Research article Received: 8 January 2014
Revised: 25 June 2014
Accepted: 29 June 2014
Published online in Wiley Online Library
(wileyonlinelibrary.com) DOI 10.1002/jms.3428
Simultaneous screening of targeted and nontargeted contaminants using an LC-QTOF-MS system and automated MS/MS library searching† S. Herrera-Lopez,a,b M. D. Hernando,c E. García-Calvo,b A. R. Fernández-Albaa,b* and M. M. Ulaszewskad,b* Simultaneous high-resolution full-scan and tandem mass spectrometry (MS/MS) analysis using time of flight mass spectrometry brings an answer for increasing demand of retrospective and non-targeted data analysis. Such analysis combined with spectral library searching is a promising tool for targeted and untargeted screening of small molecules. Despite considerable extension of the panel of compounds of tandem mass spectral libraries, the heterogeneity of spectral data poses a major challenge against the effective usage of spectral libraries. Performance evaluation of available LC-MS/MS libraries will significantly increase credibility in the search results. The present work was aimed to evaluate fluctuation of MS/MS pattern, in the peak intensities distribution together with mass accuracy measurements, and in consequence, performance compliant with ion ratio and mass error criteria as principles in identification processes for targeted and untargeted contaminants at trace levels. Matrix effect and ultra-trace levels of concentration (from 50 ng l 1 to 1000 ng l 1) were evaluated as potential source of inaccuracy in the performance of spectral matching. Matrix-matched samples and real samples were screened for proof of applicability. By manual review of data and application of ion ratio and ppm error criteria, false negatives were obtained; this number diminished when in-house library was used, while with on-line MS/MS databases 100% of positive samples were found. In our experience, intensity of peaks across spectra was highly correlated to the concentration effect and matrix complexity. In turn, analysis of spectra acquired at trace concentrations and in different matrices results in better performance in providing correct and reliable identification. Copyright © 2014 John Wiley & Sons, Ltd. Keywords: LC-MS/MS library searching; screening of targets and non-targets; trace analysis; matrix effect; relative intensity of product ions
Introduction
878
In recent years, there is a growing movement to analyze contaminants beyond target compound lists and a shift towards non-targeted or general unknown screening, independently on matrix (water, food and biological fluids) and compounds (pesticides, metabolites and (bio)transformation products). The ability to perform targeted screening on a routine basis is made possible through advancements in liquid chromatography–mass spectrometry (LC-MS) instrumentation, like the triple quadrupole (QQQ) or hybrid triple–quadrupole linear ion trap mass spectrometers, which typically offer high sensitivity (low-attomolar) and broad dynamic range. Today’s, state-of-the-art equipment permit for quantification of targets and screening of untargeted compounds in only one run by using QToF (quadrupole/time of flight mass analyzer) due to high-resolution and mass accuracy in full-scan and MS/MS experiments. The quantification of targets is usually based on mass information derived from tandem mass spectrometry data.[1–3] However, while qualitative information in the MS/MS spectra, product ions’ m/z value is essential the product ions’ intensity is significant as well to avoid false-positive identification.[4] Existing requirements from relevant European regulation dictate maximum permitted tolerances for relative ion intensities with respect to the standard.[5–7] Ion suppression is a key factor contributing to false-positive identification, as
J. Mass Spectrom. 2014, 49, 878–893
documented by a number of studies. Moreover, likelihood of false-positives may tend to be greater at trace concentration levels.[4,8,9] A larger variability of the intensity ratio leads to a questionable reproducibility and caution is required in analyte identification. In trace contaminant analysis, to ensure adequate
* Correspondence to: M. M. Ulaszewska, Fondazione Edmund Mach, Research and Innovation Center, Department of Food Quality and Nutrition, Via E.Mach 1, San Michele all’Adige, Italy. E-mail: [email protected] * Correspondence to: A. R. Fernández-Alba, IMDEA-Water (Madrid Institute for Advanced Studies-Water), University of Alcalá, 28805 Alcalá de Henares, Madrid, Spain. E-mail: [email protected] †
This article is part of the Journal of Mass Spectrometry special issue entitled “3rd MS Food Day” edited by Gianluca Giorgi.
a Pesticide Residues Research Group, European Union Reference Laboratory (EURL), Department of Chemistry and Physics, University of Almería, 04120 La Cañada de San Urbano, Almería, Spain b IMDEA-Water (Madrid Institute for Advanced Studies-Water), University of Alcalá, 28805 Alcalá de Henares, Madrid, Spain c National Institute for Agricultural and Food Research and Technology - INIA, 28040 Madrid, Spain d Fondazione Edmund Mach, Research and Innovation Center, Department of Food Quality and Nutrition, Via E.Mach 1, San Michele all’Adige, Italy
Copyright © 2014 John Wiley & Sons, Ltd.
Screening of targeted and non-targeted contaminants
J. Mass Spectrom. 2014, 49, 878–893
across broad concentration range, in standard and in matrix. Confidence in search results, in terms of reliability of ion ratios and mass accuracy, was assessed for its applicability in trace analysis of targets in real samples (i.e. effluent wastewater). The present work was also aimed to evaluate experimental data, to devise reproducible fragmentation pattern and confidence for automated identification of non-targeted analytes in MS/MS library searching.
Experimental Chemicals and reagents The following pharmaceuticals selected for this study were purchased from Sigma-Aldrich (Steinheim, Germany) with purity higher than 98%: trimethoprim, N-acetyl-4-aminoantipiryne (4AAA), citalopram, loratadine, fenofibric acid, ofloxacin and venlafaxine. Methanol and HPLC-grade acetonitrile were supplied by Merck (Darmstadt, Germany). Water used for LC-MS analysis was generated from a Direct-Q™ 5 Ultrapure Water system (Millipore Bedford, MA, USA). Working solutions of single chemicals and optimization of ion-source-dependent parameters were described elsewhere.[18] Sample preparation and LC-ESI-QTOF-MS/MS analysis In this study, a sample volume of 100 μl of river waters was analyzed, by direct injection; therefore, samples only underwent simple handling. Analytical method was developed and validated previously;[14,17,18] scheme of acquisition and processing of data is shown in Fig. 1. Briefly, sample preparation in direct injection only consists in sample (900 μl) dilution with internal standards in acetonitrile (100 μl) prior to filtration in 0.45-μm polytetrafluoroethylene filter. Labeled caffeine D5 was used as internal standard at a concentration of 100 μg l 1. Data acquisition was performed with a hybrid quadrupole timeof-flight mass spectrometer, TripleTOF 5600 System (AB SCIEX, Concord, ON, Canada) connected to an Agilent 1200 Series HPLC system (Agilent Technologies, Wilmington, DE, USA) with an electrospray interface (ESI), LC-ESI-QTOF-MS. The ion source parameters as well as the selected criteria for IDA (information dependent acquisition) operation mode were described in detail in previous studies,[14,17,18] and they are reported in Fig. 1. Concisely, the method consisted of a TOF MS survey scan followed by five IDA TOF MS/MS scans. The mass range was m/z 50–800, for both MS and MS/MS experiments. An accumulation time of 75 ms was used for each scan. Dynamic background subtraction was applied for IDA criteria, and a collision energy of 30 eV with a spread of ±10 eV was used for MS/SM scans.[18] A HPLC binary solvent delivery system equipped with a RP-C-8 analytical column (150 mm length × 4.6 mm I.D and 5-μm particle size; Zorbax Eclipse XDB-C8, Agilent Technologies, Wilmington, DE, USA) was used. Mobile phases A and B were, respectively, acetonitrile and HPLC-grade water with 0.1% formic acid present in both phases. The flow rate was 500 μl min 1. A linear gradient was set from 10% to 100% of acetonitrile in 12 min and then maintained at 100% for 5 min.[14,17,18] The data acquisition and processing in the LC-ESI-QTOF-MS system were carried out using Analyst® TF 1.5 and PeakViewTM (AB SCIEX) software. PeakView™ incorporates tools to display, filter and process IDA data. PeakView™ add-in tool ‘XIC Manager’ was used for targeted and non-targeted data processing. The XIC Manager consists of a table for defining a list of masses or formulae to generate extracted ion chromatograms (XIC) and the ability to review results for the identification of detected compounds; moreover, it allows for exploring and
Copyright © 2014 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/jms
879
processing of LC-MS data and effective application of MS/MS commercial libraries, accuracy of comparative analysis with standard spectra is critical. In library searches for target analytes, the spectral fit is obtained by matching a MS/MS pattern from the library to an experimental MS/MS spectrum, ideally at similar conditions of collision-induced decomposition (CID). To some extent, libraries, so far, are not catching up with the advances in this technology; the limited coverage of set of operating conditions as well as the availability of high-quality reference spectra is still a problem encountered in tandem mass spectral library search.[10] The most advanced mass spectrometers provide researchers tools with enhanced capabilities for identification and characterization. Approaches such those relied on accurate mass measurement, in both full scan and MS/MS modes, confer enhanced confidence and reliability in the identification of a finding.[3,11–13] However, compound identification may also be hindered by a fluctuation of MS/MS pattern, in the peak intensity distribution, depending on concentration level of the analyte especially, at ultra-trace level. In practice, professional judgment is often needed to assess whether the preliminary software finding is correct or not. For example, in pesticide residue analysis, experimental approach based on the determination of the variability of the ion ratios over time, using calibration standards to devise performance-based criteria, is recommended.[6,7] Another difficulty in LC-MS identification relates to non-target analytes in screening procedures. In water analysis, structural elucidation has been greatly facilitated by means of the determination of both, molecular formulae and product ion formulae, with minimal uncertainty using QToF.[14] In this case, the degree of similarity of the most abundant product ions in the resulting spectra is also of great importance to gain confidence in the identification. Seeking such knowledge, assessing product ion intensity fluctuation, will aid in determining the reliability of ion ratios to be used for identification. Recently, approaches to identify unknowns by aligning fragmentation trees and throughput computational analysis are being developed to overcome limitations of spectral libraries.[15,16] The automated interpretation of such data, however, is in its early stages. And, its potential applicability for large-scale compound screens needs rigorous proofs, since real fragmentation patterns of unknown compounds may not necessarily follow the known rules of fragmentation. This article shows the outputs of a simultaneous quantitative and qualitative simple and easy-to-reproduce screening of water contaminants in a single analysis. Combination of simultaneous TOF-MS and MS/MS analysis brings an answer for increasing demand of retrospective and non-targeted data analysis. However, screening methods require application of compromise conditions, such as collision energy or declustering potential rewarding as much compounds as possible, furthermore setting such as isolation mass window in collision cell, additionally affect quality of MS/MS data. Therefore, the interpretation of product ion spectra needs particular attention, especially when dealing with matrices of different complexity. The variability of the relative intensities (RI) and mass accuracy measurement in product ion spectra obtained using LC-ESI-QTOF-MS/MS (liquid chromatograph-electrospray ionization source with a hybrid QToF) were determined in order to assess the reliability of identification criteria: ion ratios and ppm errors for the trace analysis of both target and non-target water contaminants in an automated MS/MS library searching. Compounds not included ‘a priori’ in the analytical method were considered as non-targeted. The variability of the RI as well as ppm errors of the product ions (and ppm error for molecular ion) was calculated over inter-day analysis
S. Herrera et al.
Figure 1. Acquisition and data processing diagram.
processing mass spectral data using the parameters of LC retention times, accurate mass (error) and purity score obtained by matching an experimental MS/MS spectrum against library MS/MS spectrum. ABSciex commercial library of ca 1200 compounds, mainly pharmaceutical and personal care products and pesticides was used for simultanous targeted and untargeted identifications Identification tools for targeted and untargeted analysis
880
The complexity of data acquired within high resolution and mass accuracy in MS and MS/MS mode requires powerful data mining tools. We evaluated quality of MS/MS spectra, its fluctuations and variability in different matrices at different concentrations, and we applied different identification tools to verify identification reproducibility and accuracy, (see Fig. 1 representing overall workflow). We used different approaches: manual revision, inhouse library and on-line MS/MS repositories for analysis of targeted and untargeted compounds.
wileyonlinelibrary.com/journal/jms
The in-house library, consisting of a number of targeted analytes with MS/MS spectra acquired at different CE, allows the rapid and high-confidence identification of the experimentally detected molecules based on a multiparameter matching. MS/MS spectra included in library, provided by ABSciex, were acquired with QQQ detector at a CE of 30 eV in positive ionization mode. Additional MS/MS spectra of other target compounds were added to improve matching. These compounds were injected at different CE: 20, 30 and 45 eV and at four concentration levels: 1 μg l 1, 10 μg l 1, 50 μg l 1 and 1 mg l 1. The mass tolerance was set to ±0.01 Da for the precursor ions. As result of the MS/MS analysis, the lower-voltage spectrum was dominated by product ions of high m/z values including a predominant pseudomolecular ion, whereas the higher-voltage spectrum contained predominantly ions of low m/z values. Thus, the combination of both low- and high-voltage mass spectra maximized the information content of the library entry. Additionally, MS/MS spectra acquired at different concentration levels reflect
Copyright © 2014 John Wiley & Sons, Ltd.
J. Mass Spectrom. 2014, 49, 878–893
Screening of targeted and non-targeted contaminants better the MS/MS fragmentation behavior in the analysis of real samples where no information is available about concentration of compounds. During manual identification, several criteria were assumed: samples were considered as positive when ppm error did not exceed 5 and 10 ppm in both MS and MS/MS modes, respectively. Furthermore, in MS/MS mode, samples were considered as positive, when variability of RI did not exceed ±20% for ions with RI >50%; ±25% for ions with RI >20–50%; ±30% for ions with RI >10–20% and ±30% for ions 20–50%, ±25%; for ions >10–20%, ±30% and for ions 10 ppm) LC:MassBank: HC: 100%; MC: 100%; LC: HMDB: HC: 0%; MC: 0%; LC: Metlin: HC: 100%; MC: 100%; LC: -
Manual review
HC:100% MC:100% LC:100%
HC:100% MC:100% LC:80%
MC: 80% (Mean:86.9; CV: 7.11) LC: Ofloxacin Rt = 4.6
HC: 75% MC:100% LC: -
HC: 100% Mean: 79.2; CV: 6.25 MC: 66% Mean: 59.0; CV: 37.7 LC: -
Venlafaxine Rt = 5.9
Real samples (effluent wastewater, n = 11 samples b
HC:100% MC:100% LC: 60%
HC: 100% (Mean: 97.0; CV: 1.9) MC: 100% (Mean: 96.6; CV: 3.0) LC: 100% (Mean:71.5; CV:23.5)
Manual review %
MassBank HC: 100%; MC: 100%; LC: HMDB: HC: 0%; MC: 0%; LC: Metlin: HC:100%; MC:100%; LC:100% (40% with window >10 ppm) MassBank: HC: 100%; MC: 100%; LC: 100% HMDB: HC: 0%; MC: 0%; LC: 0%
83%
a
In-house library searching b 83% Mean: 89.9 CV: 7.6
On-line library searching b
Metlin: 100% (in 17% with tolerance window >10 ppm) MassBank 100% HMDB: 0%
91%
100% Mean: 88.6 CV: 2.5
Metlin: 0%. MassBank: 100% HMDB: 0%
40%
100% Mean: 77.9 CV: 10.5
Metlin 100% (in 10% with tolerance window >10 ppm) MassBank: 100% HMDB: 0%
40%
100% Mean: 87.3 CV: 6.3
Metlin: 100% (40% with window >10 ppm) MassBank: 100% HMDB: 0%
66.7%
100% Mean: 79.2 CV: 9.6
Metlin 100% (22% with window >10 ppm) MassBank: 100% HMDB: 0%
50%
100% Mean: 82.6 CV: 12.4
Metlin 100% (for 37.5% window >10 ppm) MassBank: 100% HMDB: 0%
54%
100% Mean: 75.7 CV:22
Metlin: 100% (54.5% with window >10 ppm) MassBank 100% HMDB: 0%
1
Abbreviations: 4AAA: N-acetyl-4-aminoantipiryne; CV: coefficient of variation; RI: relative intensity; LC: low concentration level (50 ng l ); MC: medium concentration 1 1 level (500 ng l ) and HC: high concentration level (1000 ng l ). a Manual review identification in %, means percentage of positive samples meeting both mass accuracy and ion ratio criterion. For mass accuracy criterion in MS/MS mode, at least four ions were considered. b Searching with online and in-house libraries (hits expressed in %) was performed for effluent samples where MS/MS spectrum was available: Trimetroprim 55% of samples; 4-AAA-100% of samples; Citalopram 91% of samples; Loratadine 91% of samples; Fenofibric acid 82% of samples; Ofloxacin 73% of samples and Loratadine 91% of samples.
887
J. Mass Spectrom. 2014, 49, 878–893
Copyright © 2014 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/jms
S. Herrera et al.
888
standard: ppm errors are calculated for each product ion, and intensity ratio are calculated between sample and standard. This is a hard and time-consuming work especially, considering huge amount of data obtained within full scan screening which otherwise can be facilitated by using on line database comparison with global MS/MS depositories such as MassBank, METLIN or HMDB, and use of inhouse libraries. Acquired MS/MS spectra of spiked matrix samples and effluent samples (trimetroprim 54.5%; 4-AAA-100%; citalopram 91%; loratadine 91%; fenofibric acid 82%, ofloxacin 80% and venlafaxine 91%) were submitted to on line depositories to identify and confirm the presence of pharmaceuticals in samples. Calculation of positive matching as well as manual review and in-house library matching was performed on similar way considering only samples with acquired MS/MS spectra. Information introduced to MassBank within ‘search by peak’ mode were m/z value of at least four product ions and their RI. In Metlin and HMDB libraries; additionally, tolerance windows for mass accuracy measurement could be set and precursor ion clearly indicated. Therefore, windows of tolerance were set for 0.02 Da for product ions and 10 ppm for precursor ion. Feedbacks given by MassBank for spiked river water at all concentration levels and effluent samples were satisfactory. For each introduced MS/MS spectrum, several proposal were received with the highest score value for our targeted compound. Selection of best hits among listed proposals was based on score value and knowledge of precursor, not selected a priori (see Table 2). It is worth to mention that correct proposals were obtained for all spiked river and effluent samples, not only with absence of information about precursor ion, but also with no mass tolerance set. Metlin feedback was also satisfactory for all spiked river and effluent samples with an important difference; that list of proposals was limited to one— our targeted compound. Key parameters for correct searching were tolerance mass set for both, precursor ion and product ions. Indeed, for LC spectra where precursor ions, for some compounds, were measured with ppm error higher than 10 ppm (4-AAA, loratadine, fenofibric acid and venflaxin), wider tolerance windows up to 20 ppm were required to get correct hits. Furthermore, appropriate proposals for targeted compounds were obtained regardless the number of ions introduced to engine search, identification was correct even for mass spectra where some product ions were absent (see Fig. 2 spectrum of trimethoprim from an effluent wastewater sample). Comparison with HMDB did not bring any results as targeted compounds of our interest were not presented in MS/MS database.The library search results, obtained with the extended version of customer database, were satisfactory for all targeted analytes at all concentrations, with purity score higher than 70% (high level of confidence), except for fenofibric acid and ofloxacin, which didn’t show any MS/MS spectra at low concentration. In real samples, targets were identified in 80–100% of cases, except trimethoprim, for which matching was satisfactory for 66% of samples. Scores obtained were always in a range of 77–100%, providing high confidence in identification. Most robust and time-consuming identification, manual review, was preformed considering two criterias: ppm error in TOF-MS (
Revised: 25 June 2014
Accepted: 29 June 2014
Published online in Wiley Online Library
(wileyonlinelibrary.com) DOI 10.1002/jms.3428
Simultaneous screening of targeted and nontargeted contaminants using an LC-QTOF-MS system and automated MS/MS library searching† S. Herrera-Lopez,a,b M. D. Hernando,c E. García-Calvo,b A. R. Fernández-Albaa,b* and M. M. Ulaszewskad,b* Simultaneous high-resolution full-scan and tandem mass spectrometry (MS/MS) analysis using time of flight mass spectrometry brings an answer for increasing demand of retrospective and non-targeted data analysis. Such analysis combined with spectral library searching is a promising tool for targeted and untargeted screening of small molecules. Despite considerable extension of the panel of compounds of tandem mass spectral libraries, the heterogeneity of spectral data poses a major challenge against the effective usage of spectral libraries. Performance evaluation of available LC-MS/MS libraries will significantly increase credibility in the search results. The present work was aimed to evaluate fluctuation of MS/MS pattern, in the peak intensities distribution together with mass accuracy measurements, and in consequence, performance compliant with ion ratio and mass error criteria as principles in identification processes for targeted and untargeted contaminants at trace levels. Matrix effect and ultra-trace levels of concentration (from 50 ng l 1 to 1000 ng l 1) were evaluated as potential source of inaccuracy in the performance of spectral matching. Matrix-matched samples and real samples were screened for proof of applicability. By manual review of data and application of ion ratio and ppm error criteria, false negatives were obtained; this number diminished when in-house library was used, while with on-line MS/MS databases 100% of positive samples were found. In our experience, intensity of peaks across spectra was highly correlated to the concentration effect and matrix complexity. In turn, analysis of spectra acquired at trace concentrations and in different matrices results in better performance in providing correct and reliable identification. Copyright © 2014 John Wiley & Sons, Ltd. Keywords: LC-MS/MS library searching; screening of targets and non-targets; trace analysis; matrix effect; relative intensity of product ions
Introduction
878
In recent years, there is a growing movement to analyze contaminants beyond target compound lists and a shift towards non-targeted or general unknown screening, independently on matrix (water, food and biological fluids) and compounds (pesticides, metabolites and (bio)transformation products). The ability to perform targeted screening on a routine basis is made possible through advancements in liquid chromatography–mass spectrometry (LC-MS) instrumentation, like the triple quadrupole (QQQ) or hybrid triple–quadrupole linear ion trap mass spectrometers, which typically offer high sensitivity (low-attomolar) and broad dynamic range. Today’s, state-of-the-art equipment permit for quantification of targets and screening of untargeted compounds in only one run by using QToF (quadrupole/time of flight mass analyzer) due to high-resolution and mass accuracy in full-scan and MS/MS experiments. The quantification of targets is usually based on mass information derived from tandem mass spectrometry data.[1–3] However, while qualitative information in the MS/MS spectra, product ions’ m/z value is essential the product ions’ intensity is significant as well to avoid false-positive identification.[4] Existing requirements from relevant European regulation dictate maximum permitted tolerances for relative ion intensities with respect to the standard.[5–7] Ion suppression is a key factor contributing to false-positive identification, as
J. Mass Spectrom. 2014, 49, 878–893
documented by a number of studies. Moreover, likelihood of false-positives may tend to be greater at trace concentration levels.[4,8,9] A larger variability of the intensity ratio leads to a questionable reproducibility and caution is required in analyte identification. In trace contaminant analysis, to ensure adequate
* Correspondence to: M. M. Ulaszewska, Fondazione Edmund Mach, Research and Innovation Center, Department of Food Quality and Nutrition, Via E.Mach 1, San Michele all’Adige, Italy. E-mail: [email protected] * Correspondence to: A. R. Fernández-Alba, IMDEA-Water (Madrid Institute for Advanced Studies-Water), University of Alcalá, 28805 Alcalá de Henares, Madrid, Spain. E-mail: [email protected] †
This article is part of the Journal of Mass Spectrometry special issue entitled “3rd MS Food Day” edited by Gianluca Giorgi.
a Pesticide Residues Research Group, European Union Reference Laboratory (EURL), Department of Chemistry and Physics, University of Almería, 04120 La Cañada de San Urbano, Almería, Spain b IMDEA-Water (Madrid Institute for Advanced Studies-Water), University of Alcalá, 28805 Alcalá de Henares, Madrid, Spain c National Institute for Agricultural and Food Research and Technology - INIA, 28040 Madrid, Spain d Fondazione Edmund Mach, Research and Innovation Center, Department of Food Quality and Nutrition, Via E.Mach 1, San Michele all’Adige, Italy
Copyright © 2014 John Wiley & Sons, Ltd.
Screening of targeted and non-targeted contaminants
J. Mass Spectrom. 2014, 49, 878–893
across broad concentration range, in standard and in matrix. Confidence in search results, in terms of reliability of ion ratios and mass accuracy, was assessed for its applicability in trace analysis of targets in real samples (i.e. effluent wastewater). The present work was also aimed to evaluate experimental data, to devise reproducible fragmentation pattern and confidence for automated identification of non-targeted analytes in MS/MS library searching.
Experimental Chemicals and reagents The following pharmaceuticals selected for this study were purchased from Sigma-Aldrich (Steinheim, Germany) with purity higher than 98%: trimethoprim, N-acetyl-4-aminoantipiryne (4AAA), citalopram, loratadine, fenofibric acid, ofloxacin and venlafaxine. Methanol and HPLC-grade acetonitrile were supplied by Merck (Darmstadt, Germany). Water used for LC-MS analysis was generated from a Direct-Q™ 5 Ultrapure Water system (Millipore Bedford, MA, USA). Working solutions of single chemicals and optimization of ion-source-dependent parameters were described elsewhere.[18] Sample preparation and LC-ESI-QTOF-MS/MS analysis In this study, a sample volume of 100 μl of river waters was analyzed, by direct injection; therefore, samples only underwent simple handling. Analytical method was developed and validated previously;[14,17,18] scheme of acquisition and processing of data is shown in Fig. 1. Briefly, sample preparation in direct injection only consists in sample (900 μl) dilution with internal standards in acetonitrile (100 μl) prior to filtration in 0.45-μm polytetrafluoroethylene filter. Labeled caffeine D5 was used as internal standard at a concentration of 100 μg l 1. Data acquisition was performed with a hybrid quadrupole timeof-flight mass spectrometer, TripleTOF 5600 System (AB SCIEX, Concord, ON, Canada) connected to an Agilent 1200 Series HPLC system (Agilent Technologies, Wilmington, DE, USA) with an electrospray interface (ESI), LC-ESI-QTOF-MS. The ion source parameters as well as the selected criteria for IDA (information dependent acquisition) operation mode were described in detail in previous studies,[14,17,18] and they are reported in Fig. 1. Concisely, the method consisted of a TOF MS survey scan followed by five IDA TOF MS/MS scans. The mass range was m/z 50–800, for both MS and MS/MS experiments. An accumulation time of 75 ms was used for each scan. Dynamic background subtraction was applied for IDA criteria, and a collision energy of 30 eV with a spread of ±10 eV was used for MS/SM scans.[18] A HPLC binary solvent delivery system equipped with a RP-C-8 analytical column (150 mm length × 4.6 mm I.D and 5-μm particle size; Zorbax Eclipse XDB-C8, Agilent Technologies, Wilmington, DE, USA) was used. Mobile phases A and B were, respectively, acetonitrile and HPLC-grade water with 0.1% formic acid present in both phases. The flow rate was 500 μl min 1. A linear gradient was set from 10% to 100% of acetonitrile in 12 min and then maintained at 100% for 5 min.[14,17,18] The data acquisition and processing in the LC-ESI-QTOF-MS system were carried out using Analyst® TF 1.5 and PeakViewTM (AB SCIEX) software. PeakView™ incorporates tools to display, filter and process IDA data. PeakView™ add-in tool ‘XIC Manager’ was used for targeted and non-targeted data processing. The XIC Manager consists of a table for defining a list of masses or formulae to generate extracted ion chromatograms (XIC) and the ability to review results for the identification of detected compounds; moreover, it allows for exploring and
Copyright © 2014 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/jms
879
processing of LC-MS data and effective application of MS/MS commercial libraries, accuracy of comparative analysis with standard spectra is critical. In library searches for target analytes, the spectral fit is obtained by matching a MS/MS pattern from the library to an experimental MS/MS spectrum, ideally at similar conditions of collision-induced decomposition (CID). To some extent, libraries, so far, are not catching up with the advances in this technology; the limited coverage of set of operating conditions as well as the availability of high-quality reference spectra is still a problem encountered in tandem mass spectral library search.[10] The most advanced mass spectrometers provide researchers tools with enhanced capabilities for identification and characterization. Approaches such those relied on accurate mass measurement, in both full scan and MS/MS modes, confer enhanced confidence and reliability in the identification of a finding.[3,11–13] However, compound identification may also be hindered by a fluctuation of MS/MS pattern, in the peak intensity distribution, depending on concentration level of the analyte especially, at ultra-trace level. In practice, professional judgment is often needed to assess whether the preliminary software finding is correct or not. For example, in pesticide residue analysis, experimental approach based on the determination of the variability of the ion ratios over time, using calibration standards to devise performance-based criteria, is recommended.[6,7] Another difficulty in LC-MS identification relates to non-target analytes in screening procedures. In water analysis, structural elucidation has been greatly facilitated by means of the determination of both, molecular formulae and product ion formulae, with minimal uncertainty using QToF.[14] In this case, the degree of similarity of the most abundant product ions in the resulting spectra is also of great importance to gain confidence in the identification. Seeking such knowledge, assessing product ion intensity fluctuation, will aid in determining the reliability of ion ratios to be used for identification. Recently, approaches to identify unknowns by aligning fragmentation trees and throughput computational analysis are being developed to overcome limitations of spectral libraries.[15,16] The automated interpretation of such data, however, is in its early stages. And, its potential applicability for large-scale compound screens needs rigorous proofs, since real fragmentation patterns of unknown compounds may not necessarily follow the known rules of fragmentation. This article shows the outputs of a simultaneous quantitative and qualitative simple and easy-to-reproduce screening of water contaminants in a single analysis. Combination of simultaneous TOF-MS and MS/MS analysis brings an answer for increasing demand of retrospective and non-targeted data analysis. However, screening methods require application of compromise conditions, such as collision energy or declustering potential rewarding as much compounds as possible, furthermore setting such as isolation mass window in collision cell, additionally affect quality of MS/MS data. Therefore, the interpretation of product ion spectra needs particular attention, especially when dealing with matrices of different complexity. The variability of the relative intensities (RI) and mass accuracy measurement in product ion spectra obtained using LC-ESI-QTOF-MS/MS (liquid chromatograph-electrospray ionization source with a hybrid QToF) were determined in order to assess the reliability of identification criteria: ion ratios and ppm errors for the trace analysis of both target and non-target water contaminants in an automated MS/MS library searching. Compounds not included ‘a priori’ in the analytical method were considered as non-targeted. The variability of the RI as well as ppm errors of the product ions (and ppm error for molecular ion) was calculated over inter-day analysis
S. Herrera et al.
Figure 1. Acquisition and data processing diagram.
processing mass spectral data using the parameters of LC retention times, accurate mass (error) and purity score obtained by matching an experimental MS/MS spectrum against library MS/MS spectrum. ABSciex commercial library of ca 1200 compounds, mainly pharmaceutical and personal care products and pesticides was used for simultanous targeted and untargeted identifications Identification tools for targeted and untargeted analysis
880
The complexity of data acquired within high resolution and mass accuracy in MS and MS/MS mode requires powerful data mining tools. We evaluated quality of MS/MS spectra, its fluctuations and variability in different matrices at different concentrations, and we applied different identification tools to verify identification reproducibility and accuracy, (see Fig. 1 representing overall workflow). We used different approaches: manual revision, inhouse library and on-line MS/MS repositories for analysis of targeted and untargeted compounds.
wileyonlinelibrary.com/journal/jms
The in-house library, consisting of a number of targeted analytes with MS/MS spectra acquired at different CE, allows the rapid and high-confidence identification of the experimentally detected molecules based on a multiparameter matching. MS/MS spectra included in library, provided by ABSciex, were acquired with QQQ detector at a CE of 30 eV in positive ionization mode. Additional MS/MS spectra of other target compounds were added to improve matching. These compounds were injected at different CE: 20, 30 and 45 eV and at four concentration levels: 1 μg l 1, 10 μg l 1, 50 μg l 1 and 1 mg l 1. The mass tolerance was set to ±0.01 Da for the precursor ions. As result of the MS/MS analysis, the lower-voltage spectrum was dominated by product ions of high m/z values including a predominant pseudomolecular ion, whereas the higher-voltage spectrum contained predominantly ions of low m/z values. Thus, the combination of both low- and high-voltage mass spectra maximized the information content of the library entry. Additionally, MS/MS spectra acquired at different concentration levels reflect
Copyright © 2014 John Wiley & Sons, Ltd.
J. Mass Spectrom. 2014, 49, 878–893
Screening of targeted and non-targeted contaminants better the MS/MS fragmentation behavior in the analysis of real samples where no information is available about concentration of compounds. During manual identification, several criteria were assumed: samples were considered as positive when ppm error did not exceed 5 and 10 ppm in both MS and MS/MS modes, respectively. Furthermore, in MS/MS mode, samples were considered as positive, when variability of RI did not exceed ±20% for ions with RI >50%; ±25% for ions with RI >20–50%; ±30% for ions with RI >10–20% and ±30% for ions 20–50%, ±25%; for ions >10–20%, ±30% and for ions 10 ppm) LC:MassBank: HC: 100%; MC: 100%; LC: HMDB: HC: 0%; MC: 0%; LC: Metlin: HC: 100%; MC: 100%; LC: -
Manual review
HC:100% MC:100% LC:100%
HC:100% MC:100% LC:80%
MC: 80% (Mean:86.9; CV: 7.11) LC: Ofloxacin Rt = 4.6
HC: 75% MC:100% LC: -
HC: 100% Mean: 79.2; CV: 6.25 MC: 66% Mean: 59.0; CV: 37.7 LC: -
Venlafaxine Rt = 5.9
Real samples (effluent wastewater, n = 11 samples b
HC:100% MC:100% LC: 60%
HC: 100% (Mean: 97.0; CV: 1.9) MC: 100% (Mean: 96.6; CV: 3.0) LC: 100% (Mean:71.5; CV:23.5)
Manual review %
MassBank HC: 100%; MC: 100%; LC: HMDB: HC: 0%; MC: 0%; LC: Metlin: HC:100%; MC:100%; LC:100% (40% with window >10 ppm) MassBank: HC: 100%; MC: 100%; LC: 100% HMDB: HC: 0%; MC: 0%; LC: 0%
83%
a
In-house library searching b 83% Mean: 89.9 CV: 7.6
On-line library searching b
Metlin: 100% (in 17% with tolerance window >10 ppm) MassBank 100% HMDB: 0%
91%
100% Mean: 88.6 CV: 2.5
Metlin: 0%. MassBank: 100% HMDB: 0%
40%
100% Mean: 77.9 CV: 10.5
Metlin 100% (in 10% with tolerance window >10 ppm) MassBank: 100% HMDB: 0%
40%
100% Mean: 87.3 CV: 6.3
Metlin: 100% (40% with window >10 ppm) MassBank: 100% HMDB: 0%
66.7%
100% Mean: 79.2 CV: 9.6
Metlin 100% (22% with window >10 ppm) MassBank: 100% HMDB: 0%
50%
100% Mean: 82.6 CV: 12.4
Metlin 100% (for 37.5% window >10 ppm) MassBank: 100% HMDB: 0%
54%
100% Mean: 75.7 CV:22
Metlin: 100% (54.5% with window >10 ppm) MassBank 100% HMDB: 0%
1
Abbreviations: 4AAA: N-acetyl-4-aminoantipiryne; CV: coefficient of variation; RI: relative intensity; LC: low concentration level (50 ng l ); MC: medium concentration 1 1 level (500 ng l ) and HC: high concentration level (1000 ng l ). a Manual review identification in %, means percentage of positive samples meeting both mass accuracy and ion ratio criterion. For mass accuracy criterion in MS/MS mode, at least four ions were considered. b Searching with online and in-house libraries (hits expressed in %) was performed for effluent samples where MS/MS spectrum was available: Trimetroprim 55% of samples; 4-AAA-100% of samples; Citalopram 91% of samples; Loratadine 91% of samples; Fenofibric acid 82% of samples; Ofloxacin 73% of samples and Loratadine 91% of samples.
887
J. Mass Spectrom. 2014, 49, 878–893
Copyright © 2014 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/jms
S. Herrera et al.
888
standard: ppm errors are calculated for each product ion, and intensity ratio are calculated between sample and standard. This is a hard and time-consuming work especially, considering huge amount of data obtained within full scan screening which otherwise can be facilitated by using on line database comparison with global MS/MS depositories such as MassBank, METLIN or HMDB, and use of inhouse libraries. Acquired MS/MS spectra of spiked matrix samples and effluent samples (trimetroprim 54.5%; 4-AAA-100%; citalopram 91%; loratadine 91%; fenofibric acid 82%, ofloxacin 80% and venlafaxine 91%) were submitted to on line depositories to identify and confirm the presence of pharmaceuticals in samples. Calculation of positive matching as well as manual review and in-house library matching was performed on similar way considering only samples with acquired MS/MS spectra. Information introduced to MassBank within ‘search by peak’ mode were m/z value of at least four product ions and their RI. In Metlin and HMDB libraries; additionally, tolerance windows for mass accuracy measurement could be set and precursor ion clearly indicated. Therefore, windows of tolerance were set for 0.02 Da for product ions and 10 ppm for precursor ion. Feedbacks given by MassBank for spiked river water at all concentration levels and effluent samples were satisfactory. For each introduced MS/MS spectrum, several proposal were received with the highest score value for our targeted compound. Selection of best hits among listed proposals was based on score value and knowledge of precursor, not selected a priori (see Table 2). It is worth to mention that correct proposals were obtained for all spiked river and effluent samples, not only with absence of information about precursor ion, but also with no mass tolerance set. Metlin feedback was also satisfactory for all spiked river and effluent samples with an important difference; that list of proposals was limited to one— our targeted compound. Key parameters for correct searching were tolerance mass set for both, precursor ion and product ions. Indeed, for LC spectra where precursor ions, for some compounds, were measured with ppm error higher than 10 ppm (4-AAA, loratadine, fenofibric acid and venflaxin), wider tolerance windows up to 20 ppm were required to get correct hits. Furthermore, appropriate proposals for targeted compounds were obtained regardless the number of ions introduced to engine search, identification was correct even for mass spectra where some product ions were absent (see Fig. 2 spectrum of trimethoprim from an effluent wastewater sample). Comparison with HMDB did not bring any results as targeted compounds of our interest were not presented in MS/MS database.The library search results, obtained with the extended version of customer database, were satisfactory for all targeted analytes at all concentrations, with purity score higher than 70% (high level of confidence), except for fenofibric acid and ofloxacin, which didn’t show any MS/MS spectra at low concentration. In real samples, targets were identified in 80–100% of cases, except trimethoprim, for which matching was satisfactory for 66% of samples. Scores obtained were always in a range of 77–100%, providing high confidence in identification. Most robust and time-consuming identification, manual review, was preformed considering two criterias: ppm error in TOF-MS (
