B American Society for Mass Spectrometry, 2019 J. Am. Soc. Mass Spectrom. (2019) 30:1733Y1741DOI: 10.1007/s13361-019-02248-w

RESEARCH ARTICLE

Mass Accuracy Check Using Common Background Peaksfor Improving Metabolome Data Quality in Chemical IsotopeLabeling LC-MS

Yunong Li, Liang LiDepartment of Chemistry, University of Alberta, Edmonton, Alberta T6G 2G2, Canada

Abstract.Chemical isotope labeling (CIL) LC-MSis a highly sensitive and quantitative method formetabolome analysis. Because of a large num-ber of peaks detectable in a sample and the needof running many samples in a metabolomics pro-ject, any significant change in mass measure-ment accuracy during the whole period of runningsamples can adversely affect the downstreampeak alignment and quantitative analysis. Herein,we report a rapid method to check the mass

accuracy of individual spectra in each CIL LC-MS run in order to flag up any run containing spectra with accuracydrift that falls outside the expected error. The flagged run may be re-run or discarded before merging with otherruns for peak alignment and analysis. Thismethod is based on the observation that some background signals arecommonly detected in almost all spectra collected in CIL LC-MS runs. A mass accuracy check (MAC) softwareprogram has been developed to first find the common background mass peaks and then use them as massreferences to calculate any mass shifts over the course of multiple sample runs. Using a metabolome dataset of324 human cerebrospinal fluid (CSF) samples and 35 quality control (QC) samples produced by CIL LC-MS, weshow that this accuracy check method can streamline the initial raw data processing for downstream analysis inmetabolomics.Keywords: Chemical isotope labeling, LC-MS, Mass accuracy, Peak alignment, Metabolome analysis,Metabolomics

Received: 10 March 2019/Revised: 24 April 2019/Accepted: 5 May 2019/Published Online: 28 May 2019

Introduction

High-performance chemical isotope labeling (CIL) liquidchromatography mass spectrometry (LC-MS) is a power-

ful tool for metabolome analysis with much improved metab-olite detectability and quantification accuracy, compared toconventional LC-MS [1–4]. CIL LC-MS uses isotope-labeling reagents to react with a common functional group ofmetabolites (e.g., amine) to analyze a chemical-group-basedsubmetabolome (e.g., all amine-containing metabolites or the

amine submetabolome [1]) and then combine the results ofdifferent submetabolomes [1–4] to achieve high-coverage anal-ysis of the whole metabolome. Many labeling reagents havebeen developed [5–11]. With rational design of the labelingreagent structures, concomitant improvements in both metaboliteseparation and MS detection can be achieved [1]. In addition,using differential isotope labeling where a reference sample (e.g.,a pooled sample) is labeled with a heavy isotope reagent andindividual samples are separately labeled with a light isotopereagent, followed by mixing the heavy labeled reference and alight labeled sample and LC-MS analysis of the mixture, accurateand precise relative quantification of concentration differences ofindividual metabolites in comparative samples can be carried outusing the peak pair intensity information [12, 13].

For CIL LC-MS, special software, such as IsoMS [12], hasbeen developed to pick peak pairs from mass spectra, filter out

Electronic supplementary material The online version of this article (https://doi.org/10.1007/s13361-019-02248-w) contains supplementarymaterial, whichis available to authorized users.

Correspondence to: Liang Li; e-mail: Liang.Li@ualberta.ca

the redundant peak pairs (e.g., adduct ions, dimers, etc.) toretain only the [M + H]+ peak pair per labeled metabolite,calculate peak pair intensity ratios, align the peak pairs basedon retention time and mass across multiple sample runs, andthen produce a matrix table of individual peak pairs vs. peakpair ratios. This table can be uploaded to statistical tools forfurther analysis such as finding the metabolites with signifi-cantly changed expression levels in two different groups ofmetabolomic samples (e.g., disease group vs. control group)[14]. Because metabolomics often requires the analysis of alarge number of comparative samples, it is not unusual tocomplete all the runs over a few days or even weeks, particu-larly for in-depth metabolome analysis. Within the long periodof running samples, some of the spectra acquired using a massspectrometer such as time-of-flight (TOF) MS may containpeak pairs with the measured masses having higher than nor-mal measurement errors. The accuracy drift may be caused byfluctuation of instrumental settings, temperature change, etc. Ifthe drift is not detected and re-calibrated back to be within therange of error tolerance, the peak pairs from these abnormalmass spectra could be misaligned with those of other samplesof the same metabolites, resulting in the addition of false peakpairs to the metabolome data (i.e., they are falsely considered tobe new peak pairs of other metabolites).

One way to address the problem is to inspect the entire set ofpeak pair data after all sample runs and data analyses arecompleted. However, this manual inspection may not be prac-tical. In addition, this approach may miss the opportunity of re-running the sample containing problematic files. Herein, wereport a method and software program that can be used topromptly check the mass accuracy of individual mass spectrato identify the problematic spectra. When the problem is de-tected, the sample can be re-run, if adequate sample remains, toresolve the problem during the sample running process. Thisprogram can also be used to check all the data files after sampleruns are finished for the whole project in order to discard anyproblematic file.

Materials and MethodsChemicals and Reagents

All the chemicals and reagents, unless otherwise stated, werepurchased from Sigma-Aldrich Canada (Markham, ON, Can-ada). In a dansylation labeling reaction, the 12C-labeling re-agent (dansyl chloride) was purchased from Sigma-Aldrich,and the 13C-labeling reagent was synthesized and purified inour lab using the procedure published previously [1]. LC-MS-grade water, methanol, and acetonitrile (ACN) were purchasedfrom ThermoFisher Scientific.

Dansylation Labeling

For the CSF metabolome analysis, 324 individual samples werecollected in a metabolomics study with a focus on discoveringbiomarkers of spinal cord injury (SCI) severity and predictive

treatment outcomes. This SCI study, with Ethics Approval fromthe University Ethics Board, will be reported elsewhere. We usethis dataset to illustrate the application and performance of themass accuracy check method and software program in a repre-sentative metabolomics study. To profile the metabolome differ-ences of these samples using dansylation LC-MS (i.e., targetingthe amine/phenol submetabolome), a pooled CSF sample wasprepared by mixing equal aliquots of each individual sample.

For dansylation labeling, in a microcentrifuge tube, 15μL ofCSF was mixed with 10 μL of H2O and 12.5 μL of ACN,followed by the addition of 12.5 μL of 250 mM sodiumcarbonate/sodium bicarbonate buffer. The solution wasvortexed, spun down, and mixed with 25 μL of freshly pre-pared 12C-dansyl chloride (DnsCl) solution (20 mg/mL, forlight labeling) or 13C-DnsCl solution (20 mg/mL, for heavylabeling). After the sample was incubated at 40 °C for 45 min,5 μL of 250 mMNaOHwas added to quench the excess dansylchloride. The solution was incubated at 40 °C for another10 min to allow the unreacted dansyl chloride to be hydrolyzedfully. Finally, 25 μL of formic acid (425 mM) in 1:1 ACN/H2Owas used to acidify the solution. A quality control (QC) samplewas prepared by mixing a 1:1 volume of light labeling andheavy labeling pooled samples. The QC sample was injectedbetween individual samples to monitor the instrument stability.A total of 35 QC injections were conducted during the LC-MSanalysis of the 324 individual human CSF samples.

LC-MS Analysis

The 12C- and 13C-labeled samples were mixed and centrifuged at20,800×g for 10 min before injecting into a Bruker CompactQTOF mass spectrometer (Billerica, MA, USA) linked to aDionex UltiMate 3000 UHPLC system. A Zorbax Eclipse PlusC18 column (2.1 mm×100 mm, 1.8 μm particle size, 95 Å poresize) from Agilent was used. Solvent A was 0.1% (v/v) LC-MS-grade formic acid in 5% (v/v) grade ACN/H2O, and solvent Bwas0.1% (v/v) LC-MS-grade formic acid in LC-MS-grade ACN. Thegradient elution profile was as follows: t = 0.0 min 20% B, t =3.5 min, 35%B, t = 18.0 min, 65%B, t = 24min, 99%B, and t =28 min, 99% B. The flow rate was 180 μL/min.

Data Processing and Analysis

Each LC-MS data file was calibrated using the sodium formatespectra collected in the beginning of the chromatogram file (seeBResults and Discussion^). The mass spectral peaks were thenexported to CSV files using Bruker DataAnalysis 4.3 software.Peak pair data were extracted from each LC-MS file by usingIsoMS [12]. The peak pairs detected in different samples werealigned to generate a CSV table which contains the retention time,accuratemass, peak intensities of each peak pair, and the peak pairintensity ratios in all samples. A zero-fill programwas used to findmissing peak pair ratio based on the rawmass peak in the originalLC-MS files [15]. In the end, peak ratios were recalculated basedon their chromatographic peaks, instead of mass spectral peaks,with IsoMS-Quant program [13]. This table could be uploaded tostatistical tools for further analysis.

1734 Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS

Results and DiscussionMass Calibration in LC-TOF-MS MetabolomeAnalysis

It is well known that TOF-MS requires frequent mass calibrationto correct mass shift caused by instrument factors such as ambienttemperature change or voltage fluctuation in order to maintain thebest possible mass accuracy [16]. In LC-TOF-MS, adding internalmass calibrants into the LC elutes can cause suppression of low-abundance ions, which reduces themetabolome coverage. A lock-mass method has been used to calibrate the analyte spectra ac-quired frequently during the LC run. This method can be imple-mented by introducing external mass calibrants into TOF-MSusing a second spray source [17]. However, not all TOF massspectrometers are equipped with dual spray sources. Anotherapproach is to use certain known background ions from a samplerun as mass calibrants. For example, polydimethylcyclosiloxanes,a group of ubiquitous contaminants of the laboratory air, werefound to be the source of background signals in nano-electrosprayMS and used for calibrating TOF-MS [18]. These background

masses have been used for calibrating Orbitrap and FT-ICR-MS[19, 20]. However, this approach is not particularly suitable forCIL LC-TOF-MS metabolome analysis, as the background sig-nals have low intensities and/or do not cover a wide range of m/z(e.g., m/z 50 to 1000) for accurate mass calibration of all metab-olites (see below).

To obtain as high mass accuracy as possible for TOF-MS,our lab employs an external mass calibration strategy byimplementing a mass calibration segment during the LC earlyelution void time for each LC-MS run. Figure 1a shows an LCchromatogram with a sodium formate calibration solutioninjected at the start of the LC run. In the first 2 min, the eluentfrom the separation column goes to the waste, and another lineof sodium formate solution is connected to the ESI source ofthe mass spectrometer. Figure 1b shows a mass spectrum ofsodium formate adducts in positive ion detection mode forcalibrating the mass range of up to m/z 1000. After datacollection, each LC-MS data file is calibrated using the cali-bration segment. Since mass accuracy is relatively stable overthe course of one sample run (< 35 min), the method is able to

0 5 10 15 20 25 30

0.0

0.5

1.0

1.5

2.0

7x10

Intens.

Sodium formate

peak

(a)

(b)

226.95141+

294.9392

362.92601+

430.91361+

498.90141+

566.88891+

634.87561+

702.86322+

770.85092+

838.8382

0

2

4

6

6x10

Intens.

200 300 400 500 600 700 800

Time (min)

m/zFigure 1. (a) LC chromatogramwith the sodium formate injection in the first 2 min. (b) Mass spectrum of sodium formate signals formass calibration in one LC-MS data file

Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS 1735

correct mass shift in each individual sample more accuratelythan traditional external mass calibration done over a longperiod of time (e.g., at 24-h or weekly intervals). However,mass shift sometimes can occur in the middle of an LC-MS run,causing an excessively large mass shift in some mass spectrawithin the sample data file. In this kind of situation, the use ofthe mass calibration segment cannot correct the mass shift inthese spectra back to be within the range of the normal errortolerance (e.g., 10 ppm), thereby adversely affecting the down-stream peak alignment.

Background Peaks as Mass References

To detect the problem of mass accuracy shift in a run, weexamined the background mass peaks in CIL LC-MS datacollected from many different types of samples using differentlabeling methods. We discovered that there were always somecommon background peaks in a submetabolome analysis. Theycould be from multiple sources, including by-products of la-beling reactions (e.g., dansyl ammonia in dansylation LC-MS).However, the signal of a background peak might vary byseveral orders of magnitudes at different retention times, mak-ing it less suitable for internal mass calibration. Moreover, foraccurate calibration of a wide mass range of up to m/z 1000,multiple calibrants of different masses covering the entire rangeshould be used, as in the case of using sodium formate clusterions as mass calibrants (Fig. 1b). We did not find any set ofbackground peaks with sufficiently high intensities at differentmass ranges for internal calibration. Thus, the background

peaks were not particularly useful for internal mass calibrationin CIL LC-MS. However, some of these common backgroundpeaks found in all mass spectra of an LC-MS run could be usedas mass references, not calibrants, to evaluate the accuracy andprecision of the mass measurement in each spectrum.

To perform mass accuracy check (MAC) based on commonbackground peaks, we have developed a software programwritten in R, which is freely available for research purposesfrom the authors upon request. The first function of MAC is tosearch the background peak masses present in all sample runsto be examined (e.g., all LC-MS runs finished in day 1 of amulti-day analysis project). Figure 2 shows the search strategy

Randomly pick 10 spectra at different

retention times, determine the masses of

peaks found in each spectrum, and

combine the mass peaks

Examine the occurrence of each mass

peak in the 10 spectra

List the mass peaks with occurrence >

80% in each individual sample

No m/z ID

1 251.0845 Dansyl ammonia (12C)

2 279.1151 Dansyl dimethyl amine (12C)

List the common background mass peaks

in all samples

Example of background peak list:

Figure 2. Workflow of the background mass peak searching.The identification of the mass peaks was done by searching thestandard database using retention time and accurate mass

Table 1. Background Mass Peaks Picked from the Human CSF Sample Data

No mz ID

1 252.0694 NA2 251.0845 Dansyl ammonia (12C)3 279.1151 Dansyl dimethyl amine (12C)4 254.0755 NA5 283.1285 Dansyl dimethyl amine (13C)6 253.0893 Dansyl ammonia (13C)7 274.0504 NA

Read the raw mass data of individual LC-

MS files of all examined samples

Extract the measured monoisotopic mass

of the background peak in each file

No RT

m/z of background

peak intensity

123 124.298 251.0847 1770

124 125.308 251.0849 2174

125 126.317 251.0852 2800

126 127.327 251.0850 3376

… … … …

2020 2039.299 251.0838 8158

2021 2040.308 251.0834 8554

Calculate the average monoisotopic mass

and standard deviation of the background

peak against the reference mass

Plot the results in all samples to evaluate

mass accuracy and precision

Example of background peak list:

Figure 3. Workflow of mass accuracy and precision check.For each backgroundmass peak, its measuredmass is extract-ed from each data file. The average measuredm/z and numberof background peak found in each sample are plotted in theresults to assist the evaluation of mass accuracy and precision

1736 Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS

and examples of the MAC program for finding the backgroundpeaks. To speed up the search of background peaks, the pro-gram picks 10 spectra randomly at different retention timesfrom the LC-MS data file and combines all the unique peakmasses into one mass list. If we assume that the commonbackground peaks are present in more than 80% of the spectraacquired within one sample (e.g., in the first LC-MS run), theprobability of each background peak not present in a randomspectrum, out of the 10 spectra examined, is 0.2. Thus, theprobability that the background peak is not present in the 10-spectra mass list is 0.210 or 1.024 × 10−7. Based on this, we canstate with confidence that the 10-spectra mass list should

include the common background mass peaks. These commonpeaks form a list of reference candidates, i.e., they can poten-tially be used as mass references for mass accuracy check.

Next, based on the list of reference candidates, the MACprogram searches each candidate mass, with a user-definedmass tolerance, against the whole LC-MS data file and calcu-lates a total number of occurrences of each mass. SupplementalTable S1 shows an example of the resultant table of referencecandidates found in a human CSFQC sample. Each mass in theBaverage.mz^ column was calculated as the average measuredmonoisotopic mass from this LC-MS file. The standard devi-ation of measuredmonoisotopic masses, peak intensity average

(a)

(b)

Figure 4. (a) Average measured monoisotopic mass of dansyl ammonia in 35 QC samples. The red dashed line indicated thetheoretical value. (b) Measured mass of dansyl ammonia in each scan in file #3

Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS 1737

(average.int), and signal-to-noise ratio average (average.sn)were calculated for each reference candidate. The Bfrequency^column shows how many times each candidate appears in allspectra, and the Boccurrence^ column gives the percentage ofeach candidate showing up in all spectra. This step was repeat-ed for each LC-MS data file in selected or all files within ametabolomics study. In the end, a common background list wasgenerated as a reference mass list (e.g., Table 1 for the CSFsample files).

Mass Accuracy and Precision Check

In the referencemass list such as that shown in Table 1, some ofthe peaks can be identified by searching the labeled standardlibrary (e.g., dansyl standard library in this case) using accuratemass and retention time matches. For identified compounds,the theoretical mass can be used as an internal reference tocheck the actual mass error as well as precision of the measuredpeak. For other unidentified peaks, the average measured mon-oisotopic mass can be used to evaluate mass precision. Figure 3shows the workflow of the MAC program for checking massaccuracy and precision with an example of dansyl ammonia, acommonly detected background in the dansylation LC-MSmethod. We can enter the theoretical m/z of dansyl ammonia,251.0849, in the reference mass selection entry of the program.An optional second reference mass can be entered. Two refer-ence masses are sufficient to detect any mass accuracy prob-lem, since mass shift usually occurs in the whole range of amass spectrum. Within a mass spectrum, if ion suppressioncauses the disappearance of one background peak, anotherbackground peak is usually present for MAC. With each in-putted mass, the program extracts its measured mass in eachspectrum using the user-defined mass tolerance and generatesan extracted mass list. The average measured monoisotopicmass and standard deviation are calculated for the peak.

Figure 4a shows a plot generated by the program from theLC-MS runs of the 35 QC samples. Each data point representsthe average monoisotopic mass and standard deviation of thedansyl ammonia peak in a sample. The average monoisotopicmass shows a mass error of within 5 ppm (0.0013 Da). Thestandard deviation within a sample gives a consistent value inmost samples except the third sample. A relatively large stan-dard deviation indicates a large mass shift within the sample.This suspicious file with potential mass accuracy problem isflagged. We can then look into the extracted peak list of thespecific file in detail as shown in Fig. 4b. Figure 4b shows thatmass shift occurred in this LC-MS run. The large mass shiftwas due to the instability of the detector power supply of themass spectrometer at the time these data were collected. Thetemperature in the laboratory was monitored and controlled to astable level. All other conditions, including LC column andMSmethods, remained the same throughout the data acquisition.As a side note, this mass shift problem occurred more frequent-ly in several subsequent weeks of operation. The power supplyto the detector was found to be faulty; replacing the powersupply corrected the problem.

We applied theMAC program to the 324 human CSF sampleswith the results shown in Fig. 5. To flag up any problematic LC-MS file, we can examine all the files and find the file with asignificantly larger or smaller mass average or a significantlylarger standard deviation. In Fig. 5, we can see seven files (markedwith red) with either a larger or smaller mass average compared tothe rest of the files. Figure 6 shows the mass shifts in one of theflagged samples (file no. 207) for two reference masses fromdansyl ammonia and dansyl dimethyl amine. With the sodiumformate mass calibration, the mass accuracy at the beginning ofthe chromatogram is well within the instrument tolerance (<10 ppm). A mass shift towards the higher mass occurred duringthe data acquisition and caused an elevated average mass andstandard deviation. Both background peaks show a similar mass

Figure 5. Mass accuracy check results for 324 human CSF samples

1738 Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS

shift in thesemass scans. It is worth noting that themeasuredmassshows spikes and dips in some of the scans. Manual inspection ofthese scans indicated that they were from the interfering peakswith similar masses to those of the background peaks (seeSupplemental Figs. S1 and S2 for two examples of theinterfering peaks). However, as shown in Fig. 6a and b, thesespike or dip scans are not detected in common scans in the twobackground peak plots, indicating that they are not from the actualmass shift (i.e., if it is a true shift, both background peaks shouldshow the mass shift in the same scan). Thus, by using two plotslike the ones shown in Fig. 6, we can readily detect the

problematic file. As a comparison, Fig. 7 shows the plots of tworeference masses from a normal file. While occasional spikes anddips in some scans are found, the measured masses are within theinstrument tolerance level (< 10 ppm). Therefore, the use ofmonoisotopic mass average and standard deviation is reliableand robust in flagging up any mass shift file in a large dataset.

Implementation of MAC in CIL LC-MS

We have implemented MAC in the CIL LC-MS metabo-lome analysis workflow for analyzing multiple samples. For

(a)

(b)

Figure 6. Mass shift in file no. 207 shown with two different background mass peaks (a) dansyl ammonia and (b) dansyl dimethylamine

Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS 1739

a typical metabolomics project of running hundreds of sam-ples, we usually run about 30 samples in the first day. MACprocessing time for 30 samples is about 15 min. Thus, afterthe first day sample running, we can check the mass accu-racy of all spectra collected in the first day. If problematicspectra are found in a sample, this sample can be re-run inday 2, if adequate sample is available for another injection.Further data analysis can then proceed after replacing theoriginal problematic sample file with the new sample file.Although mass shift during a sample run occurs less fre-quently with modern TOF-MS, it can still happen,

particularly in situations with electric interferences or powerinstability, temperature change, intermittent instrumentalfailure, etc. Running the MAC program can be a convenientstep in the data processing pipeline to capture any abnor-mality in the operation of a mass spectrometer, therebyhelping the generation of high-quality data for metabolomeanalysis. Although the MAC program is currently applied toCIL LC-MS data produced using TOF-MS, it should be appli-cable to any other LC-MS platforms, including MSwith 5 ppmor less mass accuracy, as long as common background peakscan be found in all MS data files within a run.

(a)

(b)

Figure 7. Mass shift in file no. 296 shown with two different background mass peaks (a) dansyl ammonia and (b) dansyl dimethylamine

1740 Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS

ConclusionsWe have developed a data processing program to check themass accuracy of each spectrum collected in CIL LC-MS runsof multiple samples in metabolome analysis. The programfinds the background peaks commonly detected in a set ofCIL LC-MS data files and uses two or more background peakmasses to check mass accuracy and precision of each spectrum.The MAC results are plotted to quickly detect any sample filewith mass accuracy problem. For a flagged file, more detailedinspection of the LC-MS file can be done to determine whetherthe problematic sample needs to be re-run or discarded.

AcknowledgementsThis work was supported by the Natural Sciences and Engi-neering Research Council of Canada, CIHR, Canada ResearchChairs, Canada Foundation of Innovations, Genome Canadaand Alberta Innovates. We thank Ms. Xinyun Gu for providingthe CSF data.

Compliance with Ethical Standards

Conflict of Interest The authors declare that they have nocompeting interest.

References

Y. Li, L. Li: Mass Accuracy Check in CIL LC-MS 1741

1. Guo, K., Li, L.: Differential 12C-/13C-isotope dansylation labeling andfast liquid chromatography/mass spectrometry for absolute and relativequantification of the metabolome. Anal. Chem. 81, 3919–3932 (2009)

2. Guo, K., Li, L.: High-performance isotope labeling for profiling carbox-ylic acid-containing metabolites in biofluids by mass spectrometry. Anal.Chem. 82, 8789–8793 (2010)

3. Zhao, S., Luo, X., Li, L.: Chemical isotope labeling LC-MS for highcoverage and quantitative profiling of the hydroxyl submetabolome inmetabolomics. Anal. Chem. 88, 10617–10623 (2016)

4. Zhao, S., Dawe, M., Guo, K., Li, L.: Development of high-performancechemical isotope labeling LC-MS for profiling the carbonylsubmetabolome. Anal. Chem. 89, 6758–6765 (2017)

5. Dai, W.D., Huang, Q., Yin, P.Y., Li, J., Zhou, J., Kong, H.W., Zhao,C.X., Lu, X., Xu, G.W.: Comprehensive and highly sensitive urinarysteroid hormone profiling method based on stable isotope-labeling liquidchromatography mass spectrometry. Anal. Chem. 84, 10245–10251(2012)

6. Hao, L., Johnson, J., Lietz, C.B., Buchberger, A., Frost, D., Kao,W.J., Li,L.J.: Mass defect-based N,N-dimethyl Leucine labels for quantitativeproteomics and amine metabolomics of pancreatic cancer cells. Anal.Chem. 89, 1138–1146 (2017)

7. Leng, J.P., Wang, H.Y., Zhang, L., Zhang, J., Wang, H., Guo, Y.L.: Ahighly sensitive isotope-coded derivatization method and its applicationfor the mass spectrometric analysis of analytes containing the carboxylgroup. Anal. Chim. Acta. 758, 114–121 (2013)

8. Mochizuki, T., Todoroki, K., Inoue, K., Min, J.Z., Toyo’oka, T.: Isotopicvariants of light and heavy L-pyroglutamic acid succinimidyl esters as thederivatization reagents for DL-amino acid chiral metabolomics identifi-cation by liquid chromatography and electrospray ionization mass spec-trometry. Anal. Chim. Acta. 811, 51–59 (2014)

9. Tayyari, F., Gowda, G.A.N., Gu, H.W., Raftery, D.: N-15-cholamine-asmart isotope tag for combining NMR- and MS-based metabolite profil-ing. Anal. Chem. 85, 8715–8721 (2013)

10. Wong, J.M.T., Malec, P.A., Mabrouk, O.S., Ro, J., Dus, M., Kennedy,R.T.: Benzoyl chloride derivatization with liquid chromatography-massspectrometry for targeted metabolomics of neurochemicals in biologicalsamples. J. Chromatogr. A. 1446, 78–90 (2016)

11. Yuan, W., Edwards, J.L., Li, S.W.: Global profiling of carbonyl metab-olites with a photo-cleavable isobaric labeling affinity tag. Chem.Commun. 49, 11080–11082 (2013)

12. Zhou, R., Tseng, C.-L., Huan, T., Li, L.: IsoMS: automated processing ofLC-MS data generated by a chemical isotope labeling metabolomicsplatform. Anal. Chem. 86, 4675–4679 (2014)

13. Huan, T., Li, L.: Quantitative Metabolome analysis based on chromato-graphic peak reconstruction in chemical isotope labeling liquid chroma-tography mass spectrometry. Anal. Chem. 87, 7011–7016 (2015)

14. Han, W., Sapkota, S., Camicioli, R., Dixon, R.A., Li, L.: Profiling novelmetabolic biomarkers for Parkinson’s disease using in-depthmetabolomic analysis. Mov. Disord. 32, 1710–1728 (2017)

15. Huan, T., Li, L.: Counting missing values in a metabolite-intensity dataset for measuring the analytical performance of a metabolomics platform.Anal. Chem. 87, 1306–1313 (2015)

16. Zhang, N., Fountain, S.T., Bi, H., Rossi, D.T.: Quantification and rapidmetabolite identification in drug discovery using API time-of-flight LC/MS. Anal. Chem. 72, 800–806 (2000)

17. Eckers, C., Wolff, J.-C., Haskins, N.J., Sage, A.B., Giles, K., Bateman,R.: Accurate mass liquid chromatography/mass spectrometry on orthog-onal acceleration time-of-flight mass analyzers using switching betweenseparate sample and reference sprays. 1. Proof of concept. Anal. Chem.72, 3683–3688 (2000)

18. Schlosser, A., Volkmer-Engert, R.: Volatile polydimethylcyclosiloxanesin the ambient laboratory air identified as source of extreme backgroundsignals in nanoelectrospray mass spectrometry. J. Mass Spectrom. 38,523–525 (2003)

19. Barry, J.A., Robichaud, G., Muddiman, D.C.: Mass recalibration of FT-ICR mass spectrometry imaging data using the average frequency shift ofambient ions. J. Am. Soc. Mass Spectrom. 24, 1137–1145 (2013)

20. Olsen, J.V., de Godoy, L.M.F., Li, G., Macek, B., Mortensen, P., Pesch,R., Makarov, A., Lange, O., Horning, S., Mann, M.: Parts per millionmass accuracy on an orbitrap mass spectrometer via lock mass injectioninto a C-trap. Mol. Cell. Proteomics. 4, 2010–2021 (2005)