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

Journal of Chromatography A 1656 (2021) 462495

Contents lists available at ScienceDirect

Journal of Chromatography A
journal homepage: www.elsevier.com/locate/chroma

Method development for optimizing analysis of ignitable liquid
residues using flow-modulated comprehensive two-dimensional gas
chromatography
Nadin Boegelsack a,b,∗, Kevin Hayes a,c, Court Sandau a,d, Jonathan M. Withey e,
Dena W. McMartin b, Gwen O’Sullivan a
a

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

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
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.. The three main
areas to consider for method development are column selection, modulator settings and parameter optimization. Seven column combinations with different stationary phase chemistry, column dimensions and
orthogonality were assessed for suitability based on target compound selectivity, retention, resolution,
and peak shapes, as well as overall peak capacity and area use. Using Box-Behnken design of experimentation (DoE), the effect of modulator settings such as flow ratio and loop fill capacity were evaluated using carbon loading potential, dilution effect, as well as target peak amplitude and skewing effect. The run
parameters explored for parameter optimization were oven programming, inlet pressure (column flow
rate), and modulation period. Comparing DoE approaches, Box-Behnken and Doehlert designs assessed
sensitivity, selectivity, peak capacity, and wraparound; alongside target peak retention, resolution, and
shape evaluation. Certified reference standards and simulated wildfire debris were used for method development and verification, and wildfire debris case samples scrutinized for method validation. The final
method employed a low polarity column (5% diphenyl) coupled to a semi-polar column (50% diphenyl)
and resulted in an average Separation Number (SN) exceeding 1 in both dimensions after optimization.
Separation Numbers of 18.16 for first and 1.46 for second dimension without wraparound for compounds
with at least four aromatic rings signified successful separation of all target compounds from varied matrix compositions and allowed for easy visual comparison of extracted ion profiles. Mass spectrometry
(MS) was required during validation to differentiate ions where no baseline separation between target
compounds and extraneous matrix compounds was possible. The resulting method was evaluated against
ASTM E1618 and found to be an ideal routine analysis method providing great resolution of target compounds from interferences and excellent potential for ILR classification within a complex sample matrix.
© 2021 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license

( />
1. Introduction
Ignitable liquid residue (ILR) analysis provides crucial evidence
for arson cases by determining the presence, type, and source of

∗

Corresponding author.
E-mail address: (N. Boegelsack).

an ignitable liquid, which denotes the remnants of substances or
mixtures of substances used to aid the initial development or escalation of a fire. ASTM issues the most widely used standard methods for ILRs, covering various methods of extraction and analysis
by GC–MS. ASTM E1618 bases ILR classification on their chemical
composition, including presence of analytes of interest (or groups
of analytes), equivalent n-alkane carbon range, and boiling point
(bp) ranges [1,2].

/>0021-9673/© 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( />

N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

Despite well-established ILR class profiles and standardized
methods of analysis, arson cases have amongst the lowest conviction rates in North America. The remnant nature of ILRs, combined with the presence of combustion, pyrolysis, and matrix compounds in large concentrations, amplifies the challenge of identifying and characterizing ILRs, leading to frequent false negatives [3].
ILR pattern recognition is even more challenging in wildfire samples where the physico-chemical composition of prominent matrix
compounds is very similar to that of the ILR target compounds,
leading to an increase in homolog peaks (compounds sharing the
same space in the chromatogram) [2]. In addition, accelerant application in wildfire events can be more dispersive, without pooling or other areas of high ILR concentrations to occur as is possible in structural fire scenarios, resulting in significantly greater
abundance of matrix interference. Homolog peaks can frequently

be separated by select ion analysis in MS, which is one of the reasons why MS is the preferred detection system over flame ionization detector for ILR analysis. However, if several compounds occupy the same space, and one or more of these compounds share
the same major ions, the ILR signal can no longer be distinguished
from matrix interferences.
Comprehensive
multidimensional
gas
chromatography
(GC × GC) has become a popular tool for high resolution separation of complex organic mixtures by reducing the occurrence of
overlapping homolog peaks and employing different mechanisms
of separation in a single analysis. Recent studies have highlighted
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
dimension and efficient modulation of all volatilities [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 × GC functionality and data analysis, as well as the assumption
that method development will be costly and require training as
it is considered far more complicated than in single dimension
systems.
Method development and optimization in GC × GC systems are
driven largely by the desire to fully exploit separation potential. A
number of studies have described GC × GC method optimization
[5,6,10,13,14], including method comparisons [9,11,13] as well as
the application of GC × GC to industrial fires [6]. However, much
of the literature pertains to TM systems, explores theoretical concepts, such as numerical modeling [15], and parameter optimization [16] or do not include potential matrix effects during method
development [5,9,17]. Although the 2 D separation potential is a
major advantage of GC × GC, existing method development studies

rarely provide a detailed discussion on column selection, which is
often performed on a trial and error basis until a suitable combination is found [5,6,8,9,11,13], or optimize the actual spatial distribution in the chromatogram. Instead, these studies favor theoretical
peak capacity, which assumes an evenly randomized distribution
of peaks across the available space. Lack of spatial optimization
can be partially attributed to the choice of column combination,
non-optimized column condition or non-optimized modulator settings. Relevant hardware choices, such as column combination or
flow modulator settings, are rarely discussed in detail and still primarily deduced through trial and error.
Complex method development with a high number of variables or highly interactive variables, as is the case for chromatographic analyses, benefits from design of experiments (DoE). The
two main advantages of DoE are reduction in number of experimental runs necessary and simultaneous investigation of interaction effects. Evaluation of responses is often coupled to response
surface methodology (RSM), which plots predicted responses in a

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 threedimensional model (top) and two-dimensional model as viewed from above (bottom).

multidimensional matrix and allows determination of optimum regions for variable settings at a glance.
A wide variety of experimental designs exist and have been applied to GC and GC × GC method development, including full factorial, central composite, Box-Behnken, and Doehlert uniform shell
design [14,18,19]. Each design has its own advantages and disadvantages. In this study, Box-Behnken (sampling points visualized
in Fig. 1A) and Doehlert (sampling points visualized in Fig. 1B as
3D model (top) and 2D model as viewed from above (bottom)) designs were used for analysis, with a comparative evaluation of both
models for the final parameter optimization. With the exception
of model comparisons [18], Box-Behnken has been applied to sample preparation steps most frequently [14,20] whereas Doehlert has
been predominantly applied to GC separation parameters [20,21].
Both designs are compatible with RSM and have a similar model
efficiency and accuracy for three variables [18].
In this paper, we introduce a systematic workflow for FM
GC × GC method development utilizing the benefits of DoE, and
describing in detail the most important GC × GC set-up decisions
(column choice & modulator settings) and parameter optimization
based on ILR analysis. In addition to reference standard materials,

simulated and actual wildfire samples were investigated to evaluate the proposed ILR analysis method in relation to ASTM E1618
[1] under matrix interference effects.
2. Materials & methods
2.1. Standards and reagents
Benzene (99.9%), carbon disulfide (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 material (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, d502


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

tetracosane, d8-toluene, d8-naphthalene, d10-phenanthrene, d12chrysene and d12-benz(a)pyrene) was acquired from Cambridge
Isotope Laboratories, Inc. (Tewksbury, MA, USA) and d14–1,2,4,5tetramethylbenzene (Cat#: d-0269) was purchased from CDN
isotopes (Pointe-Claire, QC, Canada).
A recovery standard was created by combining d8-naphthalene,
d10-ethylbenzene, d12–1,3,5-trimethylbenzene and d14–1,2,4,5tetramethylbenzene in methanol at a concentration of 200 ng/ml
each. A 1 ppm aromatic standard mixture containing alkanes
(49,451-U), benzene, PAHs (861,291), and toluene in DCM was prepared for the determination of RI values in comparison to a 1 ppm
standard of the deuterated Kovats-Lee retention index mix.

Accelerants were purchased in Calgary, AB, Canada and included
a composite sample of gasoline and diesel. The composite gasoline
sample was created by combining 21 gasolines of various octane
rating 87 (n = 6), 89 (n = 7), 91 (n = 6), 94 (n = 2) collected
from seven stations. The diesel composite was created by combining four diesel samples collected from four stations.
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 ± 1.5 g) and spiked with 250 μl recovery standard. Simulated ILR samples were created by additionally spiking 50 μl
each of the gasoline and diesel composite samples. All cans were
extracted in accordance with ASTM guidelines [22] at 90 °C for
16 hours.

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.

2.3. Method development

2.2. Analytical system overview

Method development in GC × GC is challenging but can be
achieved by selecting appropriate columns and instrument set-up,
and optimizing modulator and parameter settings. Main elements
to consider before method development include composition of
analytes of interest and prominent matrix compounds as well as
employing targeted versus non-targeted analysis. These inform the
practical requirements of method development and optimization,
such as stationary phase chemistry, column dimensions, modulator setup, and optimization of oven and detector parameters (e.g.
flow, temperature program etc.). A schematic overview is presented

in Fig. 3. The overall goal for method development is to maximize
selectivity and sensitivity according to the application, which for
ILR analysis translates to the separation of all target compounds
[1] from other target compounds as well as common interferences
to allow for accurate detection and quantification, pattern recognition to differentiate ILR classes [1], detector split for optimum
MS sensitivity, and appropriate run time for routine analysis (< 90
mins). Within the following sections we outline the steps taken to
develop an appropriate GC × GC method for ILR analysis using a
flow modulator.
Column selection and changes to the modulator hardware
were considered first as these relate to instrumental setup of the
GC × GC and concern the most important and difficult choices.
Modulator and parameter optimization were completed following
instrument setup.

All analyses were performed on an Agilent 7890A GC (Palo Alto,
CA), retrofitted with an Insight flow modulator (Sepsolve, Peterborough, UK) and coupled to a Markes BenchTOF-Select mass spectrometer (Llantrisant, UK). Throughout the study, the injector was
operated at 250 °C in split mode with a 25:1 ratio and 1 μl of
sample was injected via Agilent G4567A (Palo Alto, CA) autosampler. Helium was used as carrier gas with an average linear velocity
of 4.0 cm s − 1 . The MS transfer line and ion source were held at
250 °C. The electron energy applied was 70 eV and the 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 Microsoft Excel (Microsoft,
Redmond, WA). A deconvolution algorithm was used for integration, with a minimum ion count 300, minimum absolute area and
height of 10 0 0, and peak merging at 10% overlap.

2.3. GC × GC modulator
The modulator is the “heart” of a multidimensional system, its
primary function is to trap, refocus, and inject sequentially the effluent from the primary column onto the secondary column. Flow

modulators are valve-based and use differential flows to ‘fill’ and
‘flush’ a sample loop (Fig. 2) [23]. Hardware components include
first and second dimension columns (1 D & 2 D), a sample loop,
controlled supply of helium from auxiliary lines (regulated by the
Pneumatic Control Module (PCM)), and capillary bleed and transfer
lines. During the loading step, the effluent from the first column
fills the sample loop. When the modulation valve switches the
auxiliary flow, this reinjects the content of the sample loop into
the second dimension. The time taken to complete one modulation
or ‘cut’ of the first dimension is called the modulation period (PM ).
Typically, effluent from the first dimension are cut by the modulator into 2–5 modulation slices with separation on the seconddimension column occurring very fast, normally within 38 s.

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
driven by the properties of the sample, including target and matrix
compounds, and the objective of the analysis. An effective pairing
should have appropriate retention, resolution, selectivity, and peak
shape. The main choices to consider in achieving this are stationary
phase chemistry, column dimensions and order, as well as orthogonality and film thickness.
3


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

Modulator Settings

Column Selection

phase
Analytical Question,
Target Compounds,
Sample Matrix

transferline
detector 1

length

Parameter Optimization
oven
settings

column flow
Analytical Answer,
Separation of Targets,
Resolution Interference

detector
settings

i.d.

film thickness

transferline
detector 2

loop

column
order

bleed line

injector
settings

PM
auxiliary
pressure

Elements to consider:
Selectivity, Retention,
Resolution, Peak Capacity /
Area Usage, Peak Shape

Elements to consider:
Sensitivity, Peak Skewing /
Amplitude, Carbon Loading,
Dilution Effect

Elements to consider:
Sensitivity, Selectivity,
Retention, Resolution, Peak
Capacity / Area Usage, Peak
Shape, Wraparound

Achieved by:

Stationary Phase Chemistry,
Column Dimensions,
Orthogonality

Achieved by:
Flow Ratio, Loop Fill,
Detector Efficiency

Achieved by:
Oven Programming, Inlet
Pressure (Column Flow),
Modulation Period

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, stationary phase, and stationary phase orthogonality.
Combination

1

D (length m x
i.d. mm x f.t. μm)

2

1

BPX5
(25 m x 0.15 mm

x 0.25 μm)
Mega-5MS
(30 m x 0.25 mm
x 0.5 μm)
Mega Wax MS
(30 m x 0.25 mm
x 0.25 μm)
Mega Wax HT
(30 m x 0.25 mm
x 0.15 μm)
BPX5
(30 m x 0.25 mm
x 0.25 μm)
BPX5
(25 m x 0.18 mm
x 0.18 μm)
BPX5
(25 m x 0.18 mm
x 0.18 μm)

BPX50
(5 m x 0.25 mm
x 0.15 μm)
BPX50
(5 m x 0.25 mm
x 0.15 μm)
BPX50
(5 m x 0.25 mm
x 0.15 μm)
BPX50

(5 m x 0.25 mm
x 0.15 μm)
BPX50
(5 m x 0.25 mm
x 0.15 μm)
Mega Wax HT
(5 m x 0.25 mm
x 0.15 μm)
Mega-5MS
(5 m x 0.25 mm
x 0.5 μm)

2

3

4

5

6

7

D (length m x
i.d. mm x f.t. μm)

Based on expected analytes and maximum column temperatures, several column combinations were tested (Table 1). The investigation was divided into two steps, where combinations 1–5
(Table 1) were compared with the same 2 D column to choose the
most appropriate 1 D column. The chosen column was then coupled with additional 2 D columns (combinations 6 & 7) to explore

which column coupling maximized the potential performance. The
majority of applications start with a non-polar 1 D column (5%
diphenyl/95% dimethyl polysiloxane) connected to a more polar
2 D column (50% diphenyl/dimethylpolysiloxane) as this column
set separates analytes based on two mechanisms; boiling points
(1 D) and polarity ranges (2 D). Columns selected for comparison
(Table 1) were chosen based on common composition of wildfire
samples [2], system requirements for flow equilibration (see 2.3.2.),
and current routine columns used in GC–MS (non-polar, commonly
30 m length, 0.25 mm internal diameter (i.d.)).

Orthogonality
Non-polar × semi-polar

Non-polar × semi-polar

Polar ×
semi-polar
Polar ×
semi-polar
Non-polar × semi-polar

Non-polar ×
polar
Non-polar ×
non-polar

Column dimensions were chosen with expected acceptability
for routine analysis in mind. This included column lengths not
longer than 30 m to reduce sample analysis time, and thin film

thickness (< 0.5 μm) as most ILR compounds are not extremely
volatile. Narrow to medium i.d. (0.15 mm to 0.25 mm) were considered as larger i.d. columns require higher flow rates, which can
be difficult to balance in flow modulation. Orthogonality combinations, separation resulting from independent retention mechanisms
[24], were investigated for stationary phases. It is generally assumed that highly orthogonal column sets increase the separation
space occupied in the second dimension but instances of seemingly non-orthogonal sets performing better have been reported
[24]. Although stationary phase is traditionally the main measure
as it relates to selectivity and therefore separation mechanisms,
several definitions for orthogonality exist in the literature [25,26].
For this purpose, orthogonality was herein used as the traditional

4


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

stationary chemistry reference but the separation space was evaluated using peak capacity as suggested in Schure & Davis [26] together with the actual percentage of chromatographic area utilized.
Performance between column combinations was compared by
evaluating peak capacity between the solvent peak and eicosane
/ naphthalene, separation efficiency, and total percentage of 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 minimum requirement for method development, eicosane was
chosen as representative for the latest eluting compound in the
first dimension based on the test mixture detailed in ASTM E1618
[1]. Naphthalene was chosen as representative approximation for
the latest eluting compound in the second dimension based on the
target compounds listed in ASTM E1618 [1].
The area covered in both retention spaces was calculated in accordance with the Cordero method including wraparound peaks as

outlined in Eqs. (1) and (2) [27]:

A=l×w

p = me−m/n

Acovered
× 100
A potential

(1 D)

2.3.2. Modulator settings
FM systems are considered more complex to optimize than TM
systems, due to the principal of their operations and restrictions in
hardware [13,23]. Despite the importance of optimizing modulator performance, method development studies often focus only on
the “soft” parameters, solely including modulation period (PM ) in
their considerations for FM systems [11,23]. Instead, PM was considered part of the parameter optimization in this study and modulator settings concentrated on optimizing flow ratio and carbon
loading in light of the dilution effect.
Flow ratio was shown to have a major effect on peak shapes,
which in turn affects the ability for consistent integration [10].
Calculating the flush: fill volumetric ratio (see Fig. 2) in accordance with Harvey & Shellie [10], peak skewing and peak amplitude were evaluated in relation to 2 D separation and resolution
potential. Harvey & Shellie [10] recommended a flow ratio > 30
during method development to minimize the effect of peak skewing, whereas this paper assumes flow ratios < 40 as sub-optimal
based on observed differences during column selection (2.3.1).
Carbon loading was investigated two-fold: by calculating the
balance between PCM pressure and bleed line flow to ensure consistent loop fill, and by applying a Box Behnken model (see 2.3.3.)
to evaluate the x-line lengths and diameters (see Fig. 2) for optimizing detector split flow.
The flow calculator provided by Sepsolve (Peterborough, UK)
was used to calculate the required settings to balance PCM pressure and bleed line flow, as well as calculate flow rate results

for FID and MS x-lines for the 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. Therefore, the second x-line (Fig. 2) acted as bleed line for the 2 D column, and hold up time, which is a necessary consideration in a
setup with two detectors, could be disregarded in the model. Baseline settings assumed for the model were use of Helium as carrier
gas, 40 °C oven temperature, PCM pressure 27.5 psi, 1 D flowrate
0.5 ml/min, 2 D flowrate 16 ml/min, and dimensions as shown in
Fig. 2.
Root mean squared error of prediction (RMSEP) was calculated
in accordance with Eq. 9:

and w repre-

(2)

With Acovered describing the area pertaining to the column combination and Apotential as the total area in the chromatogram as calculated byEq. (1) respectively. In cases where runtimes were insufficient to elute target peaks up to KI1 2400, A% was multiplied with
the percentage of target KI1 covered and annotated as A%adj .
Separation efficiency was compared by tabulating separation
numbers across the KLI mix components according to the separation number (SN) as shown inEq. (3):

SN =

tR( j+1) − tR( j )
−1
Wh( j+1) + Wh( j )

(3)

With tR(j) and tR(j+1) representing retention times of two consecutive compounds of interest and their respective peak widths
at half height shown as Wh(j) and Wh(j+1) . Alkanes were used as
reference points for SN1 , whereas SN2 was based on LI2 reference

compounds. Since Wh was not readily available in the second dimension, width at base peak was used instead for SN2 .
After calculating retention indices for compounds of interest in
accordance with Boegelsack et al. [7], individual peak resolutions
for target compounds were calculated according toEq. (4):

RS = 1.177(RI2 − RI1 )(SN + 1 )

(4)

Where RI1 and RI2 are retention index numbers of two consecutive compounds of interest in a single dimension.
The peak capacity (n) was calculated based on Grushka [28] as
outlined in Eqs. (5) and (6):

N = 5.54

tR
Wh

2

(5)

RMSEP =

Where N is the plate number, tR is the retention time and Wh
is the width at half peak height of a given compound.
1

N 1/2 tn
n=1+

1n
4
ts

1

n ×2 n

1
I

I

yˆi − yi

2

(9)

i=1

Where I represents the number of compounds present in the
mixture, yˆ is the predicted average resolution and y is the experimentally recorded average resolution in each dimension. Twoway analysis of variance (ANOVA) with 95% level of confidence
was used to evaluate statistical significance. Standard model evaluations (multiple linear regressions and error estimates) were performed in JMP Trial 16.0.0 (SAS Institute Inc., Cary, NC, USA).
RSM was also performed in JMP Trial 16.0.0 using desired
flowrate outputs as optimal regions.

(6)

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

the retention time of eicosane and tS represents the solvent elution time. The output of the peak capacity represents the potential
number of compounds that can be fully separated within its spatial boundaries in the first dimension. To account for the second
dimension, Eqs. (5) & 6 were applied to naphthalene for values relating to the second dimension to yield 2 n. Finally, the total peak
capacity (n) was calculated as shown in Eq. 7:

n=

(8)

Where p represents the number of peaks appearing as singlets,
and m is the number of components in the sample, which was
averaged to 30 0 0 for a typical diesel on wood matrix sample to
simplify the comparison.

(1)

With l representing the length in minutes
senting the width in seconds (2 D)

A% =

Using the total peak capacity, the number of possible distinguished peaks p in a chromatogram with the same boundaries was
calculated as Eq. (8) in accordance with Bertsch [29], assuming
RS = 0.5:

2.3.3. Run parameter optimization
Optimization was completed using DoE due to the interactive
nature and number of variables considered. Run parameters cho-

(7)

5


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

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)

−1
0
1

2
5
8

1
2

3

0.18
0.25
0.32

0.18
0.25
0.32

sen as variables for the models were modulation period (PM ), flow
rate (as a function of inlet pressure), and oven ramp, as these are
generally considered to have the highest impact on GC × GC performance [13,30].
The column set consisted of a non-polar, 5% diphenyl, column
(30 m × 0.18 μm i.d. × 0.18 μm film thickness) in the first dimension and a semi-polar, 50% diphenyl, column (5 m × 0.25 μm
i.d × 0.18 μm film thickness) in the second dimension. Actual values as used for calculations can be seen in Table 3, whereas coded
values were standard for Box-Behnken (−1, 0, 1) and based on
the prescribed uniform shell design simplex laid out in Doehlert
[31] for three variables.

As column 1 had a narrow i.d., an equivalent column was
used with a shorter length to reduce solvent delay and a slightly
larger i.d. to preserve plate potential. A medium film thickness was
elected to raise the theoretical plate number while simultaneously
keeping the column bleed and retention time low for cleaner and
faster analysis. Comparing second dimension performance (column
combinations 1, 6 and 7), column combination 7 showed the greatest resolution potential in the second dimension, leading to an increased area usage. However, the potential number of distinguishable peaks performed the worst out of all three combinations.
With column combination 6 providing the best resolution in the
first dimension but the worst resolution in the second dimension,
column combination 1 offered the best overall result for further

method development.

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
reference compounds in the KLI mix. both models were compared
based on their RMSEP for results as given by eq. (9)
RSM was applied to results of the models to determine the optimum performance zones for the three variables under investigation, using maxima for A%adj. and the average peak resolutions (RS )
of each dimension as desired outputs. RSM and model evaluations
were performed in JMP Trial 16.0.0.

3.2. Stationary phase chemistry
Table 4 shows that combinations 1, 2 & 5 (non-polar × semipolar) have a greater average resolution in 1 D than combinations 3
& 4 (polar × semi-polar). Since polarity and selectivity of the phase
determine retention and interactivity of compound groups, this can
be attributed to the stationary phase (5% phenyl against wax). Additionally, the maximum operating temperature for wax columns
is lower than that for 5% phenyl columns. This restricts the potential resolution for high bp compounds, which is problematic when
identifying heavier ILRs in fire debris samples.
Separation in the first dimension is very important and prioritized in method development as the second dimension cannot
make up for any resolution lost in 1 D. One-dimensional GC method
development principles apply; therefore, linear velocity should be
operated as closely as possible to the optimum indicated from
the van Deemter curve. Other examples include oven temperature
rates contributing to the balance between peak shapes and coelutions, or inlet split ratios being considered in conjunction with
column i.d. to avoid overloading.
Looking closely at the second dimension, negative results for
SN2 avg (combinations 2 & 4 to 6) indicate that target compounds
eluted on the same plane in 2 D, as the difference between tR 2 was
smaller than the average peak width. Considering that the baseline

3. Results & discussion

3.1. Column selection
The comparative results of all column combinations are presented in Table 4. Comparing first-dimension columns (combinations 1 to 5), combinations 1 & 3 showed the greatest resolution
potential in both dimensions with the highest peak capacity and
potential number of distinguishable peaks. Although column 3 had
a higher area usage and peak capacity, it eluted less than 70% of
target compounds in the same time frame as the other columns
and resulted in extensive wraparound starting between mono- and
diaromatic compounds. Wraparound complicates data analysis as it
requires manual integration of data, making it unsuitable for routine analysis. Timely elution is another key factor for routine analysis; thus, column 1 was selected for 2 D comparison.

Table 3
Actual values for chosen parameters (modulation period, inlet pressure, oven ramp) used in model development.
Box-Behnken
Run
1
2
3
4
5
6
7
8
9
10
11
12
13

Doehlert

PM (s)

Inlet pressure (psi)

Oven ramp ( °C/min)

PM (s)

Inlet pressure (psi)

Oven ramp ( °C/min)

4
6
6
2
2
6
6
2
2
4
4
4
4

43.4
55.22
31.582
55.22

31.582
43.4
43.4
43.4
43.4
55.22
55.22
31.582
31.582

10
10
10
10
10
19
1
19
1
19
1
19
1

4
6
5
5
2
3

3
5
5
3
4
3
4

43.4
43.4
55.22
47.34
43.4
31.582
39.46
31.582
39.46
55.22
51.28
47.34
35.52

10
10
10
19
10
10
1
10

1
10
1
19
19

6


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495
Table 4
Overview of column combination performance including area percentage covered (A%), average separation (SN) and resolution (RS ) 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%

SN1 avg

RS 1 avg

SN2 avg

RS 2 avg

n

P

1
2
3
4
5
6
7

50%
25%
73%
20%
13%
49%
77%

12.8
13.1
10.3
9.8
11.5
14.8
12.6

7751
7897
6678
6022

7369
9274
7565

0.4
−0.3
0.5
−1.6
−0.8
−0.7
0.9

151
87
171
19
17
38
188

3998
2150
4805
406
927
7380
3353

1417
743

1607
2
118
1998
1226

width had to be used for this calculation, negative results were expected in light of the relatively short timeframe for modulation.
3.3. Column dimensions and order
Column order between 1 D and 2 D is an important factor for any
GC × GC system. Each column characteristics (phase, film thickness, i.d. and length) is independently important for optimal separation but must also be considered in relation to the pairing column. This is particularly important in FM systems, where it is crucial to have an appropriate flow ratio between the columns (see
3.2.1.). Each flow is affected by length and i.d. of its respective column. While TM-GC × GC follows the traditional rule of a longer
column equaling better separation albeit longer run time, FM imposes additional restrictions. A longer column still corresponds to a
higher theoretical plate number; but also leads to higher pressures
and complications in flow balancing. As a result, it is uncommon
to see columns exceeding 30 m in FM systems. When applied to
ILR analysis, shorter lengths up to 30 m proved to be sufficient for
target compound separations.
Balancing i.d.s, TM system method development often suggests
using the same i.d. in first and second dimension to heighten theoretical plate number and optimize flow [32], this is not the case
in FM systems as evident from Table 4. The 0.25 mm i.d. combinations (2, 4 & 5) do not show a great advantage in the RS 1 , and
compare poorly in area usage and RS 2 since the second-dimension
i.d. is the same diameter. This, in turn, negatively affects their ability to fully separate large numbers of peaks (see p, Table 4). In
addition to the regular considerations relating to injection method,
phase ratio, or retention; i.d. has a large impact on flow ratio and
related pressure restrictions. In FM systems, a smaller i.d. in the
first dimension will speed up analysis and allow for easier equilibration of flow (see 3.2.).

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).

ered in conjunction with the internal diameter, which plays a very
important role for flow-regulated GC × GC, column length and i.d.
between combinations 1, 6 & 7 were kept constant. Given the difference in length between 1 D and 2 D, and the subsequent amount
of time compounds spend in each dimension, it makes sense that
film thickness had a much larger impact on the 2 D separation. A
film thickness of 0.15 μm was selected in 2 D to reduce the potential for early wraparound during parameter optimization. Since
film thickness of the second dimension has a significant effect only
on 2 D separation, this can be increased at a later point as required
without affecting other development parameters.
Area usage in the second dimension is frequently not considered important in method development. Instead, phase orthogonality is favoured. Table 4 illustrates that peak capacity and area usage are closely linked. Using these variables as a measure of true
system orthogonality as suggested by Schure & Davis [26], was a
better approach for evaluating column performance than solely relying on phase orthogonality.
3.5. Modulator settings

3.4. Orthogonality and film thickness

3.5.1. Flow ratio
Fig. 4 depicts the effect of flow ratio on naphthalene. The peak
skewing effect is represented by the black dotted line and is clearly
enhanced in Fig. 4B (flow ratio = 30) compared to Fig. 4A (flow
ratio = 60). The effect on peak amplitude can also be observed as
the peak height in A is more concentrated, which is represented by
dark red area, and less drawn out, which is shown by a decrease
in green & blue areas.
As Harvey & Shellie [10] highlighted, sub-optimal flow equilibration leads to an increase in observed modulation effects, i.e.
a higher degree of peak skewing and lower peak amplitudes. Although they concluded that a flow ratio > 30 was appropriate,
Fig. 4B showed pronounced artifacts of modulation effects. In general, non-focusing modulation displays broader peak widths in
comparison to focusing modulation, which means a smaller theoretical resolution potential but increased robustness over long periods of time [12]. Avoidance of modulation effects is therefore an
important undertaking to ensure consistently well-shaped peaks

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 2 D
resolution, whereas combination 6 with high orthogonality outperformed all others in peak capacity, number of distinguishable
peaks and 1 D resolution. Therefore, it appears that the orthogonality of stationary phases is not the primary factor influencing
separation in the second dimension, confirming that true system
orthogonality is affected by a number of factors [25,27]. An important factor in the second dimension appears to be film thickness
based on combination 7.
In theory, film thickness directly affects retention (the thicker
the film, the longer compounds are retained). This was not evident
from average retention in 1 D (Table 4) where the difference between various thin films over the length of the first-dimension column was not apparent. As film thickness should always be consid7


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

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

Carbon loading to
FID

High
Low

↓

–

↓ loading/flowrate

Carbon loading to
MS

High
Low

loading/flowrate
–
↑ loading/flowrate

–

↑ loading/flowrate
–

Length of MS
x-line

Internal Diameter
of MS x-line

–

↑ loading/flowrate

↑ loading/flowrate
–
↓ loading/flowrate

–

↓ loading/flowrate
–

and maximum resolution, and a minimum flow ratio of 40 was required for this setup. Although higher flow ratios between columns
lead to better peak shapes, they must be balanced by choosing appropriate modulator dimensions to ensure consistency of flow ratio
and carbon loading of the detector [13].
3.5.2. Carbon loading & dilution affect
In FM systems, optimization of carbon loading and detector efficiency is achieved by balancing flows across the 7- and 3-port
valves (see Fig. 2).
Based on the flow calculator, a 2-m length was sufficient to balance flows across the 7-port valve for the column setup chosen in
3.1. with a 0.1-mm i.d. bleed line to ensure consistent carbon loading in 2 D. Established standard 1 D column flows, which need to be
balanced with bleed line flow for consistent loop fill (see Fig. 2),
are < 1 ml/min for capillary columns. When calculating bleed line
flow, the choice of bleed line i.d. was prioritized as it is directly

related to flow rate, whereas length is inversely related. Therefore,
choosing a 0.1-mm diameter bleed line i.d. prevented the need
for extreme length, which in turn minimized potential carrier gas
waste. An imbalance in flow ratio equilibrium in the 7-port valve
can lead to overshooting or undershooting the loop, which in turn
negatively affects carbon loading and causes loss of 2 D resolution
via peak skewing (depicted in Fig. 4).
Table 5 summarizes the variables to consider 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, specifically the relationship between MS x-line length and FID x-line i.d.
to FID and MS flowrates. The FID flowrate is represented by the
yellow-red mesh, whereas the MS flowrate is displayed as light
purple-green mesh with the optimum region where they intersect
(11 ml/min and 4 ml/min respectively). The relationship shown
in Fig. 5 correlates to the bleed line optimization. FID flow is directly related to FID diameter as shown on the y-axis, resulting in
the highest flow (red area) for the largest i.d., whereas MS length
is indirectly related to MS flow, resulting in the lowest MS flow
(gray area) for the longest x-line. These principles work vice versa,
meaning that a longer FID x-line results in decreased flow to the
FID and increased flow to the MS. Final dimensions used to achieve
the optimum ratio are listed in Fig. 2.
The model showed significance for all individual parameters,
as well as the interaction between i.d.s, and interactions between
each x-line length and respectively opposing x-line i.d. The RMSEP of the model was 1.7 with r2 = 0.98 and homogenous error
distribution. Although resulting calculated numbers were not exact when compared to the actual value output of the system, the
evaluation was performed relative to the baseline settings and the
final dilution factor was adjusted accordingly to make the use of a
flow calculator suitable.
The desired detector split flow, also referred to as dilution effect, regulates how much of the total 2 D flow is sent to the detector. A 1:4 ratio is considered ideal for MS detector efficiency

to preserve maximum sensitivity without overloading the filament

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

and introducing detector noise, which translated to the set optima
after taking into account the difference between calculated and actual values.
3.5.3. Parameter optimization
Fig. 6 shows all RSM graphs for “soft” parameter optimization,
comparing Box Behnken (top) and Doehlert uniform shell (bottom)
design. The RSM graphs show that area usage is optimized with
shorter modulation times (Fig. 6A & D). In FM systems, longer PM
risk introducing pressure inconsistencies, which lead to poor repeatability. Relating A% and PM , favouring shorter modulation was
expected as this greatly reduces the potential area to fill and increases peak stability. However, a considerable downside is an increase in wraparound, which was not accounted for in this analysis
but was evaluated manually outside of models. The Box Behnken
model (Fig. 6A) only shows one optimum (which heavily featured
wraparound from LI2 > 150), whereas Doehlert (Fig. 6D) displays a
second optimum region which did not include wraparound for LI2
≤ 450. A similar effect can be seen in relation to the inlet pressure
and area usage (A/D), as well as the second-dimension resolution
and inlet pressure (C/F). Doehlert matrices investigate values besides the model extrema that Box-Behnken models examine, which
could explain why Doehlert picked up more complex correlations
without the need for additional data manipulation.
Fig. 6B & E shows that a lower temperature ramp increased
1 D resolution. In agreement with general chromatography theory,
8


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

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 (RS 1 avg. ) against temperature ramp and inlet pressure (B, E), and average second-dimension resolution (RS 2 avg. ) against temperature ramp and
inlet pressure (C, F).

it confirmed that the first dimension can be optimized like most
traditional GC systems. The optimum range for 1 D separation lay
between −0.5 and −1 for both models, which translates to a temperature ramp of 5 °C/min to 1 °C/min. Within this range, feasibility of the total run time would dictate the chosen ramp speed.
Having a 120+ min runtime, for instance, would not be acceptable
for a commercial throughput of samples.
The relation between separation and inlet pressure (as flow rate
equivalent) showed that the entire range investigated can be considered optimal for 1 D (Fig. 6B & E), whereas a tendency for two
localized optima was expressed in 2 D (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
the medium flow rate, it was still considered part of the optimum
range based on the colouring. As the flow / pressure ratio was kept
constant between columns, an increase in 1 D is directly correlated
to an increase in 2 D. While a higher inlet pressure compresses
peaks and improves their peak shape, which in turn favours increased resolution, it can also lead to more frequent co-elutions
and a decline in p.
Both models showed the same average variance of prediction
at 0.4, but Doehlert expressed more correlations whereas Box
Behnken calculated higher model efficiencies. Model analysis requires results data to be distributed normally, which was not the
case for this study and made the results statistically insignificant.
Critical values for both models were outside of the set parameters, which can either mean that the true optimum is outside of
the set parameters or that the entire range within the parameters
is a local optimum, since modeling requires an obvious “fail” result to be able to calculate optima. In this case, the latter took

place as parameters were chosen close to the theoretical recommendations and with systematic restrictions. A theoretical success
for separation is achieved if the separation number (SN) is greater
than 1, which was met by all model points. Statistical insignificance

of both models shows that chromatographic experience can negate
the necessity for modeling when concentrating on average resolution and efficient use of chromatographic area as results. A lot of
variables related to flow, oven and injector settings already have
an optimum range recommended based on column choice and are
readily available online. These include split ratio or maximum oven
temperature or may simply be dependent on target compounds
which impact filament delay or starting oven temperature for instance.
3.4. Method development & ASTM standards
After optimization, the final method verification resulted in an
average SN > 1 in both dimensions with 18.16 for first and 1.46 for
second dimension without wraparound for compounds with LI2 of
at least 450. The goal of this method development was to allow
for ILR classification based on ASTM E1618, i.e. to provide sufficient separation of all relevant EIPs and target compounds [1] from
each other as well as separation from common interferences. The
former was successfully achieved during verification, whereas the
latter was achieved during validation.
Fig. 7 shows the relationship between the method validation
and ASTM E1618 classification requirements, where light, medium
and heavy refer to bp-based subclasses; and alkanes, cycloalkanes,
aromatics and condensed ring aromatics refer to the overall group
compositions of each class [1,7]. Identified target compounds covering a wide range of bp and polarity are highlighted by numbers,
except for numerous n-alkanes and alkyl-cyclohexanes, as well as
trans-decalin, which were omitted for image clarity.
Achieving classification in ASTM E1618 [1] is based on the visual comparison of a reference ignitable liquid to the total ion
chromatogram (TIC), extracted ion profiles (EIP) for alkane, alkene,
alcohol, aromatic, cycloalkane, ester, ketone, and polynuclear aromatic compound types, and/or a target compound chromatogram

9


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

Fig. 7. TIC of two wildfire samples used to validate the developed method highlighting ASTM E1618 ILR classification scheme groups within matrix. Target compounds identified 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 and relevant target compounds are clearly separated in Fig. 7, signifying successful method
validation.
Additional considerations specific to ILR method development
pertain to missing compounds, suitability of method evaluation,
and extraneous components. Missing compounds are not unusual
in ILR analysis, as the exposure to heat can result in loss of target compounds on the lighter end, and sample preparation techniques may exhibit preferential recovery ranges [1] or other functions such as competitive absorption. ASTM E1618 only refers to
their test mixture composed of a select few compounds to evaluate method suitability. While this may suffice for smaller adjustments on an established method, the results presented herein
clearly show that this approach is not sufficient for method development as interferences are not considered. Interferences via ex-

traneous components can consist of oxygenated compounds, paraffinic, cycloparaffinic, aromatic, or condensed ring aromatic hydrocarbons [1]. As shown in Fig. 7, their abundance can vary greatly
between compound groups depending on the matrix composition.
While they cannot be excluded from the respective EIPs, their 2 D
retention allows for easier distinction between target compound
and extraneous interference, as is shown by the separation of nalkanes, branched alkanes and alkenes addressing the example of
polyolefin or asphalt decomposition in ASTM E1618 [1].
While all target compounds listed in ASTM E1618 [1] were satisfactorily separated in the verification samples, some potential
for extraneous interferences persisted in the method validation on
wildfire samples. Fig. 8 displays three prominent instances where
compounds from an actual wildfire sample matrix could still interfere with the method. With these examples of incomplete reso-

10


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

Summarizing general FM method development recommendations, the following bullet points can be used for guidance:
•

•
•

Fig. 8. Examples of wildfire matrix interferences for ILR target compounds (italicized) with A) pentyloxirane (1), β -pinene (2) and benzaldehyde (3) eluting
near 1,2,4-trimethylbenzene (2) and 3-isopropyltoluene (5), B) limonene (8) eluting with 1,2,3-trimethylbenzene (6) and 4-isopropyltoluene (7), and C) 1,2,4,5tetramethylbenzene (9) and 1,2,3,5-tetramethylbenzene (10) (Tetris) eluting with
carveol (11).

•

•

lution and potential for co-elutions in larger concentrations, it becomes apparent that purely using standards or neat ignitable liquids may not be suitable for method development. The development of a routine analysis when expecting complex background
matrices at potentially much larger concentration than target analytes can benefit greatly from taking these into account during the
optimization process.
Although none of the compounds in Fig. 8 displayed baseline separations, a deconvolution algorithm sufficiently separated
the interfering compounds as they had distinguishable ion profiles from the target compounds, resulting in a successful method
validation. Additional systematic requirements for routine analysis method development, such as dilution factor and runtime <
90 min, were also satisfactorily met. Combining the reduction in
coeluting interfering compounds by using GC × GC with the power

of compound identification at trace level concentrations via mass
spectral libraries makes FM GC × GC–MS an ideal routine method
for ILR analysis.

•

•

A shorter 1 D column (< 30 m) with a smaller i.d. allows for
faster analysis and facilitates flow equilibration.
The 2 D column is recommended to have a smaller i.d. than 1 D.
Film thickness was found to be more important to second dimension retention than phase orthogonality, as well as easily
controlling the amount of wraparound present.
Phase orthogonality was not representative of spatial efficiency,
which could instead be evaluated by a combination of peak capacity and occupied chromatogram area.
An optimum modulator set-up can be achieved by allowing for
a flow ratio > 40 and balancing bleed line flow, carbon loading
and detector efficiency via flow calculator modeling.
Flow modeling and parameter optimization can be facilitated
by applying DoE.
The most significant parameters for run optimization are modulation period, inlet pressure (flow rate) and oven programming.

As a result, an ILR analysis on flow-modulated GC × GC–MS was
successfully optimized for visual separation of target EIP as well as
resolution of all target compounds stated in ASTM E1618 [1] from
extraneous components of wildfire matrix. The use of mass spectrometry was confirmed to be an asset in resolving matrix interferences commonly encountered in wildfire debris by applying a
deconvolution algorithm for integration of persisting homologue
compounds and the potential for group-type analysis in SIM.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to

influence the work reported in this paper.

4. Conclusion
CRediT authorship contribution statement
The suggested systematic workflow of column selection, modulator settings and parameter optimization proved to be simple and
efficient in its application. Design of experiment facilitated method
development by minimizing time and resources required for development in two of the three stages. Despite statistical insignificance in parameter optimization, this approach allowed investigation of several variables and their interactions, simultaneous optimization of first and second dimension and use of RSM for highlighting preferable parameter ranges. Applying general chromatography guidelines and recommendations after optimizing the flow
equilibrium resulted in a near-optimized method, which passed
verification requirements by sufficiently separating standards required in ASTM E1618 [1] on a simulated wildfire sample. Achieving the method verification goal after optimizing hardware components highlighted the potential benefit of including these considerations in future GC × GC-related publications, particularly for FM
systems.
During data evaluation, it became apparent that area usage, average separation efficiency and peak capacity alone as resulting
factors did not provide an accurate picture of the potential performance of the method. Additional variables were required to evaluate the tested methods, such as amount of wraparound present,
level of peak distortion in the second dimension, and separately
evaluating potentially critical co-elutions, either between target
compounds or with extraneous interfering compounds expected to
be present in the matrix. Including these variables in the development stage by analyzing simulated and actual wildfire samples
for method verification and validation respectively provided a wellrounded overview of critical parameters for method development
and their respective effects for further optimization potential.

Nadin Boegelsack: Conceptualization, Methodology, Investigation, Formal analysis, Validation, Writing – original draft, Writing
– review & editing, Visualization. Kevin Hayes: Writing – original draft, Visualization. Court Sandau: Conceptualization, Writing
– review & editing. Jonathan M. Withey: Writing – review & editing, Funding acquisition. Dena W. McMartin: Writing – review &
editing, Funding acquisition. Gwen O’Sullivan: Conceptualization,
Methodology, Investigation, Resources, Funding acquisition, Writing
– review & editing, Supervision.
References
[1] ASTM InternationalASTM E1618-14, Standard Test Method for Ignitable Liquid
Residues in Extracts from Fire Debris Samples by Gas Chromatography-Mass
Spectrometry, ASTM International, West Conshohocken, PA, 2014, doi:10.1520/
E1618-14.

[2] L.N. Kates, P.I. Richards, C.D. Sandau, The application of comprehensive twodimensional gas chromatography to the analysis of wildfire debris for ignitable
liquid residue, Forensic Sci. Int. 310 (5) (2020) 110256, doi:10.1016/j.forsciint.
2020.110256.
[3] J. Baerncopf, K. Hutches, A review of modern challenges in fire debris analysis,
Forensic Sci. Int. 244 (2014) e12–e20 11, doi:10.1016/j.forsciint.2014.08.006.
[4] N. Boegelsack, J. Withey, G. O’Sullivan, D. McMartin, A critical examination of
the relationship between wildfires and climate change with consideration of
the human impact, J Environ Prot (Irvine, Calif) 09 (05) (2018) 461–467, doi:10.
4236/jep.2018.95028.
[5] K. Nizio, J. Cochran, S. Forbes, Achieving a near-theoretical maximum in peak
capacity gain for the forensic analysis of ignitable liquids using GC×GC-TOFMS,
Separations 3 (3) (2016) 26 9, doi:10.3390/separations3030026.
[6] A.A.S. Sampat, B. Van Daelen, M. Lopatka, H. Mol, G. der Weg, G. VivóTruyols, M. Sjerps, P.J. Schoenmakers, A.C. Van Asten, Detection and characterization of ignitable liquid residues in forensic fire debris samples by comprehensive two-dimensional gas chromatography, Separations 5 (3) (2018) 43,
doi:10.3390/separations5030043.
11


N. Boegelsack, K. Hayes, C. Sandau et al.

Journal of Chromatography A 1656 (2021) 462495

[7] N. Boegelsack, C. Sandau, D.W. McMartin, J.M. Withey, G. O’Sullivan, Development of retention time indices for comprehensive multidimensional gas chromatography and application to ignitable liquid residue mapping in wildfire investigations, J. Chromatogr. A 1635 (2021) 461717, doi:10.1016/j.chroma.2020.
461717.
[8] T. Hayward, R. Gras, J. Luong, Flow-modulated targeted signal enhancement
for volatile organic compounds, J. Sep. Sci. 39 (12) (2016) 2284–2291 6, doi:10.
10 02/jssc.20160 0 054.
[9] J. Pandohee, J.G. Hughes, J.R. Pearson, O.A.H. Jones, Chemical fingerprinting of petrochemicals for arson investigations using two-dimensional gas
chromatography-flame ionisation detection and multivariate analysis, Sci. Justice 60 (4) (2020) 381–387, doi:10.1016/j.scijus.2020.04.004.
[10] P.M.A. Harvey, R.A. Shellie, Factors affecting peak shape in comprehensive
two-dimensional gas chromatography with non-focusing modulation, J. Chromatogr. A 1218 (21) (2011) 3153–3158, doi:10.1016/j.chroma.2010.08.029.

[11] C. Duhamel, P. Cardinael, V. Peulon-Agasse, R. Firor, L. Pascaud, G. SemardJousset, P. Giusti, V. Livadaris, Comparison of cryogenic and differential flow
(forward and reverse fill/flush) modulators and applications to the analysis of
heavy petroleum cuts by high-temperature comprehensive gas chromatography, J. Chromatogr. A 1387 (2015) 95–103 3, doi:10.1016/j.chroma.2015.01.095.
[12] Y. Nolvachai, L. McGregor, N.D. Spadafora, N.P. Bukowski, P.J. Marriott, Comprehensive two-dimensional gas chromatography with mass spectrometry: toward
a super-resolved separation technique, Anal. Chem. 92 (2020) 12572–12578,
doi:10.1021/acs.analchem.0c02522.
[13] J. Mommers, S. van der Wal, Column selection and optimization for comprehensive two-dimensional gas chromatography: a review, Crit. Rev. Anal. Chem.
51 (2) (2021) 183–202, doi:10.1080/10408347.2019.1707643.
[14] P.H. Stefanuto, K.A. Perrault, L.M. Dubois, B. L’Homme, C. Allen, C. Loughnane,
N. Ochiai, J.F. Focant, Advanced method optimization for volatile aroma profiling of beer using two-dimensional gas chromatography time-of-flight mass
spectrometry, J. Chromatogr.y A 1507 (2017) 45–52 7, doi:10.1016/j.chroma.
2017.05.064.
[15] P. Siriviboona, C. Tungkabureeb, N. Weerawongphroma, C. Kulsing, Direct equations to retention time calculation and fast simulation approach for simultaneous material selection and experimental design in comprehensive two dimensional gas chromatography, J. Chromatogr. A 1602 (2019) 425–431, doi:10.1016/
j.chroma.2019.05.059.
[16] X. Lu, H. Kong, H. Li, C. Ma, J. Tian, G. Xu, Resolution prediction and optimization of temperature programme in comprehensive two-dimensional gas
chromatography, J. Chromatogr. A 1086 (1–2) (2005) 175–184, doi:10.1016/j.
chroma.2005.05.105.
[17] C. Kulsing, P. Rawson, R.L. Webster, D.J. Evans, P.J. Marriott, Group-Type analysis of hydrocarbons and sulfur compounds in thermally stressed merox jet fuel
samples, Energy Fuels 9 (8978–8984) (2017) 31, doi:10.1021/acs.energyfuels.
7b01119.
[18] L.K. Skartland, S.A. Mjøs, B. Grung, Experimental designs for modeling retention patterns and separation efficiency in analysis of fatty acid methyl esters
by gas chromatography–mass spectrometry, J. Chromatogr. A 1218 (38) (2011)
6823–6831, doi:10.1016/j.chroma.2011.07.077.

[19] C. Kulsing, Y. Nolvachaib, P.J. Marriott, Concepts, selectivity options and experimental design approaches in multidimensional and comprehensive twodimensional gas chromatography, Trends Anal. Chem. 130 (2020) 115995,
doi:10.1016/j.trac.2020.115995.
[20] S.L.C. Ferreira, R.E. Bruns, E.G.P. da Silva, W.N.L. dos Santos, C.M. Quintella,
J.M. David, J.B. de Andrade, M.C. Breitkreitz, I.C.S.F. Jardim, B.B. Neto, Statistical designs and response surface techniques for the optimization of chromatographic systems, J. Chromatogr. A 1158 (1–2) (2007) 2–14 7, doi:10.1016/
j.chroma.2007.03.051.
[21] P. Araujo, S. Janagap, Doehlert uniform shell designs and chromatography, J.

Chromatogr. B 910 (2012) 14–21, doi:10.1016/j.jchromb.2012.05.019.
[22] ASTM InternationalASTM E1412-16, Standard Practice For Separation of Ignitable Liquid Residues from Fire Debris Samples by Passive Headspace Concentration With Activated Charcoal, ASTM International, West Conshohocken,
PA, 2016, doi:10.1520/E1412-16.
[23] A. Mostafa, M. Edwards, T. Górecki, Optimization aspects of comprehensive
two-dimensional gas chromatography, J. Chromatogr. A 1255 (2012) 38–55,
doi:10.1016/j.chroma.2012.02.064.
[24] B. Omais, M. Courtiade, N. Charon, J. Ponthus, D. Thiébaut, Considerations on
orthogonality duality in comprehensive two-dimensional gas chromatography,
Anal. Chem. 83 (19) (2011) 7550–7554, doi:10.1021/ac201103e.
[25] D. Ryan, P. Morrison, P. Marriott, Orthogonality considerations in comprehensive two-dimensional gas chromatography, J. Chromatogr. A 1071 (1–2) (2005)
47–53 4, doi:10.1016/j.chroma.2004.09.020.
[26] M.R. Schure, J.M. Davis, Orthogonal separations: comparison of orthogonality
metrics by statistical analysis, J. Chromatogr. A 1414 (10) (2015) 60–76, doi:10.
1016/j.chroma.2015.08.029.
[27] C. Cordero, P. Rubiolo, B. Sgorbini, M. Galli, C. Bicchi, Comprehensive twodimensional gas chromatography in the analysis of volatile samples of natural
origin: a multidisciplinary approach to evaluate the influence of second dimension column coated with mixed stationary phases on system orthogonality, J. Chromatogr. A 1132 (1–2) (2006) 268–279, doi:10.1016/j.chroma.2006.
07.067.
[28] E. Grushka, Chromatographic peak capacity and the factors influencing it, Anal.
Chem. 42 (11) (1970) 1142–1147 9, doi:10.1021/ac60293a001.
[29] W. Bertsch, Two-dimensional gas chromatography. concepts, instrumentation, and applications – part 1: fundamentals, conventional two-dimensional
gas chromatography, selected applications, J. High Resolut. Chromatogr. 22
(12) (1999) 647–665 12. doi: 10.1002/(SICI)1521-4168(19991201)22:12 647::
AID- JHRC647 3.0.CO;2-V.
[30] M. Edwards, H. Boswell, T. Górecki, Comprehensive multidimensional
chromatography, Curr Chromatogr 2 (2) (2015) 80–109 7, doi:10.2174/
2213240602666150722232236.
[31] D.H. Doehlert, Uniform shell designs, J. R. Stat. Soc. Ser. C App. Stat. 19 (3)
(1970) 231–239, doi:10.2307/2346327.
[32] M.S. Klee, J. Cochran, M. Merrick, L.M. Blumberg, Evaluation of conditions
of comprehensive two-dimensional gas chromatography that yield a neartheoretical maximum in peak capacity gain, J. Chromatogr. A 1383 (2015) 151–

159, doi:10.1016/j.chroma.2015.01.031.

12