Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of multi-platform-based CE-MS and LC-MS data

Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of... In mass spectrometry (MS)-based metabolomics, there is a great need to combine different analytical separation techniques to cover metabolites of different polarities and apply appropriate multi-platform data processing. Here, we introduce Ari- umMS (augmented region of interest for untargeted metabolomics mass spectrometry) as a reliable toolbox for multi-platform metabolomics. AriumMS offers augmented data analysis of several separation techniques utilizing a region-of-interest algorithm. To demonstrate the capabilities of AriumMS, five datasets were combined. This includes three newly developed capillary electrophoresis (CE)-Orbitrap MS methods using the recently introduced nanoCEasy CE-MS interface and two hydrophilic interaction liquid chromatography (HILIC)-Orbitrap MS methods. AriumMS provides a novel mid-level data fusion approach for multi-platform data analysis to simplify and speed up multi-platform data processing and evaluation. The key feature of AriumMS lies in the optimized data processing strategy, including parallel processing of datasets and flexible parameterization for processing of individual separation methods with different peak characteristics. As a case study, Saccharomyces cerevisiae (yeast) was treated with a growth inhibitor, and AriumMS successfully differentiated the metabolome based on the augmented multi-platform CE-MS and HILIC-MS investigation. As a result, AriumMS is pro- posed as a powerful tool to improve the accuracy and selectivity of metabolome analysis through the integration of several HILIC-MS/CE-MS techniques. Keywords Augmented data evaluation · Mid-level data fusion · Multi-platform metabolomics · nanoCEasy · Capillary electrophoresis · Hydrophilic interaction liquid chromatography Abbreviations DOE Design of experiment A.u. Arbitrary units EIC/E Extracted ion chromatogram/ AriumMS Augmented region of interest for untargeted electropherogram metabolomics mass spectrometryEOF Electro-osmotic flow BGE Background electrolytesESI Electrospray ionization BPC/E Base peak chromatogram/electropherogramFC Fold change CE Capillary electrophoresisGC Gas chromatography CWT Continuous wavelet transform HCD Higher-energy collisional dissociation DMSO Dimethyl sulfoxide HILIC Hydrophilic interaction liquid chromatography MS Mass spectrometry NAD Nicotinamide adenine dinucleotide Lukas Naumann and Adrian Haun contributed equally. PCA Principal component analysis * Christian Neusüß PC Principal component Christian.Neusuess@hs-aalen.de QTOF Quadrupole time of flight ROI Region of interest Department of Chemistry, Aalen University, Beethovenstraße 1, 73430 Aalen, Germany Vol.:(0123456789) 1 3 Naumann L. et al. RP-LC Reversed-phase liquid chromatography acidic BGEs for cation analysis [12–14] and basic BGEs for sd Standard deviation anion analysis [15–17]. In order to improve the sensitivity SD Synthetic dextrose minimal medium for metabolite analysis by CE-MS, nanoESI interfaces have SL Sheath liquid recently been used, including the porous tip interface [18]. Nanosheath–liquid interfaces are of high interest as well, due to additional flexibility and robustness [19]. Most recently, Introduction we introduced the nanoCEasy interface adding ease-of-use and the capability of valve functionality by the two-capillary Owing to the inherent chemical diversity and the large size approach (e.g., for capillary reconditioning between runs) of the metabolome, there is no universal technique that can [20–22]. be used to assess the entire metabolome, i.e., “one size does Metabolomics data evaluation is usually based on two not fit all” [1 , 2]. Nevertheless, multi-platform metabolomics major approaches: target and non-target data evaluation workflows based on mass spectrometry (MS) are able to [3]. Target-based data evaluation is hypothesis-driven and enhance metabolome coverage. focuses with high analytical sensitivity on standard mixtures Typically, scientists employ high-resolution electrospray for their assignment and interpretation, such as concentra- ionization–MS (ESI-MS) with the possibility of MS/MS tion and appearance [3, 23]. Non-target metabolomics is an experiments such as quadrupole time-of-flight (QTOF) exploratory, hypothesis-generating data evaluation workflow or Orbitrap MS [3–5]. Depending on the type of metabo- [3]. This approach is a common choice as a first step within lites to be measured (polar vs. nonpolar) and limitations a data evaluation, to capture and monitor a broad range of concerning time and sample amount, different separation molecular content and retrieve as much chemical informa- techniques can be applied for the analysis to expand the tion as possible without any prior knowledge [3]. Examples metabolome coverage [1]. These are reversed-phase liquid of multi-platform metabolomics can be found in previous chromatography (RP-LC) [3], hydrophilic interaction liq- publications [7, 24–28]. Most of them use a target/non-target uid chromatography (HILIC) [6], capillary electrophore- approach based on different data processing workflows for sis (CE) [7], and gas chromatography (GC) [8] coupled to each analytical platform. high-resolution MS [3, 9]. The analytical gold standard in Since LC-MS and CE-MS offer comprehensive infor- proteomics and metabolomics is RP-LC-MS because of its mation of the metabolome, a combined multi-platform extended dynamic concentration range, sensitivity, reten- non-targeted data evaluation based on a data fusion tion time reproducibility, and ease of use [6]. Since RP-LC approach offers a single chemometric result for enhanced does not retain very well a wide variety of highly polar and statistical prediction and metabolic coverage [29–31]. The ionizable metabolites, HILIC is a valuable alternative [6]. fusion of separation methods coupled with MS detection HILIC is driven by molecular interactions and the partition is challenging due to the multivariate nature of the data of analytes between the hydrophobic mobile phase and the (i.e., a very high variables-to-sample ratio, and shift in hydrophilic stationary phase [10]. Significant technologi- migration times during sequences) [31, 32]. Hence, an cal advances in HILIC over the last two decades, such as augmented data evaluation of comprehensive analytical the commercialization of dedicated HILIC columns, have workflows enhances the feature capacity by combining aided the implementation of HILIC in proteomics and different selectivities, thereby allowing a better charac- metabolomics [6]. Overall, this has resulted in significant terization of phenotypes. analytical improvements (e.g., sensitivity, analyte cover- Data augmentation describes the combination of sev- age, throughput, analysis speed, and resolution), and thus eral datasets into one. Three cases can be distinguished HILIC offers excellent opportunities for the analysis of [31, 33, 34]: Low-level fusion is applied before any data polar and/or ionizable metabolites [6]. reduction, and mid-level fusion after feature extraction, Since many metabolites, especially those of central car- whereas high-level data fusion combines models after bon metabolism, contain charged amino, hydroxyl, car- data analysis [35]. Mid-level data fusion is based on boxyl, and phosphate groups, they are especially suitable removing irrelevant information, such as artifacts and for CE-MS analysis [10]. Electrophoretic-driven separation noise, from each dataset. The resulting dimensionality approaches offer several advantages for the separation of reduction decreases computation time and can produce charged compounds, like efficient separation, high resolv - more robust models [36]. ing power, low solvent, and sample consumption. Since CE Region-of-interest (ROI) analysis is the approach separation is based on differences in ion mobilities [10, 11], of choice and significantly reduces both the amount different compositions of background electrolytes (BGE), of data—without loss of relevant information—and especially regarding pH, lead to different selectivities. In processing time. Only data points that have a mini- this way, CE-MS analysis has been frequently applied using mum intensity and a minimum abundance within the 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… measurement are included in an ROI [37]. Peaks are Yeast growth and sample preparation then detected, integrated, and labeled (m/z, retention time, peak area, and height) in the obtained ROIs. Vari- Production of yeast liquid cultures was carried out with ous filters (e.g., contaminant filter, isotope, and adduct Saccharomyces cerevisiae strain CEN.PK122 [42], start- filter) are then used to remove false-positive features ing from a single colony grown on SD plates. Growth took from the feature list. This method is widely used in the place in an incubation shaker in a 5 L baffle flask under web-based tools XCMS Online [38] and MetaboAnalyst controlled conditions (30 °C, 123 rpm, 16 h). SD medium [39] as well as in various software packages such as was used as a basal medium. The culture was split into two MetaboAnalystR [40] or the open-source MZmine [41]. cultures (Mock and Effect1) at 0.5 optical density. Cell line However, the focus of these programs is not on the aug- Effect1 (160 mL) contained 160 µL 35 mM halogenated mentation of different separation techniques. For exam- indole dilution in dimethyl sulfoxide (DMSO) to induce ple, it is not possible to select different preprocessing the effect. In order to determine the induced effect of the settings for different data, which is essential for different halogenated indole exactly, Mock (160 mL) as a reference separation systems. With XCMS Online and MZmine, was treated exactly the same as Effect1 (160 µL DMSO), files of different origins must be processed separately without adding the halogenated indole. Incubation at 30 °C and augmented manually afterward. and 170 rpm monitored by optical density readings every Here, we present the novel open-source AriumMS 30–60 min was performed until the inhibition of the cell (augmented region of interest for untargeted metabo- growth became apparent. Thereafter, cells were harvested lomics mass spectrometry) software to challenge the and centrifuged. The cell pellets were washed and shock- multi-platform metabolomics data analysis in combina- frozen (at −80 °C). Further sample preparation is given in tion with new methods for the analysis of polar metabo- the supplements. lites by different CE and HILIC separation techniques. AriumMS contains a universal and user-friendly toolbox, Capillary electrophoresis capable of handling multi-platform datasets. AriumMS offers automated batch processing with f lexible process- CE-ESI-MS was performed with a 7100 capillary elec- ing options and a graphical user interface. The suitability of this tool for multi-platform metabolomics is demon- trophoresis system (model no. G7100A) from Agilent Technologies (Waldbronn, Germany) coupled with an strated within a comparative study of metabolic standard mixtures and different yeast phenotypes. Therefore, the Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific, San Jose CA, USA) using the nano- metabolic standard mixtures and the yeast extracts were measured within a multi-platform approach combining CEasy interface [20]. Bare fused silica capillaries with 50/100 µm inner diameter and 360/240 µm outer diam- HILIC-MS (ESI positive/ESI negative) with three CE-MS methods applying our recently introduced nanoCEasy eter (separation/sheath liquid capillary) were obtained from Polymicro Technologies (Phoenix, AZ, USA). interface. A cationic CE-MS method was complemented by two CE-MS methods to cover a wide range of anionic Separation capillaries had a length of 90 cm and were etched with hydrof luoric acid to about an 80–100 µm metabolites. outer diameter. Three CE-MS methods have been used with the following background electrolyte (BGE) and Materials and methods sheath liquid (SL) compositions: anionic (acidic): 0.2 M formic acid pH 2.1 (BGE) and 50:50 (v/v) 2-propanol/ Materials water with 0.5% (v/v) of formic acid (SL); anionic (alkaline): 30  mM ammonium acetate pH  8.5 (BGE) The amino acid standard (1 nmol/µL in 0.1 M hydrochlo- and 50:50 (v/v) 2-propanol/water with 2.5 mM ammo- nium acetate (SL); and cationic (acidic): 1  M formic ric acid) was obtained from Agilent Technologies (Santa Clara, CA, USA). The internal standards and metabo- acid containing 10% 2-propanol pH 1.7 (BGE) and 50:50 (v/v) 2-propanol/water with 0.5% (v/v) of formic acid lites used were obtained from Sigma-Aldrich (St. Louis, MO, USA). Sugars (nucleotide sugars, phosphate sug- (SL). For each measurement, the capillary was precondi- tioned by f lushing with BGE for 5 min. For the alkaline ars) were purchased from Biosynth Carbosynth (Staad, Switzerland). Synthetic Dextrose Minimal Medium (SD, CE method, the capillary was additionally primed for 5 min applying 30 kV, and again f lushing with BGE for synthetic minimal medium) was obtained from Carl Roth (Karlsruhe, Germany). Standard materials and composi- 5  min. Samples were injected hydrodynamically with 40 mbar for 27 s (1% capillary volume). Separation was tion of the metabolomics standard can be found in the supplements. performed by applying a potential of +30 kV (cationic 1 3 Naumann L. et al. acidic and anionic alkaline BGE method) or −30  kV 54 ms accumulation time, AGC target set to “standard,” (anionic acidic BGE method) to the capillary inlet. SL 35% RF lens, and 1 micro scan. was delivered via a syringe pump (100 series, kdScien- tific, Hilliston, MA, USA) with a f low rate of 10 µL/ min, equipped with a 5  mL syringe (SGE Analytical Data evaluation and interpretation Science, Melbourne, Australia). The anionic acidic and anionic alkaline CE-MS methods were detected in ESI Data acquisition was performed using a Thermo Scientific negative mode. The cationic acidic CE-MS method was Xcalibur 4.1.50 and Orbitrap Tribrid MS Series Instrument detected in ESI positive mode. Source parameters for Control Software version 3.2 (Thermo Fisher Scientific, San Orbitrap were set to −1700 V/−2000 V/1900 V (anionic Jose CA, USA). Extraction of ion traces for the evaluation acidic/anionic alkaline/cationic acidic) spray voltage, of separation methods was done with FreeStyle 1.5.93.34 3 a.u. (arbitrary units) sheath gas, 0 a.u. aux gas, and (Thermo Fisher Scientific, San Jose CA, USA). MSconvert 300 °C ion transfer tube. 3 (ProteoWizard, Palo Alto CA, USA) [43] was used for the initial data conversion. Non-target data evaluation was Hydrophilic interaction liquid chromatography performed with AriumMS 1.0.0 (https:// github. com/ Adria nHaun/ Ar ium MS/). Software and parameters for evalua- A Dionex UltiMate 3000 (Dionex, Sunnyvale, CA, USA) tion are given in the Supporting Information (supplement high-performance liquid chromatography (HPLC) sys- Table S1 and Table S2). tem equipped with a VDSpher PUR 100 HILIC guard and separation column (4.2 × 10  mm and 150 × 3  mm, 5 µm particle size, VDS optilab Chromatographietech- nik GmbH, Berlin, Germany) heated to 30 °C was used. Results and discussion Mobile phase A was composed of H O, acetonitrile (95/5 v/v), and 5 mM ammonium acetate, and mobile phase B Study design was composed of H O, acetonitrile (5/95 v/v), and 5 mM ammonium acetate. The sample injection volume was In order to present AriumMS as a toolbox for the challenge 3 µL, and the run time was 35 min. The gradient started of multi-platform metabolomics data analysis, metabolite at 10% A, followed by a 15-min linear gradient from 10 standards and yeast extract samples were measured with five to 60% A, and hold for 5 min. Column re-equilibration analytical methods. The metabolite standard that was used was performed for 15 min at 10% A. The flow rate was contained 36 metabolites, covering important polar/ionic 300 µL/min. The LC was coupled to the Orbitrap with substance classes (mass range of 100–665 Da). The yeast the respective standard heated electrospray ionization extracts contained the metabolic information of the induced (HESI) source and sprayer. The Orbitrap source param- effect by a halogenated indole treatment. To analyze polar eters were set to 3500 V positive/negative spray voltage, and/or ionic metabolites of interest within the samples, two 50 a.u. sheath gas, 10 a.u. aux gas, 325 °C transfer tube, HILIC-MS and three CE-MS methods have been developed. and 350 °C vaporizer temperature. In order to determine optimal AriumMS data processing parameters for the generated datasets of the five analyti- Mass spectrometry cal methods, a D-optimal design of experiment (DOE) was applied for software parameter screening and optimization. For mass spectrometry, an Orbitrap Fusion Lumos mass Furthermore, the feature generation of AriumMS was vali- spectrometer (Thermo Fisher Scientific, San Jose CA, dated. This was followed by a multi-platform metabolomics USA) was used in either positive or negative ion mode, data analysis of the yeast extracts. The complete analytical with a scan range of 100–700 m/z. Resolving power was workflow is shown in Fig.  1. set to 60,000, accumulation time to 50 ms, automatic gain control (AGC) target to “standard,” 35% RF lens, and Evaluation of the analytical methods 1 micro scan. Data-dependent MS/MS experiments with 0.6 s cycle time were performed. Filters were an inten- The standard contained a total of 36 typical polar metabo- sity threshold at 2E4, exclusion after a single occurrence lites and four internal standards, including yeast metab- for 10 s, and isotope exclusion. Data-dependent MS/MS olites, amino acids, hexoses, hexose phosphates, and Orbitrap higher-energy collisional dissociation fragmen- nucleotide sugars. Anionic, cationic, zwitterionic, and tation (HCD) parameters were isolation width of 1.5 da, uncharged species were represented (Table 1). To deter- 20/35/50% HCD power, Orbitrap resolution of 30,000, mine the overall capabilities of the five different analytical 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… Fig. 1 Multi-platform metabolomics workflow overview, containing all steps of sampling, analysis, and AriumMS workflow methods regarding the number of detected analytes and co-migrating neutrals) over a period of 45 min (Table 1). duration of analysis, six repetitions of the metabolomics Using the cation CE-MS method, 17 of 36 metabolites standard were measured with each method. The five sepa - were able to be detected. Neutral metabolites, such as hex- ration methods were evaluated regarding the number of oses and caffeine, were not detected by any of the CE-MS detectable analytes, their migration time (MT)/retention methods. In summary, when the three CE-MS methods time (RT), and their separation efficiency (for further were combined, CE-MS was able to detect 32 of 36 metab- details, see supporting information, Fig. 2a–e, Table 1). olites. When all five analytical methods were applied, all These five separation methods offered overlapping and metabolites of the standard were detectable (Table  1). complementary information on the metabolite standard, Apart from the difference in selectivity, HILIC-MS and as shown in Fig. 2 and Table 1: The two HILIC methods CE-MS each had distinct advantages: HILIC-MS exhib- covered most of the metabolites, i.e., 25 of 36 in ESI+ and ited a higher retention time reproducibility and a higher 27 of 36 in ESI−, respectively, and when they were com- degree of automation, while CE-MS required a smaller bined, 30 of 36 metabolites of the standard were able to sample volume and showed more efficient separation with be detected. Some multi-carboxylic acids and basic amino sharper peaks. acids were not detected, and isomeric hexose phosphates were not baseline-separated. The selectivity of the CE-MS Non‑target data evaluation methods used was higher. The anionic alkaline CE-MS method was able to detect 30 out of 36 metabolites over a AriumMS workflow period of 30 min. Seventeen metabolites (such as neutral amino acids) co-migrated with the electroosmotic flow AriumMS was developed as a universal and user-friendly (EOF, no separation) (see Table  1). The anionic acidic computational metabolomics toolbox to tackle the chal- CE-MS method was capable of analyzing phosphates and lenge of multi-platform MS data analysis. The acronym Ari- dicarboxylic acids and covered 15 of 36 analytes (four umMS stands for augmented region of interest for untargeted 1 3 Naumann L. et al. 1 3 Table 1 List of all target analytes of the metabolomics standard Group Analyte Anion Cation ID Name HILIC CE (alkaline BGE) CE (acidic BGE) HILIC CE − + [M−H] RT ± sd AriumMS MT ± sd AriumMS MT ± sd AriumMS [M+H] RT ± sd AriumMS MT ± sd AriumMS [min] [min] [min] [min] [min] Amino acid 1 L-serine 104.0348 10.8 ± 0.1  ✓ 7.7* ± 0.3  ✓ - - 106.0504 10.8 ± 0.0  ✓ 19.5 ± 0.5  ✓ 2 L-proline 114.0555 11.6 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 116.0712 11.8 ± 0.1  ✓ 21.5 ± 0.6  ✓ 3 L-valine 116.0712 10.1 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 118.0869 10.1 ± 0.0  ✓ 19.6 ± 0.5  ✓ 4 L-threonine 118.0504 10.6 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 120.0661 10.6 ± 0.0  ✓ 20.8 ± 0.6  ✓ 5 L-cysteine 120.0119 - - 7.8* ± 0.3  ✓ - - 122.0276 13.2 ± 0.1  ✓ 22.3 ± 0.7  ✗ 6 L-leucine 130.0869 9.1 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 132.1025 9.1 ± 0.0  ✓ 20.1 ± 0.6  ✓ 7 L-isoleucine 130.0869 9.4 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 132.1025 9.4 ± 0.0  ✓ 20.4 ± 0.6  ✓ 8 L-aspartic acid 132.0297 9.9 ± 0.1  ✓ 15.9 ± 1.1  ✓ 38.6 ± 1.6  ✗ 134.0454 10.0 ± 0.0  ✓ 23.4 ± 0.7  ✓ 9 L-lysine 145.0978 - - 5.4 ± 0.1  ✓ - - 147.1134 15.5 ± 0.5  ✗ 12.7 ± 0.2  ✓ 10 L-glutamic 146.0454 10.2 ± 0.1  ✓ 14.5 ± 0.9  ✓ - - 148.0610 10.2 ± 0.0  ✓ 22.0 ± 0.7  ✓ acid 11 L-methionine 148.0433 9.3 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 150.0589 9.3 ± 0.0  ✓ 21.3 ± 0.6  ✓ 12 L-histidine 154.0617 - - 7.4 ± 0.2  ✓ - - 156.0773 - - 13.6 ± 0.3  ✓ 13 L-phenylala- 164.0712 8.7 ± 0.4  ✓ 7.8* ± 0.3  ✓ - - 166.0869 8.6 ± 0.0  ✓ 22.2 ± 0.7  ✓ nine 14 L-arginine 173.1039 - - 5.5 ± 0.1  ✓ - - 175.1196 - - 13.3 ± 0.3  ✓ 15 L-tyrosine 180.0661 8.8 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 182.0818 8.8 ± 0.0  ✓ 23.2 ± 0.7  ✓ 16 L-tryptophan 203.0821 8.1 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 205.0978 8.1 ± 0.0  ✓ 22.3 ± 0.7  ✓ Internal std 17 benzene- 141.0010 3.5 ± 0.1  ✗ - - 12.8 ± 0.2  ✗ 143.0167 - - - - sulfinic acid 18 2-nitrobenzoic 166.0140 3.1 ± 0.0  ✓ 16.3 ± 1.3  ✓ 19.2 ± 0.4  ✓ 168.0297 - - - - acid 19 methionine 180.0331 9.8 ± 0.1  ✓ 7.7* ± 0.3  ✓ - 182.0487 9.9 ± 0.0  ✓ 24.6 ± 0.8  ✓ sulfone 20 pentetic acid 392.1306 10.3 ± 0.1  ✓ 30.7 ± 5.8  ✗ - 394.1462 10.4 ± 0.0  ✓ - - Metabolites 21 succinate 117.0188 7.7 ± 1.0  ✓ - - 37.7 ± 1.5  ✓ 119.0345 - - - - 22 nicotinic acid 122.0242 6.3 ± 0.1  ✓ 17.4 ± 1.4  ✓ 19.2 ± 0.4 ✓  124.0399 6.4 ± 0.0  ✓ 20.0 ± 0.6  ✓ 23 tartaric acid 149.0086 - - - - 30.4 ± 1.0  ✓ 151.0243 - - - - 24 citrate; citric 191.0192 - - - - 32.1 ± 1.1  ✗ 193.0348 - - - - acid 25 caffeine 193.0726 - - - - - - 195.0882 4.0 ± 0.0  ✓ - - 26 ATP 505.9879 10.2 ± 0.1  ✗ 25.3 ± 2.8  ✗ - - 508.0036 10.4 ± 0.0  ✗ - - 27 NAD 662.1013 10.5 ± 0.0  ✓ 10.1 ± 0.5  ✓ 38.5 ± 1.6  ✗ 664.1170 10.6 ± 0.0  ✓ - - 28 NADH 664.1170 8.6 ± 0.0  ✓ 13.6 ± 0.9  ✗ - - 666.1327 8.7 ± 0.0  ✓ - - Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… 1 3 Table 1 (continued) Group Analyte Anion Cation ID Name HILIC CE (alkaline BGE) CE (acidic BGE) HILIC CE − + [M−H] RT ± sd AriumMS MT ± sd AriumMS MT ± sd AriumMS [M+H] RT ± sd AriumMS MT ± sd AriumMS [min] [min] [min] [min] [min] Carbohy- 29 L-fuc., 163.0607 - - 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 165.0763 - - - - drates 6-deoxy-L- gal 30 D-mannose 179.0556 6.5 ± 0.2  ✓ 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 181.0713 - - - - 31 α-D-glucose 179.0556 6.8 ± 0.2  ✓ 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 181.0713 - - - - 32 α-D-galactose 179.0556 7.5 ± 0.6  ✓ 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 181.0713 - - - - 33 D-mannose- 259.0219 10.3 ± 0.0  ✓ 21.9 ± 1.3  ✓ 15.9 ± 0.2  ✓ 261.0376 10.3 ± 0.0  ✗ - - 1-PO 34 β-D-fructose- 259.0219 10.1 ± 0.0  ✓ 19.4 ± 1.1  ✓ 15.8 ± 0.2  ✓ 261.0376 10.1 ± 0.0  ✗ - - 6-PO 35 α-D-glucose 259.0219 9.9 ± 0.0  ✓ 19.0 ± 1.1  ✓ 15.5 ± 0.2  ✓ 261.0376 9.9 ± 0.0  ✗ - - 1-PO 36 α-D-galactose- 259.0219 10.3 ± 0.0  ✓ 18.1 ± 0.9  ✓ 15.5 ± 0.2  ✓ 261.0376 10.3 ± 0.0  ✓ - - 1-PO 37 N-acetylneu- 307.0909 - - 7.7* ± 0.3  ✓ - - 309.1066 - - - - raminate 38 UDP-glucose 565.0472 7.5 ± 0.1  ✓ 14.7 ± 0.9  ✓ 11.7 ± 0.8 567.0629 7.6 ± 0.0  ✓ - - 39 GDP-L-fucose 588.0744 8.7 ± 0.3  ✓ - - - - 590.0901 9.0 ± 0.0  ✓ - - 40 CMP-N- 613.1395 9.0 ± 0.0  ✓ - - - - 615.1552 - - - - acetylneu- raminate If the peaks of the analytes fulfill the general criteria for peak detection (maximum peak width ≤ 2 min, peak intensity ≥ 5E4), then the retention time ± standard deviation (sd) is given. If an analyte is not detected by the method itself or does not fulfill the criteria for peak detection, it is labeled with “-”. If AriumMS is able to detect an analyte, it is labeled with “✓ ”, if not with “✗”. Analytes co-migrating with EOF are labeled with “*”. Analytes detected as co-migrating neutrals are labeled with “#” Naumann L. et al. metabolomics mass spectrometry. It was designed as a multi- divided by the total number of points of the peak. H is cal- tiered software for scalable (parallel processing of multiple culated for each peak, and values greater than the median sample sets) and reproducible data analysis. AriumMS con- entropy are considered as noise and discarded. The peaks of sists of a main app (AriumMS, Fig. S2a) for ROI search, the remaining features after the filtering were integrated, and alignment, and low-level data filtering, and an evaluation the obtained areas are sorted into an N x M x S matrix, where App (AriumMSEval, Fig. S2b) for feature extraction and N corresponds to the retention or migration times (rows), M augmented data analysis. To ensure a high level of MS corresponds to the m/z (columns), and S corresponds to the instrument compatibility, the open-source MS data format repeat measurement (layers). Based on the feature intensi- mzXML is used for raw data import [44]. To achieve opti- ties obtained by the integration, the features can be scaled mal results in data processing, AriumMS uses user-defined along the repeat measurements. Available scaling methods sample groups. A sample group can contain datasets of dif- include center, auto, Pareto, vast, range, and level. Constant ferent separation methods, different analytical workflows, or factors for whole groups and sample-specific factors can be multiple phenotypes of biological samples. An individual applied as well. This allows, for example, normalization to parameterization can be applied for each group, for example, the cell count of the sample or normalization to multiple depending on the different peak characteristics of the sepa- internal standards for metabolite quantitation. Logarithmic ration methods used. Files are then batch-processed group and power transformations are available as well. A guideline by group. To reduce the computation time, a crop filter can for the selection of a proper scaling method is given by van be applied to discard areas of the measurements without den Berg et al. [53]. In the next step, the user-defined groups relevant peaks. Several optional filters minimize the number are augmented by linking the data cubes along the m/z of false-positive features during the ROI phase (e.g., isotope, dimension, combining the m/z and time dimensions into one adduct, and common contaminant filter) [45, 46]. An addi- dimension. The features are now named according to the fol- tional baseline correction removes any drift across the sepa- lowing scheme: “m/z @Time, Group". If features of different ration by baseline determination over a moving window by groups have the identical mass and number of occurrences interpolation [47]. After processing the ROI stage, the data (no. of detections within the groups), they are assumed to be is automatically transferred to the AriumMSEval app. For the same and labeled accordingly. The complete flow chart automatic peak detection, the obtained ROIs are smoothed of the data processing can be found in Fig. S3. in the first step, and the difference between the smoothed ROI and the original ROI is used to estimate the noise level AriumMS parameter screening and optimization for this m/z. The second derivative of the smoothed ROI is formed, and peaks are identified by continuous wavelet In order to obtain good results with the AriumMS software transform (CWT) using the Mexican hat as the mother wave- package, we applied an efficient D-optimal DOE for soft- let [48, 49]. Since real peaks are rarely perfectly symmetric, ware parameter screening and optimization [54, 55]. The the peak boundaries are adjusted by a two-step process; first, D-optimal DOE design enables the identification of optimal via a friction border correction [50] based on the smoothed parameter settings with a lower number of required experi- peak and then via moving standard deviation border correc- ments compared to other designs. For that reason, the six tion based on the original peak. This algorithm can be found repetitions of the metabolomics standard measured by all in the supplement information (Algorithm A1). Within the v fi e analytical methods were evaluated regarding the number corrected limits, the peaks are now integrated, and the reten- of found analytes and the total number of features. Found tion time is determined. Followed by the initial feature fil- target features were defined by their m/z value and respec- tering of the AriumMSEval, possible feature filters are, for tive retention time (parameters for non-target data labeling example, minimum and maximum peak width, minimum are given in the supporting information). According to the height within an ROI, and signal-to-noise ratio (support- results, a total of 126 target features were found. The total ing information Table S2). As an additional approach, an number of target features represents the number of target information entropy peak filter adapted from Ju et al. [51] features detected by the five separation methods, including was integrated. For a Gaussian peak, all points before the internal standards and co-migrating analytes. During the maximum have a constant positive slope, and after the DOE screening, the ROI functions developed by Tauler [56] maximum, a constant negative slope; these points are called were tweaked by disabling the addition of random noise on normal points. Points that deviate from this condition are the extracted ion chromatogram/electropherogram (EIC/E), called variant points. Accordingly, the entropy of a peak since it was not required for AriumMS. By default, this algo- can be expressed by the sum of the entropy of all possible rithm added random noise on the EIC/E to remove possible events H =−p ∗ log (p) − q ∗ log (q) [51, 52], where p is gaps in the data. Here, the addition of random noise to the 2 2 the number of variant points divided by the total number EIC/E created multiple peak tips and increased the peak of points of the peak, and q is the number of normal points splitting within the six repetitions of the standard, which 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… hexose-PO HILIC (ESI -) 7 Time (min) 13 05 10 15 20 25 30 35 40 45 Time (min) hexose-PO CE (alkaline BGE, anion) Time (min) 19 05 10 15 20 25 30 35 40 45 EOF Time (min) hexose-PO CE (acid BGE, anion) 13 19 Time (min) 05 10 15 20 25 30 35 40 45 EOF Time (min) LEU/ILE HILIC (ESI +) Time (min) 12 05 10 15 20 25 30 35 40 45 Time (min) LEU/ILE CE (cation) Time (min) 22 05 10 15 20 25 30 35 40 45 Time (min) Fig. 2 Comparison of different analytical methods using the meas- and L-isoleucine (cation methods). (A) HILIC-MS anion, blue; (B) urements of the metabolomics standard that contains 40 substances. CE-MS anion alkaline, gray; (C) CE-MS anion acidic, yellow; (D) The BPC/Es are shown at A–E containing a zoom to the separation HILIC-MS cation, orange; (E) CE-MS cation, acidic of either the four hexose phosphates (anion methods) or L-leucine induced varying retention times. Peak picking and integra- ensure the comparability between repeated measurements tion of AriumMS were improved by the removal of the arti- for the same method and to counter the effects of analytical ficial distortion of the peak tips by the ROI function. variance, the chromatograms could be alternatively aligned For an initial parameter screening, 14 parameters at in time (recommended especially for CE) and m/z dimen- two levels of both stages of the software (AriumMS, Ari- sions. The mass spectra alignment shifts measured masses to umMSEval) were chosen. The DOE identified the follow - match the most common x quantile of detected masses (e.g., ing parameters as significant for further optimization: ROI 0.95). Since it aligns masses, it offers benefits for QTOF intensity threshold, mass spectra alignment, m/z error, and instruments or for low-resolution MS. In general, resolution feature occurrence l fi ter. The ROI intensity threshold den fi es and calibration of the mass spectrometer must be considered the m/z intensity cutoff limit for noise. In general, a higher for non-target data processing. Hence, the m/z error of the intensity threshold leads to a lower number of found features. ROI should be set properly; here, 0.01 Da represents 10 ppm For example, an increase of the ROI intensity threshold from at 1000 Da (upper m/z limit). One of the most important 50,000 cts. to 150,000 cts. roughly loses 10% of total fea- feature filters of AriumMS is the minimum feature occur - tures/target features (HILIC ESI). As a universal robust ROI rence, which is defined as the relative minimum of feature intensity threshold, we recommend 5–10% of the lowest base detections per group. This filter leads to a reduction of the peak chromatogram/electropherogram (BPC/E) intensity. To random noise within the MS data. If the minimum relative 1 3 Intensity Intensity Intensity Intensity Intensity Intensity Intensity Intensity Intensity Intensity Naumann L. et al. occurrence was set to 50%, the feature needs to appear in at the standard at 50% occurrence level (features were detected least three out of six measurements. For higher confidence in three of six measurements), as given in Table 1. For the of the obtained features, higher percentages for the minimum evaluation of the AriumMS peak integration algorithm, two relative occurrence were better. For example, the evaluation example datasets were chosen because of their different peak of the HILIC anion measurements showed 31 target features characteristics. These were CE-MS (alkaline BGE, anion) at 0% (≥ 1/6) occurrence, 30 target features at 50% (≥ 3/6) (Fig. 3a) and HILIC-MS (ESI−) (Fig. 3b). The peak heights occurrence, and 23 target features at 100% (6/6) occurrence. and areas of AriumMS were compared with the results of Within a D-optimal DOE optimization, further param- the manual peak integration using FreeStyle, both normal- eters were tested, and relevant parameters were optimized ized to an internal standard (supporting information). Ari- using three levels per parameter. These were minimum ROI umMS was able to find 88% (CE) and 90% (HILIC) of the size and minimum relative peak height. Minimum ROI peak height and area compared to the manual integration. size is defined as the minimum number of MS1 scans in This finding can be explained by the function of the inte - which the m/z must be present in the EIC. This parameter gration algorithm itself because the ROI intensity threshold is dependent on the processed separation method because is always subtracted from the peak. Furthermore, the peak the obtained feature peak width can differ between dif- integration of the HILIC  method had two outliers com- ferent separation methods Therefore, levels 5, 10, and 15 pared to the manual integration, caused by limit cases of were tested. HILIC required higher ROI sizes (15, broader either non-baseline separated or very broad peaks and thus peaks) and smaller CE (< 10) because of the narrower peak incorrect integration by the software. In general, the low width. In general, the minimum ROI size must be below deviation of the peak integration algorithm of AriumMS the expected peak width of each separation method. The to the manual integration demonstrates the capabilities of minimum relative peak height was significant for feature this software tool for quantitation as generally requested for filtering (AriumMSEval). This filter analyzes each ROI and metabolomics tools [58]. discards features below the relative peak height limit (%). A To test the reliability of the data processing regarding suitable value was 25%. the independence of the loaded file order and simultane- AriumMS offers the capability to define different param- ously the peak finding in general, the six data files of the eter settings for the simultaneous processing of each evalu- repeat measurements were processed three times while only ation group (different methods). Peak shapes and migration/ changing the order in which files were loaded. AriumMS retention time stability differ highly between CE and HILIC; generated similar results for all methods within the three therefore, minimum ROI group size and peak alignment file orders (Fig.  3c) except the HILIC cation. Here, three (time) were probably the key aspects and should therefore additional metabolites of the standard were found, caused be set for each group (method) individually. Especially, peak by limit cases of either non-baseline separated analytes or alignment becomes relevant for CE data due to migration very broad peaks. time shifts that can occur between replicates (cp. avg. migra- AriumMS reduces the required data post-processing tion time deviation for CE [acidic BGE, anion]: ±0.9 min, by the user significantly, compared to traditional metab- and HILIC [ESI−]: ±0.1 min). The use of effective electro- olomics software, which is typically not optimized for phoretic mobility instead of the migration time can address multi-platform analytics. The processing of the whole this issue [57] and will be implemented in AriumMS in the dataset containing five analytical methods with six meas- future. urements takes about 60 min when AriumMS was used on a consumer-grade personal computer (PC) system with Validation of the AriumMS feature generation a 6-core CPU and 32 GB RAM). Considering the com- putation power of the computer used for AriumMS, an For the validation of the feature generation of AriumMS, enterprise-grade computer (32-core CPU, 128 GB RAM) the number of found targets and their respective integration was able to reduce the required processing time to 50 min. were evaluated. The reliability of the data processing was Extensive data post-processing is not required for Ari- tested with different file orders, and the required processing umMS due to its automated data augmentation of different time of the overall workflow is given. Finally, the feature methods (groups) and the integration of related statistical generation and peak integration algorithm of AriumMS was tools, which are mandatory for multi-platform data evalu- compared with the established universal open-source plat- ation. For data evaluation, AriumMS contains advanced form MZmine 3 [41]. For this validation, the MS data of all statistical evaluation tools such as labeling of false-pos- five analytical methods were processed with the optimized itive features (false discovery rate, Benjamini–Hochberg parameter settings (Supplement Table S2). procedure) [59], data scaling options (centering, Pareto, Using the optimized data processing settings, AriumMS auto), transformation (power and log), and various plots was able to find 89% (112 of 126) of the target features of (scatter plots, volcano plots, principal component analysis 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… [PCA], and heatmaps). Because of the combination of the metabolome was analyzed here by the five different ana- feature list generation and the statistical evaluation, no lytical methods and augmented evaluation by AriumMS. In further data transfer into additional statistical software is principle, multi-platform metabolomics offers two major required, which is an advantage compared to other metabo- improvements. Firstly, it is possible to increase the analyti- lomics software. AriumMS is under active development, cal metabolome coverage, and secondly, observed metabolic and the open-source code is continuously optimized to effects are cross-evaluated by another method (if detected in further improve the required data processing times. both). Generally, the data evaluation here was based on three In order to compare the feature generation and peak inte- levels. Starting with a target evaluation, based on the feature gration algorithm of AriumMS with the established univer- list obtained by AriumMS, target features were assigned by sal open-source platform MZmine 3 [41], the data of the their m/z value and RT/MT compared to the reference values five analytical methods were processed with MZmine using of the metabolite standard, followed by a suspect evalua- optimized parameter settings and data processing options tion, where target features were assigned by m/z values of (Supplement Table S3). MZmine was able to find 87% (109 a suspect database. Moreover, in a non-targeted evaluation, of 126 target features) and AriumMS found 89% (112 of 126 the overall feature lists of Mock and Effect1 were compared. analytes) of the target features of the standard, both with The mid-level augmented data evaluation shows that the an occurrence level of 50%. MZmine found 98% (CE) and combination of all ESI negative methods (HILIC [ESI−], CE 110% (HILIC) of the peak height and area compared to the alkaline BGE, CE acidic BGE) detects 24 targets. The ESI manual integration. The comparison reveals that AriumMS positive augmentation (HILIC [ESI+, CE cation]) detects and MZmine offer similar results regarding feature genera- 19 targets (Table 2, Fig. 4a). If just one method is applied tion and peak integration, which highlights the solid foun- for the metabolome analysis, the number of found features dation of the AriumMS feature extraction for augmented is decreased because HILIC (ESI+) detects just 14, and multi-platform data analysis. CE (cation) 16 targets. Figure 4a shows a comparison of target numbers found by each method and by the augmen- tation. The total number of detectable targets in the yeast Augmented analytical workflows and data extracts was lower than in the metabolite standard due to evaluation their absence in the yeast metabolism (Table 2 “not present”) or low biological concentration. The respective number of Combination of different methods detectable target metabolites was reduced to 28 for ESI negative augmentation and 21 for ESI positive augmenta- The combination of multiple analytical methods—here, tion (Fig. 4a). CE-MS and HILIC-MS—increases the feature capacity by For the suspect evaluation, metabolites of the glycoly- their different selectivity. Each separation technique (either sis, gluconeogenesis, TCA (tricarboxylic acid) cycle, and HILIC or CE) was not able to detect all 36 metabolites of the amino acid metabolism were analytes of interest. There- standard (Fig. 4a, Table 1). The multi-platform data evalua- fore, a list containing the m/z values of relevant metabo- tion offers the possibility to increase the analytical coverage lites (Table S4) was used for a m/z search and labeling of a non-target metabolomics workflow. In an environment within the generated feature lists of the measured yeast as complex as metabolomics samples, the number of detect- extracts (labeling within a ± 0.03 m/z range). Again, the able features (target, suspect, non-target) can be expected multi-platform analysis offers a higher analytical metabo- to be higher. Hence, two combinations for the augmented lome coverage than single methods, as shown in Fig. 4b analysis were tested. Based on the AriumMS evaluation of and supplement Table S4. In addition, two mutually con- the standard, the mid-level augmented data evaluation of all firming analytical methods lead to higher confidence in three ESI negative methods enables the coverage of up to 30 suspect feature search only based on m/z values. Here, the target metabolites (w/o co-migrating metabolites). Combin- ESI negative augmentation finds 38 suspects, and the ESI ing the two ESI positive methods allows the coverage of up positive 31 suspects (Fig. 4b). The three CE methods offer to 23 target metabolites. the capability to cross-prove several of the observed sus- pect effects by the HILIC methods (Table S4). Most of Application: yeast metabolome the suspect regulation fold changes matched between the methods; a reason for varying results could be the simi- As a case study, two yeast CEN.PK122 [42] cell cultures lar m/z values of different metabolites or different matrix were analyzed, one as reference (Mock) and the other effects (for example, co-migration/elution of metabolites or treated with halogenated indole (Effect1). Adding halogen- BGE effects). This issue can be addressed with the imple- ated indole to yeast resulted in a clearly decreased growth mentation of the MS/MS confirmation and will be imple- rate, as shown in Fig. S1. The induced effect on the yeast mented soon in AriumMS and a database search. Most of 1 3 Naumann L. et al. 1.2 6.0 A B 1.0 5.0 0.8 4.0 0.6 3.0 0.4 2.0 0.2 1.0 0.0 0.0 0.00.5 1.01.5 2.0 0.01.0 2.03.0 4.05.0 Normalized area/height FreeStyle Normalizedarea/height FreeStyle height area height area CE (ESI-, alkaline BGE) CE (ESI-, acidic BGE) HILIC (ESI-) 31/33 identified 12/33 identified 29/33 identified Rep1 Rep1 Rep1 Rep2 Rep3 Rep2 Rep3 Rep2 Rep3 CE cation HILIC (ESI+) 18/33 identified 21/33 identified Rep1 Rep1 Rep2 Rep3 Rep2 Rep3 Fig. 3 Comparison of peak integration algorithms of AriumMS using repeatability of the peak finding in general for AriumMS (C). Venn the example of CE-MS anion alkaline (A) and HILIC-MS anion (B). diagrams are shown for each method Evaluation of the independence of file order and simultaneously the the targets and suspects were significantly downregulated Fig.  S4). The sum parameter of mannose-1-phosphate between Mock and Effect1, shown in Fig.  4c. An observed and galactose-1-phosphate showed a slight downregula- metabolic effect of the ESI negative augmentation was the tion with fold change (FC) of 0.9 (HILIC ESI−, indole- change in sugar metabolism (Table 2, supplement Table S4, treated effect divided by reference cell effect). The sum 1 3 Normalized area/height AriumMS AriumMS Normalizedarea/height AriumMS Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… 24 24 21 21 18 18 15 12 9 9 HILIC (ESI -) CE (alk. BGE, CE (ac. BGE, HILIC (ESI +) CE (cation)augmentation augmentation anion) anion) ESI negative ESI positive detectable targets AriumMS (metabolite std., w/o internal std., w/o co-migrating analytes) detected targets AriumMS (yeast) 14 14 HILIC (ESI -) CE (alk. BGE, CE (ac. BGE, HILIC (ESI +) CE (cation) augmentation augmentation anion) anion) ESI negative ESI positive detected suspects (yeast) 10 10 8 8 6 6 4 4 2 2 0 0 -8 -6 -4 -2 02468 -8 -6 -4 -2 0246 8 log2 fold change log2 fold change HILIC (ESI-) CE (alk. BGE) HILIC (ESI+) CE cation CE (aci. BGE) FC of 2 FC of 2FC of 0.5 FC of 0.5 critical p-value critical p-value Fig. 4 Results of the augmented data evaluation. A  Shows the num- umMS in yeast extracts is presented with black bars. Both numbers ber of detectable standard metabolites by the methods itself and by are without internal standards and co-migrating analytes. B  Shows mid-level augmented data evaluation using AriumMS presented with the number of found suspects per method and augmentation. Volcano gray bars. The number of detected targets in the yeast extracts per plots for the two augmentations are presented in (C) ESI negative and method and by the mid-level augmented data evaluation using Ari- (D) ESI positive mode parameter of fructose-6-phosphate and glucose-1-phos- (Fig.  S4a–b). On the suspect level, the appearance of phate was even more downregulated (FC 0.3, HILIC two additional disaccharides was observed (Fig. S4g–h). ESI−, Fig. S4 e–f). The CE-MS method (alkaline BGE) Hence, indole treatment may lead to lower levels of glu- showed the same downregulation of glucose-1-phosphate cose, hindering the production of mannose-1-phosphate 1 3 -log10 p No. of targets No. of suspects -log10 p Naumann L. et al. Table 2 Results of the augmented multi-platform data analysis of yeast extracts based on targets Group Analyte ESI negative augmentation ESI positive augmentation − + ID Name [M−H] HILIC/CE (alk.)/CE (aci.) [M+H] HILIC/CE Amino acid 1 L-serine 104.0348 down / down* / - 106.0504 n.a. / down 2 L-proline 114.0555 down / stable* / - 116.0712 stable / down 3 L-valine 116.0712 stable / stable* / - 118.0869 stable / stable 4 L-threonine 118.0504 down / stable* / - 120.0661 down / stable 5 L-cysteine 120.0119 - / n.a.* / - 122.0276 n.a. / n.a. 6 L-leucine 130.0869 stable / stable* / - 132.1025 knock out / stable 7 L-isoleucine 130.0869 down / n.a.* / - 132.1025 n.a. / induced 8 L-aspartic acid 132.0297 down / stable / n.a. 134.0454 down / induced 9 L-lysine 145.0978 - / stable / - 147.1134 n.a. / stable 10 L-glutamic acid 146.0454 stable / stable / - 148.0610 stable / stable 11 L-methionine 148.0433 down / knock out* / - 150.0589 down / down 12 L-histidine 154.0617 - / down / - 156.0773 - / stable 13 L-phenylalanine 164.0712 stable / up* / n.a. 166.0869 stable / stable 14 L-arginine 173.1039 - / stable / - 175.1196 - / up 15 L-tyrosine 180.0661 stable / stable* / - 182.0818 down / induced 16 L-tryptophan 203.0821 knock out / knock out* / - 205.0978 n.a. / knock out Internal std 17 benzenesulfinic acid 141.0010 used for internal standardization 143.0167 used for internal standardiza- 18 2-nitrobenzoic acid 166.0140 168.0297 tion 19 methionine sulfone 180.0331 182.0487 20 pentetic acid 392.1306 394.1462 Metabolites 21 succinate 117.0188 down /—/ n.a. 119.0345 - / - 22 nicotinic acid 122.0242 stable / stable / stable 124.0399 stable / knock out 23 tartaric acid 149.0086 - /—/ n.a. 151.0243 - / - 24 citrate; citric acid 191.0192 - /—/ n.a. 193.0348 - / - 25 caffeine 193.0726 not present 195.0882 not present 26 ATP 505.9879 n.a. / n.a. / - 508.0036 stable / - 27 NAD 662.1013 down / stable / n.a. 664.1170 stable / - 28 NADH 664.1170 n.a. / n.a. / - 666.1327 knock out / - Carbohydrates 29 L-fuc., 6-deoxy-L-gal 163.0607 - / n.a.* / n.a.* 165.0763 - / - 30 D-mannose 179.0556 knock out / knock out* / n.a.* 181.0713 - / - 31 α-D-glucose 179.0556 n.a. / n.a.* / n.a.* 181.0713 - / - 32 α-D-galactose 179.0556 n.a. / n.a.* / n.a.* 181.0713 - / - 33 D-mannose-1-PO 259.0219 stable / stable/ stable 261.0376 knock out / - 34 β-D-fructose-6-PO 259.0219 down / n.a. / stable 261.0376 n.a. / - 35 α-D-glucose 1-PO 259.0219 down / n.a. / stable 261.0376 n.a. / - 36 α-D-galactose-1-PO 259.0219 stable / stable / stable 261.0376 n.a. / - 37 N-acetylneuraminate 307.0909 not present 309.1066 38 UDP-glucose 565.0472 stable / n.a. / knock out 567.0629 n.a. / - 39 GDP-L-fucose 588.0744 not present 590.0901 not present 40 CMP-N-acetylneuraminate 613.1395 not present 615.1552 not present Number of detected targets (w/o co-migrating analytes) anion 24 cation 19 Regulation calculated by the FC (indole-treated/reference): FC < 0.50, down; 0.50 ≤ FC ≤ 2.0, stable; FC ≥ 2.00, up. Non-detected analytes are labeled as “n.a.”, undetectable analytes are labeled as “-”. Knocked-out analytes by halogenated indole treatment are labeled as “knock out,” and newly occurring analytes by treatment are labeled as “induced.” Co-migrating analytes are labeled with *. The order of the presented results is as follows: (i) augmentation ESI negative: HILIC (ESI−)/CE (alkaline, anion)/CE (acidic, anion), and (ii) augmentation ESI positive: HILIC (ESI+)/CE (cation) 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… and glucose-6-phosphate, causing increased production of of various isomeric anions such as hexose phosphates by lactose and saccharose. Lactose cannot be digested by yeast CE-MS. cells because of the lacking lactase enzyme. Furthermore, AriumMS was successfully applied for mid-level data within the two augmentations, L-tryptophan was knocked fusion to remove irrelevant information such as artifacts and out in the indole-treated samples (Effect1). noise from the entire analytical dataset (LC-MS + CE-MS) For the non-target evaluation, three PCAs were con- of yeast extracts. The results confirm the great advantage of ducted, which contain the features of Mock and Effect1 flexible parameterization for processing of individual sepa- (feature occurrence ≥ 50%). Each of the three PCAs was ration methods with different peak characteristics. Multi- a combination of an HILIC (ESI±) method with one CE platform metabolomics expands metabolome coverage and method. Figure S5 shows that the measured samples (Mock increases the confidence of the metabolic results. and Effect1) were partitioned into two major groups derived Supplementary Information The online version contains supplemen- by the treatment with halogenated indole with at least 79.5% tary material available at https://doi. or g/10. 1007/ s00216- 023- 04715-6 . of the explained variance on the first principal component (PC1). As a result, the treatment with halogenated indole has Acknowledgements Gratefully, we acknowledge Prof. Dr. Norbert Schnell and Nadine Pejs from University Aalen for generating the bio- a strong influence on the yeast metabolome. The integration logical samples and Olivia Haun for the execution of the relative DOE of several LC-MS/CE-MS techniques expands metabolome and the software optimizations for AriumMS. Thank you to Patrick coverage and increases the confidence of the metabolic Schlossbauer for valuable discussions and useful suggestions. results. This makes AriumMS a powerful tool for multi- Funding Open Access funding enabled and organized by Projekt platform metabolomics. DEAL. This study was supported by the German Federal Ministry of Education and Research within the PROmiGlykAN project (FKZ 13FH635IB6) and the associated partners of this project (Bruker Dal- tonik GmbH, Rentschler Biotechnologie GmbH, MLS GmbH). Conclusions Data availability The datasets generated during and/or analyzed during Multi-platform metabolomics by high-resolution MS based the current study are available from the corresponding author on rea- on several orthogonal separation mechanisms of CE and LC sonable request. AriumMS 1.0.0 is available at GitHub: https://git hub. com/ Adria nHaun/ Arium MS/ maximizes the metabolome coverage. The AriumMS soft- ware toolbox presented here is a powerful tool for fast untar- Declarations geted processing of these augmented datasets. AriumMS contains ROI search, preprocessing, feature detection and Competing interests The authors have no relevant financial or non- integration, false-positive filter, scaling, and transformation financial interests to disclose. followed by the augmentation and various chemometric data evaluation tools. The validation of the feature detection and Open Access This article is licensed under a Creative Commons Attri- mid-level fusion steps were successfully performed using bution 4.0 International License, which permits use, sharing, adapta- tion, distribution and reproduction in any medium or format, as long a multi-analyte standard. In AriumMS all processing steps as you give appropriate credit to the original author(s) and the source, were integrated into a single user-friendly software tool at a provide a link to the Creative Commons licence, and indicate if changes high level of e fl xibility and automatization. Further develop - were made. The images or other third party material in this article are ments will include the implementation of MS/MS spectral included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in networking in order to precisely and autonomously connect the article's Creative Commons licence and your intended use is not features detected by two or more methods. Furthermore, the permitted by statutory regulation or exceeds the permitted use, you will augmentation of spectroscopic data with chromatographic/ need to obtain permission directly from the copyright holder. To view a electrophoretic data might be interesting. copy of this licence, visit http://cr eativ ecommons. or g/licen ses/ b y/4.0/ . The AriumMS tool presented here was used to process datasets obtained from HILIC-MS and CE-MS measure- ments of a multi-analyte standard and yeast extracts. All References CE-MS methods show in general narrow peaks and benefit from the sensitivity of the nanoCEasy interface. Addition- 1. Ivanisevic J, Want EJ. From samples to insights into metabolism: ally, the valving functionality of the nanoCEasy interface uncovering biologically relevant information in LC-HRMS metab- olomics data. Metabolites. 2019. https:// doi. org/ 10. 3390/ metab enables capillary reconditioning between the injections. o9120 308. HILIC-MS and CE-MS show a large overlap with only a few 2. Comte B, Monnerie S, Brandolini-Bunlon M, Canlet C, Castelli analytes detected by only one of the methods. HILIC-MS F, Chu-Van E, Colsch B, Fenaille F, Joly C, Jourdan F, Lenuzza also covers neutral polar analytes, such as carbohydrates, N, Lyan B, Martin J-F, Migné C, Morais JA, Pétéra M, Poupin N, Vinson F, Thevenot E, Junot C, Gaudreau P, Pujos-Guillot whereas the basic BGE in particular allows the separation E. Multiplatform metabolomics for an integrative exploration of 1 3 Naumann L. et al. metabolic syndrome in older men. EBioMedicine. 2021. https:// mouse model of polycystic kidney disease. Sci Rep. 2019. https:// doi. org/ 10. 1016/j. ebiom. 2021. 103440.doi. org/ 10. 1038/ s41598- 018- 37512-8. 3. Pezzatti J, Boccard J, Codesido S, Gagnebin Y, Joshi A, Picard D, 19. Naumann L, Schairer J, Höchsmann A, Naghdi E, Neusüß C. Cap- González-Ruiz V, Rudaz S. Implementation of liquid chromatog- illary electrophoresis–mass spectrometry interfacing: principles raphy-high resolution mass spectrometry methods for untargeted and recent developments. In: Ramautar R, Chen DDY, editors. metabolomic analyses of biological samples: A tutorial. Anal Capillary electrophoresis-mass spectrometry for proteomics and Chim Acta. 2020. https:// doi. org/ 10. 1016/j. aca. 2019. 12. 062. metabolomics. Principles and applications, vol. 2021. Weinheim: 4. Jedrychowski MP, Huttlin EL, Haas W, Sowa ME, Rad R, Gygi Wiley-VCH; 2022. p. 1–33. SP. Evaluation of HCD- and CID-type fragmentation within their 20. Schlecht J, Stolz A, Hofmann A, Gerstung L, Neusüß C. nano- respective detection platforms for murine phosphoproteomics. CEasy: an easy, flexible, and robust nanoflow sheath liquid cap- MCP. 2011. https:// doi. org/ 10. 1074/ mcp. M111. 009910. illary electrophoresis-mass spectrometry interface based on 3D 5. Sentandreu E, Peris-Díaz MD, Sweeney SR, Chiou J, Muñoz printed parts. Anal Chem. 2021. https://doi. or g/10. 1021/ acs. analc N, Tiziani S. A survey of orbitrap all ion fragmentation analysis hem. 1c032 13. assessed by an R metabolist package to study small-molecule metab- 21. Naumann L, Schlossbauer P, Klingler F, Hesse F, Otte K, Neusüß olites. Chromatographia. 2018. https:// doi. org/ 10. 1101/ 257147. C. High-throughput glycosylation analysis of intact monoclonal 6. Kohler I, Verhoeven M, Haselberg R, Gargano AFG. Hydrophilic antibodies by mass spectrometry coupled with capillary electro- interaction chromatography – mass spectrometry for metabo- phoresis and liquid chromatography. J Sep Sci. 2022. https:// doi. lomics and proteomics: state-of-the-art and current trends. Micro-org/ 10. 1002/ jssc. 20210 0865. chem J. 2022. https:// doi. org/ 10. 1016/j. microc. 2021. 106986. 22. Höcker O, Knierman M, Meixner J, Neusüß C. Two capillary 7. Ibáñez C, Simó C, García-Cañas V, Gómez-Martínez A, Ferragut approach for a multifunctional nanoflow sheath liquid interface JA, Cifuentes A. CE/LC-MS multiplatform for broad metabolomic for capillary electrophoresis-mass spectrometry. Electrophoresis. analysis of dietary polyphenols effect on colon cancer cells prolifera- 2021. https:// doi. org/ 10. 1002/ elps. 20200 0169. tion. Electrophoresis. 2012. https://d oi.o rg/1 0.1 002/e lps.2 01200 143. 23. Johnson CH, Ivanisevic J, Siuzdak G. Metabolomics: beyond bio- 8. Zhang Y-Y, Zhang Q, Zhang Y-M, Wang W-W, Zhang L, Yu Y-J, markers and towards mechanisms. Nat Rev Mol Cell Biol. 2016. Bai C-C, Guo J-Z, Fu H-Y, She Y. A comprehensive automatic https:// doi. org/ 10. 1038/ nrm. 2016. 25. data analysis strategy for gas chromatography-mass spectrometry 24. Suhre K, Meisinger C, Döring A, Altmaier E, Belcredi P, Gieger based untargeted metabolomics. J Chromatogr A. 2020. https:// C, Chang D, Milburn MV, Gall WE, Weinberger KM, Mewes doi. org/ 10. 1016/j. chroma. 2019. 460787. H-W, Hrabé de Angelis M, Wichmann H-E, Kronenberg F, Adam- 9. Höcker O, Flottmann D, Schmidt TC, Neusüß C. Non-targeted LC-MS ski J, Illig T. Metabolic footprint of diabetes: a multiplatform and CE-MS for biomarker discovery in bioreactors: influence of metabolomics study in an epidemiological setting. PloS one. 2010. separation, mass spectrometry and data processing tools. Sci Total https:// doi. org/ 10. 1371/ journ al. pone. 00139 53. Environ. 2021. https:// doi. org/ 10. 1016/j. scito tenv. 2021. 149012. 25. Xuan Q, Ouyang Y, Wang Y, Liang Wu, Li H, Luo Y, Zhao X, 10. Hirayama A, Wakayama M, Soga T. Metabolome analysis based Feng D, Qin W, Chunxiu Hu, Zhou L, Liu X, Zou H, Cai C, Jiarui on capillary electrophoresis-mass spectrometry. Trends Analyt Wu, Jia W, Guowang Xu. Multiplatform metabolomics reveals Chem. 2014. https:// doi. org/ 10. 1016/j. trac. 2014. 05. 005. novel serum metabolite biomarkers in diabetic retinopathy sub- 11. Kok MGM, Somsen GW, de Jong GJ. The role of capillary electrophore- jects. Adv Sci Lett. 2020. https:// doi. or g/ 10. 1002/ adv s. 20200 sis in metabolic profiling studies employing multiple analytical techniques. 1714. Trends Analyt Chem. 2014. https:// doi. org/ 10. 1016/j. trac. 2014. 06. 004. 26. Büscher JM, Czernik D, Ewald JC, Sauer U, Zamboni N. Cross-plat- 12. Soga T, Ohashi Y, Ueno Y, Naraoka H, Tomita M, Nishioka T. Quan- form comparison of methods for quantitative metabolomics of primary titative metabolome analysis using capillary electrophoresis mass spec- metabolism. Anal Chem. 2009. https:// doi. org/ 10. 1021/ ac802 2857. trometry. J Proteome Res. 2003. https:// doi. org/ 10. 1021/ pr034 020m. 27. Li Y, Hou G, Zhou H, Wang Y, Tun HM, Zhu A, Zhao J, Xiao 13. Soga T, Kakazu Y, Robert M, Tomita M, Nishioka T. Qualitative F, Lin S, Liu D, Zhou D, Mai L, Zhang L, Zhang Z, Kuang L, and quantitative analysis of amino acids by capillary electropho- Guan J, Chen Q, Wen L, Zhang Y, Zhuo J, Li F, Zhuang Z, Chen resis-electrospray ionization-tandem mass spectrometry. Electro- Z, Luo L, Liu D, Chen C, Gan M, Zhong N, Zhao J, Ren Y, Xu phoresis. 2004. https:// doi. org/ 10. 1002/ elps. 20030 5791. Y. Multi-platform omics analysis reveals molecular signature 14. Monton MRN, Soga T. Metabolome analysis by capillary electro- for COVID-19 pathogenesis, prognosis and drug target discov- phoresis–mass spectrometry. J Chromatogr A. 2007. https:// doi. ery. Signal Transduct Target Ther. 2021. https:// doi. org/ 10. 1038/ org/ 10. 1016/j. chroma. 2007. 02. 065.s41392- 021- 00508-4. 15. Sawada H, Nogami C. Capillary electrophoresis–electrospray ioniza- 28. Ortiz-Villanueva E, Benavente F, Piña B, Sanz-Nebot V, Tauler R, tion mass spectrometry using uncoated fused-silica capillaries and Jaumot J. Knowledge integration strategies for untargeted metabo- alkaline buffer solution for the analysis of small carboxylic acids. lomics based on MCR-ALS analysis of CE-MS and LC-MS data. Anal Chim Acta. 2004. https:// doi. org/ 10. 1016/j. aca. 2003. 11. 047. Anal Chim Acta. 2017. https://doi. or g/10. 1016/j. aca. 2017. 04. 049 . 16. Yamamoto M, Ly R, Gill B, Zhu Y, Moran-Mirabal J, Britz-McK- 29. Rivera-Pérez A, Romero-González R, Garrido FA. Application of ibbin P. Robust and high-throughput method for anionic metabo- an innovative metabolomics approach to discriminate geographi- lite profiling: preventing polyimide aminolysis and capillary cal origin and processing of black pepper by untargeted UHPLC- breakages under alkaline conditions in capillary electrophoresis- Q-Orbitrap-HRMS analysis and mid-level data fusion. Food Res mass spectrometry. Anal Chem. 2016. https://doi. or g/10. 1021/ acs. Int. 2021. https:// doi. org/ 10. 1016/j. foodr es. 2021. 110722. analc hem. 6b032 69. 30. Letertre MPM, Dervilly G, Giraudeau P. Combined nuclear mag- 17. de Macedo AN, Jiwa MIY, Macri J, Belostotsky V, Hill S, Britz- netic resonance spectroscopy and mass spectrometry approaches McKibbin P. Strong anion determination in biological fluids by for metabolomics. Anal Chem. 2021. https://doi. or g/10. 1021/ acs. capillary electrophoresis for clinical diagnostics. Anal Chem. analc hem. 0c043 71. 2013. https:// doi. org/ 10. 1021/ ac402 975q. 31. Smilde AK, van der Werf MJ, Bijlsma S, van der Werff-van der Vat 18. Sánchez-López E, Kammeijer GSM, Crego AL, Marina ML, BJC, Jellema RH. Fusion of mass spectrometry-based metabolomics Ramautar R, Peters DJM, Mayboroda OA. Sheathless CE-MS data. Anal Chem. 2005. https:// doi. org/ 10. 1021/ ac051 080y. based metabolic profiling of kidney tissue section samples from a 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… 32. Dai S, Lin Z, Xu B, Wang Y, Shi X, Qiao Y, Zhang J. Metabolomics 48. Du P, Kibbe WA, Lin SM. Improved peak detection in mass spectrum data fusion between near infrared spectroscopy and high-resolution mass by incorporating continuous wavelet transform-based pattern match- spectrometry: a synergetic approach to boost performance or induce con- ing. Bioinf. 2006. https:// doi. org/ 10. 1093/ bioin forma tics/ btl355. fusion. Talanta. 2018. https:// doi. org/ 10. 1016/j. talan ta. 2018. 07. 030. 49. Wahab MF, O’Haver TC. Wavelet transforms in separation science 33. Roussel S, Bellon-Maurel V, Roger J-M, Grenier P. Fusion of aroma, for denoising and peak overlap detection. J Sep Sci. 2020. https:// FT-IR and UV sensor data based on the Bayesian inference application doi. org/ 10. 1002/ jssc. 20200 0013. to the discrimination of white grape varieties. Chemometr Intell Lab Syst. 50. Rupprecht F, Enge S, Schmidt K, Gao W, Miller R. Automating 2003. https:// doi. org/ 10. 1016/ S0169- 7439(02) 00111-9. LC-MS/MS mass chromatogram quantification: wavelet transform 34. Khaleghi B, Khamis A, Karray FO, Razavi SN. Multisensor data based peak detection and automated estimation of peak boundaries and fusion: A review of the state-of-the-art. Inf Fusion. 2013. https:// signal-to-noise ratio using signal processing methods. Biomed Signal doi. org/ 10. 1016/j. inffus. 2011. 08. 001. Process Control. 2022. https:// doi. org/ 10. 1016/j. bspc. 2021. 103211. 35. Azcarate SM, Ríos-Reina R, Amigo JM, Goicoechea HC. Data 51. Ju R, Liu X, Zheng F, Zhao X, Lu X, Zeng Z, Lin X, Xu G. handling in data fusion: methodologies and applications. Trends Removal of false positive features to generate authentic peak table Analyt Chem. 2021. https:// doi. org/ 10. 1016/j. trac. 2021. 116355. for high-resolution mass spectrometry-based metabolomics study. 36. Casian T, Nagy B, Kovács B, Galata DL, Hirsch E, Farkas A. Anal Chim Acta. 2019. https://doi. or g/10. 1016/j. aca. 2019. 04. 011 . Challenges and opportunities of implementing data fusion in pro- 52. Shannon CE. A mathematical theory of communication. Bell Syst cess analytical technology-a review. Molecules. 2022. https://doi. Tech J. 1948. https://d oi.o rg/1 0.1 002/j.1 538-7 305.1 948.t b0091 7.x. org/ 10. 3390/ molec ules2 71548 46. 53. van den Berg RA, Hoefsloot HCJ, Westerhuis JA, Smilde AK, van 37. Tautenhahn R, Böttcher C, Neumann S. Highly sensitive feature der Werf MJ. Centering, scaling, and transformations: improving detection for high resolution LC/MS. BMC Bioinf. 2008. https:// the biological information content of metabolomics data. BMC doi. org/ 10. 1186/ 1471- 2105-9- 504. Genom. 2006. https:// doi. org/ 10. 1186/ 1471- 2164-7- 142. 38. Tautenhahn R, Patti GJ, Rinehart D, Siuzdak G. XCMS Online: a 54. Smucker B, Krzywinski M, Altman N. Optimal experimental design. web-based platform to process untargeted metabolomic data. Anal Nat Methods. 2018. https:// doi. org/ 10. 1038/ s41592- 018- 0083-2. Chem. 2012. https:// doi. org/ 10. 1021/ ac300 698c. 55. Jacyna J, Kordalewska M, Markuszewski MJ. Design of experi- 39. Xia J, Psychogios N, Young N, Wishart DS. MetaboAnalyst: ments in metabolomics-related studies: an overview. J Pharm a web server for metabolomic data analysis and interpretation. Biomed Anal. 2019. https:// doi. org/ 10. 1016/j. jpba. 2018. 11. 027. Nucleic Acids Res. 2009. https:// doi. org/ 10. 1093/ nar/ gkp356. 56. Gorrochategui E, Jaumot J, Tauler R. ROIMCR: a powerful analy- 40. Chong J, Xia J. MetaboAnalystR: an R package for flexible and sis strategy for LC-MS metabolomic datasets. BMC Bioinform. reproducible analysis of metabolomics data. Bioinf. 2018. https:// 2019. https:// doi. org/ 10. 1186/ s12859- 019- 2848-8. doi. org/ 10. 1093/ bioin forma tics/ bty528. 57. González-Ruiz V, Gagnebin Y, Drouin N, Codesido S, Rudaz S, 41. Pluskal T, Castillo S, Villar-Briones A, Oresic M. MZmine 2: Schappler J. ROMANCE: A new software tool to improve data modular framework for processing, visualizing, and analyzing robustness and feature identification in CE-MS metabolomics. mass spectrometry-based molecular profile data. BMC Bioinf. Electrophoresis. 2018. https:// doi. org/ 10. 1002/ elps. 20170 0427. 2010. https:// doi. org/ 10. 1186/ 1471- 2105- 11- 395. 58. Pinu FR, Goldansaz SA, Jaine J. Translational metabolomics: 42. Entian K-D, Kötter P. 25 Yeast genetic strain and plasmid collec- current challenges and future opportunities. Metabolites. 2019. tions. In: Stansfield I, editor. Yeast gene analysis, vol. 36. 2nd ed. https:// doi. org/ 10. 3390/ metab o9060 108. Amsterdam: Elsevier; 2007. p. 629–66. 59. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a prac- 43. Chambers MC, Maclean B, Burke R, Amodei D, Ruderman DL, tical and powerful approach to multiple testing. J R Stat Soc Ser B Neumann S, Gatto L, Fischer B, Pratt B, Egertson J, Hoff K, Kes- Methodol. 1995. https://doi. or g/10. 1111/j. 2517- 6161. 1995. tb020 31.x . sner D, Tasman N, Shulman N, Frewen B, Baker TA, Brusniak M-Y, Paulse C, Creasy D, Flashner L, Kani K, Moulding C, Sey- Publisher's note Springer Nature remains neutral with regard to mour SL, Nuwaysir LM, Lefebvre B, Kuhlmann F, Roark J, Rainer jurisdictional claims in published maps and institutional affiliations. P, Detlev S, Hemenway T, Huhmer A, Langridge J, Connolly B, Chadick T, Holly K, Eckels J, Deutsch EW, Moritz RL, Katz JE, Agus DB, MacCoss M, Tabb DL, Mallick P. A cross-platform toolkit for mass spectrometry and proteomics. Nat Biotechnol. Lukas Naumann is currently a 2012. https:// doi. org/ 10. 1038/ nbt. 2377. PhD student in the research 44. Pedrioli PGA, Eng JK, Hubley R, Vogelzang M, Deutsch EW, group of Prof. Christian Neusüß Raught B, Pratt B, Nilsson E, Angeletti RH, Apweiler R, Cheung at Aalen University. He has K, Costello CE, Hermjakob H, Huang S, Julian RK, Kapp E, experience in various analytical McComb ME, Oliver SG, Omenn G, Paton NW, Simpson R, Smith techniques such as LC-MS and R, Taylor CF, Zhu W, Aebersold R. A common open representa- CE-MS. He has been working on tion of mass spectrometry data and its application to proteomics the development of analytical research. Nat Biotechnol. 2004. https:// doi. org/ 10. 1038/ nbt10 31. workflows for the characteriza- 45. Kaever A, Landesfeind M, Possienke M, Feussner K, Feussner I, tion of intact mAb glycosylation Meinicke P. MarVis-Filter: ranking, filtering, adduct and isotope by LC-MS and CE-MS and for correction of mass spectrometry data. J Biomed Biotechnol. 2012. the determination of metabolic https:// doi. org/ 10. 1155/ 2012/ 263910. changes in miRNA-overexpress- 46. Keller BO, Sui J, Young AB, Whittal RM. Interferences and con- ing cell pools. His PhD project is taminants encountered in modern mass spectrometry. Anal Chim focused on the high-throughput Acta. 2008. https:// doi. org/ 10. 1016/j. aca. 2008. 04. 043. analysis of antibody glycosyla- 47. Andrade L, Manolakos ES. Signal background estimation and tion and the related metabolome resulting from microRNA transfection baseline correction algorithms for accurate DNA sequencing. J in CHO cells. VLSI Signal Process Syst Signal Image Video Technol. 2003. https:// doi. org/ 10. 1023/B: VLSI. 00000 03022. 86639. 1f. 1 3 Naumann L. et al. Adrian Haun has been a research Dirk Flottmann became Professor scientist at the University of for analytical chemistry at Aalen Aalen in Germany since 2020. University, Germany, in April He is developing AriumMS with 2002. He employs elemental the aim of creating a powerful mass spectrometry and coupling yet user-friendly, all-in-one soft- techniques to study mainly the ware program for metabolomics distribution of trace elements in applications. His research semiconductor samples. He is focuses on the application of interested in the application of concepts from various disci- multivariate data analysis tech- plines including mathematics, niques and develops new physics and computer science in approaches to characterize the the field of analytical chemistry. metabolome in different kinds of samples to improve the non-tar- get approach. Christian Neusüß is Professor in Alisa Höchsmann is currently a the Faculty of Chemistry at PhD student in the research Aalen University. He received group of Prof. Christian Neusüß his diploma from University of at Aalen University. The subject Heidelberg and his PhD from of her master’s thesis was the University of Leipzig. His separation of cationic metabo- research interests include cou- lites using capillary electropho- pling and application of (electro- resis-mass spectrometry. Her migrative) separation techniques PhD project is related to the with mass spectrometry. He is separation of intact proteins by focusing on technical develop- capillary electrophoresis-mass ments such as nanoESI inter- spectrometry using successive faces and miniaturized two- multiple ionic-polymer layer dimensional separations in (SMIL)-coated capillaries. combination with high-resolu- tion mass spectrometry as well Michael Mohr (BSc) is currently as method development and applications for the analysis of proteins, a master’s student at Aalen Uni- metabolites and contaminants. He is co-author of more than 110 peer- versity and did his bachelor reviewed publications (h-Index = 41). thesis in the research group of Prof. Christian Neusüß during the summer term 2022. During his bachelor’s thesis, he gained expertise in the generation and non-target evaluation of aug- mented HILIC- and CE-MS metabolomic datasets of yeast samples, employing chemomet- ric and statistical tools. Also, he supported the development of AriumMS by extensive use of the software. Martin Novák (MSc) is currently study director for GC-MS, GC-MS/MS and LC-MS/MS analytics at Eurofins Agrosci- ence Services GmbH. He com- pleted his master’s thesis at Aalen University in the research group of Prof. Christian Neusüß. The subject of his master’s thesis was the analysis of anionic metabolites with CE-MS in acidic and basic background electrolytes. 1 3 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Analytical and Bioanalytical Chemistry Springer Journals

Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of multi-platform-based CE-MS and LC-MS data

Loading next page...
 
/lp/springer-journals/augmented-region-of-interest-for-untargeted-metabolomics-mass-KQhWaJwFP5

References (62)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2023
ISSN
1618-2642
eISSN
1618-2650
DOI
10.1007/s00216-023-04715-6
Publisher site
See Article on Publisher Site

Abstract

In mass spectrometry (MS)-based metabolomics, there is a great need to combine different analytical separation techniques to cover metabolites of different polarities and apply appropriate multi-platform data processing. Here, we introduce Ari- umMS (augmented region of interest for untargeted metabolomics mass spectrometry) as a reliable toolbox for multi-platform metabolomics. AriumMS offers augmented data analysis of several separation techniques utilizing a region-of-interest algorithm. To demonstrate the capabilities of AriumMS, five datasets were combined. This includes three newly developed capillary electrophoresis (CE)-Orbitrap MS methods using the recently introduced nanoCEasy CE-MS interface and two hydrophilic interaction liquid chromatography (HILIC)-Orbitrap MS methods. AriumMS provides a novel mid-level data fusion approach for multi-platform data analysis to simplify and speed up multi-platform data processing and evaluation. The key feature of AriumMS lies in the optimized data processing strategy, including parallel processing of datasets and flexible parameterization for processing of individual separation methods with different peak characteristics. As a case study, Saccharomyces cerevisiae (yeast) was treated with a growth inhibitor, and AriumMS successfully differentiated the metabolome based on the augmented multi-platform CE-MS and HILIC-MS investigation. As a result, AriumMS is pro- posed as a powerful tool to improve the accuracy and selectivity of metabolome analysis through the integration of several HILIC-MS/CE-MS techniques. Keywords Augmented data evaluation · Mid-level data fusion · Multi-platform metabolomics · nanoCEasy · Capillary electrophoresis · Hydrophilic interaction liquid chromatography Abbreviations DOE Design of experiment A.u. Arbitrary units EIC/E Extracted ion chromatogram/ AriumMS Augmented region of interest for untargeted electropherogram metabolomics mass spectrometryEOF Electro-osmotic flow BGE Background electrolytesESI Electrospray ionization BPC/E Base peak chromatogram/electropherogramFC Fold change CE Capillary electrophoresisGC Gas chromatography CWT Continuous wavelet transform HCD Higher-energy collisional dissociation DMSO Dimethyl sulfoxide HILIC Hydrophilic interaction liquid chromatography MS Mass spectrometry NAD Nicotinamide adenine dinucleotide Lukas Naumann and Adrian Haun contributed equally. PCA Principal component analysis * Christian Neusüß PC Principal component Christian.Neusuess@hs-aalen.de QTOF Quadrupole time of flight ROI Region of interest Department of Chemistry, Aalen University, Beethovenstraße 1, 73430 Aalen, Germany Vol.:(0123456789) 1 3 Naumann L. et al. RP-LC Reversed-phase liquid chromatography acidic BGEs for cation analysis [12–14] and basic BGEs for sd Standard deviation anion analysis [15–17]. In order to improve the sensitivity SD Synthetic dextrose minimal medium for metabolite analysis by CE-MS, nanoESI interfaces have SL Sheath liquid recently been used, including the porous tip interface [18]. Nanosheath–liquid interfaces are of high interest as well, due to additional flexibility and robustness [19]. Most recently, Introduction we introduced the nanoCEasy interface adding ease-of-use and the capability of valve functionality by the two-capillary Owing to the inherent chemical diversity and the large size approach (e.g., for capillary reconditioning between runs) of the metabolome, there is no universal technique that can [20–22]. be used to assess the entire metabolome, i.e., “one size does Metabolomics data evaluation is usually based on two not fit all” [1 , 2]. Nevertheless, multi-platform metabolomics major approaches: target and non-target data evaluation workflows based on mass spectrometry (MS) are able to [3]. Target-based data evaluation is hypothesis-driven and enhance metabolome coverage. focuses with high analytical sensitivity on standard mixtures Typically, scientists employ high-resolution electrospray for their assignment and interpretation, such as concentra- ionization–MS (ESI-MS) with the possibility of MS/MS tion and appearance [3, 23]. Non-target metabolomics is an experiments such as quadrupole time-of-flight (QTOF) exploratory, hypothesis-generating data evaluation workflow or Orbitrap MS [3–5]. Depending on the type of metabo- [3]. This approach is a common choice as a first step within lites to be measured (polar vs. nonpolar) and limitations a data evaluation, to capture and monitor a broad range of concerning time and sample amount, different separation molecular content and retrieve as much chemical informa- techniques can be applied for the analysis to expand the tion as possible without any prior knowledge [3]. Examples metabolome coverage [1]. These are reversed-phase liquid of multi-platform metabolomics can be found in previous chromatography (RP-LC) [3], hydrophilic interaction liq- publications [7, 24–28]. Most of them use a target/non-target uid chromatography (HILIC) [6], capillary electrophore- approach based on different data processing workflows for sis (CE) [7], and gas chromatography (GC) [8] coupled to each analytical platform. high-resolution MS [3, 9]. The analytical gold standard in Since LC-MS and CE-MS offer comprehensive infor- proteomics and metabolomics is RP-LC-MS because of its mation of the metabolome, a combined multi-platform extended dynamic concentration range, sensitivity, reten- non-targeted data evaluation based on a data fusion tion time reproducibility, and ease of use [6]. Since RP-LC approach offers a single chemometric result for enhanced does not retain very well a wide variety of highly polar and statistical prediction and metabolic coverage [29–31]. The ionizable metabolites, HILIC is a valuable alternative [6]. fusion of separation methods coupled with MS detection HILIC is driven by molecular interactions and the partition is challenging due to the multivariate nature of the data of analytes between the hydrophobic mobile phase and the (i.e., a very high variables-to-sample ratio, and shift in hydrophilic stationary phase [10]. Significant technologi- migration times during sequences) [31, 32]. Hence, an cal advances in HILIC over the last two decades, such as augmented data evaluation of comprehensive analytical the commercialization of dedicated HILIC columns, have workflows enhances the feature capacity by combining aided the implementation of HILIC in proteomics and different selectivities, thereby allowing a better charac- metabolomics [6]. Overall, this has resulted in significant terization of phenotypes. analytical improvements (e.g., sensitivity, analyte cover- Data augmentation describes the combination of sev- age, throughput, analysis speed, and resolution), and thus eral datasets into one. Three cases can be distinguished HILIC offers excellent opportunities for the analysis of [31, 33, 34]: Low-level fusion is applied before any data polar and/or ionizable metabolites [6]. reduction, and mid-level fusion after feature extraction, Since many metabolites, especially those of central car- whereas high-level data fusion combines models after bon metabolism, contain charged amino, hydroxyl, car- data analysis [35]. Mid-level data fusion is based on boxyl, and phosphate groups, they are especially suitable removing irrelevant information, such as artifacts and for CE-MS analysis [10]. Electrophoretic-driven separation noise, from each dataset. The resulting dimensionality approaches offer several advantages for the separation of reduction decreases computation time and can produce charged compounds, like efficient separation, high resolv - more robust models [36]. ing power, low solvent, and sample consumption. Since CE Region-of-interest (ROI) analysis is the approach separation is based on differences in ion mobilities [10, 11], of choice and significantly reduces both the amount different compositions of background electrolytes (BGE), of data—without loss of relevant information—and especially regarding pH, lead to different selectivities. In processing time. Only data points that have a mini- this way, CE-MS analysis has been frequently applied using mum intensity and a minimum abundance within the 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… measurement are included in an ROI [37]. Peaks are Yeast growth and sample preparation then detected, integrated, and labeled (m/z, retention time, peak area, and height) in the obtained ROIs. Vari- Production of yeast liquid cultures was carried out with ous filters (e.g., contaminant filter, isotope, and adduct Saccharomyces cerevisiae strain CEN.PK122 [42], start- filter) are then used to remove false-positive features ing from a single colony grown on SD plates. Growth took from the feature list. This method is widely used in the place in an incubation shaker in a 5 L baffle flask under web-based tools XCMS Online [38] and MetaboAnalyst controlled conditions (30 °C, 123 rpm, 16 h). SD medium [39] as well as in various software packages such as was used as a basal medium. The culture was split into two MetaboAnalystR [40] or the open-source MZmine [41]. cultures (Mock and Effect1) at 0.5 optical density. Cell line However, the focus of these programs is not on the aug- Effect1 (160 mL) contained 160 µL 35 mM halogenated mentation of different separation techniques. For exam- indole dilution in dimethyl sulfoxide (DMSO) to induce ple, it is not possible to select different preprocessing the effect. In order to determine the induced effect of the settings for different data, which is essential for different halogenated indole exactly, Mock (160 mL) as a reference separation systems. With XCMS Online and MZmine, was treated exactly the same as Effect1 (160 µL DMSO), files of different origins must be processed separately without adding the halogenated indole. Incubation at 30 °C and augmented manually afterward. and 170 rpm monitored by optical density readings every Here, we present the novel open-source AriumMS 30–60 min was performed until the inhibition of the cell (augmented region of interest for untargeted metabo- growth became apparent. Thereafter, cells were harvested lomics mass spectrometry) software to challenge the and centrifuged. The cell pellets were washed and shock- multi-platform metabolomics data analysis in combina- frozen (at −80 °C). Further sample preparation is given in tion with new methods for the analysis of polar metabo- the supplements. lites by different CE and HILIC separation techniques. AriumMS contains a universal and user-friendly toolbox, Capillary electrophoresis capable of handling multi-platform datasets. AriumMS offers automated batch processing with f lexible process- CE-ESI-MS was performed with a 7100 capillary elec- ing options and a graphical user interface. The suitability of this tool for multi-platform metabolomics is demon- trophoresis system (model no. G7100A) from Agilent Technologies (Waldbronn, Germany) coupled with an strated within a comparative study of metabolic standard mixtures and different yeast phenotypes. Therefore, the Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific, San Jose CA, USA) using the nano- metabolic standard mixtures and the yeast extracts were measured within a multi-platform approach combining CEasy interface [20]. Bare fused silica capillaries with 50/100 µm inner diameter and 360/240 µm outer diam- HILIC-MS (ESI positive/ESI negative) with three CE-MS methods applying our recently introduced nanoCEasy eter (separation/sheath liquid capillary) were obtained from Polymicro Technologies (Phoenix, AZ, USA). interface. A cationic CE-MS method was complemented by two CE-MS methods to cover a wide range of anionic Separation capillaries had a length of 90 cm and were etched with hydrof luoric acid to about an 80–100 µm metabolites. outer diameter. Three CE-MS methods have been used with the following background electrolyte (BGE) and Materials and methods sheath liquid (SL) compositions: anionic (acidic): 0.2 M formic acid pH 2.1 (BGE) and 50:50 (v/v) 2-propanol/ Materials water with 0.5% (v/v) of formic acid (SL); anionic (alkaline): 30  mM ammonium acetate pH  8.5 (BGE) The amino acid standard (1 nmol/µL in 0.1 M hydrochlo- and 50:50 (v/v) 2-propanol/water with 2.5 mM ammo- nium acetate (SL); and cationic (acidic): 1  M formic ric acid) was obtained from Agilent Technologies (Santa Clara, CA, USA). The internal standards and metabo- acid containing 10% 2-propanol pH 1.7 (BGE) and 50:50 (v/v) 2-propanol/water with 0.5% (v/v) of formic acid lites used were obtained from Sigma-Aldrich (St. Louis, MO, USA). Sugars (nucleotide sugars, phosphate sug- (SL). For each measurement, the capillary was precondi- tioned by f lushing with BGE for 5 min. For the alkaline ars) were purchased from Biosynth Carbosynth (Staad, Switzerland). Synthetic Dextrose Minimal Medium (SD, CE method, the capillary was additionally primed for 5 min applying 30 kV, and again f lushing with BGE for synthetic minimal medium) was obtained from Carl Roth (Karlsruhe, Germany). Standard materials and composi- 5  min. Samples were injected hydrodynamically with 40 mbar for 27 s (1% capillary volume). Separation was tion of the metabolomics standard can be found in the supplements. performed by applying a potential of +30 kV (cationic 1 3 Naumann L. et al. acidic and anionic alkaline BGE method) or −30  kV 54 ms accumulation time, AGC target set to “standard,” (anionic acidic BGE method) to the capillary inlet. SL 35% RF lens, and 1 micro scan. was delivered via a syringe pump (100 series, kdScien- tific, Hilliston, MA, USA) with a f low rate of 10 µL/ min, equipped with a 5  mL syringe (SGE Analytical Data evaluation and interpretation Science, Melbourne, Australia). The anionic acidic and anionic alkaline CE-MS methods were detected in ESI Data acquisition was performed using a Thermo Scientific negative mode. The cationic acidic CE-MS method was Xcalibur 4.1.50 and Orbitrap Tribrid MS Series Instrument detected in ESI positive mode. Source parameters for Control Software version 3.2 (Thermo Fisher Scientific, San Orbitrap were set to −1700 V/−2000 V/1900 V (anionic Jose CA, USA). Extraction of ion traces for the evaluation acidic/anionic alkaline/cationic acidic) spray voltage, of separation methods was done with FreeStyle 1.5.93.34 3 a.u. (arbitrary units) sheath gas, 0 a.u. aux gas, and (Thermo Fisher Scientific, San Jose CA, USA). MSconvert 300 °C ion transfer tube. 3 (ProteoWizard, Palo Alto CA, USA) [43] was used for the initial data conversion. Non-target data evaluation was Hydrophilic interaction liquid chromatography performed with AriumMS 1.0.0 (https:// github. com/ Adria nHaun/ Ar ium MS/). Software and parameters for evalua- A Dionex UltiMate 3000 (Dionex, Sunnyvale, CA, USA) tion are given in the Supporting Information (supplement high-performance liquid chromatography (HPLC) sys- Table S1 and Table S2). tem equipped with a VDSpher PUR 100 HILIC guard and separation column (4.2 × 10  mm and 150 × 3  mm, 5 µm particle size, VDS optilab Chromatographietech- nik GmbH, Berlin, Germany) heated to 30 °C was used. Results and discussion Mobile phase A was composed of H O, acetonitrile (95/5 v/v), and 5 mM ammonium acetate, and mobile phase B Study design was composed of H O, acetonitrile (5/95 v/v), and 5 mM ammonium acetate. The sample injection volume was In order to present AriumMS as a toolbox for the challenge 3 µL, and the run time was 35 min. The gradient started of multi-platform metabolomics data analysis, metabolite at 10% A, followed by a 15-min linear gradient from 10 standards and yeast extract samples were measured with five to 60% A, and hold for 5 min. Column re-equilibration analytical methods. The metabolite standard that was used was performed for 15 min at 10% A. The flow rate was contained 36 metabolites, covering important polar/ionic 300 µL/min. The LC was coupled to the Orbitrap with substance classes (mass range of 100–665 Da). The yeast the respective standard heated electrospray ionization extracts contained the metabolic information of the induced (HESI) source and sprayer. The Orbitrap source param- effect by a halogenated indole treatment. To analyze polar eters were set to 3500 V positive/negative spray voltage, and/or ionic metabolites of interest within the samples, two 50 a.u. sheath gas, 10 a.u. aux gas, 325 °C transfer tube, HILIC-MS and three CE-MS methods have been developed. and 350 °C vaporizer temperature. In order to determine optimal AriumMS data processing parameters for the generated datasets of the five analyti- Mass spectrometry cal methods, a D-optimal design of experiment (DOE) was applied for software parameter screening and optimization. For mass spectrometry, an Orbitrap Fusion Lumos mass Furthermore, the feature generation of AriumMS was vali- spectrometer (Thermo Fisher Scientific, San Jose CA, dated. This was followed by a multi-platform metabolomics USA) was used in either positive or negative ion mode, data analysis of the yeast extracts. The complete analytical with a scan range of 100–700 m/z. Resolving power was workflow is shown in Fig.  1. set to 60,000, accumulation time to 50 ms, automatic gain control (AGC) target to “standard,” 35% RF lens, and Evaluation of the analytical methods 1 micro scan. Data-dependent MS/MS experiments with 0.6 s cycle time were performed. Filters were an inten- The standard contained a total of 36 typical polar metabo- sity threshold at 2E4, exclusion after a single occurrence lites and four internal standards, including yeast metab- for 10 s, and isotope exclusion. Data-dependent MS/MS olites, amino acids, hexoses, hexose phosphates, and Orbitrap higher-energy collisional dissociation fragmen- nucleotide sugars. Anionic, cationic, zwitterionic, and tation (HCD) parameters were isolation width of 1.5 da, uncharged species were represented (Table 1). To deter- 20/35/50% HCD power, Orbitrap resolution of 30,000, mine the overall capabilities of the five different analytical 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… Fig. 1 Multi-platform metabolomics workflow overview, containing all steps of sampling, analysis, and AriumMS workflow methods regarding the number of detected analytes and co-migrating neutrals) over a period of 45 min (Table 1). duration of analysis, six repetitions of the metabolomics Using the cation CE-MS method, 17 of 36 metabolites standard were measured with each method. The five sepa - were able to be detected. Neutral metabolites, such as hex- ration methods were evaluated regarding the number of oses and caffeine, were not detected by any of the CE-MS detectable analytes, their migration time (MT)/retention methods. In summary, when the three CE-MS methods time (RT), and their separation efficiency (for further were combined, CE-MS was able to detect 32 of 36 metab- details, see supporting information, Fig. 2a–e, Table 1). olites. When all five analytical methods were applied, all These five separation methods offered overlapping and metabolites of the standard were detectable (Table  1). complementary information on the metabolite standard, Apart from the difference in selectivity, HILIC-MS and as shown in Fig. 2 and Table 1: The two HILIC methods CE-MS each had distinct advantages: HILIC-MS exhib- covered most of the metabolites, i.e., 25 of 36 in ESI+ and ited a higher retention time reproducibility and a higher 27 of 36 in ESI−, respectively, and when they were com- degree of automation, while CE-MS required a smaller bined, 30 of 36 metabolites of the standard were able to sample volume and showed more efficient separation with be detected. Some multi-carboxylic acids and basic amino sharper peaks. acids were not detected, and isomeric hexose phosphates were not baseline-separated. The selectivity of the CE-MS Non‑target data evaluation methods used was higher. The anionic alkaline CE-MS method was able to detect 30 out of 36 metabolites over a AriumMS workflow period of 30 min. Seventeen metabolites (such as neutral amino acids) co-migrated with the electroosmotic flow AriumMS was developed as a universal and user-friendly (EOF, no separation) (see Table  1). The anionic acidic computational metabolomics toolbox to tackle the chal- CE-MS method was capable of analyzing phosphates and lenge of multi-platform MS data analysis. The acronym Ari- dicarboxylic acids and covered 15 of 36 analytes (four umMS stands for augmented region of interest for untargeted 1 3 Naumann L. et al. 1 3 Table 1 List of all target analytes of the metabolomics standard Group Analyte Anion Cation ID Name HILIC CE (alkaline BGE) CE (acidic BGE) HILIC CE − + [M−H] RT ± sd AriumMS MT ± sd AriumMS MT ± sd AriumMS [M+H] RT ± sd AriumMS MT ± sd AriumMS [min] [min] [min] [min] [min] Amino acid 1 L-serine 104.0348 10.8 ± 0.1  ✓ 7.7* ± 0.3  ✓ - - 106.0504 10.8 ± 0.0  ✓ 19.5 ± 0.5  ✓ 2 L-proline 114.0555 11.6 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 116.0712 11.8 ± 0.1  ✓ 21.5 ± 0.6  ✓ 3 L-valine 116.0712 10.1 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 118.0869 10.1 ± 0.0  ✓ 19.6 ± 0.5  ✓ 4 L-threonine 118.0504 10.6 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 120.0661 10.6 ± 0.0  ✓ 20.8 ± 0.6  ✓ 5 L-cysteine 120.0119 - - 7.8* ± 0.3  ✓ - - 122.0276 13.2 ± 0.1  ✓ 22.3 ± 0.7  ✗ 6 L-leucine 130.0869 9.1 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 132.1025 9.1 ± 0.0  ✓ 20.1 ± 0.6  ✓ 7 L-isoleucine 130.0869 9.4 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 132.1025 9.4 ± 0.0  ✓ 20.4 ± 0.6  ✓ 8 L-aspartic acid 132.0297 9.9 ± 0.1  ✓ 15.9 ± 1.1  ✓ 38.6 ± 1.6  ✗ 134.0454 10.0 ± 0.0  ✓ 23.4 ± 0.7  ✓ 9 L-lysine 145.0978 - - 5.4 ± 0.1  ✓ - - 147.1134 15.5 ± 0.5  ✗ 12.7 ± 0.2  ✓ 10 L-glutamic 146.0454 10.2 ± 0.1  ✓ 14.5 ± 0.9  ✓ - - 148.0610 10.2 ± 0.0  ✓ 22.0 ± 0.7  ✓ acid 11 L-methionine 148.0433 9.3 ± 0.1  ✓ 7.8* ± 0.3  ✓ - - 150.0589 9.3 ± 0.0  ✓ 21.3 ± 0.6  ✓ 12 L-histidine 154.0617 - - 7.4 ± 0.2  ✓ - - 156.0773 - - 13.6 ± 0.3  ✓ 13 L-phenylala- 164.0712 8.7 ± 0.4  ✓ 7.8* ± 0.3  ✓ - - 166.0869 8.6 ± 0.0  ✓ 22.2 ± 0.7  ✓ nine 14 L-arginine 173.1039 - - 5.5 ± 0.1  ✓ - - 175.1196 - - 13.3 ± 0.3  ✓ 15 L-tyrosine 180.0661 8.8 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 182.0818 8.8 ± 0.0  ✓ 23.2 ± 0.7  ✓ 16 L-tryptophan 203.0821 8.1 ± 0.0  ✓ 7.8* ± 0.3  ✓ - - 205.0978 8.1 ± 0.0  ✓ 22.3 ± 0.7  ✓ Internal std 17 benzene- 141.0010 3.5 ± 0.1  ✗ - - 12.8 ± 0.2  ✗ 143.0167 - - - - sulfinic acid 18 2-nitrobenzoic 166.0140 3.1 ± 0.0  ✓ 16.3 ± 1.3  ✓ 19.2 ± 0.4  ✓ 168.0297 - - - - acid 19 methionine 180.0331 9.8 ± 0.1  ✓ 7.7* ± 0.3  ✓ - 182.0487 9.9 ± 0.0  ✓ 24.6 ± 0.8  ✓ sulfone 20 pentetic acid 392.1306 10.3 ± 0.1  ✓ 30.7 ± 5.8  ✗ - 394.1462 10.4 ± 0.0  ✓ - - Metabolites 21 succinate 117.0188 7.7 ± 1.0  ✓ - - 37.7 ± 1.5  ✓ 119.0345 - - - - 22 nicotinic acid 122.0242 6.3 ± 0.1  ✓ 17.4 ± 1.4  ✓ 19.2 ± 0.4 ✓  124.0399 6.4 ± 0.0  ✓ 20.0 ± 0.6  ✓ 23 tartaric acid 149.0086 - - - - 30.4 ± 1.0  ✓ 151.0243 - - - - 24 citrate; citric 191.0192 - - - - 32.1 ± 1.1  ✗ 193.0348 - - - - acid 25 caffeine 193.0726 - - - - - - 195.0882 4.0 ± 0.0  ✓ - - 26 ATP 505.9879 10.2 ± 0.1  ✗ 25.3 ± 2.8  ✗ - - 508.0036 10.4 ± 0.0  ✗ - - 27 NAD 662.1013 10.5 ± 0.0  ✓ 10.1 ± 0.5  ✓ 38.5 ± 1.6  ✗ 664.1170 10.6 ± 0.0  ✓ - - 28 NADH 664.1170 8.6 ± 0.0  ✓ 13.6 ± 0.9  ✗ - - 666.1327 8.7 ± 0.0  ✓ - - Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… 1 3 Table 1 (continued) Group Analyte Anion Cation ID Name HILIC CE (alkaline BGE) CE (acidic BGE) HILIC CE − + [M−H] RT ± sd AriumMS MT ± sd AriumMS MT ± sd AriumMS [M+H] RT ± sd AriumMS MT ± sd AriumMS [min] [min] [min] [min] [min] Carbohy- 29 L-fuc., 163.0607 - - 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 165.0763 - - - - drates 6-deoxy-L- gal 30 D-mannose 179.0556 6.5 ± 0.2  ✓ 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 181.0713 - - - - 31 α-D-glucose 179.0556 6.8 ± 0.2  ✓ 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 181.0713 - - - - 32 α-D-galactose 179.0556 7.5 ± 0.6  ✓ 7.8* ± 0.3  ✓ 38.9 ± 1.7  ✗ 181.0713 - - - - 33 D-mannose- 259.0219 10.3 ± 0.0  ✓ 21.9 ± 1.3  ✓ 15.9 ± 0.2  ✓ 261.0376 10.3 ± 0.0  ✗ - - 1-PO 34 β-D-fructose- 259.0219 10.1 ± 0.0  ✓ 19.4 ± 1.1  ✓ 15.8 ± 0.2  ✓ 261.0376 10.1 ± 0.0  ✗ - - 6-PO 35 α-D-glucose 259.0219 9.9 ± 0.0  ✓ 19.0 ± 1.1  ✓ 15.5 ± 0.2  ✓ 261.0376 9.9 ± 0.0  ✗ - - 1-PO 36 α-D-galactose- 259.0219 10.3 ± 0.0  ✓ 18.1 ± 0.9  ✓ 15.5 ± 0.2  ✓ 261.0376 10.3 ± 0.0  ✓ - - 1-PO 37 N-acetylneu- 307.0909 - - 7.7* ± 0.3  ✓ - - 309.1066 - - - - raminate 38 UDP-glucose 565.0472 7.5 ± 0.1  ✓ 14.7 ± 0.9  ✓ 11.7 ± 0.8 567.0629 7.6 ± 0.0  ✓ - - 39 GDP-L-fucose 588.0744 8.7 ± 0.3  ✓ - - - - 590.0901 9.0 ± 0.0  ✓ - - 40 CMP-N- 613.1395 9.0 ± 0.0  ✓ - - - - 615.1552 - - - - acetylneu- raminate If the peaks of the analytes fulfill the general criteria for peak detection (maximum peak width ≤ 2 min, peak intensity ≥ 5E4), then the retention time ± standard deviation (sd) is given. If an analyte is not detected by the method itself or does not fulfill the criteria for peak detection, it is labeled with “-”. If AriumMS is able to detect an analyte, it is labeled with “✓ ”, if not with “✗”. Analytes co-migrating with EOF are labeled with “*”. Analytes detected as co-migrating neutrals are labeled with “#” Naumann L. et al. metabolomics mass spectrometry. It was designed as a multi- divided by the total number of points of the peak. H is cal- tiered software for scalable (parallel processing of multiple culated for each peak, and values greater than the median sample sets) and reproducible data analysis. AriumMS con- entropy are considered as noise and discarded. The peaks of sists of a main app (AriumMS, Fig. S2a) for ROI search, the remaining features after the filtering were integrated, and alignment, and low-level data filtering, and an evaluation the obtained areas are sorted into an N x M x S matrix, where App (AriumMSEval, Fig. S2b) for feature extraction and N corresponds to the retention or migration times (rows), M augmented data analysis. To ensure a high level of MS corresponds to the m/z (columns), and S corresponds to the instrument compatibility, the open-source MS data format repeat measurement (layers). Based on the feature intensi- mzXML is used for raw data import [44]. To achieve opti- ties obtained by the integration, the features can be scaled mal results in data processing, AriumMS uses user-defined along the repeat measurements. Available scaling methods sample groups. A sample group can contain datasets of dif- include center, auto, Pareto, vast, range, and level. Constant ferent separation methods, different analytical workflows, or factors for whole groups and sample-specific factors can be multiple phenotypes of biological samples. An individual applied as well. This allows, for example, normalization to parameterization can be applied for each group, for example, the cell count of the sample or normalization to multiple depending on the different peak characteristics of the sepa- internal standards for metabolite quantitation. Logarithmic ration methods used. Files are then batch-processed group and power transformations are available as well. A guideline by group. To reduce the computation time, a crop filter can for the selection of a proper scaling method is given by van be applied to discard areas of the measurements without den Berg et al. [53]. In the next step, the user-defined groups relevant peaks. Several optional filters minimize the number are augmented by linking the data cubes along the m/z of false-positive features during the ROI phase (e.g., isotope, dimension, combining the m/z and time dimensions into one adduct, and common contaminant filter) [45, 46]. An addi- dimension. The features are now named according to the fol- tional baseline correction removes any drift across the sepa- lowing scheme: “m/z @Time, Group". If features of different ration by baseline determination over a moving window by groups have the identical mass and number of occurrences interpolation [47]. After processing the ROI stage, the data (no. of detections within the groups), they are assumed to be is automatically transferred to the AriumMSEval app. For the same and labeled accordingly. The complete flow chart automatic peak detection, the obtained ROIs are smoothed of the data processing can be found in Fig. S3. in the first step, and the difference between the smoothed ROI and the original ROI is used to estimate the noise level AriumMS parameter screening and optimization for this m/z. The second derivative of the smoothed ROI is formed, and peaks are identified by continuous wavelet In order to obtain good results with the AriumMS software transform (CWT) using the Mexican hat as the mother wave- package, we applied an efficient D-optimal DOE for soft- let [48, 49]. Since real peaks are rarely perfectly symmetric, ware parameter screening and optimization [54, 55]. The the peak boundaries are adjusted by a two-step process; first, D-optimal DOE design enables the identification of optimal via a friction border correction [50] based on the smoothed parameter settings with a lower number of required experi- peak and then via moving standard deviation border correc- ments compared to other designs. For that reason, the six tion based on the original peak. This algorithm can be found repetitions of the metabolomics standard measured by all in the supplement information (Algorithm A1). Within the v fi e analytical methods were evaluated regarding the number corrected limits, the peaks are now integrated, and the reten- of found analytes and the total number of features. Found tion time is determined. Followed by the initial feature fil- target features were defined by their m/z value and respec- tering of the AriumMSEval, possible feature filters are, for tive retention time (parameters for non-target data labeling example, minimum and maximum peak width, minimum are given in the supporting information). According to the height within an ROI, and signal-to-noise ratio (support- results, a total of 126 target features were found. The total ing information Table S2). As an additional approach, an number of target features represents the number of target information entropy peak filter adapted from Ju et al. [51] features detected by the five separation methods, including was integrated. For a Gaussian peak, all points before the internal standards and co-migrating analytes. During the maximum have a constant positive slope, and after the DOE screening, the ROI functions developed by Tauler [56] maximum, a constant negative slope; these points are called were tweaked by disabling the addition of random noise on normal points. Points that deviate from this condition are the extracted ion chromatogram/electropherogram (EIC/E), called variant points. Accordingly, the entropy of a peak since it was not required for AriumMS. By default, this algo- can be expressed by the sum of the entropy of all possible rithm added random noise on the EIC/E to remove possible events H =−p ∗ log (p) − q ∗ log (q) [51, 52], where p is gaps in the data. Here, the addition of random noise to the 2 2 the number of variant points divided by the total number EIC/E created multiple peak tips and increased the peak of points of the peak, and q is the number of normal points splitting within the six repetitions of the standard, which 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… hexose-PO HILIC (ESI -) 7 Time (min) 13 05 10 15 20 25 30 35 40 45 Time (min) hexose-PO CE (alkaline BGE, anion) Time (min) 19 05 10 15 20 25 30 35 40 45 EOF Time (min) hexose-PO CE (acid BGE, anion) 13 19 Time (min) 05 10 15 20 25 30 35 40 45 EOF Time (min) LEU/ILE HILIC (ESI +) Time (min) 12 05 10 15 20 25 30 35 40 45 Time (min) LEU/ILE CE (cation) Time (min) 22 05 10 15 20 25 30 35 40 45 Time (min) Fig. 2 Comparison of different analytical methods using the meas- and L-isoleucine (cation methods). (A) HILIC-MS anion, blue; (B) urements of the metabolomics standard that contains 40 substances. CE-MS anion alkaline, gray; (C) CE-MS anion acidic, yellow; (D) The BPC/Es are shown at A–E containing a zoom to the separation HILIC-MS cation, orange; (E) CE-MS cation, acidic of either the four hexose phosphates (anion methods) or L-leucine induced varying retention times. Peak picking and integra- ensure the comparability between repeated measurements tion of AriumMS were improved by the removal of the arti- for the same method and to counter the effects of analytical ficial distortion of the peak tips by the ROI function. variance, the chromatograms could be alternatively aligned For an initial parameter screening, 14 parameters at in time (recommended especially for CE) and m/z dimen- two levels of both stages of the software (AriumMS, Ari- sions. The mass spectra alignment shifts measured masses to umMSEval) were chosen. The DOE identified the follow - match the most common x quantile of detected masses (e.g., ing parameters as significant for further optimization: ROI 0.95). Since it aligns masses, it offers benefits for QTOF intensity threshold, mass spectra alignment, m/z error, and instruments or for low-resolution MS. In general, resolution feature occurrence l fi ter. The ROI intensity threshold den fi es and calibration of the mass spectrometer must be considered the m/z intensity cutoff limit for noise. In general, a higher for non-target data processing. Hence, the m/z error of the intensity threshold leads to a lower number of found features. ROI should be set properly; here, 0.01 Da represents 10 ppm For example, an increase of the ROI intensity threshold from at 1000 Da (upper m/z limit). One of the most important 50,000 cts. to 150,000 cts. roughly loses 10% of total fea- feature filters of AriumMS is the minimum feature occur - tures/target features (HILIC ESI). As a universal robust ROI rence, which is defined as the relative minimum of feature intensity threshold, we recommend 5–10% of the lowest base detections per group. This filter leads to a reduction of the peak chromatogram/electropherogram (BPC/E) intensity. To random noise within the MS data. If the minimum relative 1 3 Intensity Intensity Intensity Intensity Intensity Intensity Intensity Intensity Intensity Intensity Naumann L. et al. occurrence was set to 50%, the feature needs to appear in at the standard at 50% occurrence level (features were detected least three out of six measurements. For higher confidence in three of six measurements), as given in Table 1. For the of the obtained features, higher percentages for the minimum evaluation of the AriumMS peak integration algorithm, two relative occurrence were better. For example, the evaluation example datasets were chosen because of their different peak of the HILIC anion measurements showed 31 target features characteristics. These were CE-MS (alkaline BGE, anion) at 0% (≥ 1/6) occurrence, 30 target features at 50% (≥ 3/6) (Fig. 3a) and HILIC-MS (ESI−) (Fig. 3b). The peak heights occurrence, and 23 target features at 100% (6/6) occurrence. and areas of AriumMS were compared with the results of Within a D-optimal DOE optimization, further param- the manual peak integration using FreeStyle, both normal- eters were tested, and relevant parameters were optimized ized to an internal standard (supporting information). Ari- using three levels per parameter. These were minimum ROI umMS was able to find 88% (CE) and 90% (HILIC) of the size and minimum relative peak height. Minimum ROI peak height and area compared to the manual integration. size is defined as the minimum number of MS1 scans in This finding can be explained by the function of the inte - which the m/z must be present in the EIC. This parameter gration algorithm itself because the ROI intensity threshold is dependent on the processed separation method because is always subtracted from the peak. Furthermore, the peak the obtained feature peak width can differ between dif- integration of the HILIC  method had two outliers com- ferent separation methods Therefore, levels 5, 10, and 15 pared to the manual integration, caused by limit cases of were tested. HILIC required higher ROI sizes (15, broader either non-baseline separated or very broad peaks and thus peaks) and smaller CE (< 10) because of the narrower peak incorrect integration by the software. In general, the low width. In general, the minimum ROI size must be below deviation of the peak integration algorithm of AriumMS the expected peak width of each separation method. The to the manual integration demonstrates the capabilities of minimum relative peak height was significant for feature this software tool for quantitation as generally requested for filtering (AriumMSEval). This filter analyzes each ROI and metabolomics tools [58]. discards features below the relative peak height limit (%). A To test the reliability of the data processing regarding suitable value was 25%. the independence of the loaded file order and simultane- AriumMS offers the capability to define different param- ously the peak finding in general, the six data files of the eter settings for the simultaneous processing of each evalu- repeat measurements were processed three times while only ation group (different methods). Peak shapes and migration/ changing the order in which files were loaded. AriumMS retention time stability differ highly between CE and HILIC; generated similar results for all methods within the three therefore, minimum ROI group size and peak alignment file orders (Fig.  3c) except the HILIC cation. Here, three (time) were probably the key aspects and should therefore additional metabolites of the standard were found, caused be set for each group (method) individually. Especially, peak by limit cases of either non-baseline separated analytes or alignment becomes relevant for CE data due to migration very broad peaks. time shifts that can occur between replicates (cp. avg. migra- AriumMS reduces the required data post-processing tion time deviation for CE [acidic BGE, anion]: ±0.9 min, by the user significantly, compared to traditional metab- and HILIC [ESI−]: ±0.1 min). The use of effective electro- olomics software, which is typically not optimized for phoretic mobility instead of the migration time can address multi-platform analytics. The processing of the whole this issue [57] and will be implemented in AriumMS in the dataset containing five analytical methods with six meas- future. urements takes about 60 min when AriumMS was used on a consumer-grade personal computer (PC) system with Validation of the AriumMS feature generation a 6-core CPU and 32 GB RAM). Considering the com- putation power of the computer used for AriumMS, an For the validation of the feature generation of AriumMS, enterprise-grade computer (32-core CPU, 128 GB RAM) the number of found targets and their respective integration was able to reduce the required processing time to 50 min. were evaluated. The reliability of the data processing was Extensive data post-processing is not required for Ari- tested with different file orders, and the required processing umMS due to its automated data augmentation of different time of the overall workflow is given. Finally, the feature methods (groups) and the integration of related statistical generation and peak integration algorithm of AriumMS was tools, which are mandatory for multi-platform data evalu- compared with the established universal open-source plat- ation. For data evaluation, AriumMS contains advanced form MZmine 3 [41]. For this validation, the MS data of all statistical evaluation tools such as labeling of false-pos- five analytical methods were processed with the optimized itive features (false discovery rate, Benjamini–Hochberg parameter settings (Supplement Table S2). procedure) [59], data scaling options (centering, Pareto, Using the optimized data processing settings, AriumMS auto), transformation (power and log), and various plots was able to find 89% (112 of 126) of the target features of (scatter plots, volcano plots, principal component analysis 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… [PCA], and heatmaps). Because of the combination of the metabolome was analyzed here by the five different ana- feature list generation and the statistical evaluation, no lytical methods and augmented evaluation by AriumMS. In further data transfer into additional statistical software is principle, multi-platform metabolomics offers two major required, which is an advantage compared to other metabo- improvements. Firstly, it is possible to increase the analyti- lomics software. AriumMS is under active development, cal metabolome coverage, and secondly, observed metabolic and the open-source code is continuously optimized to effects are cross-evaluated by another method (if detected in further improve the required data processing times. both). Generally, the data evaluation here was based on three In order to compare the feature generation and peak inte- levels. Starting with a target evaluation, based on the feature gration algorithm of AriumMS with the established univer- list obtained by AriumMS, target features were assigned by sal open-source platform MZmine 3 [41], the data of the their m/z value and RT/MT compared to the reference values five analytical methods were processed with MZmine using of the metabolite standard, followed by a suspect evalua- optimized parameter settings and data processing options tion, where target features were assigned by m/z values of (Supplement Table S3). MZmine was able to find 87% (109 a suspect database. Moreover, in a non-targeted evaluation, of 126 target features) and AriumMS found 89% (112 of 126 the overall feature lists of Mock and Effect1 were compared. analytes) of the target features of the standard, both with The mid-level augmented data evaluation shows that the an occurrence level of 50%. MZmine found 98% (CE) and combination of all ESI negative methods (HILIC [ESI−], CE 110% (HILIC) of the peak height and area compared to the alkaline BGE, CE acidic BGE) detects 24 targets. The ESI manual integration. The comparison reveals that AriumMS positive augmentation (HILIC [ESI+, CE cation]) detects and MZmine offer similar results regarding feature genera- 19 targets (Table 2, Fig. 4a). If just one method is applied tion and peak integration, which highlights the solid foun- for the metabolome analysis, the number of found features dation of the AriumMS feature extraction for augmented is decreased because HILIC (ESI+) detects just 14, and multi-platform data analysis. CE (cation) 16 targets. Figure 4a shows a comparison of target numbers found by each method and by the augmen- tation. The total number of detectable targets in the yeast Augmented analytical workflows and data extracts was lower than in the metabolite standard due to evaluation their absence in the yeast metabolism (Table 2 “not present”) or low biological concentration. The respective number of Combination of different methods detectable target metabolites was reduced to 28 for ESI negative augmentation and 21 for ESI positive augmenta- The combination of multiple analytical methods—here, tion (Fig. 4a). CE-MS and HILIC-MS—increases the feature capacity by For the suspect evaluation, metabolites of the glycoly- their different selectivity. Each separation technique (either sis, gluconeogenesis, TCA (tricarboxylic acid) cycle, and HILIC or CE) was not able to detect all 36 metabolites of the amino acid metabolism were analytes of interest. There- standard (Fig. 4a, Table 1). The multi-platform data evalua- fore, a list containing the m/z values of relevant metabo- tion offers the possibility to increase the analytical coverage lites (Table S4) was used for a m/z search and labeling of a non-target metabolomics workflow. In an environment within the generated feature lists of the measured yeast as complex as metabolomics samples, the number of detect- extracts (labeling within a ± 0.03 m/z range). Again, the able features (target, suspect, non-target) can be expected multi-platform analysis offers a higher analytical metabo- to be higher. Hence, two combinations for the augmented lome coverage than single methods, as shown in Fig. 4b analysis were tested. Based on the AriumMS evaluation of and supplement Table S4. In addition, two mutually con- the standard, the mid-level augmented data evaluation of all firming analytical methods lead to higher confidence in three ESI negative methods enables the coverage of up to 30 suspect feature search only based on m/z values. Here, the target metabolites (w/o co-migrating metabolites). Combin- ESI negative augmentation finds 38 suspects, and the ESI ing the two ESI positive methods allows the coverage of up positive 31 suspects (Fig. 4b). The three CE methods offer to 23 target metabolites. the capability to cross-prove several of the observed sus- pect effects by the HILIC methods (Table S4). Most of Application: yeast metabolome the suspect regulation fold changes matched between the methods; a reason for varying results could be the simi- As a case study, two yeast CEN.PK122 [42] cell cultures lar m/z values of different metabolites or different matrix were analyzed, one as reference (Mock) and the other effects (for example, co-migration/elution of metabolites or treated with halogenated indole (Effect1). Adding halogen- BGE effects). This issue can be addressed with the imple- ated indole to yeast resulted in a clearly decreased growth mentation of the MS/MS confirmation and will be imple- rate, as shown in Fig. S1. The induced effect on the yeast mented soon in AriumMS and a database search. Most of 1 3 Naumann L. et al. 1.2 6.0 A B 1.0 5.0 0.8 4.0 0.6 3.0 0.4 2.0 0.2 1.0 0.0 0.0 0.00.5 1.01.5 2.0 0.01.0 2.03.0 4.05.0 Normalized area/height FreeStyle Normalizedarea/height FreeStyle height area height area CE (ESI-, alkaline BGE) CE (ESI-, acidic BGE) HILIC (ESI-) 31/33 identified 12/33 identified 29/33 identified Rep1 Rep1 Rep1 Rep2 Rep3 Rep2 Rep3 Rep2 Rep3 CE cation HILIC (ESI+) 18/33 identified 21/33 identified Rep1 Rep1 Rep2 Rep3 Rep2 Rep3 Fig. 3 Comparison of peak integration algorithms of AriumMS using repeatability of the peak finding in general for AriumMS (C). Venn the example of CE-MS anion alkaline (A) and HILIC-MS anion (B). diagrams are shown for each method Evaluation of the independence of file order and simultaneously the the targets and suspects were significantly downregulated Fig.  S4). The sum parameter of mannose-1-phosphate between Mock and Effect1, shown in Fig.  4c. An observed and galactose-1-phosphate showed a slight downregula- metabolic effect of the ESI negative augmentation was the tion with fold change (FC) of 0.9 (HILIC ESI−, indole- change in sugar metabolism (Table 2, supplement Table S4, treated effect divided by reference cell effect). The sum 1 3 Normalized area/height AriumMS AriumMS Normalizedarea/height AriumMS Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… 24 24 21 21 18 18 15 12 9 9 HILIC (ESI -) CE (alk. BGE, CE (ac. BGE, HILIC (ESI +) CE (cation)augmentation augmentation anion) anion) ESI negative ESI positive detectable targets AriumMS (metabolite std., w/o internal std., w/o co-migrating analytes) detected targets AriumMS (yeast) 14 14 HILIC (ESI -) CE (alk. BGE, CE (ac. BGE, HILIC (ESI +) CE (cation) augmentation augmentation anion) anion) ESI negative ESI positive detected suspects (yeast) 10 10 8 8 6 6 4 4 2 2 0 0 -8 -6 -4 -2 02468 -8 -6 -4 -2 0246 8 log2 fold change log2 fold change HILIC (ESI-) CE (alk. BGE) HILIC (ESI+) CE cation CE (aci. BGE) FC of 2 FC of 2FC of 0.5 FC of 0.5 critical p-value critical p-value Fig. 4 Results of the augmented data evaluation. A  Shows the num- umMS in yeast extracts is presented with black bars. Both numbers ber of detectable standard metabolites by the methods itself and by are without internal standards and co-migrating analytes. B  Shows mid-level augmented data evaluation using AriumMS presented with the number of found suspects per method and augmentation. Volcano gray bars. The number of detected targets in the yeast extracts per plots for the two augmentations are presented in (C) ESI negative and method and by the mid-level augmented data evaluation using Ari- (D) ESI positive mode parameter of fructose-6-phosphate and glucose-1-phos- (Fig.  S4a–b). On the suspect level, the appearance of phate was even more downregulated (FC 0.3, HILIC two additional disaccharides was observed (Fig. S4g–h). ESI−, Fig. S4 e–f). The CE-MS method (alkaline BGE) Hence, indole treatment may lead to lower levels of glu- showed the same downregulation of glucose-1-phosphate cose, hindering the production of mannose-1-phosphate 1 3 -log10 p No. of targets No. of suspects -log10 p Naumann L. et al. Table 2 Results of the augmented multi-platform data analysis of yeast extracts based on targets Group Analyte ESI negative augmentation ESI positive augmentation − + ID Name [M−H] HILIC/CE (alk.)/CE (aci.) [M+H] HILIC/CE Amino acid 1 L-serine 104.0348 down / down* / - 106.0504 n.a. / down 2 L-proline 114.0555 down / stable* / - 116.0712 stable / down 3 L-valine 116.0712 stable / stable* / - 118.0869 stable / stable 4 L-threonine 118.0504 down / stable* / - 120.0661 down / stable 5 L-cysteine 120.0119 - / n.a.* / - 122.0276 n.a. / n.a. 6 L-leucine 130.0869 stable / stable* / - 132.1025 knock out / stable 7 L-isoleucine 130.0869 down / n.a.* / - 132.1025 n.a. / induced 8 L-aspartic acid 132.0297 down / stable / n.a. 134.0454 down / induced 9 L-lysine 145.0978 - / stable / - 147.1134 n.a. / stable 10 L-glutamic acid 146.0454 stable / stable / - 148.0610 stable / stable 11 L-methionine 148.0433 down / knock out* / - 150.0589 down / down 12 L-histidine 154.0617 - / down / - 156.0773 - / stable 13 L-phenylalanine 164.0712 stable / up* / n.a. 166.0869 stable / stable 14 L-arginine 173.1039 - / stable / - 175.1196 - / up 15 L-tyrosine 180.0661 stable / stable* / - 182.0818 down / induced 16 L-tryptophan 203.0821 knock out / knock out* / - 205.0978 n.a. / knock out Internal std 17 benzenesulfinic acid 141.0010 used for internal standardization 143.0167 used for internal standardiza- 18 2-nitrobenzoic acid 166.0140 168.0297 tion 19 methionine sulfone 180.0331 182.0487 20 pentetic acid 392.1306 394.1462 Metabolites 21 succinate 117.0188 down /—/ n.a. 119.0345 - / - 22 nicotinic acid 122.0242 stable / stable / stable 124.0399 stable / knock out 23 tartaric acid 149.0086 - /—/ n.a. 151.0243 - / - 24 citrate; citric acid 191.0192 - /—/ n.a. 193.0348 - / - 25 caffeine 193.0726 not present 195.0882 not present 26 ATP 505.9879 n.a. / n.a. / - 508.0036 stable / - 27 NAD 662.1013 down / stable / n.a. 664.1170 stable / - 28 NADH 664.1170 n.a. / n.a. / - 666.1327 knock out / - Carbohydrates 29 L-fuc., 6-deoxy-L-gal 163.0607 - / n.a.* / n.a.* 165.0763 - / - 30 D-mannose 179.0556 knock out / knock out* / n.a.* 181.0713 - / - 31 α-D-glucose 179.0556 n.a. / n.a.* / n.a.* 181.0713 - / - 32 α-D-galactose 179.0556 n.a. / n.a.* / n.a.* 181.0713 - / - 33 D-mannose-1-PO 259.0219 stable / stable/ stable 261.0376 knock out / - 34 β-D-fructose-6-PO 259.0219 down / n.a. / stable 261.0376 n.a. / - 35 α-D-glucose 1-PO 259.0219 down / n.a. / stable 261.0376 n.a. / - 36 α-D-galactose-1-PO 259.0219 stable / stable / stable 261.0376 n.a. / - 37 N-acetylneuraminate 307.0909 not present 309.1066 38 UDP-glucose 565.0472 stable / n.a. / knock out 567.0629 n.a. / - 39 GDP-L-fucose 588.0744 not present 590.0901 not present 40 CMP-N-acetylneuraminate 613.1395 not present 615.1552 not present Number of detected targets (w/o co-migrating analytes) anion 24 cation 19 Regulation calculated by the FC (indole-treated/reference): FC < 0.50, down; 0.50 ≤ FC ≤ 2.0, stable; FC ≥ 2.00, up. Non-detected analytes are labeled as “n.a.”, undetectable analytes are labeled as “-”. Knocked-out analytes by halogenated indole treatment are labeled as “knock out,” and newly occurring analytes by treatment are labeled as “induced.” Co-migrating analytes are labeled with *. The order of the presented results is as follows: (i) augmentation ESI negative: HILIC (ESI−)/CE (alkaline, anion)/CE (acidic, anion), and (ii) augmentation ESI positive: HILIC (ESI+)/CE (cation) 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… and glucose-6-phosphate, causing increased production of of various isomeric anions such as hexose phosphates by lactose and saccharose. Lactose cannot be digested by yeast CE-MS. cells because of the lacking lactase enzyme. Furthermore, AriumMS was successfully applied for mid-level data within the two augmentations, L-tryptophan was knocked fusion to remove irrelevant information such as artifacts and out in the indole-treated samples (Effect1). noise from the entire analytical dataset (LC-MS + CE-MS) For the non-target evaluation, three PCAs were con- of yeast extracts. The results confirm the great advantage of ducted, which contain the features of Mock and Effect1 flexible parameterization for processing of individual sepa- (feature occurrence ≥ 50%). Each of the three PCAs was ration methods with different peak characteristics. Multi- a combination of an HILIC (ESI±) method with one CE platform metabolomics expands metabolome coverage and method. Figure S5 shows that the measured samples (Mock increases the confidence of the metabolic results. and Effect1) were partitioned into two major groups derived Supplementary Information The online version contains supplemen- by the treatment with halogenated indole with at least 79.5% tary material available at https://doi. or g/10. 1007/ s00216- 023- 04715-6 . of the explained variance on the first principal component (PC1). As a result, the treatment with halogenated indole has Acknowledgements Gratefully, we acknowledge Prof. Dr. Norbert Schnell and Nadine Pejs from University Aalen for generating the bio- a strong influence on the yeast metabolome. The integration logical samples and Olivia Haun for the execution of the relative DOE of several LC-MS/CE-MS techniques expands metabolome and the software optimizations for AriumMS. Thank you to Patrick coverage and increases the confidence of the metabolic Schlossbauer for valuable discussions and useful suggestions. results. This makes AriumMS a powerful tool for multi- Funding Open Access funding enabled and organized by Projekt platform metabolomics. DEAL. This study was supported by the German Federal Ministry of Education and Research within the PROmiGlykAN project (FKZ 13FH635IB6) and the associated partners of this project (Bruker Dal- tonik GmbH, Rentschler Biotechnologie GmbH, MLS GmbH). Conclusions Data availability The datasets generated during and/or analyzed during Multi-platform metabolomics by high-resolution MS based the current study are available from the corresponding author on rea- on several orthogonal separation mechanisms of CE and LC sonable request. AriumMS 1.0.0 is available at GitHub: https://git hub. com/ Adria nHaun/ Arium MS/ maximizes the metabolome coverage. The AriumMS soft- ware toolbox presented here is a powerful tool for fast untar- Declarations geted processing of these augmented datasets. AriumMS contains ROI search, preprocessing, feature detection and Competing interests The authors have no relevant financial or non- integration, false-positive filter, scaling, and transformation financial interests to disclose. followed by the augmentation and various chemometric data evaluation tools. The validation of the feature detection and Open Access This article is licensed under a Creative Commons Attri- mid-level fusion steps were successfully performed using bution 4.0 International License, which permits use, sharing, adapta- tion, distribution and reproduction in any medium or format, as long a multi-analyte standard. In AriumMS all processing steps as you give appropriate credit to the original author(s) and the source, were integrated into a single user-friendly software tool at a provide a link to the Creative Commons licence, and indicate if changes high level of e fl xibility and automatization. Further develop - were made. The images or other third party material in this article are ments will include the implementation of MS/MS spectral included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in networking in order to precisely and autonomously connect the article's Creative Commons licence and your intended use is not features detected by two or more methods. Furthermore, the permitted by statutory regulation or exceeds the permitted use, you will augmentation of spectroscopic data with chromatographic/ need to obtain permission directly from the copyright holder. To view a electrophoretic data might be interesting. copy of this licence, visit http://cr eativ ecommons. or g/licen ses/ b y/4.0/ . The AriumMS tool presented here was used to process datasets obtained from HILIC-MS and CE-MS measure- ments of a multi-analyte standard and yeast extracts. All References CE-MS methods show in general narrow peaks and benefit from the sensitivity of the nanoCEasy interface. Addition- 1. Ivanisevic J, Want EJ. From samples to insights into metabolism: ally, the valving functionality of the nanoCEasy interface uncovering biologically relevant information in LC-HRMS metab- olomics data. Metabolites. 2019. https:// doi. org/ 10. 3390/ metab enables capillary reconditioning between the injections. o9120 308. HILIC-MS and CE-MS show a large overlap with only a few 2. Comte B, Monnerie S, Brandolini-Bunlon M, Canlet C, Castelli analytes detected by only one of the methods. HILIC-MS F, Chu-Van E, Colsch B, Fenaille F, Joly C, Jourdan F, Lenuzza also covers neutral polar analytes, such as carbohydrates, N, Lyan B, Martin J-F, Migné C, Morais JA, Pétéra M, Poupin N, Vinson F, Thevenot E, Junot C, Gaudreau P, Pujos-Guillot whereas the basic BGE in particular allows the separation E. Multiplatform metabolomics for an integrative exploration of 1 3 Naumann L. et al. metabolic syndrome in older men. EBioMedicine. 2021. https:// mouse model of polycystic kidney disease. Sci Rep. 2019. https:// doi. org/ 10. 1016/j. ebiom. 2021. 103440.doi. org/ 10. 1038/ s41598- 018- 37512-8. 3. Pezzatti J, Boccard J, Codesido S, Gagnebin Y, Joshi A, Picard D, 19. Naumann L, Schairer J, Höchsmann A, Naghdi E, Neusüß C. Cap- González-Ruiz V, Rudaz S. Implementation of liquid chromatog- illary electrophoresis–mass spectrometry interfacing: principles raphy-high resolution mass spectrometry methods for untargeted and recent developments. In: Ramautar R, Chen DDY, editors. metabolomic analyses of biological samples: A tutorial. Anal Capillary electrophoresis-mass spectrometry for proteomics and Chim Acta. 2020. https:// doi. org/ 10. 1016/j. aca. 2019. 12. 062. metabolomics. Principles and applications, vol. 2021. Weinheim: 4. Jedrychowski MP, Huttlin EL, Haas W, Sowa ME, Rad R, Gygi Wiley-VCH; 2022. p. 1–33. SP. Evaluation of HCD- and CID-type fragmentation within their 20. Schlecht J, Stolz A, Hofmann A, Gerstung L, Neusüß C. nano- respective detection platforms for murine phosphoproteomics. CEasy: an easy, flexible, and robust nanoflow sheath liquid cap- MCP. 2011. https:// doi. org/ 10. 1074/ mcp. M111. 009910. illary electrophoresis-mass spectrometry interface based on 3D 5. Sentandreu E, Peris-Díaz MD, Sweeney SR, Chiou J, Muñoz printed parts. Anal Chem. 2021. https://doi. or g/10. 1021/ acs. analc N, Tiziani S. A survey of orbitrap all ion fragmentation analysis hem. 1c032 13. assessed by an R metabolist package to study small-molecule metab- 21. Naumann L, Schlossbauer P, Klingler F, Hesse F, Otte K, Neusüß olites. Chromatographia. 2018. https:// doi. org/ 10. 1101/ 257147. C. High-throughput glycosylation analysis of intact monoclonal 6. Kohler I, Verhoeven M, Haselberg R, Gargano AFG. Hydrophilic antibodies by mass spectrometry coupled with capillary electro- interaction chromatography – mass spectrometry for metabo- phoresis and liquid chromatography. J Sep Sci. 2022. https:// doi. lomics and proteomics: state-of-the-art and current trends. Micro-org/ 10. 1002/ jssc. 20210 0865. chem J. 2022. https:// doi. org/ 10. 1016/j. microc. 2021. 106986. 22. Höcker O, Knierman M, Meixner J, Neusüß C. Two capillary 7. Ibáñez C, Simó C, García-Cañas V, Gómez-Martínez A, Ferragut approach for a multifunctional nanoflow sheath liquid interface JA, Cifuentes A. CE/LC-MS multiplatform for broad metabolomic for capillary electrophoresis-mass spectrometry. Electrophoresis. analysis of dietary polyphenols effect on colon cancer cells prolifera- 2021. https:// doi. org/ 10. 1002/ elps. 20200 0169. tion. Electrophoresis. 2012. https://d oi.o rg/1 0.1 002/e lps.2 01200 143. 23. Johnson CH, Ivanisevic J, Siuzdak G. Metabolomics: beyond bio- 8. Zhang Y-Y, Zhang Q, Zhang Y-M, Wang W-W, Zhang L, Yu Y-J, markers and towards mechanisms. Nat Rev Mol Cell Biol. 2016. Bai C-C, Guo J-Z, Fu H-Y, She Y. A comprehensive automatic https:// doi. org/ 10. 1038/ nrm. 2016. 25. data analysis strategy for gas chromatography-mass spectrometry 24. Suhre K, Meisinger C, Döring A, Altmaier E, Belcredi P, Gieger based untargeted metabolomics. J Chromatogr A. 2020. https:// C, Chang D, Milburn MV, Gall WE, Weinberger KM, Mewes doi. org/ 10. 1016/j. chroma. 2019. 460787. H-W, Hrabé de Angelis M, Wichmann H-E, Kronenberg F, Adam- 9. Höcker O, Flottmann D, Schmidt TC, Neusüß C. Non-targeted LC-MS ski J, Illig T. Metabolic footprint of diabetes: a multiplatform and CE-MS for biomarker discovery in bioreactors: influence of metabolomics study in an epidemiological setting. PloS one. 2010. separation, mass spectrometry and data processing tools. Sci Total https:// doi. org/ 10. 1371/ journ al. pone. 00139 53. Environ. 2021. https:// doi. org/ 10. 1016/j. scito tenv. 2021. 149012. 25. Xuan Q, Ouyang Y, Wang Y, Liang Wu, Li H, Luo Y, Zhao X, 10. Hirayama A, Wakayama M, Soga T. Metabolome analysis based Feng D, Qin W, Chunxiu Hu, Zhou L, Liu X, Zou H, Cai C, Jiarui on capillary electrophoresis-mass spectrometry. Trends Analyt Wu, Jia W, Guowang Xu. Multiplatform metabolomics reveals Chem. 2014. https:// doi. org/ 10. 1016/j. trac. 2014. 05. 005. novel serum metabolite biomarkers in diabetic retinopathy sub- 11. Kok MGM, Somsen GW, de Jong GJ. The role of capillary electrophore- jects. Adv Sci Lett. 2020. https:// doi. or g/ 10. 1002/ adv s. 20200 sis in metabolic profiling studies employing multiple analytical techniques. 1714. Trends Analyt Chem. 2014. https:// doi. org/ 10. 1016/j. trac. 2014. 06. 004. 26. Büscher JM, Czernik D, Ewald JC, Sauer U, Zamboni N. Cross-plat- 12. Soga T, Ohashi Y, Ueno Y, Naraoka H, Tomita M, Nishioka T. Quan- form comparison of methods for quantitative metabolomics of primary titative metabolome analysis using capillary electrophoresis mass spec- metabolism. Anal Chem. 2009. https:// doi. org/ 10. 1021/ ac802 2857. trometry. J Proteome Res. 2003. https:// doi. org/ 10. 1021/ pr034 020m. 27. Li Y, Hou G, Zhou H, Wang Y, Tun HM, Zhu A, Zhao J, Xiao 13. Soga T, Kakazu Y, Robert M, Tomita M, Nishioka T. Qualitative F, Lin S, Liu D, Zhou D, Mai L, Zhang L, Zhang Z, Kuang L, and quantitative analysis of amino acids by capillary electropho- Guan J, Chen Q, Wen L, Zhang Y, Zhuo J, Li F, Zhuang Z, Chen resis-electrospray ionization-tandem mass spectrometry. Electro- Z, Luo L, Liu D, Chen C, Gan M, Zhong N, Zhao J, Ren Y, Xu phoresis. 2004. https:// doi. org/ 10. 1002/ elps. 20030 5791. Y. Multi-platform omics analysis reveals molecular signature 14. Monton MRN, Soga T. Metabolome analysis by capillary electro- for COVID-19 pathogenesis, prognosis and drug target discov- phoresis–mass spectrometry. J Chromatogr A. 2007. https:// doi. ery. Signal Transduct Target Ther. 2021. https:// doi. org/ 10. 1038/ org/ 10. 1016/j. chroma. 2007. 02. 065.s41392- 021- 00508-4. 15. Sawada H, Nogami C. Capillary electrophoresis–electrospray ioniza- 28. Ortiz-Villanueva E, Benavente F, Piña B, Sanz-Nebot V, Tauler R, tion mass spectrometry using uncoated fused-silica capillaries and Jaumot J. Knowledge integration strategies for untargeted metabo- alkaline buffer solution for the analysis of small carboxylic acids. lomics based on MCR-ALS analysis of CE-MS and LC-MS data. Anal Chim Acta. 2004. https:// doi. org/ 10. 1016/j. aca. 2003. 11. 047. Anal Chim Acta. 2017. https://doi. or g/10. 1016/j. aca. 2017. 04. 049 . 16. Yamamoto M, Ly R, Gill B, Zhu Y, Moran-Mirabal J, Britz-McK- 29. Rivera-Pérez A, Romero-González R, Garrido FA. Application of ibbin P. Robust and high-throughput method for anionic metabo- an innovative metabolomics approach to discriminate geographi- lite profiling: preventing polyimide aminolysis and capillary cal origin and processing of black pepper by untargeted UHPLC- breakages under alkaline conditions in capillary electrophoresis- Q-Orbitrap-HRMS analysis and mid-level data fusion. Food Res mass spectrometry. Anal Chem. 2016. https://doi. or g/10. 1021/ acs. Int. 2021. https:// doi. org/ 10. 1016/j. foodr es. 2021. 110722. analc hem. 6b032 69. 30. Letertre MPM, Dervilly G, Giraudeau P. Combined nuclear mag- 17. de Macedo AN, Jiwa MIY, Macri J, Belostotsky V, Hill S, Britz- netic resonance spectroscopy and mass spectrometry approaches McKibbin P. Strong anion determination in biological fluids by for metabolomics. Anal Chem. 2021. https://doi. or g/10. 1021/ acs. capillary electrophoresis for clinical diagnostics. Anal Chem. analc hem. 0c043 71. 2013. https:// doi. org/ 10. 1021/ ac402 975q. 31. Smilde AK, van der Werf MJ, Bijlsma S, van der Werff-van der Vat 18. Sánchez-López E, Kammeijer GSM, Crego AL, Marina ML, BJC, Jellema RH. Fusion of mass spectrometry-based metabolomics Ramautar R, Peters DJM, Mayboroda OA. Sheathless CE-MS data. Anal Chem. 2005. https:// doi. org/ 10. 1021/ ac051 080y. based metabolic profiling of kidney tissue section samples from a 1 3 Augmented region of interest for untargeted metabolomics mass spectrometry (AriumMS) of… 32. Dai S, Lin Z, Xu B, Wang Y, Shi X, Qiao Y, Zhang J. Metabolomics 48. Du P, Kibbe WA, Lin SM. Improved peak detection in mass spectrum data fusion between near infrared spectroscopy and high-resolution mass by incorporating continuous wavelet transform-based pattern match- spectrometry: a synergetic approach to boost performance or induce con- ing. Bioinf. 2006. https:// doi. org/ 10. 1093/ bioin forma tics/ btl355. fusion. Talanta. 2018. https:// doi. org/ 10. 1016/j. talan ta. 2018. 07. 030. 49. Wahab MF, O’Haver TC. Wavelet transforms in separation science 33. Roussel S, Bellon-Maurel V, Roger J-M, Grenier P. Fusion of aroma, for denoising and peak overlap detection. J Sep Sci. 2020. https:// FT-IR and UV sensor data based on the Bayesian inference application doi. org/ 10. 1002/ jssc. 20200 0013. to the discrimination of white grape varieties. Chemometr Intell Lab Syst. 50. Rupprecht F, Enge S, Schmidt K, Gao W, Miller R. Automating 2003. https:// doi. org/ 10. 1016/ S0169- 7439(02) 00111-9. LC-MS/MS mass chromatogram quantification: wavelet transform 34. Khaleghi B, Khamis A, Karray FO, Razavi SN. Multisensor data based peak detection and automated estimation of peak boundaries and fusion: A review of the state-of-the-art. Inf Fusion. 2013. https:// signal-to-noise ratio using signal processing methods. Biomed Signal doi. org/ 10. 1016/j. inffus. 2011. 08. 001. Process Control. 2022. https:// doi. org/ 10. 1016/j. bspc. 2021. 103211. 35. Azcarate SM, Ríos-Reina R, Amigo JM, Goicoechea HC. Data 51. Ju R, Liu X, Zheng F, Zhao X, Lu X, Zeng Z, Lin X, Xu G. handling in data fusion: methodologies and applications. Trends Removal of false positive features to generate authentic peak table Analyt Chem. 2021. https:// doi. org/ 10. 1016/j. trac. 2021. 116355. for high-resolution mass spectrometry-based metabolomics study. 36. Casian T, Nagy B, Kovács B, Galata DL, Hirsch E, Farkas A. Anal Chim Acta. 2019. https://doi. or g/10. 1016/j. aca. 2019. 04. 011 . Challenges and opportunities of implementing data fusion in pro- 52. Shannon CE. A mathematical theory of communication. Bell Syst cess analytical technology-a review. Molecules. 2022. https://doi. Tech J. 1948. https://d oi.o rg/1 0.1 002/j.1 538-7 305.1 948.t b0091 7.x. org/ 10. 3390/ molec ules2 71548 46. 53. van den Berg RA, Hoefsloot HCJ, Westerhuis JA, Smilde AK, van 37. Tautenhahn R, Böttcher C, Neumann S. Highly sensitive feature der Werf MJ. Centering, scaling, and transformations: improving detection for high resolution LC/MS. BMC Bioinf. 2008. https:// the biological information content of metabolomics data. BMC doi. org/ 10. 1186/ 1471- 2105-9- 504. Genom. 2006. https:// doi. org/ 10. 1186/ 1471- 2164-7- 142. 38. Tautenhahn R, Patti GJ, Rinehart D, Siuzdak G. XCMS Online: a 54. Smucker B, Krzywinski M, Altman N. Optimal experimental design. web-based platform to process untargeted metabolomic data. Anal Nat Methods. 2018. https:// doi. org/ 10. 1038/ s41592- 018- 0083-2. Chem. 2012. https:// doi. org/ 10. 1021/ ac300 698c. 55. Jacyna J, Kordalewska M, Markuszewski MJ. Design of experi- 39. Xia J, Psychogios N, Young N, Wishart DS. MetaboAnalyst: ments in metabolomics-related studies: an overview. J Pharm a web server for metabolomic data analysis and interpretation. Biomed Anal. 2019. https:// doi. org/ 10. 1016/j. jpba. 2018. 11. 027. Nucleic Acids Res. 2009. https:// doi. org/ 10. 1093/ nar/ gkp356. 56. Gorrochategui E, Jaumot J, Tauler R. ROIMCR: a powerful analy- 40. Chong J, Xia J. MetaboAnalystR: an R package for flexible and sis strategy for LC-MS metabolomic datasets. BMC Bioinform. reproducible analysis of metabolomics data. Bioinf. 2018. https:// 2019. https:// doi. org/ 10. 1186/ s12859- 019- 2848-8. doi. org/ 10. 1093/ bioin forma tics/ bty528. 57. González-Ruiz V, Gagnebin Y, Drouin N, Codesido S, Rudaz S, 41. Pluskal T, Castillo S, Villar-Briones A, Oresic M. MZmine 2: Schappler J. ROMANCE: A new software tool to improve data modular framework for processing, visualizing, and analyzing robustness and feature identification in CE-MS metabolomics. mass spectrometry-based molecular profile data. BMC Bioinf. Electrophoresis. 2018. https:// doi. org/ 10. 1002/ elps. 20170 0427. 2010. https:// doi. org/ 10. 1186/ 1471- 2105- 11- 395. 58. Pinu FR, Goldansaz SA, Jaine J. Translational metabolomics: 42. Entian K-D, Kötter P. 25 Yeast genetic strain and plasmid collec- current challenges and future opportunities. Metabolites. 2019. tions. In: Stansfield I, editor. Yeast gene analysis, vol. 36. 2nd ed. https:// doi. org/ 10. 3390/ metab o9060 108. Amsterdam: Elsevier; 2007. p. 629–66. 59. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a prac- 43. Chambers MC, Maclean B, Burke R, Amodei D, Ruderman DL, tical and powerful approach to multiple testing. J R Stat Soc Ser B Neumann S, Gatto L, Fischer B, Pratt B, Egertson J, Hoff K, Kes- Methodol. 1995. https://doi. or g/10. 1111/j. 2517- 6161. 1995. tb020 31.x . sner D, Tasman N, Shulman N, Frewen B, Baker TA, Brusniak M-Y, Paulse C, Creasy D, Flashner L, Kani K, Moulding C, Sey- Publisher's note Springer Nature remains neutral with regard to mour SL, Nuwaysir LM, Lefebvre B, Kuhlmann F, Roark J, Rainer jurisdictional claims in published maps and institutional affiliations. P, Detlev S, Hemenway T, Huhmer A, Langridge J, Connolly B, Chadick T, Holly K, Eckels J, Deutsch EW, Moritz RL, Katz JE, Agus DB, MacCoss M, Tabb DL, Mallick P. A cross-platform toolkit for mass spectrometry and proteomics. Nat Biotechnol. Lukas Naumann is currently a 2012. https:// doi. org/ 10. 1038/ nbt. 2377. PhD student in the research 44. Pedrioli PGA, Eng JK, Hubley R, Vogelzang M, Deutsch EW, group of Prof. Christian Neusüß Raught B, Pratt B, Nilsson E, Angeletti RH, Apweiler R, Cheung at Aalen University. He has K, Costello CE, Hermjakob H, Huang S, Julian RK, Kapp E, experience in various analytical McComb ME, Oliver SG, Omenn G, Paton NW, Simpson R, Smith techniques such as LC-MS and R, Taylor CF, Zhu W, Aebersold R. A common open representa- CE-MS. He has been working on tion of mass spectrometry data and its application to proteomics the development of analytical research. Nat Biotechnol. 2004. https:// doi. org/ 10. 1038/ nbt10 31. workflows for the characteriza- 45. Kaever A, Landesfeind M, Possienke M, Feussner K, Feussner I, tion of intact mAb glycosylation Meinicke P. MarVis-Filter: ranking, filtering, adduct and isotope by LC-MS and CE-MS and for correction of mass spectrometry data. J Biomed Biotechnol. 2012. the determination of metabolic https:// doi. org/ 10. 1155/ 2012/ 263910. changes in miRNA-overexpress- 46. Keller BO, Sui J, Young AB, Whittal RM. Interferences and con- ing cell pools. His PhD project is taminants encountered in modern mass spectrometry. Anal Chim focused on the high-throughput Acta. 2008. https:// doi. org/ 10. 1016/j. aca. 2008. 04. 043. analysis of antibody glycosyla- 47. Andrade L, Manolakos ES. Signal background estimation and tion and the related metabolome resulting from microRNA transfection baseline correction algorithms for accurate DNA sequencing. J in CHO cells. VLSI Signal Process Syst Signal Image Video Technol. 2003. https:// doi. org/ 10. 1023/B: VLSI. 00000 03022. 86639. 1f. 1 3 Naumann L. et al. Adrian Haun has been a research Dirk Flottmann became Professor scientist at the University of for analytical chemistry at Aalen Aalen in Germany since 2020. University, Germany, in April He is developing AriumMS with 2002. He employs elemental the aim of creating a powerful mass spectrometry and coupling yet user-friendly, all-in-one soft- techniques to study mainly the ware program for metabolomics distribution of trace elements in applications. His research semiconductor samples. He is focuses on the application of interested in the application of concepts from various disci- multivariate data analysis tech- plines including mathematics, niques and develops new physics and computer science in approaches to characterize the the field of analytical chemistry. metabolome in different kinds of samples to improve the non-tar- get approach. Christian Neusüß is Professor in Alisa Höchsmann is currently a the Faculty of Chemistry at PhD student in the research Aalen University. He received group of Prof. Christian Neusüß his diploma from University of at Aalen University. The subject Heidelberg and his PhD from of her master’s thesis was the University of Leipzig. His separation of cationic metabo- research interests include cou- lites using capillary electropho- pling and application of (electro- resis-mass spectrometry. Her migrative) separation techniques PhD project is related to the with mass spectrometry. He is separation of intact proteins by focusing on technical develop- capillary electrophoresis-mass ments such as nanoESI inter- spectrometry using successive faces and miniaturized two- multiple ionic-polymer layer dimensional separations in (SMIL)-coated capillaries. combination with high-resolu- tion mass spectrometry as well Michael Mohr (BSc) is currently as method development and applications for the analysis of proteins, a master’s student at Aalen Uni- metabolites and contaminants. He is co-author of more than 110 peer- versity and did his bachelor reviewed publications (h-Index = 41). thesis in the research group of Prof. Christian Neusüß during the summer term 2022. During his bachelor’s thesis, he gained expertise in the generation and non-target evaluation of aug- mented HILIC- and CE-MS metabolomic datasets of yeast samples, employing chemomet- ric and statistical tools. Also, he supported the development of AriumMS by extensive use of the software. Martin Novák (MSc) is currently study director for GC-MS, GC-MS/MS and LC-MS/MS analytics at Eurofins Agrosci- ence Services GmbH. He com- pleted his master’s thesis at Aalen University in the research group of Prof. Christian Neusüß. The subject of his master’s thesis was the analysis of anionic metabolites with CE-MS in acidic and basic background electrolytes. 1 3

Journal

Analytical and Bioanalytical ChemistrySpringer Journals

Published: Jul 1, 2023

Keywords: Augmented data evaluation; Mid-level data fusion; Multi-platform metabolomics; nanoCEasy; Capillary electrophoresis; Hydrophilic interaction liquid chromatography

There are no references for this article.