Anal Bioanal Chem (2011) 401:2351–2360 DOI 10.1007/s00216-011-5247-1
REVIEW
Retention indices in comprehensive two-dimensional gas chromatography Carin von Mühlen & Philip J. Marriott
Received: 24 May 2011 / Revised: 5 July 2011 / Accepted: 6 July 2011 / Published online: 9 August 2011 # Springer-Verlag 2011
Abstract The identification of compounds by using gas chromatography (GC) in samples with significant complexity comprising a range of isomeric species, where characterization is based on peak retention times and mass spectra, generates uncertainty for the analyst. This leads to identification errors. The most reliable way to confirm the identification of each compound is based on authentic standard co-injection, which in several cases is economically prohibitive, and often unachievable in the time available for analysis. Retention index procedures are important tools to minimize misidentification of compounds in conventional chromatography. The introduction of comprehensive two-dimensional GC (GC× GC) for analysis of complex samples was a decisive step to increase the analytical capacity of chromatographic techniques. For many samples, the chromatographic resolution increase leads to quantitative expansion in the number of peaks identified, compared with conventional GC analysis. Notwithstanding this improved resolution, limitations still persist in correct peak identification, which suggests the use of retention indices may assist in supporting component identification in this important technique. In this work, approaches to use of the retention index in GC×GC are discussed, based on an evaluation of the literature in this area. Interpretation of Published in the special issue Comprehensive Multidimensional Separations with guest editors James Harynuk and Philip Marriott C. von Mühlen (*) Instituto de Ciências Exatas e Tecnológicas, Universidade Feevale, ICET, RS 239, 2755 Novo Hamburgo, Rio Grande do Sul 93352-000, Brazil e-mail:
[email protected] P. J. Marriott Centre for Green Chemistry, School of Chemistry, Monash University, Wellington Rd, Clayton, VIC 3800, Australia
effective chain length data for fatty acid methyl esters in the first and second dimensions is presented. Keywords Comprehensive two-dimensional gas chromatography . Linear-temperature-programmed retention index . Retention index . Kovats index . Fatty acid methyl esters . GC×GC retention index
Introduction The complexity of isomeric forms and various chemical classes of components within the broad range of terpenes in essential oil samples means that complete separation may be largely unachievable using 1D gas chromatography (GC). In addition, identifications based on mass-spectral similarities with a reference library cannot be easily or reliably deduced when a large number of isomers with similar mass spectra are present in a sample [1]. Owing to the presence of similar structures, the degree of saturation and heteroatomic homogeneity, the use of the retention index (I) as an auxiliary tool to perform compound identification in essential oils [1, 2], forensic toxicology [3] and other areas when investigating a sample of unknown composition has been established and is widely applied in these fields. Comprehensive 2D GC (GC×GC) has been successfully applied to essential oil analysis to obtain higher resolution between peaks than in conventional GC [4–7] as predicted by statistical-overlap theory [8], and demonstrated by Davis and Samuel [9, 10]. GC×GC [11–14] is based on an initial separation in a conventional chromatographic column coupled to a modulator, where the effluent from the first dimension (1D) is collected and fractionated in small portions in the modulator device and directed to a second column. The second column (second dimension, 2D) is a short column (around 1–2 m long), where a fast separation takes
2352
place in a few seconds, so that the column is ready to accept the next fraction of effluent delivered from the modulator. In this system, the peak capacity (1nc) from the 1D column is multiplied by that of the 2D column (2nc) to derive the total GC×GC capacity, nc,tot, which implies that a higher resolution is obtained than in 1D separations and in heartcutting multidimensional GC (MDGC) [15]. In MDGC a select zone or a few selected 1D regions are cut to a 2D column then analysed under relatively classic conditions such as using a long column. By contrast, GC×GC is a continual sampling process that operates with a modulation period (PM) faster than the peak elution width at the baseline on the 1D column, and normally uses a very short 2D column to ensure retention is within the time of the PM value. As a further benefit of GC×GC, several applications have been presented in the literature demonstrating the retention structure in GC×GC [16], with clear chemical classes separated in petrochemicals [13], volatile oils [7] and other samples. Combined with reproducible retention times in 2D (2tR), even with manual injection, the opportunity for component identification based on these data should be investigated [16]. In a theoretically orthogonal chromatographic system, the separations in each dimension can be considered as independent from each other. The elution temperature (Te) in 1D, taken as the temperature of injection into the 2D column, the relative average carrier gas flow velocities, the phase ratios of the two dimensions and the solute retention factors define the quantifiable elution time on the 2D column [16]. On this basis, the peak position in the 2D separation space is defined as two separate ordinates in the 1D and 2D columns. This should allow two independent retention values to be obtained, which ideally should be translated into independent index values [16]. The information power and value of analytical information available from several retention index databases which are available for 1D GC separations [2, 17] should not simply be ignored, since this is a resource that has been found to be useful in many studies. This affirmation is supported by a number of publications [5, 7, 16, 18–26] reporting the development and evaluation of a retention index system to be applied based on GC×GC information. As will be discussed in this article, it is possible to directly use traditional 1D retention indices in GC×GC, but a few limitations are noted. Other interesting alternatives aiming to increase the information power of two dimensions of separation are under study.
Retention index The most widely used retention index systems are those known as the Kovats index (I) and the linear-temperatureprogrammed retention index (LTPRI).
C. von Mühlen, P.J. Marriott
The Kovats index [27] was introduced in 1958, to be used in isothermal separations, and it can be calculated using Eq. 1: ! 0 0 log t log t RðiÞ RðnÞ Ixy ¼ 100n þ 100 ; ð1Þ 0 0 log tRðnþ1Þ log tRðnÞ where t’R is the adjusted retention time (peak retention time minus the retention time of an unretained peak, tM), i is the analyte, n is the number of carbons of the lesser retained alkane standard, and n+1 is the number of carbons of the more retained alkane standard bracketing analyte i. The x subscript and the y superscript refer to the phase on which the index is determined and the temperature of analysis. x has a profound effect on the I value, and y exhibits some temperature dependency of the I value. Equation 1 indicates that neighbouring alkanes have a difference of 100 index units; the index value of an alkane is simply 100n, and the component i value is effectively a logarithmic interpolation within this 100-unit base. The reference standards used in 1D can be n-alkanes, although other conventional retention index standards [fatty acid methyl esters (FAME), alcohols, ketones, etc.], may be employed depending on the nature of the stationary phase on the column and the properties of analytes of interest. For instance, with FAME, it is common to use the saturated fatty acids as reference standards, and an alternative representation of the index value in this case is referred to as the effective chain length (ECL) of the FAME compound. An unsaturated FAME will have an ECL that differs from that of the saturated FAME analogue of the compound. As the name linear-temperature-programmed retention index (LTPRI) describes, this index is used for separations with a linear temperature ramp. LTPRI was firstly discussed by van den Dool and Kratz [28] in 1963 and, for that reason, it is also known as van den Dool and Kratz index (Eq. 2): tRðiÞ tRðnÞ LTPRI ¼ 100n þ 100 ; ð2Þ tRðnþ1Þ tRðnÞ where tR is the retention time, and the other parameters are as in Eq. 1. The difference between Eqs. 1 and 2 is related to the logarithmic scale and the use of adjusted retention times for I and absolute times for LTPRI. As an example of retention index application [29], the retention index of compound A (Fig. 1) was calculated, using n-alkane standard co-injection in a GC with a flame ionization detector (FID). The retention times of alkanes close to compound A (1247 s) are 1180 s for C17 and 1370 s for C18. As the chromatographic analysis was performed using a linear temperature programme, the retention index was
Retention indices in comprehensive 2D GC
3e+007
C10 C
C20
C19 11
C15
C12 C14 C13
C16
C18
C17
A
2e+007
1e+007
1st Time(s)
500
750
1000
1250
1500
1750
2000
T IC
Retention time (min)
Fig. 1 Simulated chromatogram of a sample containing compound A, and fortified with an n-alkane homologous series. Linear temperature programme conditions were used. (Modified from von Mühlen [29])
calculated on the basis of Eq. 2. The retention time of compound A is given by tR(i), using n=17 and n+1=18. The calculated LTPRI was 1735. This calculated value can be compared with published retention indices. Thus, the LTPRI value of 1735 indicates that compound A is eluted between C17 and C18, based on the value of 1700, and the value of 35 indicates that the compound is eluted closer to C17 than to C18. It is not the intention of this review to present a comprehensive overview of the history of the retention index and all retention index systems, but its intention is to discuss the approaches used by researchers to adapt the concept to the GC×GC experiment.
Direct comparison of one-dimensional retention indices The first attempt to correlate retention indices with GC×GC in essential oil analysis was based on direct comparison of the van den Dool and Kratz retention index values obtained in 1D in a GC×GC separation with those obtained in conventional 1D separations [19]. This approach was applied to Panax (ginseng) extracts by using the same type of column and similar chromatographic conditions in 1D of the GC×GC system as in a 1D system. In that experiment, 13 compounds presented retention indices within a window of 7 units when compared with the reference 1D database, although there was no standard co-injection to confirm the I value prediction. Shellie and Marriott [5] also used the same approach as an auxiliary tool to tentatively identify 65 compounds in a Pelargonium graveolens essential oil, although the same group later developed other strategies to estimate retention indices with GC×GC, which will be presented later. The question of the best way to estimate the total retention time of individual peaks is relevant here. A convenient way is to measure the total retention time (1tR + 2tR) of each peak and assume that this approximately equates to the 1tR value.
Von Muhlen et al. [7] proposed the use of a 1D reference database combined with GC×GC/time-of-flight mass spectrometry experimental retention indices, using n-alkanes as reference standards. Using that approach, a single coinjection of a standard mixture of n-alkanes was performed with the sample, and LTPRI were directly calculated from the absolute retention time of the peak apex. 1D retention times were not separately calculated. These experimentally obtained LTPRI were plotted against literature reference 1D indices, showing a linear agreement for a mixture of 30 volatile compounds and for eucalyptus (Eucalyptus dunnii) leaf essential oil components. This provides simple supporting information for peak identification using an existing database, which can be supported by 2tR data for correct peak assignment. The metric used to calculate 1I from data recorded at the detector in the GC×GC experiment immediately suggests a point of confusion for GC×GC operation. Since the ‘peak’ in GC×GC is modulated into a number of fractions, how do we decide which fraction to use for calculation of I? If the largest peak is chosen, does this actually correspond to the centre point (mode) of the 1D peak maximum? And how is this modulation point determined for reference standards and for peak i? Fig. 2 illustrates this point. The circled zone for the peak in GC×GC analysis is a generated image comprising successive discrete modulations into 2D. As 2tR increases, peak zones become broader, as expected. It may be that the geometric mean position of the zone of peak i is a reasonable estimate of the 1tR value, accounting for holdup in the modulator, and the 2tR value, but this has not been verified. Choosing the time of the major modulated peak
Alkane positions in 2D
i
axis; 2I
Signal
4e+007
C9
2I i
2D
5e+007
2353
Alkane positions in 1D
2t M
1I i
1D
axis; 1I
Fig. 2 The calculation of retention indices in the first and second dimensions, where the first dimension is defined by a lineartemperature-programmed retention index scale, and the second dimension is according to the pseudo-isothermal retention scale. 1D first dimension, 2D second dimension
2354
means that the 1D retention time does not have better precision than the modulation period timescale of the experiment. Figure 2 also illustrates that under temperatureprogrammed analysis, the alkane positions on the 1D axis (shown as line positions) should increase approximately linearly. So the 1tR value of solute i and all the alkane solutes, when translated to the 1D axis, should give a reliable estimate of 1I based on Eq. 2. Solute and alkane positions on the 2D axis are more complex. Although we cannot contrive an experiment where all the series of alkanes shown as circles in the 2D plot are eluted at the same time, we can appreciate that this 2D axis defines an almost isothermal elution. For example, if the 2D time is 6 s, and the programme rate is 5 °C/min, ΔT is only 0.5 °C during the elution of these compounds. However (as shown later) it is possible to locate alkane—or reference—peaks along the 2D axis, either as derived mathematical curves, or indeed under isothermal (or pseudo-isothermal) conditions, or using similar approaches where reference compounds are almost instantaneously delivered to 2D. So 2I is calculated according to Eq. 1. Other research in this group has addressed the question of prediction of the correct 1D elution time of a compound, based on the sequence of modulated peaks, approximated by a Gaussian distribution on the first column [30, 31]. This should allow a much more precise estimation of 1tR and improve calculation of 1I. The 2Ii value should be more readily calculated, again assuming the times of alkanes and component i are correctly identified. However, the major concern is how do we contrive to have appropriate alkanes located both earlier and later than component i at the elution position of analyte i on the second column? At this point, it is useful to recognize that 1I values are based on a linear temperature ramp, and retention reference compounds must be introduced at time t=0, so LTPRI is relevant to this value. The fast modulation and data on 2D are acquired under almost isothermal conditions, and so isothermal measurements for 2I are more appropriate.
Second dimension retention indices To obtain a direct measure of retention indices in 2D, Beens et al. [18] presented theoretical and experimental results in 1998, and included use of LTPRI to predict the separation in 1 D. To predict the separation on the 2D column, conditions were considered to approximate several short isothermal GC runs, which allowed the use of the isothermal Kovats index on that dimension. The main difficulty in directly deriving I values for the 2D separation is to obtain isothermal retention reference data for a homologous series of standard compounds across the full range of temperatures used in the experiment, since a programmed temperature ramp is used for the separation in 1D.
C. von Mühlen, P.J. Marriott
Defining pseudo-isothermal retention data for reference compounds for the 2D column over the range of temperatures used for the experiment clearly demands an innovative experimental approach. Beens et al. [18] introduced nalkanes continuously into the system, using single injection into a cold programmed-temperature vaporizer (PTV) injector to deliver these standards into the GC column. With use of this approach, an elevated baseline response was obtained for all the alkanes introduced into the 1D column, but at the modulator, the alkanes collected were then modulated into discrete peaks on the 2D column. This results in a sequence of plots referred to as isovolatility curves by the authors, corresponding to alkane retentions in the 2D column. Isovolatility curves were presented for C7–C13, as illustrated in Fig. 3. Note that the alkanes that are bled in from the PTV soon become depleted, so, for instance, the C8 trace does not extend far past the C9 retention time. The onset of each curve corresponds to the retention time of the solute on the 1D column. Clearly, the 2D ‘coverage’ by the alkanes introduced is rather limited. The ‘exponential’ tail to lesser 2tR values is logical, because as temperature increases, solute retention logarithmically decreases. If a peak is in position P, as illustrated in Fig. 3, C7 and C8 alkanes could be used as reference standards to define the retention property of P at the elution temperature of this compound. With 2tR values for C7, C8 and P of 0.73 s, 1.90 s and 1.19 s, respectively, with all three elution peaks at 1tR = 18.3 min, an estimate of the I value can be made using logarithmic interpolation (e.g. 2I~760). This approach provided an interesting guide to the problem of experimental derivation of 2I index values, but was not applied to a multicomponent sample, which offers certain complexities to interpretation of I values, as mentioned below. To calculate I, it is necessary to estimate or determine the second column dead time, 2tM. Beens et al. [18] suggested three strategies. The first was based on vapour pressure calculations for n-alkanes at different temperatures in order to achieve theoretical retention factors. Since retention factors are smaller as the temperature increases, a plot of retention factor versus temperature was constructed, and the value of 2tM (0.5 s) was obtained by extrapolation. The second strategy was extrapolation of experimental isovolatility curves (Fig. 3), achieving a 2tM value of 0.44 s. In the third strategy, the carrier gas was contaminated with a small fraction of methane. Although the authors were working with the thermal sweeper modulator, which does not effectively modulate methane (just as for all thermal modulators), to determine 2tM, they observed a small variation in peak intensity above the baseline; this was sufficient for them to determine from it the dead time of 0.66 s. Independently of the strategy adopted—although with some uncertainty according to the method used—all necessary information to calculate 2I on 2D was obtained.
2.5 1.0
1.5
2.0
n-C13
n-C12
n-C11
n-C10
n-C9
n-C8
n-C7
0.5
P
0.0
Fig. 3 Isovolatility lines obtained with continuous injection of n-alkanes, and a hypothetical compound P. (Modified from Beens et al. [18])
2355
Second dimension retention time (s)
Retention indices in comprehensive 2D GC
10
20
30
40
50
60
First dimension retention time (min) With use of this strategy, very little coverage of the separation space with reference peak retention times can be observed in Fig. 3. This makes the range of compounds that locate between the reference standards rather limited, which is one of the criteria for calculation of the retention index of a compound under isothermal temperature conditions. This approach was applied only for a mixture of 12 alkanes, so it was not tested for wide applicability on the 2D phase, but it was important to demonstrate the application of the theory, and to predict peak positions in both dimensions. Vendeuvre et al. [20] applied a similar method as a reference to identify hydrocarbons in a more complex sample, such as kerosene. One conclusion presented by the authors was that the method gave poor precision for identification of individual compounds in complex samples, since isomers from a continuous band have slightly variable and different retention times, the precision of interpolation between the bands was not good, and small differences in retention indices can lead to a false identification for compounds that are just baseline-resolved. Different strategies were adopted later by several authors [16, 23, 26, 32] to obtain 2D retention indices, applying the above-mentioned principles outlined by Beens et al. [18]. Western and Marriott [16] used repetitive, timed, sequential injections of the reference compound standard mixtures, using a conventional split/splitless injector. Thus injections were made periodically into the injector and the reference compounds had to travel through the full length of the 1D column. This approach resulted in a better use of the separation space, allowing a greater range of compounds to fit between the times of the retention limits of the standard compounds. To predict the void time on the 2D column, the extrapolation of isovolatility curves was discussed as an
alternative, using the square of the logarithm of 2D-adjusted retention time versus the number of carbons (i.e. as shown in Eq. 1). This relationship was found to be linear, and the extrapolation of the line allowed estimation of the retention time for a compound with zero carbons (C0)—the 2tM value. The isovolatility curves were obtained with an nalkane standard mixture (C12–C22), using constant carrier gas flow. The alkanes had poor retention on the 2D column (for a polar 2D column phase) and so manipulating the system to give greater 2D retention for alkanes is not straightforward. In this case, many compounds have elution times beyond that of the most retained alkane. Thus, other homologous series were used to increase the 2D retention range: 2-methyl ketones (C11–C19) and saturated FAME (C14–C20). Refinements of this procedure were presented by the same authors in 2003 [32], using also more polar homologous series (carboxylic acids and alcohols). Another strategy to expand the retention range in 2D was developed using mathematical extrapolation of the experimentally obtained isovolatility curves and showed promising results. One limitation of the sequential injection procedure presented by these authors was the presence of several large solvent peaks in the chromatogram with the same elution times as important peaks. Zhu et al. [23] used the same approach presented by Western and Marriott [16] and Adcock et al. [31], but used constant pressure instead of constant flow. Constant pressure is commonly adopted with longer columns, owing to system limitations for maintenance of higher pressures inside the injector to regulate constant flow. However, it can also be said that under conditions of constant flow, the void time in 2D constantly decreases as temperature increases, possibly also causing problems for calculation of this
2356
fibre is introduced into the GC inlet some time before the C9 compound is eluted, and all subsequent alkanes are eluted in sequence from the 1D column. Peaks broaden on the 2D column as expected. A 5 s PM setting was used here. Once the positions of all isovolatility curves have been generated, they should be applicable to all subsequent analyses using this column set and conditions. Thus, all analyses can be referred to the 2D retention map of alkanes that has been set up for this column. Later in the analysis, the C15–C18 components in the C15–C20 injected mixture show little separation, so any analyte peaks that might appear in this region will have poorly defined retention index values; the likelihood of analytes being eluted here is unlikely since they will have greater polarity and retention than the lower molar mass alkanes. A mixture containing 25 standard fragrances was used as the test mixture with the isovolatility curves, as illustrated in Fig. 5. Dotted lines of best fit join the same alkane. It is observed that the lines of alkanes converge for lighter alkanes at higher Te, and this suggests that 2I value precision is likely to be poor for poorly retained solutes. Although this approach offered additional data owing to the combined information from both 2D columns, and the solvent-free injection approach improves data presentation, the authors also discussed the same limitations presented earlier: some compounds in the sample are outside the region of isovolatility curves, and compounds that wrapped around cannot be interpolated between neighbouring alkanes. On the basis of this experience, a second strategy was developed, to expand the separation space beyond the modulation period by running multiple isothermal plateau
C11
4
C12
C14
C13
C15
C16
C 17
C13
3
16
C18
17 C19 C20
15
12
C10
14
18
13
11
10 9
C9
8 7
2
19
0
10
20
30
40
C15-C20
C14-C20
C13-C 20
C12-C 20
C12-C 19
C11-C 18
C10-C 16
C11-C 17
C10-C14
C10-C 15
C9-C13
C9-C11
1
C9-C12
6
C9-C11
Second dimension retention time (s)
required parameter. Since tM is a critical parameter for all GC analyses, and should be no less so in GC×GC, a validated method for estimating it, by measurement or calculation, should be sought. When constant pressure was used, the void time was not the same for all temperatures, and another way to calculate the void time in 2D was necessary. The authors presented a mathematical equation to calculate the void time under those conditions, based on pressure, temperature, mobile phase viscosity and column dimensions. They also suggested the conversion of GC× GC retention indices over different column temperature conditions, showing calculation accuracies better than 1.0 retention index unit. This improved accuracy can be related to the use of constant pressure instead of constant flow. This development was applied to a cigarette essential oil sample, and to characterize flavour compounds in Chinese liquor Moutai [24]. The second application was performed with two different column sets, but retention indices were not presented in the experimental peak tables. Pang et al. [26] used n-alkanes and saturated FAME as homologous series to generate expanded isovolatility curves with a (nonpolar×polar) column set. In this work, the continuous injection method described by Western and Marriott [16] and constant carrier gas pressure and void time calculations in 2D presented by Zhu et al. [23] were used. This method was applied to search for volatile unknowns in a tobacco leaf extract fraction using a GC× GC-FID system. To overcome the limitation imposed by the presence of significant solvent peaks spread throughout the separation space, Bieri and Marriott [22] proposed the use of solidphase microextraction (SPME) to obtain a solvent-free introduction of the injected standards (alkanes were used for 2tR data) into the GC columns, eliminating solvent peaks. In addition, the authors used a GC×GC system that splits the primary column effluent towards two 2D columns, to increase identification capabilities based on relative retention since two sets of index values are available. A polar column was used in 1D (SolGel-Wax) using n-alcohols (C6–C20, C22) as retention standards (LTPRI) in the first column for 1I, and both a nonpolar (BPX5) and a mediumpolarity (BP10) column for 2D, using n-alkanes (C9–C20) as reference standards (2I). An example chromatogram generated by this strategy is shown in Fig. 4. The absence of solvent in SPME sampling can be appreciated—otherwise a solvent peak would arise for each injected sample. The peak zones for the alkanes are favourable, and superimposed on the plot is a line for the alcohols (C6–C22) introduced at time t=0, which allows these to be used for 1I values. Note also that the alkanes can be selected to give good coverage of the maximum 2tR time (5 s) in this example, so the whole 2D space is essentially covered. Referring to Fig. 4, the fifth SPME-sampled alkane mix comprises C9–C13. The SPME
C. von Mühlen, P.J. Marriott
50
22 20
60
First dimension retention time (min)
Fig. 4 Contour plot for a sequential injection of n-alkanes (solid lines) and alcohols (dotted lines) using a polar column in the first dimension and a nonpolar column in the second dimension. (Reprinted with permission from Bieri and Marriott [22]. Copyright 2011 American Chemical Society)
Retention indices in comprehensive 2D GC
Second dimension retention time (s)
C11
C16
C17 C18
C13
1
4
C15
C14
C12
C19 C20
C10
5a
19 6
4
2
3
5b
14
17b 21a 22b 22a
7
C9
9
11 13 15
16
23 24
25
21b
18
12
3
2
17a
20
10
1 8*
0 0
10
20
30
40
50
60
First dimension retention time (min)
Fig. 5 Isovolatility lines from the chromatogram presented in Fig. 4, for an allergenic compound sample. (Reprinted with permission from Bieri and Marriott [22]. Copyright 2011 American Chemical Society)
analyses with a fast temperature ramp between each plateau temperature. With this approach, the 2tR times were extrapolated for subsequent homologues beyond those in the injected series, and were reasonably accurately predicted. Thus, even when a 5 s PM value was used, it was possible to experimentally obtain isovolatility curves with retention times up to 11 s, allowing the interpolation of wrapped-around peaks. Comparison between LTPRI calculated using a single 1D column and with a 2D column coupled to the 1 D column, but with the modulator off, demonstrated a small deviation between results (maximum of 14 LTPRI units). The precision of the retention index obtained in that way is smaller than the precision obtained for the 1D index, as pointed out by the authors. The poor retention of some reference compounds on the 2D axis results in a small separation space between homologous alkanes, and leads to difficulty in accurate interpolation of retention. On the basis of the experience described above, Bieri and Marriott [33] proposed a double injection system, using the second injector of the chromatograph to directly introduce n-alkanes into the 2D column (Fig. 6). An empty transfer line column segment was introduced from the second injector directly to the modulator, bypassing the 1D column. Two Y-piece connectors were used to direct flow either to the first detector to generate 1I independent of 2D or via the modulator to generate modulated peaks, as illustrated in Fig. 6. The alkanes no longer need to traverse the 1D column, and so they can be added at the start of the 2 D column easily and rapidly. It is noteworthy that most of these studies were directed to applying the GC×GC method with added reference compounds under precisely the same conditions as used for the GC×GC experiment, but this can present technical implementation challenges, although these are not insurmountable.
With this system, 1D LTPRI values were obtained via the transfer line FID response. These values had minimal variations in index data (less than 1 index unit). On the other hand, LTPRI obtained after the 2D column had index variations of the order of 17 units. Experimentally, n-alkanes were injected into the first column at time t=0 for 1I, and successive solventless SPME injections were made into the second injector for 2I data. When the SPME fibre was inserted into the injector, n-alkanes were instantaneously desorbed and almost immediately introduced into the 2D column. With this system it was possible to easily introduce more reference compounds, allowing extrapolation of reference retention times. This strategy increases the elution range covered by reference compounds. There is an interesting possibility for deriving temperaturevariable retention indices for compounds by making repeated injections of a compound into the system via SPME and the second injector, and constructing an isovolatility curve for that compound. Any change in the relative positions of the alkane reference and solute isovolatility curves will mean that the solute exhibits temperature variability in its retention index. This can be done in a single temperature-programmed analysis, as opposed to multiple isothermal analyses. This approach also suggests a new study, which was alluded to by Bieri and Marriott [33], where retention indices can be readily derived for the 2D column in MDGC. By use of the second injector, and preferably with SPME sampling, reference compounds can be introduced directly into the 2D column, where they can be cryotrapped and released at the prevailing oven temperature into the 2D column. Solute(s) from the first column can be co-trapped in the cryotrap and also simultaneously introduced into the 2D column. In a recent study, Zhao et al. [34] implemented an approach not unlike that of Bieri and Marriot, where different retention times were established for 2tR values of alkanes under different temperature programme conditions.
Inj 1
Det 2
Det 1
Inj 2
TL 2D
M
Y
Y
5
2357
1D
2 x 3-way connectors
Fig. 6 Comprehensive 2D gas chromatography (GC×GC) modified system to obtain independent retention indices in the first and second dimensions proposed by Bieri and Marriott. TL transfer line, Inj injector, Det detector, M modulator. (Redrawn on the basis of the schematic concept in [33])
2358
6
205
2D
ECL
200
tR (s)
5
195
C20:0
4
C20:0
3
190 185
2
180
1
175 1
0
2
3
4
5
7
6
170
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Injection time (min) Fig. 7 Example of fatty acid methyl ester (FAME) compounds (C16:0-C20:0) injected at incrementally increased isothermal temperatures into the second dimension column of a GC×GC experiment. This establishes the retention times of the FAME in a temperatureprogrammed GC×GC experiment at the respective temperatures used for the reference data. ECL effective chain length
Response (pA)
As temperature increases, FAME are eluted progressively earlier. With these data in hand, a GC×GC analysis of a FAME sample can be conducted (Fig. 8) and the reference data can be overlaid on the GC × GC 2D plot, choosing the appropriate Te for the reference data. This allows the nowfamiliar isovolatility curves to be constructed. Interpolation allows ECL values to be estimated for all the FAME components. As expected, saturated FAME should have a whole-number ECL (i.e. nC18 FAME should have ECL of 18.00), and the overlaid curve serves as a confirmation of the correspondence of data. Selected ECL data are given in Table 1. Some comments will assist in understanding the C16:0
80 60 40 20
C18:0
C17:0
C20:0
5 C19:0 C20:0
3 2
C18:0
C16:0
2D
(s)
C17:0
ECL (2ECL)
4
R
FAME compounds are widely interpreted on the basis of index values [37, 38], which are often referred to as ‘effective chain lengths’ (ECL). This is based on saturated FAME, where the index for saturated FAME is simply the carbon number; the ECL is calculated in an analogous manner to I values, but is not multiplied by the factor of 100. Usually, on a nonpolar phase the saturated FAME is eluted later than its unsaturated analogues, whereas it is eluted first on a polar phase [37, 38]. ECL values may be calculated in the same manner as detailed already for alkane indices, but they refer strictly to FAME. In the study of Yang [39, 40] saturated FAME standards were introduced by solventless SPME sampling in the same instrument described in Fig. 6. Figure 7 presents the results for repeat injections of a C16–C20 reference mix at various isothermal settings in the GC×GC system. The modulator was programmed to release the introduced standards at each temperature setting.
210
C20:0
2t
Fatty acid methyl ester retention indices
Oven T °C
7
2D
This allowed a model for the observed 2tR properties to be derived, termed a 2tR–Te function. A plot of 2D column T versus adjusted 2tR values is very much akin to the plots shown in Fig 5. The method allowed estimation of 2I values for a MegaMix sample of analytes, for each of six different temperature-programmed experiments. Seeley and Seeley [25] proposed a mathematical model to predict retention times in GC×GC using 1D I data as reference values. According to these authors, the precision (relative retention time in 2D) obtained with this model (10%) was slightly lower than that achieved with previous models (5% [18], 10% [20] and 2% [21]). On the other hand, it was possible to predict relative retention times in 2 D using this model, without the use of isovolatility curves. This model was tested for 139 volatile compounds, showing consistent results, although real samples were not tested in this work. Recently, Seeley et al. [35] used another model with solvation parameters and physical–chemical information about compounds, and chromatographic column information [36], aiming to predict the position of hydrocarbons within the separation space in GC×GC. Seeley et al. [35] used this model to predict retention indices of compounds in both 1D and 2D columns. The model was tested for 54 compounds in four different stationary phase combinations. Deviations between calculated peak positions and experimental retention times were of the order of 1% for 1 D and 5% for 2D. The most important limitation of this model, according to the authors, was the existence of limited information about chemical compounds in the literature, which limits the application of the model. The model did not completely take into account changes in the retention order as a function of the temperature.
C. von Mühlen, P.J. Marriott
1 0 25
30
35 1t
40
45
50
R (min)
Fig. 8 GC×GC result for a FAME sample superimposed with saturated FAME reference data acquired at the elution temperature at which reference data were obtained. This defines the isovolatility curves of the saturated FAME
Retention indices in comprehensive 2D GC
2359
Table 1 First dimension (1D) and second dimension (2D) retention indices based on saturated fatty acid methyl esters (FAME) and calculated as effective chain length (ECL) values determined for a range of FAME compounds
unsaturated FAME are eluted progressively later than their unsaturated analogues [37].
Compound
Conclusions
GC×GC ECL BPX5 1 a D
a
FCL BP20 Db
b
2
1
Da
2
Db
C16:1 C16:0 C17:1 C17:0 C18:3n6 C18:3n3
15.77 15.99 16.80 17.02 17.77 17.53
16.10 15.91 17.30 16.87 19.17 18.93
−0.23 −0.01d −0.20 0.02c −0.23 −0.47
0.10 −0.09c 0.30 −0.13c 1.17 0.93
C18:2n6c C18:2n6t C18:1n9c C18:1n9t C18:0
17.70 17.80 17.78 17.84 18.04
18.59 18.67 18.14 18.18 17.96
−0.30 −0.20 −0.22 −0.16 0.04d
0.59 0.67 0.14 0.18 −0.04c
GC×GC comprehensive 2D gas chromatography, FCL fractional chain length a
Data according to the van den Dool and Kratz method
b
Data according to the Kovats method
c
These values should be zero, but fitting the curves to the exact peak position in the mixture leads to some inaccuracy, and deviation from the expected whole-number ECL value
data in Fig 8 and Table 1. The saturated FAME in the mixture should lie on the isovolatility curves for their respective reference compounds (i.e. C16:0 in the sample should lie on the curve for C16:0). This is approximately so (Table 1), but the data are not matched exactly, so an expected residual error arises. The extent of difference from the corresponding saturated Cn:0 FAME position is less than 0.04 for 1D ECL and less than 0.13 for 2D ECL. Given the lower precision for 2ECL data (due to the narrow interpolation range) and the inaccuracy in translating the isovolatility curves to the GC×GC experiment, some increased uncertainty is expected. C16:1 and C17:1 are eluted with smaller 1tR than their Cn:0 analogues, and with marginally greater 2tR than their respective Cn:0 isovolatility lines, so have smaller 1D ECL, but slightly larger 2D ECL. Of the C18 FAME, C18:3n3 is eluted earliest (lowest 1tR; 1D ECL 17.53) but does not have the greatest 2D ECL (18.93). This is found for C18:3n6, with 2D ECL of 19.17 (note therefore that the elution position of C18:3n6 is a little higher than the C19:0 isovolatility curve). Data are also given in terms of the fractional chain length, which is the difference in ECL values between an unsaturated FAME and the target FAME with the same number of carbons. The results seem logical and in agreement with expected ECL data, for instance with the observation that on a polar-phase column,
A variety of studies have presented options and directions for the use of retention indices as an analytical tool that aids peak identification, by deriving retention information obtained for 1 D and 2D columns in GC×GC. The alternatives presented different possibilities, with either mathematical or instrumental approaches, depending on the strategy adopted. Retention models based on solvation parameters and analyte properties can also be used as important tools to build databases and identification software. The use of isovolatility curves permits correlation of existing 1D index databases and retention information in 2D for GC×GC, with somewhat variable analytical accuracy. However, this approach still presents a time-consuming data acquisition process, and is clearly not as practical as the method for generation of conventional 1D retention indices—essentially just (co)injection of a mixture of reference standards. For laboratories used to working with retention indices on a daily basis, and therefore for whom the I value is an important identification or confirmatory tool, it is important to provide an interpretation of how the GC×GC technique can be used to acquire I data, the instrumental approaches and modifications required to derive such values, and whether first principles (e.g. solvation data) are reliable methods to determine 2I. Although an accurate experimental retention index value is unlikely on the 2D column, owing to the small retention differences of reference compounds, it will be interesting to gauge if the combined availability of 1I and 2I might provide an information/identification power not possible using 1D GC. Future work will be required to establish this. Once a 2D ‘map’ of the retention plane in GC×GC is established, it should be possible to overlay experimental data on this map to obtain the required 2I values. If it is reduced to a simply computer-based metric, this will likely be a convenient and effective way to estimate 2I. Experimental methods described for generation of indices in GC×GC should be suited to MDGC 2I calculations, and simplified experimental methods to obtain variable-temperature I values have been proposed. On the basis of the alternatives presented here, development of user-friendly methods or software-integrated models and databases for retention indices in both dimensions in GC× GC requires some innovative approaches that are still emerging as practical methods, and new contributions in this area are expected soon. The aim will be to bring to GC×GC all the inherent value and knowledge acquired over the decades where retention indices have played an important role in GC. It is anticipated that improved identification of compounds will result for this new and exciting mode of GC.
2360
References 1. Sandra P, Bicchi C (1987) Capillary gas chromatography in essential oil analysis. Huethig, New York 2. Adams RP (1989) Identification of essential oils by ion trap mass spectroscopy. Academic, New York 3. Zeeuw RA, Franke JP, Maurer HH, Pfleger K (1992) In: Report XVIII of the DFG Commission for Clinical-Toxicological Analysis. Special issue of the TIAFT Bulletin. Wiley-VCH, Weinheim, p 409 4. Dimandja JMD, Stanfill SB, Grainger J, Patterson DG Jr (2000) J High Resolut Chromatogr 23:208–214 5. Shellie R, Marriott PJ (2003) Analyst 128:879–883 6. Zhu S, Lu X, Dong L, Xing J, Su X, Kong H, Xu G, Wu C (2005) Anal Chim Acta 545:224–231 7. von Muhlen C, Zini CA, Caramão EB, Marriott PJ (2008) J Chromatogr A 1200:34–42 8. Davis JM, Giddings JC (1983) Anal Chem 55:418–424 9. Davis JM (2005) J Sep Sci 28:347–359 10. Davis JM, Samuel C (2000) J High Resolut Chromatogr 23:235– 244 11. Marriott PJ, Shellie R (2002) Trends Anal Chem 21:573–583 12. Ong RCY, Marriott PJ (2002) J Chromatogr Sci 40:276–291 13. von Mühlen C, Zini CA, Caramão EB, Marriott PJ (2006) Quim Nova 29:765–775 14. von Muhlen C, Zini CA, Caramão EB, Marriott PJ (2007) Quim Nova 30:682–687 15. Simmons MC, Snyder LR (1958) Anal Chem 30:32–35 16. Western RJ, Marriott PJ (2002) J Sep Sci 25:832–838 17. Castello G (1999) J Chromatogr A 842:51–64 18. Beens J, Tijssen R, Blomberg J (1998) J Chromatogr A 822:233– 251 19. Shellie R, Marriott PJ, Huie CW (2003) J Sep Sci 26:1185–1192
C. von Mühlen, P.J. Marriott 20. Vendeuvre C, Bertoncini F, Thiébaut D, Martin M, Hennion MC (2005) J Sep Sci 28:1129–1136 21. Lu X, Kong HW, Li HF, Ma CF, Tian J, Xu GW (2005) J Chromatogr A 1086:175–184 22. Bieri S, Marriott PJ (2006) Anal Chem 78:8089–8097 23. Zhu SK, Lu X, Qiu YQ, Pang T, Kong HW, Wu CY, Xu GW (2007) J Chromatogr A 1150:28–36 24. Zhu SK, Lu X, Ji KH, Guo KF, Li YL, Wu CY, Xu GW (2007) Anal Chim Acta 597:340–348 25. Seeley JV, Seeley SK (2007) J Chromatogr A 1172:72–83 26. Pang T, Zhu S, Lu X, Xu GW (2007) J Sep Sci 30:868–874 27. Kovats E (1958) Helv Chim Acta 1915–1932 28. van den Dool H, Kratz PD (1963) J Chromatogr 11:463–471 29. von Mühlen C (2009) Sci Chromatogr 1(3):21–28 30. Xie LL, Marriott PJ, Adams M (2003) Anal Chim Acta 500:211–222 31. Adcock JL, Adams M, Mitrevski BS, Marriott PJ (2009) Anal Chem 81:6797–6804 32. Western RJ, Marriott PJ (2003) J Chromatogr A 1019:3–14 33. Bieri S, Marriott PJ (2008) Anal Chem 80:760–768 34. Zhao Y, Zhang J, Wang B, Kim SH, Fang A, Bogdanov B, Zhou Z, McClain C, Zhang X (2011) J Chromatogr A 1218:2577–2583 35. Seeley JV, Libby EM, Edwards KAH, Seeley SK (2009) J Chromatogr A 1216:1650–1657 36. Arey JS, Nelson RK, Xu L, Reddy CM (2005) Anal Chem 77:7172–7182 37. Harynuk J, Wynne PM, Marriott PJ (2006) Chromatographia 63: S61–S66 38. Harynuk J, Vlaeminck B, Zaher P, Marriott PJ (2006) Anal Bioanal Chem 386(3):602–613 39. Yang MY (2007) Simultaneous dual column retention indices of fatty acid methyl esters in comprehensive two-dimensional gas chromatography. Honours thesis, RMIT University 40. Yang MY, Marriott PJ Unpublished results