Method development for optimizing analysis of ignitable liquid residues using flow-modulated comprehensive two-dimensional gas chromatography

The abundance and composition of matrix compounds in fire debris samples undergoing ignitable liquid residue analysis frequently leads to inconclusive results, which can be diminished by applying comprehensive two-dimensional gas chromatography (GC × GC). Method development must be undertaken to fully utilize the potential of GC × GC by maximizing separation space and resolution.

journalhomepage:www.elsevier.com/locate/chroma

chromatography

Nadin Boegelsacka,b,∗, Kevin Hayesa,c, Court Sandaua,d, Jonathan M Witheye,

Dena W McMartinb, Gwen O’Sullivana

a Department of Earth and Environmental Sciences, Mount Royal University, 4825 Mount Royal Gate SW, Calgary, AB Canada, T3E 6K6

b Department of Civil, Geological and Environmental Engineering, University of Saskatchewan, 57 Campus Drive, Saskatoon, SK Canada, S7N 5A9

c Manchester Metropolitan University, Ecology & Environment Research Centre, Chester Street, Manchester, U.K., M1 5GD

d Chemistry Matters Inc., 104-1240 Kensington Rd NW Suite 405, Calgary, AB Canada, T2N 3P7

e Department of Chemistry and Physics, Mount Royal University, 4825 Mount Royal Gate SW, Calgary, AB Canada, T3E 6K6

a r t i c l e i n f o

Article history:

Received 24 May 2021

Revised 29 July 2021

Accepted 23 August 2021

Available online 26 August 2021

Keywords:

Design of experiment (DoE)

Response surface methodology (RSM)

Multidimensional Analysis

GC × GC-TOFMS

Fire Debris

ILR

a b s t r a c t

Theabundanceandcompositionofmatrixcompoundsinfiredebrissamplesundergoingignitableliquid residueanalysisfrequentlyleadstoinconclusiveresults,whichcanbediminishedbyapplying compre-hensivetwo-dimensionalgaschromatography(GC× GC).Methoddevelopmentmustbeundertakento fullyutilizethe potentialofGC× GCbymaximizingseparationspaceandresolution Thethreemain areastoconsiderformethoddevelopmentarecolumnselection,modulatorsettingsandparameter opti-mization.Sevencolumncombinationswithdifferentstationaryphasechemistry,columndimensionsand orthogonality wereassessedforsuitabilitybasedontargetcompound selectivity, retention,resolution, andpeakshapes,aswellasoverallpeakcapacityandareause.UsingBox-Behnkendesignof experimen-tation(DoE),theeffectofmodulatorsettingssuchasflowratioandloopfillcapacitywereevaluated us-ingcarbonloadingpotential,dilutioneffect,aswellastargetpeakamplitudeandskewingeffect.Therun parametersexploredfor parameteroptimizationwereovenprogramming,inletpressure (columnflow rate),and modulationperiod.Comparing DoEapproaches,Box-Behnkenand Doehlertdesignsassessed sensitivity,selectivity, peakcapacity, andwraparound;alongside target peakretention,resolution, and shapeevaluation.Certifiedreferencestandardsandsimulatedwildfiredebriswereusedformethod de-velopmentandverification,andwildfiredebriscasesamplesscrutinizedformethodvalidation.Thefinal methodemployedalowpolaritycolumn(5%diphenyl)coupledtoasemi-polarcolumn(50%diphenyl) andresultedinanaverageSeparationNumber(SN)exceeding1inbothdimensionsafteroptimization SeparationNumbersof18.16forfirstand1.46forseconddimensionwithoutwraparoundforcompounds withatleastfouraromaticringssignifiedsuccessfulseparationofalltargetcompoundsfromvaried ma-trixcompositionsand allowedforeasyvisualcomparison ofextracted ionprofiles.Massspectrometry (MS)wasrequiredduringvalidation todifferentiateionswherenobaselineseparationbetweentarget compoundsandextraneousmatrixcompoundswaspossible.Theresultingmethodwasevaluatedagainst ASTME1618andfoundtobeanidealroutineanalysismethodprovidinggreatresolutionoftarget com-poundsfrominterferencesandexcellentpotentialforILRclassificationwithinacomplexsamplematrix

© 2021TheAuthors.PublishedbyElsevierB.V ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/)

1 Introduction

Ignitable liquidresidue(ILR)analysisprovidescrucialevidence

for arson casesby determining thepresence, type, andsource of

∗ Corresponding author

E-mail address: nboegelsack@mtroyal.ca (N Boegelsack)

an ignitable liquid,which denotesthe remnantsof substancesor mixturesofsubstancesusedtoaidtheinitialdevelopmentor esca-lationofafire.ASTMissuesthemostwidelyusedstandard meth-ods forILRs, covering variousmethods of extractionandanalysis

by GC–MS.ASTME1618 basesILR classificationon their chemical composition,includingpresence ofanalytesofinterest (orgroups

of analytes),equivalent n-alkane carbon range, andboiling point (bp)ranges[1,2]

https://doi.org/10.1016/j.chroma.2021.462495

0021-9673/© 2021 The Authors Published by Elsevier B.V This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ )

Despite well-established ILR class profiles and standardized

methods of analysis, arson cases have amongst the lowest

con-viction ratesinNorthAmerica Theremnant natureofILRs,

com-binedwiththepresenceofcombustion,pyrolysis,andmatrix

com-poundsinlargeconcentrations,amplifiesthechallengeof

identify-ingandcharacterizingILRs,leadingtofrequentfalsenegatives[3]

ILR patternrecognitionisevenmore challenginginwildfire

sam-pleswherethephysico-chemicalcompositionofprominentmatrix

compounds is very similar to that of the ILR target compounds,

leadingto anincrease inhomologpeaks(compoundssharingthe

same spacein thechromatogram)[2].In addition,accelerant

ap-plication inwildfireeventscanbemoredispersive, without

pool-ing orother areas ofhigh ILR concentrationsto occur as is

pos-sible in structural fire scenarios, resulting in significantly greater

abundance of matrix interference.Homolog peakscan frequently

beseparatedbyselectionanalysisinMS,whichisoneofthe

rea-sonswhyMSisthepreferreddetectionsystemoverflame

ioniza-tion detectorfor ILRanalysis.However, if severalcompounds

oc-cupy thesamespace, andone ormoreofthesecompounds share

thesamemajorions,theILRsignalcannolongerbedistinguished

frommatrixinterferences

Comprehensive multidimensional gas chromatography

(GC × GC)hasbecomea populartool forhighresolution

separa-tion of complex organic mixtures by reducing the occurrence of

overlapping homologpeaks andemploying differentmechanisms

ofseparation inasingleanalysis.Recent studieshavehighlighted

the potential application for fire debris analysis [2,4,5,6,7] The

majority of these GC × GC studies have been performed on

thermally modulated (TM) systems [2,5,6], but flow-modulated

(FM) GC × GC has been gaining increased popularity due to its

financial accessibility and robust peak parameters in the second

dimensionandefficientmodulationofallvolatilities[8,9,10,11,12]

Integration of GC × GC into routine analysis and commercial

laboratories has been slow due to the perceived complexity of

GC× GCfunctionalityanddataanalysis,aswellastheassumption

that method development will be costly and require training as

it is considered far more complicated than in single dimension

systems

MethoddevelopmentandoptimizationinGC× GCsystemsare

drivenlargelybythedesiretofullyexploitseparationpotential.A

number ofstudies havedescribedGC× GC method optimization

[5,6,10,13,14], including method comparisons [9,11,13] as well as

the applicationofGC× GC toindustrialfires[6].However,much

oftheliterature pertainstoTMsystems,explorestheoretical

con-cepts, such asnumericalmodeling [15],andparameter

optimiza-tion[16]ordonotincludepotentialmatrixeffectsduringmethod

development [5,9,17] Although the 2D separation potential is a

majoradvantageofGC× GC,existingmethoddevelopmentstudies

rarelyprovideadetaileddiscussiononcolumnselection, whichis

oftenperformedonatrialanderrorbasisuntilasuitable

combina-tionisfound[5,6,8,9,11,13],oroptimizetheactualspatial

distribu-tion inthe chromatogram.Instead,thesestudies favortheoretical

peak capacity,which assumes an evenly randomized distribution

of peaks across the available space Lack of spatial optimization

can be partially attributedto the choice ofcolumn combination,

non-optimizedcolumnconditionornon-optimizedmodulator

set-tings Relevant hardwarechoices,such ascolumn combinationor

flowmodulatorsettings,arerarelydiscussedindetailandstill

pri-marilydeducedthroughtrialanderror

Complex method development with a high number of

vari-ables or highlyinteractive variables, asis thecasefor

chromato-graphic analyses, benefits fromdesign of experiments (DoE).The

two main advantages ofDoE are reduction in numberof

experi-mental runsnecessary andsimultaneous investigation of

interac-tion effects Evaluationof responses is oftencoupledto response

surfacemethodology (RSM),which plotspredictedresponses ina

Fig 1 Visualisation of Box-Behnken (A) and Doehlert (B) matrix with center point

(0) in dark gray and sampling points in light gray Two of the three sampling planes for Box-Behnken are shown in light blue Doehlert model is represented as three- dimensional model (top) and two-dimensional model as viewed from above (bot- tom)

multidimensionalmatrixandallowsdeterminationofoptimum re-gionsforvariablesettingsataglance

Awidevarietyofexperimentaldesignsexistandhavebeen ap-pliedtoGCandGC× GCmethoddevelopment, includingfull fac-torial,centralcomposite,Box-Behnken,andDoehlertuniformshell design [14,18,19].Each designhas its own advantagesand disad-vantages In this study, Box-Behnken (sampling pointsvisualized

inFig.1A) andDoehlert(sampling pointsvisualizedin Fig.1Bas 3Dmodel(top)and2Dmodelasviewedfromabove(bottom)) de-signswereusedforanalysis,withacomparativeevaluationofboth models for the final parameter optimization.With the exception

ofmodelcomparisons[18],Box-Behnkenhasbeenappliedto sam-plepreparationstepsmostfrequently[14,20]whereasDoehlerthas beenpredominantly appliedto GCseparation parameters[20,21] Bothdesigns are compatiblewithRSMandhave asimilar model efficiencyandaccuracyforthreevariables[18]

In this paper, we introduce a systematic workflow for FM

GC× GC methoddevelopmentutilizing thebenefits of DoE,and describingindetailthemostimportantGC× GCset-updecisions (columnchoice&modulatorsettings)andparameteroptimization basedonILRanalysis.Inadditiontoreferencestandard materials, simulatedandactual wildfiresampleswereinvestigatedto evalu-ate theproposed ILRanalysis method inrelation to ASTM E1618 [1]undermatrixinterferenceeffects

2 Materials & methods

2.1 Standards and reagents

Benzene(99.9%),carbondisulfide(CS2,99.9%), d8-naphthalene (Cat#: AC174960010), d10-ethylbenzene (AC321360010), dichloromethane (DCM, 99.9%), methanol (99.9%) and toluene (99.9%) were obtained from Fisher Scientific (Ottawa, ON, Canada) A C7–C30 saturated alkanes certified reference mate-rial (Cat#: 49,451-U), PAH Mix 3 certified reference material (Cat#:861,291), and d12–1,3,5-trimethylbenzene (Cat#: 372,374–

1 G) were purchased from Sigma Aldrich (Supelco, Bellefonte,

PA, USA) Deuterated Kovats-Lee retention index mix (KLI mix, consisting of d22-decane, d32-pentadecane, d42-eicosane,

d50-tetracosane, d8-toluene, d8-naphthalene, d10-phenanthrene,

d12-chrysene and d12-benz(a)pyrene) was acquired from Cambridge

Isotope Laboratories, Inc (Tewksbury, MA, USA) and

d14–1,2,4,5-tetramethylbenzene (Cat#: d-0269) was purchased from CDN

isotopes(Pointe-Claire,QC,Canada)

Arecoverystandardwascreatedbycombiningd8-naphthalene,

d10-ethylbenzene, d12–1,3,5-trimethylbenzene and

d14–1,2,4,5-tetramethylbenzene inmethanol ata concentration of200 ng/ml

each A 1 ppm aromatic standard mixture containing alkanes

(49,451-U),benzene,PAHs(861,291),andtolueneinDCMwas

pre-paredforthedeterminationofRIvaluesincomparisontoa1ppm

standardofthedeuteratedKovats-Leeretentionindexmix

AccelerantswerepurchasedinCalgary,AB,Canadaandincluded

acompositesampleofgasolineanddiesel.Thecompositegasoline

sample wascreatedby combining 21 gasolinesof various octane

rating 87 (n = 6), 89 (n = 7), 91 (n = 6), 94 (n = 2) collected

fromsevenstations.Thedieselcompositewascreatedby

combin-ingfourdieselsamplescollectedfromfourstations

Controlled burns were conducted on pine and cedar blend

woodchips (Pestell, New Hamburg, ON, Canada), charring them

to a 50% burn Metal quart cans (Uline, Edmonton, AB, Canada)

were filled to approximately 80% with controlled burn material

(41.5 g ± 1.5g) and spiked with250 μl recovery standard

Sim-ulated ILR samples were created by additionally spiking 50 μl

each ofthegasoline anddiesel compositesamples.Allcanswere

extracted in accordance with ASTM guidelines [22] at 90 °C for

16hours

2.2 Analytical system overview

AllanalyseswereperformedonanAgilent7890AGC(PaloAlto,

CA),retrofittedwithanInsightflowmodulator(Sepsolve,

Peterbor-ough, UK) andcoupled to a Markes BenchTOF-Select mass

spec-trometer(Llantrisant, UK).Throughoutthestudy,theinjector was

operated at 250 °C in split mode with a 25:1 ratio and1 μl of

samplewasinjected viaAgilentG4567A(PaloAlto, CA)

autosam-pler.Heliumwasusedascarriergaswithanaveragelinearvelocity

of4.0cm − 1.The MStransferlineandionsourcewere heldat

250 °C The electron energyapplied was70 eV andthe scanned

mass range was 50–400 m/z in electron ionization mode Data

was acquired and processed using the ChromSpace software (V

1.5.1, Sepsolve, Peterborough, UK)and MicrosoftExcel(Microsoft,

Redmond, WA) Adeconvolution algorithm wasused for

integra-tion,withaminimumioncount300,minimumabsoluteareaand

heightof1000,andpeakmergingat10%overlap

The modulatoristhe“heart” ofa multidimensionalsystem,its

primary functionistotrap,refocus,andinjectsequentiallythe

ef-fluentfromtheprimarycolumnontothesecondarycolumn.Flow

modulators are valve-based andusedifferential flowsto ‘fill’ and

‘flush’ a sampleloop (Fig.2) [23].Hardware components include

first and second dimension columns (1D & 2D), a sample loop,

controlled supplyofheliumfromauxiliarylines(regulatedbythe

PneumaticControlModule(PCM)),andcapillarybleedandtransfer

lines During theloading step,the effluentfromthe first column

fills the sample loop When the modulation valve switches the

auxiliary flow, this reinjects the content of the sampleloop into

theseconddimension.Thetimetakentocompleteonemodulation

or‘cut’ofthefirstdimensioniscalledthemodulationperiod(PM)

Typically, effluent from the first dimension are cut by the

mod-ulator into2–5modulation sliceswithseparationon the

second-dimensioncolumnoccurringveryfast,normallywithin38s

Fig 2 FM modulator hardware setup with column and loop dimensions Loop fill

and flush flow directions are depicted as solid and dashed arrows respectively

Method development in GC × GC is challenging but can be achievedby selectingappropriatecolumnsandinstrumentset-up, andoptimizingmodulatorandparametersettings.Mainelements

to consider before method development include composition of analytesof interest andprominentmatrix compounds aswell as employingtargetedversusnon-targetedanalysis.Theseinformthe practical requirements of methoddevelopment andoptimization, such asstationary phase chemistry,column dimensions, modula-torsetup, andoptimizationofovenanddetectorparameters (e.g flow,temperatureprogrametc.).Aschematicoverviewispresented

inFig.3.Theoverallgoalformethoddevelopmentistomaximize selectivity andsensitivity accordingto the application, which for ILR analysis translates to the separation of all target compounds [1]fromothertargetcompoundsaswellascommoninterferences

to allow foraccurate detection andquantification, pattern recog-nition to differentiate ILR classes[1], detector split for optimum

MSsensitivity,andappropriateruntimeforroutineanalysis(<90 mins).Withinthefollowingsectionsweoutlinethestepstakento develop an appropriate GC× GCmethod forILRanalysisusing a flowmodulator

Column selection and changes to the modulator hardware were consideredfirst astheserelate toinstrumental setup ofthe

GC × GC and concern the most important and difficult choices Modulatorandparameter optimizationwere completed following instrumentsetup

2.3.1 Selection of column set

Choice of column sets is one of the most important steps in method development Dictating selectivity of the method, it is drivenbythepropertiesofthesample,includingtargetandmatrix compounds,andtheobjectiveoftheanalysis.Aneffectivepairing shouldhaveappropriateretention,resolution,selectivity,andpeak shape.Themainchoicestoconsiderinachievingthisarestationary phasechemistry,columndimensionsandorder,aswellas orthog-onalityandfilmthickness

Elements to consider:

Selectivity, Retention, Resolution, Peak Capacity / Area Usage, Peak Shape

Achieved by:

Stationary Phase Chemistry, Column Dimensions, Orthogonality

Elements to consider:

Sensitivity, Peak Skewing / Amplitude, Carbon Loading, Dilution Effect

Achieved by:

Flow Ratio, Loop Fill, Detector Efficiency

Elements to consider:

Sensitivity, Selectivity, Retention, Resolution, Peak Capacity / Area Usage, Peak Shape, Wraparound

Achieved by:

Oven Programming, Inlet Pressure (Column Flow), Modulation Period

Analytical Question,

Target Compounds,

Sample Matrix

Analytical Answer, Separation of Targets, Resolution Interference

Column Selection Modulator Settings Parameter Optimization

phase length

i.d

film thickness

column order

loop bleed line

transferline detector 2

transferline detector 1

detector settings

PM

column flow oven

settings

injector settings

auxiliary pressure

Fig 3 Workflow steps applied to method development from left to right, detailing practical considerations underneath each step

Table 1

Column pairings installed in first and second dimension including physical dimensions, station- ary phase, and stationary phase orthogonality

Combination 1 D (length m x

i.d mm x f.t μm)

2 D (length m x i.d mm x f.t μm)

Orthogonality

(25 m x 0.15 mm

x 0.25 μm)

BPX50 (5 m x 0.25 mm

x 0.15 μm)

Non-polar × semi-polar

(30 m x 0.25 mm

x 0.5 μm)

BPX50 (5 m x 0.25 mm

x 0.15 μm)

Non-polar × semi-polar

(30 m x 0.25 mm

x 0.25 μm)

BPX50 (5 m x 0.25 mm

x 0.15 μm)

Polar × semi-polar

(30 m x 0.25 mm

x 0.15 μm)

BPX50 (5 m x 0.25 mm

x 0.15 μm)

Polar × semi-polar

(30 m x 0.25 mm

x 0.25 μm)

BPX50 (5 m x 0.25 mm

x 0.15 μm)

Non-polar × semi-polar

(25 m x 0.18 mm

x 0.18 μm)

Mega Wax HT (5 m x 0.25 mm

x 0.15 μm)

Non-polar × polar

(25 m x 0.18 mm

x 0.18 μm)

Mega-5MS (5 m x 0.25 mm

x 0.5 μm)

Non-polar × non-polar

Based on expected analytes and maximum column

tempera-tures,severalcolumncombinationswere tested(Table1).The

in-vestigation was divided into two steps, wherecombinations 1–5

(Table1)werecomparedwiththesame2Dcolumntochoosethe

most appropriate 1D column The chosen column wasthen

cou-pled withadditional2Dcolumns(combinations 6&7) toexplore

whichcolumncouplingmaximizedthepotentialperformance.The

majority of applications start with a non-polar 1D column (5%

diphenyl/95% dimethyl polysiloxane) connected to a more polar

2D column (50% diphenyl/dimethylpolysiloxane) as this column

set separates analytes based on two mechanisms; boiling points

(1D) and polarity ranges (2D) Columns selected for comparison

(Table1) were chosen based oncommoncomposition ofwildfire

samples[2],systemrequirementsforflowequilibration(see2.3.2.),

andcurrentroutinecolumnsusedinGC–MS(non-polar,commonly

30mlength,0.25mminternaldiameter(i.d.))

Column dimensions were chosen with expected acceptability for routine analysis in mind This included column lengths not longer than 30 m to reduce sample analysistime, andthin film thickness (< 0.5μm) asmost ILR compounds are not extremely volatile.Narrowtomedium i.d.(0.15 mmto0.25mm)were con-sideredaslargeri.d.columnsrequirehigherflowrates,whichcan

bedifficulttobalanceinflowmodulation.Orthogonality combina-tions,separationresultingfromindependentretentionmechanisms [24], were investigated for stationary phases It is generally as-sumedthathighlyorthogonalcolumnsetsincreasetheseparation space occupied in the second dimension butinstances of seem-ingly non-orthogonal sets performing better have been reported [24].Although stationaryphaseis traditionallythemain measure

as it relates to selectivity and therefore separation mechanisms, severaldefinitionsfororthogonalityexistintheliterature [25,26] Forthispurpose,orthogonalitywashereinusedasthetraditional

eval-uatedusingpeakcapacityassuggestedinSchure&Davis[26]

to-getherwiththeactualpercentageofchromatographicareautilized

Performance betweencolumn combinations wascompared by

evaluating peak capacity betweenthe solvent peak and eicosane

/ naphthalene,separation efficiency, and total percentageof area

used Flow rates were changed for individual columns in order

to keep the flow ratio stable for comparable results (see 2.3.2)

As minimumrequirementformethoddevelopment, eicosane was

chosen as representative for the latest eluting compound in the

firstdimension basedonthetestmixturedetailedinASTME1618

[1] Naphthalenewaschosen as representative approximation for

thelatestelutingcompoundintheseconddimensionbasedonthe

targetcompoundslistedinASTME1618[1]

Theareacoveredinbothretentionspaceswascalculatedin

ac-cordancewiththeCorderomethodincludingwraparoundpeaksas

outlinedinEqs.(1)and(2)[27]:

With l representing the length in minutes (1D) and w

repre-sentingthewidthinseconds(2D)

A%= A co v ered

WithAcovereddescribing the areapertaining to the column

com-binationandApotentialasthetotalareainthechromatogramas

cal-culatedbyEq.(1)respectively.Incaseswhereruntimeswere

insuf-ficienttoelutetargetpeaksuptoKI12400,A%wasmultipliedwith

thepercentageoftargetKI1coveredandannotatedasA%adj

Separation efficiency was compared by tabulating separation

numbers acrosstheKLImix componentsaccordingtothe

separa-tionnumber(SN)asshowninEq.(3):

SN= t R ( j+1) − t R ( j )

W h ( j+1)+W h ( j )− 1 (3)

With tR(j) andtR(j+1) representing retentiontimes of two

con-secutive compounds ofinterest and their respective peak widths

at half height shownas Wh(j) andWh(j+1) Alkanes were used as

referencepointsforSN1,whereasSN2 wasbasedonLI2 reference

compounds SinceWh wasnot readilyavailable inthesecond

di-mension,widthatbasepeakwasusedinsteadforSN2

After calculatingretentionindicesforcompoundsofinterestin

accordance withBoegelsack etal.[7],individual peak resolutions

fortargetcompoundswerecalculatedaccordingtoEq.(4):

R S =1.177(R I2− R I1) (SN+1) (4)

WhereRI1 andRI2areretentionindexnumbersoftwo

consec-utivecompoundsofinterestinasingledimension

Thepeakcapacity(n)wascalculatedbasedonGrushka[28]as

outlinedinEqs.(5)and(6):

N=5.54t

R

W h

2

(5)

WhereNistheplatenumber,tRistheretentiontimeandWh

isthewidthathalfpeakheightofagivencompound

1n=1+N41/ 21nt n

Where 1n is the peak capacity of the first dimension, tn is

the retention time ofeicosane andtS representsthe solvent

elu-tiontime.Theoutputofthepeakcapacityrepresentsthepotential

numberofcompounds thatcan befullyseparatedwithinits

spa-tial boundariesin the firstdimension To account forthe second

dimension,Eqs.(5)&6wereappliedtonaphthaleneforvalues

re-latingtothe seconddimension toyield2n Finally,thetotalpeak

capacity(n)wascalculatedasshowninEq.7:

Using the total peak capacity, the number of possible distin-guishedpeakspinachromatogramwiththesameboundarieswas calculated as Eq (8) in accordance with Bertsch [29], assuming

RS=0.5:

Whereprepresentsthenumberofpeaksappearingassinglets, and m is the number of components in the sample, which was averaged to 3000for a typical diesel on wood matrixsample to simplifythecomparison

2.3.2 Modulator settings

FMsystemsareconsideredmorecomplextooptimizethanTM systems,duetotheprincipaloftheiroperationsandrestrictionsin hardware [13,23] Despite the importance of optimizing modula-torperformance,methoddevelopmentstudiesoftenfocusonlyon the“soft” parameters,solely includingmodulationperiod (PM) in their considerationsforFM systems[11,23].Instead,PM was con-sideredpartoftheparameteroptimizationinthisstudyand mod-ulator settings concentrated onoptimizing flow ratio andcarbon loadinginlightofthedilutioneffect

Flow ratiowasshown tohave a major effecton peak shapes, which in turn affects the ability for consistent integration [10] Calculating the flush: fill volumetric ratio (see Fig 2) in accor-dancewithHarvey &Shellie [10], peak skewingandpeak ampli-tude were evaluated in relation to 2D separation and resolution potential Harvey & Shellie [10] recommended a flow ratio > 30 duringmethoddevelopmenttominimizetheeffectofpeak skew-ing,whereas thispaperassumes flowratios < 40assub-optimal basedonobserveddifferencesduringcolumnselection(2.3.1) Carbon loading was investigated two-fold: by calculating the balancebetweenPCMpressureandbleedlineflowtoensure con-sistentloopfill,andby applyingaBoxBehnkenmodel(see2.3.3.)

toevaluate thex-line lengthsanddiameters (seeFig 2)for opti-mizingdetectorsplitflow

The flow calculator provided by Sepsolve (Peterborough, UK) wasusedto calculatethe requiredsettings tobalance PCM pres-sure and bleed line flow, as well as calculate flow rate results forFID andMS x-lines forthe 4-factor Box Behnken model (see Table 2) Only one detector was in use for this setup as ASTM E1618 [1] prescribes the use of MS as detection method There-fore,thesecond x-line(Fig 2)actedasbleedlineforthe2D col-umn, and holdup time, which is a necessary consideration in a setupwithtwodetectors,couldbedisregardedinthemodel Base-linesettingsassumedforthemodelwereuseofHeliumascarrier gas, 40°C oven temperature, PCM pressure 27.5 psi, 1Dflowrate 0.5ml/min,2D flowrate16 ml/min,anddimensions asshownin Fig.2

Rootmeansquarederrorofprediction(RMSEP)wascalculated

inaccordancewithEq.9:

RMSEP=



1

I

I



i =1



ˆ

y i − y i 2

(9)

Where Irepresents the number ofcompounds presentin the mixture, y is the predicted average resolution and y is the ex-perimentallyrecordedaverageresolutionineachdimension Two-way analysis of variance (ANOVA) with 95% level of confidence wasused toevaluatestatisticalsignificance.Standardmodel eval-uations(multiplelinearregressionsanderrorestimates)were per-formedinJMPTrial16.0.0(SASInstituteInc.,Cary,NC,USA) RSM was also performed in JMP Trial 16.0.0 using desired flowrateoutputsasoptimalregions

2.3.3 Run parameter optimization

Optimization wascompleted usingDoE due tothe interactive nature andnumberofvariables considered Runparameters

cho-Table 2

Corresponding coded and actual values used for calculating flow rates of FID and MS x-lines in 4-factor Box Behnken model

Coded value l FID x-line (m) l MS x-line (m) i.d FID x-line (mm) i.d MS x-line (mm)

senasvariablesforthemodelsweremodulationperiod(PM),flow

rate(asafunctionofinletpressure),andovenramp,astheseare

generallyconsideredto havethehighestimpactonGC× GC

per-formance[13,30]

The columnsetconsistedof anon-polar, 5%diphenyl, column

(30 m × 0.18 μm i.d × 0.18 μm film thickness) in the first

di-mensionandasemi-polar,50%diphenyl,column(5m× 0.25μm

i.d× 0.18μmfilmthickness)intheseconddimension.Actual

val-uesasusedforcalculationscanbeseeninTable3,whereascoded

values were standard for Box-Behnken (−1, 0, 1) and based on

the prescribed uniform shell design simplex laid out inDoehlert

[31]forthreevariables

2.3.4 In addition to the evaluation tools described in 2.3.1., average

Sn and Rs for both dimensions were calculated for modeling using

based on their RMSEP for results as given by eq (9)

RSMwasappliedtoresultsofthemodelstodeterminethe

op-timum performancezonesforthethreevariablesunder

investiga-tion,usingmaximaforA%adj.andtheaveragepeakresolutions(RS)

ofeachdimensionasdesiredoutputs.RSMandmodelevaluations

wereperformedinJMPTrial16.0.0

3 Results & discussion

3.1 Column selection

The comparative results of all column combinations are

pre-sented in Table 4 Comparing first-dimension columns

(combina-tions 1 to5), combinations1& 3showed thegreatestresolution

potential inboth dimensionswith thehighest peakcapacity and

potentialnumberofdistinguishablepeaks.Althoughcolumn3had

a higherarea usage andpeak capacity,iteluted lessthan70% of

target compounds in thesame time frame asthe other columns

andresultedinextensivewraparoundstartingbetweenmono-and

diaromaticcompounds.Wraparoundcomplicatesdataanalysisasit

requiresmanual integrationofdata,makingitunsuitablefor

rou-tineanalysis.Timelyelutionisanotherkeyfactorforroutine

anal-ysis;thus,column1wasselectedfor2Dcomparison

As column 1 had a narrow i.d., an equivalent column was usedwitha shorterlength toreduce solventdelayandaslightly largeri.d.topreserveplatepotential.Amediumfilmthicknesswas electedtoraisethetheoreticalplatenumberwhilesimultaneously keepingthecolumnbleedandretentiontimelow forcleanerand fasteranalysis.Comparingseconddimensionperformance(column combinations1,6and7),columncombination7showedthe great-estresolutionpotentialintheseconddimension,leadingtoan in-creasedarea usage.However,thepotentialnumberof distinguish-able peaks performed the worst out of all three combinations Withcolumn combination6 providing the bestresolution in the firstdimensionbuttheworst resolutioninthesecond dimension, column combination1 offered the best overall result for further methoddevelopment

3.2 Stationary phase chemistry

Table4 showsthat combinations1,2 & 5(non-polar × semi-polar)haveagreateraverageresolutionin1Dthancombinations3

&4(polar× semi-polar).Sincepolarityandselectivityofthephase determineretentionandinteractivityofcompoundgroups,thiscan

beattributedtothestationaryphase(5%phenylagainstwax) Ad-ditionally,the maximum operatingtemperature forwax columns

islowerthanthatfor5%phenylcolumns.Thisrestrictsthe poten-tialresolutionforhighbpcompounds,whichisproblematicwhen identifyingheavierILRsinfiredebrissamples

Separation in the first dimension is very important and pri-oritizedin methoddevelopmentasthe second dimensioncannot makeupforanyresolutionlostin1D.One-dimensionalGCmethod developmentprinciplesapply; therefore,linearvelocityshouldbe operated as closely as possible to the optimum indicated from thevanDeemtercurve.Otherexamplesincludeoventemperature rates contributing to the balance between peak shapes and co-elutions,orinletsplitratiosbeingconsideredinconjunctionwith columni.d.toavoidoverloading

Looking closely at the second dimension, negative results for

SN2 avg (combinations 2& 4to 6)indicate that target compounds elutedonthesameplanein2D,asthedifferencebetweentR2was smallerthantheaveragepeakwidth.Consideringthatthebaseline

Table 3

Actual values for chosen parameters (modulation period, inlet pressure, oven ramp) used in model development

Run

P M (s) Inlet pressure (psi) Oven ramp ( °C/min) P M (s) Inlet pressure (psi) Oven ramp ( °C/min)

Table 4

Overview of column combination performance including area percentage covered (A%), av- erage separation (SN) and resolution (R S ) in both dimensions, total peak capacity for the system (n) and the potential number of distinguishable peaks (p) appearing as singlets in

an average gasoline/diesel mix + wood matrix sample

Column combination A% SN 1

avg R S1avg SN 2

avg R S2avg n P

widthhadtobeusedforthiscalculation,negativeresultswere

ex-pectedinlightoftherelativelyshorttimeframeformodulation

Columnorderbetween1Dand2Disanimportantfactorforany

GC × GC system Each column characteristics (phase, film

thick-ness, i.d.andlength)isindependentlyimportantforoptimal

sep-arationbutmustalsobeconsideredinrelationtothepairing

col-umn.ThisisparticularlyimportantinFMsystems,whereitis

cru-cial to have an appropriate flow ratio betweenthe columns (see

3.2.1.).Eachflowisaffectedbylengthandi.d.ofitsrespective

col-umn While TM-GC × GC followsthe traditionalrule ofa longer

columnequaling betterseparationalbeitlonger runtime,FM

im-posesadditionalrestrictions.Alongercolumnstillcorrespondstoa

highertheoreticalplatenumber;butalsoleadstohigherpressures

andcomplications inflow balancing.As a result,it isuncommon

to seecolumns exceeding 30min FM systems.When appliedto

ILRanalysis,shorterlengthsupto30mprovedtobesufficientfor

targetcompoundseparations

Balancing i.d.s,TMsystemmethoddevelopmentoftensuggests

usingthesamei.d.infirstandsecond dimensiontoheighten

the-oretical platenumberandoptimizeflow[32],thisisnot thecase

in FMsystems asevident fromTable4.The 0.25mm i.d

combi-nations(2, 4&5) donotshow agreat advantageintheRS1,and

comparepoorlyinareausageandRS2sincethesecond-dimension

i.d.isthesamediameter.This,inturn,negativelyaffectstheir

abil-ity to fully separate large numbers of peaks (see p, Table 4) In

additiontotheregularconsiderationsrelatingtoinjectionmethod,

phaseratio,orretention;i.d hasalargeimpactonflowratioand

related pressure restrictions In FM systems, a smalleri.d inthe

first dimensionwillspeedup analysisandallowforeasier

equili-brationofflow(see3.2.)

3.4 Orthogonality and film thickness

Looking at area usage and resolution in both dimensions for

column combinations in Table 4, combination 7, which had no

orthogonality, outperformed all others in regard to area and 2D

resolution, whereas combination 6 with high orthogonality

out-performed all others inpeak capacity, number ofdistinguishable

peaks and 1D resolution Therefore, it appears that the

orthogo-nality of stationary phases is not the primary factor influencing

separation in the second dimension, confirmingthat true system

orthogonalityisaffectedbyanumberoffactors[25,27].An

impor-tant factor inthe second dimension appearsto be film thickness

basedoncombination7

In theory,film thickness directlyaffects retention (the thicker

thefilm,thelongercompoundsareretained).Thiswasnotevident

from average retention in 1D (Table 4) where the difference

be-tweenvariousthinfilmsoverthelengthofthefirst-dimension

col-umnwasnotapparent.Asfilmthicknessshouldalwaysbe

consid-Fig 4 Comparison of the flow ratio effect on peak skewing (black dotted line) and

amplitude enhancement as shown on naphthalene with flow ratios of 60 (A) and

30 (B)

eredinconjunctionwiththeinternaldiameter,whichplaysavery importantroleforflow-regulatedGC× GC,columnlengthandi.d betweencombinations1,6&7werekeptconstant.Giventhe dif-ferenceinlengthbetween1Dand2D,andthesubsequentamount

oftimecompoundsspendineach dimension,itmakessense that filmthickness hadamuch largerimpact onthe 2Dseparation.A film thickness of 0.15 μm wasselected in 2D to reduce the po-tentialforearlywraparoundduringparameteroptimization.Since filmthicknessoftheseconddimensionhasasignificanteffectonly

on2Dseparation,thiscanbeincreasedatalaterpointasrequired withoutaffectingotherdevelopmentparameters

Area usage inthe second dimension isfrequently not consid-eredimportantinmethoddevelopment.Instead,phase orthogonal-ityisfavoured.Table4illustrates thatpeak capacityandarea us-agearecloselylinked Usingthesevariables asa measureoftrue systemorthogonality assuggestedby Schure &Davis [26], wasa betterapproachforevaluatingcolumnperformancethansolely re-lyingonphaseorthogonality

3.5 Modulator settings 3.5.1 Flow ratio

Fig.4depictstheeffectofflowratioonnaphthalene.Thepeak skewingeffectisrepresentedbytheblackdottedlineandisclearly enhanced in Fig.4B (flow ratio= 30) compared to Fig 4A(flow ratio=60).Theeffectonpeakamplitudecanalsobeobservedas thepeakheightinAismoreconcentrated,whichisrepresentedby darkredarea, andlessdrawnout, whichisshownby adecrease

ingreen&blueareas

As Harvey & Shellie [10]highlighted, sub-optimal flow equili-bration leads to an increase in observed modulation effects, i.e

a higherdegree ofpeak skewingandlower peak amplitudes Al-though they concluded that a flow ratio > 30 was appropriate, Fig.4Bshowedpronouncedartifactsofmodulationeffects.In gen-eral, non-focusing modulation displays broader peak widths in comparisontofocusingmodulation,whichmeans asmaller theo-reticalresolutionpotential butincreasedrobustnessoverlong pe-riodsoftime[12].Avoidanceofmodulationeffectsisthereforean important undertaking to ensure consistently well-shaped peaks

Table 5

Dependency table summarizing the impact of increasing the length and internal diameter of transfer lines to MS and FID on the carbon loading to the detector (dilution effect)

Carbon Loading / x-line Flowrate

Impact Quantifier

Length of FID x-line

Internal Diameter

of FID x-line

Length of MS x-line

Internal Diameter

of MS x-line Carbon loading to

FID

loading/flowrate

ing/flowrate

–

Carbon loading to

MS

Low ↑ load-

ing/flowrate

ing/flowrate

–

andmaximumresolution,andaminimumflowratioof40was

re-quiredforthissetup.Althoughhigherflowratiosbetweencolumns

leadtobetterpeakshapes,theymustbebalancedbychoosing

ap-propriatemodulatordimensionstoensureconsistencyofflowratio

andcarbonloadingofthedetector[13]

3.5.2 Carbon loading & dilution affect

InFMsystems,optimizationofcarbonloadinganddetector

ef-ficiency is achieved by balancing flows across the 7- and3-port

valves(seeFig.2)

Basedontheflowcalculator,a2-mlengthwassufficientto

bal-ance flowsacrossthe7-portvalveforthecolumnsetupchosen in

3.1.witha0.1-mmi.d.bleedlinetoensureconsistentcarbon

load-ingin2D.Establishedstandard1Dcolumnflows,whichneedtobe

balanced withbleed lineflow forconsistent loop fill(see Fig.2),

are<1ml/minforcapillarycolumns.Whencalculatingbleedline

flow, thechoice of bleed linei.d was prioritized asit is directly

relatedtoflowrate,whereaslengthisinverselyrelated.Therefore,

choosing a 0.1-mm diameter bleed line i.d prevented the need

forextremelength,whichinturnminimizedpotential carriergas

waste.An imbalanceinflowratioequilibriuminthe 7-portvalve

canleadtoovershootingorundershootingtheloop,whichinturn

negatively affectscarbonloadingandcausesloss of2Dresolution

viapeakskewing(depictedinFig.4)

Table 5 summarizes the variables toconsider when balancing

flows across the 3-port valve and the respective impact of their

changes on each detector flow rate Fig 5 shows an RSM graph

as one example of the relationships outline in Table 5,

specifi-cally therelationshipbetweenMSx-linelengthandFIDx-linei.d

to FID andMS flowrates.The FID flowrate isrepresented by the

yellow-red mesh, whereas the MS flowrate is displayed as light

purple-greenmeshwiththeoptimumregionwheretheyintersect

(11 ml/min and 4 ml/min respectively) The relationship shown

in Fig.5correlates to thebleed lineoptimization.FID flowis

di-rectlyrelatedtoFIDdiameterasshownonthey-axis,resultingin

thehighestflow(redarea)forthelargesti.d.,whereas MSlength

is indirectly related to MS flow, resulting in the lowest MS flow

(grayarea)forthelongestx-line.Theseprinciplesworkviceversa,

meaning that alongerFID x-line resultsindecreasedflow tothe

FIDandincreasedflowtotheMS.Finaldimensionsusedtoachieve

theoptimumratioarelistedinFig.2

The model showed significance for all individual parameters,

aswell astheinteractionbetweeni.d.s,andinteractions between

each x-line length andrespectively opposing x-line i.d The

RM-SEP ofthe modelwas 1.7with 2 = 0.98andhomogenous error

distribution Although resulting calculatednumbers were not

ex-act whencompared totheactual valueoutput ofthesystem,the

evaluationwasperformedrelativetothebaselinesettingsandthe

finaldilutionfactorwasadjustedaccordinglytomaketheuseofa

flowcalculatorsuitable

The desireddetectorsplit flow, alsoreferred toasdilution

ef-fect, regulates how much ofthe total 2Dflow is sent to the

de-tector A 1:4 ratio is considered ideal for MS detector efficiency

topreservemaximumsensitivitywithoutoverloadingthefilament

Fig 5 Flow model RSM displaying relationship between FID x-line i.d and MS x-

line length with effect on MS x-line flow (light purple-dark green, optimum region displayed in light purple) and FID x-line flow (light yellow-red, optimum region displayed in red & dark orange)

andintroducingdetectornoise,whichtranslatedtothesetoptima aftertakingintoaccountthedifferencebetweencalculatedand ac-tualvalues

3.5.3 Parameter optimization

Fig.6showsallRSMgraphsfor“soft” parameteroptimization, comparingBoxBehnken(top)andDoehlertuniformshell(bottom) design The RSM graphs show that area usage isoptimized with shortermodulationtimes(Fig.6A& D).InFMsystems,longerPM risk introducing pressure inconsistencies, which lead to poor re-peatability.RelatingA%andPM,favouringshortermodulationwas expectedasthisgreatly reduces thepotential area to filland in-creasespeakstability.However,aconsiderabledownsideisan in-creaseinwraparound,whichwasnotaccountedforinthisanalysis butwasevaluated manually outsideofmodels.The Box Behnken model(Fig.6A)only showsoneoptimum (whichheavily featured wraparoundfromLI2>150),whereasDoehlert(Fig.6D)displaysa secondoptimumregionwhichdidnotincludewraparoundforLI2

≤ 450.Asimilareffectcanbeseeninrelationtotheinletpressure andareausage (A/D),aswell asthesecond-dimensionresolution andinlet pressure (C/F) Doehlert matricesinvestigate values be-sidesthemodelextremathatBox-Behnkenmodelsexamine,which could explainwhyDoehlert pickedup morecomplexcorrelations withouttheneedforadditionaldatamanipulation

Fig 6B & E shows that a lower temperature ramp increased

1Dresolution.Inagreement withgeneralchromatographytheory,

Fig 6 Response surface plots for Box Behnken (A-C) and Doehlert (D-F) models showing responses for adjusted area% against modulation and inlet pressure (A, D), average

first-dimension resolution (R S1avg ) against temperature ramp and inlet pressure (B, E), and average second-dimension resolution (R S2avg ) against temperature ramp and inlet pressure (C, F)

it confirmedthat the firstdimension can be optimizedlike most

traditional GCsystems.The optimum rangefor 1Dseparationlay

between−0.5and−1forbothmodels,whichtranslatestoa

tem-perature ramp of 5°C/min to 1 °C/min.Within thisrange,

feasi-bilityofthetotal runtime woulddictatethechosen rampspeed

Havinga120+minruntime,forinstance,wouldnotbeacceptable

foracommercialthroughputofsamples

Therelationbetweenseparationandinletpressure(asflowrate

equivalent)showedthattheentirerangeinvestigatedcanbe

con-sideredoptimalfor1D(Fig 6B& E),whereas atendencyfortwo

localized optima was expressed in 2D (Fig 6C & F) This trend

was expressed more clearly in the Doehlert model (F) than the

Box Behnken model (C) Although the low point existed around

themediumflowrate,itwasstillconsideredpartoftheoptimum

rangebasedonthecolouring.Astheflow/pressureratiowaskept

constantbetweencolumns,anincreasein1Disdirectlycorrelated

to an increase in 2D While a higher inlet pressure compresses

peaks and improves their peak shape, which in turn favours

in-creased resolution, it can also lead to more frequent co-elutions

andadeclineinp

Both models showed the sameaverage variance of prediction

at 0.4, but Doehlert expressed more correlations whereas Box

Behnken calculated higher model efficiencies Model analysis

re-quires resultsdatatobe distributednormally, whichwasnotthe

caseforthisstudyandmadetheresults statisticallyinsignificant

Critical values for both models were outside of the set

parame-ters, whichcan eithermean that the trueoptimum isoutside of

thesetparametersorthattheentirerangewithintheparameters

is a local optimum,since modeling requires an obvious“fail”

re-sult to be able to calculate optima In this case, the latter took

place as parameters were chosen close to the theoretical

recom-mendations andwithsystematicrestrictions.Atheoreticalsuccess

forseparationisachievediftheseparationnumber(SN)isgreater

than1,whichwasmetbyallmodelpoints.Statisticalinsignificance

ofbothmodelsshowsthatchromatographicexperiencecannegate thenecessityformodelingwhenconcentratingonaverage resolu-tionandefficientuseofchromatographic areaasresults.Alotof variables relatedto flow, oven andinjector settings alreadyhave

anoptimumrangerecommendedbasedoncolumnchoiceandare readilyavailableonline.Theseincludesplitratioormaximumoven temperature or may simply be dependent on target compounds whichimpact filamentdelayorstartingoven temperaturefor in-stance

3.4 Method development & ASTM standards

Afteroptimization,thefinal methodverificationresultedinan averageSN>1inbothdimensionswith18.16forfirstand1.46for seconddimensionwithoutwraparoundforcompoundswithLI2of

atleast 450.The goalof thismethod developmentwas to allow for ILR classification based on ASTM E1618, i.e to provide suffi-cientseparationofallrelevantEIPsandtargetcompounds[1]from eachother aswell asseparation fromcommoninterferences The formerwassuccessfullyachievedduringverification, whereas the latterwasachievedduringvalidation

Fig 7 shows the relationship between the method validation andASTME1618classification requirements,wherelight,medium andheavyrefertobp-basedsubclasses;andalkanes,cycloalkanes, aromaticsandcondensedringaromaticsrefertotheoverallgroup compositionsofeachclass [1,7].Identifiedtargetcompounds cov-eringawiderangeofbpandpolarityarehighlightedbynumbers, exceptfornumerousn-alkanesandalkyl-cyclohexanes, aswellas trans-decalin,whichwereomittedforimageclarity

Achievingclassification in ASTME1618 [1]isbased onthe vi-sual comparison of a reference ignitable liquid to the total ion chromatogram(TIC),extractedionprofiles(EIP)foralkane,alkene, alcohol,aromatic,cycloalkane,ester, ketone, andpolynuclear aro-matic compoundtypes,and/or a target compound chromatogram

Fig 7 TIC of two wildfire samples used to validate the developed method highlighting ASTM E1618 ILR classification scheme groups within matrix Target compounds iden-

tified include n-alkanes (along solid line), toluene (1), Three Musketeers including p-xylene (2), Castle Group including o- & m-ethyltoluene (3) and 1,3,5-trimethylbenzene (4), indane (5), Gang of Four (1,3-&1,4-diethylbenzene, 3-&4-propyltoluene, n-butylbenzene, 1-ethyl-3,5-dimethylbenzene, 6), tetramethylbenzenes (Tetris, 7), methylindanes (8), naphthalene (9), methylnaphthalenes (Twin Towers, 10), and Five Fingers including ethylnaphthalenes (11), and dimethylnaphthalenes (12)

(TCC) of the sample[1].All EIP groups andrelevant target

com-poundsareclearlyseparatedinFig.7,signifyingsuccessfulmethod

validation

Additional considerations specific to ILR method development

pertain to missing compounds, suitability of method evaluation,

andextraneous components.Missingcompounds are notunusual

in ILR analysis, asthe exposure to heat can resultin lossof

tar-get compounds on thelighter end, and samplepreparation

tech-niques mayexhibit preferentialrecovery ranges[1]orother

func-tions such as competitive absorption ASTM E1618 only refers to

their test mixture composed ofa selectfew compounds to

eval-uate method suitability While this may suffice for smaller

ad-justmentson anestablished method,theresultspresentedherein

clearlyshowthatthisapproachisnotsufficientformethod

devel-opment asinterferences are not considered Interferences via

ex-traneouscomponentscanconsistofoxygenatedcompounds, paraf-finic,cycloparaffinic,aromatic,orcondensedring aromatic hydro-carbons[1].As showninFig.7,their abundancecan varygreatly betweencompoundgroupsdependingonthematrixcomposition Whilethey cannot beexcluded fromtherespectiveEIPs, their2D retention allows for easier distinction between target compound andextraneous interference,asis shownby the separationof n-alkanes,branched alkanes andalkenes addressingthe exampleof polyolefinorasphaltdecompositioninASTME1618[1]

WhilealltargetcompoundslistedinASTME1618[1]were sat-isfactorily separated in the verification samples, some potential forextraneousinterferencespersistedinthemethodvalidationon wildfiresamples.Fig.8 displaysthree prominentinstances where compounds from an actual wildfire sample matrixcould still in-terferewiththemethod.Withtheseexamplesofincomplete