S I p c A D a A A K G D Q 1 ( o d f fi t R n o r s t G t d a u m f ( p t i m 0 d Journal of Chromatography A, 1255 (2012) 190– 195 Contents lists available at SciVerse ScienceDirect Journal of Chromatography A j our na l ho me p ag e: www.elsev ier .com/ locate /chroma hort communication ntegration parameters and their effects on quantitative results with two-step eak summation quantitation in comprehensive two-dimensional gas hromatography . Paulina de la Mata, Katie D. Nizio, James J. Harynuk ∗ epartment of Chemistry, University of Alberta, Edmonton, AB T6G2G2, Canada r t i c l e i n f o a b s t r a c t rticle history: vailable online 23 February 2012 eywords: C×GC Quantification of comprehensive two-dimensional gas chromatographic (GC×GC) data in a commercial software package is examined. ChromaTOF® (Leco Instruments) includes a smoothing step prior to data integration. Improper choice of integration parameters can introduce errors in excess of ±10% and as high as 60% in the total peak area. Herein we demonstrate the critical nature of the smoothing parameters in this software, particularly the expected peak width in the second dimension (2w) which should be C pro ata processing uantification errors verified as part of a QA/Q . Introduction Comprehensive two-dimensional gas chromatography GC×GC) is widely applicable for the analysis of complex mixtures f small organic molecules and has been applied in a number of ifferent fields, such as petroleum [1], environmental analysis [2], ood and flavours [3], metabolomics[4], among others. GC×GC was rst demonstrated by Liu and Phillips in 1991 [5] and employs wo GC columns of different selectivity, coupled by a modulator. eaders unfamiliar with the technique are referred to any one of umerous reviews in the literature [6–13]. GC×GC provides enhanced sensitivity and detectability of peaks ver 1D GC, in the first part due to the secondary separation which emoves interfering chemical signals (decreases noise), and in the econd part due to the zone compression effected by the modula- or (this effect is lessened in pneumatic modulators). Consequently, C×GC can often be used to quantify analytes at levels lower than hose accessible to 1D GC. However, the modulation process intro- uces new sources of error that must be considered [14]. The nalyte peak is fractionated into a number of slices during the mod- lation process, resulting in a base peak (i.e. the tallest peak) and ultiple sub-peaks. The results of integration depend on several actors beyond the mass of analyte present: the modulation ratio MR), defined as the ratio of 1w to the modulation period (PM) [15], hase (�) or position of the analyte peak relative to the modula- ion events [16], shifts in 1D retention time (1tR), and peak width n both the first and second dimensions (1w and 2w). One of the ost commonly used software packages for GC×GC quantification ∗ Corresponding author. Tel.: +1 780 492 8303; fax: +1 780 492 8231. E-mail address:
[email protected] (J.J. Harynuk). 021-9673/$ – see front matter © 2012 Elsevier B.V. All rights reserved. oi:10.1016/j.chroma.2012.02.044 tocol. © 2012 Elsevier B.V. All rights reserved. is ChromaTOF® (Leco Instruments) which relies on a two-step inte- gration method. The purpose of this work is to demonstrate the critical nature of the integration parameter of “expected 2w” which is much more important than indicated by the software documen- tation. 2. Experimental 2.1. Simulated GC×GC data Simulated GC×GC chromatograms were generated by a lab- oratory written model using MATLAB (v.7.120.635; Mathworks, Natick, USA). This model generates a 1D Gaussian peak of specified width (here 12 s) and simulates its modulation to generate a series of 2D Gaussian peaks of specified width (here varied depending on the experiment). For this work, MR = 4 and � = 0◦ were used to simulate the data. Normally distributed random noise of a magnitude similar to that observed in typical GC×GC data was added to the baseline. The data were then converted into a format that could be imported into ChromaTOF® (v.4.33; Leco Instruments, St. Joseph, MI) using another custom-written MATLAB script. 2.2. Real GC×GC data 2.2.1. Materials and reagents Alkane standards dodecane and tetradecane were purchased from Eastman Organic Chemicals (Rochester, NY), undecane from Fisher Scientific (Ottawa, ON) and tridecane from Sigma Aldrich (Oakville, ON). A series of alcohols ranging from 1-undecanol to 1-tetradecanol were purchased from Sigma Aldrich with the excep- tion of 1-dodecanol (Eastman). A series of ketones ranging from dx.doi.org/10.1016/j.chroma.2012.02.044 http://www.sciencedirect.com/science/journal/00219673 http://www.elsevier.com/locate/chroma mailto:
[email protected] dx.doi.org/10.1016/j.chroma.2012.02.044 A. Paulina de la Mata et al. / J. Chromatogr. A 1255 (2012) 190– 195 191 Table 1 Simulated data (italics represent when actual 2w matches 2w used for data processing, bold values represent when absolute value %D is greater than 10%). Simulated concentration Actual 2w (s) Data processing 2w (s) No. of peaks detected Total peak area Total peak area absolute D% S/Nb S/Nb absolute D% High 1 1 3 75 759 0 137 100 0.5 3 74 522 1.6 106 53.6 0.3 2 61 491 18.8 83 20.3 0.1 1 46 760 38.3 61 11.6 0.1 1 0 0 100 0 100 0.5 1 51 924 31.2 696 17.4 0.3 3 75 726 0.3 700 18.0 0.1 3 75 515 0 593 0 Med 1 1 3 38 189 0 69 0 0.5 2 32 138 15.8 54 21.4 0.3 1 24 541 35.7 42 39.7 0.1 0 0 100 0 100 0.1 1 0 0 100 0 100 0.5 2 32 101 15.7 346 18.1 0.3 3 38 381 1 346 18.1 0.1 3 38 074 0.8 293 0 Low 1 1 1 12 729 0 36 0 0.5 1 13 409 5.3 28 20.2 0.3 1 11 658 8.4 22 38.5 0.1 1 1920 84.9 4 87.9 0.1 1 0 0 100 0 100 0.5 2 16 596 12.7 174 18.5 0.3 2 16 141 15.1 175 19.0 0.1 3 19 012 0 147 0 S 2 S t c p t k t C v 2 w I m P c o o t t R t o l o t 2 C e s i ( 0 0 f /Nb, signal to noise ratio of the modulated base peak. -undecanone to 2-tridecanone (C11–C13) purchased from Fisher cientific and a series of alkyl phosphates comprising trimethyl-, riethyl-, triisopropyl-, tripropyl-, and tributyl phosphate pur- hased from Sigma Aldrich. Trihexyl- and trioctyl phosphate were urchased from Alfa Aesar (Ward Hill, MA). Two sample mix- ures were prepared: one containing the alkanes, alcohols, and etones in concentrations ranging from 10 to 100 �g mL−1 and he other containing the alkyl phosphates at 50 and 125 �g mL−1. HROMASOLV® grade hexane (Sigma Aldrich) was used as a sol- ent. .2.2. Instrumental conditions Analysis of the mixture of alkanes, alcohols, and ketones as conducted on a Pegasus 4D-GC×GC-TOFMS system (Leco nstruments) equipped with a liquid nitrogen cryogenic quad jet odulator. A 10 m × 0.18 mm, 0.18 �m RTX-5 (Restek, Bellefonte, A) column in 1D and a 1.5 m × 0.10 mm, 0.1 �m Rxi-17 (Restek) olumn in 2D. All injections were performed in triplicate using 1 �L f sample with an Agilent 7683B Series autosampler and a split ratio f 100:1. The inlet temperature was 250 ◦C. Helium was used as he carrier gas with a flow rate of 1.1 mL min−1. The primary oven emperature program was 60 (held for 1 min)–160 ◦C at 3 ◦C min−1. elative to the primary oven, the secondary oven was programmed o have a constant offset of +15 ◦C and the modulator a constant ffset of +30 ◦C. A modulation period of 3 s was used. Data were col- ected from m/z 30–250 at a rate of 100 spectra s−1. A solvent delay f 110 s was used. The detector voltage was −1275 V, the ion source emperature was 200 ◦C, and the MS transfer line temperature was 40 ◦C. Analysis of the alkyl phosphates was conducted using a onsumable-free cryogenic quad jet GC×GC system (Leco) quipped with a CFT Deans switch (Agilent Technologies, Missis- auga, ON) for post-column detector selection and both a flame onization detector (FID) and a nitrogen phosphorus detector NPD). The column configuration consisted of a 10 m × 0.18 mm, .2 �m RTX-5 (Restek) column in 1D and a 0.5 m × 0.18 mm, .18 �m Rxi-17 (Restek) column in 2D. All injections were per- ormed in triplicate using 1 �L of sample with an Agilent 7683B Series autosampler and a split ratio of 50:1. The inlet temper- ature was 250 ◦C. Helium was used as the carrier gas with a ramped pressure program of 15.56 psi (held for 3.50 min)–29.27 psi at 0.31159 psi min−1 calculated using the Agilent Deans Switch Cal- culator software. The primary oven temperature program was 40 (held for 3.50 min)–260 ◦C at 5 ◦C min−1. The secondary oven was programmed to have a constant offset of +25 ◦C and the modulator a constant offset of +40 ◦C relative to the primary oven. A modu- lation period of 3 s was used for the first 1098s of the run while a modulation period of 2 s was used for the remainder. Initially, 2D column effluent (i.e. solvent) was directed to the FID by the Deans switch. After 3.75 min, effluent was directed to the NPD for detec- tion of the alkyl phosphates. The pneumatic zone controlling the Deans switch had a ramped pressure program of 4.44 psi (held for 3.50 min)–6.68 psi at 0.0509 psi min−1 calculated using the Agilent Deans Switch Calculator. Both detectors were kept at 325 ◦C. Data were acquired at a rate of 100 Hz. 2.3. Software Both the simulated and real data were processed using ChromaTOF® (v.4.33; Leco). The baseline offset was set through the middle of the noise (0.5), the minimum S/N for the base peak and the sub-peaks were set at 10 and 3, respectively, for the pro- cessing of real data. For the processing of simulated data, these parameters were set at 3. Data were auto smoothed by the soft- ware. The setting for expected 1w was 12 s for the simulated data and alkyl phosphates and 9 s for the mixture of alkanes, alcohols, and ketones. 3. Results and discussion The two-step integration algorithm used by ChromaTOF® relies on the identification of individual 2D peaks which are subsequently integrated and combined with other sub-peaks deemed to originate from the same compound on the basis of retention coordinates and, if available, mass spectral matching criteria. As part of the data interpretation it is recommended to smooth the data and to 192 A. Paulina de la Mata et al. / J. Chromatogr. A 1255 (2012) 190– 195 Table 2 Real data (italics represent when actual 2w matches 2w used for data processing, bold values represent when absolute value %D is greater than 10%). Concentration (ppm) Compound Actual 2w (s) Data processing 2w (s) No. of peaks detected (1/2/3)a Avg total peak area Avg total peak area absolute D% Avg S/Nb Avg S/Nb absolute D% 100 Undecane 0.3 1.0 2/2/3 35 416 0.3 557 31.0 0.5 3/3/3 35 540 0.7 509 19.7 0.3 3/3/3 35 304 0.0 425 0.0 0.1 3/3/3 35 006 0.8 325 23.6 1-Undecanol 0.3 1.0 2/2/2 5856 14.3 63 28.7 0.5 2/2/3 5581 8.9 58 19.0 0.3 3/2/2 5125 0.0 49 0.0 0.1 2/2/2 3921 23.5 38 23.0 2-Undecanone 0.3 1.0 3/2/1 35 435 3.8 364 35.6 0.5 2/3/3 34 639 1.5 323 20.7 0.3 3/3/3 34 124 0.0 268 0.0 0.1 3/3/3 29 776 12.7 198 26.0 125 Trimethyl phosphate 1 1.0 9/9/10 5 173 276 0.0 1335 0.0 0.5 9/8/9 4 939 761 4.5 1081 19.0 0.3 9/9/9 4 439 675 14.2 892 33.2 0.1 7/6/6 2 932 964 43.3 685 48.7 Tripropyl phosphate 0.5 1.0 4/5/4 9 749 072 0.5 6958 20.8 0.5 4/4/4 9 702 971 0.0 5758 0.0 0.3 3/3/3 9 610 073 1.0 4789 16.8 0.1 3/3/3 9 079 508 6.4 3686 36.0 Trioctyl phosphate 0.3 1.0 6/5/7 5 989 347 0.7 5098 38.3 0.5 7/7/9 6 001 377 0.9 4382 18.8 0.3 7/7/6 5 950 140 0.0 3687 0.0 0.1 6/6/7 5 834 218 1.9 2856 22.6 50 Undecane 0.3 1.0 2/2/3 14 310 7.6 267 19.8 0.5 2/2/3 14 512 6.3 259 16.2 0.3 3/3/3 15 488 0.0 223 0.0 0.1 2/3/3 14 707 5.0 172 22.5 1-Undecanol 0.3 1.0 1/2/1 12 463 15.0 134 47.4 0.5 3/2/1 12 237 12.9 121 32.8 0.3 2/1/1 10 837 0.0 91 0.0 0.1 1/1/1 9364 13.6 73 19.5 2-Undecanone 0.3 1.0 3/2/2 2040 57.3 24 33.1 0.5 3/3/3 1901 46.5 21 21.3 0.3 3/3/3 1297 0.0 18 0.0 0.1 3/2/3 969 25.3 12 30.9 Trimethyl phosphate 1 1.0 8/7/8 3 778 000 0.0 1067 0.0 0.5 10/8/5 3 506 247 7.2 866 18.8 0.3 6/5/5 3 009 380 20.3 715 33.0 0.1 6/6/6 2 114 811 44.0 548 48.6 Tripropyl phosphate 0.5 1.0 4/4/5 4 832 254 1.4 2924 19.5 0.5 4/4/4 4 764 367 0.0 2448 0.0 0.3 4/4/5 4 610 636 3.2 2041 16.6 0.1 3/4/5 4 214 884 11.5 1567 36.0 Trioctyl phosphate 0.3 1.0 5/6/6 2 947 152 1.2 2677 37.0 0.5 6/6/6 2 940 348 1.0 2316 18.5 0.3 6/6/6 2 911 423 0.0 1954 0.0 0.1 5/5/5 2 830 427 2.8 1511 22.7 S d run a s s w c t w [ fi c d t d M t B o /Nb, Signal to noise ratio of the modulated base peak. a 1/2/3: Number of peaks detected (base peak plus sub-peaks) for run 1, run 2, an utomatically select the number of data points to be used for the moothing window [17]. The smoothing that is applied uses a Gaus- ian smoothing function [17]. The optimal size of the smoothing indow depends on 2w and the data rate. While the manner of the alculation of the smoothing window is not specified, it is likely 1/3 o 1/4 of the number of data points expected across the peak, as one ould use when applying a Gaussian-shaped Savitsky-Golay filter 18] to chromatographic data. It is expected that an inappropriate lter width would impact the smoothing and subsequent quantifi- ation. Consequently, the expected 2w is expected to be a critical ata processing parameter for a two-step integration algorithm hat incorporates a smoothing step, as in ChromaTOF®. To study the differences between GC×GC data processed using ifferent 2w, simulated data were used so that error sources such as R, 1w, 2w, �, and shifts in 1tR could be controlled. Real data were hen used to confirm the results obtained from the simulated data. oth the simulated and real data were evaluated using parameters f total peak area (the sum of the areas of the largest peak, called 3, respectively. the base peak and the other component peaks, called sub-peaks) [17], signal-to-noise ratio (S/N) of the modulated base peak, and the number of included sub-peaks. The differences between the values calculated using the actual 2w (chosen by the investigator upon viewing the 2w of the peak in the actual chromatogram) and an imprecise 2w were measured as a percentage using Eq. (1): D% = ∣∣∣ ( Xp Xa × 100 ) − 100 ∣∣∣ (1) where D% is the difference in peak areas expressed as a percentage; Xp is the total peak area obtained with an imprecise 2w and Xa is the total peak area obtained with the actual 2w. Hence the data processed with the actual 2w will have a value of 0 for D%. If the value obtained from processing with the imprecise 2w is greater than that obtained with the actual 2w, D% > 0% if the area with the imprecise 2w is less than when the actual 2w is used D < 0%. A. Paulina de la Mata et al. / J. Chromatogr. A 1255 (2012) 190– 195 193 F A) Exp p eaks; 3 f T p w s i t t ( p d w t 3 a p l ig. 1. Different data processing for 50 ppm trioctyl phosphate, actual 2w of 0.3 s. ( eak (14-B) and 5 sub-peaks; (C) Expected 2w of 0.1 s: base peak (14-B) and 4 sub-p .1. Simulated GC×GC data Peaks of differing 1D magnitude were simulated to evaluate dif- erences between analytes with higher and lower concentrations. he modulation was simulated such that � = 0◦ for the modulated eaks and two values of 2w (1.0 and 0.10 s) were chosen. All data ere processed using expected 2w input into the data processing oftware of 1, 0.5, 0.3, and 0.1 s. When the actual 2w was 1 s the D% ncreased as the 2w used for processing decreased and vice versa for he data with an actual 2w of 0.1 s. The results show differences in otal peak area greater than 10% regardless of the overall magnitude concentration) of the peak (Table 1). However, in some cases the arameters of total peak area and S/N of the base peak presented ifferences up to 100%. These differences are a result of the soft- are not identifying all of the sub-peaks or in some cases missing he peak altogether. .2. Real GC×GC data Data collected from a mixture of alkanes, alcohols, and ketones t concentrations of 50 and 100 �g mL−1 and a mixture of alkyl hosphates at 50 and 125 �g mL−1 were used to verify the simu- ations. The alkanes, alcohols, and ketones were found to have an ected 2w of 1.0 s: base peak (18-B) and 4 sub-peaks; (B) Expected 2w of 0.3 s: base (D) Expected 2w of 0.05 s: base peak (12-B) and 3 sub-peaks. actual 2w of 0.3 s while the alkyl phosphates had varying 2w as seen in Table 2. As with the simulated data, the real data were processed using 2w of 1, 0.5, 0.3 and 0.1 s. Significant deviations in total peak area measured by D% were observed, though they were in general not as extreme as with the simulated data. The alkyl phosphate peaks have very high areas, compared to analytes analysed on the TOFMS. This likely in part due to how ChromaTOF® interprets the signal from the NPD vs. the TOFMS. Additionally, for the real data, the phase of the peaks was not controlled, and the differences between actual and expected 2w were not as extreme as in the simulated data. Nevertheless, there were significant errors in area and in S/N, especially with the badly tailing phosphate peaks (most notably trimethyl phos- phate). Fig. 1 shows the chromatogram of a sample of 50 �g mL−1 tri- octyl phosphate (actual 2w of 0.3 s) processed using expected 2w of 1, 0.3, 0.1 and 0.05 s. In each case, the software identifies a dif- ferent number of sub-peaks (denoted npeak − nsubslice) surrounding the base peak (denoted npeak − B) depending on the 2w used for data processing. When the actual 2w was used for data process- ing (Fig. 1B), the software distinguished one base peak with five sub-peaks. When the processing 2w was increased or decreased the software identified a fewer number of sub-peaks. 194 A. Paulina de la Mata et al. / J. Chromatogr. A 1255 (2012) 190– 195 Fig. 2. (A) 5 ppm series of phosphates; (B) peak table when data processed with 2w of 0.1 s; (C) peak table when data processed with 2w of 0.3 s. Fig. 3. 1 ppm trioctyl phosphate, actual 2w of 0.3 s. (A) Expected 2w of 1.0 s; (B) Expected 2w of 0.5 s; (C) Expected 2w of 0.3 s; (D) Expected 2w of 0.1 s. hrom l t s s T s a c o r d 4 r s r i p i w t a G m t s w t [ [ [ [ [ [15] W. Khummueng, J. Harynuk, P.J. Marriott, Anal. Chem. 78 (2006) 4578–4587. A. Paulina de la Mata et al. / J. C Choosing an imprecise width also affects the number of ana- ytes identified by the software (Fig. 2). With a 2w of 0.1 s (Fig. 2B) he software identified 4 analytes; however, with a 2w of 0.3 s the oftware identified 6 different analytes (Fig. 2C). Table 2 presents elected results for high and low concentration analytes on the OFMS and NPD, some of which have symmetrical peak shapes and ome of which tail badly. In general, the errors introduced to peak rea by improper choice of expected 2w are more severe for lower oncentration peaks and tailing peaks. Regardless of the peak shape r intensity this parameter will have a serious impact on the S/Nb eported. Fig. 3 depicts changes in the noise and peak profiles for ata processed with different expected 2w. . Conclusions There are many situations in GC×GC data (phase, modulation atio, univariate vs. multivariate detection, tailing in either dimen- ion) which can arise, and a complete study remains for future esearch. In the interim, we have demonstrated that expected 2w s a critical integration parameter when data smoothing is incor- orated into the data interpretation routine. Improper choice of 2w mpacts the number of sub-peaks identified in a chromatogram as ell as the number of analyte peaks found. This then affects the otal peak area and S/N for an analyte. It is important to choose s precise a 2w as possible when quantifying data. In quantitative C×GC analyses one must ensure that the 2w observed in a chro- atogram is within a narrow tolerance of the expected 2w used in he calibration in order to ensure the validity of the results. Finally, while these results were obtained using ChromaTOF®, imilar results would be expected from any data interpretation soft- are incorporating window-based smoothing of the raw data prior o signal integration. [ [ [ atogr. A 1255 (2012) 190– 195 195 Acknowledgements The authors would like to acknowledge NSERC (Canada) and AITF (Alberta) for providing financial support for this research in the form of research grants to J. Harynuk and scholarships for K. Nizio. Support for acquiring the GC×GC-TOFMS system was provided in part by the University of Alberta, Leco Canada, CFI, and the Province of Alberta’s SEGP. References [1] M. Adahchour, J. Beens, R.J.J. Vreuls, U.A.T. Brinkman, TrAC, Trends Anal. Chem. 25 (2006) 726–741. [2] O. Panić, T. Górecki, Anal. Bioanal. Chem. 386 (2006) 1013–1023. [3] F. Gogus, M.Z. Ozel, D. Kocak, J.F. Hamilton, A.C. Lewis, Food Chem. 129 (2011) 1258–1264. [4] A.W. Culbertson, W.B. Williams, A.G. McKee, X. Zhang, K.L. March, S. Naylor, S.J. Valentine, LC GC N. Am. (2008) 74–81. [5] Z. Liu, J.B. Phillips, J. Chromatogr. Sci. 29 (1991) 227–231. [6] J.B. Phillips, J. Beens, J. Chromatogr. A 856 (1999) 331–347. [7] W. Bertsch, J. High Res. Chromatogr. 23 (2000) 167–181. [8] T. Górecki, J. Harynuk, O. Panić, J. Sep. Sci. 27 (2004) 359–379. [9] M. Adahchour, J. Beens, R.J.J. Vreuls, U.A.T. Brinkman, TrAC, Trends Anal. Chem. 25 (2006) 438–454. 10] M. Adahchour, J. Beens, R.J.J. Vreuls, U.A.T. Brinkman, TrAC, Trends Anal. Chem. 25 (2006) 540–553. 11] G. Semard, M. Adahchour, J.-F. Focant, Comprehensive Analytical Chemistry, Elsevier, 2009, pp. 15–48. 12] P.Q. Tranchida, G. Purcaro, P. Dugo, L. Mondello, TrAC, Trends Anal. Chem. 30 (2011) 1437–1461. 13] T. Górecki, M. Edwards, A. Mostafa, Anal. Bioanal. Chem. 401 (2011) 2335–2349. 14] J.J. Harynuk, A.H. Kwong, P.J. Marriott, J. Chromatogr. A 1200 (2008) 17–27. 16] R.C.Y. Ong, P.J. Marriott, J. Chromatogr. Sci. 40 (2002) 276–291. 17] Leco Corporation, ChromaTOF® Software Instruction Manual, St. Joseph, MI, Version 4.4x ed., April 2011. 18] A. Savitzky, M.J.E. Golay, Anal. Chem. 36 (1964) 1627–1639. Integration parameters and their effects on quantitative results with two-step peak summation quantitation in comprehensiv... 1 Introduction 2 Experimental 2.1 Simulated GC×GC data 2.2 Real GC×GC data 2.2.1 Materials and reagents 2.2.2 Instrumental conditions 2.3 Software 3 Results and discussion 3.1 Simulated GC×GC data 3.2 Real GC×GC data 4 Conclusions Acknowledgements References