WIXLIB

LC-MS method validation.QC and QAGeorgios TheodoridisAristotle University Thessaloniki, Greece1AcknowledgmentCollaboratorsProf. Ian Wilson, ICLAUTh GroupAssist. Prof. M. Witting, HelmholtzAssist. Prof. Helen Gika Assist. Prof. A. Agapiou, Cyprus Univ.C. Virgiliou, PhDO. Deda, PhDBiomic-AUThA. Pechlivanis, PhD Prof. V. MougiosA. Orfanidis, PhDProf. G. Arsenos, Vet.C. Zisi, PhDAssoc. Prof. N. Raikos, MDPhD studentsO. BegouD. DiamantidouA. LioupiM. MarinakiA. PesiridouA. KrokosAll MSc studentsAssoc. Prof. K. Sarafidis, MDAssoc. Prof. G. Sianos MDAssist. Prof. G. Giannakoulas, MDProf. M. Roilidis, MDFoodOmics GRProf. N. Thomaidis, Athens UnivProf. M. Mikros, Athens UnivProf. A. Spyros, Crete Univ.G. Tsangaris, Athens Academy andthe FoodOmicsGR RI Consortium21

BIOMIC AUTh Interdisciplinary Effort AUTh17 Staff from 8 schools 25 young researchersOmics with emphasis on metabolomicsNew laboratory facility 250 m2 http://biomic.web.auth.gr/LC-MS/MSGC-MS/MSAnd NMRDatabasesSpectra librariesStatistical softwarePathway analysis 100 publications in metabolomics3FoodOmicsGR-RI Open Research Infrastructure 2019-2021 7 Greek Universities and Institutes AUTh, Athens Univ., Crete Univ, Ioannina Univ.,Aegean Univ., Agric. Univ. Athens, Intern. Univ., Academy AthensScopeMap regional foods: comprehensive characterization of foodsSupport nutritional studies, studies on food authenticity and controlof geographical origin.Provide hard molecular data on nutritional value 450 man months for young researchers 70 Staff from 25 Disciplines (plant growth, animal husbandry, milkScience, pharmacognosy, toxicology, analytical chemistry, food chem.Food technology, informatics, biochemistry proteomics, NGS,Elemental metabolomics42

The elements What is metabonomics?“the quantitative measurement of the dynamicmultiparametric response of living systems topathophysiological stimuli or genetic modification”J. Nicohlson et al. 1999Metabolomics is the "systematic study of the uniquechemical fingerprints that specific cellular processes leavebehind" - specifically, the study of their small-moleculemetabolite profiles”B. Daviss, 20055Metabonomics/Metabolomics The determination of thewhole metaboliccomplement of a cell,tissue, organism, sample Correlation withphysiology, stress, origin,other factors It is a holistic approach Why? “The whole is greaterthan the sum of itsparts”63

Source: Considerations in the design of clinical and epidemiological metabolic phenotyping studiesG Theodoridis et al 2013, ebook Metabolic profiling in clinical applications. doi:10.4155/EBO.13.4877Targeted MetabolomicsKey challenges Ion suppression issues No blanks available No agreed strategy for validation Metabolites are there! Unlike in pesticideanalysis: Urine 140 determined, cell culture 100 metabolites . How to quantify ? Surrogate matrix, standardaddition method, external calibration ?Sample Preparation Strategies for the Effective Quantitation of HydrophilicMetabolites in Serum by Multi-Targeted HILIC-MS/MS.Tsakelidou et al Metabolites, 201784

Signal spans to several orders of magnitude9More issues Labelled internal standards not alwaysavailable If available, Not easy to handle Not easy to organise the calibrants Wide range of linear ranges dependingon the matrix and condition Whatever the cost: worth the effort105

WorkflowHypothesisExperimental designRaw data setCollect samplesLC-MS analysis:Short: 10-12 min/ sampleFull scan 100-900 amuExport as peak list files:SAMPLE – Mass/RtDevelop specificassay for themarkersNMR, GC-MSOrbi TRAP-MSIdentify/validatebiomarkersInvestigate ions thatcontribute to the groupseparationMultivariate statisticsPCA, PLS-DA, O-PLS11Pipeline for untargeted LCLC-MS basedMetabolomicsBiological questionStudy protocolSample collectioncollection- storagestorage- treatmentLCLC-MS AnalysisData preprocessingDeconvolution, normalisationMultivariate statisticsData interpretation126

Obvious sources of error Changes in chromatographic properties (retention,peak shape, resolution, selectivity). Changes in mass accuracy Changes in sensitivity (e.g. source contamination)Danger if instrument variation is misinterpreted as beingdue to biological process Extraction, inappropriate. Extraction edits thesample and thus the metabolic profile13Control Methods Internal standards? Test Mixtures (external standards ca. 40 metabolites ofdifferent classes). Quality controls (QCs)? Replicates?Intervention How long can a batch be? 24h24h-48h48h-72h?When to stop and clean the source?When to change pre or analytical column?How can I put data together ? Extract all samplestogether?147

Quality controls/internal standards For LC-MS in metabon/lomics internal standards (IS) are not as useful as inconventional quantitative methods. Suggestion: 5 or more Deuterated ISrepresentative of important classes Test Mixture. Data (XIC) used to control the stability of the LC and MS (rt andintensity). This is used to chop bad data sets not to prove that a dataset isuseful. Data NOT to be used in PCA. QCs – use of a standard/representative BIOLOGICAL sample:-an aliquot of each test sample to make a“ master mix” or pooled sample.For very large sample sets QCs can be made by aliquoting a number of testsamples only (not all).QC should be used for its own batch only for PCA and XIC. Peak Tables from QCsstored and used for comparison.-QC samples run before, during and after analysis to asses system suitability Replicate injections, All ? Selected samples? For large datasets 1 inj-2inj- or3inj? Suggestion: First screen one injection of all samples and repeats (1x) ofup to 20%. Repetitions of the whole batch in second stage to verify the finding(including the data processing).15QC approachtest samples QC sampletestmixtestmixQC QC QC QC QC 10 test samples QC 10 test samples QC QCData scrutiny: Conventional XIC and MVA.Gika, JPR 2007, 2012, Guy 2008, Kamleh 2012 168

Sample AnalysisUPLC-MSTandem MSQC-formatDDATestmixMSE1st QCLast QCQCPooled GroupsSample ExtremesTandem MSConditioningSamplesDilutiontestQCDilution( 2x) 7 QC injectionsMultipleextraction-blanksData scrutiny: Conventional XIC and MVA.Gika, JPR 2007, 2012, Guy 2008, Kamleh 2012 [Vorkas PA et al, Anal Chem, 2015]17Sample AnalysisUPLC-MSTandem MSTestQC-formatDDAMS Emix1st QCLast QCQCPooled GroupsSample ExtremesTandem MSConditioningSamplesQCDilution( 2x) 7 QC injections Know your methodUse methods from literature Run authentic standards Unbiased tandem MS acquisitionRun Data Dependent Acquisition (DDA) extraction-blanksRun a test mix Check RTAccurate massFrequent adducts Run sample preparation blanksMSE(Use pooled/QC samples – pooled groups/extremes) Run Pos and Neg modes with the same gradient189

Retention time19(PCA-X)PCA scores plotQC cluster806034012t[2]200-20-40-60-80-50050t[1]2010

PCA plot :Analytical variability – poor c9qc8qc6-20qc5qc4-40qc3qc2 qc1-60-80-100qc-1000100SIMCA-P 11 - 15/06/200718:37:03Note: PCA plot can be misleadingIn case where test samples have small differences principal components represent small proportion ofthe variability in the data and QCs appear more scattered.21Time series plot2211

Plot profile of ions103.7/5.7841.4/9.08.E 087.E al6.E 085.E 08QC12QC13QC12QC13110120QC44.E 08QC1QC23.E 08QC1QC2QC32.E 08QC5QC4QC6QC141.E 080.E 00010203040708090100130run order23QC Peak Table: Check ions with high CV% valuesQC8 QCxCV%m1 t1104512801147 13218m2 t2876757120 85035237823952547 23676 QC7 QC6mx tx1000900CV Time (min)2412

Metabonomic Study Diseased vs Control populationRaw data, TIC across all samples QC samples25PCACheck for run order effects2613

Metabonomic Study Diseased vs Control populationQC Run orderOutlier? Problem inthe system?System stabilization ?27Raw data, Log (intensity),4 randomly selected peaks2814

Log (Intensity)4 randomly selected peaksBoxplot shows test sample intensities QC samples29Other proposed filters 80% ruleNot all researchers agree 50% rule System suitability testing(i) m/z error of 5 ppm compared to theoretical mass, (ii)retention time error of 2% compared to the definedretention time, (iii) peak area equal to a predefinedacceptable peak area 10% and (iv) symmetrical peakshape with no evidence of peak splitting System suitability blank and process blank samples3015

Intensity vs reproducibility31Day to day variation-Over 20h run (60 human urine samples) repeated for 5 days.How reproducible the data are?Obvious day-to day effectappears vertically in t2although only 6% of totalvariance in datat1 shows sample to samplevariation horizontally3216

Day to day variationPCA “control model”TuesdayMonday-both day-to-day and within day variation33DayDay-toto-day variation-Start up variation each day-4th and 5th batch higher reproducibility3417

QC data offers ways to correctbatch effectBroadhurst et al Metabolomics 201835PrePre-processing of data may have a profound effect onthe resultsLC-MS raw data, software parameters– Unfolding and alignment issues– Mass matching and time windowing-Normalisation (to total signal, to constant peak to baselinenoise etc): danger, may reverse the direction of trends!Eg. If major peak absent then other peaks increase in apparentimportance-Scaling-Improve reproducibility in the analytical methodWhen we cannot repeat experiment (ethical issues, small samplevolume cost) we may ‘rescue’ the data by- OPLS-DA to “See through” unwanted variation- spectral filtersBUT overall it can be GIGO3618

/ The consortium’s mission is to engage the metabolomicscommunity to communicate and promote the development,dissemination and harmonization of best QA/QC practices inuntargeted metabolomics through the following objectives:To identify, catalog, harmonize and disseminate QA/QC bestpractices for untargeted metabolomics.To establish mechanisms to enable the metabolomics communityto adopt QA/QC best practices.To promote and support systematic training in QA/QC bestpractices for the metabolomics community.To encourage the prioritization and development of referencematerials applicable to metabolomics ortia/mQACC/Reference and Test Material Working GroupDissemination of Current QA/QC PracticesWorking Group Think Tank on Quality Assurance and Quality Control for UntargetedMetabolomics Studies (October 2017)Workshops: Metabolomics 2018 and 2019 Workshop: QA and QC inUntargeted MetabolomicsThink Tank on Quality Assurance & Quality Control for UntargetedMetabolomics StudiesQuality Control in Untargeted MetabolomicsQuality Assurance PanelConferencesWebex meetings3819

/Built 1300Built 4900039Synopsis Metabolomics is coming of age for theregulators (best practice guidelines, discussions, few papers) LC-MS is going to be a major part of thesolution (and the problem!) We need to be very cautious with data(instrument variation dependant) Always validate data before statistical analysis(Easy to produce statistical artifacts based on bad quality data!)4020