental formula assignment greatly improved. Using RIA filtering for Orbitrap MS data, 97.1 (68 out of 70) and 93.4% (76 out of 81) of metabolites investigated were assigned the correct elemental formulas in positive and negative ion modes, respectively. Among those metabolites, 71.4 (50 out of 70) and 75.3% (61 out of 81) had only a single correct elemental formula assignment in positive and negative modes, respectively. For FT-ICR MS, 100.0 (77 out of 77) and 95.7% (90 out of 94) of metabolites were assigned the correct elemental formula, and 76.6 (59 out of 77) and 79.8% (75 out of 94) were assigned a single correct elemental formula in positive and negative ion modes, respectively.

Though RIA for ¹⁸O₁ isotopes was generally less accurate than for ¹³C₁ isotopes, their application could still help determine the best candidate elemental formulas. The number of oxygen atoms (no.O_exp) was calculated, and no.O_exp ± 1, when available, was used to further filter elemental formula candidates. Due to the limited number of detected ¹⁸O₁ isotopes, the improvement on elemental formula assignment was limited for Orbitrap datasets. In positive ion mode, 97.1% (68 out of 70) of metabolites investigated were assigned the correct elemental formula, with 74.3% (52 out of 70) metabolites having a single correct elemental formula assignment. Results remained unchanged for the negative mode dataset. For FT-ICR MS, the performance of elemental formula assignment improved, especially in negative ion mode. 100.0 (77 out of 77) and 97.9% (92 out of 94) of metabolites were assigned the correct elemental formula, among which 79.2 (61 out of 77) and 89.4% (84 out of 94) had a single correct formula in positive and negative ion modes, respectively. This improvement was attributed to more detected ¹⁸O₁ isotopes and relatively more reliable RIA measurements using FT-ICR MS. A summary of elemental formula assignments with various filtering criteria under different experimental conditions is shown in Figure 2.

Figure 2.

Summary of elemental formula assignment with various filtering criteria (mass accu.: only a mass error threshold of <0.001 m/z or 3 ppm was applied; add C13: add no.C_exp ± 1 filter based on ¹³C₁ RIA calculation; add O18: further apply no.O_exp ± 1 based on ¹⁸O₁ RIA calculation). Data were collected using Orbitrap MS in positive (a) and negative (b) ion modes and FT-ICR MS in positive (c) and negative (d) ion modes. For Orbitrap MS, 500 K resolution and an AGC target of 1 × 10⁵ were used. For FT-ICR MS, 8 M transient size and 0.025 s accumulation time were used. ESI: electrospray ionization; NI: not identified or correct elemental formula was not listed in the top 3 candidates; top 3/multiple: correct elemental formulas were listed in top 3 candidates with multiple assignments; 1/multiple: assigned with correct elemental formula as the top 1 candidate with multiple assignments; and 1/single: a single correct formula assignment.

As higher resolution was observed to negatively affect RIA accuracy and sensitivity with the Orbitrap platform, we also evaluated elemental formula assignment performance using the same criteria but at lower resolution settings (120 and 240 K, 1 × 10⁵ AGC target). A summary of elemental formula assignment results at different resolution settings with the Orbitrap MS system is given in Figure 3. Contrary to expectations, the performance of elemental formula assignment at 120 K resolution was better than at 240 K. Though the percentage of correctly identified metabolites was highest at 500 K, it should be noted that the number of detected metabolites species was lower at 500 K than at 120 and 240 K (in positive ion mode: 120 K: 72, 240 K: 72, 500 K: 70; in negative ion mode: 120 K: 88, 240 K: 86, 500 K: 81). These results were attributed to the reduction in signal abundances at higher resolution settings, as previously observed. However, lower resolution at 120 and 240 K was not sufficient for discriminating other minor isotopic signals, such as ¹³C₂ and ¹⁸O₁ at m/z above 250 and 350, respectively,³³ which can cause inaccurate ¹⁸O₁ RIA measurements.0

Figure 3.

Summary of elemental formula assignment results with Orbitrap MS at 120 K (a, b), 240 K (c, d), and 500 K (e, f) resolution settings. Data were collected in positive (a, c, e) and negative (b, d, f) ion modes. In all cases, the AGC target setting of 1 × 10⁵ was used.

Taken together, the optimal experimental settings for the Orbitrap MS platform were a resolution setting of 240 K and AGC target at 1 × 10⁵ for detection and identification purposes. As the m/z values of the majority of the metabolites tested were below 350, this resolution setting was sufficient for separating minor isotopes such as ¹³C₂ and ¹⁸O₁ and sensitive enough for detecting the investigated metabolites with excellent performance for elemental formula elucidation. At 500 K resolution, the performance for correct elemental formula assignment appeared to be comparable, even with increased RIA errors. However, caution should be exercised in terms of the possibility of signal loss and higher RIA errors for less abundant species. For the FT-ICR MS platform, high performance in assigning the correct elemental formula as the top 1 candidate (100% in positive ion mode and 92.5% in negative ion mode) can be achieved solely based on its ultrahigh mass accuracy. Addition of RIA information further enhanced annotation performance due to the relatively lower RIA errors while still maintaining high resolving power. These results suggest that, whenever possible, the use of both platforms in combination may offer the best results in terms of coverage and identification.

CONCLUSIONS

Mass accuracy and ¹³C₁ and ¹⁸O₁ RIA measurements produced by high-field Orbitrap ID-X and 12T solariX FT-ICR platforms were evaluated in the context of elemental formula assignments in metabolomics. A total of 104 metabolites were tested in neat solvent and a complex biological extract matrix. For the Orbitrap MS platform, high mass accuracy (<1 ppm) was achieved with the combination of internal and external calibration. Results indicated that high resolution (500 K) and AGC target (5 × 10⁵) settings resulted in higher mass errors and a lower number of detected metabolite species. At the 500 K resolution setting, RIA errors greatly increased for metabolites with lower abundance (<1.0 × 10⁶). Therefore, it is recommended that 240 K resolution and 1 × 10⁵ AGC target settings are used for general non-targeted metabolomics studies with lower molecular weight metabolites. When higher resolving power is needed, as with very complex mixtures, or when near-isobaric spectral interferences might be present, caution should be taken with RIA accuracies for low abundance species due to the associated signal losses. For the 12 T FT-ICR MS platform, ultrahigh mass accuracy (<0.2 ppm) was obtained using external calibration, together with high performance in assigning the correct elemental formula as the top 1 candidate solely based on accurate mass measurements. This performance, however, was achieved at the expense of slower spectral acquisition rates (~1 Hz). Reliable RIA measurements for ¹³C₁ and ¹⁸O₁ isotopes were obtained while maintaining high resolving power. For both tested platforms, ¹³C₁ and ¹⁸O₁ RIA measurements could improve annotation performance by reducing the number of candidate elemental formulas. To further utilize the advantages of the test platforms, UPLC-Orbitrap MS could be firstly applied for fast metabolic profiling of the samples; fractions of interested metabolites could then be infused into the FT-ICR MS platform for obtaining enhanced isotopic fine structures using its ultrahigh resolving power. Taken together, our study provides a systematic evaluation of two of the most advanced FTMS analytical platforms currently used in metabolomics studies.