Tpa

Pa

b

a

ARRAA

KGEWCRP

1

npfmwsaotie

T

h0

Journal of Chromatography A, 1364 (2014) 223–233

Contents lists available at ScienceDirect

Journal of Chromatography A

j o ur na l ho me page: www.elsev ier .com/ locate /chroma

wo-dimensional gas chromatography/mass spectrometry, physicalroperty modeling and automated production of component maps tossess the weathering of pollutants

atrick M. Antlea, Christian D. Zeiglera, Dimitri G. Livitzb, Albert Robbata,∗

Tufts University Department of Chemistry, United StatesUniversity of Massachusetts, Amherst, United States

r t i c l e i n f o

rticle history:eceived 3 June 2014eceived in revised form 7 August 2014ccepted 10 August 2014vailable online 16 August 2014

eywords:C × GC/MSnvironmental forensicseathering

omponent mapsetention indexAH

a b s t r a c t

Local conditions influence how pollutants will weather in subsurface environments and sediment, andmany of the processes that comprise environmental weathering are dependent upon these substances’physical and chemical properties. For example, the effects of dissolution, evaporation, and organic phasepartitioning can be related to the aqueous solubility (SW), vapor pressure (VP), and octanol–water par-tition coefficient (KOW), respectively. This study outlines a novel approach for estimating these physicalproperties from comprehensive two-dimensional gas chromatography–mass spectrometry (GC × GC/MS)retention index-based polyparameter linear free energy relationships (LFERs). Key to robust correlationbetween GC measurements and physical properties is the accurate and precise generation of retentionindices. Our model, which employs isovolatility curves to calculate retention indices, provides improvedretention measurement accuracy for families of homologous compounds and leads to better estimatesof their physical properties. Results indicate that the physical property estimates produced from thisapproach have the same error on a logarithmic-linear scale as previous researchers’ log–log estimates,

yielding a markedly improved model. The model was embedded into a new software program, allowingfor automated determination of these properties from a single GC × GC analysis with minimal modeltraining and parameter input. This process produces component maps that can be used to discern themechanism and progression of how a particular site weathers due to dissolution, organic phase parti-tioning, and evaporation into the surrounding environment.

© 2014 Elsevier B.V. All rights reserved.

. Introduction

Fossil fuel pollution is the result of a number of factors, includingaturally occurring and unintended seepages; collection, trans-ort and storage activities; and incomplete combustion. Whenossil fuels and their by-products are released into the environ-

ent, they are subject to a number of weathering factors. Theseeathering factors include physical (evaporation, adsorption, dis-

olution, and emulsification), biological (microbial degradation),nd chemical (photo- and oxidative degradation) processes, allf which can significantly change the chemical composition of

he substance over time. Understanding how local environmentsmpact weathering is critical to determining whether the localcosystem is capable of remediation, i.e., the natural attenuation

∗ Corresponding author at: 62 Talbot Avenue, Medford, MA 02155, United States.el.: +1 6176273474.

E-mail address: Albert.Robbat@Tufts.edu (A. Robbat).

ttp://dx.doi.org/10.1016/j.chroma.2014.08.033021-9673/© 2014 Elsevier B.V. All rights reserved.

of pollution effects. Because site-specific weathering processes candramatically change the chemical composition of fossil fuel mix-tures, even at the isomer level [1], it is important to assess thesechanges as a function of each component’s physical and chemi-cal properties [2]. Once known, one can use this information todetermine if natural attenuation is sufficient to reduce pollutantimpact on the environment or if active remediation is required. Tomake this determination, the compositional effects of dissolution,organic phase partitioning, and evaporation must be known; eachof which one can examine by studying the aqueous solubility (SW),octanol–water partition coefficient (KOW), and vapor pressure (VP)of sample components, respectively [3,4].

The measurement of aqueous solubility and octanol–waterpartition coefficient of hydrophobic fossil fuel components suchas benzene, polycyclic aromatic hydrocarbons (PAH) and sulfur

heterocycles (PASH) and their substituted homologues is time-consuming, challenging, and susceptible to error [5]. For thesereasons, gas chromatographic (GC) retention indices (RIs) are oftenused to estimate these properties [6,7]. GC provides the means to

224 P.M. Antle et al. / J. Chromatogr

Fe

ntieueibge[tSfBpLir[pt

psor[pctcea(tnaa

rcptcwntcs

ca

employed to evaluate if the accuracy of physical property estimates

ig. 1. Schematic representation of a component map; arrows correspond to weath-ring processes.

ot only estimate these properties, but also to assess the extento which natural attenuation has occurred, all without the need todentify each sample component or directly measure their prop-rties. Both 1-dimensional GC and 2-dimensional GC × GC can besed to estimate SW, KOW, and VP via linear or logarithmic freenergy relationships (LFERs). GC × GC is often used in weather-ng studies because it provides a visual depiction of the differencesetween fresh and weathered sample chromatograms, and ortho-onal column pairings provide the means to generate LFERs tostimate these properties simultaneously [4]. Arey and coworkers8,9] derived an empirical expression for the isothermal parti-ion coefficient (K), then used this information to estimate VP,W, and the air–water and octanol–water partition coefficientsrom the 1st and 2nd dimension retention indices of diesel fuel.ased on the assumption that the partitioning of the solute isrimarily controlled by size and polarizability, a two-componentFER, with the 1st dimension retention index (RI1D) conveyingnformation about size and the 2nd dimension polarizability, envi-onmental researchers have used GC × GC to produce contour maps10,11] and air and water mass transfer models [12], and estimatehase-transfer properties, phospholipid membrane–water parti-ion coefficients and corresponding narcosis toxicity [13].

This line of research is not without its challenges – for exam-le, the calculation of meaningful 2nd dimension retention indices,ince either the 2nd dimension hold-up time (tM,2D) or the retentionf bracketing compounds at different temperatures across the sepa-ation space is required. Researchers have examined column bleed14] and employed continuous injections of an unretained com-ound [15] to assess tM,2D, and created “hypothetical” bracketingompounds [8,9] and isovolatility curves [16]. Isovolatility curves,he result of continuous solute elution over a prolonged period,an provide the means to obtain retention information at differ-nt elution temperatures by creating curved elution lines that cutcross the 2nd dimension separation space. Unlike time-of-flightTOF) mass filters, only recently have quadrupole mass spectrome-ers provided the scan speeds sufficient for invariant spectra acrossarrow 2nd dimension peaks and, in turn, accurate quantitativenalysis of components of complex mixtures, such as PAH, PASHnd their alkylated homologs in coal tar and crude oil [17].

Fig. 1 shows a weathering map that can be used to informemedial decisions and to delineate site-specific weathering pro-esses [10]. The axes correspond to volatility and solubility androvide important sample information. For example, sites that con-ain a large number of volatile and highly soluble compounds willontinue to pose risk to the environment, as opposed to highlyeathered sites in which only non-volatile and insoluble compo-ents remain. If these compounds are also biologically inaccessible,he pollution no longer poses risk to the environment. This is espe-ially important since local weathering conditions, even at the sameite, may attenuate pollution differently.

In this study, we report for the first time the use of isovolatilityurves to generate retention indices in both GC × GC dimensionst every point in 2D space. In contrast with other studies that

. A 1364 (2014) 223–233

employed alkane standards to estimate the properties of aromatictargets, RIs are calculated and physical properties are estimatedusing 2-, 3-, 4-, and 5-ring PAH as bracketing compounds. Since PAHare mutagenic [18], carcinogenic [19], and persistent [20] organicpollutants, they serve as important model compounds for thisstudy. We and others have shown that when target analytes, in thiscase alkylated PAH and PASH, are bracketed by structurally similarcompounds, accurate measurement of their separation is obtainedunder linear temperature-programmed conditions [21–24]. This,in turn, leads to more robust linear free energy relationships andmore accurate physical property estimates. This property estima-tion process has been encoded in a new software program, allowingfor automated determination of physical properties from one anal-ysis, with minimal model training and parameter input. Since coaltar is predominantly aromatic [25], it serves as an ideal model mix-ture to test this hypothesis, especially considering our experiencewith analysis of C1- to C4-alkylated PAH and PASH homologues[17,26–30].

2. Experimental

2.1. Standards and reagents

Airgas (Salem, NH, USA) supplied the ultra-high purity heliumand nitrogen used in this study. Chromatography-grade tolueneand dichloromethane were purchased from Sigma–Aldrich (St.Louis, MO, USA). The 16 EPA priority pollutant PAH, diben-zothiophene, and internal standards 1.4-dichlorobenzene-d4,naphthalene-d8, phenanthrene-d10, chrysene-d12, and perylene-d12 were obtained from Restek (Bellefonte, PA, USA). Supelco(Bellefonte, PA, USA) supplied the base/neutral surrogatespike mix (2-fluorobiphenyl, nitrobenzene-d5, p-terphenyl-d14) as well as a number of neat standards: anthracene,benzo(b)thiophene, fluorene, fluoranthene, hexylbenzene, pyrene,1-phenyloctane, 1-methylnaphthalene, 2-methylnaphthalene, and1,7-dimethylnaphthalene. Also purchased were neat standards ofn-decylbenzene and 2,6-dimethylnaphthalene from Ultra Scien-tific (North Kingstown, RI, USA) and Crescent Chemical (Islandia,NY, USA), respectively.

2.2. Samples and sample preparation procedure

Pure coal tar and impacted soils were obtained from a utility inIllinois. We modified EPA method 3550C [30–32]: 15 g of samplewas spiked with surrogate mix and sonicated for 10 min in 8 mLof 50% (v/v) toluene/dichloromethane (Branson Ultrasonics, Dan-bury, CT), and the procedure was repeated eight times to obtainmaximum extraction efficiency. Activated copper and anhydroussodium sulfate were used to eliminate elemental sulfur and waterfrom the extracts, which were concentrated under a stream of nitro-gen prior to the addition of 10 �g/mL of internal standards.

2.3. Instrumentation

GC × GC/MS analyses were performed using an Agilent Tech-nologies (Santa Clara, CA, USA) 6890/5975C GC/MS with Gerstel(Mülheim an der Ruhr, Germany) MPS2 autosampler and CIS6injector. Since most sample components in coal tar are aro-matic members of homologous families, instrumental conditionswere chosen to maximize utilization of 2D separation spaceand to increase data granularity. Following the example of Areyand coworkers [8,9], two different GC × GC/MS methods were

was method-dependent. These included differences in columnmanufacturer and size, split ratio, temperature program, flowrate, and modulation time; instrumental parameters for both

P.M. Antle et al. / J. Chromatogr

Table 1GC × GC method parameters.

Method A Method B

GC parametersInjection mode 20:1 split SplitlessInjection

volume (�L)1 1

Injectiontemperatureor PTVprogram

−20 ◦C, 12 ◦C/min to320 ◦C, hold for 5 min

−20 ◦C, 12 ◦C/min to320 ◦C, hold for 5 min

Flow 1.2 mL/min, constantflow

1.0 mL/min, constantflow

Carrier gas HeliumTransfer line

temperature300 ◦C

Column 1 30 m × 0.25 mm × 0.25 �mRxi-5SilMS (Restek)

30 m × 0.25 mm × 1.0 �mDB-5 UI (Agilent)

Column 2 1.0 m × 0.25 mm × 0.25 �mRxi-17SilMS (Restek)

1.5 m × 0.18 mm × 0.36 �mRxi-17SilMS (Restek)

Modulationtime (s)

12 8

Hot jettemperature(◦C)

310

Cryogen flow(L/min)

15

Temperatureprogram

60 ◦C for 1 min,4.5 ◦C/min to 320 ◦C, holdfor 5 min, −20 ◦C/min to300 ◦C, hold for 5 min

60 ◦C for 1 min,6.5 ◦C/min to 320 ◦C, holdfor 5 min

Run time (min) 65.77 76.77MS parametersSolvent delay

(min)10 17

Mass range(m/z)

50–300 50–350

Scan rate(scans/s)

23.7 19.8

Quadrupoletemperature(◦C)

150 150

Ion source 230 230

Gult(wsM

utuS3T3

2

u(lbut

t (T) = × × (�(T)) (2)

temperature(◦C)

C × GC/MS methods are found (Table 1). Columns were connectedsing Restek press-fit connectors or VICI Valco (Houston, TX, USA)

ow mass external union connectors. The GC × GC cryogenic andhermal modulation hardware was provided by Zoex CorporationHouston, TX, USA). GC Image (Lincoln, NE, USA) supplied the soft-are to create the three-dimensional chromatograms. Unknown

ample components were identified using Ion Analytics (Andover,A, USA) spectral deconvolution software.2D retention indices were confirmed by analyzing standards

nder isothermal conditions; i.e., we compared our GC × GC RI2Do 1D GC/MS RI results for the same compounds on the same col-mn. These 1-dimensional GC/MS analyses were performed using ahimadzu (Columbia, MD, USA) GC2010/QP2010+ instrument on a0 m × 0.25 mm × 0.25 �m RXI-17MS column provided by Restek.hese runs were performed at temperatures between 100 and00 ◦C at 25 ◦C increments. Methane was used to measure tM.

.4. Software programming and functionality

The automated physical property program was constructedsing the R and Octave software environments within MATLABMathworks, Natick, MA, USA). The software is available for down-

oad in supporting information. Basic functionality is describedelow, and a point-by-point list of instructions and illustrative fig-res is found in the supplemental information. The program wasrained through input of the SW, VP, and KOW values and retention

. A 1364 (2014) 223–233 225

times obtained from the bracketing compound isovolatility curves,see Fig. 2A. As fitting the correct curve to the 2D chromatogramis essential for obtaining accurate retention indices, the MATLABcurve fitting toolbox was utilized to test a number of possible mod-els. During the training step of each regression, the curve toolbox’s“fit” function was used to fit a curve based on a specific model (lin-ear, quadratic, power, etc.), with the x and y values obtained fromthe 1D and 2D retention times of the isovolatility curve bracketingcompounds. The resulting curve was stored and the goodness offit recorded. This process is repeated for all curve types, and thebest fit is selected for further use. In order to use these isovolatilitycurves for retention index calculation and physical property esti-mation, a second spreadsheet was used to input the 1st and 2nddimension retention times and literature physical property val-ues of training compounds that spanned the retention space andphysical property estimation limits of the sample, see Fig. 2B. Thissheet also contained the retention times of the user-defined targetcompounds.

Using this data, the program compares the RT from the train-ing compounds to those of the isovolatility curves to determinewhich bracketing standards to use for each training compound.Next, it calculates the 1D and 2D retention indices for each trainingcompound and regresses the known physical properties against theRI values. The literature vapor pressures, octanol–water partitioncoefficients, and aqueous solubilities of these training compoundsare used to construct free energy relationships that output VP, KOW,and SW estimates for each point in the 2D chromatogram basedsolely upon retention indices, see Fig. 2C. These estimates are usedto “map” the entire chromatogram. Partition lines for both vaporpressure and aqueous solubility are evaluated in half-minute inter-vals across the 2D chromatogram to delineate sections of similarproperties. The two resulting surfaces are plotted as a contour mapon top of the 2D chromatogram, with the contour lines correspond-ing to physical property log integer values. The model then outputsVP, KOW, and SW estimates for each target compound. Any com-pound in the sample can be evaluated using this approach. If thephysical property literature values of compounds in the sampleare known, the model-generated estimates are compared to pre-viously measured values. In this study, however, there was limitedopportunity for these comparisons, given the paucity of KOW andSW literature for coal tar components beyond the 16 PAH listed byEPA as target compounds [33–37]. In cases where literature val-ues are available, the program generates an error analysis (modelversus literature value estimate) for each LFER. This error analy-sis is displayed in the output spreadsheet and each compound isplotted on the component map with a user-defined color that cor-responds to the magnitude of the error. Importantly, only a singleGC × GC analysis is required to produce the contour map and thecorresponding property estimates.

3. Theory

As outlined by Curvers and expressed in Eq. (1), retention, r,as a function of temperature, T, is controlled by two factors: (1)thermodynamics, as described by the bracketed terms; and (2) fluiddynamics, as described by the t0(T) term:

r(T) =∫ TR

T0

dT

t0(T)[1 + (exp(�S/R)/�) exp(�H/RT)

] (1)

where: ((P3 − 1)

) (128 × L2

)

0

P4 − 2P2 + 1 3 × po × d2c

and � is the column phase ratio, P is the column pressure ratio(in/out), po is the column outlet pressure, L is the column length,

226 P.M. Antle et al. / J. Chromatogr. A 1364 (2014) 223–233

F pertyt param

dgEs(t[tWmTsiciviptiict

piic[isiAtvea

ig. 2. Input and output software process used to produce component maps and prohe model, (B) graphic user interface used to input file from (A) and select modeling

c is the column inner diameter, and �(T) is the dynamic carrieras velocity [38]. If a mass spectrometer is used, the first term ofq. (2) simplifies to an inverse linear function by assuming P = pinuch that P4

i− 2P2

i+ 1 ≈ P4

iand P3 − 1 ≈ P3. The middle term in Eq.

2) is a combination of constants, and, as noted by Ettre, the �(T)erm is a linear function within the typical GC temperature range39]. Therefore, by approximation, Eq. (2) is a ratio of linear func-ions, and Eq. (1), in turn, the integral of an inverse linear function.

hen P is constant throughout a run, which is common in isother-al separations, the integral of the fluid dynamic term from T0 to

R (e.g., the fluid hold up time change versus temperature) is a con-tant; this mathematical prediction is borne out in practice as tM

s unchanged with constant pin and temperature. Retention, in thisase, is entirely dependent on the thermodynamics of the analytetself. In contrast, when column temperature and/or inlet pressurearies throughout the separation, the integrated fluid dynamic terms a non-linear function. As such, and as is the case in temperature-rogrammed, constant flow GC, retention is intimately tied to bothhermodynamics and fluid dynamics, the latter of which is exper-mentally condition-dependent. These effects are readily observedn temperature-programmed GC analyses, as the elution order ofritical pairs can be reversed with changes to system flow andemperature programming parameters [40].

Despite this fundamental relationship between fluid dynamicarameters and retention behavior, in practice, the use of retention

ndices allows for the repeatable measurement of retention behav-or. This is observed in inter- and intra-laboratory retention indexomparisons, wherein indices typically vary by only a few percent21–23,41]. This small error is partially attributable to unavoidablenconsistencies between chromatographic systems (cold and hotpots, variability in dead volumes and stationary phase crosslink-ng, etc.) and slight differences in absolute mass on column.lthough changing fluid dynamic factors can reverse elution orders,

he corresponding retention indices do not change enough to pre-ent a good regression of elution behavior versus physical propertystimation, especially with respect to the considerable error associ-ted with traditional measurements of the physical property [5]. In

estimates. (A) Retention times and literature physical property values used to traineters, and (C) output of property estimates, visualization map, and error analysis.

order for fluid dynamic effects to significantly affect the retentionbehavior of compounds such that changes in the retention indexpreclude regression of physical properties, dramatic (and unrealis-tic) instrumental conditions would be necessary.

RIs have been shown to correlate with the SW, VP, and KOW of agiven compound, and LFERs have been used to estimate these prop-erties for families of compounds based on their RI [8,42–48]. Thesefree-energy relationships typically follow the linear or logarithmicforms shown below:

X = a1Y1 + a2Y2 + · · · + anYn + c (3)

X = a1 log Y1 + a2 log Y2 + · · · + an log Yn + c (4)

where X is the physical property to be determined for a com-pound, Yn is either the retention index or a previously known ordetermined physical property of that compound, and an are fittedconstants [49].

In addition to the fluid dynamic factors discussed above, reten-tion is dependent on multiple thermodynamic factors, includingthe partial molar enthalpy of solution, partial molar entropy of solu-tion, and partial molar isobaric heat capacity [21]. Predictably, therelationship between temperature and the thermodynamic vari-able of interest changes dramatically based on the chemistry of themolecule [50–52]. Assumptions are often made about the temper-ature dependence, or equivalency, of each of these parameters forboth the compound of interest and reference compound [53–57].Care must be taken to minimize the error brought about by theseassumptions through the selection of appropriate reference com-pounds that undergo the same intermolecular reactions as thecompounds of interest in the stationary phase (i.e., mechanisti-cally they must be the same) [58]. When comparing a family ofcompounds, such as alkylated PAH, the underlying thermodynamicparameters change with size and structure such that the relative

separation remains unchanged despite differences in chromato-graphic conditions (temperature, heating rate, phase ratio, etc.).However, when reference and analyte compounds behave differ-ently, measurable effects in retention indices and, in turn, physical

P.M. Antle et al. / J. Chromatogr. A 1364 (2014) 223–233 227

wo ho

phctomcAos

t

k

Fig. 3. Comparison of thermodynamic parameters of t

roperty modeling are found. To date, however, researchers whoave estimated physical properties often use alkanes as referenceompounds for analyses of polycyclic aromatic hydrocarbons, evenhough alkanes exhibit different intermolecular behavior than PAHn the same stationary phase [59]. Fig. 3 compares relevant ther-odynamic parameters as a function of temperature for selected

ompounds in two homologous series – PAH and alkanes [60–63].s seen in the figure, differences are found between the two homol-gous series, both in the parameters themselves as well as thelopes of the functions.

The activity coefficient at infinite dilution (Fig. 3A) can be relatedo retention behavior by recasting the capacity factor, k, as follows:

= tr − tm

tm= TRnS

�∞i,s

p0iVM

(5)

mologous compound families, alkanes and aromatics.

where R is the gas constant, T the temperature, ns the total num-ber of moles in the stationary phase, �∞

i,sthe activity coefficient

at infinite dilution, p0i

the vapor pressure of the pure solute, andVM the volume of the stationary phase. As Beens et al. [64] state:“there are only two compound-dependent factors that affect reten-tion. These are the vapour pressure of the pure solute, p0

i, which is

an exponential function of the temperature and the activity coef-ficient of the solute in the stationary phase.” Therefore, when thesolute and reference compound activity coefficients do not scalewith temperature in a uniform manner (see Fig. 3A), any accu-racy in LFER regressions is empirical; useful under only a single

set of instrumental conditions. In contrast, LFER-based vapor pres-sure estimation amongst a homologous series, i.e. where �∞

i,sand

�∞i,x

scale uniformly with temperature, results in differences of only∼4%, which is in line with the measurement error common in

2 atogr. A 1364 (2014) 223–233

cLltAtt(Goaweht

dLpaObctc

4

msowTtcsucptrt

4

p[

L

wgitbab1fiact

Fig. 4. Naphthalene and phenanthrene isovolatility curves used to calculate 2nddimension retention index. While the 1st dimension retention index of point X

28 P.M. Antle et al. / J. Chrom

hromatographic measurements [65]. In other words, producingFERs by regressing homologous compounds against one anothereads to measurement error on the same order of magnitude ashe chromatographic measurements these estimates are based on.s the activity coefficient at infinite dilution is governed by parti-

ion thermodynamics, it is intimately related to the aforementionedhermodynamic parameters heat capacity (Fig. 3B) and free energyFig. 3C). As seen in the figures, these functions are linear across theC temperature range for the homologous series, but the slopesf the functions are not equivalent, which precludes the use oflkane retention to estimate the physical properties of aromaticsith good accuracy, and vice versa. The thermodynamic factors are

xplainable on a molecular basis. For example, PAH have a muchigher affinity for H-bonding and Van der Waals (VdW) interactionshan alkanes [58].

Chemical and structural differences between analytes and stan-ards can dramatically effect retention indices, and in turn, theFERs derived thereof. In this way, the assumptions outlined aboverovide the method with an obvious limitation – that is, it is onlypplicable for homologous series of compounds in the same family.n the other hand, since the thermodynamic similarities of the PAHrackets lead to improvements in RI reproducibility for aromaticompounds in general when compared to alkane-based methods,he same brackets should lead to reliable estimates of all aromaticomponents found in complex samples such as coal tar.

. Results and discussion

GC × GC offers a distinct advantage in the chromatographic esti-ation of the physical properties of analytes in complex samples,

ince orthogonal column pairings provide simultaneous estimationf multiple properties for hundreds of components in a single run,ithin time scales similar to those of one-dimensional analyses.

his is particularly helpful in environmental forensics, as many ofhe processes that comprise weathering are related to the physi-al properties of the constituents of the mixture. For example, thepecific physical properties examined in this study – aqueous sol-bility, vapor pressure, and octanol–water partition coefficient –an be utilized to study water washing, evaporation, and organichase partitioning, respectively. Our goal is to correlate each ofhese properties with GC retention, and the key to producing aobust correlation is the accurate and precise generation of reten-ion indices.

.1. 1st Dimension retention index

As illustrated in Fig. 4, the 1st dimension linear temperature-rogrammed retention index (LTPRI1D) is calculated using Eq. (6):21]

TPRI1D = LTPRIB +[(

RTX − RTB

RT(B+1) − RTB

)× 100

](6)

here RT refers to the 1st dimension retention times of tar-et compound X and bracketing compounds B and (B + 1), e.g.,n the case of Fig. 4, naphthalene and phenanthrene, respec-ively, and LTPRIB is the retention index of the earlier-elutingracket compound. The bracketing compound retention indicesre: naphthalene (200), phenanthrene (300), chrysene (400), andenzo(g,h,i)perylene (500). As discussed in Theory, although thest dimension separation is temperature-programmed and there-ore dependent on fluid and thermodynamics, the corresponding

ndices are still capable of providing precise retention indicesnd good correlations to thermodynamic data under controlledonditions [66]. Despite two vastly different experimental condi-ions (differing column lengths, stationary phase film thicknesses,

can be calculated based on RT1D, isovolatility curves (and extrapolation of thephenanthrene isovolatility curve) are required for calculation of the 2nd dimensionretention index.

carrier gas flow rates, and temperature programs), we obtained 1stdimension retention index precision of ±0.4%, which is excellent.This result is consistent with past studies [21–23,67,68] and is animprovement in precision compared to previous GC × GC propertyestimation studies that employed alkane-bracketed LTPRI [8,9] orretention times [13].

4.2. 2nd Dimension retention index

In contrast to the temperature-programmed 1st column separa-tion, each 2nd dimension separation is effectively isothermal, thus:[69]

RI2D = RIB +[(

log(RTX/RTB)log(RT(B+1)/RTB)

)× 100

](7)

Since isothermal retention indices are directly related to ln K(the equilibrium constant for the solute partitioning between twophases), which, in turn, is related to Gibbs free energy, the use of ahomologous series and associated indices is not necessary. In thisway, any retention index system (or capacity factor k) can be used toproduce valid thermodynamic data, with the caveat that one must:(a) correct for temperature since each ln K corresponds to its own2nd dimension modulation temperature and (b) determine tM,2D,which is not constant throughout the separation. Beyond thesechallenges, there has also been historical difficulty in generatinga “meaningful” 2nd dimension retention index in GC × GC analy-ses, as outlined in the review by von Muehlen and Marriott [14].Briefly, depending on the homologous series chosen, e.g. alkanes,alcohols, amines or FAMEs (a comprehensive list of RI schema canbe found in Castello et al., 1999) [70], target compounds may notelute within the retention window of the brackets. For example,in Fig. 4, compound X does not elute in the 2nd dimension reten-tion window of naphthalene and phenanthrene (i.e., if not for theisovolatility curves, neither bracket compound would elute at thesame 1st dimension time as compound X), so their 2nd dimensionretention times cannot be compared. While one can analyze the tar-get compound and the brackets using isothermal GC on the samestationary phase as the second column, or continuously introducethe bracket compounds by means of extra plumbing, we developeda simpler method to accurately measure RI2D, in which compoundX can be compared to B and B + 1 at the same elution temperature.

Our method of addressing this issue came from the realiza-tion that we could take advantage of the GC × GC modulationprocess by having the cold jet serve as a secondary inlet. The

modulator employed in our system passes a stream of cryogenicgas across a capillary column loop, as shown in Fig. S1. Periodi-cally, a secondary stream of hot gas with much higher flow rateis pulsed across the same portion of the GC column to desorb the

P.M. Antle et al. / J. Chromatogr

Table 2Retention indices for PAH and alkyl PAH on a DB-17 column as calculated fromisothermal GC/MS data and GC × GC 2nd dimension isovolatility curves.

Compound GC/MSisothermalRI

GC × GCcalculatedRI

% Error

1-Methylnaphthalene 220.70 219.85 0.42,6-Dimethylnaphthalene 231.19 231.18 0.02-Ethylnaphthalene 229.06 231.36 1.02-Methylnaphthalene 216.36 216.01 0.2Acenaphthylene 246.86 246.24 0.3Anthracene 300.98 303.22 0.7Fluoranthene 341.72 343.71 0.6Fluorene 265.86 265.49 0.1n-C10 benzene 253.01 251.13 0.7

shisaTsGdcdee2

i

TL

TL

n-C8 benzene 222.14 224.49 1.1Pyrene 351.46 354.54 0.9

ample. By overloading the modulator, some of the analyte thatas already eluted through the first column is “delayed” upon

ntroduction to the second column; the resulting chromatogram isimilar to the ones produced using the method employed by Bierind Marriott [71], except that no additional plumbing is required.he process essentially generates dozens, if not hundreds, of con-tant temperature chromatograms for that compound within oneC × GC run. The end result is a 2D chromatogram marked by easilyefined isovolatility bracket curves, see Fig. 4, which shows distincthromatographic peak tails with asymptotically decreasing secondimension retention times. This is to be expected, since the analytexperiences a continual increase in modulation temperature during

ach injection onto the second column, and, as such, the resultingnd retention time decreases with each subsequent pulse.

In those cases in which the elution line did not extend far enoughn a particular dimension to provide “coverage” for a particular

able 3inear free energy relationships (LFERs) used to estimate physical properties and their co

Solute property GC × GC method LFER coefficient aRI1D +

a

VP

A −0.027

B −0.026

A (Column 2) −B (Column 2) −

KOWA 0.10 −B 0.08 −

SWA −0.12

B −0.09

able 4iterature and estimated PAH and alkyl PAH physical properties.

Compound log VP (atm) log SW

Literature Calculated % Error Literatu

1,3-Dimethylnaphthalene −4.515 −4.574 1.3 −4.2911,4-Dimethylnaphthalene −4.550 −4.650 2.2 −4.1361-Ethylnaphthalene −4.479 −4.461 0.4 −4.1691-Methylnaphthalene −4.055 −4.009 1.1 −3.7172,6-Dimethylnaphthalene −4.473 −4.499 0.6 −4.1672-Ethylnaphthalene −4.381 −4.461 1.8 −4.2912-Methylnaphthalene −4.046 −3.933 2.8 −3.653Acenaphthene −4.939 −4.876 1.3 −3.966Benzo[a]fluorene −7.865 −7.980 1.5 −5.048Benzo[a]pyrene −9.677 −10.013 3.5 −6.338Benzo[b]fluoranthene −10.391 −9.972 4.0 −6.599Dibenz[a,h]anthracene −11.606 −11.277 2.8 −7.690Fluoranthene −7.179 −7.290 1.5 −5.177Fluorene −5.303 −5.291 0.2 −4.157n-C10 benzene −5.774 −5.668 1.8 −7.958n-C8 benzene −4.828 −4.763 1.3 −6.459Pyrene −7.232 −7.549 4.4 −5.187

. A 1364 (2014) 223–233 229

compound (see phenanthrene in Fig. 4), we used MATLAB’s curvefitting functions to extrapolate the isovolatility line across the 2Dretention space. The best fit for isovolatility curves was found in anexponential function of the form:

RT2D = A × e−b×RT1D (8)

To determine the minimum number of points for an accept-able fit, we varied the number of 1D and 2D bracket compoundretention times and used Eq. (8) to estimate the 2nd dimensionretention times for each point on the isovolatility curve. For allbracketing standards, b was approximately 0.1, meaning the fit-ted curves were of similar shape. With as few as five consecutivepoints along the isovolatility curve, the exponential functions wegenerated predicted a given compound’s RT2D at any RT1D within5% (0.5 s). As the isovolatility curves are not themselves exact expo-nential functions, this level of accuracy is excellent. Nonetheless,given the importance of obtaining accurate RT2D, an isovolatilitycurve of 20 points or more is recommended, as this produces 2nddimension RT accuracy >99%.

Based on these results, we examined the accuracy of the 2nddimension retention indices by comparing the calculated RI2D fromthe isovolatility curves against isothermally measured indices, seeTable 2. Excellent agreement was obtained; the calculated and mea-sured indices were within 1.5 units, or 0.6%, of one another. Thesefindings confirm the validity of our approach for calculating 1st and2nd dimension retention indices and are comparable to those of thetemperature-corrected ln K approach, which requires burdensomethermodynamic calculations and training measurements for each

unique experimental condition [8,9].

The advantage gained by these retention tails is twofold: inaddition to knowing precisely when a compound elutes on the sec-ond column at a number of 1D elution temperatures, the highly

rresponding statistics.

bRI2D + c Standard error r2

b c

1.9 0.10 0.991.8 0.11 0.98

0.0067 −2.9 0.02 0.990.0025 −3.6 0.10 0.980.09 1.0 0.17 0.980.07 1.1 0.19 0.950.11 −0.54 0.24 0.950.08 −0.28 0.18 0.95

(mol L−1) log KOW

re Calculated % Error Literature Calculated % Error

−4.297 0.1 4.420 4.414 0.1 −4.326 4.6 4.373 4.441 1.6 −4.183 0.3 4.397 4.314 1.9 −3.486 6.2 3.870 3.708 4.2 −4.476 7.4 4.333 4.562 5.3 −4.284 0.2 4.380 4.399 0.4 −3.564 2.4 3.915 3.771 3.7 −4.034 1.7 3.920 4.202 7.2

−5.371 6.4 5.373 5.444 1.3 −6.089 3.9 6.040 5.872 2.8 −6.353 3.7 6.000 5.980 0.3

−7.308 5.0 6.750 6.647 1.5 −4.840 6.5 5.195 4.970 4.3 −4.319 3.9 4.180 4.458 6.6 −7.613 4.3 7.350 7.256 1.3 −6.414 0.7 6.300 6.210 1.4 −4.832 6.8 4.994 4.973 0.4

230 P.M. Antle et al. / J. Chromatogr. A 1364 (2014) 223–233

Fig. 5. Aqueous solubility (A) and octanol–water partition coefficient (B) contour maps derived from the analysis of a coal tar-impacted soil sample. Black (vertical) linesc compp

rwisdfp

4

crpo

orrespond to vapor pressure, red (contour) lines to either SW (A) or KOW (B). All

roperty is <0.5.

eproducible tail also permits the estimation of when a compoundould have eluted on the second column if it had eluted earlier than

t otherwise did. In other words, we can estimate the 2nd dimen-ion retention time of any compound both before and after its 1stimension elution time. With the ability to determine RI1D and RI2Dor the entire separation space, we can now estimate the physicalroperties of every compound in the sample.

.3. Estimation of physical properties

A compound’s vapor pressure defines the volatility of a

ompound, which affects its transport and partitioning in envi-onmental matrices. It can be used to determine a compound’sresence in the atmosphere, aqueous media, soil, and soil-boundrganics. A partition process dependent only on VdW interactions

ound dots are colored teal because the absolute error of each estimated physical

can be adequately predicted for a group of compounds using onlya single-parameter LFER [58] and, as such, vapor pressure valuesfor each component in coal tar was determined using the followingsingle-parameter LFER:

VP = aRI1D + c (9)

LFER regression statistics are found in Table 3. The excellentcorrelation between vapor pressure and nonpolar column RI isconsistent with models developed by others [8,9,13], but in thisstudy we obtained a lower standard error of ∼10%. Vapor pres-sure estimates based on our model were accurate to within 1.8%

of literature values, on average, see Table 4, and are more accu-rate than those produced by 1D RI-based boiling point estimates[72]. We also estimated vapor pressure using only 2nd dimensionretention indices, as retention on the DB-17 column is also partially

P.M. Antle et al. / J. Chromatogr. A 1364 (2014) 223–233 231

F overls

gfraapps

mteawirafsscr

tficdoma

l

wrfsA

ig. 6. Vapor pressure (vertical) and aqueous solubility (contour) lines from Fig. 5ame line possess the same physical properties.

overned by VdW interactions, and found results similar to thoserom the 1st column indices, see Table 3. The agreement of theseesults despite the differing phenyl content of these two columnsnd corresponding changes in molecular interactions suggests thatny methylphenyl-based column can be used to estimate the vaporressure of aromatic solutes. In the same vein, a column with nohenyl content will likely provide superior results for a homologouseries of alkanes.

When examining analyte partitioning that is dependent onultiple types of intermolecular interactions, one would assume

hat a polyparameter LFER would provide improved results, withach parameter corresponding to a type of intermolecular inter-ction. In contrast, others have discovered for compound classeshose H-bond interactions are proportional to their van der Waals

nteractions, a single-parameter LFER will provide results compa-able to a two-parameter LFER [58]. Although proportional VdWnd H-bonding interactions are characteristic of PAH [58], weound that polyparameter LFERs were markedly superior to theingle-parameter LFERs for KOW and SW; e.g., the single-parametertandard error was on average 3–6 times higher than that of theorresponding two-parameter LFER. Therefore, only the polypa-ameter results are presented.

The remaining two physical chemical properties examined inhis study, aqueous solubility and octanol–water partition coef-cient, also play important roles in examining the behavior ofhemicals in the environment. These properties can be used toetermine bioaccumulation factors and partition coefficients withrganic carbon, and a reliable assessment of fate in the environ-ent requires accurate values for both properties [73]. In each case,

two-parameter LFER was employed:

og SW or log KOW = aRI1D + bRI2D + c (10)

here a, b, and c are dependent on the property estimated. LFER

egression statistics are found in Table 3. Results were excellentor SW and KOW, producing comparable correlation coefficients butmaller standard errors when compared to other studies [8,9,13].queous solubilities and the octanol–water partition coefficients

aid onto a GC × GC chromatogram of the same sample; compounds that lie on the

were within 4.2% and 2.3% of literature values, see Table 4. In all,the average error for the three estimated parameters across all com-pounds was 3.0% for two different GC × GC operating conditions.The model used to obtain these estimates is based on log-linear freeenergy relationships, which correspond to more accurate estimatesthan the log–log relationships reported in other work.

4.4. Component maps

The component maps in this study are similar to thoseintroduced by Arey and coworkers [11,12] for diesel fuel. Given theKOW and SW estimates discussed above, the resulting map offers amore accurate delineation than those previously reported. To cre-ate the map, the LFERs outlined in Eqs. (9) and (10) were used tooverlay physical property estimates onto corresponding retentiontimes, extrapolated from RIs, which can be used to estimate theproperties of any compound that elutes within the bounded reten-tion space. In the SW and KOW component maps shown in Fig. 5Aand B, the blue lines trace the isovolatility curves, while the red andblack contour lines designate the areas of the chromatogram thatcorrespond to a specific value of one of the three physical propertiesestimated in this study. Each dot on the chromatogram representsa compound identified in the coal tar sample whose physical prop-erties have been estimated based solely on retention indices, usingthe LFERs found in the top-left corner of the map. If the absoluteerror of each estimated property is less than 0.5 (see Table 4 forunits), the dot is colored teal. This is the case for each compoundfound in the figure. Had any property estimate’s error been between0.5 and 1, the corresponding compound’s dot would be colored red.In each case, the dot color (teal or red) darkens as the threshold isapproached.

In the figure, the black lines correspond to vapor pressure,and as this LFER is based solely on RI1D, the lines are vertical,

relating only to the 1st dimension retention time. For example, 1,3-dimethylnaphthalene, which is labeled in the SW component map,Fig. 5A, falls approximately halfway between the black lines labeled-4 and -5, and has an estimated log vapor pressure of −4.57 atm.

2 atogr

TegapvitoIbv2e

mtiolsipact−otialftiiwahacpctp

5

itigpiwppphsaoos

[

[

[

[

[

[

[

[

[

[

[

[

32 P.M. Antle et al. / J. Chrom

his compares well with the literature value of −4.52 atm. Since SWstimates are based upon retention values from both chromato-raphic dimensions (see Eq. (10)), the corresponding lines curvecross the chromatogram. As with the previous example, any com-ound that lies on the red line labeled -7 has an estimated SWalue of −log 7 mol L−1. For 1,3-dimethylnaphthalene, we see thatt lies in the area almost exactly one-third of the way betweenhe lines labeled -4 and -5, corresponding to an estimated log SWf −4.30 mol L−1, which differs from the literature value by 0.1%.n the bottom chromatogram, Fig. 5B, benzo[a]fluorene is locatedetween red lines 5 and 6, with an estimated log KOW of 5.44. Thisalue differs from the literature value by 0.07 (1.3%). Given 1st andnd dimension retention times, VP, SW, and KOW can be readilystimated for every compound found in the 2D chromatogram.

Fig. 6 shows the contour lines overlaid onto the 2D chro-atogram of an environmentally exposed soil and illustrates how

his component map can be used to assess physical weatheringn a more informed manner. Note that maps created in this studynly allow for the investigation of physical processes and not bio-ogical and chemical processes. Visually, we can still see that theample has only minimally weathered, since many of the aliphat-cs and a number of the C1- and C2-alkylated naphthalenes areresent, but the additional value of the component map is that itllows assignment of volatility and solubility values to all sampleomponents. For example, a wide range of organics, whose respec-ive log VP and log SW are between −4 and −10 atm and −3 and7 mol L−1, have not been lost to the environment. An examinationf these compounds suggests not only that the sample will con-inue to weather, but also by which physical weathering process,.e., dissolution and/or evaporation. For example, both carbazolend C4-alkylated biphenyls, identified in the figure by A and B,ie on the same volatility line, but their aqueous solubilities dif-er by two orders of magnitude. The biphenyls are far more likelyo be affected by water-washing than carbazole. If the same sites analyzed later and C4-biphenyl is present but carbazole is not,t could be inferred that the sample and, thus, the site, was water-

ashed to a greater extent than volatilized. Similarly, the biphenylsnd benzo[a]pyrene (B and C) lie on the same SW contour line, butave widely differing volatilities. The former is far more likely to beffected by evaporation than the latter. Assessments such as thesean be made for any point in the chromatogram, whether the com-ound has been identified or not, over the lifetime of the site, fromontamination to final remediation. Specific to this site, it is clearhat volatile and soluble compounds remain and further release ofollutants is likely.

. Conclusion

The outcome of this research supports hazardous waste sitenvestigation and cleanup projects. In addition to providing quanti-ative measurements of pollutant concentrations, GC × GC/MS datan conjunction with our model discerns the mechanism and pro-ression of how a site weathers due to dissolution, organic phaseartitioning, and evaporation caused by the local environment. Our

mproved method for obtaining retention information, if combinedith LFER-based approaches to estimating phospholipid–waterartition coefficients and, in turn, bioaccumulation [13], shouldrovide insight into how contaminants weather due to biologicalrocesses. In addition, correlating weathering processes with Abra-am solvation parameters [74] and other molecular descriptorshould further inform forensic investigations, such as longitudinal

ssessments of how weathering patterns change over time. More-ver, this approach can be used by researchers in a wide rangef disciplines, including toxicology (bioaccumulation and toxicitytudies) and restoration (ecological and urban planning studies).

[

. A 1364 (2014) 223–233

Acknowledgements

The authors appreciate the support of Zoex, GC Image, Gers-tel, Shimadzu, Agilent and Ion Analytics in this study. This workwas partially funded by the Electric Power Research Institute underagreement EP-P39203/C17417.

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at http://dx.doi.org/10.1016/j.chroma.2014.08.033.

References

[1] M.C. Kennicutt II, The effect of biodegradation on crude oil bulk and molecularcomposition, Oil Chem. Pollut. 4 (1988) 89–112.

[2] G.S. Douglas, A.E. Bence, R.C. Prince, S.J. McMillen, E.L. Butler, Environmen-tal stability of selected petroleum hydrocarbon source and weathering ratios,Environ. Sci. Technol. 30 (1996) 2332–2339.

[3] P.M. Sherblom, P.M. Gschwend, R.P. Eganhouse, Aqueous solubilities, vaporpressures, and 1-octanol–water partition coefficients for c9–c14 linear alkyl-benzenes, J. Chem. Eng. Data 37 (1992) 394–399.

[4] R.K. Nelson, B.M. Kile, D. Plata, S.P. Sylva, X. Li, C.M. Reddy, R.B. Gaines, G.S.Frysinger, S.E. Reichenbach, Tracking the weathering of an oil spill with com-prehensive two-dimensional gas chromatography, Environ. Forensics 7 (2006)33–44.

[5] K. Ruzicka, B. Koutek, M. Fulem, M. Hoskovec, Indirect determination ofvapor pressures by capillary gas–liquid chromatography: analysis of the ref-erence vapor-pressure data and their treatment, J. Chem. Eng. Data 57 (2012)1349–1368.

[6] G. Tse, H. Orbey, S.I. Sandler, Infinite dilution activity coefficients and Henry’slaw coefficients for some priority water pollutants determined by relative gaschromatographic method, Environ. Sci. Technol. 26 (1992) 2017–2022.

[7] D. Mackay, R.S. Boethling, Handbook of Property Estimation Methods for Chem-icals, CRC Press, Boca Raton, FL, 2000.

[8] J.S. Arey, R.K. Nelson, L. Xu, C.M. Reddy, Using comprehensive two-dimensionalgas chromatography retention indices to estimate environmental partitioningproperties for a complete set of diesel fuel hydrocarbons, Anal. Chem. 77 (2005)7172–7182.

[9] J.S. Arey, R.K. Nelson, L. Xu, C.M. Reddy, Using comprehensive two-dimensionalgas chromatography retention indices to estimate environmental partition-ing properties for a complete set of diesel fuel hydrocarbons, correction, Anal.Chem. 79 (2007), 4736-4736.

10] G.S. Frysinger, R.B. Gaines, C.M. Reddy, GC × GC – a new analytical tool forenvironmental forensics, Environ. Forensics 3 (2002) 27–34.

11] J.S. Arey, R.K. Nelson, C.M. Reddy, Disentangling oil weathering using GC × GC,part 1: chromatogram analysis, Environ. Sci. Technol. 41 (2007) 5738–5746.

12] J.S. Arey, R.K. Nelson, D.L. Plata, C.M. Reddy, Disentangling oil weathering usingGC × GC, part 2: mass transfer calculations, Environ. Sci. Technol. 41 (2007)5747–5755.

13] A.P. Tcaciuc, R.K. Nelson, C.M. Reddy, P.M. Gschwend, Estimating phos-pholipid membrane–water partition coefficients using comprehensive two-dimensional gas chromatography, Environ. Sci. Technol. 46 (2012) 3449–3456.

14] C. von Muehlen, P.J. Marriott, Retention indices in comprehensive two-dimensional gas chromatography, Anal. Bioanal. Chem. 401 (2011) 2351–2360.

15] M.S. Klee, L.M. Blumberg, Measurement of retention in comprehensive two-dimensional gas chromatography using flow modulation with methane dopant,J. Chromatogr. A 1217 (2010) 1830–1837.

16] R.J. Western, P.J. Marriott, Retention correlation maps in comprehensive two-dimensional gas chromatography, J. Sep. Sci. 25 (2002) 831–838.

17] P.M. Antle, C.D. Zeigler, Y. Gankin, A. Robbat, New spectral deconvo-lution algorithms for the analysis of polycyclic aromatic hydrocarbons andsulfur heterocycles by comprehensive two-dimensional gas chromatogra-phy/quadrupole mass spectrometery, Anal. Chem. 85 (2013) 10369–10376.

18] B. Mahadevan, L.A. Courter, W.M. Baird, Mutations induced by polycyclicaromatic hydrocarbons, in: C.L. Valon (Ed.), New Developments in MutationResearch, Nova Science Publishers, Inc., Hauppauge, NY, 2007, pp. 19–40.

19] J. Jacob, The significance of polycyclic aromatic hydrocarbons as environmen-tal carcinogens. 35 years research on PAH-a retrospective, Polycyclic Aromat.Compd. 28 (2008) 242–272.

20] P.W. Sammarco, S.R. Kolian, R.A.F. Warby, J.L. Bouldin, W.A. Subra, S.A. Porter,Distribution and concentrations of petroleum hydrocarbons associated withthe bp/deepwater horizon oil spill, Gulf of Mexico, Mar. Pollut. Bull. 73 (2013)129–143.

21] C.E. Rostad, W.E. Pereira, Kovats and Lee retention indices determined by gas

chromatography/mass spectrometry for organic compounds of environmentalinterest, J. High Resolut. Chromatogr. 9 (1986) 328–334.

22] M.L. Lee, D.L. Vassilaros, C.M. White, M. Novotny, Retention indices forprogrammed-temperature capillary-column gas chromatography of polycyclicaromatic hydrocarbons, Anal. Chem. 51 (1979) 768–773.

atogr

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[berg, Solution thermodynamics of some slightly soluble hydrocarbons in water,J. Chem. Eng. Data 28 (1983) 197–200.

P.M. Antle et al. / J. Chrom

23] D.L. Vassilaros, R.C. Kong, D.W. Later, M.L. Lee, Linear retention index systemfor polycyclic aromatic compounds: critical evaluation and additional indices,J. Chromatogr. 252 (1982) 1–20.

24] F.R. Gonzalez, A.M. Nardillo, Retention index in temperature-programmed gaschromatography, J. Chromatogr. A 842 (1999) 29–49.

25] G.E. Totten, S.R. Westbrook, R.J. Shah (Eds.), Fuels and Lubricants Handbook:Technology, Properties, Performance, and Testing, ASTM, International, WestConshohocken, PA, 2003.

26] A. Robbat Jr., N.M. Wilton, A new spectral deconvolution – selected ion moni-toring method for the analysis of alkylated polycyclic aromatic hydrocarbonsin complex mixtures, Talanta 125 (2014) 114–124.

27] C.D. Zeigler, A. Robbat Jr., Detection of alkylated polycyclic aromatic hydrocar-bons: a comprehensive approach, Environ. Sci. Technol. 46 (2012) 3935–3942.

28] C.D. Zeigler, N. Wilton, A.J. Robbat, Toward the accurate analysis of C1 to C4

polycyclic aromatic sulfur heterocycles, Anal. Chem. 84 (2012) 2245–2252.29] A. Robbat Jr., A. Kowalsick, J. Howell, Tracking juniper berry content in oils and

distillates by spectral deconvolution of gas chromatography/mass spectrome-try data, J. Chromatogr. A 1218 (2011) 5531–5541.

30] P.M. Antle, C.D. Zeigler, N.M. Wilton, A. Robbat, A more accurate analysis ofalkylated PAH and PASH and its implications in environmental forensics, Int. J.Environ. Anal. Chem. (2013) 332–347.

31] United States Environmental Protection Agency, SW-846 Method 3550C: Ultra-sonic Extraction, Washington, DC, 1996.

32] K.D. Bartle, M.L. Lee, S.A. Wise, Modern analytical methods for environmentalpolycyclic aromatic-compounds, Chem. Soc. Rev. 10 (1981) 113–158.

33] Y.D. Lei, R. Chankalal, A. Chan, F. Wania, Supercooled liquid vapor pressures ofthe polycyclic aromatic hydrocarbons, J. Chem. Eng. Data 47 (2002) 801–806.

34] J.J.H. Haftka, J.R. Parsons, H.A.J. Govers, Supercooled liquid vapour pressuresand related thermodynamic properties of polycyclic aromatic hydrocarbonsdetermined by gas chromatography, J. Chromatogr. A 1135 (2006) 91–100.

35] R.S. Pearlman, S.H. Yalkowsky, S. Banerjee, Water solubilities of polynucleararomatic and heteroaromatic compounds, J. Phys. Chem. Ref. Data 13 (1984)555–562.

36] D. Mackay, W.Y. Shiu, K.C. Ma, Illustrated Handbook of Physical–Chemical Prop-erties and Environmental Fate for Organic Chemicals: Polynuclear AromaticHydrocarbons, Polychlorinated Dioxins, and Dibenzofurans, Lewis Publishers,Chelsea, MI, 1992.

37] W. Lai, C. Song, Temperature-programmed retention indices for GC and GC/MSanalysis of coal- and petroleum-derived liquid fuels, Fuel 74 (1995) 1436–1451.

38] J. Curvers, J. Rijks, C. Cramers, K. Knauss, P. Larson, Temperature programmedretention indices: calculation from isothermal data. Part 1: theory, J. High Res-olut. Chromatogr. 8 (1985) 607–610.

39] L.S. Ettre, Viscosity of gases used as the mobile phase in gas-chromatography,Chromatographia 18 (1984) 243–248.

40] R.J. Pell, H.L. Gearhart, Elution order inversions observed on using differentcarrier gas velocities in temperature programmed gas-chromatography, J. HighResolut. Chromatogr. Commun. 10 (1987) 388–391.

41] C.M. White, J. Hackett, R.R. Anderson, S. Kail, P.S. Spock, Linear temperatureprogrammed retention indices of gasoline range hydrocarbons and chlori-nated hydrocarbons on cross-linked polydimethylsiloxane, J. High Resolut.Chromatogr. 15 (1992) 105–120.

42] T.M. Letcher, P.K. Naicker, Determination of vapor pressures using gas chro-matography, J. Chromatogr. A 1037 (2004) 107–114.

43] D.J. Hamilton, Gas chromatographic measurement of volatility of herbicideesters, J. Chromatogr. 195 (1980) 75–83.

44] R. Hackenberg, A. Schutz, K. Ballschmiter, High-resolution gas chromatographyretention data as basis for the estimation of KOW values using PCB congenersas secondary standards, Environ. Sci. Technol. 37 (2003) 2274–2279.

45] S.F. Donovan, New method for estimating vapor pressure by the use of gaschromatography, J. Chromatogr. A 749 (1996) 123–129.

46] R.P. Schwarzenbach, P.M. Gschwend, D.M. Imboden, Environmental OrganicChemistry, John Wiley & Sons, New York, NY, 1993.

47] M. Hoskovec, D. Grygarova, J. Cvacka, L. Streinz, J. Zima, S.P. Verevkin, B. Koutek,Determining the vapour pressures of plant volatiles from gas chromatographicretention data, J. Chromatogr. A 1083 (2005) 161–172.

48] D.W. Hawker, D.W. Connell, Octanol water partition-coefficients of polychlo-rinated biphenyl congeners, Environ. Sci. Technol. 22 (1988) 382–387.

49] M.M. Miller, S. Ghodbane, S.P. Wasik, Y.B. Tewari, D.E. Martire, Aqueoussolubilities, octanol water partition coefficients, and entropies of melting ofchlorinated benzenes and biphenyls, J. Chem. Eng. Data 29 (1984) 184–190.

50] W. Shiu, K. Ma, Temperature dependence of physical and chemical properties ofselected chemicals of environmental interest. I. Mononuclear and polynucleararomatic hydrocarbons, J. Phys. Chem. Ref. Data 29 (2000) 41–130.

51] W. Shiu, K. Ma, Temperature dependence of physical and chemical prop-erties of selected chemicals of environmental interest. II. Chlorobenzenes,

[

. A 1364 (2014) 223–233 233

polychlorinated biphenyls, polychlorinated dibenzo-p-dioxins, and dibenzo-furans, J. Phys. Chem. Ref. Data 29 (2000) 387–462.

52] A. Robbat, N.P. Corso, P.J. Doherty, D. Marshall, Multivariate relationshipsbetween gas chromatographic retention index and molecular connectivityof mononitrated polycyclic aromatic hydrocarbons, Anal. Chem. 58 (1986)2072–2077.

53] F.R. Gonzalez, Application of capillary gas chromatography to studies on solva-tion thermodynamics, J. Chromatogr. A 1037 (2004) 233–253.

54] R.C. Castells, E.L. Arancibia, A.M. Nardillo, Regression against temperature ofgas chromatographic retention data, J. Chromatogr. 504 (1990) 45–53.

55] F. Aldaeus, Y. Thewalim, A. Colmsjö, Prediction of retention times of poly-cyclic aromatic hydrocarbons and n-alkanes in temperature-programmed gaschromatography, Anal. Bioanal. Chem. 389 (2007) 941–950.

56] K. Goss, R.P. Schwarzenbach, Empirical prediction of heats of vaporization andheats of adsorption of organic compounds, Environ. Sci. Technol. 33 (1999)3390–3393.

57] B. Koutek, J. Cvacka, L. Streinz, P. Vrkocova, J. Doubsky, H. Simonova, L. Feltl,V. Svoboda, Comparison of methods employing gas chromatography reten-tion data to determine vapour pressures at 298 k, J. Chromatogr. A 923 (2001)137–152.

58] K. Goss, R.P. Schwarzenbach, Linear free energy relationships used to evaluateequilibrium partitioning of organic compounds, Environ. Sci. Technol. 35 (2001)1–9.

59] W. Spieksma, R. Luijk, H.A.J. Govers, Determination of the liquid vapourpressure of low-volatility compounds from the kovats retention index, J. Chro-matogr. A 672 (1994) 141–148.

60] D. Gruber, D. Langenheim, J. Gmehling, W. Moollan, Measurement of activitycoefficients at infinite dilution using gas-liquid chromatography. 6. Results forsystems exhibiting gas–liquid interface adsorption with 1-octanol, J. Chem.Eng. Data 42 (1997) 882–885.

61] R.A. Alberty, C.A. Gehrig, Standard chemical thermodynamic properties ofalkane isomer groups, J. Phys. Chem. Ref. Data 13 (1984) 1173–1197.

62] R.A. Alberty, A.K. Reif, Standard chemical thermodynamic properties of poly-cyclic aromatic-hydrocarbons and their isomer groups. 1. Benzene series, J.Phys. Chem. Ref. Data 17 (1988) 241–253.

63] F.D. Rossini, Selected Values of Physical and Thermodynamic Properties ofHydrocarbons and Related Compounds, Comprising the Tables of the Amer-ican Petroleum Institute Research Project 44 Extant as of December 31, 1952,American Petroleum Institute, Carnegie Press, Pittsburgh, PA, 1953.

64] J. Beens, R. Tijssen, J. Blomberg, Prediction of comprehensive two-dimensionalgas chromatographic separations: a theoretical and practical exercise, J. Chro-matogr. A 822 (1998) 233–251.

65] S. Nigri, R. Oumeddour, B. Daoud, Determination of vapor pressures usingextrapolation of adjusted retention times, Chromatographia 61 (2005)167–171.

66] O. Engkvist, P. Borowski, A. Bemgard, G. Karlstrom, R. Lindh, A. Colmsjo, On therelation between retention indexes and the interaction between the solute andthe column in gas–liquid chromatography, J. Chem. Inf. Comput. Sci. 36 (1996)1153–1161.

67] C.D. Zeigler, M.M. Schantz, S. Wise, A. Robbat Jr., Mass spectra and retentionindexes for polycyclic aromatic sulfur heterocycles and some alkylated analogs,Polycyclic Aromat. Compd. 32 (2012) 154–176.

68] P.M. Antle, C.D. Zeigler, A. Robbat Jr., Retention behavior of alkylated polycyclicaromatic sulfur heterocycles on immobilized ionic liquid stationary phases,J. Chromatogr. A (2014), http://dx.doi.org/10.1016/j.chroma.2014.08.010 (inpress).

69] H. van Den Dool, P. Dec Kratz, A generalization of the retention index systemincluding linear temperature programmed gas-liquid partition chromatogra-phy, J. Chromatogr. 11 (1963) 463–471.

70] G. Castello, Retention index systems: alternatives to the n-alkanes as calibra-tion standards, J. Chromatogr. A 842 (1999) 51–64.

71] S. Bieri, P.J. Marriott, Dual-injection system with multiple injections for deter-mining bidimensional retention indexes in comprehensive two-dimensionalgas chromatography, Anal. Chem. 80 (2008) 760–768.

72] W.P. Eckel, T. Kind, Use of boiling point Lee retention index correlation for rapidreview of gas chromatography–mass spectrometry data, Anal. Chim. Acta 494(2003) 235–243.

73] W.E. May, S.P. Wasik, M.M. Miller, Y.B. Tewari, J.M. Brown-Thomas, R.N. Gold-

74] M.H. Abraham, Hydrogen-bonding 27. Solvation parameters for functionally-substituted aromatic-compounds and heterocyclic-compounds, fromgas–liquid-chromatographic data, J. Chromatogr. 644 (1993) 95–139.