VU Research Portal
Analysis of complex samples using multidimensional gas chromatography and selective detection Stee, L.L.P.
2017
document version Publisher's PDF, also known as Version of record
Link to publication in VU Research Portal
citation for published version (APA) Stee, L. L. P. (2017). Analysis of complex samples using multidimensional gas chromatography and selective detection.
General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ?
Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
E-mail address: [email protected]
Download date: 06. Oct. 2021
Analysis of complex samples using multidimensional gas chromatography and selective detection
Leo L.P. van Stee
ISBN: 978-94-028-0561-1
VRIJE UNIVERSITEIT
Analysis of complex samples using multidimensional gas chromatography and selective detection
ACADEMISCH PROEFSCHRIFT
ter verkrijging van de graad Doctor aan de Vrije Universiteit Amsterdam, op gezag van de rector magnificus prof. dr. V. Subramaniam, in het openbaar te verdedigen ten overstaan van de promotiecommissie van de Faculteit der Exacte Wetenschappen op woensdag 12 april 2017 om 15.45 uur in de aula van de universiteit, De Boelelaan 1105
door
Leonard Laurens Petrus van Stee
geboren te Willemstad, Curaçao promotoren: prof.dr. U.A.Th. Brinkman prof.dr.ir. J.G.M. Janssen
Aan mijn ouders, Aletta, Tobias en Lara
Contents
List of abbreviations ...... 11
1 1. Introduction ...... 13 1.1 General introduction ...... 13 1.2 Scope of the thesis ...... 15
2 2. Gas chromatography with atomic emission detection: a powerful and versatile technique ...... 19 2.1 Introduction ...... 19 2.2 Applications ...... 21 2.3 Conclusions ...... 39
3 3. Identification of non-target compounds using gas chromatography with simultaneous atomic emission and mass spectrometric detection (GC–AED/MS): analysis of municipal wastewater ...... 41 3.1 Introduction ...... 41 3.2 Experimental...... 42 3.3 Results and discussion ...... 44 3.4 Conclusions ...... 52
4 4. Comprehensive two-dimensional gas chromatography with atomic emission detection (GC×GC–AED) and correlation with mass spectrometric detection: principles and application in petrochemical analysis ...... 53 4.1 Introduction ...... 53 4.2 Experimental...... 54 4.3 Results and discussion ...... 56 4.4 Application: petrochemical analysis ...... 60 4.5 Conclusions ...... 65
5 5. Evaluation of the combined use of biomimetic and solid-phase extraction techniques for the screening of organic micropollutants in wastewater ...... 67 5.1 Introduction ...... 67 5.2 Materials and methods ...... 70
5.3 Results and discussion ...... 73
6 6. Toxicity identification and evaluation of inland waters: Use of semi-permeable membrane devices and solid-phase extraction for the wide-range screening of microcontaminants in surface water by GC–AED/MS ...... 83 6.1 Introduction ...... 83 6.2 Experimental...... 84 6.3 Results ...... 88 6.4 Conclusions ...... 97
7 7. Responses in sediment bioassays used in the Netherlands: can observed toxicity be explained by routinely monitored priority pollutants? ...... 103 7.1 Introduction ...... 103 7.2 Materials and methods ...... 104 7.3 Results ...... 107 7.4 Discussion ...... 109 7.5 Conclusions ...... 111
8 8. Comprehensive two-dimensional gas chromatography (GC×GC) measurements of volatile organic compounds in the atmosphere...... 113 8.1 Introduction ...... 113 8.2 Experimental...... 115 8.3 Results and discussion ...... 119 8.4 Conclusions ...... 130
9 9. Peak detection methods for GC×GC: an overview ...... 137 9.1 Introduction ...... 137 9.2 GC×GC: principles and visualisation ...... 138 9.3 Classification of peak detection methods ...... 140 9.4 Methods based on 1D peak detection ...... 141 9.5 Multivariate techniques ...... 153 9.6 Graphical drain method ...... 159 9.7 Conclusions ...... 161 10 10. Peak clustering in GC×GC–MS based on theoretical calculation of two-dimensional peak shapes: the 2DAid approach ...... 165 10.1 Introduction ...... 165 10.2 Theory ...... 166 10.3 Experimental...... 170 10.4 Results and discussion ...... 172 10.5 Conclusions ...... 178
11 11. GC×GC–ToF-MS using various sets of column combinations for the non-target screening of 150 micro-contaminants detected in surface water ...... 179 11.1 Introduction ...... 179 11.2 Experimental...... 180 11.3 Results and discussion ...... 181 11.4 Conclusions ...... 185
12 12. References ...... 187
Samenvatting ...... 199
List of publications ...... 203
Dankwoord ...... 205
List of abbreviations
ABS ...... acrylonitrile-butadiene-styrene AED ...... atomic emission detection AES ...... atomic emission spectrometer AFC ...... alternative fluorocarbons ALS ...... alternating least squares BDE ...... brominated diphenyl ether CB ...... chlorobiphenyl CDS ...... chromatography data system CFC ...... chlorofluorocarbon CIC ...... compound-independent calibration CSV ...... comma separated values CWA...... chemical warfare agents DBT ...... dibenzothiophene ECD ...... electron-capture detector EI ...... electron ionisation EPA ...... environmental protection agency FCC ...... fluidised catalytic cracking FID ...... flame ionisation detector FPD ...... flame photometric detector GC ...... gas chromatography GC×GC ...... comprehensive two-dimensional gas chromatography GC–GC ...... heart-cutting two-dimensional gas chromatography GPC ...... gel-permeation chromatography GRAM ...... generalised rank annihilation method HPLC ...... high-performance liquid chromatography i.d...... internal diameter IR ...... infrared LC ...... liquid chromatography LLE ...... liquid–liquid extraction LOD ...... limit of detection LVI ...... large-volume injection m/z ...... mass-to-charge ratio MCO ...... medium cycle oil MIP ...... microwave-induced plasma MS ...... mass spectrometer or spectrometry MVA...... multivariate analysis NMR ...... nuclear magnetic resonance OHP ...... organohalogen pesticides OPP ...... organophosphorus pesticides P&T ...... purge and trap PAH ...... polyaromatic hydrocarbons PARAFAC ...... parallel factor analysis PASH ...... polyaromatic sulphur heterocycles PBDE ...... polybrominated diphenyl ether PCB ...... polychlorinated biphenyl PDA ...... photo diode array
PFPD ...... pulsed flame photometric detector PTFE ...... polytetrafluoroethylene PTV ...... programmed-temperature vaporisation PVC ...... polyvinyl chloride Py–GC ...... pyrolysis–gas chromatography RI ...... retention index RSD ...... relative standard deviation RTL ...... retention time locking S/N ...... signal-to-noise ratio SBSE ...... stir-bar sorptive extraction SCD ...... sulphur chemiluminescence detector SDE ...... simultaneous steam distillation/solvent extraction, Likens-Nickerson SFE ...... supercritical fluid extraction SLE ...... solid–liquid extraction SPE ...... solid-phase extraction SPMD ...... semi-permeable membrane device SPME ...... solid-phase micro extraction TCA ...... trichloroanisole TIC ...... total ion chromatogram ToF-MS ...... time-of-flight mass spectrometer WWTP ...... wastewater treatment plant
12 1 1. Introduction
1.1 General introduction A decade ago, a Ph.D. thesis from the Department of Analytical Chemistry and Applied Spectroscopy of this university opened with the statement ‘The excellent separation capability of capillary gas chromatography (GC), the high speed of analysis and the ease of combination with a variety of selective detectors—with the mass spectrometer (MS) taking the first place these days—all contribute to make GC the analytical method of choice whenever GC-amenable compounds have to be determined’ [1]. This statement is still true today, but we should immediately add that, in the meantime, distinct progress has been made in all the core areas, i.e. with regard to sample preparation, analytical separation and (selective) detection. To quote a few examples, selective sample preparation has received much attention when determining one or a few target compounds or a specific class of analytes. On the other hand, if general screening is requested—i.e. if non-target compounds are an essential part of the study—a non- selective approach has to be used. Here, one main concern is to effect the isolation of the original, i.e. the intact, compounds and to avoid metabolisation or other types of analyte transformation during sample handling and, also, during further processing. The latter caveat essentially rules out the use of derivatisation (and related) procedures which—to mention another serious disadvantage—invariably increase the time of analysis and frequently cause problems, either due to interferences present in the reagents used or formed as by-products in side-reactions with other compound classes contained in the samples to be analysed1. In other words, resolving one’s problem now has to sought via improved separation and/or detection capabilities. In the field of capillary GC separation, the past two decades have seen the emergence of comprehensive two-dimensional GC (GC×GC), initially for research alone, but now also increasingly for routine application. Here, the considerable gains made with regard to sample fractionation and overall separation efficiency—without
1 With the analytes of interest usually being present at the (ultra-)trace level, other sample constituents present at much higher concentrations can create serious interferences even though they are converted at a minimum level of, say, a few per cent.
1. Introduction
materially increasing the GC run time!—have been amply demonstrated for a wide variety of sample types and many different classes of compounds such as organohalogen microcontaminants, flavours and fragrances, aerosol constituents and essentially all classes of compounds present in petrochemicals. Actually, now that GC×GC has become a mature technique, one complaint is that the amount of information becoming available per analysis and, consequently, per working day, is so large that—rather than the analytical procedure itself—it is data handling that frequently is the main stumbling block. Here, proper peak detection, analyte identification and analyte quantification are the three areas of most common concern. As for analyte detection in the most wide sense of the word, devices such as the electron-capture detector (for all halogen-containing microcontaminants) and the flame- ionisation detector (for most studies of petrochemicals) still have a unique role to play. Besides these common detectors, the use of very specific detectors such as the atomic emission detector (AED) with its unique element-selective characteristics can be very rewarding even though the costs are higher. However, these are the exceptions rather than the rule which, simply, says that MS detection is to be recommended in virtually all instances—and specifically whenever large numbers of, almost invariably, partly overlapping analyte peaks/spots show up, and/or whenever unambiguous identification is required for, e.g., legal or commercial purposes. Selfevidently, MS detection is also essential if non-target analyses have to be performed. Here, the relatively poorly investigated field of air and aerosol analysis is an illustrative example. Today, it usually is the time-of-flight mass spectrometer (ToF-MS) which is installed at the outlet of a GC×GC separation system. Its rapid data acquisition, general flexibility of operation and advantageous cost/benefit ratio are appreciated by all workers in the field. Commercial hardware is now available to successfully study even unusually complex sample types and/or classes of analytes, with the cigarette smoke problem as one example and the in-depth study of the ubiquitous polyhalogenated alkanes as another one. It is in such areas that rapid and reliable peak detection and identification still are a problematic issue which requires further study.
In the present thesis, most of the problems and problem areas briefly discussed above, have been addressed—both in terms of method and instrument development and in terms of real-life application. In addition, two topics—AED-based detection and peak picking/detection in GC×GC—have been reviewed. Overall, the experimental results which are presented and the discussions which are included, clearly illustrate the substantial progress that has been made in the past ten to fifteen years and, specifically, the wide applicability range of the analytical techniques applied. Simultaneously they
14 1. Introduction also indicate where further improvement is necessary or, at least, desirable—or, in other words, which aims should be pursued in the near future. 1.2 Scope of the thesis Chapter 2 opens with a short review of the development of atomic emission detection coupled to gas chromatography (GC–AED) and the unique properties of that combination. The first commercially available AED was launched in 1989; it used a helium microwave plasma with temperatures in the order of 2500 K. At these temperatures molecules are atomised and the atoms excited; the light emitted during transition to a lower energy level is analysed spectrophotometrically. The AED can selectively detect essentially all elements that are present in GC-amenable compounds, e.g. C, H, O, S, P, Cl, Br, F, Si, Hg and Sn. Since the measured responses originate from the individual atoms that make up the compounds, AED can be used for compound- independent calibration and, often, partial molecular formulas can be proposed. In the review, emphasis is on real-life—e.g., food and environmental—applications for non-metals such as sulphur, phosphorus, nitrogen and the halogens, and on the potential of combined AED/mass spectrometric (AED/MS) detection.
Chapter 3 describes the parallel use of AED and MS detection. For the analysis of non- target compounds, MS has been the favoured technique for many years. The most commonly used ionisation technique in GC–MS is electron ionisation at 70 eV, and over several decades the distinct fragmentation patterns have been collected in large libraries. Spectra of non-target compounds can be matched against library spectra in an automated process and so provide tentative identification or give clues as regards the class of compound. Furthermore, the isotopic distribution in the mass spectra can give (partial) information on the elemental composition and the presence of specific functional groups. Combined AED/MS information further expands the options for identification by providing element-specific information. Here, an important requirement is a high correlation between the data, i.e., the recorded retention times for both AED and MS should preferably be the same or at least have a constant offset throughout the chromatogram. The study describes a set-up with parallel AED/MS detection that takes into account the (varying) gas flows in the different heated zones to obtain as high a correlation as is possible. The technique was used to analyse sewage water influents and effluents of a treatment plant. As an example, in one water sample, seven non-target chlorinated compounds were identified, and comparison of the GC profiles of the influents and effluents showed a 85–90% removal of almost all of the GC-amenable compounds.
15 1. Introduction
As discussed above, comprehensive two-dimensional gas chromatography (GC×GC) provides a high separation efficiency that is ideal for application in non-target analysis.
Chapter 4 reports the combined use of GC×GC–AED and GC×GC–time-of-flight-MS (ToF-MS). One problem that has to be solved to enable the successful use of AED detection in GC×GC is the generation of a sufficiently high sampling frequency to obtain enough data points across a chromatographic peak. This condition was met by proper adaptation of the dimensions of the transfer lines and the gas flow rates. Analyses were performed on separate GC×GC instruments, and high-sulphur petroleum products were used as samples. Data correlation by means of graphical scaling and overlay was studied and shown to be successful. One interesting result was the unambiguous recognition of the presence of a class of N-containing analytes (dimethylcarbazoles), where proper correlation of the AED and ToF-MS data turned out be crucial for identification.
In the studies described in Chapters 5–7 some of the above techniques were applied to the analysis of surface and pore water samples collected by Rijkswaterstaat, then the Institute for Inland Water Management and Waste Water Treatment (RIZA, Lelystad). In previous studies of these types of sample the long- and short-term toxicities were examined with various bioassays such as Daphnia magna and Chironomus riparius. However, despite much research, part of the toxicity could not be explained by contaminants determined during routine monitoring. One hypothesis is that lipophilic compounds which are present at only very low concentrations in water, can bioaccumulate, which would explain the high toxicity. However, because of their very low concentrations, these compounds are usually not picked up during routine water monitoring. One way to solve this problem, is to use so-called biomimetic extraction. Here (Chapter 5), a device containing a hydrophobic material that acts as an accumulation phase is placed in the water sample for several days up to weeks. The process mimics analyte accumulation—and, thus, concentration build-up—in biological species; it has the added advantage that the compounds are not metabolised, as is the case in samples taken from living organisms. Different biomimetic samplers were evaluated for use under various conditions in the field as well as the laboratory; they were also compared with conventional solid-phase extraction (SPE). Fouling was found to be a serious problem when sampling (highly) polluted water samples under dynamic, i.e. field, conditions. However, no such problems occurred when using biomimetic extraction under non-flow conditions in the laboratory, and the combination of this approach and conventional SPE turned out to give successful results for a very wide log P-range of microcontaminants. As a follow-up (Chapter 6), an automated screening procedure was developed for over 400 industrial, agrochemical and household chemicals. In samples taken at various
16 1. Introduction locations in the main Dutch river systems, some 150 target compounds were detected at the low-ng/l to low-µg/l level. Next to these, several brominated and chlorinated non- target compounds were detected, and provisionally identified, by means of GC–AED screening. In a further study (Chapter 7), suspended matter and sediments from the river Meuse were studied. The results of a series of standard acute bioassays and standard chronic tests were compared with those obtained in the course of the present work. In many cases, the toxicity observed in the standard assays could largely be explained by the presence, and concentrations, of known persistent priority pollutants—mainly heavy metals and polycyclic aromatic hydrocarbons. Agricultural run-off of pesticides, which are not routinely measured in sediments, may explain part of the remaining differences. In other experiments, pore water samples from Lake IJssel sediments yielded a further list of microcontaminants, such as phthalates, decanes, cosanes and fragrances. However, their contribution to the effects measured in the bioassays used, was found to be negligible. In summary, although much analytical and conceptual progress has been made and many more microcontaminants have been detected, the core problem of the ‘toxicity gap’ has, as yet, not fully been solved.
In Chapter 8, GC×GC was used to analyse volatile organic compounds (VOCs) in air. These compounds, either originating from human activity or natural sources, play an important role in e.g. smog and ozone formation. Due to the wide variety of sources, these VOCs form a complex mixture comprising many different classes of compounds. Data obtained from in situ GC×GC–FID and laboratory GC×GC–ToF-MS analyses, combined with retention indices obtained from the literature, were used to identify about 250 compounds, and quantification was performed for selected analytes. The application demonstrates the advantage of the high separation capacity of GC×GC for non-target analysis, specifically in ‘unbalanced samples’, i.e. samples with large differences in concentration of the compounds detected. To give an example, in air samples taken at a station in the island of Crete, some 650 peaks were identified with signal/noise ratios of >100 and mass-spectral similarities of >850.
With the very large amount of data being generated in each GC×GC run, data analysis has become a topic of distinct interest. Chapter 9 reviews one specific aspect, i.e., the currently available methods for (individual) peak detection. Four main types of detection method are distinguished—user-supervised peak(let) summation, automated peaklet clustering, multivariate methods and graphical methods. The principles of the various approaches and required input and type of output are discussed in light of analytical requirements such as target analysis and general screening/non-target analysis. Several
17 1. Introduction
published real-life applications are briefly discussed. One main conclusion is that for more or less routine and flexible procedures—which is what most analysts are in need of—commercial tools and toolkits should invariably be used.
Chapter 10 discusses the merits of an in-house developed method for 2D peak integration in GC×GC–MS data. The method is based on peaklet clustering and uses relatively simple chromatographic/thermodynamic laws to calculate the theoretical shape of a 2D peak in order to define the area in which the peaklets of each individual compound can be expected to show up. In the delimited area the peaklets are then clustered based on analyte-identity information obtained from mass spectral library searching, and the responses of the peaklets are summed to obtain a single 2D peak response. It is demonstrated that the 2D peak shapes can be accurately predicted and that clustering and further processing can reduce the final 2D peak list to a manageable size.
Chapter 11 reports the application of the method reported in Chapter 10 to the detection of 150 organic microcontaminants representing nine classes of compounds, with GC×GC–MS using five different GC column combinations. As was to be expected, with such a large number of analyte classes (and individual compounds), none of the column combinations could provide complete group separations. However, analysis of the experimental results—with peak overlap as a main criterion—gives a rapid and efficient indication of the available 2D separation space, and of the column combination to be recommended for a specific separation problem.
18 2 2. Gas chromatography with atomic emission detection: a powerful and versatile technique
2.1 Introduction Atomic emission detection (AED) is a sensitive as well as selective detection technique for capillary gas chromatography (GC) which provides very valuable element-selective information. The well-defined and identifiable electron transitions in excited atoms or ions render atomic spectroscopy the best element-selective method available to the analyst. Since its first analytical use—the visual identification of salts by introducing a sample into a flame—for a long time, atomic spectrometry remained in the domain of the inorganic chemist. The situation changed dramatically in the early 1990s, after the introduction of an atomic emission spectrometer (AES) which was compatible with capillary GC, and so became the tool of many organic, environmental and analytical chemists. The first use of GC–AES was reported in 1965 [2;3]. With an argon microwave- induced plasma (MIP), limits of detection (LODs) in the pg/s range were achieved for several elements, but the selectivities against carbon were very poor. The introduction of
the Beenakker TM010 cavity [4;5] was a major breakthrough. Because of the better energy transfer to the discharge tube, it allowed the operation of a stable helium plasma at atmospheric pressures. Helium has the advantage over argon that there are fewer emission lines of diatomic species formed by recombination of analyte atoms with contaminants present in the gas and incomplete degradation of the analyte molecules. The cavity was first used in conjunction with GC in 1978 [6]; LODs were in the 2–60 pg/s range and selectivities were much better than observed with other types of plasma. A modified version of this cavity was included in the Hewlett-Packard system—usually called an AED rather than an AES—launched in 1989. Today, many hundreds of papers have been published on GC–AED with the HP5921A (and its successors, the Agilent G2350A, and the 2370AA, marketed by JAS). It is not the intention of this paper to present a complete overview of the many technical developments introduced since 1965 and/or of all real-life applications. For this, the reader should consult reviews on, e.g., element-selective detection in chromatography
Published as: L.L.P. van Stee, U.A.Th. Brinkman, J. Chromatogr. A 1186 (2008) 109 2. GC–AED a powerful technique
[7], speciation of, especially, mercury-, tin- and lead-containing compounds [8-10] and environmental and other applications [11-13]. The main goal of the present review is to show the versatility and practicability of GC–AED to solve a wide variety of problems in trace-level organic analysis. Because of the detailed information available in the literature on (speciation studies of) organo-metal and organo-metalloid compounds, the focus will be on the monitoring of non-metallic elements. For the convenience of readers not familiar with the technique, the main characteristics of AED detection are briefly discussed below. AED performance. GC–AED provides simultaneous multi-channel detection for up to four elements with excellent LODs of 1–30 pg/s for many important elements, response linearities of typically 3–5 orders of magnitude and element vs. carbon selectivities of the same order of magnitude. The high selectivity helps to maintain analyte detectability at its standard level with even complex samples. Selected data are shown in Table 2.1. Table 2.1: Analytical characteristics of AED detection for selected elementsa Element Wavelength Setb LOD Selectivity over carbon (nm) (pg/s) (× 10-3) N 174.2 1 15–50 2–5 S 180.7 1 1–2 5–20 C 193.1 1 0.2–1 - P 178.1 2 1–3 5–8 C 495.8 3 15 - H 486.1 3 1–4 - Cl 479.5 3 25–40 3–10 Br 478.6 3 30–60 2–6 F 685.6 4 60–80 20–50 O 777.2 5 50–120 10–30 Si 251.6 6 1–7 30 Hg 253.7 6 0.1–0.5 250 Pb 261.4 6 0.2–1 300 Sn 270.7 6 1 300 N 388.3 7 10 >5 a. Data collected from various sources; b. Arbitrary order As regards multi-element detection, depending on the application—i.e., on the elements of interest—several runs are often required to cover the complete set of elements which, of course, increases the time of analysis. This is due to the fact that the optics of the AED are designed to realise the high resolution required to distinguish certain elements, e.g. chlorine (479.5 nm) and bromine (478.6 nm). Because of this, during one run, detection is possible within a window of only 20–25 nm out of the total 160–800 nm that the AED can cover. In addition, elements such as e.g. phosphorus require special make- up gases and/or flow conditions (see Table 2.1). Throughout the text, the wavelength used to measure an element refers to the wavelengths in Table 2.1. In case more
20 2. GC–AED a powerful technique wavelengths are given in Table 2.1 or when the wavelength is not included in the table, the wavelength will be indicated by subscript (e.g. C193). A most rewarding aspect of AED detection is the so-called universal or compound- independent calibration (CIC). The high temperatures of the MIP-type plasma cause an essentially complete breakdown of all analyte molecules into their constituent atoms. Consequently, the response per mass unit of an element is more or less independent of the structure of the analyte of interest. As a result, quantification for a whole series of compounds can be based upon data recorded for a single analyte containing the common hetero-atom; if reference compounds are not available, a related compound can be used. In addition, elemental ratios and, thus, (partial) molecular formulae can be calculated. In experimental practice, these tools are frequently used (see [14] for a review), and with marked success. In many instances, AED is combined with mass spectrometric (MS) detection to acquire even more detailed information. 2.2 Applications 2.2.1 Food and drinks The commercially available HP 5921A AED was first used for pesticide analysis by Wylie and Oguchi [15]. They developed a method to detect 27 pesticides in apple extracts. By using the traces of nine elements (LODs, 0.1–75 pg/s) to derive molecular formulae and combining these with retention data, proper identification of 20 pesticides was achieved. Three pesticides were correctly identified in an apple spiked much below the EPA’s maximum residue limit. A literature overview of pesticide analysis in food is given in Table 2.2; some selected studies are discussed in more detail below. In an early study, the decomposition of two fungicides applied to strawberries was monitored for several weeks by means of the Cl, N and S traces [16]. Next, the same group published an extensive study on the detection of 385 pesticides in fruits and vegetables, giving LODs for each of the main hetero-atom channels for each individual pesticide. Generally speaking, the high selectivity of the AED was found to outweigh its lower sensitivity (for e.g. halogen atoms), compared with electron-capture detection (ECD). The element traces of the most interesting hetero-atoms, N, Cl, S, P, Br and F, were found to be almost free of disturbances caused by co-eluting matrix constituents. Only in the case of rather notorious vegetables such as onion, leek and garlic, did interferences occur in the S trace. The high selectivity allowed the sample preparation procedure to be simplified, and the feasibility to screen for pesticide residues in plant foodstuffs down to the 10 ng/g level (ca. 10 g samples; 6% of final extract injected) was demonstrated [17]. The results of this extensive study were in full agreement with those of an earlier study by Lee et al. [18], who compared AED with flame photometric
21 2. GC–AED a powerful technique
(FPD), nitrogen\phosphorus (NPD) and ECD detection for the analysis of pesticides in twelve types of fruits and vegetables, and emphasised that the high selectivity of AED detection could obviate the need for clean-up in most cases. Table 2.2: GC–AED analysis of pesticides in food products Sample(s) Techniques Comments Refs. Apple LLE Organo-P (OPP) and -halogen (OHP) pesticides; empirical [15] formula calculation; 0.3–0.7 mg/kg spikes Vegetables, fruits LLE Comparison with FPD, NPD and ECD; 0.2 mg/kg spikes [18] Fruits, vegetables LLE OPPs, OHPs, carbamates, metal-containing pesticides [19] Rice grains SFEa OPPs at 10 µg/kg, used over a two-week period with up to 70 [20] samples/day Onion, radish, potato S19 Snapshots used to prevent false positives due to matrix [21] methodb compounds Fruits, vegetables S19 385 pesticides; LOD, 10 µg/kg [17] Fruits S19, LVIc AED vs. ECD and NPD; AED LODs of same order as ECD or [22] NPD (0.02–0.9 mg/kg) Strawberry S19 9-week monitoring of decomposition of two fungicides applied [16] to strawberry in the field Honey LLE Acaracides such as amitraz and decomposition products; plus [23-25] ECD, NPD and MS. Fruits, vegetables LLE Validation with large set of pesticides and fruits. GC–AED and [26] –MS; partial formulae calculation Honey SPME 16 organochlorines, OPPs and pyrethrins; LODs 0.02–10 µg/kg [27] a. Supercritical fluid extraction b. LLE, GPC and fractionation on silica column c. Large-volume injection In a more recent study [26], fruit and vegetable extracts were screened for over 400 pesticides using an experimental database. Retention-time locking (RTL) was used to match GC–AED and GC–MS retention times. Samples were analysed for S, N, P and Cl; possible pesticides were suggested by database search and identified by GC–MS. For blind spikes of extracts, the database suggested 22 out of 26 pesticides as matches; 19 were identified by GC–MS. Jiménez and co-workers [23-25] devoted several papers to the determination of acaricides (which are used to treat beehives against mites). In early work on four acaricides, the authors found that GC–AED is more sensitive (C193 LODs, 0.05–0.5 ng/g), but less selective, than GC–ECD/NPD [23]. In another study [25], combined AED and MS detection was used to detect degradation products—a challenging task because for such products standards often are not available. By using the Cl and N traces, the degradation of chlordimeform in spiked honey was followed over a period of 28 weeks. Two degradation products were shown to be present which were, next, identified by means of GC–MS. Very recently, direct sampling of honey with an SPME fiber [27] was shown to be a simple method for residue analysis (16 organochlorines, organo-Ps and pyrethrins) in honey. Element-selective detection by means of the Cl, Br and S channels gave LODs of 0.02–10 ng/g. The method can be used for routine analysis. However,
22 2. GC–AED a powerful technique because of the complexity of the matrix, for reliable quantification standard addition had to be applied. AED has also been employed for the determination of compounds other than pesticides in food products (Table 2.3). In this context, the high selectivity and sensitivity of the S channel is a major advantage because of the low odour thresholds and flavour characteristics of many S-containing compounds. Headspace SPME was used to study the release of 4-mercapto-4-methylpentan-2-one, a volatile thiol which is a potent contributor to wine aroma, from its non-volatile L-cysteine precursor. The fermentation temperature as well as the yeast strain selected were found to provide important tools to enhance or modulate the final wine aroma [28]. Table 2.3: GC–AED analysis of non-pesticides in food and beverages Sample(s) Techniques Comments Refs. Wine Headspace Aroma compound 4-mercapto-4-methylpentan-2-one [28] SPME Wine, corks P&T 2,4,6-Trichloroanisole [29] Garlic headspace Static headspace 15 S-containing compounds; plus MS [30] Water, beer, coffee P&T 13 S-containing targets with potential for off-flavour [31] Wine P&T Dimethyl sulphide in red wines [32] Cured ham SDEa Profiling of volatiles in complex flavour isolate; plus GC– [33] MS, –NPD, –FPD and –FIDb Onion, garlic P&T Analysis of onion and garlic headspace [34] a. Likens-Nickerson, simultaneous steam distillation/solvent extraction b. Flame ionisation detection In another paper [29], the high selectivity of the Cl channel was used to detect trichloro- anisole (TCA) in wine and cork stoppers. TCA is a common off-flavour in many food products and is formed by microbial degradation of chlorophenols. GC–MS in the selected ion monitoring mode and GC–ECD can both be used for trace-level TCA studies, but GC–AED enables easier quantification. With purge-and-trap (P&T) for sample preparation, LODs of 25 pg/g and 5 ng/l were obtained for corks and wine, respectively. Trace amounts of TCA were detected in two out of fifteen corks analysed. Several other studies [30-32] also illustrate the extremely good performance of S- based GC–AED, for example in the analysis of various volatile (di)sulphides and related compounds in wine, beer and coffee powder down to less than 1 ng/l. In one such study [30], S-based GC–AED was used to locate fifteen volatile S-containing compounds in various types of garlic sample. The C/H/S ratios and the subsequently recorded mass spectra were successfully used to identify all but two of them. 2.2.2 Water On the basis of many national and international directives, threshold values for the concentration of organic micro-contaminants in raw water used for the production of drinking water, and the final product itself, typically are in the sub-µg/l range. Trace
23 2. GC–AED a powerful technique
enrichment is therefore a key issue and, also, the percentage of the analytes contained in the sample taken, that is injected onto the chromatographic column. Today, non-selective solid-phase extraction (SPE) on a copolymer sorbent and, occasionally, solid-phase micro-extraction (SPME) have largely superseded classical liquid–liquid extraction (LLE) procedures. There is no need to go into details regarding the various procedural improvements introduced in the early 1990s. These can be summarised as follows: classical 1-μL injections out of a 1-ml extract representing a 1-l sample (e.g. [35]) were increasingly replaced by large-volume injections of 10–100 μl (with proper venting ensuring there are no adverse effects such as flame-outs [36]) and, subsequently, by off- line and, finally, on-line SPE–GC–AED procedures. With the last-named approach, when the entire amount of analytes contained in the sample is injected, LODs (P channel) for a series of organo-P pesticides were found to be 30–150 ng/l (off-line SPE; 10-ml samples), 10–30 ng/l (on-line SPE; 10 ml) and as low as 1–2 ng/l (on-line; 100 ml) [37]. For obvious reasons, it is the SPE-based options which are used in most modern ultra-trace-level GC–AED studies. The studies quoted above mainly deal with semi-volatile target analytes (Table 2.4), but the determination of volatile compounds by means of P&T or static headspace has also attracted attention. In one paper [38], sixteen volatile organic compounds (VOCs) were analysed by means of P&T–GC–AED (using a dryer to prevent plasma destabilisation caused by water vapour carried over from the P&T unit by the purge gas). Element-selective detection (C, H, Cl, Br) was in most instances fully successful and empirical formulae calculated generally reliable. LODs were in the 30–400 pg/l range for 5-ml samples AED/MS detection. With several of the examples discussed earlier, it was indicated that the potential of GC–AED can be considerably enhanced if complementary structural information is provided by MS detection. This approach is used in close to half of the research papers cited in this review. A particularly rewarding strategy—especially for non-target compounds (Table 2.5)—is firstly to screen for certain elements of interest and, next, use MS for identification. The elemental composition data then help to reduce the length of the hit list and, hence, simplify the MS identification process. Careful matching of the two sets of chromatographic data is extremely important when using this approach. If two different GC systems are used, this can be achieved by retention-index (RI) based data correlation or by matching retention times with RTL (cf. [26]).
24 2. GC–AED a powerful technique
Table 2.4: Target analysis of various compounds in water using GC–AED Sample(s) Techniques Comments Refs. Water, beverages, P&T 10 volatile organohalogens and chlorophenols; LODs, [39-41] soil 0.05–0.5 µg/l Surface water SPME, SPE Small set of pesticides; LODs, 0.5–5 µg/l (1-l sample) [35] Surface water, LLE 145 targets (volatiles, haloethers, chlorobenzenes, nitro- [42] suspended matter compounds, anilines) with GC–MS and complementary AED, NPD and ECD Surface water SPE, LLE, LVI LODs for OPPs: 0.1–0.5 µg/l (10-ml sample) [36;43] Surface and On-line SPE, On-line SPE–GC–AED with 100 µl LVI; P186 LOD for [37] wastewater LVI OPPs: 1–30 ng/l Surface water SBSEa 8 OPPs (including polar fenamiphos); P186 LODs: 0.8– [44] 15 ng/l Surface and reagent LLE Large set of N-containing herbicides; compound- [45] water independent calibration (CIC); N LODs: 30–200 ng/l (3-l sample) Drinking water Derivztn., SPE 8 chlorophenols; LODs: 0.05–0.2 µg/l [46] Distilled water P&T 16 volatiles; CIC and empirical formula determination; [38] LODs: 0.03–0.4 µg/l Aqueous solution P&T Photocatalysis of methyl-t-butyl ether; six transformation [47] products identified Water LLE 14 pesticides; LODs: 0.04–0.2 µg/l (25-ml samples with [48] 5-µl splitless injections) Snow, glacial ice SPE, LLE Chloroacetates [49] Wastewater SPME Metazochlor [50] a. Stir-bar sorptive extraction Even so, with very complex samples, the proper matching of element-selective peaks with those in the crowded full-scan mass chromatogram easily creates problems. The best method to obtain the required high AED/MS data correlation, is to use a single GC and split the eluent for parallel AED/MS detection. After an early report by Hooker et al. [51], this approach was re-introduced in 1999. Firstly, a protocol for the (non-target) screening of hetero-atom-containing microcontaminants was designed by still using two separate GC systems, and applied to tap and waste water. Because of the use of on-line SPE–GC, 10–50 ml samples sufficed to reach LODs as low as 20–500 ng/l [52]. Next, fully integrated SPE–GC with parallel AED/MS detection was presented. This reduced the earlier observed retention time differences of up to 9 s, to 0.5 s or less. The practicability of the approach was demonstrated for river water and vegetable extracts and compound-independent calibration showed good agreement with MS quantification [53]. An improved set-up was used to analyse influents and effluents from a sewage treatment plant. In such complex samples many peaks of interest will go unnoticed in full-scan GC–MS because they are obscured by much larger co-eluting interferences and/or background signals. This is demonstrated in Fig. 2.1 which shows blow-ups of parts of the elemental traces, and the total ion chromatogram (TIC) as well as relevant mass traces of an influent sample. It is clear that even the small peak of chlorpyrifos
25 2. GC–AED a powerful technique
(spiked at 1 µg/l) would probably have gone unnoticed during non-target analysis and the same is true for alachlor. From among several non-target compounds which were identified, two are included in the figure. Peak 11 was identified as 5-chloro-2-(2,4- dichlorophenoxy) phenol or triclosan, a disinfectant, and peak 5 as tris(2-chloroethyl) phosphate. In both instances, no peak was visible in the TIC [54].
5 9 N P
Cl Cl
m/z=249 m/z=314
TIC TIC 11.4 12.4 12.5 13.5
8
N 11
Cl Cl
m/z=288 m/z=237
TIC TIC
13.1 14.1 12.1 13.1 Time/min Fig. 2.1: Blow-ups of parts of GC–AED/MS chromatograms showing obscured peaks of chlorinated compounds in the TIC trace and the corresponding extracted ion chromatograms and AED elemental traces. Peak assignment: (5) tris(2-chloroethyl) phosphate; (8) alachlor; (9) chlorpyriphos; (11) 5-chloro-2-(2,4- dichlorophenoxy) phenol [54].
In another study both SPE and SPMD (semi-permeable membrane devices) were used in order to extend the compound polarity range usually associated with the former technique, in an automated GC–MS-based screening study of river water for over 400 organic micro-contaminants [55]. Biomimetic SPMD, essentially a thin-walled polyethylene tube filled with a fatty substance, mimicks the absorption in fat-containing aquatic organisms such as fish, which is the basis of so-called bioconcentration. When the various AED traces were carefully searched for relevant—i.e. mainly chlorinated and brominated—‘unknowns’, several Br-containing compounds were detected and tentatively identified in an estuarine water sample (Fig. 2.2). Some of these are reported
26 2. GC–AED a powerful technique to be naturally formed by marine organisms [56], which explains their absence in river water more upstream.
Br 211 Br O Br 212 Br Br Br C C 213 NH2 H N Br H Br N 210 H O H Br
TIC 8 10 12 14 16 18 20 22 24 26 28
Fig. 2.2: Full-scan GC–MS chromatogram (bottom) and GC–AED bromine trace (top) of an SPMD extract of estuarine river water. For compound 213, two isomers were identified [55]. Table 2.5: Analysis of non-target compounds in water by combined GC–AED and GC–MS Sample(s) Techniques Comments Refs. Surface water SPE Trace pollutants and pesticides; selection rules and [57;58] formula calculation Wastewater On-line SPE, LVI On-line SPE; partial formula calculation; LODs: 0.05 [52] (tapwater)–0.5 µg/l (wastewater) Rain, snow Various Identification of chlorinated compounds by (Py)–GC–MS [59-61] and –AED Surface water SPE, LVI Parallel AED/MS detection; LODs: 0.02–0.5 µg/l [53] Wastewater Continuous LLE Chlorinated sulphur compounds in pulp-mill effluent with [62] GC–MS and –AED Raw water LLE Cl and Br containing semi-volatile compounds after [63] chlorination procedure; plus MS Coal wastewater LLE and SFEa Analysis of highly contaminated waste water and [64;65] sediment by (Py)–GC–MS and –AED Wastewater SPE Parallel AED/MS detection; comparison of treatment [54] plant influent and effluent water Seawater LLE Products of benzothiophene photooxidation [66] a. Supercritical fluid extraction 2.2.3 Soils and sediments The isolation of micro-contaminants from soils and sediments is more difficult than from water because of stronger interaction with the matrix. In addition, background interferences due to humic substances are much higher. Therefore, after solid–liquid extraction (SLE), clean-up with gel permeation chromatography (GPC) is usually necessary to make the final method more reliable and robust. An overview of studies using GC–AED for this type of analyses is given in Table 2.6; some illustrative examples are briefly discussed below. In a study [67] of a highly polluted harbour sediment sample, some 120 ‘major’ peaks were observed by GC–MS screening in the TIC mode. For all but ten compounds
27 2. GC–AED a powerful technique
high-match-quality provisional identification was found to agree with the elemental information obtained by GC–AED. In two (out of these ten) cases, AED enabled positive identification; in all others, additional information was provided but full characterisation was not possible. In addition, by comparing the Cl-trace AED results and a reconstructed PCB-targeted ion chromatogram, the presence of several ‘non-PCB’ compounds was revealed—one of these, nonachlorodiphenyl ether, being reported for the first time in a marine sediment [67]. In a related paper [68], the same group of authors showed that the selectivity of the AED was much higher than that of ECD and even high-resolution MS detection due to matrix problems. For reliable analysis of the highly polluted sample, the raw extract had to be subjected to GPC to remove elemental sulphur. The relative insensitivity of AED-based Cl detection was overcome by using 15 g of dry sample; individual CBs could then be detected down to the 5–20 µg/kg level [68]. As regards pesticides, Bernal et al. [69] determined LODs and retention indices on GC–AED and GC–MS for 181 phytochemicals in standard solutions. Using liquid extraction and SPE clean-up, 90 samples were analysed revealing the presence of about 30 different pesticides in total. In another study [70] soil samples were first cleaned-up with an acidic solution before extraction, and various pesticides could be detected with LODs in the low µg/kg range. One research group used the combined AED/MS strategy to study naturally occurring halogenated compounds. The presence of several chlorinated aromatic substructures in macromolecular organic matter derived from various types of decaying plant material and soil was established after chemical [71] and thermal [72] degradation. Among the compounds detected in both studies were several chlorophenols and substituted chlorobenzoic acid methyl esters. Table 2.6: GC–AED analysis of soils and sediments Sample(s) Techniques Commentsa Refs. Spiked soil SLE Retention data of 181 pesticides; validated recoveries for 11 [69] spiked pesticides Marine sediment SLE Non-target organic micropollutants [67] Marine sediment SLE Comparison of GC–MS, –ECD and –AED for detection of [68;73] PCBs Soil SLE 10 pesticides in soil; LODs: 2–5 µg/kg soil [70] Sediments, sludges Pyrolysis Differentiation of sludges; measurement of adsorbed/bound [74;75] chlorine ratio Soil, decaying plant Chemical Natural chlorinated matter in plant material and soil [71;72] matter degradation, pyrolysis Soil SLE Natural small halogenated compounds in forest soil [76] Marine sediment SLE Identification of sulphur compounds in marine sediments [77] a. In all studies except [70], GC–MS was used additionally.
28 2. GC–AED a powerful technique
In a related study [76], low-molecular-weight halogenated compounds were shown to be present as such—i.e., not as part of macromolecules—in forest soil collected in the vicinity of a fungus. Fourteen compounds were detected, and identified as halogenated and methoxylated benzaldehydes, benzoic acids and benzenes. The chlorinated compounds were found to be present in concentrations of up to 20 mg/kg of soil; the three brominated compounds in the set were present in the soil at 0.1–2 mg/kg levels. For both classes, additional GC–MS data were used to achieve unambiguous identification. Some selected structures are given in Fig. 2.3 which shows the three relevant AED traces. According to the authors, this is the first report on the natural occurrence of low-molecular-weight brominated compounds in terrestrial soil [76].
Fig. 2.3: GC–AED of neutral components in a sample collected at the arc of a Lepista nuda fairy ring. The carbon (496 nm), chlorine (479 nm) and bromine (478 nm) traces are shown. The internal standard (IS) was 1- chlorotetradecane (100 ng/µl) [76].
29 2. GC–AED a powerful technique
2.2.4 Air and gaseous samples Several studies in which GC–AED has contributed to the detection of contaminants in air and gases are included in Table 2.7. Becker et al. [78] used GC–AED to detect S-containing (thiaarenes or PASHs) polyaromatic hydrocarbons (PAHs) in the workplace air of an aluminium melting facility. The S trace revealed the presence of over 130 S-containing compounds. Some fifty of these were identified as thiaarenes by utilising complementary information derived from GC–MS. Three benzonaphthothiophene isomers were found to account for 35–40% of the total PASH concentration. The main advantages of AED detection were that (i) compound-independent response allowed reliable quantification, even in the absence of many standards, and (ii) PAHs did not interfere in the detection. The LODs were as low as a few ng/m3 based on a sampling volume of less than 1 m3. The merits of SPME-based air sampling with subsequent GC–AED were evaluated for organoleptic volatile sulphur compounds such as methanethiol and dimethyl sulphide. Although SPME is an attractive means of sampling since the fiber itself acts as a passive sampler and no equipment such as pumps and flow controllers are required, and LODs in the 5–50 nL/m3 range could be achieved, the authors emphasise that the technique is suitable only for qualitative analysis: several rather serious experimental problems (humidity, fiber quality, loss of analytes) made accurate quantification impossible [79]. Table 2.7: Analysis of air and other gaseous samples by GC–AED Sample(s) Techniques Commentsa Refs. Spiked air SPME Organo-S volatiles; LODs 4–50 ppt; storage stability [79] found to be low Workplace air Glass fiber filter in Detection of PAHs and PASHs in air of aluminum [78;80;81] personal sampler melting plant Landfill and Canister sampling VOCs and siloxanes in landfill and sewage gas with [82] sewage gas parallel GC–AED and GC–MS (split after the injector to two analytical columns) Landfill gas Sorbent trap S-containing volatiles [83] Gas standards Loop injection Fluorocarbons; additional ECD and MS; AED superior [84] to ECD, especially for Cl-free compounds (Table 2.8) Air Cryogenic Field application of GC–AED for analysis of [85] trapping atmospheric sulphur gases; LODs: 0.3 nmol/m3 a. In all studies except [85], GC–MS was used additionally. AED-based detection of fluorine is poorer than that of most other elements (cf. Table 2.1). However, GC–AED has been shown to be the preferred choice for the detection of alternative fluorocarbons (AFCs) in air. AFCs were developed as refrigerant coolant fluids which can replace the ozon-depleting chlorofluorocarbons (CFCs). The latter compounds can be detected sensitively by means of GC–ECD. However, with the AFCs, especially those not containing chlorine, the detectability is 100–1000-fold poorer. As is
30 2. GC–AED a powerful technique illustrated in Table 2.8, the detection problem can be overcome by using GC–AED instead. The LODs using the F-channel are only twice as high as those found by means of GC–MS; however, with the latter technique the high element selectivity is lost. Based on predicted atmospheric concentrations, one can estimate that 1–10 l samples are required for proper monitoring [84]. Table 2.8: LODs (ppt, v/v) of CFC and selected AFCs with GC and various detectorsa Compound ECD AED MS CFC
CCl2F2 0.2 0.5 0.2
Cl-containing AFCs
CHClF2 10 0.3 0.2
CHCl2CF3 2 0.2 0.1
CHClFCF3 6 0.3 0.2
Cl-free AFCs
CHF2CF3 1100 0.4 0.3
CH2FCF3 4100 0.4 0.2
CH3CHF2 nd 0.5 0.2 a. Data for 1-l samples; nd, not detected [84] 2.2.5 Petrochemical samples Most applications of GC–AED in the field of petrochemical samples focus on the detection of S- and N-containing compounds. For three reasons they cause concern: the presence of polluting SO2 and NOx in exhaust gases, the reduction of catalyst life-time due to poisoning and an adverse effect on product stability [86]. Catalytic hydrotreating is the subject matter of many AED-based studies although other methods such as chemical, photochemical and bacterial processes have also been studied (Table 2.9). A typical example obtained by split injection of an untreated medium cycle oil (MCO) is shown in Fig. 2.4. The major S-containing species found were dibenzothiophene (DBT), monomethylated (MDBT) and dimethylated (DMDBT) DBT.
Small amounts of C3- and C4-alkylbenzothiophenes were also detected. N-containing species found included carbazole, mono- and dimethylcarbazoles, and di- and trimethylindoles [87]. The effects of various types of catalytic hydrotreatment are shown in Fig. 2.5. Single-stage hydrotreatment removed 75% of sulphur, but an additional 30-min treatment only resulted in slight improvement, probably due to strong inhibition by the products and/or catalyst deactivation. After renewal of the hydrogen atmosphere to remove H2S and NH3 (the products of the process) and addition of fresh catalyst (stage B, Fig. 2.5), a 97% reduction was achieved. Even in the latter stage, nitrogen could only be reduced by 66%, which illustrates the relative difficulty of hydrodenitrogenation.
31 2. GC–AED a powerful technique
Fig. 2.4: C-, S- and N-trace chromatograms of an MCO recorded by GC–AED. Cn: normal paraffin with n, number of C; Cz, carbazole; C1-Cz, monomethyl carbazole; C2-Cz, dimethyl carbazole [87].
Fig. 2.5: Hydrotreatment configuration (left) and comparison of sulphur species in an MCO as a function of reaction method (right) [87].
32 2. GC–AED a powerful technique
Table 2.9: GC–AED analysis of petrochemical samples Sample(s) Techniques Comments Refs. Coal Pyrolysis Thermal desorption followed by pyrolysis in one system [88] Shale oil SPE, LCa PASHs; RIs of 93 thiophenes on three GC columns [89;90] Light cycle oil None Large group of thiophenes; plus GC–MS [91] Light cycle oil Dilution Identification of 90 organo-S; structure vs. retention time [92] Crude oil, tar balls SPE PASH distribution; relative abundance used to distinguish [93;94] sample types; plus MS Gasoline Derivztn. Alkenes after bromination and selective detection (Br); plus MS [95]
Coal SFE Elemental sulphur (S8) [96] Various On-line LC Heterocyclic polyaromatics by LC–GC–AED and –MS [97] Crude oil, coal tar SPE, LC Identification of PASHs; three GC columns, including liquid [98] crystalline phase; plus MS Crude oil Derivztn. Phenols and alcohols as ferrocenecarboxylic acid esters; Fe [99;100] detection Naphtha Oxidation Selective oxidation to differentiate thiophenes and other [101] sulphides FCC product None Identification of sulphur compounds with GC×GC–AED and [102] GC×GC–MS Various Various Effects of catalytic hydrotreating; plus MS [86;87;103 -110] Vacuum gas oil SPE Photochemical desulphurisation and denitrogenation; plus MS [111] Vacuum gas oil SPE Desulphurisation and denitrogenation by methylation and [112] precipitation as tetrafluoroborates; plus MS Various Various Bacterial desulphurisation; plus MS [113-115] a. Liquid chromatography. Wiwel et al. [103] used the 388-nm nitrogen line to assess compositional changes of N- containing compounds in diesel-range gas oils during hydrotreating. Some 60 N- containing compounds were identified. By applying pure-silica-based SPE trace enrichment, the authors could quantify individual nitrogen compounds down to the 50 μg N/l level. In many studies, a fractionation step is included to improve the analytical performance. On-line fractionation by LC–GC–AED and LC–GC–MS was used by Lewis et al. [97] to detect and quantify polycyclic aromatic compounds in e.g. fuels and their combustion products. The use of multidimensional chromatographic techniques proved effective in producing on-line fractions of compounds of a particular chemical class. Moreover, detection of trace-level species was much improved. Among the compounds identified were several carbazoles, benzothiophenes and substituted dibenzothiophenes. Another interesting application is the rapid determination of alkylphenols in non- polar samples such as crude oils. Derivatisation with ferrocenecarboxylic acid chloride and subsequent GC–AED at 302 nm for Fe-selective detection was found to yield an LOD of 0.2 pg Fe. Consequently, sample amounts of less than 1 mg suffice for the ng/g detection of the target analytes. The total sample preparation takes 45 min, which is
33 2. GC–AED a powerful technique
much less than with conventional procedures. As an application, twenty C0–C3- alkylphenols were quantified in shale oil and crude oil [100]. Another study that combines the selectivity of AED detection and derivatisation used bromination to determine alkenes in complex mixtures of aromatic and saturated hydrocarbons. Detection of the formed dibromoalkanes using the Br channel was found to be essentially free of interferences, with sub-ng LODs [95]. AED was used as a detector for comprehensive two-dimensional gas chromatography (GC×GC) by van Stee et al. [102]. The performance of GC×GC–AED was evaluated with a pesticide standard, and as a real-life sample a fluidised catalytic cracking (FCC) product was analysed. An overlay of the S and C traces with some well known classes of S-containing compounds is shown in Fig. 2.6. Analyte identification was possible by combining GC×GC–AED and GC×GC–MS data, and demonstrated for a compound not part of the quoted classes which was tentatively identified as phenathro[4,5-bcd]thiophene.
9.80 FCC product BNTs 8.40 DBTs 7.00
5.60
BTs 4.20
2.80
1.40 Cx 11 14 18 22 26 30 n-alkane 0.00 100 500 900 1300 1700 2100 2500 2900 3300 3700 4100 4500 Fig. 2.6: GC×GC–AED chromatogram of an FCC product. The S and C channels are represented in orange and blue, respectively. The boxes indicate some known classes of S-containing compounds: BT, benzothiophenes; DBT, dibenzothiophenes; BNT, benzonaphtothiophenes [102]. 2.2.6 Synthetic polymers Pyrolysis–GC (Py–GC) is often used to unravel polymer composition and there are several papers in which AED detection is shown to be of significant help in this regard (Table 2.10). To quote an example, polyvinyl alcohol (PVA) is a water-soluble synthetic resin used amongst others in various types of adhesives to which a small amount of polyacrylamide is often added. Analysis is difficult because separation cannot easily be achieved, and most non-destructive spectrometric techniques suffer from a lack of
34 2. GC–AED a powerful technique sensitivity or from interference. Wang [116] showed that Py–GC–AED and Py–GC–MS enable the detection of 1% polyacrylamide by monitoring the fragments by means of N trace AED (Fig. 2.7). The same author also used Py–GC–AED to detect copolymerised acrylamide [117] and the widely used co-monomers of fumaric and itaconic acid [118], with analysis of the acids involving derivatisation with a primary amine to form a cyclic imide. In another study, Wang [119] used the same analytical approach to identify the type of brominated polymeric flame retardants used in thermoplastics by means of peak- pattern recognition through a halogen-element AED trace.
Fig. 2.7: N trace AED pyrograms (plus blow-ups of relevant ranges) of polyvinylalcohol with 1% polyacrylamide (left) and without polyacrylamide (right) [116]. Table 2.10: Analysis of synthetic polymers by pyrolysis GC–AEDa Sample(s) Techniques Comments Refs. PVA Py–GC Polyacrylamide in PVA by Py–GC [116] Latex Off-line Copolymerised acrylamide in latex [117] pyrolysis Latex Derivztn., Py– Copolymerised fumaric and itaconic acids by derivatisation [118] GC with amines Various Py–GC Brominated polymeric flame retardants in various [119] thermoplastic resins Various Off-line Effect of brominated flame retardants and PVC on thermal [120] pyrolysis degradation of ABS Epoxy resins Py–GC Cured epoxy resins; plus GC–FT-IR [121] Adhesive tape Py–GC Characterisation of adhesives [122] Chitin/PVC blend Py–GC Thermal degradation products of chitin/PVC blend [123] a. In all studies GC–MS was used in addition to GC–AED Today, more and more plastic waste is accumulating, which poses serious problems to the environment. An efficient way to recover the material is pyrolysis, since only 10% of the energy content of the waste is used to convert the scrap into valuable hydrocarbon products. Brebu et al. [120] used GC–MS and GC–AED (C, O, N, Cl and Br traces) to study the products formed during off-line pyrolysis of acrylonitrile-butadiene-styrene
35 2. GC–AED a powerful technique
(ABS) and brominated ABS as individual polymers or mixed with polyvinyl chloride (PVC). Thermal degradation at 450ºC led to pyrolysis oils rich in valuable benzene derivatives but, also, significant amounts of halogenated phenols and benzenes, and N- containing organics. Catalyst development to remove these compounds from liquid products is therefore indicated. 2.2.7 Chemical warfare agents (CWAs) One application area in which GC–AED is increasingly being used, is that of the detection and identification of CWAs and/or related compounds (see Table 2.11). In one recent study [124], a sludge sample from an old ton container used to store chemical warfare material, was extracted and derivatised with 1,3-propanedithiol to look for the presence of lewisite-1. This compound was indeed detected by GC–AED (Cl and As traces) with subsequent identity confirmation by means of MS. Three isomers of lewisite-3 and two isomers of a dimer were also detected. The same strategy was used to study ozone-based surface decontamination of equipment exposed to CWAs; the persistent nerve gas VX was selected as CWA material. Combined AED/MS allowed screening of the VX removal, detection and provisional identification of three reaction products, and the proposal of a reaction scheme. In another study [125], the same group identified a major and several minor P-containing compounds in a munition shell by the combined use of GC–AED, GC–IR–MS, LC–MS and stand-alone NMR. Fifty compounds were detected in a block of yperite (or sulphur mustard) fished up from the Baltic in 1997. Thirty compounds were identified using AED and MS data. Not too surprisingly, most of these contained sulphur and/or chlorine, and a few, arsenic. The final product of hydrolysis, thiodiglycol, was not detected: presumably, as a water- soluble compound, it was leached into the water phase [126]. Amongst the chemicals that can be found in CWA-related waste are very reactive acylating species such as cyanogen chloride, phosgene and chloroformates. Because they are hydrolysed very easily, it is extremely difficult to perform sampling under conditions which do not cause loss of analytes. Schoene et al. [127] demonstrated the use of in situ sampling/derivatisation and subsequent analysis by GC–AED and GC–MS. The compounds were derivatised with dibutylamine that was applied as a coating on the solid-phase material inside the sorbent tube. Standard atmospheres were sampled and empirical formula calculation together with MS data was used to identify the compounds. For ten compounds a 25–90% recovery was found, but it was as low as 8% for one compound after 14-day storage of the tube. As indicated by the authors, optimisation of sampling and storage conditions may be required for several individual compounds.
36 2. GC–AED a powerful technique
Table 2.11: GC–AED analysis of CWAs and related compoundsa Sample(s) Techniques Comments Refs. Munition shell Various Characterisation of a yellow liquid from shell using many [125] analytical techniques Water, wipes, soils SPE, Quantitation of alkyl methylphosphonic acids by [128] derivatisation derivatisation Water, soil, toxic Various Various CWAs and degradation products; plus GC–IR [129;130] waste Yperite waste SLE Sulphur mustard and transformation products [126] CWAs LLE Products of sulphur mustard with decontamination fluids [131;132] Spiked extracts SLE, Development of derivatisation method for Adamsite [133] derivatisation Toxic waste LLE, Lewisite-1 and related compounds; empirical formula [124] derivatisation calculation Standard atmospheres Derivatisation Acylating gases and vapours [127] on sampling tube a. In all studies except [131;132] GC–MS was used in addition to GC–AED 2.2.8 Biological samples GC–AED has also been used for a wide variety of biological samples and compounds of interest (Table 2.12). Some examples are highlighted below. GC–AED was used to detect the methylsulphonyl metabolites of CBs and DDT in porpoise [134] and grey seal tissue [135]. The LODs of 0.5 ng/g were adequate for these samples; because of the high selectivity introduced by monitoring both the S and Cl channels, sample clean-up could be simplified. Complementary MS data were used to confirm the identification of the target compounds. The highest methylsulphonyl-CB concentrations were found in liver (0.15–0.5 µg/g lipid wt.); they represented about 2% of the total CB concentration. Polybrominated diphenyl ethers (PBDEs) are one of several classes of brominated compounds which are extensively used as flame retardants. Johnson and Olson [136] used GC–AED to detect and quantify BDEs and CBs in fish tissue. The high Br/Cl selectivity is clearly demonstrated in Fig. 2.8. The results also show that tetra and penta isomers were the major BDEs present. It is interesting to note that total BDE concentrations detected in a fish from a remote spring-fed river were 1000-fold lower than in a fish from a river in an urbanised area, i.e. 1.4 and 1250 µg/kg wet weight, respectively.
37 2. GC–AED a powerful technique
Fig. 2.8: Cl (top) and Br (bottom) GC–AED traces of fish tissue showing CBs and BDEs, respectively. Surrogate standards (S.S.), tetrachloro-m-xylene (TMX), decachlorobiphenyl (DCB), and dibromooctafluorobiphenyl (DBOB) were added to sample prior to extraction. Total CB and BDE concentrations were 130 and 150 µg/kg, respectively [136].
Another interesting application of GC–AED is in isotope analysis. Small differences in the properties of isotopes can be exploited to detect stable isotope-labelled compounds. One practical example is the molecular band of CO, which is formed inside a plasma 12 13 when using O2 and H2 as reactant gases. The second-order lines of CO and CO are 342.574 and 341.712 nm, respectively, whence a 0.86 nm difference. An algorithm, ‘SUPPRESS’, developed by Quimby et al. [137] calculates the real-time contributions from 12C and 13C, and is able to produce chromatograms only showing peaks that have 13C enrichment higher than the natural abundance (1.1%). Boukraa et al. [138] used this method to detect metabolites of caffeine in human urine (Fig. 2.9). When using standards and complementary MS detection, at least ten metabolites could be detected and identified. In a completely different type of application, elemental white phosphorus was determined in ducks found in an estuarine salt marsh located in an artillery range. Because the phosphorus was submerged, it remained as elemental P in the sediment and was subsequently taken in by the sediment-feeding ducks. The advantage of using AED is that a stable organo-P compound such as triethyl phosphate can be used as a standard, instead of elemental phosphorus which is more difficult to prepare. The highest content of phosphorus was ca. 6 mg in the gizzard of one duck [139].
38 2. GC–AED a powerful technique
b a
Fig. 2.9: GC–AED profiles obtained from a urine extract: (a) before caffeine intake, (b) from the urine of a subject taking [13C]-caffeine. Profiles are the results of ‘SUPPRESS’ data processing [138]. Identities of some relevant compounds: (1) Caffeine; (3) 1,7-dimethylxanthine; (3') 3,7-dimethylxanthine; (7) 1-methylxanthine.
Table 2.12: GC–AED analysis of biological samples Sample(s) Techniques Comments Refs. Fish oil, cow fat SLE PCB-contaminated and non-contaminated samples; linearity [140] and CIC evaluated; LOD 0.15 mg/kg; plus ECD Grey seal tissues SLE, GPC 24 methyl sulphonyl PCBs using S and Cl channels; plus [135] ECD Harbour porpoise SLE, dialysis, PCBs, DDTs and methyl sulphone metabolites in various [134] GPC tissues; plus MS Mouse liver LLE Method development for detection of low mol. wt. silicones [141;142] in tissue; plus MS Duck gizzard SLE Quantification of elemental P in duck gizzard [139] Fish tissues SLE PBDEs; extremes found: 1.4 and 1250 µg/kg, in fish caught [136] in remote and urbanised area respectively (Fig. 2.8) Human urine LLE, 13C-labelled caffeine metabolites in human urine [138;143; derivatisation 144] Bacteria None Differentiation of bacteria using Py–GC–AED and statistical [145;146] pattern recognition Plant leaves SLE, Long-chain Halowaxes in halophytes; plus MS [147] derivatisation Essential oil Steam S-containing compounds in essential oil of Tagetes; plus GC– [148] distillation MS and FT-IR
2.3 Conclusions Combining a GC separation on-line with AED detection provides a powerful means to screen for the presence of hetero-atom-containing organic micro-contaminants in a wide variety of complex samples. Selectivity is excellent and LODs are in the low pg/s range for many elements of interest. The detection technique enables the calculation of (partial) molecular formulae and (semi-) quantification can be achieved even for non- target analytes by means of compound-independent calibration. If simultaneous AED- plus-MS detection is used, unambiguous identification is possible in many instances.
39 2. GC–AED a powerful technique
The selected applications discussed above, and many others of a similar nature published in the scientific literature, as well as those dealing with the detection—and, frequently, speciation—of organometals and organometalloids [8-10] convincingly demonstrate that GC–AED is a powerful hyphenated technique that is to be recommended especially for the analysis of complex and/or highly contaminated samples. In this context, a distinct advantage of AED over MS is that non-target screening can be carried out for individual elements.
40 3 3. Identification of non-target compounds using gas chromatography with simultaneous atomic emission and mass spectrometric detection (GC–AED/MS): analysis of municipal wastewater
3.1 Introduction The concern over the presence of micropollutants in the aquatic environment causes a growing demand for methods of identification of organic contaminants which are present at relatively low concentrations in water. In environmental analysis, gas chromatography (GC) has long been used in many standard target methods and for sample screening purposes. Since almost all pesticides, and many of the organic compounds of the US Environmental Agency (EPA) priority pollutants list contain heteroatoms, it is clear that element selectivity is a desirable characteristic for environmental analysis. The most common element-selective detectors like the nitrogen–phosphorus detector (NPD) or the electron-capture detector (ECD) still detect several elements and the element-to-carbon selectivity sometimes is modest. The atomic emission detector (AED), on the other hand, can detect all elements except helium separately. This classifies it as the most versatile selective detector. Regardless of the selectivity of a detector, and the use of retention data on two columns with different stationary phases, state-of-the-art confirmation of target compounds, and identification of unknowns, requires a detector providing structural information. In this respect, mass spectrometry (MS), which is sensitive, offers several complementary modes of operation, and is compatible with GC, is the technique of choice. The combined use of AED and MS detection allows selective screening for compounds containing (a) specific element(s) and, subsequently, provisional identification. Several studies have demonstrated that AED/MS detection indeed is a powerful technique in the analysis of unknowns [23;57;62;67;149;150]. However, in these studies separate chromatographic systems were used for GC–MS and GC–AED
Published as: L.L.P. van Stee, P.E.G. Leonards, R.J.J. Vreuls, U.A.Th. Brinkman, Analyst 124 (1999) 1547 3. Analysis of wastewater with GC–AED/MS
data acquisition, while one important aspect of the use of complementary data— especially when analysing complex samples—is the distinct need of a very close correspondence of the retention times of both systems. For optimum results, the set-up of the two systems should be identical. However, especially when using detectors operating at different pressures, retention indices should be used to minimise errors in peak allocation [150]. Another approach is to adjust the pressure programme of one of the GC systems in such a way that the resulting retention times exactly match those on the other system. With software developed by Hewlett-Packard (Retention Time Locking software) this can be achieved in an automated fashion [151;152]. However, some operational aspects and limitations have to be considered. First, five runs at different pressure programmes have to be performed to lock the systems, and regular checks have to ensure that they maintain locked. More importantly, very precise electronic pressure control is essential, so RTL is only applicable using the latest models GC and software which are not currently available in our department. A more elegant and even better solution, is to connect both detectors in parallel to one chromatographic system. In such a set-up, very small differences of 0.5–1 s between the recorded AED and MS retention times can be obtained [153]. In this work, the applicability of GC–AED/MS to screen for and identify heteroatom-containing compounds in a complex sample, in this case waste water, was studied. The water received by municipal sewage treatment plants mainly consists of domestic wastewater, run-off stormwater from roofs and roads, and often also industrial wastewater. Such influents typically contain over a hundred compounds with concentrations ranging from 0.5 to 5700 µg/l [154]. The practicality of two AED/MS strategies was evaluated. One strategy is to use the AED as a screening tool and, subsequently, to use the MS for their identification. The alternative is to use the AED elemental composition data to include or exclude compounds from the list of compounds provisionally identified by MS library searching in the first step. 3.2 Experimental 3.2.1 Chemicals HPLC-grade methyl acetate (J.T. Baker, Deventer, the Netherlands) was freshly distilled before use. HPLC-grade water was prepared by distillation of demineralised water. A solution of fluometuron, trifluralin, metobromuron, alachlor, chlorpyrifos and bromophosmethyl at concentrations of 0.8, 0.75, 0.75, 1.6, 1.0 and 1.1 µg/l, respectively, in methyl acetate was used as a standard; the last three pesticides are chlorine- containing. All pesticides were obtained from Riedel-de Haen (Seelze, Germany) and were of >99% purity.
42 3. Analysis of wastewater with GC–AED/MS
3.2.2 Equipment A Hewlett-Packard (Palo Alto, CA, USA) Model 5890 Series II gas chromatograph, a Hewlett-Packard 5971A mass selective detector and a Hewlett Packard 5921A atomic emission detector were used for GC–AED/MS detection. A Hewlett Packard 7673 autosampler was used for the introduction of 5 µl of sample extract. The pressure- programmable on-column injector was connected to 1 m × 0.53 mm i.d. diphenyltetramethyldisilazane (DPTMDS)-deactivated fused silica tubing (BGB Analytik, Zürich, Switzerland) which was connected to the GC column (HP5-MS, 24 m × 0.25 mm i.d., 0.25 µm film). A glass press-fit Y-piece at the end of the GC column split the flow through two pieces of deactivated fused silica tubing to the AED (0.25 mm i.d., 65.5 cm in the transfer line + 30 cm in the oven) and the MS (0.1 mm i.d., 28.3 cm in transfer line + 1.5 cm in the oven). Data acquisition and analysis was performed on a Pentium 150 MHz computer by running the G1701AA 3.00 and G2360AA 03.03 Hewlett Packard Chemstations simultaneously. Mass spectral libraries used were Wiley 6 and Nist 98. Solid-phase extraction was performed on a PROSPEKT sample preparation system of Spark Holland (Emmen, the Netherlands). The PROSPEKT system consists of three Rheodyne six-port valves, an automated cartridge exchanger and a solvent delivery unit consisting of a solvent selection valve and a single-piston LC pump. Two separate stainless-steel six-port valves were used to allow drying of the cartridge with nitrogen and transfer of the preconcentrated sample to a glass vial with a glass 250-µl insert. Methyl acetate for desorption was delivered by a Phoenix CU20 (Carlo Erba Instrumentazione, Milan, Italy) syringe pump. 3.2.3 Experimental conditions Gas chromatography. Helium (99.999%, Praxair, Oevel, Belgium) was used as carrier gas at 2.0 ml/min in the constant-flow mode. The oven temperature was 56°C during injection. After 4 min the temperature was programmed to 280°C at 20°C/min and finally held at 280°C for 6 min. The on-column injector was used in the oven track mode
(T = Toven + 3C). In all analyses 5 µl of sample were injected. Atomic emission detection. The sensitivity for nitrogen was increased by changing the usual flow-rate of 30 ml/min make-up to 20 ml/min (elongated dwell time of analytes in the plasma). The make-up and reagent gases used were: oxygen (99.999%, Hoekloos, Schiedam, the Netherlands) at 1.8 bar, hydrogen (99.999%, Hoekloos) at 4.1 bar, 10% methane /
90% N2 (Hoekloos) at 4.8 bar and helium (Praxair) at 4.1 bar. The make-up flow was set at 20 ml/min and the cavity pressure at 10 kPa. The temperatures of the heated zones
43 3. Analysis of wastewater with GC–AED/MS
were: transfer line, 300°C; cavity, 300°C and water, 65°C. The solvent venting was switched on 1 min after injection and switched off 4.5 min after injection. The spectrometer was purged at a flow-rate of 4 l/min. Standard recipes for elemental detection were used for all analyses. Mass spectrometry. Spectra were recorded over a scan range of m/z 45–350 amu at a scan rate of 2.3 scans/s and an electron energy of 70 eV. The temperatures of the heated zones were transfer line, 300°C and quadrupole, 188°C. Sample preparation. Influent and effluent samples from a municipal sewage treatment plant were filtered through a glass fibre filter (type GF/D, Whatman International, Maidstone, UK) and next through a 0.45 µm membrane filter (type HA, Millipore, Etten-Leur, the Netherlands). Aliquots of the filtered samples were spiked at the 1 µg/l level by adding 50 µl of the pesticide standard to 50 ml of sample. For SPE, the SPE cartridges were first conditioned with 2.5 ml of HPLC-grade water at a flow rate of 2.5 ml/min. Next, 25 ml of the water sample were preconcentrated on the 10 mm × 2 mm i.d. cartridge which contained 15–25 µm PLRP-S copolymer (LC Separations, Hendrik Ido Ambacht, the Netherlands) at a rate of 2.5 ml/min. After washing with 1.6 ml of HPLC-grade water at 2.5 ml/min, the cartridge was dried for 30 min with nitrogen at 50 ml/min. Finally, the analytes were eluted with 100 µl of methyl acetate into a vial with a glass 250-µl insert. 3.3 Results and discussion 3.3.1 Instrument set-up and detection limits The limits of detection (LOD) for the most common elements in GC-amenable compounds in GC–AED can be classified in three ranges, viz. low (2–5 pg; C193 nm, P, S); medium (30–100 pg; N, Cl, Br) and high (180–350 pg; F, O). Absolute elemental LODs obtained with the instrument at hand, together with literature data for comparison, are presented in Table 3.1. Comparison of several sets of results show that our AED
(which is a fairly old instrument) performs well for most elements, but not for the C193 and F channels. From the data presented in Table 3.1 one can calculate that LODs for pesticides will generally be from 50 pg to 5 ng, depending on the number of heteroatoms and the channel being monitored. More specifically, the LOD for compounds containing 10– 50% (w/w) of chlorine will be in the order of 0.1–0.5 ng on the chlorine channel. It is more difficult to give typical values for LODs in GC–MS, since much depends on the fragmentation pattern of the analytes of interest, and on the number of diagnostic ions required for reliable detection/identification. Still, it seems correct to say that the LODs or, rather, the limits of identification, for quadrupole MS instruments as used in the
44 3. Analysis of wastewater with GC–AED/MS present study, also are in the 0.1–0.5 ng range. That is, an AED/MS split ratio of about 50:50 is a good compromise. For practical reasons, viz. the required lengths of transfer lines and available dimensions of fused silica tubing, a configuration as described in Experimental was used. The set-up chosen provided a constant AED/MS split of about 60:40 over the temperature range used. Table 3.1. Elemental detection limits of GC–AED in this work compared with literature values LOD (pg)a LOD (pg/s)b Element Wavelength (nm) this work this work Ref. [155] Ref. [156] Ref. [157] Ref. [153] C 193.1 10 5 0.5 1 0.2 3.7 O 777.2 225 100 75 120 50 79 N 174.2 25 10 7 50 15 26 S 180.7 2 1 1.7 2 1 1.0 Cl 479.5 70 45 39 40 25 18 Br 478.6 100 45 75 60 30 - H 486.1 20 5 2.2 4 - 4.7 P 177.5 2 1 1.5 1 3 0.7 F 685.6 350 170 40 80 60 - a. LOD calculated using S/N=3. b. LOD using S/N=2, divided by full width at half height of peak. Values were converted to S/N=2 for Ref. [153]. Values of Ref. [155] are equivalent to AED Specification guide (Jan. 1990). Table 3.2 shows analyte LODs obtained for five pesticides (fluometuron is not included because of poor elution) by means of full-scan MS and AED for a 25-ml water sample. These illustrate the performance of the entire procedure, i.e. including SPE and the eluent split, and demonstrate that the present split ratio is indeed suitable. Table 3.2. LODs in g/l (S/N=3) for the spiked pesticides using a 25-ml samplea AED channel Compound Full-scan C H N O P S Cl Br F MS Trifluralin 0.4 0.03 0.6 0.3 2 - - - - 3 Metobromuron 0.3 0.04 0.5 0.6 3 - - - 0.6 - Alachlor 0.2 0.03 0.3 0.8 3 - - 0.7 - - Chlorpyrifos 0.2 0.05 0.7 0.5 3 0.03 0.04 0.2 - - Bromophosmethyl 0.6 0.06 2.0 - 3 0.04 0.05 0.5 0.4 - a. Determined from the analysis of influent sample spiked at 1 µg/l or, if necessary, spiked HPLC water (C and H) or by calculation (O and F) from the elemental detection limits from Table 3.1. 3.3.2 Identification of unknowns in complex samples When the retention data of both detectors were examined it was found that there was a constant difference in recorded retention times of 1.9 s during the whole run. The MS software allows overlaying of AED chromatograms, and only a small adaptation in the MS overlay macro that shifts one of the chromatograms in time, was required to eliminate this retention time difference. In the following discussion only the corrected retention times will be quoted. After overlaying, all the functions of the MS software are still operational; this means that a mouse click on an AED-recorded peak suffices to
45 3. Analysis of wastewater with GC–AED/MS
retrieve the MS spectrum for that compound. A major advantage of a single over two separate data systems is that, next to the retention times, the peak shapes can easily be compared. This allows a better recognition of co-elution and, hence improved background subtraction. A typical example of GC–AED/MS results from the analysis of an influent sample is shown in Fig. 3.1. When the full-scan MS chromatogram is studied in detail, some 150 peaks with S/N>3 can be counted. It is interesting to note that there are several distorted, or co-eluting peaks, which show a closely similar profile in the MS and several AED traces (see, e.g., peaks in 10–13 min range).
MS
C
H
O 3 9 2 6 8 10 1 4 5 7 11 Cl
Br
S
N
P 5 6 7 8 9 10 11 12 13 14 15 Time/min Fig. 3.1: Full-scan MS chromatogram and all recorded AED traces of influent water sample spiked at the 1 µg/l level with six pesticides as described in experimental. Sample volume, 25 ml; 5 µl of extract equivalent with 1.25 ml sample injected. Peaks Nos. 8, 9 and 10 originate from the three chlorinated pesticides spiked at 1 µg/l. See Table 3.3 for names of numbered chlorine peaks.
AED as screening tool. To illustrate the screening/identification strategy, the environmentally relevant chlorinated compounds were studied in more detail. In the spiked influent sample eleven chlorinated compounds were detected by GC–AED above the S/N=3 level (Fig. 3.1). Three of these (Nos. 8, 9 and 10) can be attributed to the chlorinated spiked pesticides; the other spiked pesticides will not be discussed. Detailed
46 3. Analysis of wastewater with GC–AED/MS
GC–full-scan MS and the Cl, P and N chromatograms are shown in Fig. 3.3. The figure clearly demonstrates the screening potential of the system. For example, the spiked chlorpyrifos produces a distinct peak (No. 9) in the chlorine trace. However, the small peak in the MS trace would probably have gone unnoticed if only MS detection would have been used. After background subtraction using the baseline of the Cl trace, MS library searching indeed confirmed the identity as chlorpyrifos (Fig. 3.2A).
A Cl EtO N P O Cl 5 Abundance 9 97 EtO S 199 N Cl P
65 125 258 314 47 286 Cl m/z 40 80 120 160 200 240 280 320 Cl Abundance 97 199 m/z=249 m/z=314 258 314 65 125 286 47 243 TIC TIC m/z 40 80 120 160 200 240 280 320 11.4Time/min 12.4 12.5Time/min 13.5 B O ClH2C H2COP O CH2 CH2Cl Abundance 8 O 63 CH2 CH2Cl 143 249 N 83 99 205 223 11 49 161 235 Cl m/z 40 60 80 100 120 140 160 180 200 220 240 Cl Abundance 63 249 143 205 m/z=288 m/z=237 99 221 45 81 161 235 m/z 40 60 80 100 120 140 160 180 200 220 240 TIC TIC 13.1Time/min 14.1 12.1Time/min 13.1 Fig. 3.2 A: spectrum (top), library spectrum (bottom) and structure of chlorpyrifos (spiked at Fig. 3.3: Blown-up GC–MS and GC–AED 1 µg/l, peak No. 9 in Fig. 3.1 and Fig. 3.3). B: chromatograms (cf. Fig. 3.1) showing obscured spectrum (top), library spectrum (bottom) and peaks of chlorinated compounds in the MS TIC structure of tris(2-chloroethyl) phosphate (peak chromatogram with their corresponding extracted No. 5 in Fig. 3.1 and Fig. 3.3). ion chromatograms and AED elemental traces.
As an example of a non-spiked compound, peak No. 5 clearly indicates the presence of a chlorinated compound, which also showed a response in the P trace. Again, partly obscured by other peaks, the small peak visible in the MS full-scan chromatogram was not readily recognised. After background subtraction of the MS spectrum at the retention time of the chlorine peak, the compound was identified by MS library searching as the flame retardant tris(2-chloroethyl) phosphate (Fig. 3.2B). Although the MS hit quality was rather low, the additional data about elemental composition helped to confirm the correctness of the identification: the P trace shows a peak with exactly the same shape and at the same retention time as the extracted ion chromatogram (m/z=249) of the provisionally identified compound. Other examples of chlorinated compounds that are
47 3. Analysis of wastewater with GC–AED/MS
essentially obscured in the full-scan MS traces are peaks No. 8 and No. 11. The latter peak is seen to coelute with a compound with a slightly higher retention and cannot be recognised as a separate peak in the full-scan MS trace at all. However, an MS library search of the background-subtracted spectrum allowed identification as 5-chloro-2-(2,4- dichlorophenoxy)phenol (Triclosan). Using the present strategy, seven of the eight non- spiked, i.e. unknown, compounds could be identified. Relevant details are given in Table 3.3. Table 3.3. Identity of the chlorinated compounds identified in the influent sample of figures 3.1, 3.2 and 3.6 Peak No. Compound Ret. time Use or origin (min) 1 trichlorophenol 9.53 depends on isomer 2,3 dimethyl-chlorophenol 9.80 - 4 unknown 10.50 - 5 tris(2-chloroethyl) phosphate (Fyrol CEF) 11.92 flame retardant 6 tris(2-chloroisopropyl) phosphate (Fyrol PCF) 12.08 flame retardant 7 isomer of Fyrol PCF 12.15 flame retardant 8 alachlor a 12.63 herbicide 9 chlorpyriphos a 13.02 insecticide 10 bromophosmethyl a 13.20 insecticide 11 5-chloro-2-(2,4-dichlorophenoxy)phenol (Triclosan) 13.63 antiseptic, disinfectant a. spiked at 1 g/l From among the above examples, peaks Nos. 8 and 11 demonstrate most convincingly the distinct need for a close correspondence of the retention times. Because no (small) peaks or shoulders show up for these compounds in the full-scan MS chromatogram, the use of separate GC–MS and GC–AED systems without retention time locking can easily result in an erroneous assignment of the peaks. The compounds that do produce peaks in the full-scan MS chromatogram have retention time differences of only 1.5 s with the Cl peaks; however, these are not the chlorinated compounds. It could be argued that such small time differences will be no problem during manual identification of chlorinated compounds because the chlorine isotope pattern should be recognisable in some of the spectra around the retention time indicated by the AED. However, coelution with other compounds can make recognition of the isotope pattern very difficult. Furthermore, subtraction of the spectrum of coeluting compounds, rather than ‘background’, often leads to loss of the isotope pattern, especially at low responses. This is underlined by the fact that no chlorine isotope pattern could be recognised for compound No. 4 of Fig. 3.1. As an example of the identification of compounds which contain no heteroatoms with a distinct isotope pattern, N channel traces were examined. Some of the N- containing compounds which were readily identified by MS library searching were indole, methyl anthranilate, nicotine and caffeine. A more complex example is shown in Fig. 3.4. The peak at a retention time of 11.2 min in the nitrogen trace, which was initially considered to be a single peak, was also observed in the full-scan MS (not
48 3. Analysis of wastewater with GC–AED/MS shown) and the C193 chromatograms. However, the first MS library searches resulted in hits with low match qualities. After close examination of some AED and extracted ion traces, it was found that the peak in the C193 GC–AED chromatogram comprised three closely eluting compounds.
C193
N174 m/z=127
P178 m/z=99
m/z=156
11.10 11.14 11.18 11.22 11.26 Time/min Fig. 3.4: Element traces and extracted ion chromatograms of three co-eluting compounds (peak No. 17 in Fig. 3.5). The top trace shows the AED carbon trace which is equivalent to the MS TIC (not shown). The N trace and m/z=127 indicate the position of tetraacetylethylenediamine (TAED); P178 and m/z=99 tributyl phosphate; and the compound with fragment m/z=156 was identified as dihydromethyljasmonate.
Response 22
20 15
2 6 16 8 17 21 1 7 23 4 10 19 9 11 14 3 13 18 5 12
14
17 910 13 16 23
910 13 16
7 8 9 10 11 12 13 14 15 16 17 Time/min Fig. 3.5: GC–full-scan MS chromatograms of influent (top), effluent (centre) and HPLC water (bottom) shown on the same scale. Concentration levels in the influent are between 10 and 90 (dodecanoic acid, peak No. 15) µg/l. Sample volume, 25 ml; 5 µl of extract injected (equivalent with 1.25 ml sample). For the names of the identified compounds, see Table 3.4.
Because of the small but noticeable differences in retention times, it was possible to carry out background subtractions. Subsequently, the compounds could be identified as tetraacetylethylenediamine, tributyl phosphate and dihydromethyljasmonate. As will be
49 3. Analysis of wastewater with GC–AED/MS
obvious from Fig. 3.4 in two of these instances, the close similarity of a compound- specific extracted ion trace and the N or P trace could be considered an extra confirmation. This example clearly shows the benefits of the capability to overlay AED and MS chromatograms so that peak shapes can be compared.
MS screening. As an alternative strategy, MS library searching was used to identify some twenty of the most abundant compounds. Since the levels of these are between 10– 90 µg/l in the influent, the 1 µg/l pesticide spikes are not included (Fig. 3.5 and Table 3.4). Identification was not straightforward in all instances and regularly required manual spectral subtraction. However, even then the elemental composition data provided by GC–AED did not help to shorten the MS library-search hit list. The main cause is that essentially all candidate compounds on the MS hit list contain only C, H and O—which is not unexpected in view of the close similarity of the full-scan MS, C, H and O traces of Fig. 3.1. In other words, for the identification of the more abundant compounds, AED detection is not really helpful because its strongest point— discriminating analytes on the basis of a heteroatom response—cannot be put into effect. Calculation of elemental formulae, another valuable aspect in the case of heteroatoms, is not a real option now either because of (i) serious co-elution of compounds which all contain, at least, C and H, and (ii) the unsatisfactory nature of H (non-linearity) and O (compound-dependent response) calibration plots [158;159]. In summary, comparison of the AED-first and MS-first screening strategies unambiguously shows the superiority of the former approach. It easily reveals the presence of sample constituents (cf. Fig. 3.3 and Fig. 3.4) which are completely obscured in conventional, abundance-orientated approaches. 3.3.3 Identified compounds and water treatment efficiency The identity of peaks Nos. 1–23 of the GC–MS traces of Fig. 3.5 reveals that most of the abundant compounds present in the influent water are plasticisers, fragrances and fatty acids (Table 3.4). Four of these, peaks No. 9, 10, 13 and 16, were present in essentially all samples analysed and are background interferences from the sample preparation module or the analytical system. For example, the isomers of 2,2,4-trimethyl-1,3- pentanediol monoisobutyrate probably originate from the SPE cartridge or the membrane filter. More importantly, inspection of the influent and effluent chromatograms generally showed a high efficiency of the treatment procedure (Fig. 3.5). Calculations for the total set of GC-amenable compounds showed that 85–90% of these were removed during the treatment. It is interesting, however, to indicate that the use of AED next to full-scan MS detection again helps to reveal additional, and somewhat unexpected information. When
50 3. Analysis of wastewater with GC–AED/MS the chlorine traces of the influents and effluents are compared (Fig. 3.6), it appears that the treatment is not sufficient to remove all of the chlorinated compounds. The concentration of the chlorinated compounds are between 0.5 and 2 µg/l in the influent, and between 0.25 and 1 µg/l in the effluent. The chlorinated phenols and the disinfectant 5-chloro-2-(2,4-dichlorophenoxy)phenol (peaks Nos. 1–3 and 11) are effectively removed during treatment. However, the concentrations of the flame retardants Fyrol CEF and PCF (peaks Nos. 5–7) remain virtually the same. Furthermore, a peak is visible which elutes slightly later than the unidentified compound No. 4.
Response
6 2,3
1 11 4 5 7
6
5 7
9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 14 Time/min Fig. 3.6: Chlorine traces of non-spiked influent (top) and effluent (bottom) sample. Sample volume, 25 ml; 5 µl of extract injected (equivalent with 1.25 ml sample). For the names of identified compounds see Table 3.3.
Whether the inadequate removal depends on a poor biodegradability or a high solubility of the compounds concerned in water is beyond the scope of this paper. It is noteworthy however that, since the response to chlorine is virtually independent of molecular structure, the AED provides direct information on the total organic chlorine level of aqueous (and other) samples—the prerequisite being an adequate SPE recovery and GC- amenability. In the present instance, with the average recovery taken equal to that of chlorpyrifos, total levels of 6.7 and 2.6 µg/l Cl were calculated for the influent and effluent, respectively.
51 3. Analysis of wastewater with GC–AED/MS
Table 3.4: Identity of the most abundant compounds in the MS full-scan chromatograms of Fig. 3.5 Peak No. Compound Ret. Time Use or origin (min) 1 1,8-cineole 6.76 fragrance 2 dihydromyrcenol 7.17 fragrance 3 tetrahydrolinalol 7.40 fragrance 4 linalol 7.43 fragrance 5 camphor 7.84 plasticiser 6 menthol 8.08 fragrance 7 2-(2-butoxyethoxy)ethanol 8.19 solvent 8 -terpineol 8.23 fragrance 9, 10 2,2,4-trimethyl-1,3-pentanediol monoisobutyrate (2 isomers) 9.40, 9.53 ------11 decanoic acid 9.62 fats and oils 12 dimethyl phthalate 10.07 plasticiser 13 2,6-(tert-butyl)-4-hydroxy-4-methyl-2,5-cyclohexadien-1-one 10.17 ------14 triisobutyl phosphate 10.42 plasticiser 15 dodecanoic acid 10.82 fats and oils 16 diethyl phthalate 10.90 plasticiser 17 tributyl phosphate 11.20 plasticiser 18 tridecanoic acid 11.30 fats and oils 19 tetradecanoic acid 11.84 fats and oils 20 diisobutyl phthalate 12.35 plasticiser 21 butyl isobutyl phthalate 12.58 plasticiser 22 dibutyl phthalate 12.82 plasticiser 23 tris(butoxyethyl) phosphate 14.76 plasticiser
3.4 Conclusions
This study shows that GC with simultaneous AED/MS detection is an excellent way to acquire reliable information on the nature of heteroatom-containing non-target compounds present at the trace level in complex samples. The main prerequisite is that the AED and MS retention times are the same to within 1 s across the whole chromatogram. The required strategy is to use GC–AED for screening and, next, GC–MS for identification/confirmation purposes. The examples provided in this study clearly show that compounds/peaks which would go unnoticed with MS detection only, can be traced in a routine fashion if the combined AED/MS approach is used. The, admittedly, limited data for influent/effluent wastewater indicate that the proposed GC–AED/MS procedure can be a useful tool to monitor the removal of GC– amenable pollutants in treatment plants. Monitoring of specific element channels can simultaneously provide semi-quantitative information on the decrease of, e.g., the total organic sulphur or chlorine contents.
52 4 4. Comprehensive two-dimensional gas chromatography with atomic emission detection (GC×GC–AED) and correlation with mass spectrometric detection: principles and application in petrochemical analysis
4.1 Introduction In the first years after its introduction in the 1990s, comprehensive two-dimensional gas chromatography (GC×GC) was mainly used in petrochemical analysis. More recently, the proven advantages of the technique have sparked interest in many other fields, such as environmental, food and air analysis and, at present, over one hundred papers have been published in this area (see, e.g., [160-162]). However, the range of detectors that has been used so far, is limited. One main reason is the small width of the second- dimension chromatographic peaks, which can be as narrow as 100–300 ms at their base. Proper registration—which includes accurate peak apex location, an important issue in the identification process—then requires detectors with small dead volumes and sufficiently high acquisition rates. The obvious early choice was the flame ionisation detector (FID) which operates at frequencies of, typically, 50–200 Hz and, moreover, is the detector in use for most petrochemical analyses. Recently, a miniaturised electron- capture detector, the Agilent µ-ECD, has been shown to be a successful selective detector for the GC×GC analysis of a wide variety of halogenated compounds [163-165]. More importantly, with the introduction of the fast acquiring time-of-flight mass spectrometer (ToF-MS), the possibility of structure-related detection, i.e., identification, has been created [160;166;167]. In previous work in our group, the use of atomic emission detection (AED) data complementary to mass spectrometric (MS) information for the identification of unknowns was studied in some detail [54;153]. Specifically, such data were found to be highly useful for, on the one hand, confirmation of high-quality mass spectral library hits (elemental composition) and, on the other hand, to act as a powerful aid for the reduction of the number of candidate compounds when the confidence levels of the library hits are
Published as: L.L.P. van Stee, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, J. Chromatogr. A 1019 (2003) 89 4. Petrochemical analysis with GC×GC–AED/MS
low (selective element traces). The individual element traces also yield rewarding results when a general screening for compounds containing a specific element, e.g., Cl or Br, is the main goal of the exploratory study: the subsequent MS analysis can then be targeted very precisely at the ‘hot spots’. Although the resolution of GC×GC is—as experience has meanwhile abundantly demonstrated—much higher than that of conventional one-dimensional GC, the same experience has taught that many separation problems still are not solved at all. In other words, the added potential of AED detection is most welcome. In this paper we explore, for the first time, whether combining GC×GC with, in principle, much too slow AED detection holds any promise for the future. Petrochemical analysis is the application area used for testing. 4.2 Experimental 4.2.1 Materials The pesticide standard was prepared by dissolving analytical-grade pesticides obtained from various sources in ethyl acetate (J.T. Baker, Deventer, the Netherlands). An n- alkylbis(trifluoromethyl)-phosphinesulphide mixture (so-called M-series; alkyl C6–C18; Nordion, Helsinki, Finland) was added to the pesticide mixture, with a final concentration of 50 ng/μl for all compounds. A crude oil and a fluidised catalyst cracking (FCC) product, sulphur content, 2.3 and 1.8 wt.%, respectively, were a gift from an oil refinery plant. 4.2.2 Instrumentation and methods An HP5980/II gas chromatograph equipped with an HP7673 autosampler and an HP5921A atomic emission detector was used for all AED experiments (Hewlett- Packard, Wilmington, DE, USA). The first- and second-dimension columns were a 15 m × 0.25 mm i.d. × 0.25 μm J&W DB-1 (Agilent, Amstelveen, the Netherlands) and a 0.6 m × 0.1 mm i.d. × 0.1 μm BPX-50 (SGE, Milton Keynes, UK) column, respectively. The second-dimension column was connected to the AED via a 0.7 m × 0.25 mm i.d. deactivated fused silica capillary (BGB Analytik, Anwil, Switzerland) kept at 280ºC. The cavity and cooling water were kept at 300ºC and 60ºC, respectively. The AED helium make-up flow was set at 20 ml/min. Standard wavelengths and reagent gas pressures were used for the three sets of elements: set 1, C 496 nm; H 486 nm; Cl 479 nm; Br 478 nm; set 2: C 193 nm; S 181 nm; N 174 nm; set 3: P 178 nm (also see Section 4.3). GC×GC–MS was performed with an HP6890 gas chromatograph and a Leco (Mönchengladbach, Germany) Pegasus II time-of-flight mass spectrometer equipped
54 4. Petrochemical analysis with GC×GC–AED/MS
with an electron ionisation (EI) ion source. The multi-channel plate voltage was 1900 V, the EI energy 70 eV, the acquisition rate 50 Hz and the mass range 50–450 amu. The first- and second-dimension columns were the same as used for the AED set-up; however, instead of a deactivated fused silica connection, the final 0.3 m of the 0.9 m second-dimension column was used as transfer line and kept at 280ºC. For analyte identification, in-house developed software was used which eliminated part of the manual operations earlier required. The MS software that has to be used to record the data, has not been designed to handle GC×GC data; the data are stored linearly in time. Until now, the conversion of the 2D into linear 1D time(s) was performed manually. Next, the experimental spectrum at that time had to be retrieved manually from the MS software. With the new software this procedure is automated: by clicking the peak of interest in the GC×GC chromatogram, the MS software is triggered to display the experimental mass spectrum and the library hit.
spectrometer A. liq. CO2 B. make-up He & window purge reagent gas 30 ml/min 50 ml/min valve valve oven wall
2nd dim. col. plasma ~ 1ml/min
press-fit
1st dim. col. 2nd dim. col. BPR ferrule purge vent GA 20 ml/min cavity vent 61 ml/min
Fig. 4.1: (A) Dual-stage liquid CO2 jet modulator. The analytes are immobilised on the head of the second- dimension column which is kept under slight tension by the spring action of the brackets. (B) Schematic showing the gas flows in the AED plasma. Out of the 50 ml/min of helium which is supplied, 20 ml/min is used as plasma make-up gas and 20 ml/min for column venting.
In both systems an in-house constructed dual-stage liquid CO2 jet modulator was used to modulate the first-dimension column eluent (Fig. 4.1A). Modulation takes place by immobilising the analytes by a stream of expanding liquid CO2, and subsequent mobilisation through heating by the hot air of the GC oven when the stream of CO2 is switched off. The jet on the left-hand side is used to prevent breakthrough of the analytes during the time that the right-hand jet is switched off. The electric valves are actuated by a programmable laboratory-built controller. In all experiments the jets were switched alternatingly every 4 s. Since re-injection on the second-dimension column takes place when the right-hand valve is closed, the
55 4. Petrochemical analysis with GC×GC–AED/MS
modulation period is 8 s. Analyte immobilisation occurs in the first part of the second- dimension column. About 5 cm of this column protrudes from the right-hand side of the modulator to allow for easy connection with the first-dimension column; another 5 cm is located within the modulator. Therefore, 0.5 m of the overall 0.6 m second-dimension column located in the oven, is effectively used for separation. For further details, one should consult [168]. Hot-split 1-µl injections were performed at a constant temperature of 260ºC using an Optic II programmable injector with a multi-capillary liner (ATAS, Veldhoven, the Netherlands). The GC columns were operated in the constant-pressure mode using helium (99.999%, Praxair, Oevel, Belgium) at 200 kPa as carrier gas. For the analyses of the pesticides standard a split ratio of 1:50 was used; the oven start temperature was 80ºC (no hold) with a programming rate of 5ºC/min to 280ºC. The petrochemical samples were injected using a split ratio of 1:200; the oven start temperature was 60ºC (no hold) with a subsequent programming rate of 3ºC/min to 280ºC. 4.3 Results and discussion 4.3.1 Relevant AED characteristics The commercial AED uses a 50 W microwave generator and a re-entrant cavity to focus the energy into a 1 mm i.d. fused silica tube in which a plasma is sustained by a steady flow of helium make-up gas. A spectrometer employing a diffraction grating and a movable photodiode array views the plasma axially and can detect the emitted radiation in the 160–800 nm region with a 0.1 nm resolution at 400 nm. However, out of this total range only 20-nm wide segments can be measured during one run (also see Section 4.3.4). For a detailed description, one should consult [155;169]. 4.3.2 Dead volume For a detector to be applicable in GC×GC it has to be capable to record very narrow peaks. The acquisition rate and the detector volume or, when using make-up gases, the effective dead volume, are the key parameters. A GC peak with a width of 300 ms at the baseline, will occupy a volume of approx. 5 μl at a flow of 1 ml/min. To avoid undue extra-column peak broadening, the detector volume should be several times smaller than this peak volume. With 12 μl the AED detector volume is much too large, were it not for the make-up gas flow that is necessary to maintain a stable plasma. In the HP5921A the plasma is completely enclosed (Fig. 4.1B), which prevents atmospheric back-diffusion into the plasma. This permits stable plasma operation at helium flows as low as 5 ml/min. However, in experimental practice a make-up flow of about 30 ml/min is used to prevent excessive peak tailing caused by reactions of the
56 4. Petrochemical analysis with GC×GC–AED/MS
ionised analyte products on the discharge tube wall and/or dead-volume effects [155]. Another factor that seriously reduces the effective dead volume is the column venting system (Fig. 4.1B) which consists of a continuous flow of helium at 20 ml/min that purges the area around the GC column outlet. To contrast the impression that the use of a higher make-up gas flow is purely beneficial, one should realise that (i) the analytes are diluted by the make-up gas and (ii) the higher linear speed reduces the residence time in the plasma. As a consequence, the analyte detectability is adversely affected and an optimum has to be selected depending on the type of application. For example, when there is no sample-size limitation the amount introduced onto the column can be increased (lower split ratio or larger injection volume) and the make-up flow increased to avoid peak broadening due to dead volume. However, this scenario only holds fully true when the acquisition rate is no limitation. 4.3.3 Acquisition rate The GC peaks should be wide enough to allow a sufficient number of sampling periods at the maximum acquisition rate of the AED, which is 10 spectra/s. Although the minimum number of data points across a peak is three for a triangular peak—not really considered good chromatographic practice, but acceptable for qualitative analysis—at least five data points are required for an adequate description and the elimination of noise and spiking problems. With this minimum requirement, the peak width at the base should cover at least four sampling periods. For an acquisition rate of 10 Hz, this means that the peaks should have a baseline width of at least 400 ms. In other words, for peaks that can be as narrow as 100–300 ms when using GC×GC under optimal conditions, some (intended) broadening may be required for proper registration using the AED. Actually, the transfer line that connects the GC and the AED may be of some 'help' here in that it will cause some peak broadening. To enhance the effect, a 0.25 mm instead of a 0.1 mm i.d. transfer line was used. Peak widths also depend on the second-dimension retention time. Although temperature programming is used and the first- and second-dimension columns are immersed in the same oven, the change in temperature during the very fast separations on the second-dimension column is so small that these separations can be regarded as isothermal. As a consequence, peaks eluting early in the second dimension are the narrowest ones. Experience showed that under the experimental conditions used in this study, such peaks typically had a baseline width of approximately 500 ms (with some broadening caused by minor tailing): that is, six data points could be used to describe the peak. In other words, the limited, but more or less deliberately introduced additional band broadening compared with GC×GC–FID (where values of about 150 ms have been
57 4. Petrochemical analysis with GC×GC–AED/MS
reported [168]) has created conditions which allow the AED to be used as a detector for GC×GC. 4.3.4 Data recording and presentation A phenomenon that complicates the use of AED detection in conventional 1D-GC as well as GC×GC is that several runs are required to unravel the elemental composition of the analyte of interest. This is caused by the fact that (i) a very high spectral resolution is required and (ii) the plasma has to be doped with reagent gases to increase selectivity and prevent peak tailing for some specific elements. Briefly, to enable the simultaneous recording of several wavelengths, a photo diode array (PDA) is used which, in the HP5921A, is a 12-mm long array with 211 diodes. The instrument has been designed to provide 0.1 nm resolution at 400 nm, which implies that a window of 211×0.1 or 20 nm is spanned by the PDA. Actually, depending on the position in the 160–800 nm focal plane, the window varies between 10 and 30 nm. Most importantly, simultaneous measurement of several elements is possible only when their emission lines are within the same window [156;169]. As regards doping of the plasma, a few examples may serve to illustrate the problems. Elements which form refractory oxides such as phosphorus or boron require hydrogen, while oxygen has to be used when carbon, hydrogen, chlorine and bromine are determined. In addition, even with hydrogen as a reagent gas, the phosphorus trace shows extensive tailing unless a very high make-up flow of 180, instead of the conventional 30, ml/min is used. For the pesticide analysis, three runs were performed to record the C, H, Cl, Br, S, N and P traces. Each run included at least one element which enabled the detection of the marker compounds, i.e. the M-series, viz. C for set 1, S for set 2 and P for set 3. Typical results are presented in Table 4.1. These are highly encouraging as regards the most critical parameter to be studied when combining data from several runs is an issue, viz. mutual differences in second-dimension retention times. The differences are seen to be small, i.e. 0.05–0.10 s in most cases, with a few exceptions of 0.10–0.15 s. Similar results were reported in a study on pesticides, where differences of 0.01–0.20 s were found for a large majority of all analytes [170]. One should add that repeatability data in another study [171] indicate that further improvement should be possible.
58 4. Petrochemical analysis with GC×GC–AED/MS
Table 4.1: Test analytes used in the study and their second-dimension retention times 2nd dimension retention time (s) Analyte Formula C Cl Br N S P
M1 C6H13PS(CF3)2 0.53 0.62 0.63
M2 C8H17PS(CF3)2 0.55 0.70 0.51
M3 C10H21PS(CF3)2 0.62 0.62 0.62
M4 C12H25PS(CF3)2 0.55 0.66 0.58
M5 C14H29PS(CF3)2 0.60 0.64 0.62
M6 C16H33PS(CF3)2 0.57 0.66 0.59
M7 C18H37PS(CF3)2 0.67 0.69 0.65
P1 Diazinon C12H21N2O3PS 1.96 2.05 1.99 1.94
P2 Fenchlorphos C8H8Cl3O3PS 2.25 2.29 2.29 2.21
P3 Fenitrothion C9H12NO5PS 2.96 2.98 2.99 2.92
P4 Malathion C10H19O6PS2 2.67 2.69 2.65
P5 Chlorpyrifos C9H11Cl3NO3PS 2.26 2.28 2.26 2.24 2.21
P6 Bromophos C8H8BrCl2O3PS 2.46 2.48 2.47 2.54 2.46
P7 Bromophos-ethyl C10H12BrCl2O3PS 2.14 2.20 2.20 2.16 2.05
P8 Azinphos-methyl C10H12N3O3PS2 4.94 5.07 5.10 4.98
P9 Pyrazophos C14H20N3O5PS 3.08 3.21 3.20
P10 Coumaphos C14H16ClO5PS 3.68 3.76 3.75 3.62
A colour-plot presentation for four elements is given in Fig. 4.2. Since the M-series compounds all exhibit closely similar second-dimension retention times because identical functional groups, but not the alkyl chains, determine the interaction, the marker spots show up at regular intervals in the lower part of the 2D plots and can be used for unambiguous alignment of various GC×GC–AED runs. Of course, a distinct disadvantage of the presentation of separate plots is that, with more complex samples, it becomes increasingly difficult to rapidly discern the combination of elements that is present in a specific analyte of interest. Fig. 4.3 illustrates how a much improved result can be obtained by means of a procedure which requires little data processing. For this figure, the colour plots of the various elements were overlaid while using a slight offset in the direction of the Y-axis. Obviously, when adequate enlargement is used and 3–4 element traces are combined per frame, a fully satisfactory result is obtained and the (qualitative) element composition of each analyte can immediately be read from the spot colours.
59 4. Petrochemical analysis with GC×GC–AED/MS
8 7 Carbon 6
5 4
3 2
1 0
8 7 Chlorine 6
5 4
3 2 1 0
8
7 Sulphur 2nd dim. retention time(s) retention dim. 2nd 6 5 4
3 2 1 Fig. 4.3: Combined presentation of elemental 0 composition data of pesticides present in rectangle indicated in Fig. 4.2. Slight shifts of 8 every 'element layer' in Y-axis direction prevent 7 Nitrogen loss of visual information. For peak 6 identification, see P1–P7 in Table 4.1. 5
4
3 2
1 0 400 600 800 1000 1200 1400 1600 1800 2000 1st dim. retention time (s) Fig. 4.2: GC×GC–AED of mixture of pesticides and marker compounds; the latter show up at 2nd dimension retention times of about 0.5 s. For red rectangle in Carbon frame, see Fig. 4.3. 4.4 Application: petrochemical analysis Desulphurisation of fuels receives much attention today because of the urgent need to reduce the adverse effects of fuel combustion on the environment. The development of novel catalysts requires detailed knowledge of the molecular structures of the sulphur- containing compounds present in crude oil and intermediate products. The detectors of first choice for such analyses are selective detectors such as the sulphur
60 4. Petrochemical analysis with GC×GC–AED/MS
chemiluminescence detector (SCD), the pulsed flame photometric detector (PFPD) and the AED. These detectors all show limits of detection (LOD) of 0.5–2 pg S/s [172]. With 1D-GC, such selective screening provides a rapid provisional classification of the main S-containing compounds present in a sample. However, as is vividly illustrated in, e.g., a study by Amorelli et al. [173] on the use of GC–AED, there is a serious overlap of the several analyte classes of interest (Fig. 6 of that study). Improved performance requires the use of extensive and time-consuming sample work-up and/or fractionation (see, e.g., [89;174;175]) or hyphenated analyses [176] which is not really attractive. In other words, another solution is of distinct interest.
9.80 Crude oil S
BNTs 8.40 Rx S 7.00 DBTs Rx 5.60
4.20 S BTs
2ndt dim.retentiontime (s) 2.80 Rx
1.40 Cx 11 14 18 22 26 30 n-alkane 0.00 100 700 1300 1900 2500 3100 3700 4300 4900
9.80 FCC product Fig. 4.5 Fig. 4.6 BNTs 8.40 DBTs 7.00
5.60 Fig. 4.8 ?
BTs 4.20
2ndt dim.retentiontime (s) 2.80
1.40 Cx 11 14 18 22 26 30 n-alkane 0.00 100 500 900 1300 1700 2100 2500 2900 3300 3700 4100 4500 1st dim. retention time (s) Fig. 4.4: GC×GC–AED of (top) a crude oil, and (bottom) an FCC product. The sulphur (orange) and carbon (blue) channels are indicated. For further explanation, see text.
61 4. Petrochemical analysis with GC×GC–AED/MS
9.10 Fig. 4.6 AED C193 7 A. MS TIC 8.20 4 6 5 7.30 12 area of 6.40 33 BNTs
5.50 area of DBTs
4.60 1600 2000 2400 2800 3200 3600 B. AED element: 7.70 MS TIC Fig. 4.6 7 C (Blue) 6 S (Red) 7.10 4 5 N (Green) 2ndt dim.retentiontime (s) 6.50 area of BNTs 12 5.90 time(s) retention dim. 2nd 3 3 5.30 area of DBTs
4.70 C. MS extr. ions 1900 2300 2700 3100 3500 3900 m/z: 1st dim. retention time (s) 202 (Blue) 208 (Red) Fig. 4.5: Retention-time correlation between (top) 195 (Green) AED C-channel and (bottom) TIC-MS chromatogram by stretching-plus-shifting and modulation-time manipulation. Marker compounds indicated: 1. DBT; 2. Phenanthrene; 3. Me-DBT; 4. Pyrene; 5. Me-pyrene; 6. BNT; 7. 1st dim. retention time Chrysene Fig. 4.6: Correlation and identification of compounds in selected part (see rectangle indicated in Fig. 4.5) of the FCC product. The AED signals (B) were used to locate the compounds in the TIC MS chromatogram (A). Extracted ion traces (C) were then constituted, viz. for dimethylcarbazoles (m/z 195) and pyrene (m/z 202) and, tentatively phenanthro[4,5- bcd]thiophene (m/z 208). For details, see text.
To investigate the practicality of GC×GC–AED to separate the main classes of S- containing compounds—aliphatic and aromatic thiols, alkylated benzothiophenes (BTs), dibenzothiophenes (DBTs) and benzonaphthothiophenes (BNTs)—a crude oil and an FCC product were analysed. Next to lower-boiling compounds, a crude oil contains high-boiling alkanes and heavily alkylated ring structures which, depending on the source, may be dominant. FCC is used to convert long (>C12) into shorter ( 62 4. Petrochemical analysis with GC×GC–AED/MS continuous bands with their intensity increasing at higher retention times. In the FCC product, however, the high-boiling sulphur compounds are absent and moderately alkylated (C1–C6) aromatic sulphur compounds dominate. The widely different carbon profiles illustrate the effect of the FCC conversion to short carbon chains. One group of sulphur-containing compounds of interest are those situated between the DBTs and BNTs, which are indicated by a question mark because they do not belong to any of the afore mentioned groups. These compounds were also detected, but not identified, in a GC×GC–SCD study by Wang et al. [177]. Further study of a class of unknowns as mentioned in the previous paragraph requires the combined, and correlated, use of AED and MS detection. The main difficulty of correlating the two sets of data is the difference in column outlet pressure, with the MS being held at vacuum and the AED at near-atmospheric pressure. Since the inlet pressure was the same in both systems, the pressure at the point of connection of the two columns is different; hence, the first- and second-dimension retention times are affected. Consequently, straightforward correlation is not possible. A solution was found by shifting and stretching one of the chromatograms along the first-dimension axis, and by also changing the modulation time used for converting the one-dimensional chromatogram into a two-dimensional plot. Although the modulation time actually used during the GC×GC analysis is constant, the modulation time used for graphical conversion can be varied. This causes a change of angle relative to the X-axis. After such manipulation had been successfully performed for the carbon channel of the AED and the MS total ion chromatogram (TIC)—as is demonstrated in Fig. 4.5, where several marker compounds are indicated—the same parameters were used to convert other AED channels and extracted ion chromatograms. Since nitrogen can be monitored under the same conditions (i.e. in the same run) as sulphur, this channel was also recorded. Although the number of nitrogen-containing compounds that was detected, was only about ten, they form an interesting group for a first test of the AED vs. MS correlation. This is because compounds containing an uneven number of nitrogen atoms have an uneven molecular mass and therefore show a rather 'unique' mass spectrum amongst most other compounds which possess even molecular masses. Some of these nitrogen compounds elute in the same region as do the unknown sulphur compounds mentioned before. The rather blurred TIC MS chromatogram displayed in Fig. 4.6A clearly reveals that a search for unknowns indeed requires the targeted information on individual elements shown in Fig. 4.6B. In-house developed software, in which several steps of the earlier, manual [170], procedure had been automated (cf. Experimental section) was used for the subsequent identification. 63 4. Petrochemical analysis with GC×GC–AED/MS All N-containing compounds in the area shown in Fig. 4.6B turned out to be dimethylcarbazoles. One example is shown in Fig. 4.7A; similarly high match factors as indicated here (914) were found for the other compounds. As a further confirmation of identity, manual checking showed a good mutual match of the peak shapes of the relevant individual ion traces. A. B. x H N NH2 NH2 phenanthro[4,5-bcd]thiophene 1,8-Anthracenediamine 2,3-Dimethylcarbazole S Fig. 4.7: (A) Mass spectrum of one of the nitrogen peaks of Fig. 4.6, and corresponding library spectrum. (B) Mass spectrum of S-containing peak in Fig. 4.6. Initial identification had poor fit. For explanation of subsequently proposed structure, see text. The S-containing compound of interest in the 2D plane of Fig. 4.6B is the red-coloured spot in the top right-hand corner of the AED chromatogram; its base peak was at m/z 208. To emphasise the need for proper alignment, it is interesting to note that, prior to this procedure, the peak was erroneously identified as 1,8-anthracenediamine (m/z 208, but no sulphur and incorrect fragment ions; similarity 610). After alignment, additional information was derived from the elution of pyrene (C16H10; m/z 202; blue spot in Fig. 4.6C) in the close 2D-vicinity of the S-containing unknown, which suggests similar aromaticity. On the basis of the combined information now available, the number of molecular formulae could be drastically limited, with unsaturated C14H8S compounds as the most likely candidates. To all probability the analyte of interest is phenanthro[4,5- bcd]thiophene (Fig. 4.7B). The m/z 163 ion indicates the loss of CHS+ (45 amu) which confirms the presence of sulphur, while the doubly charged ion (at m/z 104) is typical for aromatic compounds. 64 4. Petrochemical analysis with GC×GC–AED/MS Since phenanthro[4,5-bcd]thiophene is not commonly reported in oil fractions, a literature study was performed to substantiate the identification. The non-alkylated and methylated analogues were found in coal liquids and shale oils [89;178] and in the workplace air of an aluminium melting plant [78], the up to C3-alkylated compounds in coal [179], and the up to C7-alkylated congeners in vacuum gas oil [175]. Finally, extracted ion GC×GC chromatograms of masses associated with the alkylated forms of phenanthro[4,5-bcd]thiophene such as m/z 222, 236 and 250 revealed that these compounds are also present in the samples analysed in this study. This is demonstrated in Fig. 4.8. 7.2 MS extr. ions 7.0 6.8 6.6 6.4 6.2 2nd dim. retention time(s) retention dim. 2nd 6.0 2710 2870 3030 3190 3350 3510 1st dim. retention time (s) Fig. 4.8: GC×GC–MS extracted ion chromatogram of FCC product. Traces shown are m/z 208 (red), with the spot in the upper left-hand corner being the tentatively identified phenanthro[4,5-bcd]thiophene, and the alkylated congeners m/z 222 (green), 236 (blue) and 250 (yellow). The corresponding area in the GC×GC– AED chromatogram is indicated in Fig. 4.4. 4.5 Conclusions GC×GC can be on-line combined with AED detection to provide valuable element- selective information. Minor adaptations of the transfer line dimensions and the gas flow rates, and the use of the highest acquisition rate available (10 Hz) suffice to record at least five or six data points even for peaks rapidly eluting from the second-dimension column. The use of 'element overlays' allows an easy screening for compounds featuring specific combinations of elements. Combination of GC×GC–AED and GC×GC–ToF- MS information is highly useful when screening for selected types—e.g. S-containing— of unknowns. Ordered structures, which typically indicate the presence of sets of related compounds, considerably help to solve identification problems. More sophisticated studies, with higher AED/MS correlation, will be possible by using a precisely 'tuned' restriction capillary in front of the MS inlet or by including both detectors in one GC×GC set-up. 65 5 5. Evaluation of the combined use of biomimetic and solid-phase extraction techniques for the screening of organic micropollutants in wastewater 5.1 Introduction Today’s industrial society faces the presence of many xenobiotic chemicals in the environment, emitted directly or indirectly via industrial, agricultural or consumer activities. Some of these compounds are highly persistent due to their lipophilic character and resistance to biotransformation. Others are more polar and are easily degraded in the environment and degradation products can be produced which sometimes are more persistent than the parent compounds [180]. A wide variety and large number of organic micropollutants are present in the environment; however, most studies focus on a rather limited set of compounds. Developing a wider-ranging screening strategy for organic micropollutants is, therefore, an important topic. This is demonstrated by, e.g., a study of Hendriks et al. [181], where only a small part (10– 20%) of the toxicity of freshwater could be explained by the compounds identified using a conventional target approach. A considerable number of unknown compounds present in the water is probably responsible the bulk of the toxicity. Today, most screening studies focus on the determination of medium-polar compounds in the environment. Less attention, has so far, been paid to the development of screening methods for non-polar compounds with a log octanol–water partitioning coefficient (log Kow or log P)>3–4. The analytical problem with non-polar compounds— especially compounds which have log P>5—are the low concentrations present freely dissolved in water. The consequence is that large volumes of sample must be handled in order to obtain final concentrations within the instrumental detection range. Screening studies are usually carried out by applying solid-phase extraction (SPE), using cartridges packed with relatively hydrophobic sorbents. The profiles typically obtained with such an ‘exhaustive’ extraction procedure show that a wide range analytes can be covered but that results in the high log P range are unsatisfactory. Biomimetic 5. Biomimetic extraction of wastewater extraction was developed to mimic the bioconcentration process of biota by means of partitioning of the target compounds between the aquatic phase and a surrogate material which mimics the fatty tissue of an organism. The resulting profile mimics the profile in an organism [182]. The striking difference between the two types of profile is schematically presented in Fig. 5.1. An excellent review of so-called passive samplers was published by Stuer-Lauridsen a few years ago [183]. It provides much information on the history of these samplers, the different designs and their application. From among the various types of sampler, the semi-permeable membrane device (SPMD) developed by Huckins et al. [184;185] is most frequently used: there are over a hundred published studies. The SPMD is constructed from a lay-flat polyethylene tube, which acts as a membrane, and contains the triglyceride triolein which is spread out as a thin film within the tube, thus creating a high surface-to-volume ratio. The SPMD is commercially available, and standardised methods for the entire procedure of use have been published [186]. The advantages of biomimetic extraction compared to conventional trace- enrichment procedures are (i) the ability to handle large volumes of water, (ii) the relative simplicity of the method, (iii) the enrichment of only freely dissolved and bioavailable organic micropollutants from the aqueous phase, and (iv), time integrative sampling. Compared to the alternative technique of biomonitoring organisms, it is a distinct advantage that biomimetic extraction is not affected by biotransformation processes. It is, therefore, a suitable technique for the extraction of potentially bioaccumalable compounds from the environment. Another type of biomimetic sampler is based on Empore disks [182;187]. Usually, these disks are applied in conventional, exhaustive, SPE procedures. They consist of sorbent particles (e.g. with C18 functional groups) that are embedded in a thin disk of PTFE fibres. Due to the flat geometry the disk has a high surface-to-volume ratio which favors its use as a passive equilibrium sampling device. Since the disks are commercially available and proper disk holders can be constructed fairly easily, the Empore approach is a promising technique. In the present study, the two biomimetic techniques briefly introduced above were compared. This was done primarily in order to test their main characteristics (see Table 5.1) and, as importantly, the relevance of the information provided. 68 5. Biomimetic extraction of wastewater conc. SPE extract Table 5.1: Characteristics of the two biomimetic techniques* Parameter SPMD Empore device LogP Analytical procedure Longer (due to GPC) Short Kinetic equilibrium Slow Faster conc. Biomimetic extract Analyte mass per device High Low Preparation of device Laborious Simple * As usually reported in the literature LogP Fig. 5.1: Typical concentration vs. log P profiles of organic micro-pollutants (in water) obtained by means of SPE and biomimetic procedures. In the second place, we wanted to assess whether the combined use of a biomimetic technique and a conventional SPE-based procedure will provide information on organic micropollutants covering a very wide range of log P values. In order to provide practically useful information, municipal wastewater was selected as sample type. For instrumental analysis, GC with simultaneous mass spectrometric and atomic emission detection (GC–AED/MS) was used. The main advantage of this hyphenated system is that mass spectra and elemental composition data are acquired at the same time. Atomic emission data also enable a search for compounds with specific heteroatoms in complex matrices [54]. One further aspect was addressed. A common disadvantage of biomimetic studies such as discussed here, is the technical problem of proper deployment of the devices used in the water to be monitored: proper position, right anchoring, prevention of outside interference, etc. Therefore, parallel to the field tests another experiment was run. Large water samples (ca. 10 l) were taken at the site of deployment, and an Empore-disk-type extraction was, next, performed in the laboratory. Under these conditions, the mass of C18 bonded sorbent (ca. 7 mg [187]) is very small compared to the water sample (10 l). One can therefore expect a biomimetic extraction profile, because the reduction of the analyte concentration will be only limited [182]. For this experiment, the Empore approach was preferred over the use of an SPMD. In the latter case, the mass of the lipophilic phase is, relatively speaking, much larger (ca. 0.3 g) and there is a real risk of exhaustion, i.e. of creating an exhaustive rather than a biomimetic profile. The two aspects of interest with the in-laboratory technique—with which the time- integrative nature is, admittedly, lost—were the comparison of the shape of the various 69 5. Biomimetic extraction of wastewater biomimetic profiles and the relative ease of operation, e.g. with regard to biofouling and, also, analyte detectability. 5.2 Materials and methods 5.2.1 Sample site Wastewater was collected at a municipal wastewater treatment plant (WWTP) located in the surroundings of Amsterdam, the Netherlands. This plant receives domestic wastewater and rainwater. The WWTP is equipped with a primary clarification system, an aeration tank and a final clarification system (post-sedimentation step). The final effluent is discharged into a river. 5.2.2 Solid-phase extraction Wastewater samples (1 l) were taken after the primary clarification system, and will be referred to as influent wastewater. Sampling prior to this step was impossible from a technical point of view, and was not very useful because the water was not homogenous at the inflow point. Water samples were also collected after the post-sedimentation, and will be referred to as effluent water. Samples were collected at the start and end of the biomimetic sampling period. The water samples were disinfected by adding 1 ml of silver nitrate solution (>99%, Sigma, Zwijndrecht, the Netherlands; 1 mg/ml water) to 1 l of water, and filtered through a 2 µm pore size glass-fiber GD1M filter (Whatman, Maidstone, UK) on top of a 0.45 µm pore size HA cellulose ester filter (Milipore, Etten-Leur, the Netherlands). Organic micropollutants were extracted from 25 ml filtered water using SPE on a PLRP-S cartridge (Polymer Laboratories, Church Stretton, UK) using a Prospekt sampling system (Spark, Emmen, the Netherlands). PLRP-S was chosen because this material has higher extraction efficiencies for medium-polar to polar compounds than C18 material. For details about the set-up of the Prospekt system, see van Stee et al. [54]. The micropollutants were eluted from the PLRP-S cartridge with 100 µl of methyl acetate. This extract was analysed using GC combined with simultaneous AED and MS detection (see Section 5.2.5). Distilled water was used as a blank and processed in the same way as the influent and effluent samples. 5.2.3 Empore disk sampling Empore devices were prepared as described by Van Loon et al. [187]. Briefly, Empore disks (diameter 47 mm, J.T. Baker, Deventer, the Netherlands) were precleaned with methanol and hexane. After cleaning, 13 mm diameter disks were cut, placed in a home- 70 5. Biomimetic extraction of wastewater made disk holder and stored for no longer than 24 h in airtight metal cans. Next, influent and effluent wastewater were sampled, 2 × 10 l in glass bottles, and disinfected with 5 ml silver nitrate (1 mg/ml) to prevent degradation of compounds by microorganisms. An Empore device was placed in each 10-l bottle for one week, and for a second week in a new batch (10 l) of water, this to prevent depletion of the compounds in the aqueous phase of more than 25% [182]. The water was continuously stirred by means of a magnetic stirrer. All experiments were performed in the dark at room temperature. The Empore disks were collected at the end of the experiment, cleaned with a tissue and extracted for 24 h in a glass reaction tube with 1 ml of cyclohexane (J.T. Baker) that was stirred several times per day. Next, the cyclohexane was collected, and the extraction was repeated. The extracts were combined and evaporated to approx. 200 µl under a stream of nitrogen. After the addition of an internal standard (2,4,5-trichlorotoluene; Ultra-Scientific, UK) the final extract was weighed. The samples were analysed by GC– AED/MS. A blank disk was extracted in the same way as the samples. Empore devices were also placed directly in the influent and effluent wastewater streams, and kept there for either for 14 or 48 days. After such field sampling, the disks were cleaned and extracted in the same way as described above. 5.2.4 SPMD field sampling Thin-walled low-density layflat polyethylene tubing (Brentwood plastics, Brentwood, MO, USA) was used for the preparation of the SPMDs according to Booij et al. [188]. The tubing was precleaned, to remove analytical interferences, with hexane–ethyl acetate (1:1, v/v, both ultra-resi grade, J.T. Baker) for 4 days using fresh solvents each day. The SPMDs were 33 cm long, 2.5 cm wide and had a wall thickness of 70 µm. One end of the polyethylene tubing was heat-sealed and 0.29 g of 99% triolein (Sigma, St. Louis, MO, USA) was added to the bag, the remaining air was removed and the other end of the bag was also sealed. SPMDs were stored in airtight metal cans for a maximum period of one day. Next, the SPMDs were put in stainless-steel racks and placed in the influent and effluent streams of the WWTP for either 14 or 48 days, simultaneously with the Empore devices. The water temperature during the field exposure was approx. 22°C. After sampling, the SPMDs were cleaned with a tissue and with distilled water. The extraction of the SPMDs was performed by dialysis against 50 ml hexane–ethyl acetate (1:1, v/v). After 24 h a fresh batch of solvent was used, and the process repeated. The combined extracts were concentrated to approx. 5 ml using a Kuderna-Danish device, and quantitatively transferred and concentrated to ca. 0.1 ml in a tapered sample vial by repetitive transfer and evaporation under a stream of nitrogen, without allowing complete dryness. The whole extract was then separated into two fractions with gel permeation chromatography (Phenogel 5, Phenomenex, 300 mm length, 7.8 mm i.d.). 71 5. Biomimetic extraction of wastewater Hexane–ethyl acetate (60:40, v/v) at a flowrate of 1.5 ml/min was used as eluent. The triolein-containing fraction (6.3–9.0 min) was discarded, and the 9.0–33 min fraction was collected and further concentrated to ca. 0.3 ml as described above. Internal standard (2,4,5-trichlorotoluene) was added, and the final extract was analysed by GC– AED/MS. Triolein-containing SPMDs, stored for only one day in the metal can, were used as blanks, and were processed in the same way as the samples. 5.2.5 GC–AED/MS Sample injection was performed by a programmed temperature vaporisation (PTV) injector (Optic 2; ATAS, Cambridge, UK); 5 µl of the sample extract were injected. The PTV was used in the splitless mode, with an initial temperature of 66°C and a temperature ramp of 6°C/s to 265°C. After 2 min the split vent was opened. The column was an HP-5 (30 m × 0.25 mm i.d. × 0.25 µm film; Agilent Technologies, Amstelveen, the Netherlands) with helium as carrier gas at a constant flow of 2 ml/min. The eluent was split with a glass press-fit Y-splitter to the MS and the AED (for details, see [54]). The temperature programming of the GC oven was 77°C at injection held for 4 min, then to 140°C at 10°C/min which is held for 5 min, then to 250°C at 5°C/min with a hold for 5 min, and finally to 310°C at 10°C/min which is held for 10 min. An HP MSD 5972 quadrupole mass spectrometer was used in the full-scan (m/z 50– 450) electron ionisation (EI) mode. The ion source and transfer line temperatures were 188°C and 300°C, respectively. Eight elements (C, H, Cl, Br, P, N, O and F) were analysed using the AED 5921A (for details about AED settings, see [54]). 5.2.6 Identification and quantification For quantification authentic standards were used if available. Several standards, containing in total some 300 compounds, were kindly provided by the Institute for Inland Water Management and Waste Water Treatment (RIZA). The compounds in the standards have been identified on one or more occasions in Dutch surface water and mainly are compounds used in industry and/or agriculture. Additionally, a set of 15 standard solutions containing in total some 140 semi-volatile compounds, used for EPA Method 8270, was purchased from J.T. Baker. The standards were diluted if necessary (10 ng/µl level) and measured prior to the sample sequence. The standards were used to build the target tables in the Chemstation software using their retention times and two or more target ions. Several compounds were present in more than one standard mixture; some 300 compounds were added as targets to the quantification tables. The quantification of selected compounds was used to compare the extraction methods; in these cases the correct integration of the peaks 72 5. Biomimetic extraction of wastewater was manually checked. For qualitative purposes, the mass spectra and retention times of the standards were used for confirmation of identity. Unknown compounds were identified using library searching in the NBS54K (NIST, Gaithersburg, MD, USA), Wiley (6th ed., Pallisade, NewField, NY, USA) and HPPest- library (HP, Palo Alto, CA, USA). The library search was based on probability-based matching [189] as implemented in the Chemstation software. AED-elemental profiles were used as complementary information to select or exclude candidate compounds from the library hit list. Furthermore, AED data, more particular the Cl and Br traces, were used to locate halogenated compounds in the MS data, followed by tentative identification using mass spectral library searching. Details and examples of this strategy are described in Chapter 3. Log P values were retrieved from experimental data from several sources or calculated using the programme ClogP (BioByte, Claremont, CA, USA). 5.3 Results and discussion 5.3.1 General characteristics Generally speaking, the Empore-based approach was a simple and straightforward technique, and no additional clean-up or fractionation was required, not even when analysing influent water. Cleaning of the disks presented no difficulties and there were almost no analytical interferences. Although absolute amounts trapped on the Empore disk are smaller compared to the SPMD, analyte detectability did not create problems in the present study. Should there be a need to improve the detection limits, then one could also use an entire disk of 47 mm diameter that we used to cut the 13 mm disks from. With the SPMDs, the experimental results were also fully satisfactory. However, the various disadvantages recorded in Table 5.1 all clearly showed up. The precleaning and filling-with-triolein and, moreover, the GPC clean-up—which is necessary to avoid overloading the GC column with triolein which is always present in some amount in the final extract—make the technique quite time-consuming. In addition, more analytical skills are required. The GPC step caused some loss of more polar compounds such as, e.g., several organophosphate esters. However, the recoveries of the relevant non-polar compounds such as phthalates, ethers, organochlorine pesticides, polycyclic aromatic compounds, polychlorinated biphenyls invariably were higher than 85%. With the conventional SPE procedure which was carried out using a Prospekt system, there were no technical or analytical problems at all. 73 5. Biomimetic extraction of wastewater 5.3.2 Deployment time Usually, biomimetic devices are deployed in the field for several weeks up to months. In this study we compared the effect of sampling for 2 or 7 weeks on the final concentrations of the microcontaminants in the extracts. These experiments were conducted in consecutive order, i.e. the first two weeks were not integrated in the 7-week samples. In other words, concentrations in the 7-week sample can be lower than in the 2- week sample. Table 5.2: Concentrations of selected compounds (mg/g fat), and their ratio, after 2 and 7 weeks deployment of the biomimetic devices in the field effluent Empore disk SPMD deployment (weeks) deployment (weeks) 2 7 Ratio 2 7 Ratio Analyte* conc. conc. conc. conc. Chlorpyrifos 0.2 0.2 0.9 0.3 0.4 0.8 ADBI 0.4 0.4 1.2 0.3 0.1 2.2 Pyrene 0.1 0.1 1.1 0.8 0.8 1.0 AHMI 1.0 1.0 1.0 0.8 0.4 1.9 ATII 0.7 0.8 0.9 1.4 1.6 0.9 Acetylcedrene 2.4 1.2 2.0 2.5 1.1 2.3 AHTN 7.8 8.7 0.9 9.4 10.3 0.9 HHCB 11.9 11.3 1.1 19.1 15.8 1.2 *ADBI: 4-Acetyl-6-t-butyl-1,1-dimethylindane; AHMI: 5-Acetyl-1,1,2,3,3,6-hexamethylindane; HHCB: 1,3,4,6,7,8-Hexahydro-4,6,6,7,8,8-hexamethylcyclopenta-gamma-2-benzopyran; AHTN: 7-Acetyl-1,1,3,4,4,6- hexamethyl tetrahydronaphthalene; ATII: 5-Acetyl-3-isopropyl-1,1,2,6-tetramethylindane. This would simply indicate that analyte concentrations(s) were lower in that period of time. The results for a selected group of compounds are shown in Table 5.2. For six of the eight compounds, the concentrations are very similar after 2- and 7-week deployment (ratio 0.8–1.2). Although there will be exceptions—as for the analytes showing an about 2-fold difference—probably specifically for compounds present in very small concentrations, 2-week sampling seems adequate for most purposes. The role of biofouling will be discussed below. 5.3.3 Qualitative aspects and biofouling For a comparison of the different extraction methods only the 2-week samples were used. The total ion MS chromatograms are shown in Fig. 5.2, with the AED chlorine traces (for identification see Table 5.3) displayed in grey. When the influent water chromatograms are compared, large differences are observed. Firstly, the three biomimetic profiles of the influent water (Fig. 5.2c-e) show much larger mutual differences than do those of the effluent water (Fig. 5.2f-h). 74 5. Biomimetic extraction of wastewater Fig. 5.2: GC–MS total ion chromatograms of the SPE and biomimetic influent and effluent samples. The AED chlorine traces are shown in grey (for compound names see Table 5.3). Biofouling may well play a significant role her. Both the Empore disk and the SPMD were mounted in stainless-steel holders with holes of approx. 4 × 4 mm. Although the samplers were deployed after the primary settler, the amount of particulate matter and microbiota in the water was still very large. Even after 2 weeks, the samplers were 75 5. Biomimetic extraction of wastewater covered with a layer of slimy material which clearly obstructed the water flow across the Empore disks or SPMDs within the device. Since the Empore disk holder was the smallest device with the least holes, this effect was probably the strongest in this case. The SPMD sampling cage was rather large, therefore relatively many more holes remained unobstructed which allowed water to reach the SPMD. Besides biofouling of the sampler cages, biofouling of the Empore disk or SPMD itself can also influence the uptake of compounds. Whether using larger cages to hold the Empore disk can improve results will require further study. The strongest effect of the assumed biofouling can be seen when the laboratory biomimetic influent (Fig. 5.2c) extract is compared with the corresponding field sample (Fig. 5.2d). The high concentrations and large number of compounds found in the former case are not due to an artefact but indeed represent microcontaminants extracted from the sample. Especially the differences observed for the chlorine-containing compounds initially raised the suspicion of accidental contamination. However, the two Empore laboratory experiments were conducted in parallel, so if contamination played a role, the pertinent compounds should have been visible in the MS traces of both extracts. A typical example of the potential of parallel AED/MS detection is demonstrated by the detection of chlorine-containing compounds (Fig. 5.2 and Table 5.3). The AED data can be used to confirm chlorine-containing targets found in the MS data, but, above all, it is especially useful to discover additional compounds. The targets comprised some 300 compounds detected and identified over the course of many years of surface water monitoring. This specific collection of targets, and the absence of 13 of the 23 chlorine- containing compounds found, clearly shows the discovery potential of AED/MS. Although many unknowns were not readily identified by MS library searching, the combined information of AED/MS, retention index, and expert spectral interpretation can be used to extend the list of important target compounds. 76 5. Biomimetic extraction of wastewater Table 5.3: Chlorine containing compounds detected in various 2-week samples. Compounds in normal type were identified by MS target analysis and confirmed by AED; compounds in italic type were non-targets located by AED and tentatively identified by mass-spectral library searching Compound Ret. time Name (min) 1 5.24 1,4-Dichlorobenzene 2 8.03 2,4-Dichlorophenol 3 10.03 4-Chloro-3-methylphenol 4 11.59 4-Chloro-3,5-dimethylphenol 5 12.03 unknown 1 6 12.45 3,4-Dichloroaniline 7 13.13 unknown 2 (monochloro) 8 14.50 unknown 3 (dichloro) 9 16.95 unknown 4 (trichloro) 10 17.19 unknown 5 (chloro-bromo) 11 17.61 unknown 6 (trichloro) 12 18.06 unknown 7 (trichloro, 1,2-dimethoxy-3,4,5-trichlorobenzene ?) 13 19.79 Hexachlorobenzene 14 20.19 Pentachloromethoxybenzene 15 21.40 beta-Hexachlorocyclohexane 16 21.62 Trichloroethyl phosphate 17 22.31 Tris(2-chloroisopropyl) phosphate 18 22.72 Tris(2-chloroisopropyl) phosphate (isomer) 19 24.82 o-Benzyl-p-chlorophenol (chlorophene) 20 26.12 unknown 8 (dichloro) 21 26.28 Chlorpyrifos 22 28.88 Triclosan 23 29.18 Triclosan methyl 5.3.4 Comparison of SPE vs. biomimetic and field vs. laboratory biomimetic extraction A comparison of SPE and biomimetic extraction based on data for individual compounds was performed with the influent samples since the amount and variety of compounds is larger than in the effluent. First, the peaks representing the main part of the total area under the peaks were compared for SPMD and SPE. All peaks in the chromatograms were integrated using a non-target approach and sorted on area. Next, the total areas were calculated, and the highest peaks representing 50% of the total area were identified either by a match with one of the standards or by library searching and examination of the mass spectra. The results are shown in Tables 5.4 and 5.5. 77 5. Biomimetic extraction of wastewater Table 5.4: Compounds comprising 50% of the Table 5.5: Compounds comprising 50% of the total area under the peaks in the SPE influent total area under the peaks in the SPMD influent sample field sample Compound Conc. log P Compound* Conc. log P (µg/l) (mg/g fat) Caffeine 28.4 -0.1 Squalene 6.9 14.1 2-(2-Butoxyethoxy)ethanol 10.7 0.6 Octadecanoic acid, butyl ester 5.1 9.7 Menthol 5.8 3.3 Sulphur (S8) 4.4 - Uroterpenol (tentative) 4.6 - HHCB 4.0 6.3 Nicotine 4.4 1.2 Acetylcedrene 3.3 5.4 9-Octadecenamide 3.9 6.5 Hexadecanoic acid 2.6 7.2 -Terpineol 3.8 3.3 Hexyl salicylate 2.3 - 2-Phenoxyethanol 3.8 1.2 Hexenyl salicylate 2.1 4.6 Tetraacetylethylenediamine 2.8 -2.4 Benzyl salicylate 2.0 4.3 Diethyl phthalate 2.3 2.4 Butyl benzyl phthalate 1.5 4.7 Linalool 2.3 3.0 Triclosan 1.4 4.8 Dihydromyrcenol 2.1 3.5 AHTN 1.4 6.3 Butylated hydroxytoluene 2.1 5.1 Octanoic acid 1.2 3.1 Octadecanoic acid, butyl ester 2.0 9.7 Hexadecanoic acid, butyl ester 1.1 8.7 Acetophenone 1.8 1.6 Dibutylphthalate 1.0 4.5 Oxindol 1.7 1.2 Diisobutyl phthalate 0.9 4.1 Octylphenol triethoxylate 1.4 - Phenanthrene 0.8 4.5 Octylphenol tetraethoxylate 1.3 - Hexyl cinnamic aldehyde 0.5 1.7 Methyl dihydrojasmonate 1.3 3.0 Indole 0.4 2.1 Diisobutyl phthalate 0.5 4.1 * For abbreviations see Table 5.2. From the quoted tables, the difference in type of compounds sampled rapidly becomes clear. Only two compounds (shown in italics) were found with both methods. In general the log P values of the top 50% of compounds found in the SPE extract are in the range of 0–3. The top 50% of compounds in the SPMD sample generally show log P values between 3 and 8. Overall, in these two lists of top 50% compounds, the combined methods cover a log P range of 0–14. Here it is noteworthy that there are only a few exceptions; three compounds with high log P values (5.1–9.7) show up in the top 50% SPE list, and three compounds with log P values as low as 1.7–3.1 are included in the SPMD list. A detailed explanation of these exceptions cannot easily be provided, also because relatively high concentrations of individual compounds may play a role here. Another interesting comparison was that of laboratory and field biomimetic extraction profiles. For this comparison, a selection of twenty fragrances with a wide range of log P values (2.7–6.4, Fig. 5.3) was used. These compounds were selected since they are mostly used in common household products, and therefore the concentration ratios between the individual fragrances can be assumed to be fairly constant over time in a mixture of wastewater from several hundred thousands of households. From the selected group of compounds, the SPE method is seen to extract compounds with log P values up to 4.2. What is interesting to note is that the Empore 78 5. Biomimetic extraction of wastewater disk method extracts compounds with a much wider range of log P values, from 2.7 up to 6.4, with the profiles of the two methods being rather similar in the region up to log P ca. 4. However, the relative amounts of the high log P (>4.5) compounds are much higher in the laboratory Empore extract than in the field Empore extract. Since the laboratory Empore extraction uses a sample taken at a single point in time, and the field Empore extraction is cumulative in time this can cause profiles to be different. Then again, when constant concentration ratios between the compounds are assumed, this difference in profiles cannot be explained by the time integrative nature of the field deployment. Most probably, the observed differences are caused by biofouling in the field experiment. If the flow of water is reduced by solid particles present in the water shortly after the experiment is started, say in the first few days, the accumulation of compounds by the Empore disk is strongly hindered. The maximum equilibrium concentrations of compounds with log P <4 may then have already been reached. However, the high-log P compounds that are present at much lower concentrations in the water phase require more time to accumulate, and the fouling thus creates an adverse effect. Such obstruction does not occur in the laboratory experiment, and here the high-log P compounds are extracted efficiently. For the effluent sample, with which biofouling does not seem to play a role, no real differences between the laboratory and field experiments can be observed, as is clear from the profiles of Figs. 5.3c and 5.3d. The resemblance between the field and laboratory experiment, and with the expected biomimetic profile (Fig. 5.1), suggests that for both the field and laboratory conditions, the Empore disk extraction is a suitable method for the extraction of the latter class of compounds. 5.3.5 SPMD vs. Empore devices: absolute amounts Table 5.6 lists quantitative data for two main groups of compounds that were detected— fragrances and PAHs—in effluent samples. The amounts found are presented in two ways, as pg/g fat and as the total amount of analyte in the device. When the absolute amounts per device are considered, it is obvious that the SPMDs sequester larger amounts per device which can be beneficial when follow-up experiments like toxicity studies need to be performed. Although for laboratory experiments the use of larger biomimetic devices will disturb the biomimetic profile (becoming close to exhaustive SPE extraction), for field experiments one can scale-up the size of the Empore disks. In this study, the ratio in the amount of total fatty substance between the 0.3-m SPMDs and the 13-mm diameter Empore disk is about 42. Full-sized 47-mm Empore disks contain about 13 times more fat compared to the 13-mm disks, and reduce this ratio to about 3. 79 5. Biomimetic extraction of wastewater Such considerations, and the use of multiple devices, should be taken into consideration if follow-up experiments are planned. a. Influent Field Empore SPE Peak Peak area b. Influent Laboratory Empore SPE Peak Peak area c. Effluent Field Empore SPE Peak Peak area d. Effluent Laboratory Empore SPE Peak Peak area 2.7 3.0 3.0 3.3 4.2 4.6 4.8 5.1 5.9 6.3 log P Fig. 5.3: Differences in extraction profiles of twenty common fragrances with varying log P values sampled by SPE and 2-week field and laboratory biomimetic extraction. In the effluent, the concentrations of compounds with log P<4.2 are about 50–1000 times lower than in the influent. Therefore, the bars of compounds extracted by SPE barely show up. 80 5. Biomimetic extraction of wastewater Table 5.6: Concentrations and absolute amounts for SPMD and Empore, effluent samples Concentration Absolute amount per device (pg) Compound* Empore SPMD Empore SPMD (pg/g C18) (pg/g fat) Fragrances ABDI 6000 2000 40 600 AHMI 10000 6000 100 2000 Acetylcedrene 30000 20000 200 5000 AITI 10000 9000 70 3000 HHCB 200000 100000 1000 40000 AHTN 100000 60000 800 20000 PAHs Naphthalene 90 200 1 70 Fluorene 100 60 1 20 Phenanthrene 400 300 3 100 Fluoranthene 700 1000 5 400 Pyrene 2000 5000 10 2000 Benzo[a]anthracene 300 1000 2 300 Chrysene 80 200 1 60 * For abbreviations see Table 5.2. 5.3.6 Conclusions The combined use of SPE and biomimetic extraction allows the analysis of microcontaminants in complex water samples with a very wide polarity range. As expected, the biomimetic methods proved to be successful in the extraction of apolar compounds that are present in low concentrations. Two and seven week field deployment were compared, and results were similar which indicates two week sampling to be adequate for a general screening study. A serious disadvantage was the influence of biofouling, which blocked the flow of water when influent water was sampled in the field. However, one of the main findings was that a small Empore disk placed in a 10-l water sample and analysed in the laboratory yields a biomimetic extraction profile that was, at least for effluent water, proven to be very similar to the field extract. Although the time-integrative nature is lost in this way, apolar compounds can now be confidently determined. Biofouling now does not cause problems, not even with influent wastewater, and the method requires only one trip to the sampling site. 81 6 6. Toxicity identification and evaluation of inland waters: Use of semi-permeable membrane devices and solid-phase extraction for the wide-range screening of microconta- minants in surface water by GC–AED/MS 6.1 Introduction It is well-known that surface water in industrialised countries contains many tens, if not hundreds of men-made compounds at or above the low-ng/l level. In most instances, from among these microcontaminants, only a selected group is monitored on a regular basis. As regards surface water in the Netherlands, recent studies have shown that only part of the toxicity can be attributed to compounds determined by instrumental (e.g. gas or liquid chromatography) techniques [181]. In order to come to a better understanding, the Institute of Inland Water Management and Waste Water Treatment (RIZA) recently started an ‘unknown compounds’ project. The aim of the present study, which is part of that project, was to determine the identity and concentration range of a large set of organic micropollutants present in the major rivers of the Netherlands. Subsequently, the results of this study will be related to toxicity parameters of the same, or similar, samples determined by other scientists and used to select additional priority pollutants and devise improved methods for future surface water monitoring. In the present study, two extraction techniques were used which both effect the enrichment of microcontaminants which are present at the trace level, and are complementary in that they are aimed at compounds in different polarity ranges. Solid- phase extraction (SPE) was used to extract medium polar and polar organic compounds with, typically, log Kow<3. A technique using semi-permeable membrane devices (SPMDs) was utilised to isolate (bioaccumulative) apolar compounds with log Kow>3. In view of the detailed attention in the literature to SPE-based procedures for environmental analysis, both off-line and on-line (see e.g. refs. [190] and [191]), the SPE approach does not require discussion here. Published as: L.L.P. van Stee, P.E.G. Leonards, W.M.G.M. van Loon, A.J. Hendriks, J.L. Maas, J. Struijs, U.A.Th. Brinkman, Water Res. 36 (2002) 4455 6. Screening of surface water Compounds with a high log Kow and a low solubility in water are easily adsorbed to sediment and suspended matter. The concentration of the dissolved fraction is therefore difficult to determine by means of conventional techniques such as liquid–liquid extraction or SPE. In the literature several methods are described to measure the apolar fraction. One approach is to analyse controlled colonies of zebra mussels (Dreissena polymorpha) or wild biota, e.g. fish. Disadvantages of such methods are the extensive clean-up required to remove lipids and proteins, and the problems caused by biological variability. Of course, the apolar compounds can also be analysed directly in sediments or suspended matter. Although the analyte profiles found in suspended matter generally are a good representation of those found in biota, other adsorbed compounds (e.g. humic acids) easily interfere with the analysis and a proper clean-up is therefore required [192]. A third method to sample apolar compounds is to mimic bioaccumulation with an artificial device. The first extensively evaluated biomimetic extraction procedure was developed about a decade ago by Huckins et al. [184]. The original design of their SPMD consisted of a 90-cm long piece of layflat polyethylene tube containing 0.9 g of triolein forming a thin film inside the tube. More recently, another biomimetic method was developed based on the use of Empore extraction disks, which consist of C18- bonded silica particles embedded in a PTFE disk [182]. Because of its higher sampling capacity, the SPMD method is preferred in this study. The biomimetic methods have the advantage that the extracts are almost free from matrix interferences and the procedures are relatively simple. In addition, biomimetic extraction can be used when it is difficult to collect sediment or suspended matter, e.g. in open seas. All SPE and SPMD extracts were analysed using gas chromatography (GC) and screened for the presence of 430 target compounds, mainly agrochemicals and industrial and household chemicals. In addition, the use of simultaneous atomic emission and mass spectrometric detection (AED/MS) allowed the detection and identification of compounds not present in the target set. The GC–AED/MS technique is especially useful for the detection of heteroatom-containing compounds such as organohalogens [54]. Further, the complementarity of the two extraction methods and the feasibility of the SPMD technique for routine monitoring were investigated. The origin of several microcontaminants is briefly discussed in view of the future selection of additional priority pollutants. 6.2 Experimental 6.2.1 Sampling locations SPMDs were deployed and water samples were taken at five sampling locations (Fig. 6.1). At Lobith (LOB), the river Rhine enters the Netherlands, and at Eijsden (EIS), the 84 6. Screening of surface water river Meuse. The largest part of the water carried by the rivers Rhine and Meuse debouches into the North Sea close to Maassluis (MSL). The fourth sampling location was at Schaar van Ouden Doel (SOD), in the estuary of the river Scheldt. A fifth sampling location was Noordwijkerhout (NWK), which is located in an agricultural area and is not part of one of the major river systems. 6.2.2 Semi-permeable membrane devices Thin-film (70 µm) low-density layflat polyethylene tubing (Brentwood Plastics, Brentwood, MO, USA) was used to prepare the SPMDs according to Booij et al. [188]. The tubing was precleaned—to remove interferences—by 4 times 24-h extraction at room temperature with hexane–ethyl acetate (1:1, v/v). Both solvents were of Ultra Resi Grade and were obtained from J.T. Baker (Deventer, the Netherlands). After heat-sealing one end of the tubing, 0.29 g of triolein (99%, Sigma, St. Louis, MO, USA) was pipetted into the tubing and after removing the remaining air the other end of the tubing was also sealed. SPMDs were freshly prepared and stored in airtight metal cans, maximally 24 h before field exposure. Noordwijkerhout Maassluis Lobith Schaar van Ouden Doel Eijsden Fig. 6.1: Sampling locations. The SPMDs were exposed to water in cages of stainless-steel grating which allowed a free flow of water around the SPMDs. The deployment times were 21, 21, 28, 24 and 21 days for LOB, EIS, MSL, SOD and NWK, respectively. A blank was prepared by processing an SPMD which was stored for one day in a metal can. After sampling, the SPMDs were cleaned with a tissue and distilled water. The extraction of the accumulated compounds was performed by 2 × 24-h dialysis with 50 ml hexane–ethyl acetate (1:1, 85 6. Screening of surface water v/v). The combined extract (ca. 100 ml) was concentrated to approx. 5 ml using a Kuderna–Danish evaporator, and subsequently to 0.1 ml under a stream of nitrogen. To remove any triolein still present the extract was purified by gel permeation chromatography (GPC) on a 30 cm × 7.8 mm i.d. Phenogel 5 column (Phenomenex, Torrance, CA, USA). Hexane–ethyl acetate (6:4, v/v) at a flow rate of 1.5 ml/min was the GPC eluent. The 9.0–33.0 min fraction which contained all target compounds except triolein (elution 6.3–9.0 min) was collected, and subsequently concentrated to 0.3 ml under a stream of nitrogen and analysed using GC–AED/MS. 6.2.3 SPE extraction SPE extraction was performed according to the procedures of Roghair et al. [193] and Collombon et al. [194]. Briefly, 150 ml of a 1:1 (v/v) mixture of cleaned macroporous polystyrene–divinylbenzene copolymer—XAD-4, particle size 250–840 μm (Rohm & Haas, Antwerp, Belgium)—and macroporous polymethylmethacrylate—XAD-8, particle size 250–420 μm (Supelco, Zwijndrecht, the Netherlands)—were added to 80 l of a non- filtered water sample. After 24 h extraction under stirring the XAD was sieved and dried at room temperature under a gentle stream of air for 18 h. The sorbed compounds were eluted with 150 ml of acetone and the eluate was evaporated to a volume of approx. 1 ml by means of Kuderna–Danish distillation. The concentrate was subsequently added to 75 ml of water and the remaining acetone was evaporated by purging for 30 min with nitrogen. An aliquot of 7.5 ml of the aqueous concentrate was extracted with 3 × 2 ml of dichloromethane and the extract evaporated to a volume of 50 μl. After the addition of 0.5 ml hexane–ethyl acetate (1:1, v/v) and weighing, a 5-μl aliquot was analysed by GC– AED/MS. A blank was prepared by extracting 7.5 l of mineral water (Spa, Belgium). 6.2.4 Standards Seven standard solutions containing in total some 300 compounds were supplied by RIZA. These are all compounds which have been identified on one or more occasions in Dutch surface water and they mainly are compounds used in industry and/or agriculture. A set of 15 standard solutions containing in total some 140 semi-volatile compounds, used for EPA Method 8270, was purchased from J.T. Baker. Diluted standard solutions (2–10 µg/l) were analysed by GC–AED/MS using 5 µl injections. The results were used to set up the quantification table in the MS Chemstation software. 6.2.5 GC–AED/MS A Hewlett-Packard (Palo Alto, CA, USA) Model 5890 Series II gas chromatograph, a Hewlett-Packard 5971A mass selective detector and a Hewlett Packard 5921A atomic 86 6. Screening of surface water emission detector were used for GC–AED/MS. A Hewlett Packard 7673 autosampler was used for the introduction of 5 µl sample extracts. An Optic 2 (Ai Qualitek, Cambridge, UK) programmed temperature vaporiser (PTV) was connected to the analytical column (HP5-MS, 24 m × 0.25 mm id, 0.25 µm film). Deactivated glass multicapillary liners (1289712, ATAS, Eindhoven, the Netherlands) were used and replaced after 30–40 injections. A metal capillary outlet splitter (type VSOS; SGE, Ringwood, Australia), which was kept at 300ºC using a thermally isolated heater, was used to split the eluent (approx. 1:1) and lead it through two pieces of deactivated fused silica tubing to the AED (0.25 mm i.d., 65.5 cm in the transfer line + 30 cm in the oven) and MS (0.1 mm i.d., 28.3 cm in transfer line + 1.5 cm in the oven). Data acquisition and analysis was performed on a Pentium 150 MHz computer by running the G1701AA 3.00 and G2360AA 03.03 Hewlett Packard Chemstations simultaneously. Gas chromatography. Helium (99.999%; Praxair, Oevel, Belgium) was used as carrier gas. The Optic 2 was used in the splitless injection mode and 5 µl of sample were injected at a temperature of 77°C and the injection liner was heated to 265°C at 6°C/s; after 90 s the split valve was opened. The column head pressure was linearly changed from 166 kPa at the start, to 285 kPa at the end of the oven temperature programme. The split flow was 50 ml/min and the septum purge 10 ml/min at a pressure of 166 kPa. The oven temperature was programmed as follows for the SPMD samples. After a 4-min hold at 77°C the temperature was linearly increased at 10°C/min to 140°C and held there for 5 min. Next, the temperature was raised to 250°C at 5°C/min and held at 250°C for 5 min. Finally, the temperature was increased to 290°C at 10°C/min and held for 10 min. For the analysis of the SPE extracts the initial temperature was 50°C rather than 77°C. Atomic emission detection. The make-up and reagent gases were: oxygen (99.999%; Hoekloos, Schiedam, the Netherlands) at 1.8 bar, hydrogen (99.999%, Hoekloos) at 4.1 bar, 10% methane / 90% N2 (Hoekloos) at 4.8 bar and helium (99.999%; Praxair) at 4.1 bar. The make-up flow was set at 20 ml/min and the cavity pressure at 10 kPa. The temperatures of the heated zones were: transfer line, 300°C; cavity, 300°C and water, 65°C. The solvent venting was switched on 1 min after injection and switched off 4.5 min after injection. The spectrometer was purged with nitrogen at a flow-rate of 4 l/min. Standard recipes for elemental detection were used for all analyses. Mass spectrometry. Spectra were recorded over a scan range of m/z 45–350 amu at a scan rate of 2.3 scan/s and an electron energy of 70 eV. The temperatures of the heated zones were: transfer line, 300°C and quadrupole, 188°C. The following global settings were used for MS target analysis: correlation window, 0.018 min; non-reference window, 0.2 min. The integration parameters were: initial area reject, 1 area count; initial peak width, 0.02 min; shoulder detection, off; initial threshold, 7.7. Compound 87 6. Screening of surface water settings were: extraction window width, 0.5 min; quant signal, target ion with 20% relative uncertainty; response, by area; identification, best qualifier value; subtraction, no subtraction. 6.3 Results 6.3.1 Identification criteria used in target analysis When setting up the calibration in the Chemstation software, one target m/z value has to be set for each target compound. Furthermore, between one and three qualifier ion m/z values, including their relative intensities, can be selected for each target compound. During automated data analysis the extracted ion chromatogram of the target ion is searched for peaks in a specified time window after which the fit qualities for each peak found are calculated on the basis of the intensities of the qualifier ions. Depending on the selected value of the identification parameter (Ident By), the returned result is the best matching peak in the specified time window. In our case Ident By was set to Best Qualifier Value, in which case the result is the peak with the highest fit quality (Q value) with regard to the relative intensities of the qualifier ions. Furthermore, a target signal status (Cal Pk Found) is returned; its value depends on the percentage of relative uncertainty allowed in the qualifier ion intensities (set to 20% in our study). The value of Cal Pk Found is 0 when the target ion is not found. When the target ion is found, but the relative intensity of one or more of the qualifier ions deviates more than the allowed uncertainty (20% in this case) Cal Pk Found is 1, meaning the peak is flagged as “found but not qualified”. If the relative intensities deviate less than the allowed uncertainty, Cal Pk Found is 2 and the peak is flagged as “found and qualified”. Because the number of compounds in most sample extracts was very large, it is extremely time-consuming to manually check all the hits. Therefore, a number of early runs was checked manually and, on the basis of the results obtained, the following set of empirical rules was made for the decision of discarding or accepting certain hits. All peaks indicated by the Chemstation software as Cal Pk Found 2 were accepted as correct hits. Hits with a fit quality larger than 80% and with areas larger than 100,000 counts were also accepted without further checking. Hits that were not included in the above mentioned groups and with a fit quality larger than 40% and areas of over 40,000 counts were subjected to manual inspection. These selection criteria were applied as follows. After running the quantification function of the MS Chemstation software, all the quantification results were transferred to the spreadsheet program Excel. Using an Excel macro, the compounds that required manual checking were then selected and manually checked in the Qedit Quant Result mode of the MS Chemstation program. Where these rules were followed, the number of compounds that required manual checking was, on 88 6. Screening of surface water average, 10–20 per sample. Corrections for compounds present in the blank were made by discarding the result if the concentration was not at least twice the concentration in the blank. Since the toxicity studies referred to earlier have to be conducted in aqueous samples, while GC is not really compatible with water injections, with the SPE approach two extraction steps were performed. First, the XAD SPE procedure was performed to produce a concentrated water sample. This concentrate was subsequently extracted by means of LLE with dichloromethane. Previous experiments showed that the analyte recoveries for the entire procedure generally are in the range of 40–70% [195]. However, they can be much lower, i.e. less than 10%, for compounds with a high Henry’s law constant (log H>2) such as the tri- to pentachlorobenzenes. Because precise recovery data are known for only a small proportion of the 430 target compounds in the present study, all analyte concentrations in the SPE procedure were calculated without any correction for loss of recovery. This means that the experimental results will often be lower than the actual concentrations, although general experience has taught us that differences will not easily be larger than 2-fold. With the SPMD procedure, on the other hand, the concentrations are calculated in ng/g fat which can be interpreted as close to a worst-case scenario because of the lack of biotransformation in SPMDs. The obvious conclusion is that the quantitative results should be regarded with some caution: the values indicate concentration ranges rather than precise values. Accordingly, all figures in the Appendix are presented with one significant number. The analysis of real samples caused a problem that was not expected during method development. During manual revision as described above, it was found that the mass spectral similarity indicated the presence of several compounds in the target list but at shifted retention times. After careful study it was clear that these compounds all eluted 0–3 min before the large di-isobutyl phthalate peak (25). However, not all peaks in this time window were shifted. After correction, six compounds could be added to the list: simazine (4), atrazine (5), Fyrol PCF (139), Fyrol CEF (140), N-butylbenzenesulphon- amide (57) and caffeine (127). Most shifts were 0.1–0.3 min, but one was as high as 0.75 min. The apolar fragrances 16–18 did not show any shift of retention time although they also eluted just before the phthalate peak. Obviously, only the polar and medium polar compounds are affected. This clearly is an effect of overloaded, non-linear, chromatography in which the adsorption isotherms of the shifted compounds are influenced by the presence of the large phthalate peak. The literature on this subject is limited. Grob describes the influence of adjacent peaks on the peak shapes of other compounds [196], and shifts of retention time are briefly discussed by Poole and Poole [197]. Whether a compound is influenced by an adjacent overloaded peak depends on 89 6. Screening of surface water many physicochemical parameters of the compound in question and the analytical column; in the present case the effect appears to depend on their polarity. In summary, whenever a target analysis is performed one should be aware that false negatives may occur due to shifts of retention time whenever there are one, or a few, excessively large peaks. 6.3.2 Comparison of SPMD and SPE extractions: number of identified compounds One of the interesting aspects of this research project is the possibility to compare SPMD and SPE extractions. Although SPE extraction is related to the set of microcontaminants present in the river water at one particular time, and the SPMD procedure to a cumulative view over a one-month period, these samples are a good set to study the nature of the compounds which are detected with both methods. When the full-scan GC–MS chromatograms (Fig. 6.2) recorded after SPE and SPMD extraction are compared, it is evident that the SPE method extracts more compounds which elute in the first 15 min of the chromatogram. Since an apolar GC column was used, one can conclude that these are relatively volatile analytes. The number of compounds in the high boiling range, specifically in the 20–40 min time window, is greater in the SPMD extracts. From among the 430 target compounds, a total number of 116 was found by using SPE extraction of the five samples. With SPMD extraction 60 compounds were found. An overview of the number of compounds identified in every sample is shown in Table 6.1. Table 6.1: Number of compounds identified in SPE and SPMD extracts Location SPE and/or only SPE only SPMD SPMD LOB 103 59 26 EIS 95 55 16 MSL 104 68 22 SOD 85 49 23 NWK 73 32 24 90 6. Screening of surface water SPMD 70 Triolein 17 18 69 57 29 16 68 72 Response 15 71 20 510 15 20 25 30 35 40 Time (min) b* SPE 25 26 121 126 140 29 111 139 a* 57 18 22 138 Response 17 510 15 20 25 30 35 40 Time (min) Fig. 6.2: Full-scan GC–MS chromatograms obtained for SPE and SPMD extracts of river Meuse water sampled at Eijsden. For peak numbers, see Appendix at the end of this chapter. * Peaks a (benzothiazole) and b (tris(butoxyethyl) phosphate) are not included in the Appendix because of blank correction according to identification criteria for target analysis. The data show that there is an area of overlap, i.e. compounds that are detected in both types of extract. More importantly, the complementary nature of the two techniques is clearly evident, with ‘only SPMD’ contributing some 20–25, and ‘only SPE’ some 45– 55 compounds per sample which were not detected with the alternative procedure. When comparing SPE and SPMD extraction, the log Kow values are a relevant parameter of the analytes. Most hydrophobic-based SPE procedures tend to extract compounds over a wide polarity range. However, because of the low water solubility of the more hydrophobic compounds, their concentrations are generally very low and it is therefore difficult to determine them in the aqueous phase. As regards SPMDs, they can sample compounds with a low log Kow, but their capacity to isolate compounds rapidly increases with increasing hydrophobicity. This is reflected by the effective sampling rate which is the volume of water sampled per day by a standard, 1-m long SPMD containing 1 g of triolein. The effective sampling rate falls below 1 l/day for compounds with a log Kow lower than 4, and has a maximum of 5 l/day for compounds with log Kow values between 5 and 6. Because the SPMDs are deployed for 20–30 days, the large volume of water sampled allows the detection of apolar compounds which are present at very low concentrations. To visualise the influence of the hydrophobicity effect, in Fig. 6.3 the number of compounds detected by each method, N, is plotted against the log Kow value. 91 6. Screening of surface water 100 SPE 80 SPMD 60 N 40 20 0 -3-2 -1 0 1 2 3 4 5 6 7 8 log K ow Fig. 6.3: Number of compounds detected (N) in the five samples plotted against their log Kow values. The latter were combined per unit of log Kow . As expected, the SPE method extracts compounds over a wide polarity range and the SPMDs primarily in the range log Kow 4–7. There is one complicating factor here. In the present, as in many SPE procedures, the water samples were not filtered. This will result in a considerable additional extraction of many compounds adsorbed to suspended matter, which is typically present at 25 mg/l in Dutch surface waters. It has been reported that compounds with log Kow<4 are present for 90% or more in the dissolved phase, while compounds with log Kow>5 are largely bound to suspended matter [192;193]. Therefore, it can be expected that most compounds with log Kow>5 that are found in the SPE extracts were extracted or desorbed from the suspended matter. 6.3.3 Comparison of SPMD and SPE extraction: nature of compounds The Appendix summarises the results of the target analysis of the SPE and SPMD extracts. The classification of the microcontaminants was based on their usage (e.g. pesticides, fragrances) and their nature (e.g. phenols, PAHs). That is, one should keep in mind that some compounds might as well have been listed under a different heading. The log Kow values included in the table are either experimental data from several sources (indicated by an asterisk in the Appendix) or data calculated using the program ClogP (BioByte, Claremont, CA, USA). The most abundant compound in the SPE extract is di-isobutylphthalate (25)*, which makes up about 80–90% of the total concentration of compounds detected. Since it was later suspected to largely be a contaminant originating from tubing used to siphon the sample, this finding should not be considered representative. Other compounds present at relatively high concentrations of, occasionally, up to or above 1 µg/l, are the phthalates 26 and 29, the surfactant Surfynol 104 (134), phenol (30) and several glycol * For compound numbering see Appendix 92 6. Screening of surface water dimethyl ethers (120, 129, 135). It also becomes clear that essentially all modern pesticides, which are relatively polar, can only be detected using SPE extraction. Only the herbicide triallate (3) was detected twice using SPMD, but not with SPE. However, this can also be due to the grab-sampling nature of the SPE method, as against the integrative nature of the SPMD procedure. The most abundant compounds detected after SPMD extraction were the PAHs, a number of phthalate esters and several polycyclic synthetic fragrances. Because of their bioaccumulation potential and their volume of use, these fragrances are a most relevant group of compounds to monitor and several studies have been conducted to determine the toxicological consequences [198-202]. Their presence has been reported in surface water, sea [203] and sewage [204] water, aquatic species [205;206] and human adipose tissue [207;208]. Fig. 6.4 shows the extracted ion chromatograms (EIC) of typical m/z values of some fragrances in the target list that were detected using the mass spectral target analysis. O *17 *17 O Response (18) O (17) *17 O *17 (15) *17*17 (16) m/z 258 x 0.14 O m/z 246 (14) m/z 244 x 0.14 19 20 21 22 23 24 25 Time (min) Fig. 6.4: Extracted ion chromatograms of EIS SPMD extract showing the profiles of the polycyclic synthetic fragrances. The peak intensity in the 23–23.5 min range was scaled down 7-fold in the m/z 244 and 258 traces. Products related to HHCB (17) are indicated by an asterisk (see text). Five polycyclic synthetic fragrances were detected, the highest concentrations being found for HHCB (17) and AHTN (18). Several peaks (indicated in Fig. 6.4 by asterisks) showed mass spectra similar to that of HHCB. These compounds were also present in 93 6. Screening of surface water the authentic HHCB standard, but in different proportions. Several of these peaks can be assigned to side products formed during HHCB synthesis. Differences in relative concentration levels were also reported for human adipose tissue [207] and aquatic species [206]. Concentrations of analytes freely dissolved in water cannot readily be calculated for the SPMD extracts. However, most fragrances were also detected using SPE at concentrations from 6 to 270 ng/l. Other studies have reported concentrations of up to 6 µg/l of HHCB in heavily polluted waters [202]. 6.3.4 Comparison of different sampling locations The sampling points LOB and EIS are at Lobith and Eijsden where the rivers Rhine and Meuse, respectively, enter the Netherlands (Fig. 6.1). Because of industrial, agricultural and other activities in the basins of these rivers in Germany and Belgium, high levels of microcontaminants can be expected here. Two other sampling locations, MSL and SOD, are located in tidal estuaries and a dilution effect caused by seawater can, consequently, be expected. The sampling point NWK is located in an agricultural area which is not directly connected with the major river systems and can be considered a reference point. Generally speaking, for most classes of compounds, the number of analytes detected at NWK was indeed lower than at the other sampling locations; exceptions were the ubiquitous alkanes, PAHs and plasticisers (cf. Appendix). Notable examples are the pesticides, phenols and anilines which are essentially absent at NWK, and the much lower concentrations recorded for the fragrances HHCB (17) and AHTN (18). Finally, an interesting compound is the drug carbamazepine (118), which was found at concentration levels of 10–25 ng/l at LOB and MSL. This probably reflects the higher population density of the Rhine compared with the Meuse basin—not emissions due to losses of production plants (if any) in the former region. As shown by the data of Table 6.1 and the Appendix, the results for the other locations—and specifically for LOB, EIS and MSL—are, generally speaking, fairly similar. However, despite this general similarity, there are some noticeable differences between the estuarine and up-river locations which were found when the AED bromine trace was studied. This will be discussed in the next section. 6.3.5 Potential of AED screening An interesting group of microcontaminants to study in some detail are the halogenated compounds. The use of simultaneous AED/MS detection easily allows the detection and identification of heteroatom-containing compounds [54]. When the AED chlorine channel chromatograms of the five SPMD samples were studied, up to 50 chlorinated compounds were detected in the SOD, EIS, MSL and LOB samples. As an example the LOB sample was studied in more detail (Fig. 6.5). 94 6. Screening of surface water In the SPE extract four chlorinated compounds were identified that were not in the target list. Two of these (200, 202) were found to be chlorinated anilines. The structure of the dichlorodimethoxybenzene (201) cannot be deduced from the mass spectrum; a likely candidate is 1,4-dichloro-2,5-dimethoxybenzene, which is used as a soil fungicide. In the SPMD extract eleven peaks were identified as PCBs (206-209) which are known to accumulate well in SPMDs; the number of chlorine substituents ranged from three to six. In addition, peaks were visible in the chlorine trace which were due to the target compounds hexachlorobutadiene (123) and penta- and hexachlorobenzene (53, 55). Furthermore, next to the disinfectant Triclosan (38), the degradation product Triclosan methyl (205) was detected. As for brominated compounds, none were detected at EIS. At NWK and LOB, 5 to 10 compounds were detected at very low levels; higher levels were found at SOD and MSL. The structures of the compounds tentatively identified by mass spectral library searching (Wiley, 6th ed., Pallisade, NewField, NY, USA) are shown in Fig. 6.6; their names are given in Table 6.2. 140 20 18 16 14 12 Response 10 139 (isomers) 34 203 8 202 40 1 200 201 204 6 4 205 coeluting w ith c 55 10 38 a: 206 b: 207 b 8 c c: 208 c d: 209 a b b c 123 a d d 6 53 b Response 4 10 15 20 25 30 35 Time (min) Fig. 6.5: Chlorine channel AED traces of the SPE (top) and SPMD (bottom) extracts of LOB river water. Peaks that are not labelled could not be identified. Peak numbers 200–209 indicate non-target compounds (see Table 6.2). 95 6. Screening of surface water Br Br 211 Br O Br 212 Br Br C C 213 NH2 H N Br H Br N 210 H O Response H Response 8 10 12 14 16 18 20 22 24 26 28 Time (min) Fig. 6.6: Full-scan GC–MS chromatogram (bottom) and AED bromine trace of the SPMD extract of MSL river water. The tentatively identified brominated compounds 210–213 are indicated (see Table 6.2). The presence of relatively high concentrations of brominated compounds at MSL where a large part of the river water entering the Netherlands at LOB and EIS (where the concentrations were low) leaves the country, is of interest. It implies that their source of emission is in the Netherlands. Literature study revealed several possible sources for the brominated compounds. Dibromoacetamide is a known degradation product of the biocide 2,2-dibromo-3-nitrilopropionamide which is used as an antibacterial agent in cooling water systems [209]. Tribromophenol is used as a flame retardant in plastics. For the bromoindoles no applications were found. Table 6.2: Tentatively identified halogenated compounds No. Compound 200 Dichloroaniline 201 Dichlorodimethoxybenzene 202 Trichloroaniline 203 Dichloromethoxytoluene 204 p-Methylsulphonylbenzyl chloride 205 4-Chloro-1-(2,4-dichlorophenoxy)-2-methoxy-benzene (Triclosan methyl) 206 Trichlorobiphenyl 207 Tetrachlorobiphenyl 208 Pentachlorobiphenyl 209 Hexachlorobiphenyl 210 Dibromoacetamide 211 Bromoindole 212 Tribromophenol 213 Dibromoindole Here, one should consider that some 1600 naturally occurring organobromine compounds are known today [56]. This suggests that the bromoindoles are not of anthropogenic origin. Both blue-green and red algae are indeed known to contain a wide range of bromoindoles and even bromophenols are naturally occurring compounds. The fact that most organisms containing organobromines are of marine origin and that, in the 96 6. Screening of surface water present study, the highest concentrations were found in the brackish waters of the SOD and MSL samples, adds to the suggestion that these compounds are of biological origin. 6.4 Conclusions The complementary use of SPMD and SPE as analyte isolation techniques in combination with GC–AED/MS is a valuable strategy for the screening of large numbers of organic micropollutants. A wide polarity and volatility range is covered and, next to the target analytes of interest, heteroatom-containing unknowns can be detected and provisionally identified. Semi-quantification of the results is also possible. In the present study, some 150 compounds were found in a series of surface water samples covering the major river systems of the Netherlands. Their concentrations were in the low-ng/l to low-µg/l range. Pesticides, fragrances, PAHs, plasticisers, phenols and anilines were among the classes of compounds most frequently detected. The development of a proper identification strategy for the MS target analysis was found to be a critical factor for the correct identification and quantification. The currently chosen parameter settings allowed the automated target analysis of 430 compounds with a low risk of false positives. Finally, it should be noted that during occasional non-target mass spectral library searching the observed match qualities were often very low. One explanation, of course, is that the library, although extensive, has a finite size or, in other words, for many of the unknowns, there will be no corresponding entries in the library. However, another confounding factor is that, at the low concentration levels studied, co-elution of several compounds may be expected to be a frequently encountered phenomenon. In other words, for an extensive non-target analysis one will have to use a technique with much higher separation power. As recent attempts have shown, GC combined with time-of- flight mass spectrometry (GC×GC–ToF-MS) is a good candidate for such an approach [167;170]. Work in this area is currently in progress in our group. 97 6. Screening of surface water * * * * * * * * * * 2.7 3.1 2.2 1.8 2.2 3.3 3.0 5.5 5.7 4.8 6.3 0.8 1.6 3.4 2.4 4.0 1.2 4.1 logP 19.9 20.95 21.32 23.49 12.63 16.61 23.73 30 10 200 500 40 500 500 400 600 600 20 300 600 700 1000 20 200 400 500 700 800 2000 60 60 40 100 800 2000 9.47 Ret t LOB EIS MSL SOD NWK Ret t Ret (min) NWK SOD MSL EIS LOB t Ret 13.45 17.16 23.67 10.53 11.06 11.34 22.66 23.64 24.03 26.01 10.07 15.45 16.98 19.31 21.02 21.26 26.60 7 20 30 30 30 30 40 50 200 40000 2 7 10 80 70 50 20 50 80 20 80 100 O NWK SOD 30000 P SPMD SPE 9 20 50 40 30 60 70 80 90 40 100 100 100 300 50000 5 10 60 30 30 10 10 20 30 60 I MSL EIS 200 100 200 400 300 100 60000 Concentration (ng/l water) Concentration (ng/g fat) Concentration (ng/l water) Concentration 7 4 7 60 90 300 200 200 300 200 100 50000 90-43-7 78-40-0 84-66-2 77-93-0 84-69-5 AR LOB CASRN 8-0042.137* 3.7 28.01 4 122-34-9 886-50-0 617-94-7 464-48-2 131-11-3 126-71-6 126-73-8 331- 092.446* 4.6 22.74 9 * 2.6 20 23.92 200 6.1 30 23.24 1194-65-6 80 50 2303-17-5 20 800 1912-24-9 10 4000 3000 25.78 6 100 60 2216-51-5 300 200 1222-05-5 1506-02-1 62-964 81 . * 2.7 * 3.1 * 2.1 28.12 28.43 29.85 40 70 5 6 20 26225-79-6 51218-45-2 67129-08-2 10482-56-1 13171-00-1 15323-35-0 32388-55-9 value) ow Nature and concentrationsof compounds identified SPE in and SPMDextracts from five surface water samples : : Pesticides o-Phenylphenol Fragrances Plasticizers Name Dichlobenil (2,6-Dicholorobenzonitrile) ester) Simazine (6-Chloro-N,N'-diethyl-1,3,5-triazine-2,4-diamine) diamine) methanesulfonate) methylethyl)acetamide acetamide) alpha,alpha-dimethyl- Benzenemethanol, Camphor L-(-)-Menthol l-alpha-Terpineol 4-Acetyl-6-t-butyl-1,1-dimethylindan (ADBI) 5-Acetyl-1,1,2,3,3,6-hexamethyl (AHMI) indan Acetyl cedrene (HHCB) 7-Acetyl-1,1,3,4,4,6-hexamethyl (AHTN) tetrahydronaphthalene acid, triethylPhosphoric ester 1,2-Benzenedicarboxylic acid, ester dimethyl acid, tris(2-methylpropyl) Phosphoric ester 1,2-Benzenedicarboxylic acid, diethyl ester acid, tributyl Phosphoric ester 1,2,3-Propanetricarboxylic triethyl acid, 2-hydroxy-, ester 1,2-Benzenedicarboxylic acid, ester bis(2-methylpropyl) 1 2 Triallate (Bis(1-methylethyl)carbamothioic acid S-(2,3,3-trichloro-2-propenyl)3 4 Atrazine (6-Chloro-N-ethyl-N'-(1-methylethyl)-1,3,5-triazine-2,4-diamine) 5 Terbutryn (N-(1,1-dimethylethyl)-N'-ethyl-6-(methylthio)-1,3,5-triazine-2,4- 6 (2-Ethoxy-2,3-dihydro-3,3-dimethyl-5-benzofuranol Ethofumesate 7 Metolachlor (2-Chloro-N-(2-ethyl-6-methylphenyl)-N-(2-methoxy-1- 8 Metazachlor (2-Chloro-N-(2,6-dimethylphenyl)-N-(1H-pyrazol-1-ylmethyl)- 9 Appendix (*: experimentallogK 10 11 12 13 14 15 16 1,3,4,6,7,8-Hexahydro-4,6,6,7,8,8-hexamethylcyclopenta-gamma-2-benzopyran 17 18 19 20 21 22 23 24 25 98 6. Screening of surface water * * * * * * * * * * * * * * * * * * * * * * * * 4.5 4.7 4.6 7.6 1.5 2.3 2.4 2.5 3.1 2.9 3.3 4.1 4.8 4.0 1.5 4.0 2.0 1.6 1.9 4.6 1.6 4.2 1.6 4.0 logP 7.7 32.9 6.59 25.79 33.69 35.94 17.98 28.63 10.33 11.55 11.32 80 30 20 400 900 6 10 400 200 500 100 7 7 3 90 20 900 300 200 2 700 200 400 100 500 3 3 10 20 500 700 7.62 6.83 8.58 9.10 9.46 Ret t LOB EIS MSL SOD NWK Ret (min) t Ret NWK SOD MSL EIS LOB t Ret 28.31 35.26 36.04 38.30 10.66 10.99 10.99 11.04 11.42 12.83 10.42 12.11 14.18 30 30 20 700 1000 5 6 8 20 20 O NWK SOD P SPMD SPE 20 30 20 60 70 100 100 1000 2000 2 20 40 80 70 20 20 40 70 20 I MSL EIS 600 200 3000 Concentration (ng/l water) Concentration (ng/g fat) Concentration (ng/l Concentration water) 6 3 10 40 30 200 400 100 100 2000 57- 02 04 51 . * 2.7 84-74-2 85-68-7 15.12 40 88-69-7 20 98-54-4 20 40 95-76-1 95-93-2 98-86-2 98-95-3 95-94-3 92-52-4 AR LOB CASRN 0-182 .717* 1.7 * 3.5 17.73 20 9.07 115-86-6 117-81-7 20.30 108-95-2 105-67-9 108-68-9 20 526-75-0 3 120-83-2 8 118-79-6 20 7 100-61-8 122-39-4 100-52-7 120-82-1 103-83-3 140-29-4 101-84-8 770-35-4 3380-34-5 (continued) Phenols Anilines compounds based aromatic Other N,N-Dimethylbenzylamine Name 1,2-Benzenedicarboxylic acid, dibutyl ester 1,2-Benzenedicarboxylic ester acid, phenylmethyl butyl acid, ester triphenyl Phosphoric 1,2-Benzenedicarboxylic acid, bis(2-ethylhexyl) ester Phenol 2,4-dimethyl-Phenol, 3,5-dimethyl-Phenol, 2,3-dimethyl-Phenol, 2,4-dichloro-Phenol, 2-(1-methylethyl)-Phenol, 4-(1,1-dimethylethyl)-Phenol, 2,4,6-tribromo-Phenol, 5-chloro-2-(2,4-dichlorophenoxy)-Phenol, (Triclosan) Benzene, 1,2,4,5-tetramethyl- Benzaldehyde Benzene, 1,2,4-trichloro- 1-phenyl- Ethanone, Benzene, nitro- Benzene, 1,2,4,5-tetrachloro- Benzeneacetonitrile ether Diphenyl 2-Propanol, 1-phenoxy- 1,1'-Biphenyl 26 27 28 29 30 31 32 33 34 35 36 37 38 Benzenamine, N-methyl- 39 Benzenamine, 3,4-dichloro- 40 Benzenamine, N-phenyl- 41 42 43 44 45 46 47 48 49 50 51 52 Appendix 99 6. Screening of surface water * * * * * * * * * * * * * * * * * * * * * * * * * * 5.2 2.2 5.7 3.2 2.5 2.8 3.3 2.9 3.9 3.9 3.2 3.9 3.9 4.2 4.5 4.5 5.2 4.9 5.8 5.8 5.8 6.1 6.1 6.6 6.8 6.6 5.0 6.5 6.1 7.6 8.6 9.2 logP 7.96 9.93 14.23 19.57 17.56 21.92 10.18 10.27 12.76 13.54 16.51 22.07 21.82 27.87 28.85 34.63 34.76 39.31 39.54 40.95 46.36 46.48 47.13 3 6 9 6 30 40 20 50 30 20 700 500 200 400 100 1000 4000 3000 1000 6 6 6 3 50 20 40 40 10 80 200 100 300 300 1000 3 3 20 80 30 20 30 20 40 70 90 500 200 500 200 3000 5000 2000 5 2 2 20 20 20 10 10 70 20 40 40 100 500 200 600 400 600 200 1000 7 6 20 70 10 30 40 10 90 20 10 60 20 30 40 600 400 100 3000 3000 1000 7.71 9.78 Ret t LOB EIS MSL SOD NWK Ret t Ret (min) NWK SOD MSL EIS LOB t Ret 18.87 20.47 24.73 37.96 15.57 16.32 19.16 24.31 30.21 31.17 11.38 12.89 16.69 19.74 9 5 40 90 10 10 40 10 30 70 3 1 4 1 40 20 100 O NWK SOD P SPMD SPE 2 8 7 4 60 30 20 10 10 10 30 40 60 200 1 8 4 4 2 80 40 10 10 20 20 I MSL EIS 300 Concentration (ng/l water) Concentration (ng/g fat) Concentration (ng/l water) Concentration 8 9 5 4 50 60 10 40 20 100 300 91-20-3 95-13-6 91-57-6 90-12-0 83-32-9 86-73-7 85-01-8 56-55-3 50-32-8 53-70-3 AR LOB CASRN 608-93-5 134-62-3 118-74-1 119-61-9 791-28-6 275-51-4 208-96-8 120-12-7 206-44-0 129-00-0 218-01-9 205-99-2 207-08-9 193-39-5 191-24-2 124-18-5 112-40-3 629-50-5 629-62-9 544-76-3 3622-84-2 1120-21-4 (continued) Polyaromatic hydrocarbons (PAHs) hydrocarbons Polyaromatic Alkanes Name Benzene, pentachloro- Benzamide, N,N-diethyl-3-methyl- Benzene, hexachloro- Benzophenone N-butyl- Benzenesulfonamide, triphenyl- oxide, Phosphine Naphthalene 1H-Indene 2-methyl- Naphthalene, 1-methyl- Naphthalene, Azulene Acenaphthylene 1,2-dihydro- Acenaphthylene, 9H-Fluorene Anthracene Phenanthrene Fluoranthene Pyrene Benz[a]anthracene Chrysene Benzo[b]fluoranthene Benzo[k]fluoranthene Benzo[a]pyrene Indeno[1,2,3-cd]pyrene Dibenz[a,h]anthracene Benzo[ghi]perylene n-Decane n-Undecane n-Dodecane n-Tridecane n-Pentadecane n-Hexadecane 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 Appendix 100 6. Screening of surface water * * * * * * * * * * * * * * 9.7 1.6 1.9 1.0 2.0 2.0 0.0 2.1 2.1 2.6 1.2 2.6 2.2 1.4 4.1 3.2 3.4 2.5 10.2 10.2 10.8 10.7 11.3 11.8 12.3 12.9 10.2 13.4 13.9 14.5 12.9 15.0 15.5 16.0 logP 38.6 8.72 9.83 22.3 30.57 33.64 37.05 40.48 10.78 14.47 16.86 3 6 60 50 30 300 6 20 20 20 400 200 50 100 20 300 200 6 80 10 90 400 4.74 7.48 7.77 Ret t LOB EIS MSL SOD NWK Ret t Ret (min) NWK SOD MSL EIS LOB t Ret 22.54 22.62 25.02 25.16 27.28 29.32 31.18 33.03 34.73 36.14 36.38 37.89 39.49 40.91 42.69 44.87 11.84 12.04 12.25 12.40 12.87 13.06 13.64 14.16 14.57 15.29 17.22 19.69 24.80 35.29 2 7 60 50 70 30 70 20 40 30 10 20 200 200 100 200 7 6 4 8 4 1 30 20 60 300 700 200 O NWK SOD P SPMD SPE 8 4 80 30 70 40 40 10 20 20 10 20 10 20 60 10 10 30 10 10 200 4 9 50 10 10 40 30 20 80 10 90 20 40 I MSL EIS 100 400 200 Concentration (ng/l water) Concentration (ng/g fat) Concentration (ng/l water) Concentration 9 9 9 8 20 30 20 20 20 20 10 30 10 30 90 30 20 90 40 30 100 600 95-16-9 91-22-5 91-63-4 54-11-5 91-64-5 AR LOB CASRN 629-78-7 593-45-3 638-36-8 629-92-5 112-95-8 629-94-7 629-97-0 638-67-5 123-95-5 646-31-1 629-99-2 630-01-3 111-02-4 593-49-7 630-02-4 630-03-5 583-61-9 108-75-8 105-60-2 119-65-3 120-72-9 491-35-0 132-64-9 615-22-5 260-94-6 298-46-4 1921-70-6 2379-55-7 14667-55-1 (continued) Heterocyclic aromatic compounds aromatic Heterocyclic Name n-Heptadecane 2,6,10,14-tetramethyl-Pentadecane, n-Octadecane 2,6,10,14-tetramethyl-Hexadecane, n-Nonadecane n-Eicosane n-Heneicosane n-Docosane n-Tricosane Octadecanoic ester acid, butyl n-Tetracosane n-Pentacosane n-Hexacosane 2,6,10,14,18,22-Tetracosahexaene, 2,6,10,15,19,23-hexamethyl-, n-Heptacosane n-Octacosane n-Nonacosane Pyridine, 2,3-dimethyl- Pyridine, 2,4,6-trimethyl- Pyrazine, 2,3,5-trimethyl- Benzothiazole Quinoline Caprolactam Isoquinoline 1H-Indole Quinoline, 2-methyl- Nicotine Quinoline, 4-methyl- Quinoxaline, 2,3-dimethyl- Coumarin Dibenzofuran Benzothiazole, 2-(methylthio)- Acridine Carbamazepine 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 Appendix 101 6. Screening of surface water * * * * * * * * 1.2 5.1 0.2 4.8 1.1 2.8 8.8 0.8 0.0 0.1 1.6 2.1 3.6 0.1 2.4 0.1 2.3 1.4 3.4 -0.4 -2.2 logP 8.35 10.13 11.93 22.06 21.46 25.69 20 10 20 60 20 7 60 20 20 10 20 30 6.31 6.50 7.33 7.48 Ret t LOB EIS MSL SOD NWK Ret t (min) t Ret NWK SOD MSL EIS LOB t Ret 10.01 10.29 26.01 11.60 11.72 12.47 12.64 12.89 14.65 16.70 19.44 20.61 24.85 24.19 28.50 3 40 20 200 500 300 7 30 80 80 20 20 10 900 100 500 100 200 200 600 300 1000 O NWK SOD P SPMD SPE 90 30 80 40 40 80 30 600 200 500 200 400 600 400 1000 70 60 10 60 10 I MSL EIS 200 100 500 200 100 700 400 1000 Concentration (ng/l water) Concentration (ng/g fat) Concentration (ng/l water) Concentration 80 40 20 300 200 300 100 100 200 200 600 200 400 500 1000 2000 87-68-3 58-08-2 83-33-0 84-65-1 AR LOB CASRN 111-96-6 556-67-2 632-22-4 102-54-5 541-02-6 126-54-5 112-49-2 761-65-9 126-86-3 143-24-8 483-63-6 115-96-8 865- 02 02 01.13 01.13.9 16.41 30 5131-66-8 30 19.31 20 20 3735-92-0 50 20 20 6846-50-0 6145-73-9 56-742 04 01.12.1 14.01 10 10581-38-1 40 30 20 25265-77-4 10543-57-4 (continued) Miscellaneous compounds Miscellaneous Name 2-Propanol, 1-butoxy- Diethylene glycol ether dimethyl Cyclotetrasiloxane, octamethyl- Urea, tetramethyl- 1,3-Butadiene, 1,1,2,3,4,4-hexachloro- 2,2,6,6-tetramethyl-4-piperidone monohydrate Ferrocene decamethyl- Cyclopentasiloxane, Caffeine 2,4,8,10-tetraoxaspiro[5.5]undecane Triethylene glycol ether dimethyl acid,Carbamodithioic ester dimethyl-, methyl 1-Indanone N,N-dibutyl- Formamide, (+,-,meso)2,4,7,9-tetramethyl-5-decyne-4,7-diol Tetraethylene glycol dimethyl ether (N-Ethyl-N-(2-methylphenyl)-2-butenamide) Crotatmiton Tetraacetylethylenediamine (3:1) 1-Propanol, 2-chloro-, phosphate (3:1) 2-chloro-,Ethanol, phosphate 9,10-Anthracenedione 119 120 121 122 123 124 125 126 127 128 129 130 131 132 with acid, 2,2,4-trimethyl-1,3-pentanediol Propanoic 2-methyl-, monoester 133 134 135 acid, Propanoic 2-methyl-, 2,2-dimethyl-1-(1-methylethyl)-1,3-propanediyl ester 136 137 138 139 140 141 Appendix 102 7 7. Responses in sediment bioassays used in the Netherlands: can observed toxicity be explained by routinely monitored priority pollutants? 7.1 Introduction In the 1960s and 1970s the Rhine and Meuse rivers in the Netherlands were heavily polluted with various inorganic and organic substances. This led to considerable sediment contamination with persistent micropollutants. As a consequence of emission reductions the quality of the water of both rivers has improved significantly since the 1980s and new layers of sediments which are deposited are cleaner than before [210]. However, in many river localities such as floodplains and downstream areas where contaminants were deposited in the past, poor benthic invertebrate communities are still found and sediments do provoke toxic effects in bioassays [211;212]. Until recently, in the Netherlands attention was focused on approx. 50 persistent so- called priority substances which were the main cause of (acute) toxicity of Rhine and Meuse river water. This toxicity has decreased over the last two decades, but toxic effects are still observed occasionally. In surface waters, a lot of these observations cannot be contributed to known substances. In the 1990s more than 89% of the toxicity in the lipophilic fraction of the water could not be explained by compounds identified with standard GC–MS analysis [213]. The principal objective of this study was to assess the toxicity of a large number of sediments with different degrees of pollution using standard bioassays and to substantiate if the observations could be explained with results from these routine chemical sediment analyses. In addition, samples from five of the most polluted localities were used to evaluate the sensitivity of additional and novel bioassays and extracts of pore water from these five sediments was subjected to further chemical screening to detect other possible toxicants. Adapted from: J. Lahr, J.L. Maas-Diepeveen, S.C. Stuijfzand, P.E.G. Leonards, J.M. Druke, S. Lucker, A. Espeldoorn, L.C.M. Kerkum, L.L.P. van Stee, A.J. Hendriks, Water Res. 37 (2003) 1691 7. Relation of ecotoxicity with detected pollutants 7.2 Materials and methods 7.2.1 Sampling and pre-treatment Sediments were collected at twenty different sites in the Netherlands (Fig. 7.1) in various types of water and with different degrees of pollution (Table 7.1). At each sampling site at least five subsamples from the top 10 cm of the sediments were taken using an Ekman–Birge grab (surface area 0.0225 m2). After transport to the laboratory they were stored at 4°C. Prior to testing and analysis, the subsamples from each site were mixed, thoroughly homogenised and sieved over a mesh width of 500 µm to remove larger debris. After the sediments had been left to settle for a few days, the above-standing water was removed by decantation. For silty sediments, pore water was prepared by 17.5-min centrifugation at 3,500 rpm at 5°C until visibly clear water was obtained. In the case of sandy samples, pore water was prepared by using a high-air-pressure sediment press equipped with 0.45 µm pore-size glass-fiber filters. Lake IJssel North Sea GERMANY Markermeer THE NETHERLANDS Ketelmeer Eemmeer Vossemeer Drontermeer Brabantse Biesbosch Wolderwijd Puttershoek Twenthekanaal Lobith Haringvliet Gameren Rhine Dommel SWTP Borgharen Volkerak Amer Dommel Meuse BELGIUM Hollands Diep Neerpelt Eijsden Fig. 7.1: Locations in the Netherlands where sediments were sampled. 104 7. Relation of ecotoxicity with detected pollutants Table 7.1: Characteristics of sampling sites and sedimentsa Site Location Type of site Type of Pollution Polluting substances code sediment classc VM Vossemeer lake fine sand 0 - MM Markermeer lake silty sand 0 - WO Wolderwijd lake sandy silt 0 - EM Eemmeer lake sandy silt 0 - TW Twenthekanaal man-made channel silty sand 2 PAHs, PCBs VK Volkerak/Zoommeer lake sandy silt 2 PAHs, PCBs HV Haringvliet lake silt 2 Cd, Hg, Cu, PAHs, PCBs, HM Helmond SWTP stream near SWTP silty sand 3 CB 52 KMb Ketelmeer lake sandy silt 3 PAHs, PCBs HD Hollands Diep lake ND 3 HCB PTb Puttershoek lowland river sandy silt 3 Hg, Cu, HCB, PCBs GM2 Gameren 2 secondary river silty sand 3 HCB, PCBs, DDT channel DMN Dommel Neerpelt stream coarse sand 4 Cd GM1 Gameren 1 secondary river sandy silt 4 Zn channel DMSb Dommel SWTP stream near SWTP fine sand 4 Cd, Cu, Zn (Hg, Ni) BBb Brabantse Biesbosch river creek silt 4 Zn (Cd, PCBs) AMP Amer Power Plant inland harbor silt 4 Zn (Cd, Cu, CB 153) AM12 Amer loc. 12 lowland river sandy silt 4 Zn (Cd, PAHs, PCBs) LO Lobith inland harbor silt 4 Zn (Cd, Hg, Cu, Ni, PAHs, PCBs) BOb Borgharen lowland river sandy silt 4 Zn (Cd, Cu, PAHs, CB 153, DDT) a. The table shows the pollutants which are responsible for the designated class. For class 4 sediments, class 3 pollutants are additionally shown between parentheses. SWTP = Sewage Water Treatment Plant; ND = not determined. b. sediments used for additional investigations. c. Pollution class 0, not polluted; class 1, slightly polluted; class 2, moderately polluted; class 3, heavily polluted; class 4, extremely polluted. 7.2.2 Physical and chemical analysis Routine analysis of persistent priority micropollutants in the sediments and pore waters was performed using different methods such as atomic absorption spectrometry, gas chromatography (GC) and column liquid chromatography (LC). The parameters included heavy metal content (As, Cd, Cr, Cu, Hg, Pb, Ni and Zn), sixteen common polycyclic aromatic hydrocarbons (PAHs), various persistent organochlorine pesticides (OCPs: DDT and its derivates, dieldrin and related compounds, etc.), two chloro- benzenes, seven polychlorinated biphenyls (PCBs), mineral oil content and extractable organic halogenated hydrocarbons (EOX). The routine analyses were all conducted by a STERLAB-qualified laboratory. The degree of pollution with priority substances was classified from 0 (clean) to 4 (heavily polluted) on the basis of the chemical analyses according to the method prescribed by the Ministry of Transport and Public Works and using the software program WABOOS version 0.4 ([214] and references cited therein). 105 7. Relation of ecotoxicity with detected pollutants From five sediments (see Table 7.1) extra pore water was recovered. These liquid samples were subjected to analysis of the same eight heavy metals as in sediments and to additional screening for organic substances by GC with simultaneous atomic emission and mass-selective detection (GC–AED/MS) [55]. A sample of distilled water was processed and analysed simultaneously as a control. 7.2.3 Bioassays The effect of pore water on the luminescence of the bacterium Vibrio fischeri was measured using the Microtox®-test. Pore water was also used for the Thamnotoxkit F™ with the fairy shrimp Thamnocephalus platyurus. The bioassays lasted 24 h. No Observed Lethal Concentrations (NOLCs) were derived using Dunnett’s hypothesis test available from the TOXCALC software package. Median lethal concentrations (LC50) were calculated with the trimmed Spearman–Karber method from the same software. A reproduction test with the water flea Daphnia magna in pore water was used as a chronic bioassay. The highest test concentration which caused no significant effect on the number of juveniles was called the NOEC for reproduction (NOECR). The NOEC for survival (NOECS) was defined as the highest pore water concentration which caused 20% mortality or less. The LC50 of the adult water fleas was calculated similarly to the two microbiotests (previous paragraphs). Based on different criteria for each assay, effects observed in routine bioassays were classified as: 0, no effect; +, moderate effect; ++, strong effect. For further details the reader is referred to [214]. 7.2.4 Toxic Unit (TU) analysis and critical body burdens In order to identify micropollutants which contributed to effects observed in the standard bioassays, the results of chemical analyses of the sediments were compared to the known toxicity of the various individual priority substances. Toxicity values were collected from publications by the Dutch program setting integrated environmental quality standards. These were completed by data from, amongst others, the AQUIRE component of the internet-based ECOTOX database and from the TerraTox™ toxicity database ([214] and references cited therein). Assuming that the effects of pollutants with a similar mode of action are mainly additive, individual toxic unit values were added to estimate the toxicity of groups of compounds (concentration-addition model). These TU values for metals, PAHs or organochlorine compounds may explain observed toxic effects when no effects of individual substances are expected. 106 7. Relation of ecotoxicity with detected pollutants 7.3 Results 7.3.1 Degree of pollution The pollution classes of the twenty sediments are given in Table 7.2. Sediments ranged from unpolluted (class 0) to heavily polluted (class 4). As expected, the cleanest sedi- ments were found in the Lake IJssel area. The most polluted sites were found in the Rhine flood plain areas at Gameren and Brabantse Biesbosch, in the Lobith harbour, in the Amer (Meuse) and in the small stream Dommel. PAHs, PCBs and hexachloro- benzene (HCB) were often the predominant pollutants in moderately polluted (class 2) sediments. Heavy and extreme pollution was in most cases caused by zinc and/or cadmium. Table 7.2: Toxicity of pore water compared to results of Toxic Unit (TU) analysesa. Site Pollu- Vibrio fischeri Thamnocephalus Daphnia magna chronic code tion acute (Microtox) platyurus acute class (Thamnotoxkit F)b TU TU TU TU TU TU TU Effect metals PAHs OCPs Effect metals Effect metals PAHsc OCPs VM 0 0 0.01 0.01 0.00 0 0.06 0 0.12 0.01 0.00 MM 0 0 0.02 0.00 0.00 0 0.04 0 0.12 0.00 0.00 WO 0 0 0.02 0.00 0.00 + 0.07 0 0.13 0.02 0.00 EM 0 0 0.02 0.00 0.00 0 0.07 ++ 0.15 0.02 0.00 TW 2 + 0.03 0.05 0.00 0 0.07 0 0.21 0.25 0.00 VK 2 0 0.05 0.06 0.00 0 0.14 0 0.32 0.34 0.00 HV 2 0 0.10 0.00 0.00 0 0.22 0 0.49 0.25 0.00 HM 3 ++ 0.05 0.03 0.00 0 0.09 0 0.27 0.16 0.00 KM 3 0 0.11 0.11 0.00 0 0.26 0 0.66 0.60 0.01 HD 3 + 0.12 0.09 0.00 + 0.24 + 0.60 0.31 0.01 PT 3 0 0.14 0.13 0.00 + 0.32 + 0.79 0.61 0.01 GM2 3 0 0.11 0.26 0.00 0 0.29 0 0.63 0.61 0.02 DMN 4 0 0.05 0.03 0.00 0 0.13 0 0.81 0.08 0.00 GM1 4 0 0.15 0.07 0.00 0 0.31 0 0.80 0.37 0.01 DMS 4 ++ 0.40 0.08 0.00 + 0.82 0 2.68 0.33 0.00 BB 4 0 0.15 0.08 0.00 + 0.30 + 0.90 0.34 0.00 AMP 4 0 0.16 0.05 0.00 +d 0.31 +d 0.93 0.22 0.00 AM12 4 0 0.14 0.11 0.00 0 0.30 0 0.89 0.50 0.01 LO 4 0 0.23 0.24 0.00 0 0.48 + 1.23 1.21 0.01 BO 4 0 0.17 0.09 0.00 + 0.28 + 0.92 0.44 0.13 a. Not or very slightly toxic = 0, moderately toxic = +, highly toxic = ++. The sum of the toxic units, TU, was calculated per group of substances with a similar mode of action. b. no toxicity data found for PAHs and + OCPs, c. including QSAR estimations, d. NH4 during testing exceeded threshold concentrations for toxic effects 7.3.2 Standard bioassays and TU analysis The quantitative results of the routine bioassays (not shown) were classified as slightly, moderate or high, and are compared to results of TU analyses in Table 7.2. The quantitative results and the NOECs and NOLCs used for this TU analysis can be found in [214]. 107 7. Relation of ecotoxicity with detected pollutants The standard Microtox tests detected effects in four out of twenty sediment pore waters. The results did not seem to correlate well with the general degree of pollution with priority substances (Table 7.2). A rather complete set of toxicity data was collected for the assay, but none of the observed effects could be properly explained with the TU analysis, i.e. calculated TU values were below 1. At the location Dommel sewage water treatment plant (SWTP) the combined effects of metals may have been partially responsible for the observed effect (TUmetals = 0.40), in the other 4 samples that showed an effect TU for any of the groups of pollutants was below 0.15. Based on the number of sediments in which effects occurred, the crustaceans T. platyurus (acute) and D. magna (chronic) were the most sensitive organisms in the assays. The results of these two tests were also in reasonable agreement with each other. Both assays detected mostly moderate effects in the pore waters from Hollands Diep, Puttershoek, Brabantse Biesbosch and Borgharen. In the samples from Amer Power + Plant high ammonia concentrations (60–100 mg/l NH4 ) in the pore water during testing may have been responsible for the observed toxicity to T. platyurus. In the Dommel SWTP sediment that caused a moderate effect, TUmetals was 0.82, almost 1 (a combination of notably copper, lead and zinc toxicity). The effects in samples from Hollands Diep, Puttershoek, Brabantse Biesbosch and Borgharen can perhaps be partially explained by combined effects of metals (TUmetals between 0.24 and 0.31), but metal toxicity does not seem to have contributed to the observed moderate effect at Wolderwijd (TUmetals was only 0.07). There were no data on the toxicity of organic substances to the Thamnotoxkit F. There are stronger indications that the effects in the chronic Daphnia assay may have been at least partially caused by a combination of metals. TUmetals ranged from 0.60 to 1.23 in 7 of the 8 sediments in which effects were found in the corresponding pore water. PAHs may have (partially) contributed to the toxicity of Puttershoek and Lobith sediments to D. magna (TUPAHs of 0.61 and 1.21 respectively), whereas + ammonia concentrations were again high (30–100 mg/l NH4 ) in the sample from Amer Power Plant during testing. An impact of copper and other metals on D. magna in the chronic test was expected for the sediment pore water from Dommel SWTP (TUmetals = 2.68), but no adverse effects were observed. The strong toxicity of Eemmeer pore water to D. magna could not be substantiated on the basis of the current TU analyses. There were hardly any differences between the TU analysis results for Daphnia and PAHs with or without the inclusion of QSAR results. Although a significant number of data on the toxicity of organochlorine compounds to some of the species in the bioassays were found, the levels in the collected sediments were most likely too low to have caused effects in the bioassays. Hardly any data were 108 7. Relation of ecotoxicity with detected pollutants found on the toxicity of individual PCBs in the bioassays. Their contribution to the observed effects could therefore not be assessed, but it is very unlikely that at the measured concentrations these substances have contributed to the impact of pore water in these invertebrate bioassays, even in combination. 7.3.3 Additional chemical analyses in selected pore water samples The results of the extra chemical screening and effect modelling exercise are only briefly described (data not shown). Thirty-five compounds were identified by GC–MS/AED in the pore water of the five polluted sediments selected for further investigations (see previous sections). Phthalates were found in relatively large amounts in all five pore waters (only small amounts occurred in clean control samples). Oil components composed of long carbon chains (decanes and cosanes) were also frequently detected. For only 5 of these 35 compounds long-term toxicity data for D. magna were found. Another 5 NOECs could be derived by extrapolation from acute toxicity data. The remaining NOECs were estimated with a QSAR [215]. A similar number of data was found for V. fischeri. For decanes and cosanes toxicity data were found for none of the organisms, and due to the high Kow values of these compounds it was not possible either to estimate reliable toxicity levels with QSAR-modeling. When applying the concentration-addition model, toxicity of the identified compounds to the organisms was found to be negligible. Taking the extra QSAR data for D. magna into account, the observed effects could still not be properly explained by the additionally measured compounds. Estimates for critical body burdens in crustaceans for several compounds, notably phthalates, 2,4 bis-tert-butylphenol and two musk fragrances (HHCB and AHTN, see [55]), came close to or even exceeded the critical NOEC of 0.1 mmol/kg wet weight (data not shown). 7.4 Discussion 7.4.1 Test performance Effects were detected with all routine bioassays, but on the whole the acute T. platyurus test (Thamnotoxkit F) and the chronic D. magna reproduction test, seemed the most sensitive standard tests to monitor sediment/pore water toxicity. With the exception of Ketelmeer, the Thamnotoxkit and chronic Daphnia tests detected toxicity in pore water from the five polluted sediments which were used for additional screening. 109 7. Relation of ecotoxicity with detected pollutants 7.4.2 Explanation of effects None of the results for single tests seemed to correlate in a general way with the pollution classes deduced from routine chemical analyses of priority substances in the sediments from the present study. However, when the whole test battery is regarded, it follows that most class 3 and 4 sediments (heavily polluted) caused moderate to strong effects in one or more bioassays. Exceptions were Ketelmeer (class 3), Gameren 2 (class 3) and Dommel Neerpelt (class 4), for which not a single effect was observed. For Ketelmeer and Gameren 2 these results are in agreement with the results of TU analyses, which did not yield indications that detrimental effects might occur. However, for Dommel SWTP a high TU ratio was calculated for the chronic pore water bioassay with D. magna (2.68), but no effect was observed. Such differences may be due to altered bioavailability. In other instances effects were observed, that could not be properly explained by TU analysis of the concentrations of priority pollutants in the sediment. Bioluminescence of V. fischeri in the Microtox assay was moderately to highly reduced in samples from Twenthekanaal, Helmond SWTP, Hollands Diep and Dommel SWTP without an obvious cause (although metals may have been partially responsible for the toxicity of Dommel SWTP pore water). It is interesting to note that three of these sites concern smaller inland waters where sediment samples were either taken next to sewage water treatment plants (Dommel, Helmond) or in a water body under influence from (treated) domestic and industrial discharges (Twenthekanaal). Both types of effluent sources point to organic substances as the most likely culprits. The Microtox test is indeed relatively sensitive to organic substances, for instance to PAHs and mineral oil. However, according to our calculations PAH levels were not high enough to explain the effects in the three sediments. The effects on V. fischeri of the Twenthekanaal, Helmond SWTP and Dommel SWTP sediments seem to have been caused by other, unidentified, but probably organic compounds. Moreover, it can not be excluded that confounding factors other than the ones that were measured, for instance sulphide may have contributed to the toxicity of these sediments in the Microtox. Apparently clean sediments (class 0) from Wolderwijd and Eemmeer provoked a considerable impact in bioassays with various invertebrates. These results were not supported by TU analysis of routine chemistry data. Eemmeer and Wolderwijd are large inland lakes situated in the Lake IJssel area between the mainland and the newly created polder areas of Flevoland (Fig. 7.1). On both sides of the lakes there are vast areas of pastures and cereal crops where pesticides are used extensively. Water from the mainland site drains into a stream called the Eem which empties in Lake Eemmeer. 110 7. Relation of ecotoxicity with detected pollutants Water from Flevoland can be discharged in Wolderwijd by a nearby pumping station. Both locations may therefore be prone to contamination by pesticides from agricultural run-off, aerial deposition and discharges from sewage water treatment plants. Additional chemical analysis of pore waters prepared from five of the sediment samples yielded more identified substances such as 4-tert-butylphenol, phthalates, decanes, cosanes and fragrances, but TU analysis for these compounds with V. fischeri and D. magna showed that their contribution to toxic effects was negligible. Modelling of the body burdens did not seem to work for the higher alkane chains identified and for the phthalate DEHP because the estimations were unrealistically high. The alkane molecules may be too large to pass cell membranes and do therefore probably not accumulate in animal tissues. Phthalates are not persistent and can be biotransformed [216]. Molecular size and biotransformation are not incorporated in the model [217]. 7.5 Conclusions The results of this study show that sediment toxicity in heavily polluted (class 3 and 4) sediments could partially, but in most cases not fully, be explained by comparing chemical analysis results of the most common (priority) pollutants with available toxicity data on these substances for the standard test species used. Metals probably contributed most to the toxicity of the sediments and in some cases PAHs. Organochlorine pesticides, chlorobenzenes and PCBs seemed much less important. The present study seems to confirm an earlier study on sediment quality in the Rhine–Meuse delta in which an attempt to relate effects found in bioassays to sediment contaminant levels was made as well. In one study, only 29% of the variation could be explained by a set of 50 contaminants [218]. Peeters et al. [219] found that bioassay responses were correlated with pollutants in the sediment from the Rhine–Meuse delta in the Netherlands, but they also found significant correlations between bioassay results and other factors such as erosion at the sampling site and the type of sediment. There are many other possible reasons why the observed effects in the present study could not be fully explained. The possibility of combination toxicity mechanisms other than concentration-addition, for instance synergy, and an in-depth assessment of the bioavailability of compounds (besides normalisation of concentrations) have not been taken into account in the present work. Finally, it cannot be excluded that substances other than those routinely screened in Dutch sediments (metals, PAHs, OCPs and PCBs) or detected by GC–MS/AED, are partly responsible for toxic effects. It can also be argued, on the other hand, that pore water assays overestimate the toxicity of the more soluble compounds in sediments like metals and ammonia compared to the relatively less soluble organic compounds. The organic compounds may also sorb onto the walls of the testing containers thereby reducing the exposure of test organisms 111 7. Relation of ecotoxicity with detected pollutants and the true equilibrium concentrations that exist in in situ pore water do not necessarily occur in pore water that is removed from the sediment. Given the complex relation between sediment chemistry and observed toxicity, and the fact that reliable data sets on toxicity of individual compounds still remain incomplete, one may conclude that at present bioassays with environmental samples are still needed in addition to chemical analysis to assess the true toxicity of sediments. When routinely measured pollutants cannot sufficiently explain the toxicity of sediments by toxic unit analysis, approaches such as Toxicity Identification and Evaluation and bioassay-directed fractionation (e.g. refs. 67–72 in [214]) may be further used to identify the true cause of the observed effects. . 112 8 8. Comprehensive two-dimensional gas chromatography (GC×GC) measurements of volatile organic compounds in the atmosphere 8.1 Introduction Depending on their physical and chemical properties, volatile organic compounds (VOCs) play a wide variety of important roles in the atmosphere. Reactive hydrocarbons and their intermediate products have been recognised as precursors of tropospheric O3, organic acids, and organic aerosols [220-224]. If present at a high concentration in the boundary layer, O3 is toxic for humans and vegetation. Some VOCs themselves, especially those from anthropogenic sources, endanger human health directly [225]. Organic acids contribute to the acidification of precipitation, while some organic species may form aerosols which as potential cloud condensation nuclei can affect weather and climate [221;223]. Long-lived VOCs can be transported to remote areas, where they may influence chemical and physical properties of the remote atmosphere. Some VOCs, particularly halocarbons, have very long lifetimes, so that they may be transported into the stratosphere and act there as destroyers of stratospheric O3 [226]. Many halocarbons are strong greenhouse gases as well [227]. For the aforementioned reasons, the accurate measurement of volatile organic components in ambient air is an important aspect of atmospheric science. Gas chromatography (GC), in combination with flame ionisation detection (FID) or mass spectrometry (MS), has been used in many studies to measure atmospheric VOCs (e.g. [228-230]). However, conventional GC often fails to separate components in complex samples to a satisfactory degree, being limited by the separation power of a single column. Severe peak overlap in single column chromatography causes difficulties in identification and inaccuracy in quantification. Peak overlap has another consequence for air analysis. Due to the strong differences in abundance between components, a large number of components with relatively lower mixing ratios may be completely masked by the enhanced baseline, so that they are not visible on the conventional chromatograms [231]. These components can be very reactive ozone precursors, intermediate products Published as: X. Xu, L.L.P. van Stee, J. Williams, J. Beens, M. Adahchour, R.J.J. Vreuls, U.A.Th. Brinkman, J. Lelieveld, Atmos. Chem. Phys. Discuss. 3 (2003) 1139 8. GC×GC atmospheric analysis of photochemical reactions, tracers of specific processes and hence of interest for atmospheric chemists. Even though their individual abundances are low, quantification of these species may be of importance for understanding atmospheric processes such as ozone formation. Since its invention, the novel technique of comprehensive two-dimensional gas chromatography (GC×GC) has been developed to separate and analyse complex samples, such as, petroleum, flavours, and environmental samples. This technique employs two coupled columns of different selectivity and subjects the entire sample to a two-dimensional separation. Effluent from the first column is modulated to produce sharp chemical pulses, which are rapidly separated on the second column. A separation plane is produced by the two orthogonal retention time axes for both columns. Usually, the first column contains a nonpolar stationary phase, and the second column a polar stationary phase. This combination allows components to be independently separated, first according to their volatility, and then according to their polarity. In comparison to conventional single column gas chromatography, GC×GC has a much higher peak capacity, because the entire plane of a GC×GC chromatogram can be used for separation. Other advantages of GC×GC include enhanced sensitivity due to analyte refocusing, true background around resolved peaks, more reliable identification due to two retention times and due to well ordered bands of compound groups [162;232-237]. The key element in a GC×GC system is the modulator, which compresses segments of the effluent from the primary column and reinjects them onto the secondary column. Different types of modulator have been designed and shown to be capable of making GC×GC measurements. In their pioneer work, Liu and Phillips [18] and Phillips and Xu [11] used an on-column two-stage thermal modulator, which is heated by a resistive film painted onto the capillary surface and cooled by ambient air. This modulator is difficult to operate and has only a short lifetime. A similar thermal modulator using a wire, instead of the painted film, has higher durability, but sluggish thermal response [19]. A more robust thermal modulator using a rotating heated sweeper showed good performance; however, the operation temperature of the GC oven must be about 100°C lower than the maximum allowed temperature of the stationary phase in the modulation capillary [20]. Instead of using heating, Kinghorn and Marriott [21] developed a modulator using cooling. This modulator regularly traps and releases solutes from the first column by moving a cryogenic trap back and forth along the second column. While achieving good performance, the prototype of this modulator showed problems with ice buildup, which were overcome in modified designs [22;23]. Both the heated sweeper technique and the moving cryogenic trap technique have a common drawback, i.e. frequent breakage of capillaries by moving parts in the system. More recent development 114 8. GC×GC atmospheric analysis of the modulation technique is the jet-cooled modulator, which uses no moving parts. Adapted from the jet-cooled thermal modulator [24], a jet-cooled/heated modulator was reported by Ledford [238]. The modulator employs two cold and two hot nitrogen jets that are pulsed to alternately cool and heat two spots at the front end of the second column for focusing and remobilising analytes eluting from the first column. While this type of modulator allows excellent modulation of compounds even as volatile as methane, use of liquid nitrogen is limited by its availability and requires bulky facilities for storage and insulation. A practical solution has already been demonstrated by Beens et al. [25], who used a CO2-cooled jet modulator and obtained sharp (about 30 ms) second-dimension peaks. Apart from the thermal modulators, valve based modulators can also be used to make GC×GC measurements [239-241]. However, the valve-based modulators send only a part of the effluent from the first column to the second column because they use the so-called heartcutting technique. Therefore, use of the valve-based modulation technique is limited to relatively concentrated samples. The GC×GC technique is now being applied to a wide variety of complex sample measurements. The potential of GC×GC to ambient air measurements was first demonstrated by Lewis et al. [231]. It was shown that some 550 individual components could be separated in urban air. In this study we have applied GC×GC to the in situ measurements of VOCs at a more remote site, approximately days transport time from sources. This paper describes the instrumental setup as well as the identification and quantification techniques. An atmospheric chemistry analysis and interpretation of the quantitative data are given in a separate paper [242]. 8.2 Experimental 8.2.1 Site The in situ measurements were performed during the Mediterranean INtensive Oxidant Study (MINOS) project in August 2001 (summarised by Lelieveld et al. [243]). Atmospheric VOCs were observed at Finokalia, Crete, a groundbased station (35°19′ N, 25°40′ E; 130 m above sea level) established by the University of Crete. Crete is located roughly in the middle of the Eastern Mediterranean, about 400 to 1000 km away from the coasts of Greece and Turkey. The wind was steady and northerly throughout the campaign. The 7.4 m/s average wind speed corresponds to a transport time of 0.5–1 d from continental coastal sources to measurement point. 115 8. GC×GC atmospheric analysis 8.2.2 GC×GC system The measurement system used for the in situ observation is depicted schematically in Fig. 8.1. The whole system consists of a flow controller and a thermal desorber (both from Markes International, Pontyclun, UK), and a gas chromatograph (GC6890, Agilent, Wilmington, DE, USA), equipped with a flame ionisation detector (FID) and jet- modulated GC×GC parts (Zoex, Lincoln, NE, USA). The sampling and desorption control software (Markes International, Pontyclun, UK) and ChemStation (Agilent, Wilmington, DE, USA) installed on a personal computer controlled the sampling/thermal desorption system and the GC, respectively. For the GC×GC modulation, a homemade multipurpose device (V25) was used as a pulse generator, which can be easily synchronised with the GC. The main benefit of such synchronisation is that the second-dimension retention times do not drift randomly and consistent geometries can be achieved from run to run. Fig. 8.1: Schematic of the thermal desorber GC×GC-FID system. The left part shows an air server that contains a sampling manifold and a mass flow controller (MFC). A Nafion dryer can be used for removing moisture from ambient air. The middle part shows the thermal desorber in the trap desorption step. The arrows give flow directions of carrier gas (helium). During online sampling the carrier gas flows in the reversed direction. The solid and dotted lines show flow paths with and without gas flow, respectively. SV, NV, PT, and MFC represent solenoid valve, needle valve, pressure transducer, and mass flow controller, respectively. The right part shows the GC×GC system with its controlling units. In the real design, the hot-jet tubes are orthogonal to the cold-jet tubes. 116 8. GC×GC atmospheric analysis The thermal desorber can be set either to the online mode or to the 2-stage desorption mode. In the online mode ambient air is drawn through a link tube and collected directly onto the cold trap (quartz, 12 mm, 2 mm i.d.) of the thermal desorber and analysed immediately after the sampling (for more details, see Section 2.3). The cold trap contains two beds of sorbent, i.e. Tenax TA and Carbograph, supported by quartz wool. The sorbent beds of the cold trap are cooled by a 2-stage Peltier cell. A minimum of −10°C can be reached at an ambient temperature as high as 30°C. If set to the 2-stage desorption mode, the thermal desorber can transfer volatile compounds from a sample tube into the cold trap for focusing and subsequent injection. In this case, the link tube is replaced by the sample tube (for more details, see Section 8.2.4). A split flow is used during the desorption and the injection. The discharged flow is filtered using a charcoal filter of the same size as the sample tube. The front and rear ends of the sample tube (or the link tube in the case of online sampling), the cold trap, and the charcoal filter are connected through adapters to the solenoid valves and to the heated valve (200°C), respectively. All connections are sealed using Viton O-rings. PTFE filters are inserted into the adapters to prevent particles from being carried into the valves. Two needle valves and a mass flow controller are used to measure and control the desorption and split flows. Table 8.1: GC×GC parameters and conditions First column DB-5, 30 m × 0.25 mm i.d. × 1 µm df (5% phenyl)methylpolysiloxane) 50–200°C at 2.5°C/min Second column Carbowax, 1 m × 0.1 mm i.d. × 0.1 µm df (polyethylene glycol) 30–180°C at 2.5°C/min Modulator Jet-cooled and -heated Cold jet tubes i.d. 2.7 mm, axial distance between cold jets 7.8 mm Hot jet tubes i.d. 4.2 mm, axial distance between hot jets 8.2 mm Cold jet flows 10 l/min N2, hot jet flows 70 l/min N2. Modulation time 6 s, upstream pulse duration 0.3 s, downstream pulse duration 0.3 s, pulse delay 0.4 s Carrier gas Helium (99.9999%, filtered using water, hydrocarbon and oxygen traps), 276.8 kPa A DB-5 column (Agilent, Waldbronn, Germany) and a Carbowax column (Quadrex, Woodbridge, CT, USA) are used as the first- and the second-dimension columns, respec- tively. Detailed column parameters and operation conditions are listed in Table 8.1. Roughly 1 m of the first column is used as the transfer line from the thermal desorber to the GC. This transfer line is protected by a heated sleeve (PTFE tubing covered with silicone foam rubber insulation). The sleeve temperature is set to 200°C to ensure no retention of compounds of interest in the transfer line. A jet-cooled and heated modulator is used for the GC×GC modulation. The modulator consists of two cold jet tubes installed in an evacuated outer casing and two hot jet tubes. Both cold jet tubes lie parallel to each other as do the hot jet tubes, but the cold jet tubes are orthogonal to the hot jet tubes. Both cold jets and hot jets are nitrogen 117 8. GC×GC atmospheric analysis gas from a Dewar (120 l, max. 4 bar, Linde, Dortmund, Germany), with the cold jet gas being conductively cooled by passing it through copper tubing coiled in a cryogenic trap (liquid N2) and the hot jets being heated by a heater at the tube outlet. More details about the modulator are given in Table 8.1. No extra modulation tube is used. Modulation is performed on the second column, at a distance of about 10 cm from the connection with the first column. The middle segment (about 75 cm) of the second column is housed in a separate chamber, which can be heated and cooled independently. The end segments (about 25 cm) of the column are exposed to the air bath in the main oven of the GC. 8.2.3 In situ measurement During the in situ measurements the sampling system was set to its on-line mode. The on-line measurement includes five steps, i.e. leak test, link tube purging, on-line sampling, trap purging, and injection. During the leak test every part of the flow path of the thermal desorber is pressure tested, without heat or carrier gas flow. The pressure is measured by a pressure transducer. If the measured pressure drops more than 5% in 30 s, the leak test fails, and the other steps are not conducted. If the leak test is passed, the sampling system is purged using sample gas at a rate of 50 ml/min for 4 min, with no gas flowing through the cold trap. After the purging the ambient air (or helium in the case of the blank measurement) is sampled directly onto the cold trap. The sampling flow is usually set to 50 ml/min. A Nafion dryer connected to the air server can be used to continuously remove water vapour in the air sample stream. However, Nafion dryers can partially remove some VOCs as well, especially the oxygenated hydrocarbons. For this reason, the dryer was offline for most of the time during the campaign. To avoid trapping significant amounts of water, the trapping temperature of the cold trap was set to 10°C, which was adequate for the targeted C7−C14 compounds. The sampling took 60 or 80 min, corresponding to a sample volume of 3 or 4 l. After the sampling the cold trap was purged using helium for 4 min, and heated to 250°C in less than 5 s and held at this temperature for 5 min to inject the focused compounds. During the injection, the direction of the flow through the cold trap is reversed from that during sampling (see Fig. 8.1), so that heavier compounds on the Tenax bed (rear) have no contact with the stronger sorbent Carbograph (front). The flow from the cold trap was split, with 2.5 ml/min being directed to the GC and 5 ml/min to the charcoal filter. Chromatographic signals from the in situ measurements were detected by the FID attached to the GC. The data acquisition frequency was set to 100 Hz, which was high enough for the measurements, allowing 10 points per peak even for those with small peak widths. FIDs have proven to be robust detectors capable of high frequency data acquisition necessary for GC×GC measurements. Since the response factor of the FID 118 8. GC×GC atmospheric analysis depends mainly on the carbon number of compounds, the peaks of compounds for which no standard is available, can be calibrated relative to other compound of the same carbon number. This advantage becomes more significant when all isolated peaks should be quantified, as is intended in ambient air studies. For identification, cartridge samples were collected at Finokalia and analysed in the laboratory with a time-of-flight mass spectrometer (ToF-MS) as described in the following section. 8.2.4 Laboratory GC×GC–ToF-MS measurement In addition to the on-line measurements, several cartridge samples were collected using sample tubes with DiffLok caps (Markes International, Pontyclun, UK). The sample tubes (Silcosteel, 89 mm, 5 mm i.d.) are packed in sequence of increasing retention capacity with Tenax TA, Carbopack B, and Carboxen 1000, effectively trapping VOCs down to propane. Ambient air was drawn through the tubes at 100 ml/min for one or two hours using a calibrated air pump (FL-1001, CHEMATEC, Roskilde, Denmark). After sampling, the tubes were stored in an isolated box for 3 months before analysis The analysis of the cartridge samples was performed on the same thermal desorption and GC×GC system under the same conditions as in the online measurements. For the detection a ToF-MS (Pegasus II, Leco, St. Joseph, MI, USA) was used. A transfer line (ca. 20 cm) was used to connect the second column of the GC×GC system and the ToF- MS. The ToF-MS was controlled by another personal computer. Spectra with a mass range of m/z 35–300 were collected and stored at a rate of 100 Hz. A detailed description of this system is given elsewhere [166]. To analyse a cartridge sample, the sample tube was positioned in the desorption oven (i.e. the link tube position in Fig. 8.1). A leak test was done, without heating the desorption oven. After a successful leak test, the sample tube was purged using helium at a rate of 50 ml/min for 4 min to remove moisture and oxygen, with no carrier gas flowing through the cold trap. After the purging the sample tube was heated to 280°C at 20°C/s and held at this temperature for 5 min to desorb the organic compounds. During the desorption the cold trap was set to 10°C to focus compounds from the cartridge sample. The desorption and split flows were set to 10 ml/min and 5 ml/min, respectively. After the desorption the cold trap was purged and heated to inject the focused compounds as in online measurements. 8.3 Results and discussion 8.3.1 GC×GC chromatograms Fig. 8.2 shows three example GC×GC chromatograms. Fig. 8.2a and Fig. 8.2b are typical ambient air and blank chromatograms from the in situ measurements using 119 8. GC×GC atmospheric analysis GC×GC–FID. Fig. 8.2c is a total ion count (TIC) chromatogram from the GC×GC–ToF- MS analysis of a cartridge air sample collected at Finokalia, Crete, during the MINOS campaign. The dark spots with white boundary are the major peaks, while the red and white spots are the medium and small peaks, respectively. At first glance, there seem to be roughly two hundred peaks on the in situ air sample chromatogram (Fig. 8.2a), but the peak density is actually much higher. There are a lot of small peaks, especially in the lower bands, as can be seen in the insert of Fig. 8.2a. Including the small peaks, there are approximately 30 peaks in the small area of the chromatogram. The total number of peaks may be well above 500. Under the aforementioned conditions, the optimum separation range is between 10 min and 55 min, corresponding roughly to C7−C14 n-alkanes. In the first 10 min the separation is bad because of overloading and eluting temperatures that are too high for the very volatile species. Therefore, data from that retention range are not shown in the figure. The lower right-hand corner of the separation area is usually crowded with peaks of heavy compounds that are believed to originate from column bleeding or artifacts from other parts of the system. The long tailing peaks, the so-called “flying comets”, are also believed to be caused by system artifacts, e.g. degradation products of stationary phases and sorbents. They elute from the first column above certain temperatures and produce consecutive peaks of which the retention time in the second dimension shifts towards t=0 as the oven temperature rises. These artifact peaks affect the quantification of peaks overlapping on them and removal of these tailing peaks should be one of the tasks in future system improvements. 120 8. GC×GC atmospheric analysis Fig. 8.2: Example GC×GC chromatograms. (a) and (b) are FID chromatograms from the MINOS field measurements of an air sample and a blank sample, respectively. (c) is a TIC chromatogram from the GC×GC– ToF-MS analysis of a cartridge sample collected at Finokalia, Crete, during the MINOS campaign. In comparison with some GC×GC chromatograms in the literature (e.g. [244;245], the peaks on the chromatogram shown in Fig. 8.2 seems less ordered. This is not caused by the conditions used for the analysis, but rather a result of the different nature of an air sample compared with petrochemical samples. As will be discussed in Section 8.3.3, 121 8. GC×GC atmospheric analysis atmospheric VOCs cover many classes of compounds. Some of the compounds may come from anthropogenic emissions associated with fossil fuel use. Other airborne compounds are photoproducts, biogenic in origin, or products from industries unrelated to fuel processing. Hence a wider range of compounds can be expected in the air. The differences in polarity between different classes of compounds can be large or small, depending on degree of oxidation. The situation is further complicated by different degree of branching and by compounds with two or more functional groups. As a consequence, distinct bands tend to be masked by compounds that are scattered between the bands. Nevertheless, some bands are still visible, for example, the lowermost alkane band and the bands of aromatic hydrocarbons. These aid orientation considerably over 1D-GC. There are several peaks below the alkane band. These peaks are so-called “wrap- arounds”. Because of higher polarity, the retention times of some compounds are longer than the modulation period, so that they elute during the second or third period after being injected onto the second column. Wrap-arounds, if overlapping other peaks, may affect the quantification. They can be removed by using a longer modulation period. On the other hand, a longer modulation period leads to a lower resolution on the first column. In this work a modulation period of 6 s is used as a compromise. The chromatogram of the blank sample (Fig. 8.2b) contains much fewer peaks than that of the air sample. However, it is not as clean as expected. Besides the column bleeding and sorbent degradation, slight contaminations on the O-rings, filters, etc., of the thermal desorber, may have been responsible for the peaks in the blank chromatogram. After the MINOS campaign the thermal desorber was cleaned by the manufacturer. The blank level has been improved significantly since then. In spite of the issues mentioned above, the advantages of using GC×GC over conventional 1D-GC are obvious. Many compounds that would overlap on a 1D-GC chromatogram are well separated by GC×GC, as can be seen in Fig. 8.2a. This is especially important for the peaks of medium and small sizes since they otherwise would be masked by the enhanced baseline or major peaks of the 1D chromatogram. In addition, in the GC×GC measurement the artifact peaks, such as the long tailing peaks, influence only the peaks with which they coincide in both dimensions, not all peaks in the first dimension retention ranges that they cover, which would be the case in 1D-GC. The basic features of the TIC chromatogram (Fig. 8.2c) are similar to those of the FID chromatogram of the air sample (Fig. 8.2a), but there are some differences between both chromatograms. On the TIC chromatogram there seems to be a band 0.4 s above the lowermost band (i.e. the alkane band). This is a result of double modulation caused probably by a disturbance to the temperature of the downstream cold spot when the 122 8. GC×GC atmospheric analysis upstream hot jet is fired. It seems that the double modulation only occurs in the measurements of relatively concentrated samples. No double modulation was observed in the in situ measurements because the samples were less concentrated. Except for the double modulation problem, the TIC chromatogram also shows a higher noise level than the FID chromatogram. Another difference is that many small peaks of polar compounds are not visible on the TIC chromatogram. Possible reasons for that are the loss of these compounds during storage and/or low sensitivity of ToF-MS to these compounds 8.3.2 Identification Cartridge samples of volatile organics in ambient air were measured using the thermal desorption GC×GC–ToF-MS system. All conditions were as listed in Table 8.1, except for the carrier gas (helium) pressure, which was increased to 289.4 kPa to compensate the effect of the transfer line between the GC and the ToF-MS. Data from one of the cartridge samples were analysed using the ChromaTOF software from LECO for tentative initial identification. The sample (see the chromatogram in Fig. 8.2c) was collected at Finokalia on the 12th of August 2001 between 00:35 and 2:35 local time. Since a big biomass burning plume influenced the site in the period from the 8th to the 12th of August 2001, the sample was expected to contain more components than samples from other periods. The ChromaTOF software uses a deconvolution algorithm which mathematically separates partially coeluting peaks. The first step in the identification process was mass spectral library matching of the deconvoluted peaks against the NIST library (NIST ’98, National Institute of Standards and Technology, Gaithersburg, MD, USA). Because of less interference from the background and coeluting compounds the higher spectral purity of the deconvoluted peaks makes the library matching more reliable. Detailed examples of this method are described in [166;170]. To allow both narrow and broad peaks being recognised by the software, three data analyses were performed for expected peak widths of 100, 300 and 2000 ms. Results of the tentative identifications were reported in peak tables, containing compound name, formula, retention time, similarity, signal-to-noise ratio (S/N), etc. The retention times were corrected for the time difference between the start of the ToF-MS data acquisition and the start of the sample injection, and then converted to the first dimension and second dimension retention times based on the modulation period of 6 s. Starting with tens of thousands of peaks recognised in the data processing, several rules were applied to reduce this set of peaks. First, all peaks with a mass-spectral match (i.e. similarity) lower than 800 were discarded. Next, a selection based on the S/N was made. An important factor regarding the S/N is on which basis it is calculated; in mass spectrometry one can either choose a selected ion chromatogram or the total ion 123 8. GC×GC atmospheric analysis chromatogram. For the present processings the so-called “unique mass” was chosen to calculate the S/N. During the deconvolution process described shortly above, the algorithm looks for masses (m/z values) that distinguish the peak in hand from other coeluting compounds or background signals, and designates this as the “unique mass”. One has to realise that the S/N value based on this m/z does not say anything about the intensity in the total ion chromatogram or comparable FID chromatogram because the selected m/z has a variable relative intensity within the mass spectrum in hand. However, the advantage of the method is that many small peaks can be recognised which are coeluting with other compounds or are engulfed in the (chemical) noise of the total ion chromatogram. Arbitrarily, a minimal S/N of 100 for the unique mass was chosen to reduce the set of peaks further to about 650 peaks which were subjected to additional confirmation using retention indices. Linear retention indices on the first column (DB-5) were calculated using tr( x) tr(n) RI 100 n x tr(n1) tr(n) where RIx is the retention index of component x; n is the carbon number of the last n- alkane eluting before component x; tr(x) , tr(n), and tr(n+1) are the retention times of component x, the preceding n-alkane with carbon number n, and the next n-alkane with carbon number n+1, respectively. The calculation was done for components eluting between pentane and tetradecane. Because the solutes from the first column are refocused in the GC×GC measurement, it is possible that the first-dimension retention times, i.e. the peak apices of the solutes, are shifted back or forth, relative to the retention times from a single-column separation. However, the shift should be less than one modulation period, i.e. 6 s, corresponding to an average error of 3.2 index unit. To verify the tentative identifications, the measured RI values were compared with literature RI values determined on appropriately similar stationary phases. RI values of some compounds are not included in the available literature. Therefore, the model of Zenkevich [246] was used to predict RIs from the boiling points and taxonomic parameters of the compounds. Since the contemporary level of interlaboratory reproducibility of experimental RI determination is about 10 units [246], it seems reasonable to allow a disagreement of 20 units. between RIs from this work and those from the literature or from the prediction. If the index disagreement exceeded 20 units, the compound was considered not confirmed and therefore discarded from the table. In some cases, mainly arising from the three different data processing results put together, several peaks were recognised as the same compound which all complied with the rules. 124 8. GC×GC atmospheric analysis In these cases, the identification with the highest S/N was chosen; if the S/N made no clear distinction, the peak with the highest spectral similarity was selected. Table 8.3 lists components that have been tentatively identified by the software and confirmed by the RI comparison. The compounds are classified as acyclic alkanes, cyclic alkanes, acyclic alkenes, cyclic alkenes, aromatic hydrocarbons, oxygenated aromatics, alcohols, aldehydes, ketones, esters, nitriles, halogenated hydrocarbons, and miscellaneous. In total 235 compounds have been confirmed. More than half of the confirmed compounds are hydrocarbons, with alkanes, alkenes, and aromatic hydrocarbons contributing 31%, 10%, and 15%, respectively. Nearly one-third of the compounds are oxygenated species, including alcohols, aldehydes, ketones, esters, and oxygenated aromatic compounds. Other compounds, such as nitriles, halogenated hydrocarbons, and some miscellaneous species, make only a small contribution to the total number of the confirmed compounds. Retention times in both dimensions are given. Not all confirmed compounds in Table 8.3 show up on the GC×GC chromatograms from the field measurements. The main reasons for that are: (1) the identification has been based on the GC×GC–ToF-MS measurement of a cartridge sample that is more concentrated than the online samples for the field measurements; (2) the sensitivity of ToF-MS is different from that of FID; (3) some compounds may have entered the sample tube during storage. Of the 235 confirmed compounds, 150 show up in the optimised separation range (C7–C14) on the chromatograms from the in situ measurements, suggesting that they were present in the atmospheric boundary layer at Finokalia during the MINOS campaign. While most of the peaks are well separated, there are a few overlaps, even with GC×GC separation. The overlaps are mainly caused by structural isomers (e.g. p-xylene / m-xylene), and in a few cases, also by quite different compounds having similar retention times on both columns (e.g. 1,2,4-trimethylbenzene/octanal). Some of the overlapping peaks may be resolved in the future by using different column combinations. Although the identification based on mass spectrometry together with RIs results in a high level of confidence of correct identification, one should realise that in some cases there is no complete certainty. If several isomers of a compound exist that elute closely together and produce mutually similar spectra, there is a chance that the compound is identified as another isomer, and that the isomer is (erroneously) confirmed because of the similar RI values of the various isomers. In the results presented in Table 8.3 a typical class in which this might occur are the branched alkanes. Although the GC×GC–ToF-MS measurement has led to the successful identification of 150 peaks (cf. above), the identification process is still being continued because about 500 peaks show up in the GC×GC–FID chromatogram with S/N>10. Some of these 125 8. GC×GC atmospheric analysis unknown peaks which are in the list of unconfirmed initial identifications, obviously are the best candidates for a continued search. The presence in the atmosphere of the organic compounds listed in Table 8.3 is not surprising. In fact, most of the compounds have already been detected in previous observations at urban and remote sites mainly using GC–MS (e.g. [228;247;248]). Due to direct emissions of organic species from anthropogenic and biogenic sources and photochemical production of secondary organic species, the lower atmosphere always contains a large number of VOCs, even in the Antarctic [249]. Given the numerousness of atmospheric VOCs, nonselective detection, such as FID, is preferred for the simultaneous quantification of the VOCs. GC×GC–FID, which combines the high separation power of GC×GC with the universal nature and robustness of FID, is very suitable for the routine measurements of atmospheric VOCs. 8.3.3 Quantification As in conventional GC, determination of peak sizes is necessary for quantifying the analytes of interest. Integration and chemometric analysis are the two commonly used methods to quantify the sizes of GC×GC peaks. The chemometric method utilises the multivariate techniques, such as the generalised rank annihilation method (GRAM) (see e.g. [240]). The accuracy, precision, and lower detection limit in the quantification can be improved using the GRAM method. Another advantage of using this method is the quantification of partially overlapped peaks. In spite of these advantages, the GRAM method is not used in this work, because the requirements for its successful use were not fully met. Under the conditions used in this work, the data density for the first column is lower than 4 points per peak for most peaks. Some peaks only cover 1 or 2 separation periods. Such small peaks cannot be reliably quantified by GRAM though it was able to analyse peaks with a data density down to 3 points per peak by appropriate interpolation and retention time correction [250]. In addition, some peaks do not show near symmetrical ellipse boundaries, suggesting that the data matrix does not fit a bilinear structure, on which the GRAM analysis relies. Instead of chemometric analysis, integration and subsequent calibration have been done for some well resolved n-alkane and aromatic compounds, which are contained in the standard mixture used in this study. The integration of GC×GC peaks was done using integration software called Blob from the Zoex Corporation. Prior to the integration, chromatogram data (retention time and FID signal) collected by the ChemStation software are saved in files of comma separated values (CSV) format. The CSV files are then read by the Blob software. 2D chromatograms with colour coded peaks are created automatically, based on the given numbers of the modulation periods and the data points of each period. Background signals can be subtracted from the chromatograms using the corresponding command. 126 8. GC×GC atmospheric analysis Peak (or blob) volumes are calculated, after the user draws polygons around the peaks of interest and types in peak names. Integration reports, containing peak names, apex positions, peak height, peak volumes, etc., can be created for samples chosen by the user. The software often finds two or more peaks within a polygon. In this case it integrates all peaks in the polygon and reports data of all peaks. Among the peaks within a polygon, there is usually a major one. The volume of the major peak is usually more than one order of magnitude larger than those of the minor ones. Therefore, the major peaks found in different polygons are considered to be the peaks of interest. The volumes of the major peaks are used for the concentration calculations. While the software can integrate well-resolved medium or large peaks, it is not able to integrate some small peaks. The colour tables available in the software cannot make the small peaks visible on the chromatogram, even if the colour scale is reduced. In addition, the software cannot automatically integrate all GC×GC peaks that it finds on a chromatogram, but relies on polygons drawn by the user. If more than several tens of peaks need to be integrated simultaneously, drawing polygons is not only time consuming and laborious, but it can lead to loss of overview and hence wrong positioning of polygons. These are major issues that need to be addressed in the future development of the integration software. External calibrations were made to obtain masses of individual components in the samples. A standard gas mixture (Apel-Riemer Environmental, Denver, CO, USA) was used for the calibrations. The standard contains 74 C2−C11 hydrocarbons in nitrogen, with mixing ratios ranging from 0.14 to 12.35 ppbv. Multipoint measurements of standard were made in the laboratory using the same sampling and analysis methods as in the field. Fig. 8.3 shows an example of calibration curves. The peak volume of toluene, as integrated using the Blob software, shows a linear dependence on the mass of toluene. The regression coefficient is close to unity, suggesting that the linearity of the relationship between the peak volume and the mass is very good. This was also found to be true for the other calibrated components (R2 from 0.9832 to 0.9998). The good linear correlation between the peak volume and the mass of analytes suggests, that while providing strong separation power, GC×GC is also a competent technique for quantitative measurements, as shown by Beens et al. [251]. 127 8. GC×GC atmospheric analysis Fig. 8.3: Calibration curve for toluene. The mass of toluene was calculated from the sampled volume of the standard mixture and the mixing ratio of toluene in the mixture. The peak volume values were obtained using the integration software. During the MINOS campaign calibrations were made approximately once every five days. Only two-point calibrations were made, so as not to detract from the measurement frequency of atmospheric VOCs. The mixing ratio of any analyte in the ambient air sample x was calculated as: ( x b )Vs Cx Cs ( s b )Vx where Cx and Cs are the mixing ratio of the analyte in the air sample and the standard sample, respectively; γx, γs, and γb are peak volumes of the analyte corresponding to the air sample, the standard sample, and the blank sample, respectively; Vx and Vs are the volumes of the air sample and the standard sample, respectively. Blank levels were observed approximately once every three days, by simulating the air sampling using helium, i.e. passing helium through the air sample pathway in the air server and in the thermal desorber (see Fig. 8.1), and analyzing VOCs focused in the cold trap. The accuracy of the measurements depends on the systematic errors of the peak integration, the sample volumes, and the standard. The integration error depends on the peak size and its relative contribution is small for the middle and large peaks. The error in the volume determination is about 1%. The error of the standard concentrations is 2%. Therefore, the accuracy is estimated to be about 5% for the already quantified compounds. The precisions for 20 quantified hydrocarbons range from 5% to 28%, as estimated from the relative (1σ) standard deviations of the compounds in the field calibrations. The detection limit of the GC×GC method is theoretically much better than that of the 1D-GC method since the modulation makes the peaks sharper and the baseline cleaner. In this study the enhancement of the peak height by the GC×GC modulation is estimated to be 20–60 times, based on the peak width for most isolated peaks (0.1–0.3 s) and the modulation period (6 s). Even if the baseline were as noisy as that of the 1D-GC, 128 8. GC×GC atmospheric analysis the detection limit would have been improved 20–60 fold. However, for some compounds this improvement was not practically reached during the MINOS campaign, because the blank levels and their variations were relatively high. The situation was significantly improved during the second half of the campaign after changing the cold trap and one of the filters of the thermal desorber. Therefore, the detection limit is estimated separately for the first and second half of the campaign, based on the (2σ) standard deviation of the blank values. The detection limit was between 0.2 and 35 pptv in the first half of the campaign and between 0.2 and 12 pptv in the second half of the campaign. The detection limit can be further improved by obtaining a cleaner sampling and injection system. As an example of quantitative results, Table 8.2 lists the mixing ratios and LODs of some compounds found in an online measurement during the MINOS campaign. The peak positions of these compounds are marked on the chromatogram shown in Fig. 8.2a. Toluene, ethylbenzene, and xylenes are important aromatic compounds. These compounds are released into the atmosphere mainly through the use of gasoline and solutions containing them. Biomass burning also emits certain amounts of these compounds. In polluted areas the mixing ratios of these compounds are usually at the ppb level, while in remote areas they decrease significantly, due to the dilution and photochemical degradation during the transport [229;252]. The mixing ratios of the aromatic hydrocarbons listed in Table 8.2 coincides with the remoteness of the Finokalia site. On the other hand, they also suggest that the anthropogenic impact on the air chemistry at the site may still be important, considering the high reactivity of the compounds. More detailed analysis and interpretation of the hydrocarbon data are presented in Xu et al. [242]. Table 8.2: Mixing ratios and LODs (pptv) of selected compounds in an air sample measured during the MINOS campaign. The chromatogram of the sample is shown in Fig. 8.2a. Peak Compound Mixing ratio LODa No. 1 Toluene 130 15 2 Ethylbenzene 22 9 3 p/m-Xylene (co-elution) 33 17 4 o-Xylene 48 21 5 Benzonitrileb 6 a. Estimated for the first half of the campaign b. Calibrated using toluene standards and response factors from Katritzky et al. [253]. One of the interesting results of this study is the complete separation of benzonitrile from other compounds. If conventional capillary GC had been used, benzonitrile would not have been detected, because it would have been completely masked by a column bleed compound with the same first-dimension retention time as that of benzonitrile. The peak of this interfering compound, eluting just before decane in the lowest band in Fig. 8.2a, 129 8. GC×GC atmospheric analysis is usually two orders of magnitude higher than that of benzonitrile. The ToF-MS software has identified the compound as octamethylcyclotetrasiloxane, but it has not yet been confirmed. Benzonitrile is used as a solvent and chemical intermediate in the pharmaceutical, dyestuffs and rubber industries (US National Toxicology Program, http://ntpserver.niehs.nih.gov). Except for the industrial sources, biomass burning also emits benzonitrile [254;255]. Reaction with OH radical is probably the main sink for atmospheric benzonitrile. The OH lifetime of benzonitrile is about 10 days. Less is known about the atmospheric budget of this compound. Since no benzonitrile standard was available for the present work, benzonitrile peaks were indirectly calibrated using toluene standards and the FID response factors for toluene (1.17) and benzonitrile (0.91) [253]. The mixing ratio of benzonitrile was lower than 5 pptv for most of the time during the campaign, but increased significantly during the biomass burning events. A comparison of benzonitrile data with acetonitrile data from the proton-transfer-reaction mass spectrometry (PTR-MS) measurements [256] shows a positive correlation (r=0.3, n=81) between benzonitrile and acetonitrile, a marker of biomass burning. More studies are necessary to estimate the source and sink strengths of benzonitrile and to know the role and usefulness of this compound in atmospheric chemistry. 8.4 Conclusions The novel GC×GC technique provides very high peak capacity and enhanced sensitivity, hence is an ideal tool for the simultaneous measurements of atmospheric VOCs. During the MINOS campaign the technique was successfully applied to the in situ measurement of atmospheric VOCs at the Finokalia ground station. GC×GC chromatograms from the measurements show hundreds of peaks, suggesting that even at the remote site ambient air is a very complex mixture which cannot be separated to a satisfactory degree by conventional GC. Indeed VOC concentrations determined by 1D-GC methods from highly complex samples such as biomass burning must be viewed with caution [257] because of the number of potential interferences seen here in ambient air. Multidimensional separation techniques, such as GC×GC, appear inevitable if atmospheric VOCs should be simultaneously measured to a detailed extent. A three-dimensional system coupling a GC×GC system with a ToF-MS was used for the identification of compounds in the air samples collected at Finokalia. About 650 identified two-dimensional peaks show significant S/N ratios (>100) and high spectra similarities (>800). So far, 235 of the identifications have been confirmed by an independent identification method, i.e. the retention index comparison. Of the 235 confirmed compounds, 150 show up in the C7−C14 range on the chromatogram from the in situ measurement. However, at least as many peaks are still unknown. To identify these unknown peaks is one of the future tasks. 130 8. GC×GC atmospheric analysis Quantification of GC×GC measurements is rather simple once the peak volumes are reliably integrated. For effective integration of hundreds of 2D peaks the integration software has to be improved significantly. The accuracy and precision of the GC×GC– FID measurements in this work is comparable to conventional GC–FID measurements. Because of higher blank values the detection limit for some hydrocarbons did not show significant improvement over conventional GC, although the sensitivity of the GC×GC system is about 20–60 times higher than that of conventional GC. A very clean sampling and injection system is required to really achieve the low detection limits that the GC×GC technique can provide. Table 8.3: Compounds tentatively identified by ToF-MS and confirmed by RI comparison 1 2 Compound tR tR Simi- Obsd. Lit. RI Lit. Found in field (min) (s) larity RI ref. measurements Acyclic alkanes Pentane 4.5 4.31 929 500.0 2,2-Dimethylbutane 5.0 4.67 871 526.3 528.5 [258] 2,3-Dimethylbutane 5.6 1.03 890 557.9 558.7 [259] 3-Methylpentane 6.0 1.04 903 579.0 578.6 [258] Hexane 6.4 1.04 943 600.0 2,2-Dimethylpentane 7.0 1.53 827 618.2 620.5 [258] 2,4-Dimethylpentane 7.1 1.05 884 621.2 625.8 [258] 2,2,3-Trimethylbutane 7.4 1.05 933 630.3 631.4 [258] 3,3-Dimethylpentane 8.1 0.66 899 651.5 650.5 [258] 2-Methylhexane 8.5 0.72 913 663.6 662.9 [258] 3-Methylhexane 8.8 0.76 930 672.7 672.2 [258] 2,3-Dimethylpentane 8.9 1.08 862 675.8 665.0 [258] Heptane 9.7 0.73 910 700.0 y 2,4-Dimethylhexane 11.1 1.11 880 729.2 731.7 [258] y 2,3-Dimethylhexane 12.5 1.13 935 758.3 757.9 [258] y 3-Ethyl-2-methylpentane 12.6 1.15 859 760.4 759.7 [258] y 2-Methylheptane 12.7 1.14 914 762.5 764.1 [258] y 4-Methylheptane 12.8 1.14 944 764.6 765.6 [258] y 3-Methylheptane 13.1 1.14 931 770.8 772.1 [258] y 3-Ethylhexane 13.2 1.14 934 772.9 775.0 [258] y Octane 14.5 1.18 849 800.0 y 2,4-Dimethylheptane 15.9 1.18 934 822.2 823.1 [258] y 4-Methyloctane 18.5 0.78 951 863.5 863.7 [258] y 3-Methyloctane 19.0 0.80 916 871.4 871.4 [258] y 2,4,6-Trimethylheptane 19.2 0.80 837 874.6 874.5 [259] Nonane 20.8 0.81 942 900.0 y 2,6-Dimethyloctane 23.0 0.85 827 933.3 921.6 [258] y 3-Ethyl-2-methylheptane 23.6 0.85 839 942.4 951.2 [258] y 4-Ethyloctane 24.5 0.86 832 956.1 965.0 [258] 4-Methylnonane 24.9 0.85 880 962.1 961.6 [258] y 3-Methylnonane 25.5 0.87 884 971.2 970.5 [258] y 2,2,4,6,6-Pentamethylheptane 27.2 0.87 921 997.0 990.2 [259] y Decane 27.4 0.88 917 1000.0 y Undecane 34.1 0.95 887 1100.0 y 2-Methylundecane 38.2 1.00 825 1165.1 1164.5 [259] y Dodecane 40.4 1.02 955 1200.0 y Tridecane 46.5 1.08 950 1300.0 y 3-Methyltridecane 50.6 1.11 852 1371.9 1371.1 [259] y Tetradecane 52.2 1.14 960 1400.0 y 131 8. GC×GC atmospheric analysis 1 2 Compound tR tR Simi- Obsd. Lit. RI Lit. Found in field (min) (s) larity RI ref. measurements Cyclic alkanes trans-1,2-Dimethylcyclopropane 4.8 3.55 879 515.8 503.0 [246] cis-1,2-Dimethylcyclopropane 4.8 4.67 828 515.8 523.6 [246] Cyclopentane 5.7 2.79 846 563.2 566.5 [260] Ethylcyclobutane 7.3 0.64 870 627.3 628.4 [246] Methylcyclopentane 7.3 1.05 878 627.3 625.6 [260] trans-1,3-Dimethylcyclopentane 9.2 3.17 826 684.8 687.9 [261] cis-1,3-Dimethylcyclopentane 9.3 1.09 914 687.9 691.5 [261] y Isopropylcyclobutane 9.4 1.10 934 690.9 696.4 [246] y trans-1,2-Dimethylcyclopentane 9.4 4.54 822 690.9 693.3 [261] y cis-1,2-Dimethylcyclopentane 10.8 1.11 898 722.9 717.9 [259] y Methylcyclohexane 10.9 1.12 944 725.0 718.1 [259] y Ethylcyclopentane 11.3 1.12 905 733.3 733.7 [246] y Norbornane 12.4 0.75 839 756.3 754.3 [246] y 1,2,4-Trimethylcyclopentane 13.4 1.16 848 777.1 779.2 [261] y cis-1,3-Dimethylcyclohexane 13.7 1.16 914 783.3 782.2 [261] y trans-1,4-Dimethylcyclohexane 13.8 0.75 884 785.4 783.9 [261] cis-1-Ethyl-2-methylcyclopentane 14.3 0.75 854 795.8 814.9 [246] trans-1-Ethyl-3-methylcyclopentane 14.3 1.17 896 795.8 789.4 [246] y trans-1,2-Dimethylcyclohexane 14.8 1.19 908 804.8 799.1 [246] y 1,4-Dimethylcyclohexane 14.9 1.18 815 806.3 796.6 [246] (1α, 2α, 3β)-1, 2, 3-Trimethylcyclo- 15.1 1.17 810 809.5 795.1 [246] pentane cis-1-Ethyl-2-methylcyclopentane 16.2 1.19 813 827.0 814.9 [246] cis-1,2-Dimethylcyclohexane 16.7 1.21 894 834.9 821.7 [246] Ethylcyclohexane 16.9 1.20 938 838.1 829.0 [246] Propylcyclopentane 16.8 1.20 856 836.5 825.7 [246] 1,1,4-Trimethylcyclohexane 17.2 1.21 824 842.9 834.8 [246] 1,2,4-Trimethylcyclohexane 17.3 0.79 869 844.4 859.6 [246] 1,2,3-Trimethylcyclohexane 19.7 0.81 829 882.5 872.1 [259] cis-1-Ethyl-3-methylcyclohexane 20.4 0.82 910 893.7 885.6 [246] y cis-1-Ethyl-4-methylcyclohexane 20.6 0.82 896 896.8 899.8 [246] y cis-1-Ethyl-2-methylcyclohexane 21.8 0.85 878 915.2 912.2 [246] y Propylcyclohexane 23.2 0.85 840 936.4 929.2 [259] y cis-Decahydronaphthalene 32.2 0.98 875 1071.6 1068.9 [246] y Acyclic alkenes (Z)-1,3-Pentadiene 4.7 1.03 869 510.5 529.8 [246] 2-Methyl-2-butene 4.8 4.27 932 515.8 523.0 [262] 1-Hexene 6.1 1.03 833 584.2 592.1 [262] 2-Methyl-1-pentene 6.1 2.79 836 584.2 591.0 [262] (E,Z)-2,4-Hexadiene 7.6 0.69 845 636.4 651.1 [262] 5-Methyl-1-hexene 7.7 1.06 811 639.4 659.3 [262] 1-Heptene 9.4 0.17 880 690.9 691.9 [262] 3-Heptene 10.1 1.12 823 708.3 703.9 [262] 3-Methyl-1-heptene 12.0 1.12 837 747.9 757.2 [246] 3-Methyleneheptane 13.4 1.16 838 777.1 787.1 [246] 2-Methyl-1-heptene 13.9 0.75 858 787.5 783.9 [262] 1-Octene 14.0 1.18 924 789.6 790.1 [228] (E)-2-Octene 14.4 1.17 827 797.9 807.2 [262] (E)-3-Octene 15.4 1.19 850 814.3 801.5 [262] trans-4-Decene 27.0 0.87 841 993.9 991.4 [261] Diisoamylene 28.0 0.88 830 1009.0 998.2 [246] 1-Tetradecene 51.7 1.14 920 1391.2 1392.3 [228] y 132 8. GC×GC atmospheric analysis 1 2 Compound tR tR Simi- Obsd. Lit. RI Lit. Found in field (min) (s) larity RI ref. measurements Cyclic alkenes Methylenecyclohexane 11.7 1.15 801 741.7 754.1 [263] 1-Methylcyclohexene 13.2 1.18 862 772.9 771.0 [262] 4-Ethenylcyclohexene 16.9 1.29 817 838.1 833.0 [263] 5-Ethylidene-bicyclo[2.2.1]hept-2-ene 22.2 0.94 940 921.2 922.6 [263] y α-Pinene 23.5 0.88 952 940.9 941.0 [264] y Camphene 24.7 0.92 822 959.1 953.0 [264] y 3-Carene 28.7 0.96 881 1019.4 1034.0 [265] Limonene 29.9 1.01 924 1037.3 1039.1 [228] y Aromatic hydrocarbons Benzene 8.4 1.28 981 660.6 660.1 [228] Toluene 13.1 0.96 946 770.8 767.6 [228] y Ethylbenzene 18.8 1.03 968 868.3 861.5 [228] y 1,3/4-Dimethylbenzene 19.4 1.05 975 877.8 869.3 [228] y Phenylethyne 19.6 1.92 810 881.0 875.9 [228] Styrene 20.5 1.32 953 895.2 892.9 [228] y 1,2-Dimethylbenzene 20.7 1.12 959 898.4 895.6 [228] y (1-Methylethyl)-benzene 22.8 1.05 971 930.3 919.0 [266] y 2-Propenylbenzene 24.2 1.21 927 951.5 954.2 [246] y Propylbenzene 24.7 1.09 976 959.1 949.0 [266] y 1-Ethyl-3-methylbenzene 25.3 1.10 960 968.2 962.6 [259] y 1-Ethyl-4-methylbenzene 25.4 1.11 967 969.7 958.0 [266] y 1,3,5-Trimethylbenzene 25.7 1.13 961 974.2 963.0 [266] y 1-Ethyl-2-methylbenzene 26.6 1.15 971 987.9 975.0 [266] y α-Methylstyrene 26.6 1.31 893 987.9 980.0 [228] y 1,2,4-Trimethylbenzene 27.4 1.17 960 1000.0 993.4 [259] y (2-Methylpropyl)-benzene 28.4 1.08 915 1014.9 1007.9 [259] y 1-Methyl-3-(1-methylethyl)-benzene 29.2 1.11 950 1026.9 1023.1 [259] y 1-Methyl-2-(1-methylethyl)-benzene 29.3 1.11 964 1028.4 1034.0 [266] y 1,2,3-Trimethylbenzene 29.6 1.24 957 1032.8 1023.1 [259] y Indane 30.6 1.29 949 1047.8 1036.1 [267] y 1,3-Diethylbenzene 31.1 1.14 955 1055.2 1052.6 [246] y 1,2-Diethylbenzene 31.2 1.14 917 1056.7 1060.9 [246] y 1-Methyl-3-propylbenzene 31.3 1.13 959 1058.2 1052.6 [246] y 1,4-Diethylbenzene 32.1 1.17 929 1070.1 1051.0 [266] y 1-Ethyl-2,4-dimethylbenzene 33.2 1.20 968 1086.6 1075.0 [266] y 4-Ethyl-1,2-dimethylbenzene 33.6 1.21 971 1092.5 1086.7 [246] y 1-Ethyl-2,3-dimethylbenzene 33.7 1.22 968 1094.0 1106.9 [246] y 2-Ethyl-1,3-dimethylbenzene 34.1 1.24 849 1100.0 1090.5 [246] y 1,2,4,5-Tetramethylbenzene 36.0 1.30 953 1130.2 1119.6 [246] y 1,3-Diethyl-5-methylbenzene 37.3 1.20 863 1150.8 1142.6 [246] 1,2,3,4-Tetrahydronaphthalene 39.1 1.42 904 1179.4 1166.3 [259] Naphthalene 40.6 2.06 975 1203.2 1186.9 [267] y 2-Methylnaphthalene 47.5 1.99 945 1317.5 1310.0 [266] y Oxygenated aromatics Benzaldehyde 25.1 2.48 974 965.2 960.2 [228] y Phenol 26.0 3.10 956 978.8 979.9 [228] y Benzoic acid methyl ester 34.1 2.04 943 1100.0 1101.9 [267] y Benzofuran 27.8 2,00 933 1006.0 995.9 [267] y Benzeneacetaldehyde 30.7 2.58 956 1049.3 1053.0 [265] y Acetophenone 32.2 2.45 977 1071.6 1068.4 [228] y 4-Methylbenzaldehyde 32.5 2.25 918 1076.1 1085.9 [228] y Acetic acid phenylmethyl ester 38.4 2.15 838 1168.3 1164.0 [265] α,α-Dimethylbenzenemethanol 33.4 3.16 926 1089.6 1089.1 [246] y 1-(3-Methylphenyl)-ethanone 39.3 2.25 930 1182.5 1175.2 [246] 1-(4-Ethylphenyl)-ethanone 44.9 2.06 854 1273.8 1282.2 [246] α-Oxobenzeneacetic acid methyl ester 48.5 4.27 864 1335.1 1325.3 [246] 133 8. GC×GC atmospheric analysis 1 2 Compound tR tR Simi- Obsd. Lit. RI Lit. Found in field (min) (s) larity RI ref. measurements Alcohols Isopropyl alcohol 4.8 0.88 945 515.8 522.2 [246] 2-Methyl-2-propanol 5.0 0.82 929 526.3 527.8 [246] 2-Butanol 6.5 1.09 878 603.0 589.5 [246] 2-Methyl-1-propanol 7.1 1.40 885 621.2 610.0 [266] y 1-Butanol 8.2 1.59 955 654.5 650.1 [260] y 1-Pentanol 12.6 1.92 921 760.4 764.0 [265] y (S)-2,5-Dimethyl-2-hexanol 18.1 1.21 836 857.1 859.2 [246] y 1-Hexanol 18.6 1.98 863 865.1 860.1 [267] y (E)-2-Hexen-1-ol 19.9 2.15 859 885.7 887.0 [265] (Z)-2-Hexen-1-ol 19.9 2.15 859 885.7 871.8 [246] 3-Heptanol 20.3 1.48 837 892.1 883.4 [268] y 1-Heptanol 25.2 1.93 879 966.7 966.7 [267] y 1-Octen-3-ol 25.9 1.82 825 977.3 982.0 [246] 1-Octanol 32.0 1.84 910 1068.7 1067.9 [267] y 1-Nonanol 38.6 1.76 805 1171.4 1169.2 [267] y 1-Decanol 44.9 1.71 855 1273.8 1269.9 [267] y 1-Undecanol 50.7 1.69 916 1373.7 1371.1 [267] y Aldehydes 2-Methyl-2-propenal 5.8 0.79 936 568.4 575.8 [246] Butanal 6.2 0.78 956 589.5 573.6 [228] Pentanal 9.4 1.34 863 690.9 696.0 [228] y 2-Methylpentanal 11.9 0.91 821 745.8 747.8 [246] y 2-Methyl-2-pentenal 12.6 1.01 838 760.4 760.8 [246] y 3-Methyl-2-butenal 13.7 1.53 801 783.3 800.5 [246] y Hexanal 14.5 1.02 922 800.0 799.4 [228] y (E)-2-Hexenal 17.7 1.37 901 850.8 854.0 [265] y 2-Ethylhexanal 24.4 1.07 911 954.5 963.1 [246] y (Z)-2-Heptenal 24.5 1.43 855 956.1 957.0 [265] y Octanal 27.6 1.17 932 1003.0 1003.8 [269] y (E)-2-Octenal 31.3 1.44 856 1058.2 1060.0 [265] y Nonanal 34.3 1.24 956 1103.2 1103.3 [228] y (E)-2-Nonenal 38.4 1.29 801 1168.3 1167.5 [246] Decanal 40.8 1.28 948 1206.6 1207.0 [228] y Undecanal 47.0 1.33 967 1308.8 1309.7 [228] y Ketones 2-Butanone 6.3 0.81 933 594.7 600.5 [228] 2-Pentanone 9.1 0.91 902 681.8 689.0 [270] y Methyl isobutyl ketone 11.3 1.37 958 733.3 722.1 [268] y 2-Methyl-3-pentanone 12.0 0.91 893 747.9 742.0 [246] y 3-Methyl-2-pentanone 12.1 0.93 920 750.0 749.6 [246] y 3-Hexanone 13.7 0.97 926 783.3 767.6 [268] y 2-Hexanone 13.9 1.04 941 787.5 798.0 [270] y Cyclopentanone 14.2 1.36 965 793.8 788.9 [246] y 2-Cyclopenten-1-one 16.7 2.37 803 834.9 822.4 [246] y 3-Ethyl-2-pentanone 16.9 1.01 907 838.1 838.8 [246] 2-Methylcyclopentanone 17.2 1.25 897 842.9 836.0 [270] y 3-Methyl-2-hexanone 17.3 1.04 908 844.4 842.9 [246] y 4-Methyl-2-hexanone 17.4 1.05 850 846.0 847.0 [246] y 3-Methylcyclopentanone 17.5 1.32 874 847.6 855.8 [246] (R)-(+)-3-Methylcyclopentanone 17.5 1.32 832 847.6 849.7 [246] 3-Heptanone 19.8 1.08 974 884.1 869.2 [268] y 2-Heptanone 20.0 1.14 953 887.3 890.0 [270] y Cyclohexanone 20.6 1.46 974 896.8 890.8 [267] y 2-Methyl-2-cyclopenten-1-one 21.3 1.86 854 907.6 926.4 [246] 2,2,4,4-Tetramethyl-3-pentanone 21.9 0.95 854 916.7 910.3 [246] 1-Cyclopentylethanone 23.0 1.29 933 933.3 919.2 [246] y 4-Methyl-2-heptanone 23.2 1.12 958 936.4 920.7 [246] y 134 8. GC×GC atmospheric analysis 1 2 Compound tR tR Simi- Obsd. Lit. RI Lit. Found in field (min) (s) larity RI ref. measurements 1-Octen-3-one 25.9 1.27 893 977.3 979.0 [265] y 6-Methyl-5-hepten-2-one 26.4 1.34 906 984.8 985.3 [228] y 2-Octanone 26.7 1.21 948 989.4 999.0 [265] y 2-Nonanone 33.4 1.25 913 1089.6 1090.0 [266] y (1R)-(+)-Norinone 37.4 1.63 862 1152.4 1155.1 [246] y 2-Decanone 40.0 1.30 955 1193.7 1196.5 [246] y 2-Undecanone 46.1 1.34 906 1293.4 1292.2 [267] y Esters Acetic acid methyl ester 5.0 0.72 941 526.3 511.0 [266] Ethyl acetate 6.7 1.20 806 609.1 612.0 [270] 2-Methyl-2-propenoic acid methyl 10.2 0.94 872 710.4 696.0 [266] ester Nitriles 2-Propenenitrile 4.9 1.48 964 521.1 511.1 [246] Propanenitrile 5.9 1.58 858 573.7 587.5 [246] Pentanenitrile 13.2 1.51 895 772.9 772.7 [246] Hexanenitrile 19.2 1.60 922 874.6 871.4 [246] y Benzonitrile 26.6 3.07 977 987.9 983.4 [228] y Octanenitrile 32.8 1.62 620 1080.6 1081.7 [267] Halogenated hydrocarbons 1,1-Dichloroethene 4.9 0.66 938 521.1 511.0 [228] Dichloromethane 5.1 1.28 918 531.6 521.9 [228] 1,1,1-Trichloroethane 7.8 1.17 950 642.4 645.1 [228] y Trichloroethylene 9.8 1.33 950 702.1 691.0 [266] 1-Chloropentane 12.3 0.83 843 754.2 754.9 [246] y Tetrachloroethylene 15.5 1.32 881 815.9 811.3 [228] y Chlorobenzene 17.8 1.28 970 852.4 846.5 [228] y 1-Chlorohexane 18.1 0.92 932 857.1 849.0 [268] y Tribromomethane 20.3 2.44 834 892.1 892.3 [271] y 1,4-Dichlorobenzene 28.4 1.51 930 1014.9 1016.5 [228] y 1,3-Dichlorobenzene 28.8 1.61 965 1020.9 1004.1 [268] y 1,2-Dichlorobenzene 28.9 1.60 961 1022.4 1027.0 [266] y Miscellaneous Dimethyl disulfide 12.0 1.05 803 747.9 742.0 [265] 3-Furaldehyde 16.5 4.06 877 831.7 829.0 [264] y 4-Hydroxy-4-methyl-2-pentanone 17.1 2.26 921 841.3 842.0 [266] y 2-Butylfuran 20.3 1.01 853 892.1 893.0 [270] y 2-Butoxyethanol 21.1 2.05 932 904.5 890.0 [266] y 1-(2-Furanyl)-ethanone 21.5 3.10 908 910.6 904.0 [270] y 2-Pentylfuran 27.0 1.07 932 993.9 994.0 [270] y Eucalyptol 30.2 1.03 930 1041.8 1030.0 [265] y Benzothiazole 43.1 3.20 936 1244.3 1240.0 [270] y 135 9 9. Peak detection methods for GC×GC: an overview 9.1 Introduction The use of comprehensive two-dimensional gas chromatography (GC×GC) is rapidly becoming more widespread also outside academia, especially since much improved hardware is now commercially available. Basically, GC×GC has three main benefits compared to one-dimensional GC (1D-GC): (i) if a sample contains a number of classes of structurally related compounds, a group-type separation can often be effected; (ii) in most cases a better separation between individual compounds can be realised; (iii) if GC×GC can separate the analytes of interest from the bulk of the sample matrix, sample preparation can be simplified. Petrochemistry was the first area in which GC×GC was widely applied, and today the technique is regularly used, e.g., to monitor and improve refinery processes. While for petrochemicals improved group-type separation is the main benefit, the higher separation efficiency by itself has been found to be extremely valuable in many other areas: GC×GC is increasingly being used today for the analysis of complex environmental, food and biological samples. The chemical diversity of the compounds in these samples is often large and the resulting chromatograms are therefore usually less structured than petrochemical samples; consequently, in most cases, the quantification of individual compounds rather than groups is required. As will be explained in more detail below, in GC×GC the signal of an individual compound is split into several peaklets (also called pulses or modulations) which have to be combined to reconstruct the complete GC×GC peak. Automating this combination process, i.e. 2D peak detection, is a challenging problem—especially when large sample series and many analytes are involved. Indeed, peak detection—which is the subject matter of the present chapter—is a much more complex problem in GC×GC than in 1D- GC. Other chemometric techniques such as retention time alignment, feature finding— i.e. finding distinct differences in large series of samples for e.g. metabolomics studies, visualisation, optimisation, background correction, etc.—are also more complex than in Published as: L.L.P. van Stee, U.A.Th. Brinkman, Trends Anal. Chem. 83 (2016) 1 9. Peak detection methods for GC×GC 1D-GC. For a discussion of techniques that deal with this wide range of topics, the reader is referred to publications such as [272-280] and references therein. 9.2 GC×GC: principles and visualisation In multidimensional gas chromatography, a second column is connected to the outlet of the first column, to enable a second separation with different selectivity. A device that can trap and release compounds eluting from the first column is inserted between the two columns. In one type of multi-dimensional GC, so-called heart cutting (GC–GC), a single fraction of typically 30–60 s of length eluting from the first column is trapped and re-analysed on the second column; the remaining sections of the chromatogram are diverted via a waste port. The chromatogram of the second separation has the same format as a normal 1D chromatogram and data analysis can be performed using standard methods. If there are a few key components in a sample which cannot easily be separated from each other and/or interfering matrix compounds, GC–GC is a rewarding strategy. However, there are two main disadvantages: (i) during trapping of the fraction of interest the resolution achieved on the first column is lost, and (ii) the procedure rapidly becomes very time-consuming and complicated if more than a few fractions have to be re- analysed; that is, non-target screening is very impractical. Injector Detector Main oven Carrier 2nd oven (optional) gas 1st dimension 2nd dimension column column Modulator Fig. 9.1: Schematic of a GC×GC system. For details, see text. In contrast to GC–GC, in which usually only a single fraction is re-injected onto the second column, in GC×GC many fractions comprising the entire 1D chromatogram are re-analysed, whence the often used term ‘comprehensive two-dimensional gas chromatography’, or GC×GC. The typical set-up of such a system is shown in Fig. 9.1; the device used to repeatedly trap, concentrate and, next, release compounds eluting from the first dimension column is called the modulator. The length of the trapped fractions is chosen such that loss of the first-dimension resolution is minimised. Consequently, the sampling rate has to be high enough to allow several measurements to be made across each first-dimension peak. Peak widths in the first-dimension separation—which is usually carried out on a conventional capillary column of 15–30 m length—typically are 10–30 s; the second separation should therefore be completed in 2– 138 9. Peak detection methods for GC×GC 8 s if one wants to have at least four sampling points across a 1D peak. Such fast separations can only be realised on short capillary columns of 1–2 m length, and preferably with a narrow bore and a thin film in order to maximise the plate number. All sample constituents are now subjected to two separations based on mutually different retention principles. A schematic illustration is shown in Fig. 9.2a and b: for each individual compound injected onto the first-dimension column several discrete peaklets show up in the second-dimension chromatograms. As regards data analysis, this peak- into-peaklet splitting is the major difference between 1D-GC and GC×GC. c. Trapped fractions a. b. Demodulation Visualisation as contour plot d. 7.5 8.0 8.5 024 68 024 6 8 1st dimension retention time (min) nd 2 dimension retention times (s) Fig. 9.2: Schematic depiction of a GC×GC separation. Three peaks showing partial overlap after the first-di- mension separation (a) are modulated (8 s modulation time) and detected as several discrete peaklets for each compound (b). The result of demodulation and visualisation is shown in c and d, respectively. The result of a GC×GC analysis is a collection of many short second-dimension chromatograms. In principle, each second-dimension run can be stored as a separate data file. However, since the basics of most software were developed for 1D chromatography, it is common practice to record all short chromatograms sequentially in a single file. The raw GC×GC data can of course be visualised as a conventional chromatogram (see Fig. 9.2b). However, interpretation will then be very difficult because, from the raw data, one cannot easily tell which peaklets belong to the same compound. In addition, any structure present in the chromatogram that can aid interpretation is not visible. These features only show up after so-called demodulation. Basically, demodulation implies cutting of the sequential GC×GC data into sections of the length of the modulation time. These sections are then stacked next to each other: the 1D data (a vector) are transformed into a 2D matrix. After conversion, the 2D data can be visualised as a contour or surface plot—in the early years by using general-purpose visualisation software, nowadays often by using dedicated GC×GC software packages. Examples of such software are ChromaTOF (Leco, St. Joseph, MI, USA), HyperChrom (Thermo Fischer Scientific, Waltham, MA, USA), Chromsquare (Shimadzu, Kyoto, Japan or Chromaleont, Messina, Italy) and 139 9. Peak detection methods for GC×GC GCImage (Zoex, Houston, TX, USA). To our knowledge, no direct comparisons of features of these packages such as speed of processing, visualisation, and data management have been published in the open literature. The main reason for this may well be that such a comparison rapidly loses its value because of the frequency with which updates of the software are published. GC×GC hardware, operational procedures and applications have been reviewed in several papers (e.g. [160;232;237;281]) and will not be extensively discussed here. 9.3 Classification of peak detection methods One of the key processes in chromatographic data analysis is peak detection, i.e. distinguishing analyte responses from each other and from the background signal. In the literature, this process is also referred to as ‘peak resolution’, ‘peak identification’, ‘peak integration’, ‘peak picking’ or 'deconvolution'. Based on the approach used, the methods for peak detection in GC×GC can be divided in four classes, and the general flow of data processing is shown in Fig. 9.3. It is important to note that the demodulation and visualisation steps (cf. Section 9.2) are performed for all methods to produce the 2D plot that is used in the graphical interface for all peak detection methods. It therefore holds the central position in the figure, and the procedure is indicated by bold lines. The demodulated data are also used as input for the multivariate and graphical methods that will be discussed in Sections 9.5 and 9.6. They are fundamentally different from the methods based on 1D peak detection (Section 9.4), where the first step is conventional peak(let) detection and integration, in most cases with algorithms included in the chromatography data system (CDS) used to record the chromatogram. The peaklet list is then further processed by either integrated algorithms, or exported to external software. For all methods, the concept of peak detection will be discussed as well as their merits and demerits, and application range. 140 9. Peak detection methods for GC×GC GC×GC raw data Legend nd (Sequentially stored 2 dimension chromatograms) Process Input or output Only performed for summation and clustering methods (9.4) 9.4.1 1D peak integration 9.2 1D peak detection Demodulation (9.2) 1D peaklet list Visualisation GCxGC matrix (2nd dimension 1D to 2D time conversion 2D plot chromatograms stacked (used in interface side-by-side) in all methods) peaklet list with 1 2 tR and tR 9.4.2 9.4.3 9.5 9.6 User supervised Automated peaklet Multivariate methods Graphical method peak(let) summation clustering Fig. 9.3: Flow of data-processing and classification of the different methods of peak detection. Centrally placed is the 2D-plot used for all approaches in the user interface. On the left side are the methods based on 1D peak(let) detection, on the right side the multivariate and graphical methods that use the converted data as input. Details of subsequent processing (in the bottom squares) are discussed in the separate sections. 9.4 Methods based on 1D peak detection 1D peak-detection and integration routines are well known, as are their flaws and merits; they are included in almost all CDSs. Subsequent summation of the areas of the peaklets, will result in the total peak area of the selected compound(s). However, automated selection of peaklets belonging to a single compound is not trivial and will be discussed later. Historically, the 1D-integration-based methods were first applied to group-type analyses, usually with in-house developed software. First all peaklets are integrated by 1D-software. The 2D-plot is then used to manually draw/define polygons around peak(let)s which belong to the same group. Finally, the 1D-peaklet areas of all peaklets inside each polygon are summed. Pertinent user-supervised summation methods will be discussed in Section 9.4.2. These methods can also be used for individual compounds: there is no fundamental difference between selecting a group or an individual compound by drawing a polygon. 141 9. Peak detection methods for GC×GC However, if many individual targets have to be integrated, the method becomes laborious, especially if one has to correct for retention-time shifts in large series of samples. Obviously, in these cases and even more so when a screening study has to be performed, for which it is desirable to have a method that can generate a list of all 2D peaks in the chromatogram, an automated method to cluster all peaklets originating from the individual compounds into 2D peaks, is required. In this chapter, the automated methods are referred to as automated clustering methods (Section 9.4.3) in contrast to the user-supervised polygon method that will be referred to as a user-supervised summation method (Section 9.4.2). 9.4.1 1D processing of chromatographic data The basic, but frequently not straightforward, process of 1D peak detection aims at finding the start, end and apex of each peak. Once these are known, the peak area can be calculated through integration. Only the two most applied methods used in conjunction with GC×GC data are described here. Single-channel data. From among the single-channel (or univariate) detectors, the flame ionisation detector (FID) is still widely used because of its robustness, excellent CH- based sensitivity, almost unit response for hydrocarbons, and, consequently, wide application range. The two most commonly applied methods of peak detection for this type of data are geometric methods and, more sophisticated, curve fitting [282;283]. However, only the former have been used for GC×GC. The geometric methods are the classical methods based on triangulation; they were implemented in electronic integrators in the 1960s and, later, in computer software. Apices and borders are usually determined from sign changes of the first, second and/or higher order derivatives of the chromatographic signal. Vertical lines in the valleys (perpendicular drop) are commonly used to separate the peaks, but tangent skimming can be useful to obtain more accurate results for small peaks on the tail of a much larger peak. These methods are still used extensively in 1D and 2D chromatography, and errors can be recognised and corrected during (visual) inspection of the results. However, in complex 2D chromatograms, with thousands of peaklets per chromatogram, such evaluation will become unmanageable. Another factor that requires attention is the variation in peak widths. For reliable results the peak width is an important input parameter for most software based on the geometric methods. Since in temperature- programmed GC×GC the temperature changes very little during each second-dimension run, each run can be considered isothermal. As a consequence peak widths will broaden during each second-dimension run which can hinder optimal peak detection. For further details of these methods, the interested reader should consult e.g. [284-286]. 142 9. Peak detection methods for GC×GC Multi-channel data. Several multi-channel detectors are available for use with GC (e.g. the atomic emission detector (AED) and the pulsed flame photometric detector (PFPD)) but in the following text we will only discuss the most commonly applied, the mass spectrometer (MS). MS data can be processed as single-channel data when, e.g., the total ion chromatogram (TIC) or a single m/z trace is used as input. However, most GC×GC–MS studies published so far utilise the Leco (Mönchengladbach, Germany) ChromaTOF software, sometimes combined with in-house developed software. The ChromaTOF software includes a proprietary 1D deconvolution algorithm that is based on m/z traces that maximise simultaneously. Unfortunately, details about averaging, baseline correction and peak detection are not available. Peak integration is performed by integrating and subsequent summing of the deconvoluted m/z traces. The result of the deconvolution and integration is a peaklet list with peak start, apex and end, retention times and ‘corrected’ peak areas. Next to this, a ‘pure’ mass spectrum, which features only these m/z values, is generated. Moreover, the software also selects a ‘unique’ m/z that shows the least interference from adjacent compounds. For an overview of this type of 1D-deconvolution methods, the reader is referred to [287;288]. The 1D peaklet lists generated by either the geometric or deconvolution method are used as input for the methods described in the next two sections. 9.4.2 User-supervised summation by polygon method With user-supervised summation, the peaklets that have to be summed are selected using manually defined areas. The retention-time windows of the target compounds, the so- called template, are specified by manually drawing geometric shapes around groups or individual compounds of interest. Apart from the quality of the initial 1D integration (cf. Section 9.4.1), the accuracy of the final result depends only on the precision with which the shapes have been drawn. Fig. 9.4: Example of a polygon method used to quantify different groups of toxaphene congeners [164]. 143 9. Peak detection methods for GC×GC The ease with which the geometric shapes can be defined and transformed is mainly a case of software design. For example, it is preferable that next to rectangles and ovals, irregular shapes such as splines or polygons, as depicted in Fig. 9.4, can be drawn as well. Another issue is the adjustment of the template to correct for run-to-run retention- time shifts or shifts caused by gradual column ageing. A convenient feature for this is the use of ‘magnetic’ polygons [289] that define a mesh-grid without any gaps. Once all the polygons have been manually drawn, adaptation of one polygon automatically reshapes the adjoining polygons. Another useful feature is the option to stretch and shift the entire template [289], as illustrated in Fig. 9.5. Fig. 9.5: Linear transformation of the integration template. The dashed lines represent the original template, the solid lines the template after transformation [289]. The template method is implemented in various types of in-house developed software, and also in some proprietary and some partly disclosed software [164;251;289] and used successfully on a routine basis by many laboratories in the petrochemical industry. It is also implemented in the commercially available software HyperChrom, Chromsquare, ChromaTOF and GCImage. An overview of various applications of the above mentioned commercial software is included in a review by Pierce et al. [275]. It is possible to use the template method for analyses of individual compounds; however, when the aim is to analyse all compounds in a chromatogram this is very laborious. Without a graphical user interface, user-supervised summation of GC×GC–MS data can also be performed manually in a spreadsheet software. This requires that the tentative identities from a mass spectral library search are available, and the data are 1 2 demodulated so that the first- and second-dimension retention times, tR and tR , respectively, of each peaklet are known. Results obtained by careful sorting and 144 9. Peak detection methods for GC×GC combining of the peaklets are most satisfactory [170;290], but the procedure is laborious and time-consuming. In other words, efficient utilisation of the technique requires automation, which is discussed in the next section. 9.4.3 Automated peaklet clustering In order to carry out automated clustering of the peaklets that make up the 2D peaks of compounds of interest, an algorithm is required that can distinguish the individual 2D peaks, i.e., that can perform 2D peak detection. Here one should note that most GC×GC analyses are carried out with the first- and second-dimension columns situated in the same, temperature-programmed, oven. Temperature programming is common practice in many GC separations and allows analysis over a wide volatility range in an efficient 2 manner. However, the increasing temperature results in a slightly lower tr for each next peaklet belonging to the same compound. As a result, the shape of the 2D peak made up from the individual peaklets, is not fully symmetrical (not bilinear, see Section 9.5.1). Four clustering methods are discussed below. Peters’ method. Peters et al. [291] described an algorithm to detect 2D peaks in GC×GC–FID data. After peaklet detection by processing the raw data as a regular 1D chromatogram, the decision whether to cluster the peaklets or not is based on (i) the overlap criterion, i.e. the differences in their second-dimension retention times, and (ii) the unimodality criterion, i.e. the peak profile in the first-dimension should show only one maximum. Fig. 9.6: Schematic of theoretical mutual positions of peaklets (represented as rectangles) belonging to a single 2D peak. The boxes indicate the start and end of the peaklets in both dimensions, the dots the peak apices [291]. For explanation, see text. For the overlap criterion (see Fig. 9.6), peaklet A is defined as the last peaklet of the existing 2D cluster and peaklet B as the candidate peaklet to merge. Depending on the adjacent peaklet regions considered, five different situations are distinguished: (a) both peaklets start and end at the same 2D retention times (b) peaklet A starts later than peaklet B and it also ends later (c) peaklet B starts later than peaklet A and it also ends later 145 9. Peak detection methods for GC×GC (d) peaklet B starts later than peaklet A, but it ends earlier (e) peaklet A starts later than peaklet B, but it ends earlier. In temperature-programmed GC×GC, the regions of adjacent peaklets should in principle obey the trend of Fig. 9.6b because of the increase in temperature between successive second-dimension chromatograms. However, in practice, the peak regions may show all variations displayed in Fig. 9.6, that can be caused by peak overloading (fronting), tailing or matrix effects. Clustering proceeds as follows. The ratio of overlap, OV, is calculated as OV=b/a. A threshold, ThrOV, is selected by the user. If OV is greater than ThrOV, the candidate peaklet B is accepted to be subjected to the unimodality criterion. If not, this candidate peaklet is rejected and the algorithm proceeds to the next candidate peaklet. If a peaklet is fully encompassed by the region of the other peaklet (Fig. 9.6d and e), the candidate peaklet B is always accepted. The 3-step process is visualised in Fig. 9.7. In order to reliably trace the first- dimension profile of a peak for the unimodality criterion, at least 7–10 data points are required. Since, in practice, typically only 3–5 data points (peaklets) are recorded, an interpolation algorithm, detailed in [291], is applied to increase the number of data points describing the 1D profile (shown as smaller sized dots in step 2 of Fig. 9.7). 146 9. Peak detection methods for GC×GC Step 1: 2 Depending on the overlap, adjacent peaklets are grouped together (cf. Fig. 9.6) into three groups. 1 3 Step 2: According to the unimodality criterion, the groups are split into separate 2D peaks. Since after point c another maximum is detected, group 1 is split into two peaks at the extrapolated minimum between c and d. Step 3: Finally, four peaks are detected, consisting of the peaklets (a-c), (d-f), 3 (g-k) and peaklets at bottom right. Fig. 9.7: Steps used in automated clustering (2D peak detection) in GC×GC–FID (adapted from [291]). For explanation, see text. The authors found that the algorithm was able to correctly cluster peaklets in most cases. As regards complications, applying the unimodality criterion to a distorted peak can result in the reporting of two peaks whilst only one is present. On the other hand, disregarding this criterion can cause two compounds to be reported as a single 2D peak, as would be the case for the peaklets in cluster 1 of Fig. 9.7. However, such problems are not unique to processing of 2D data—they can also occur during the processing of 1D 147 9. Peak detection methods for GC×GC data. An improved Peters’ method was published several years later [292] by one of the co-authors of the original study. Here, the statistical method of Bayesian inference is used to compute the most likely arrangement of the peaklets that belong to individual 2D peaks. Although promising results were obtained, overlap is not dealt with, neither do other methods for unimodal data such as from an FID. According to the author, in highly crowded chromatograms the user should therefore be careful to distinguish the concepts of peak and compound [292]. The methods described above for clustering GC×GC–FID peaklets are based on peaklet positions and intensities, i.e. expected profiles in both dimensions and mutual positions of the peaklets in the 2D chromatogram. In the case of GC×GC–MS, additional mass-spectral information can be used to help reach a decision on the clustering of peaklets. Two relevant methods are discussed below. ChromaTOF software method. A much used GC×GC–MS clustering method is the one incorporated in the Leco ChromaTOF software. The first step is 1D peaklet detection that is based on finding m/z traces that maximise simultaneously (cf. Section 9.4.1) which results in a collection of deconvoluted peaklet areas and mass spectra. Next, the algorithm in the ChromaTOF software directly compares the spectra of adjacent deconvoluted peaklets to reach a decision on whether to cluster or not. A limit can be set on the mutual similarity requirement for peaklets to be clustered. Similar to the deconvolution algorithm, the algorithm for clustering is proprietary and is not described in detail. Most probably, a unimodality criterion is included as well. Details about the shape of the ‘search area’ and the sequence in which the peaklets are selected and processed are not available. The ChromaTOF software is used as data system in many studies. A large part of these—and many GC×GC studies in general—are of a qualitative nature, in which the improved separation and mass spectral information is used to identify compounds in a large variety of samples. Widespread use of this software is clear from the many references in literature reviews (e.g. [275;293-295]). As an illustration, a selected number of recent studies using the approach is shown in Table 9.1. 148 9. Peak detection methods for GC×GC Table 9.1: Selected recent GC×GC applications using ChromaTOF software for data processing. Applications Refs. Food, fragrances, vegetable oils and fats Pesticides in food [296] (Semi-)volatiles in wine [297-300] Volatiles from pineapple pulp [301] Antimicrobial food components [302] Chemical signature of ecstasy volatiles [303] Volatile spoilage metabolites in fermented cucumbers [304] Elaidic and vaccenic acids in foods [305] Characterisation of foodborne pathogen bacteria profiles [306] Volatile constituents in wine and brewed coffee [307] Biological/biota Free D-amino acids in serum and urine [308] Alkylnaphthalenes in mussels [309] Organobrominated compounds in bluefin tuna [310] Phytosterol oxidation products in human plasma [311] Evaluation of non-target/metabolomic data analysis in plasma samples [312] Postprandial plasma metabolome analysis [313] Organohalogens/environmental PCB separation on polysiloxane and an ionic liquid column [314] Degradation of tar oil in soil [315] Semivolatile organic compounds in aerosol particles [316] Priority and emerging contaminants in waste and river water [317] Environmental forensic investigations [318] Petrochemical/biofuel Oligomerisation products of Fischer-Tropsch derived light alkenes [319] Qualitative and quantitative analysis of pyrolysis oil [320] Naphthenic acids in petroleum related samples [321-323] Rapeseed oil methyl ester pyrolysis [324] PAHs in heavy fuel oil [325] Oxygenated compounds in a direct coal liquefaction middle distillate [326] Organic sulphur compounds in coal bitumen [327] The ChromaTOF software can perform automated non-target peak finding in entire 2D chromatograms, and some studies also report its use for quantitative 2D data analysis [312;328-336]. However, the time consumption of the procedures and/or specific problems are usually not described in detail. In one study [336] the consequences of duplicate peak finding and a method to filter them out using additional software is discussed. In another study, de la Mata et al. [286] investigated the effect of several integration parameters that can be set in the ChromaTOF software. The main parameter that was investigated was the ‘expected peak width’ of the peaklets. It was found that setting this width close to the experimentally found widths results in the most accurate detection and integration; large differences could cause peaks to be missed, especially at low signal 149 9. Peak detection methods for GC×GC intensities. According to the authors, smoothing (which was set to automatic, but is related to the expected peak width) plays an important role, and similar results could be expected from any data-interpretation software incorporating window-based smoothing. Koek et al. [312] performed a comparison of the quantitative data obtained from the analysis of mouse liver samples with an Agilent 5973 GC–MS, and a Leco Pegasus III GC×GC–MS. A single quality control (QC) sample pooled from six different liver samples was injected 5 times in between 24 study samples (not discussed here). The GC–MS data of the QC sample were processed with a target method for 175 compounds2, and peak integration was checked/corrected manually. These quantitative data were used as reference. GC×GC–MS data were analysed as follows. First, a table of all 2D peaks present in the chromatogram was generated by fully automated non-target analysis with the ChromaTOF software of one selected run of the QC sample. Then all peaks related to solvents and reagents were deleted from the table, and the retention times and spectra of the remaining 1025 peaks were used to build a target method in ChromaTOF. Peaks that could be identified based on their spectra and retention times were labelled as such, but most peaks were added as unknown targets. Then peak detection was performed with the ChromaTOF target method for all five QC runs, and the results were used without further manual correction of the integration or peaklet merging. The RSDs of the MS response of target compounds that were found with both 1D and 2D GC×GC–MS (total 107 targets) were compared. For the majority (70) of the metabolites, similar RSDs were found with both methods (<10% difference in RSDs), although generally the values of the RSDs for the GC×GC–MS data were slightly higher than those obtained with the GC–MS processing. It was concluded that such a semi- automated approach is feasible with the ChromaTOF software although the time required by the technician was about 50 h, with an additional 60 h of computing time (mostly overnight) for the 29 samples. Faster computers and software optimisation will be needed to improve the feasibility of the approach [312]. An advantage of GC×GC was the much larger number of peaks detected, which resulted in the finding of several extra candidate biomarkers in the subsequent feature-finding principal component analysis (not discussed here) of the study samples. ChromSquare. Another software program that is commercially available was developed by Chromaleont Srl (Messina, Italy). The initial version of the software [337] was aimed at LC×LC, but GC×GC and LC×GC versions are also available now, for both univariate 2 The targets were selected on the basis of their presence in the analysed samples from an in-house library containing the mass spectra and tR of over 600 authentic standards, over 100 annotated metabolites (spectral match with NIST library) and over 200 unknown metabolites that were measured in blood products in numerous previous studies [312]. 150 9. Peak detection methods for GC×GC and MS detectors. Depending on the license type the programme can process the machine-independent data format netCDF (AIA/ANDI), the native data format used by Shimadzu (Kyoto, Japan), or both. Some of the options include the visualisation as 2D and 3D plots, quantification using calibration curves, and identification of compounds using MS spectral searches with additional software linked to ChromSquare. The LC×LC version employs a triangularisation method for peaklet integration (see [337;338] for details). Whether this method is also used in the GC×GC versions cannot be found in the literature. The method for clustering the peaklets is proprietary, but as described on the supporting web page [339], retention times and spectra are used to decide whether a peak (peaklet) belongs to a certain spot (2D peak). The usual workflow and examples of the graphical user interface are also described in [340]. The software has been used in more than a dozen published studies, in many cases mainly for visualisation and MS identification, e.g. for lipids in milk [341], roasted coffee volatiles [342], and essential oils [343]. Some papers discuss flow modulation and gas velocity in GC×GC [344-347]. In some studies quantification was performed as well. A large number of compounds were quantified (as % of total peak area) in the headspace of white truffle [348], and volatiles in different types of Marsala wines were also quantified [349]. In another paper [350], GC×GC–MS/MS was used for the quantification of preservatives in essential oils including calibration curves and analysis in triplicate. Repeatabilities for the peak areas of the lowest calibrator were <5% RSD. In the last study the authors commented that generation of peak areas was done in a simple and rapid manner. Unfortunately, no further details on the number of peaklets and the correct clustering procedure are discussed in any of the quantitative studies. Hyperchrom. Automated 2D peak integration based on 1D peak integration and clustering is also available as the Hyperchrom software. However, except for one reference [351] that describes that the algorithm is based on automated clustering, no further detailed information can be found in the open literature about this mode of operation. 2DAid method. Another GC×GC–MS peaklet clustering method, 2DAid, is based on theoretical, i.e. calculated, 2D peak shapes and compound identities found after mass- spectral library searching [352]. Briefly, after 1D GC–MS data processing using the ChromaTOF deconvolution algorithm, the resulting peaklet list is imported into home- written software. The raw GC×GC–MS data, saved as netCDF, are also loaded by this software. 151 9. Peak detection methods for GC×GC 5.60 a. 4-Nitro-aninline b. 4.80 4-Bromo-aninline 4.00 3.20 2.40 100% 1.60 0.9% 0.2% 0.80 2-Chloro-phenol 0.04% 0.00 200 600 1000 1400 1800 2200 2600 Fig. 9.8: (a) Peak shape prediction (parallel lines) in a contrast-enhanced extracted-ion chromatogram of a mixture of phenols and anilines. The squares indicate the most intense peaklet. (b) A normal-contrast depiction of the chromatogram including predicted peak shapes. Subsequent clustering proceeds as follows. First, the most intense peaklet in the entire chromatogram is selected automatedly. In the next step, using the known oven temperature, carrier gas velocities and hold-up times, the positions of other sub-peaklets possibly belonging to this ‘master’ peaklet are calculated using basic GC principles, and a polygon encompassing the calculated 2D peak shape is drawn around the master peaklet; this is illustrated in Fig. 9.8 where the polygons are reduced to two parallel lines. Next, the identities of the peaklets experimentally found within the borders of the calculated 2D peak shape are compared with that of the master peaklet using the CAS numbers and match qualities of the library hits. The first library hit of either the master or one of the sub-peaklets is not necessarily the correct hit. Therefore, for the master peaklet the first hit is selected by default, but for each next peaklet the first three to five hits are compared with the master peaklet in order to achieve correct clustering. An upper limit can be set to the allowed difference in spectral match quality. After the sub-peaklets have been combined with the main peaklet—i.e. when the 2D peak has been defined—they are marked as no longer eligible for further clustering. Now, the next highest peaklet in the chromatogram is selected and the process is repeated until no more peaklets are available. When no sub-peaklets are found to be available for clustering with the master, that single peaklet is accepted as the one defining the 2D peak. 152 9. Peak detection methods for GC×GC The algorithm proved to be successful in clustering the peaklets, and in combination with a built-in filter that sets a lower limit on the area of the 2D peaks that are finally reported, the data become manageable even when a large number of peaks are present. In one application, when 34 anilines and phenols (5 ng each) were analysed, the 2000 peaklets found were reduced to 490 2D peaks after clustering; subsequent filtering reduced the final list to 40 peaks. In another experiment, with a mixture of 362 compounds analysed, the 3000 peaklets found were clustered to 1400 2D peaks (most analytes were modulated 2–4 times), and when several were filters applied (e.g. column bleed, peak size) the list was reduced to 450 compounds, with only a few of the 362 compounds missing in the final list. Further details are presented in Chapter 10 of this thesis. It may be good to add that the main advantage of the ChromaTOF over the 2D Aid method is that it directly compares the deconvoluted mass spectra of the peaklets instead of the indirect method of matching CAS numbers of the library hits. On the other hand, a disadvantage of the ChromaTOF method is that the secondary processing, i.e. the clustering, cannot be performed as an independent task. This means that the time consuming 1D deconvolution procedure has to be executed every time the clustering parameters are changed, whilst the 2DAid procedure treats the clustering as an independent process which takes only a few seconds. The authors’ own experience and information provided by other users indicates that clustering can go wrong with certain peaks, both with ChromaTOF [353] and 2DAid. Designing tools that can, then, efficiently review the results, ‘flag’ possible problem areas, and help to correct the errors, are needed. Unfortunately, this topic is hardly ever discussed in the open literature. 9.5 Multivariate techniques Quite a number of papers describe deconvolution and/or quantification of GC×GC data with multivariate analysis techniques (MVA). These techniques are usually based on the so-called parallel factor analysis (PARAFAC) model. PARAFAC-based methods can be applied to find individual differences (contributing factors) in a set of samples that are measured in a ‘crossed fashion’. A single GC×GC–FID measurement results in a 2D landscape of responses of individual compounds at different retention times, but a PARAFAC method cannot be applied to these data. However, when several related samples are measured in a crossed fashion, e.g. a time series of samples from a fermentation process, the combined GC×GC–FID chromatograms contain information on several or many compounds that vary in concentration, and it are these variations that PARAFAC methods can distinguish from the background signal and other peaks. 153 9. Peak detection methods for GC×GC Importantly, the PARAFAC methods make no assumption regarding the chromatographic peak shape. Consequently, if two or more peaks co-vary in exactly the same way in concentration, this set of chromatographic peaks will be modelled as a single factor, not as separate peaks. This can be avoided by division of the chromatograms in smaller sections, as will be discussed in Section 9.5.3. In chemometric terms, data from a single GC×GC–FID run, in which a matrix (a 1 2 two-dimensional array) is indexed by tR and tR, and the elements of the matrix contain the response, are two-way data. When the data of two or more related GC×GC–FID runs are combined, they are arranged in a cube (three-dimensional array) and are called three- way data. PARAFAC methods cannot be applied to two-way data. Another example of three-way data is a single GC×GC–MS run (except if there is only single-ion monitoring): the data are also arranged in a cube with many 2D chromatograms for each m/z value. PARAFAC data analysis can now distinguish the individual 2D peaks by the mutual differences in the responses in the m/z profiles of the 2D peaks, i.e. their mass spectra. However, as is the case with co-varying concentrations in the fermentation example, when two or more peaks have identical mass spectra, they cannot be distinguished as separate 2D peaks without dividing the chromatogram into smaller sections. 9.5.1 The PARAFAC model The structure of demodulated data from a single GC×GC chromatogram is a matrix 1 2 indexed by tR and tR with responses in each matrix element. A single 2D peak within a chromatogram can be represented by a small sub-matrix limited by the retention times of the 2D peak. However, the PARAFAC model uses a more compact way to describe the data. As a consequence, the measured data can usually not be modelled exactly (the requirements for optimal modelling will be discussed below). In the PARAFAC model, the shape of a 2D peak is modelled by two vectors, of which the outer product most closely fits the actual peak shape (Fig. 9.9). 154 9. Peak detection methods for GC×GC 1 i ( tR ) ) R t 2 0 0 0 0 0 1 5 9 5 1 ai 0 0 0 0 0 0 0 0 0 0 0 j (j 1 0 0 0 0 0 1 5 9 5 1 1 0 0 0 0 0 1 5 9 5 1 2 0 0 0 0 0 2 10 18 10 2 4 0 0 0 0 0 4 20 36 20 4 5 0 0 0 0 0 5 25 45 25 5 9 0 0 0 0 0 9 45 81 45 9 1 0 0 0 0 0 1 5 9 5 1 (5×9=45, 0 0 0 0 0 0 0 0 0 0 0 etc.) b j Fig. 9.9: Description of how the outer product of two vectors (a and b, a so called dyad) models a 2D profile in the PARAFAC model. The outer product results in a matrix with the signal intensity in each matrix element. In mathematical notation3, the outer product of the two vectors in Fig. 9.9 can be written as: Χ xij aibj (i 1,...,I;j 1,...,J ) (Eq. 9.1) The shape of a 2D peak of an analyte present in a series of chromatograms, is modelled only once. Only the variation in the response(s) is incorporated for each analyte in each chromatogram. This is illustrated in Fig. 9.10. r=2 r=1 Sample 1 r = 1 r = 2 r = 3 k=1 = + + + E Sample 2 r=3 factor (r) k=2 r=1 ckr 1 2 3 1 3 5 0 2 2 0 4 sample (k) sample Fig. 9.10: Illustration of how four individual peaks in two chromatograms are modelled in a PARAFAC model. Two peaks (r=1), present in both chromatograms, originate from the same analyte and the shape is modelled only once. The intensity of each 2D peak (factor) is modelled into the vector c for each kth sample/chromatogram. The peaks from the other analytes (r=2, r=3) are modelled separately. E (or eijk) describes the residual signal (see text). The equation of the full PARAFAC model is: R Χ xijk airbjrckr eijk (Eq. 9.2) r1 3 i, j, k are indices of the arrays, and I, J, K are the sizes of the arrays. Scalars are indicated by italic (e.g. ai), and vectors by bold lower-case characters (e.g. a). Bold upper-case letters (e.g. X) are used for two- dimensional arrays (matrices) and underlined bold upper-case letters (e.g. X) for three-dimensional arrays. 155 9. Peak detection methods for GC×GC It is the summation of all (R) ‘modelled’ 2D peaks (factors) with their shapes (aibj) th and relative concentrations (ck) in each (k ) run. This equation also introduces the residual array, eijk, an additional 3D array that contains the residual signal in each run that is not described by the sum of the factors, e.g., the baseplane, spurious peaks and noise. However, the residual can also include the 2D profile of a peak that was not recognised/modelled during the model-fitting procedure. Optimal data structure for model fit. Experimental data that are suitable to be optimally fitted by the PARAFAC model, should meet certain requirements. Firstly, for multiple chromatograms, since the shape and position of each contributing factor is modelled only once, the retention times in both dimensions have to be constant in all chromatograms in the data set. Pre-processing via retention-time alignment is often used to achieve this (see [250;354-356] for a detailed discussion). In addition, each peaklet of 2 a particular 2D peak should have the same tR. This requires a constant temperature of the second-dimension column during the elution of the 2D peak, which will result in a peak parallel to the first-dimension axis [357]. Actually, most GC×GC analyses are run under temperature-programmed conditions and the requirement, consequently, is not met. Finally, when the shapes of the measured profiles for a particular compound vary between different samples, e.g. due to overloading, the model found can contain two, instead of one, factor(s) modelling the same analyte. In other words, optimal data for fitting should be trilinear, i.e. linearly scalable in all three dimensions. As mentioned earlier, three-way data are also obtained from a single GC×GC–MS run4. In this case, the peak shapes in the separate m/z traces can be modelled with PARAFAC. Contrary to the case above, retention-time shifts are not an issue since each ion chromatogram is recorded in the same single GC×GC run. Furthermore, if a ToF-MS instrument is used, skewing does not occur and each ion trace will have the same shape irrespective of the analyte concentrations, except at the saturation and noise levels. Except for the influence of temperature programming of the oven, a single GC×GC–MS analysis will result in almost perfect trilinear data. PARAFAC modelling applied to a GC×GC–MS chromatogram can result in useful deconvoluted MS spectra and relative responses for each compound from the summed intensities of the m/z traces. For application of PARAFAC to both types of the above data, one should consult Section 9.5.3. 9.5.2 PARAFAC algorithms The previous section described the PARAFAC model and the way in which the experimental data can be presented as a collection of modelled 2D profiles and 4 Except when single-ion monitoring is used. 156 9. Peak detection methods for GC×GC responses. However, calculation of such a model by applying appropriate algorithms (sometimes also referred to as methods) is an entirely different matter. Two algorithms are mainly used in applications to GC×GC data: Generalised Rank Annihilation Method (GRAM) and PARAFAC-Alternating Least Squares (ALS). In short, GRAM is a fast non-iterative algorithm using eigenvectors, and is designed to handle only two samples, usually a standard (addition) sample, and an incurred sample. PARAFAC-ALS (somewhat confusingly often abbreviated in the literature as PARAFAC rather than as ALS) is a slower, iterative method, capable of handling larger sets of samples. It uses the residual term eijk in Eq. 9.2 to find the optimum fit by minimising the sum of the squared residuals. For an in-depth discussion of these complex procedures, the reader should consult e.g. [358-362]. One more important aspect in ALS, (prior) knowledge about the number of factors involved, will be discussed in the next sections. 9.5.3 Application of PARAFAC-ALS to GC×GC analyses Most early papers [239;363-366] on the application of PARAFAC-based approaches to GC×GC can be considered as proof-of-principle demonstrations. Typically, studies deal with a limited number (2–4) of typically poorly separated compounds; some studies apply isothermal separation conditions. Encouraging results were reported in terms of, e.g., the purity of the mass spectra and of the accuracies and precisions of the analyte concentrations determined even under low-resolution conditions. However, analyte concentration ratios often are rather close to unity and individual concentrations (signal- to-noise ratios) much higher than can be expected in real-life situations. In most of the quoted studies, the known number of peaks was defined manually as input for the model. Alternatively, chromatographically separated target analytes—e.g., six compounds in seven perfume samples analysed with GC×GC–FID—were processed in subsections, each containing a single analyte [355]. In other words, as regards the subject matter of the present review, peak detection, these studies do not provide much relevant information. Firstly, for non-target analysis, such input is not available, and secondly, the challenge (next to increasing the complexity of the sample types studied) is to find a way to detect the number of compounds in an automated fashion. The proof of principle of a semi-automated detection of the correct number of 2D peaks in a single GC×GC–MS run was described for the determination of bromobenzene in hexane [367]. The authors used mass spectral comparisons of the peaks to detect ‘overfitting’: a series of models, using an increasing number of factors (n), were fit to the actual data; when overfitting occurs, a single peak will be modelled as two adjacent peaks with very similar spectra, and the correct number of factors for the model is then determined as n-1. In subsequent studies [368;369] this method was further generalised 157 9. Peak detection methods for GC×GC to deal with larger sections or even entire GC×GC–ToF-MS chromatograms. Amongst others, it was applied to the headspace of a urine sample collected via SPME, that contained many unknowns. A large region of the chromatogram was subdivided into 55 sections. Computer processing time was between about 4.5 and 8 h depending on the approach used (cf. below), and about 50 compounds were reported. The method was called “fairly straightforward” by the authors, who state that care is needed in selecting appropriate subsection sizes. They also mention that significant savings in analysis time can be realised by excluding subsections in which no peaks are present, and that computation time could conceivably be reduced significantly by using massive parallel computing in which each processor deals with a single subsection5. 9.5.4 Non-trilinear models Non-trilinear modelling has been studied by some researchers to avoid the requirements of PARAFAC on trilinearity. PARAFAC2 [355;372;373] is a modified version of PARAFAC and is not limited by shifts in tR and change of peak shape [374]. This is effected by replacing one of the two vectors (a or b in Eqs. 9.1 and 9.2) by a set of 2 vectors; differences in tR and shape between runs can then be modelled separately. Similar to PARAFAC, the PARAFAC2 model is usually fitted using the ALS algorithm. A disadvantage of PARAFAC2 is that it is inherently more susceptible to noise [375]. The second non-trilinear method discussed here is multivariate curve resolution- ALS (MCR-ALS). This technique is designed for bilinear data, and can be applied to GC×GC data by ‘unfolding’ the data matrices into separate vectors. When applied to multiple-run GC×GC–FID data [376] retention-time alignment is usually required. However, with multiple-run GC×GC–MS data, this is not necessary since in MCR-ALS the algorithm can use the MS spectra to correlate the data [377]. This was demonstrated for PAHs in heavy-fuel oil samples where MCR-ALS could deconvolute overlapping peaks even when the spectra were only slightly different [325]. MCR-ALS is less stringent on trilinearity, but, similar to PARAFAC-ALS, its application is usually restricted to small areas of GC×GC chromatograms and initial estimates of the number of compounds (see e.g. [378]) are required as well before iterative analysis. Other drawbacks are potential failures to correct local minima and the requirement of advanced user intervention [374]. The studies mentioned above and some others [379;380] report the advantages and progress in this field. However, as is the case 5 There are several recent applied-type publications from the same group [331;370;371]. However, in these studies no essential further developments or improvements of the PARAFAC approach are reported. 158 9. Peak detection methods for GC×GC with PARAFAC, no dedicated GC×GC software is available and application to entire chromatograms by a general chromatographer is not a possibility yet. 9.6 Graphical drain method Similar to the multivariate methods, the graphical drain method is applied to the 2D form of GC×GC data obtained after demodulation of the raw data. However, the algorithm for 2D peak detection is entirely different and conceptually easy to understand: this so- called drain algorithm is an inversion of the watershed algorithm ([381] and references therein). The algorithm can be understood conceptually by picturing the 2D chromatogram as a relief map with more intense peaks having higher elevation. The surface is placed under enough 'water' to cover the point of highest elevation. Then the water is 'drained' and peaks appear as 'islands', and are labelled with a unique identification number. As the draining proceeds, the border between 2D peaks is established when the water between two parts of land is eliminated [381]. With a single exception [382], studies on the development and use of the graphical drain method in GC×GC refer to the commercially available software GCImage [272;381;383;384]. Before actual peak detection, a pre-processing step that is based on the structural and statistical properties of the GC×GC data, is used to remove the background [384]. It is also important to note that interpolation or resampling may be used during certain steps in the visualisation process. However, the quantification is based on the raw data (after background removal) [338]. In certain cases, the data may be smoothed to minimise oversegmentation; that is, to prevent a single peak being detected as multiple peaks. This 2 can occur when the difference in tr between two peaklets belonging to one 2D peak is too large and thus creates a saddle; on the other hand, too much smoothing can lead to undersegmentation [280]. 2 The probability of failure, such as e.g. oversegmentation, caused by differences in tr values between peaklets that belong to a single compound (e.g. due to temperature programming and instrumental variation) was studied in detail by Vivó-Truyols and Janssen [385] and was found to be 15–20% under normal conditions. In response, Latha et al. [386] showed that the rate of failure is much lower when shift correction is applied. However, this correction cannot be performed as an automated step in GCImage data processing. In both studies the watershed method was also compared to the clustering method of Peters et al. [291]. One of the drawbacks of the watershed routine, according to the authors of this critical study [385], is that it does not impose the condition of continuity for each peaklet, which implies that the signal of a substance may appear and disappear several times during the course of its elution. 159 9. Peak detection methods for GC×GC Another conclusion of the authors is that neither their own nor the drain method can be considered perfect under all conditions. The user-friendly graphical user interface of the GCImage software allows the display of chromatograms and MS spectra in various formats, and the performance of MS library searches and quantification of peaks. For example, Fig. 9.11b shows the visualisation of a single chromatogram were the apices of the detected peaks are indicated by white circles. Another option is the so-called ‘peak region matching’, which combines the watershed procedure with an approach that bears similarity with the polygon method in the user-supervised summation method (Section (9.4.2). However, in the present case an automated procedure is used to define one polygon per individual 2D peak. Moreover, the automated process can be based on multiple (cumulative) chromatograms, and thus allows a sort of target approach for all (cumulative) unknowns in each separate chromatogram [272;387]. An example of the visualisation of such a template is shown in Fig. 9.11a. a. b. Fig. 9.11: a. Cumulative chromatogram for nine samples of roasted hazelnuts and the regions of detected peaks that form the template (see text) shown as white polygons. b. Example of the visualisation of a single chromatogram with the apices of the detected peaks indicated by white circles (the inset represents the area bordered by the dotted lines, figures adapted from [388]). In several studies, GCImage is used for complete chromatographic quantitative data processing of hundreds of peaks (e.g. [388;389]). However, the performance of the watershed routine as regards segmentation errors, processing time and accuracy/precision is not discussed in these publications. It is therefore difficult to evaluate the performance of GCImage from these real-life studies. Several other examples of studies that employ, at least in part, GCImage for chromatographic processing are shown in Table 9.2. 160 9. Peak detection methods for GC×GC Table 9.2: Selected GC×GC applications using GCImage software for (partial*) chromatographic data processing Applications Refs. Food Non-target analysis of volatiles in hazelnuts (SPME) [388] Target and non-target analysis of volatiles in roasted coffee (SPME) [387] Analysis of dark chocolate (SPME) [390] Volatiles and semi-volatiles in milk powders (SPME) [391] Analysis of virgin olive oil (SPME)* [392] Analysis of essential oils of two species of Piperaceae (pepper) [389] Biological/biota Metabolomics on cancer tumor cells using GC×GC–high-resolution-MS (HRMS) [393] Analysis of rosewood essential oil [394] Analysis of volatiles in Eucalyptus clones using headspace SPME* [395] Analysis of Cannabis sativa extracts* [396] Organohalogens/environmental Organohalogens in various environmental samples using GC×GC–HRMS and MS/MS [397] Organohalogens in a sediment core sample using GC×GC–HRMS* [398] Qualitative analysis of airborne particulate matter using GC×GC–MS [399] Petrochemical/biofuel Qualitative and quantitative analysis of pyrolysis bio-oils [400] Investigation of oil weathering [401] Quantitative analysis of biodiesel* [402] * For example only visualisation and/or qualitative analysis. 9.7 Conclusions The above review discusses the main methods for automated target and, especially, non- target detection of individual peaks in GC×GC. The automated methods can be classified in three main groups: (1) automated peaklet clustering, (2) multivariate methods and (3) graphical techniques. The first group is based on 1D peak(let) detection, while the other two use the converted 2D data as input. Recommending one of the above techniques as providing the best results is not an easy matter, should such a preferred group exist at all. One problem is that a variety of parameters has to be considered, e.g. ease of handling, adaptability/flexibility, the possibility to check the general reliability of the results, suitability for feature finding (relevant in metabolomics) and the (commercial) availability of the software. It will be clear that not all of these parameters will have the same importance when addressing different types of analytical problem. Actually, speed of processing should also be included in this list. However, insufficient information is available in the open literature to discuss this aspect. 161 9. Peak detection methods for GC×GC In general, all methods will perform well for the majority of 2D peaks in a chromatogram that are mutually well separated. Difficulties can be expected with co- eluting peaks, especially when response ratios are large and/or MS spectra are highly similar. Therefore, optimal separation is a key issue when setting up a GC×GC method. However, with most complex samples co-elution cannot be avoided and flexibility during the subsequent data processing is essential to obtain reliable results. All methods based on 1D peaklet detection (Group 1) and subsequent automated clustering share the advantage that they are familiar to most chromatographers. A main benefit is that the integration start and end of a peaklet can be inspected visually and adjusted if necessary. In addition, the clustering of the peaklets can be inspected as well and, by adjusting the parameters, the 2D clustering parameters can be optimised to minimise errors. With the, often available, general knowledge about sample composition, only specific problem areas will then still require additional checks. The multivariate methods (Group 2) deal with the data in their true 2D form and the peak detection algorithm makes maximal use of the information available in the 2D structure. Almost all early publications describe the application of PARAFAC-ALS or GRAM to rather small parts of an entire chromatogram, but recent publications demonstrate progress towards a more non-target exploratory technique. However, a distinct drawback compared with the other two methods is that it cannot be applied to data from a single sample. Moreover, the methods are of an academic rather than applied nature: with no commercial products having been made available, there will be few analysts who will attempt to implement these methods on the basis of the descriptions presented in the open literature. The graphical drain method (Group 3) is applied to the GC×GC data in their 2D form and is conceptually easy to understand. With the appropriate software the boundaries of the detected peaks can be inspected visually and when errors are observed, the overall peak detection parameters can be adapted and optimised. Next to this, specific parameter sets can be used for selected areas and the integration can also be corrected manually. The automated peaklet clustering and the graphical methods share two more advantages. Firstly, they can be applied to the data of a single sample. This is essential for non-target screening to determine the global composition of an individual sample. Secondly, in contrast to the multivariate methods, both methods are available as part of commercial software programmes (which increasingly also include options such as feature finding6). Consequently, support from the manufacturers—and information exchange during user meetings or in fora—will be available. This will help to minimise 6 See e.g. [272-280] and references therein. 162 9. Peak detection methods for GC×GC start-up costs and reduce start-up time—important aspects today both in academia and, specifically, in more applied-type environments such as industry, hospitals and monitoring agencies. In summary, the present review illustrates that the development of GC×GC data analysis and its tools has already come a long way. Today, a wide variety of complex analytical problems can be handled both efficiently and reliably—that is, provided the analysts working with the technique have received a dedicated (but, in our experience, not overly long) training. On the other hand, the technique—or, rather, the various approaches presently on the market—are not fully mature as yet: further research is required, especially to validate the performance of the available methods and to compare them—and, here, academia should take the leading role—in terms such as listed in the early part of this section. Creating a wealth of innovative applications and helping to further widen the field will be the reward of workers active in this field. 163 10 10. Peak clustering in GC×GC–MS based on theoretical calculation of two-dimensional peak shapes: the 2DAid approach 10.1 Introduction After many years of development in the field of GC×GC hardware, several types of robust and user-friendly instruments are now available, and recording a GC×GC chromatogram is a straightforward procedure. This is also true for the visualisation of the chromatograms and the qualitative evaluation of small sets of data, a problem often encountered in daily practice. On the other hand, quantitative analysis of many individual compounds can be more problematic. Briefly, the specific problem with individual-compound data analysis is that each compound is usually modulated several times and in the raw data emerges as several peaklets (individual peaks in the one-dimensionally recorded data) of which the areas have to be summed to obtain the total signal of the 2D peak for that compound. There are three groups of methods for (semi-)automated 2D peak detection. Two of these are based on the raw data converted to its 2D representation, the third is largely based on processing of the 1D data. The first group comprises the so-called drain or watershed algorithm and is applied in the commercial GCImage software. The algorithm can be understood by considering the 2D chromatogram as a relief map with larger values having higher elevation. The surface is then submerged, and water is gradually 'drained'. During draining, an algorithm determines the borders between the peaks as soon as there is no more water dividing them [381]. The second group of methods is based on chemometric multivariate analysis techniques [241;355;365;403-405]. In most published literature these methods are applied to selected areas of multiple chromatograms and are shown to be useful for fast data analysis in e.g. process chemistry. Progress is continually being reported, and with increasing computer power, faster processing of entire chromatograms should be possible [275;368]. Published as: L.L.P. van Stee, U.A.Th. Brinkman, J. Chromatogr. A 1218 (2011) 7878 10. Peaklet clustering The third group of methods is based on the raw data that—with any GC×GC system—are always recorded as a 1D chromatogram and are made-up of the individual peaklets. The advantage is that the peaklet areas are determined using proven 1D peak integration techniques, with their known merits and flaws. In software employing this method, a 2D plot is used as well, but only for visualisation and interaction with the data. In general GC×GC practice, this is one of the most used methods, in combination with manual selection of peaks or entire areas that are integrated. However, for automated peak detection, algorithms are required to recognise and combine the peaklets that constitute individual 2D peaks. Since a couple of years the Leco ChromaTOF software includes such an algorithm that is proprietary. So far, Peters et al. [291] are the only workers who developed and described such an algorithm for GC×GC–FID data in the published literature. This study will focus on a method for clustering the individual peaklets, i.e., on devising an automated non-target screening method in which a peak list of manageable length is created by means of adequate combining/clustering of all peaklets belonging to individual compounds, plus the use of several selective filters. The clustering algorithm is incorporated in a home-made software application called 2DAid that includes visualisation functions for the chromatograms and MS spectra. Emphasis in the present chapter will be on the clustering algorithm. Some of the other features will be discussed in more detail in Chapter 11. 10.2 Theory As described in Chapter 9, GC×GC analyses are usually carried out with the first- and second-dimension columns situated in the same temperature-programmed oven. Because of the temperature programming, the consecutive peaklets of a single compound elute at slightly higher temperatures, which results in (slightly) different second-dimension 2 retention times ( tr) for each next peaklet. As a result, the shape of the 2D peak made up from the individual peaklets, is not symmetrical. The method explored in this study uses basic chromatographic laws describing the variation of retention time with temperature to calculate the expected 2D peak shapes, thereby defining the area in which the peaklets belonging to a single compound can be expected. Subsequently, mass-spectral library searching of the experimentally recorded peaklets in this area is used to select, and combine, them into a single 2D peak. One parameter required to calculate the shape of a 2D peak is the hold-up time in the 2 second-dimension column ( tm). First, we will describe the method for calculating this parameter and, next, the 2D peak shape prediction and the clustering algorithm. For the entire column ensemble, the inlet and outlet pressures are known: the column-head pressure and the vacuum of the MS, respectively. Moreover, for each 166 10. Peaklet clustering segment the dimensions and temperatures are known. Consequently, for each segment a Poiseuille equation (see e.g. [406;407] and references therein) can be described. A solution for the set of equations is found by demanding the same mass flow of carrier gas in each segment. When solved, all pressures at the junctions are found, and the flow and hold-up time for each segment can be calculated. The results will be reported in Section 10.4.1. 10.2.1 2D peak shape in temperature-programmed GC×GC A 2D peak consists of a number of peaklets that are equidistantly spaced on the first- dimension time axis with intervals equal to the modulation time used (see Chapter 9). The second-dimension retention time of each peaklet of a compound depends on the temperature, carrier flow and (thermodynamic) properties of the compound. The basic equations describing the relation between these parameters are given below. In GC, the relation between the retention time (tr), the hold-up time (tm) and the capacity factor (k) is given by [408;409]: k t r t m (Eq. 10.1) t m k is related to the temperature and the thermodynamic properties of a compound as follows: B ln k C (Eq. 10.2) T in which T is the absolute temperature. The enhalpy of solvation (ΔH), entropy of solvation (ΔS), universal gas constant (R) and phase ratio of the column (β) are included via: H B (Eq. 10.3) R and S C ln (Eq. 10.4) R Although B and C are dependent on the temperature [408], variations due to the temperature are more or less cancelled out, and good predictions can be made based on one set of values for B and C [409]. It is not our aim to calculate the retention times and peak shapes for compounds solely based on reported values of B and C and the known experimental conditions (T and tm). Such data are scarce and they depend on the nature of the stationary phase used. 167 10. Peaklet clustering 2 Rather, the idea behind the 2D peak-shape prediction described here is to use the tr from the most intense peaklet of a compound as an experimental value. If C is known (see below), this value and the experimental conditions can then be used to estimate a value for B which is valid for the specific compound and column used. Next, the value of B is 2 used to predict (whence the accent) tr' for the other peaklets of this compound. From the oven programme the temperature at each point in time is known, and the hold-up time can be calculated as described in the previous section. Next, k can be calculated (Eq. 10.1). This leaves two unknowns (B and C) in Eq. 10.2, which equation cannot be solved without choosing a fixed value for one of the parameters. Although C is compound- and column-type dependent, it generally varies between 8 and 20 [408;409]. In the present study, several values of C were selected to verify whether it is permitted to use a single value to appropriately describe the experimental 2D peak shapes of the majority of compounds. Once the value of C has been selected, the value of B of a specific compound can be 2 calculated according to Eq. 10.5, using tr from the most intense peaklet, the calculated 2 value of tm and the oven temperature: 2 2 2 B T (ln(( tr tm )/ tm ) C) (Eq. 10.5) 2 B and the other parameters can now be used to predict tr' for the other peaklets of this compound at different temperatures: 2 2 tr ' tm (1 exp(B T C)) (Eq. 10.6) Eq. 10.6 describes a plot in the GC×GC chromatogram in which the peaklets of one 2 1 compound are expected to elute at tr' and tr the latter being directly related to T by the oven temperature programme. The area that will be used to search for peaklets belonging to a single compound is defined by adding a user-defined margin in the vertical direction on both sides of the plot. In Section 10.4, the calculated and experimental peak shapes 2 will be compared, and the robustness against changing values of tm and C will be discussed. 10.2.2 Peak clustering As mentioned above, the area defined by the expected 2D peak shape will be used as the ‘search area’ to cluster the peaklets belonging to one compound based on the tentative identities obtained from mass spectral library searching. Usually, results from library searching are presented as a list of possible candidates that are sorted on the quality of the match between the unknown and the library spectrum. In GC–MS screening, it is important to have several hits available during manual interpretation of the results. In 168 10. Peaklet clustering Import list of all peaklets such a situation, knowledge of the nature of, and other constituents present in, the sample, Sort on height combined with the expertise of the technician, can help to select the best Select (next) highest candidate from the list, which is not eligible peaklet necessarily the hit with the best match Determine search area quality. for selected peaklet For the clustering algorithm presented here, multiple hits were also used as will be Loop through every peaklet in search area described later. The 2DAid software imports these library search results directly from ‘link filter’ results generated by the CDS software. The Q>limits? further input required comprises the full-scan Y chromatogram in netCDF format, the Add to selected highest peaklet modulation time, the oven temperature programme, the expected peak width, and the 2 values of tm and C. All peaklets in N search area The clustering can then be started and evaluated? proceeds as indicated in Fig. 10.1. First, all Y individual peaklets are sorted on peaklet area. Mark highest and all Then, the peaklet with the largest area in the clustered peaklets as non- eligible entire chromatogram is selected. Using the calculations described above, the 'search area' associated with this peaklet is then Y Eligible determined. peaklets remaining? Then, the first five library hits of every peaklet within this area are compared with N the first library hit of the most intense peaklet ‘post filter’ ('master peaklet') based on the CAS numbers Q and total peak area of the hits. However, there is a possibility >limits? that very low-quality library matches are Y clustered to the main peak in this way. Add the 2D peak Therefore, an adaptable filter ('link filter') is to the final list used to evaluate the match quality of a library Fig. 10.1: Flow diagram of the clustering hit before it is clustered to the master peak. In and filtering method. this filter, threshold values can be set for the similarity, reversed and probability qualifiers. 169 10. Peaklet clustering When a peaklet passes the filter, it is clustered to the master peaklet. Subsequently, the next adjacent peaklet in the search area is evaluated. After all peaklets in the search area have been processed, the master peaklet, and all peaklets clustered to it, constitute a 2D peak, with a total area equal to the sum of the peaklet 'peak areas'. All these peaklets are then marked as 'non-eligible' for further processing, and the entire procedure is restarted by selecting the next most intense eligible peaklet in the entire chromatogram, until all peaklets are marked non-eligible. Post filtering. A number of filters based on peak area, identity and match quality can be applied to the final list of 2D peaks and are useful to reduce the number of 2D peaks reported. Although in a non-target analysis the compounds of most interest are not necessarily the largest in intensity, a filter based on the total area (i.e. the summed areas of the peaklets) can help to reduce the number of peaks to a manageable quantity. Besides a lower limit, an upper limit can be set as well. In this way, several regimes can be inspected separately. One can set up a non-target analysis in which, first, the most intense peaks are inspected, and subsequently the peaks with medium and low intensity. Another filter that is incorporated into 2DAid allows the exclusion of 2D peaks based on their identity (CAS number). In this way, compounds originating from e.g. column bleeding and solvents can be eliminated from the final results. 10.3 Experimental Methods. GC×GC analyses were performed in a system consisting of an HP6890 gas chromatograph equipped with an HP7673 autosampler (Agilent Technologies, Palo Alto, CA, USA) used for 1-µl injections. An Optic II programmable injector (ATAS, Veldhoven, the Netherlands) fitted with a multi-capillary liner was used for hot-split injections at 260ºC, with a split ratio of 1:10. After injection the oven was held at 50ºC for 3 min and then programmed to 280ºC at a rate of 5ºC/min and held for 3 min. The column ensemble consisted of a 10 m × 0.25 mm i.d. × 0.25 µm DB-5 column (J&W Scientific, Folsom, CA, USA) coupled in series using a press-fit connector to a 1 m × 0.1 mm i.d. × 0.1 µm BPX50 column (SGE, Ringwood, Australia) of which the last 27 cm are situated in the heated transfer line to the mass spectrometer. The carrier gas was helium (99.999%, Praxair, Oevel, Belgium) with the pressure linearly programmed from 191–300 kPa (absolute) during the temperature ramp of 50–280ºC. The time-of-flight mass spectrometer used was a Leco Pegasus II equipped with an electron ionisation (EI) ion source (Leco, St. Joseph, MI, USA). The ion source and transfer line temperatures were 200 and 280ºC, respectively. The multi-channel plate voltage was 1900 V and the EI energy 70 eV; mass spectra were recorded at a rate of 50 Hz over the mass range of 50–450 amu. Modulation was performed using a dual-stage 170 10. Peaklet clustering liquid CO2 jet modulator which is described elsewhere [168]. Generally, a modulation time of 6 s was used. System control and data recording were performed using the Pegasus (version 1.40, 8 Dec 2000) chromatography data system (CDS). The February 1998 version of the NIST (Gaithersburg, MD, USA) Mass Spectral Library database was used for mass spectral library matching. The following settings were used for data processing. Baseline track, default; baseline offset, 0.5; smoothing, 8; peak width, 1 s; number of library hits to return for each peak, 10; allowed mass of library hit, 40–500 amu; minimum abundance of base ion, 0; minimum similarity match before name is assigned, 600; area/height calculation, TIC; maximum number of peaks to find, 3000. Materials. A mixed standard (5 ng/µl) of 16 phenols and 18 anilines, both with various alkyl, halogen and nitro substituents, was prepared in ethyl acetate from the stock solutions of the pure standards obtained from various suppliers. Another, more complex, mixed standard was kindly provided by the National Institute for Inland Water Management and Waste Water Treatment (RIZA, Lelystad, the Netherlands) and diluted in ethyl acetate (5 ng/µl). It contained compounds that have been detected at least once in surface water in the Netherlands analysed regularly over a period of many years by the RIZA institute. Many of the compounds have several different functional groups and cannot easily be classified. For about 140 compounds that could be classified, their class and the number of compounds are shown in Table 10.1 Table 10.1: Examples of classes of compounds, individual representatives and number of compounds in each class in the test mixture* Class No. Examples Alcohols 14 alkyl alcohols, menthol, 2-phenoxyethanol Alkylbenzenes 18 trimethylbenzenes (3), n-butylbenzene, 1-methyl-4-tert-butylbenzene Alkylpyridines 12 methylpyridines (3), dimethylpyridines (5), 2-vinylpyridine Anilines 23 aniline, methylanilines (3), dichloroanilines (2), 4-methyl-2-nitroaniline Phenols 25 phenol, dimethylphenols (6), 2-tert-butyl-4-methoxyphenol, 2-chloro-5-methylphenol Glycol ethers 13 2-butoxyethanol, 2-(2-butoxyethoxy)ethyl acetate, (2-methoxymethylethoxy)propanol Heterocyclic aromatics 15 nicotine, benzothiazole, indole, acridine Polyaromatic hydrocarbons 10 indene, fluorene, phenanthrene, pyrene Triazines 8 atrazine, propazine, terbutryn * Numbers of isomers shown in brackets. To indicate the variety of compounds in each class, some representative examples are listed. For some compounds, several isomers are seen to be present (e.g. 6 dimethylphenols). Retention index data (for conventional 1D-GC) were available for all compounds. 171 10. Peaklet clustering 10.3.1 Tools used for manual identification/confirmation In order to check the results of the automated procedure, the locations of all compounds were confirmed and/or identified in a semi-automated way using the tools briefly described below. Manual inspection. When in 2DAid a spot in the 2D chromatogram is selected with the computer pointing device, the mass spectrum at that location is displayed together with its library hits. This allows visual inspection of the spectrum and expert decisions regarding the identity. Automated calculation of retention indices (RI). As mentioned before, the 1D-RIs of the compounds in the test mixture were available. They were used, amongst others, to accurately identify the isomers, which in some cases had almost identical spectra. The mixture contained a series of C9–C29 n-alkanes; after manual selection, their locations were stored and used for the automated calculation of the 1D-RIs for all peaks in the chromatogram. Ion ratio chromatograms. A selection method based on the ratios of m/z intensities in a single spectrum was implemented in 2DAid to allow the selective visualisation of peaks in a process similar to that described by Welthagen et al. [410]. Briefly, for each m/z in the spectrum, a range of relative intensities was specified to which the spectrum should comply before a signal is plotted. Usually two or three m/z values with their intensity ranges proved to be sufficient to build a selective filter. 10.4 Results and discussion 10.4.1 Carrier gas flow rate and second-dimension hold-up time Table 10.2 shows the results of the flow calculations at 50 and 280ºC, that is at the start and end of the oven temperature programme. The total flow (converted to standard conditions, 25ºC, 1 atm) varies between 0.8 and 1.1 ml/min in the course of the temperature programme. The key parameter for which the calculation was made is the hold-up time in the second-dimension column (segment 2). This is seen to vary between 0.3 and 0.5 s, and these values will be used below to make further calculations. The parameter values found with our calculations were also compared with values obtained with the Leco GC×GC Column Calculator (software that is supplied with later versions of the Leco ChromaTOF software), and with results described by Beens et al. [411], and were found to be in close agreement. Some other results of the calculations are of interest as well. First among these is the large pressure drop across the piece of column located within the MS transfer line. This 172 10. Peaklet clustering is caused by its considerable length, narrow diameter and the higher viscosity of the carrier gas at the high temperature of the transfer line. The pressure drop across this segment alone is about 2–3 times larger than along the second-dimension column inside the oven, which shows the importance of treating it as a separate column segment when calculating flows. Another interesting result are the linear velocities of the carrier gas; although these were not specifically optimised, they show a good overlap with the optimum value ranges [411;412]. Table 10.2: Calculated pressures, linear speeds and hold-up times in the various column segments a b Column segment T l d pi po ū Flow tm (ºC) (m) (mm) (kPa) (kPa) (cm/s) (ml/min) (s) Oven 50ºC 1 50 10 0.25 191 175 16 0.78 62.7 2 50 0.6 0.1 175 127 120 0.78 0.5 3 280 0.27 0.1 127 0 369 0.78 0.07 Oven 280ºC 1 280 10 0.25 300 265 24 1.06 41.9 2 280 0.6 0.1 265 147 199 1.06 0.3 3 280 0.27 0.1 147 0 429 1.06 0.07 a. Segments 1 and 2, first- and second-dimension columns; segment 3, transfer capillary, invariably held at 280ºC. b. Converted to flow at 1 atm and 25ºC; pressures are absolute. 10.4.2 Calculated and experimental 2D peak shape As described in Section 10.2.1, the 2D peak shape is calculated according to Eq. 10.6. T 2 is known from the temperature programme, tm has been calculated above and is in the range of 0.3–0.5 s. For the only variable left, C, a value has to be chosen to enable solving Eq. 10.6. Literature values vary between 20 and 8 [408;409], and the initial values tested were around the average of 14. The influence of this parameter on the calculated peak shape was studied by varying its value and visually evaluating the match between experimental and calculated peak shapes. By using 2D peaks that show considerable tailing in the first dimension, the calculated and experimental peak shapes can be compared over an extended time window in the first dimension. Therefore, for this comparison a standard containing several phenols and anilines (5 ng/µl each) was used. Fig. 10.2a shows the selected ion chromatogram of m/z 65, representing the + cyclopentadienyl ion C5H5 present in the spectra of some of the test analytes in the mixture. In this plot the level at which a signal results in coloured pixels is set very low; the 2D peaks are therefore visible over a wide range in the first dimension, even though the signal intensities decrease strongly after the first few peaklets. This is illustrated by the (relative) per cent heights of the peaklets of 4-bromo-aniline that are indicated in Fig. 173 10. Peaklet clustering 10.2a, and also by comparing with Fig. 10.2b, which shows the same selected ion chromatogram using conventional plot settings. 2 tr (s) 5.60 a. 4-Nitro-aniline b. 2-Nitro-aniline 4.80 4-Bromo-aniline 4.00 3.20 2.40 100% 1.60 0.9% 0.2% 0.80 2-Chloro-phenol 0.04% 0.00 200 600 1000 1400 1800 2200 2600 st 1 1 dimension retiontion time ( tr) (s) Fig. 10.2: a. High-contrast plot of extracted-ion chromatogram (m/z 65) of a mixture of phenols and anilines. 2 The paired red lines show the calculated peak shapes for tm=0.4 s and C=13 for three selected analytes. For 4-bromo-aniline, the heights of several peaklets are indicated as percentage of the master peaklet. b. The chromatogram of m/z 65 plotted using conventional plot settings. 2 Using chromatograms such as shown in Fig. 10.2a, the effect of the values for C and tm on the fit of the peak shapes was studied. The match between the calculated and experimental peak shapes was found to be best with values of 0.4 s (the average of the 2 calculated values) for tm, and 13 for C. With these values, the calculated peak shapes match the actual peak shapes of the selected compounds over at least 15 min in the first dimension. This is amply sufficient since by far the largest part of the total mass of a compound elutes in the first 3–5 peaklets. Most rewardingly, the calculated peak shapes show a good match independent of their position in the 2D chromatogram even though the angles of the slopes which depend on the position in the chromatogram are very different1. Similar results were obtained for all other anilines and phenols. 2 In order to show the effect of varying tm and C values in some more detail, the match for the first fifty peaklets of the 2-nitro-aniline peak was studied with values for C 2 and tm deviating from the optimal values. The influence of C is demonstrated in Fig. 10.3a and b. Even with a deviation of 2 from the optimum of –13, the first fifty peaklets 2 lie within the predicted area. The same holds true for tm: even for values outside the calculated range of 0.3–0.5 s, the first fifty peaklets are all within the calculated area (c). 1 Earlier versions for the grouping algorithm, which are not described here, were based on linear equations with parameters depending on the position in the chromotogram. None of these methods could encompass the peak shapes as well as the algorithm based on chromatographic principles. 174 10. Peaklet clustering 4.60 2 C= C= tm= -13 -13 0.4 s 4.20 -11 -15 0.6 s -9 -17 0.2 s 2 2 3.80 tm= 0.4 s tm= 0.4 s C= - 13 3.40 3.00 2.60 2.20 1.80 a b c 1.40 770 950 1130 770 950 1130 770 950 1130 2 Fig. 10.3: Influence of variation of C and tm on the accuracy of the predicted 2D peak shape, with values 2 below and above the optimal values (red) for C and tm; extracted ion chromatogram (m/z 138) of a mixture of phenols and anilines showing the peak of 2-nitro-aniline. 2 It can be concluded that with fixed values for tm and C, one can reliably calculate the 2D peak shapes for compounds in different regions of a 2D chromatogram; relatively large deviations from the optimal values do not seriously affect the match in terms of predicted peak shape, irrespective of the angle of the leading edge of the 2D peak of interest. 10.4.3 Applications Example No. 1. As a first application of the entire clustering and filtering procedure, the mixture of 34 anilines and phenols, some of which were used above to demonstrate the predicted and experimental peak shapes, was studied. 1D data processing was performed using the Pegasus II software, with the maximum number of peaks to find set at 2000 and a minimum required S/N of 502. The other parameters were as reported in Section 10.3. The results so obtained were subsequently post-processed with 2DAid. If processing was limited to using the clustering method, the 2000 peaklets were now reported as 490 2D peaks, a reduction of more than 75%. These 490 peaks are shown in Fig. 10.4 which shows the visual output of 2DAid. The ‘contours’ of the 2D peaks show up as blue lines and the peak apices are indicated by green rectangles. Actually, Fig. 10.4a is still very crowded, and 490 peaks is over 10-fold more than expected with only 34 compounds injected. Actually, most of these peaks are due to (chemical) noise and 2 During the 1D chromatographic deconvolution process, the Pegasus software determines a unique m/z for each peaklet to distinguish them from background and adjacent peaklets; this m/z is also used to calculate the S/N. In this way numerous peaklets with very low intensity and no visible signal in the total ion chromatogram are detected. 175 10. Peaklet clustering are not visible in the total ion chromatogram (cf. earlier footnote). To further reduce the number of 2D peaks, additional filters were applied. 5.6 a. 5.6 b. 4.8 4.8 4.0 4.0 3.2 3.2 2.4 2.4 1.6 1.6 0.8 0.8 0.0 0.0 200 600 1000 1400 1800 2200 2600 200 600 1000 1400 1800 2200 2600 Fig. 10.4: Apex plots of the standard containing 34 anilines and phenols. a. Plot containing 490 clusters, and their contours, after applying only the cluster step. b. 40 apices remaining after applying additional filters, with the 34 anilines and phenols indicated by red dots, and the 6 other compounds by black rectangles. In the present example, with amounts of 5 ng injected of each compound, the most intense peaks most likely are the target analytes. When the lower peak area limit was set at 1% of the highest peak, 92 of the initial 490 2D peaks were reported and, with a limit of 4%, only 56 peaks were reported. After filtering out some solvent impurities and column bleed compounds by using the identity filter, 40 compounds were reported; their apices are shown in Fig. 10.4b. In other words, the filtering operation is markedly successful and the peak list is now reduced to a manageable length. Actually, all 34 target compounds were found to be present in the final 2D peak list. Among the six other compounds were common contaminants such as phthalates, and a few unknowns which most likely are impurities from the standards or reaction products formed in the mixed standard. Example No. 2. To further investigate the efficiency of clustering and filtering in 2DAid, the method was next applied to a complex mixture containing 362 analytes belonging to a wide variety of compound classes (see Table 10.1). For this application, the same strategy was used as for Example No. 1, with a maximum number of peaklets of 3000. When only clustering was applied, the 3000 peaklets were reduced to about 1400 (2D) peaks. This about 50% reduction can be called satisfactory but, admittedly, it is less than the 75% reduction obtained with the phenols and anilines. This is primarily due to the fact that, on average, the percentage of compounds that exhibit strong tailing in the first dimension is smaller than with the 176 10. Peaklet clustering phenols and anilines. About 20% of the compounds were found as a single peaklet, 70% as clusters of 2–4 peaklets, and the rest as clusters of 5–12 peaklets. Although one or two modulations across the width of 1D peak does not meet the modulation criterion [413;414] of a true multidimensional separation, it is often the consequence of practical considerations such as (i) the total analysis time and (ii) the fact that the number of compounds showing wrap-around should be limited because they complicate the data analysis. For a general screening analysis of a sample containing compounds with a large boiling point and polarity range, it is customary to use temperature programmed GC in order to perform the analysis within one hour in order to achieve a sufficiently high sample throughput. 2D peak shapes acquired under these conditions are not bilinear, all peaklets of one 2 compound do not have the same tm. However, also under optimal conditions, when bilinearity is met, the calculated peak shape will fit the experimental 2D peak shape. This is substantiated by the almost horizontal calculated peak shape for the solvent peak in the bottom left of Fig. 10.4. Moreover, under optimal conditions, when the modulation criterion is met, the clustering step will cause a much larger reduction from the peaklet to the 2D peak list since each 2D peak will comprise more peaklets. In practice, a balance has to be found between the total run-time, the number of modulations across the 1D peaks, and the number of wrapped-around compounds. Therefore, in many (published) wide-screening analytical GC×GC studies, the modulation criterion is not met, at least for a part of the compounds. Theoretically this could form a problem for specific pairs of peaks, but the overall separation of all sample constituents, especially when using MS detection, is generally better compared to 1D- GC. Table 10.3: Data reduction using different filters* Filter No. of peaks LECO data processing 3000 (peaklets) Clustering only 1400 Limit on area (>0.4% of area of most abundant peak) 800 Solvent and column bleed filter 750 Match factor filter: Sim>600 550 Sim>700 450 * For explanation, see text. After applying the clustering, the filters were applied. When a lower limit of >0.4% of the area of the most abundant peak was applied, the number of peaks was reduced to about 800. Excluding compounds to be attributed to column bleeding, and solvent peaks, caused a further reduction to around 750 peaks. A very substantial reduction was obtained by introducing mass-spectrometric Similarity (Sim) match factor filtering. With a, rather modest, threshold value of 600, around 550 peaks remained. This figure was 177 10. Peaklet clustering reduced to 450 peaks when a more realistic Sim match factor of 700 was applied. The results are summarised in Table 10.3. In order to evaluate the results of the automated clustering, the 2D positions of all 362 compounds in the standard mixture were interactively determined using the partially automated MS- and RI-based methods introduced in Section 10.3.1, and compared with the 450 peaks in the final list. From the 362 target compounds, 350 were found to be present in the final peak list. As regards the remaining twelve analytes, all of these were difficult to locate because of low ion intensities, although half of them could be properly located manually. Interestingly, with a search area of plot line ± 0.1 s in vertical direction, there was not even one problem caused by the merging of two or more compounds into a single 2D peak. Such problems only started to occur for margins of ± 0.3 s or wider. An important aspect of the proposed strategy is that clustering and filtering are performed as a post-processing, and are completed in 10–20 s on a normal desktop computer. The advantage is that the first step of 1D peak processing in the Pegasus software, which is a complex and time-consuming process (30–60 min), has to be performed only once. Because of this, it is no problem to use parameter settings which cause a large number of low-intensity peaklets also to be integrated and reported. Then, the abundance of data produced, is subjected to the rapid post-processing and the parameters are easily optimised to come to a final peak list that contains a manageable number of 2D peaks. 10.5 Conclusions For temperature-programmed GC×GC, the use of chromatographic laws describing the relation between the capacity factor of an analyte and the enthalpy and entropy factors enables accurate prediction of the 2D peak shapes using the position of the master, i.e. the most intense, peaklet as a starting point. Rather large deviations from the optimal values of the relevant parameters have no adverse effect on the prediction of the relevant initial part of the 2D peaks. Consequently, the proposed procedure, 2DAid, can be applied to real-life data, which can be subject to variations of the flow velocity during the run, and variation of compound-specific thermodynamic properties. The clustering of individual peaklets based on their tentative identity found by mass- spectral library searching within the area of the predicted peak shape results in the automated determination of the 2D peak apices and a significant reduction of the amount of data. Subsequent filtering on the basis of relative signal intensities, elimination of column bleed and organic solvent peaks, and mass-spectral Sim matching further reduces the peak list to a manageable size. Two examples—one of which involves several hundreds of non-target analytes—demonstrate the practicality of the proposed procedure. 178 11 11. GC×GC–ToF-MS using various sets of column combinations for the non-target screening of 150 micro-contaminants detected in surface water 11.1 Introduction In a recent study of our group (Chapter 10) [352], a method was presented to facilitate the non-target analysis of data generated in comprehensive two-dimensional gas chromatography with time-of-flight mass spectrometric detection (GC×GC–ToF-MS). Two examples were given to demonstrate the potential of the proposed approach, called 2DAid. In this communication, the strategy (Pegasus-cum-2DAid) is used to facilitate data analysis in the evaluation of the potential of different column, i.e. stationary-phase, combinations for the separation of a wide variety of chemical compounds. Optimisation of the stationary phase combination is described in the published literature for a number of analytical challenges. However, most such studies deal with one or a few specific groups of analytes (e.g. [415;416]) and only a few (e.g. [417]) describe the differences in performance of such combinations for chemicals of (widely) different classes. In the present study, a non-polar DB-1 first-dimension column was combined with five different second-dimension columns to study the separation of compounds regularly encountered during the monitoring of surface water. In order to create a close-to-real-life situation, the 362-compound test mixture of our previous study [352] was used again. All compounds in this mixture were detected at least once in surface waster in the Netherlands, analysed over a period of many years by government laboratories. Many of the organic micro-contaminants in the mixture have several different functional groups and cannot easily be classified. Obviously, they were less suitable candidates for the present, chemical-class-separation-based study. Some 140–150 for which classification did not create problems were therefore used as target analytes. Classes of such compounds abundantly present in the mixture included alkylpyridines, polyaromatic hydrocarbons, alcohols, anilines, phenols, glycol ethers, alkylbenzenes and various types of heterocyclics. 11. Column combinations Experimental details of the GC×GC–ToF-MS and a brief summary of the 2DAid- based data handling and interpretation procedure are given in the Experimental section. For a more detailed discussion of the procedure, and the occasional use of additional identification tools, the reader should consult [352]. 11.2 Experimental All analyses were performed on a GC×GC system consisting of a HP6890 gas chromatograph equipped with an HP7673 autosampler (Agilent Technologies, Palo Alto, CA, USA) and a dual-stage liquid CO2 jet modulator. Modulation times of 6–10 s were used. An Optic II programmable injector (ATAS, Veldhoven, the Netherlands) fitted with a multi-capillary liner was used for hot-split 1-µl injections (split ratio, 1:10) at 260°C. For all studies, a 15 m × 0.25 mm i.d. × 0.25 µm df DB-1 first-dimension column was used, which was combined with 1 m × 0.1 mm i.d. second-dimension columns. The five second-dimension columns used were BPX35, BPX50, HT8, BP20 and BPX70 (SGE, Ringwood, Australia). The film thickness was 0.1 µm for all phases except BPX70, which had a 0.2 µm coating. The column-head pressure was regulated using the electronic pressure control of the HP6890; the column ensembles were operated in the constant-flow mode using helium (99.9999%; Praxair, Oevel, Belgium). Since the firmware is not designed to handle two columns of different dimensions in series, the resulting flow deviated from the programmed flow of 1.3 ml/min; the actual flows are reported in [352]. After injection, the GC oven was held for 3 min at 50C and then programmed to 280C at a rate of 5°C/min, with a final 3-min hold. Detection was done with a Pegasus II time-of-flight mass spectrometer equipped with an electron-ionisation (EI) ion source (Leco, St. Joseph, MI, USA). The multi- channel plate voltage was 1900 V and the EI energy, 70 eV. Mass spectra were recorded at a rate of 50 Hz over the mass range 50–400 Da, using Pegasus software. The ion source and transfer line temperatures were 200 and 280°C, respectively. For mass-spectral-based library searching, a set comprising only the spectra of the compounds in the test mixture was selected from the February 1998 version of the NIST (Gaithersburg, MD, USA) Mass Spectral Library database. The mass spectra of a small number of compounds were not present in this database; for these compounds, the spectra recorded during the present study were added. The Pegasus software uses the NIST search algorithm to match spectra of unknowns to those in the library. This algorithm returns three qualifiers, Similarity, Reversed and Probability, to indicate the degree to which the spectrum of an unknown and a library spectrum match. Limits are set on these qualifiers by both the Pegasus and the 2DAid software to filter out low- quality matches. 2DAid reduces the final list of peaks by clustering the peaklets originating from one compound and by applying additional filters. Briefly, the highest 180 11. Column combinations peaklet, the master peaklet, of a 2D peak is selected and from its retention time and the oven temperature a thermodynamic variable is calculated, which in turn is used to predict the positions of any other peaklets due to this compound. If peaklets at these positions match the identity of the master peaklet, they are clustered together into a single 2D peak. The mixed standard was kindly provided by the National Institute for Inland Water Management and Waste Water Treatment (RIZA, Lelystad, the Netherlands). A dilute solution of 5 ng/µl (per compound) was prepared in ethyl acetate. For about 140 compounds that could be classified, their class and the number of compounds are shown in Table 11.1. To indicate the variety of compounds in each class, some representative examples are listed. For some compounds, several isomers are seen to be present (e.g. 6 dimethylphenols). Retention index data (for conventional 1D-GC) were available for all compounds. Table 11.1: Examples of classes of compounds, individual representatives and number of compounds in each class in the test mixture* Class No. Examples Alcohols 14 alkyl alcohols, menthol, 2-phenoxyethanol Alkylbenzenes 18 trimethylbenzenes (3), n-butylbenzene, 1-methyl-4-tert-butylbenzene Alkylpyridines 12 methylpyridines (3), dimethylpyridines (5), 2-vinylpyridine Anilines 23 aniline, methylanilines (3), dichloroanilines (2), 4-methyl-2-nitroaniline Phenols 25 phenol, dimethylphenols (6), 2-tert-butyl-4-methoxyphenol, 2-chloro-5-methylphenol Glycol ethers 13 2-butoxyethanol, 2-(2-butoxyethoxy)ethyl acetate, (2-methoxymethylethoxy)propanol Heterocyclic aromatics 15 nicotine, benzothiazole, indole, acridine Polyaromatic hydrocarbons 10 indene, fluorene, phenanthrene, pyrene Triazines 8 atrazine, propazine, terbutryn * Numbers of isomers shown in brackets. 11.3 Results and discussion 11.3.1 Initial studies In most GC×GC studies, a non-polar first-dimension column is combined with a (semi-) polar or shape-selective second-dimension column. Since a large majority of the GC- amenable constituents of most environmental samples is of a non-polar or only slightly polar nature, this conventional set-up helps to create an efficient, boiling-point-based first separation—which is combined with an essentially independent, functionality-based second separation. This has the dual advantage that a large part of the GC×GC space will be available for the separation of interest, and ordered structures frequently show up for groups of related compounds, which facilitates tentative identification of unknowns. In line with this approach, a non-polar DB-1 stationary phase was selected for the first-dimension separation. From among the five second-dimension columns, BPX35 and 181 11. Column combinations BPX50 are medium-polar phases similar to 35 and 50% phenylsiloxane columns, respectively. Phenyl phases have a weak to moderate hydrogen bonding capacity, basically show no dipole interactions, and - interactions between phenyl groups of the solute and the stationary phase play a distinct role. HT8 has a polarity comparable to 8% phenylsiloxane, and is a specialty phase with a boron-modified backbone. It is mainly used for the analysis of polychlorinated biphenyls since it has the unique property to preferentially interact with ortho-substituted congeners [418]. The BPX70 stationary phase is very polar and consists of polysilphenylene-siloxane with 70% cyanopropyl. Its hydrogen bonding capacity is moderate but it exhibits strong dipole interactions. BP20 is a polyethylene glycol column which shows strong hydrogen-bonding interaction, which causes high retention for e.g. alcohols. As Table 11.2 indicates, the 1D retention indices (RIs) of five analytes from several different classes of compounds are up to 400–500 units higher on the second-dimension stationary phases selected for our study, than on a non-polar BP1 (closely similar to DB-1) phase. Table 11.2: RI data of five compounds representing different chemical classes* Stat. Benzene Butanol 2-Pentanone Nitropropane Pyridine phase BP1 647 646 666 707 722 HT8 680 673 728 796 780 BPX35 728 726 763 862 848 BP20 947 1153 998 1217 1185 BPX70 1067 1219 1170 1365 1300 * SGE product information. The first experimental results showed that, with a modulation time of 8 s, the number of peaks showing wrap-around was very high when using the polar BPX70 or BP20 stationary phases in the second dimension. This phenomenon can easily complicate data interpretation because compounds now do not elute during their own but, instead, during (a) later modulation cycle(s). Further experiments showed that a modulation time of 25 s would be required to prevent most of the wrap-around for these two columns. However, with such a long modulation time, the resolution effected during the first-dimension separation will be largely lost. Therefore, the analyses carried out on the DB1×BPX70 and the DB1×BP20 column sets were performed using a GC×GC system housing a small oven to separately heat the second-dimension column (for practical information on such a system, see e.g. [245;281]). By setting the temperature of this oven 25ºC above the main-oven temperature and using a modulation time of 10 s, the degree of wrap-around was significantly reduced. 182 11. Column combinations 11.3.2 Separations on different column combinations For the comparison of the separation potential of the five column combinations, the combined Pegasus-cum-2DAid strategy was used for the 150 target analytes, which represented thirteen chemical classes. For the convenience of the reader, the experimental results shown in Fig. 11.1a–d are presented as apex plots, with the coloured symbols indicating the different classes of target compounds, and grey dots the 200-odd (multi-functional) non-classified compounds. Fig. 11.1a–d shows the apex plots of only four out of the five combinations studied. Although some small differences could be observed between the DB1×BPX35 and the DB1×BPX50 column combinations, the general distributions of the test compounds over the 2D plane were very similar for these two sets; therefore, no plot is shown for the former one. In all plots, the horizontal axes show the 1D-RI values, with an upper limit of 2350. The test mixture did contain a few alkanes and several other compounds eluting up to an RI of 2900, but almost all compounds had RI < 2300. As for the vertical axes, for BPX50 and HT8 they correspond with the modulation times used. For BP20 and BPX70, for which the plots were recorded using a 10-s (instead of an 8-s) modulation time, the axes are extended to allow plotting of compounds displaying wrap-around. Wrap-around was recognised during manual inspection on the basis of unduly high second-dimension peak widths; such peaks were plotted with the modulation time added to the experimental second-dimension retention time. Even with the relatively limited number of classes of compounds plotted in colour in Fig. 11.1, the four chromatograms are seen to be extremely crowded, which adversely affects visualisation. However, the plots do provide a general impression of the efficiency with which the 2D space is used. It is obvious that with BPX70 and, even more so, BP20, as second-dimension column, the available separation space is used more fully than with the other three columns tested (as mentioned above, the results obtained with BPX35 were closely analogous to those found with BPX50). One may tentatively conclude that for samples containing a wide-ranging—and largely unknown—mixture of compounds with many different functionalities, the use of DB1×BP20 or DB1×BPX70 column sets is a recommendable strategy. In addition, these two column combinations seem to provide a better separation for volatile compounds. Going beyond such a statement by discussing a variety of specific group-type separations is, to our opinion, of little value in the general ‘mixture-of-unknowns’ situation. Such an approach should be pursued only when a few particular classes of compounds are of key interest, as is, for example, encountered in studies of organohalogen micro-contaminants [165;419]. 183 11. Column combinations 8 14 a. BPX50 b. BPX70 7 12 6 10 5 8 4 dim. ret. time(s) dim.ret. 6 nd 3 2 4 2 1 2 0 0 550 1050 1550 2050 550 1050 1550 2050 1st dim. RI 16 6 c. BP20 d. HT8 14 5 12 4 10 8 3 6 2 4 1 2 0 0 550 1050 1550 2050 550 1050 1550 2050 Legend Alcohols PAHs Other aromatic based compounds Alkylbenzenes Phenols Alkanes Alkylpyridines Triazines Nitro-containing compounds Glycol ethers Anilines Not classified Ketones Heterocyclic aromatic compounds Fig. 11.1: GC×GC apex plots of 13 classes of compounds present in the test mixture: DB1×second-dimension column sets indicated—grey dots represent all non-classified compounds. Broken lines indicate wrap-around correction (see text). The visual inspection discussed above provides a satisfactory preliminary comparison of the various column sets. However, applying a numerical method to semi-quantitatively evaluate the separation efficiency (in terms of either resolution or peak overlap) is, of course, to be preferred. Fairly sophisticated methods have been reported in the literature [420-422]. Unfortunately, because of the temperature-programming nature of essentially all GC×GC analyses, and the resulting sloping peak shapes, rather complicated mathematical operations are required to calculate the (total) degree of overlap. We therefore used another, simpler, approach to calculate the total overlap of all 2D peaks in a chromatogram (also see Giddings’ paper on optimum separation [423]). 184 11. Column combinations When keeping in mind that under conventional temperature-programmed experimental conditions, 2D peaks usually consist of only two or three peaklets and ‘stepped’ or discrete signals rather the well-formed elliptical spots show up [421;422], it is justified to present these peaks as rectangles with roughly the size of the experimentally observed peaks and positioned parallel to the time axes. In order to be on the safe side, first- × second-dimension dimensions of 24 × 0.4 s2 were selected. The total overlap was calculated by summing the overlap (in s2) of all pairs of rectangles, an operation that takes only a few seconds. The results which are shown in Table 11.3 confirm the preliminary assessment based on visual inspection of the GC×GC chromatograms of Fig. 11.1. By far the best use of the 2D space is found for the DB1×BP20 combination, with DB1×BPX70 in second place. BPX50 and BPX35 are seen to yield virtually identical results, which underscores our earlier decision not to include the latter phase in Fig. 11.1. Finally, the DB1×HT8 column set clearly shows the poorest performance of all combinations tested. Table 11.3: Overlap calculation using rectangle strategy 2nd-dim. column Overlap (s2) BP20 460 BPX70 660 BPX50 860 BPX35 880 HT8 1350 One note of warning should be added. It is clear that the DB1×BP20 and DB1×BPX70 column sets can be considered good first choices for the analysis of complex GC- amenable mixtures of unknowns. If, however, high-boiling compounds play a major role, the upper temperature limit of 260ºC of these columns may cause problems. Both columns have to be used with a temperature offset of some 20ºC above the main oven temperature. There is, therefore, a temperature limit of about 240ºC for the first- dimension separation. 11.4 Conclusions Analysis of more or less unspecified 'mixtures of unknowns' is a challenging task. One aspect of interest when devising a proper GC×GC-based strategy will be to select a proper column combination, specifically a suitable second-dimension column. The present study, which uses some 150 micro-contaminants as test compounds, shows that applying the recently proposed 2DAid approach (for data handling and interpretation) and a simple semi-quantitative method for the calculation of peak overlap provides useful results. This indicates that further testing and evaluation of the approach should be performed. 185 11. Column combinations As regards the screening of the five column sets, combining a polar rather than a medium-polar second-dimension column with a non-polar first-dimension column is the preferred approach when mixtures of unknowns have to be analysed. Such a GC×GC system should comprise an extra oven for the second column to avoid too much wrap- around occurring. 186 12 12. References [1] M. Adahchour, Design, optimization and use of GC×GC systems for food analysis, Ph.D. Thesis, VU University Amsterdam, The Netherlands, 2006. [2] A.J. McCormack, S.C. Tong, W.D. Cooke, Anal. Chem. 37 (1965) 1470. [3] C.A. Bache, D.J. Lisk, Anal. Chem. 37 (1965) 1477. [4] C.I.M. Beenakker, Spectrochim. Acta. Part B 31 (1976) 483. [5] C.I.M. Beenakker, Spectrochim. Acta. Part B 32 (1977) 173. [6] B.D. Quimby, P.C. Uden, R.M. Barnes, Anal. Chem. 50 (1978) 2112. [7] P.C. Uden, J. Chromatogr. A 703 (1995) 393. [8] R. Lobinski, F.C. Adams, Spectrochim. Acta. Part B 52 (1997) 1865. [9] I. Rodriguez Pereiro, A. Carro Díaz, Anal. Bioanal. Chem. 372 (2002) 74. [10] J.C.A. Wuilloud, R.G. Wuilloud, A.P. Vonderheide, J.A. Caruso, Spectrochim. Acta. Part B 59 (2004) 755. [11] S. Pedersen-Bjergaard, T. Greibrokk, J. High Resolut. Chromatogr. 19 (1996) 597. [12] H. Bagheri, M. Saraji, U.A.Th. Brinkman, in D. Barceló (Editor), Sample handling and trace analysis of pollutants. Elsevier, Amsterdam, the Netherlands, 1999, Ch. 5. [13] L.L.P. van Stee, U.A.Th. Brinkman, H. Bagheri, Trends Anal. Chem. 21 (2002) 618. [14] J.T. Andersson, Anal. Bioanal. Chem. 373 (2002) 344. [15] P.L. Wylie, R. Oguchi, J. Chromatogr. 517 (1990) 131. [16] M. Linkerhägner, H.-J. Stan, Dtsch. Lebensm.-Rundsch. 89 (1993) 143. [17] H.-J. Stan, M. Linkerhägner, J. Chromatogr. A 750 (1996) 369. [18] S.M. Lee, P.L. Wylie, J. Agric. Food Chem. 39 (1991) 2192. [19] K.C. Ting, P. Kho, J. Assoc. Offic. Anal. Chem. 74 (1991) 991. [20] Z.V. Skopec, R. Clark, P.M. Harvey, R.J. Wells, J. Chromatogr. Sci. 31 (1993) 445. [21] H.-J. Stan, M. Linkerhägner, J. High Resolut. Chromatogr. 16 (1993) 539. [22] M. Linkerhägner, H.-J. Stan, Z. Lebensm. Unters. Forsch. 198 (1994) 473. [23] J.J. Jiménez, J.L. Bernal, J. Atienza, Chromatographia 42 (1996) 130. [24] J.J. Jiménez, J.L. Bernal, M.J. Delnozal, L. Toribio, J. Atienza, J. High Resolut. Chromatogr. 20 (1997) 81. [25] J.J. Jiménez, J.L. Bernal, L. Toribio, M.J. del Nozal, M.T. Martín, J. Chromatogr. A 946 (2002) 247. [26] J. Cook, M. Engel, P. Wylie, B. Quimby, J. Assoc. Offic. Anal. Chem. 82 (1999) 313. [27] N. Campillo, R. Peñalver, N. Aguinaga, M. Hernández-Córdoba, Anal. Chim. Acta 562 (2006) 9. [28] K.S. Howell, J.H. Swiegers, G.M. Elsey, T.E. Siebert, E.J. Bartowsky, G.H. Fleet, I.S. Pretorius, M.A. de Barros Lopes, FEMS Microbiol. Lett. 240 (2004) 125. [29] N. Campillo, N. Aguinaga, P. Viñas, I. López-García, M. Hernández-Córdoba, J. Chromatogr. A 1061 (2004) 85. [30] D. Deruaz, F. Soussan-Marchal, I. Joseph, M. Desage, A. Bannier, J.L. Brazier, J. Chromatogr. A 677 (1994) 345. [31] C. Gerbersmann, R. Lobinski, F.C. Adams, Anal. Chim. Acta 316 (1995) 93. [32] H.B. Swan, J. Food Comp. Anal. 13 (2000) 207. [33] D.W. Baloga, G.A. Reineccius, J.W. Miller, J. Agric. Food Chem. 38 (1990) 2021. 12. References [34] S. Junyapoon, A.B. Ross, K.D. Bartle, B. Frere, A.C. Lewis, M. Cooke, J. High Resolut. Chromatogr. 22 (1999) 47. [35] R. Eisert, K. Levsen, G. Wunsch, J. Chromatogr. A 683 (1994) 175. [36] E.C. Goosens, D. de Jong, G.J. De Jong, F.D. Rinkema, U.A.Th. Brinkman, J. High Resolut. Chromatogr. 18 (1995) 38. [37] Th. Hankemeier, A.J.H. Louter, F.D. Rinkema, U.A.Th. Brinkman, Chromatographia 40 (1995) 119. [38] I. Silgoner, E. Rosenberg, M. Grasserbauer, J. Chromatogr. A 768 (1997) 259. [39] N. Campillo, P. Viñas, I. López-García, N. Aguinaga, M. Hernández-Córdoba, J. Chromatogr. A 1035 (2004) 1. [40] N. Campillo, P. Viñas, I. López-García, N. Aguinaga, M. Hernández-Córdoba, Talanta 64 (2004) 584. [41] N. Campillo, N. Aguinaga, P. Viñas, I. López-García, M. Hernández-Córdoba, Anal. Chim. Acta 552 (2005) 182. [42] R. Gotz, O.H. Bauer, P. Friesel, K. Roch, Chemosphere 36 (1998) 2085. [43] F.D. Rinkema, A.J.H. Louter, U.A.Th. Brinkman, J. Chromatogr. A 678 (1994) 289. [44] S. Mothes, P. Popp, R. Wennrich, Chromatographia 57 (2003) S249-S252. [45] N.L. Olson, R. Carrell, R. Cummings, R. Rieck, S. Reimer, J. Assoc. Offic. Anal. Chem. 78 (1995) 1464. [46] I. Rodriguez, M.I. Turnes, M.C. Mejuto, R. Cela, J. Chromatogr. A 721 (1996) 297. [47] A. Agüera, M. Mezcua, D. Hernando, S. Malato, J. Cáceres, A. Fernández-Alba, Int. J. Environ. Anal. Chem. 84 (2004) 149. [48] P. Viñas, N. Campillo, I. López-García, N. Aguinaga, M. Hernández-Córdoba, J. Chromatogr. A 978 (2002) 249. [49] L. von Sydow, H. Bordn, A. Grimvall, Chemosphere 39 (1999) 2479. [50] A. Wenner, M. Wortberg, J. High Resolut. Chromatogr. 21 (1998) 661. [51] D.B. Hooker, J. DeZwaan, Anal. Chem. 61 (1989) 2207. [52] Th. Hankemeier, J. Rozenbrand, M. Adahchour, J.J. Vreuls, U.A.Th. Brinkman, Chromatographia 48 (1998) 273. [53] H.G.J. Mol, Th. Hankemeier, U.A.Th. Brinkman, LC-GC Int. 12 (1999) 108. [54] L.L.P. van Stee, P.E.G. Leonards, R.J.J. Vreuls, U.A.Th. Brinkman, Analyst 124 (1999) 1547. [55] L.L.P. van Stee, P.E.G. Leonards, W.M.G.M. van Loon, A.J. Hendriks, J.L. Maas, J. Struijs, U.A.Th. Brinkman, Water Res. 36 (2002) 4455. [56] G.W. Gribble, Chem. Soc. Rev. 28 (1999) 335. [57] H. Frischenschlager, C. Mittermayr, M. Peck, E. Rosenberg, M. Grasserbauer, Fresenius J. Anal. Chem. 359 (1997) 213. [58] H. Frischenschlager, M.L. Peck, C. Mittermayr, E. Rosenberg, M. Grasserbauer, Fresenius J. Anal. Chem. 357 (1997) 1133. [59] K. Laniewski, H. Borén, A. Grimvall, Environ. Sci. Technol. 32 (1998) 3935. [60] K. Laniewski, H. Borén, A. Grimvall, M. Ekelund, J. Chromatogr. A 826 (1998) 201. [61] K. Laniewski, H. Borén, A. Grimvall, Chemosphere 38 (1999) 393. [62] S. Pedersen-Bjergaard, T.N. Asp, J. Vedde, G.E. Carlberg, T. Greibrokk, Chromatographia 35 (1993) 193. [63] R.J.B. Peters, E.W.B. de Leer, J.F.M. Versteegh, J. Chromatogr. A 686 (1994) 253. [64] J. Pörschmann, F.D. Kopinke, M. Remmler, K. Mackenzie, W. Geyer, S. Mothes, J. Chromatogr. A 750 (1996) 287. [65] J. Pörschmann, F.D. Kopinke, G. Balcke, S. Mothes, J. Microcol. Sep. 10 (1998) 401. [66] F. Traulsen, J.T. Andersson, M.G. Ehrhardt, Anal. Chim. Acta 392 (1999) 19. [67] S. Pedersen-Bjergaard, S.I. Semb, J. Vedde, E.M. Brevik, T. Greibrokk, Chemosphere 32 (1996) 1103. [68] S. Pedersen-Bjergaard, S.I. Semb, J. Vedde, E.M. Brevik, T. Greibrokk, Chromatographia 43 (1996) 44. [69] J.L. Bernal, M. del Nozal, M.T. Martin, J.J. Jiménez, J. Chromatogr. A 754 (1996) 245. [70] P. Viñas, N. Campillo, I. López-García, N. Aguinaga, M. Hernández-Córdoba, J. Agric. Food Chem. 51 (2003) 3704. 188 12. References [71] C. Flodin, E. Johansson, H. Borén, A. Grimvall, O. Dahlman, R. Morck, Environ. Sci. Technol. 31 (1997) 2464. [72] C. Flodin, M. Ekelund, H. Borén, A. Grimvall, Chemosphere 34 (1997) 2319. [73] S. Pedersen-Bjergaard, S.I. Semb, E.M. Brevik, T. Greibrokk, J. Chromatogr. A 723 (1996) 337. [74] P. Faure, F. Vilmin, R. Michels, E. Jarde, L. Mansuy, M. Elie, P. Landais, J. Anal. Appl. Pyrol. 62 (2002) 297. [75] E. Jarde, F. Vilmin, L. Mansuy, P. Faure, J. Anal. Appl. Pyrol. 71 (2004) 553. [76] O. Hjelm, H. Borén, G. Öberg, Chemosphere 32 (1996) 1719. [77] A. Louati, B. Elleuch, P. Sandra, F. David, A. Saliot, J. Dagaut, J. Oudot, J. Microcol. Sep. 13 (2001) 90. [78] G. Becker, U. Nilsson, A. Colmsjö, C. Östman, J. Chromatogr. A 826 (1998) 57. [79] C. Haberhauer-Troyer, E. Rosenberg, M. Grasserbauer, J. Chromatogr. A 848 (1999) 305. [80] G. Becker, A. Colmsjö, Anal. Chim. Acta 376 (1998) 265. [81] G. Becker, A. Colmsjö, C. Östman, Environ. Sci. Technol. 33 (1999) 1321. [82] M. Schweigkofler, R. Niessner, Environ. Sci. Technol. 33 (1999) 3680. [83] S. Junyapoon, K.D. Bartle, A.B. Ross, M. Cooke, Int. J. Environ. Anal. Chem. 82 (2002) 47. [84] S.R. Sousa, S.E. Bialkowski, Anal. Chim. Acta 433 (2001) 181. [85] H.B. Swan, J.P. Ivey, J. High Resolut. Chromatogr. 17 (1994) 814. [86] S. Shin, K. Sakanishi, I. Mochida, D.A. Grudoski, J.H. Shinn, Energy Fuels 14 (2000) 539. [87] S. Shin, H. Yang, K. Sakanishi, I. Mochida, D.A. Grudoski, J.H. Shinn, Appl. Catal. A 205 (2001) 101. [88] M.P.M. van Lieshout, H.G. Janssen, C.A. Cramers, G.A. van den Bos, J. Chromatogr. A 764 (1997) 73. [89] J.T. Andersson, B. Schmid, J. Chromatogr. A 693 (1995) 325. [90] B. Schmidt, J.T. Andersson, Anal. Chem. 69 (1997) 3476. [91] G.A. Depauw, G.F. Froment, J. Chromatogr. A 761 (1997) 231. [92] H. Du, Z. Ring, Y. Briker, P. Arboleda, Catal. Today 98 (2004) 217. [93] A.H. Hegazi, J.T. Andersson, M.S. El-Gayar, Fuel Process. Technol. 85 (2004) 1. [94] A.H. Hegazi, J.T. Andersson, M.A. bu-Elgheit, M.S. El-Gayar, Chemosphere 55 (2004) 1053. [95] N.R. Hardas, R. Adam, P.C. Uden, J. Chromatogr. A 844 (1999) 249. [96] P.K.K. Louie, R.C. Timpe, S.B. Hawthorne, D.J. Miller, Fuel 72 (1993) 225. [97] A.C. Lewis, S.A. Askey, K.M. Holden, K.D. Bartle, M.J. Pilling, J. High Resolut. Chromatogr. 20 (1997) 109. [98] S.G. Mössner, S.A. Wise, Anal. Chem. 71 (1999) 58. [99] J. Rolfes, J.T. Andersson, Anal. Comm. 33 (1996) 429. [100] J. Rolfes, J.T. Andersson, Anal. Chem. 73 (2001) 3073. [101] A. Stumpf, K. Tolvaj, M. Juhász, J. Chromatogr. A 819 (1998) 67. [102] L.L.P. van Stee, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, J. Chromatogr. A 1019 (2003) 89. [103] P. Wiwel, K. Knudsen, P. Zeuthen, D. Whitehurst, Ind. Eng. Chem. Res. 39 (2000) 533. [104] S.D. Sumbogo Murti, K. Sakanishi, O. Okuma, Y. Korai, I. Mochida, Fuel 81 (2002) 2241. [105] S.D. Sumbogo Murti, H. Yang, K.H. Choi, Y. Korai, I. Mochida, Appl. Catal. A 252 (2003) 331. [106] S.D. Sumbogo Murti, K.H. Choi, K. Sakanishi, O. Okuma, Y. Korai, I. Mochida, Fuel 84 (2005) 135. [107] S.D. Sumbogo Murti, K.H. Choi, Y. Korai, I. Mochida, Appl. Catal. A 280 (2005) 133. [108] T. Kabe, A. Ishihara, H. Tajima, Ind. Eng. Chem. Res. 31 (1992) 1577. [109] D.D. Link, P. Zandhuis, Fuel 85 (2006) 451. [110] P. Steiner, E.A. Blekkan, Fuel Process. Technol. 79 (2002) 1. [111] Y. Shiraishi, T. Hirai, I. Komasawa, Ind. Eng. Chem. Res. 40 (2001) 293. [112] Y. Shiraishi, T. Hirai, I. Komasawa, Ind. Eng. Chem. Res. 40 (2001) 3398. [113] F.L. Li, P. Xu, C.Q. Ma, L.L. Luo, X.S. Wang, FEMS Microbiol. Lett. 223 (2003) 301. [114] T. Furuya, Y. Ishii, K.i. Noda, K. Kino, K. Kirimura, FEMS Microbiol. Lett. 221 (2003) 137. [115] S. Guobin, X. Jianmin, Z. Huaiying, L. Huizhou, J. Chem. Technol. Biotechnol. 80 (2005) 420. [116] F.C.Y. Wang, J. Chromatogr. A 753 (1996) 101. [117] F.C.Y. Wang, B. Gerhart, Anal. Chem. 68 (1996) 3917. [118] F.C.Y. Wang, J.G. Green, B.B. Gerhart, Anal. Chem. 68 (1996) 2477. [119] F.C.Y. Wang, Anal. Chem. 71 (1999) 2037. 189 12. References [120] M. Brebu, T. Bhaskar, K. Murai, A. Muto, Y. Sakata, M. Uddin, Chemosphere 56 (2004) 433. [121] U. Fuchslueger, H.J. Grether, M. Grasserbauer, Fresenius J. Anal. Chem. 349 (1994) 283. [122] S. Nakamura, M. Takino, S. Daishima, Anal. Sci. 16 (2000) 627. [123] H. Sato, S. Tsuge, H. Ohtani, K. Aoi, A. Takasu, M. Okada, Macromolecules 30 (1997) 4030. [124] J.R. Stuff, W.R. Creasy, A.A. Rodriguez, H.D. Durst, J. Microcol. Sep. 11 (1999) 644. [125] M.D. Brickhouse, W.R. Creasy, B.R. Williams, K.M. Morrissey, R.J. O'Connor, H.D. Durst, J. Chromatogr. A 883 (2000) 185. [126] M. Mazurek, Z. Witkiewicz, S. Popiel, M. Sliwakowski, J. Chromatogr. A 919 (2001) 133. [127] K. Schoene, H.J. Bruckert, J. Steinhanses, Fresenius J. Anal. Chem. 345 (1993) 688. [128] W.R. Creasy, A.A. Rodriguez, J.R. Stuff, R.W. Warren, J. Chromatogr. A 709 (1995) 333. [129] W.R. Creasy, J. Stuff, B. Williams, K. Morrissey, J. Mays, R. Duevel, H.D. Durst, J. Chromatogr. A 774 (1997) 253. [130] W.R. Creasy, M.D. Brickhouse, K.M. Morrissey, J.R. Stuff, R.L. Cheicante, J. Ruth, J. Mays, B.R. Williams, R. O'Connor, H.D. Durst, Environ. Sci. Technol. 33 (1999) 2157. [131] S. Popiel, Z. Witkiewicz, T. Nalepa, J. Hazard. Mater. 123 (2005) 269. [132] S. Popiel, Z. Witkiewicz, A. Szewczuk, J. Hazard. Mater. 123 (2005) 94. [133] K. Schoene, H.J. Bruckert, H. Jurling, J. Steinhanses, J. Chromatogr. A 719 (1996) 401. [134] K. Karlson, R. Ishaq, G. Becker, P. Berggren, D. Broman, A. Colmsjö, Environ. Pollut. 110 (2000) 29. [135] K. Janák, G. Becker, A. Colmsjö, C. Östman, M. Athanasiadou, M. Valters, Å. Bergman, Environ. Toxicol. Chem. 17 (1998) 1046. [136] A. Johnson, N. Olson, Arch. Environ. Contam. Toxicol. 41 (2001) 339. [137] B.D. Quimby, P.C. Dryden, J.J. Sullivan, Anal. Chem. 62 (1990) 2509. [138] M.S. Boukraa, D. Deruaz, A. Bannier, M. Desage, J.L. Brazier, J. Pharm. Biomed. Anal. 12 (1994) 185. [139] J.J. Johnston, D.A. Goldade, D.J. Kohler, J.L. Cummings, Environ. Sci. Technol. 34 (2000) 1856. [140] J. Hajšlová, P. Cuhra, M. Kempny, J. Poustka, K. Holadova, V. Kocourek, J. Chromatogr. A 699 (1995) 231. [141] E.D. Lykissa, S.V. Kala, J.B. Hurley, R.M. Lebovitz, Anal. Chem. 69 (1997) 4912. [142] S.V. Kala, E.D. Lykissa, R.M. Lebovitz, Anal. Chem. 69 (1997) 1267. [143] W. Elbast, F. Besacier, D. Deruaz, J.L. Brazier, J. Chromatogr. B 690 (1997) 115. [144] W. Elbast, M.S. Caubet, D. Deruaz, J.L. Brazier, Anal. Lett. 32 (1999) 3111. [145] S. Voisin, F.N. Renaud, J. Freney, M. de Montclos, R. Boulieu, D. Deruaz, J. Chromatogr. A 863 (1999) 243. [146] S. Voisin, D. Deruaz, J. Freney, F.N.R. Renaud, Res. Microbiol. 153 (2002) 307. [147] V. Grossi, D. Raphel, Phytochem. 63 (2003) 693. [148] C. Bicchi, C. Frattini, G. Pellegrino, P. Rubiolo, V. Raverdino, G. Tsoupras, J. Chromatogr. 609 (1992) 305. [149] F. David, P. Sandra, J. High Resolut. Chromatogr. 14 (1991) 554. [150] Th. Hankemeier, J. Rozenbrand, M. Adahchour, J.J. Vreuls, U.A.Th. Brinkman, Chromatographia 48 (1998) 273. [151] L.M. Blumberg, M.S. Klee, Anal. Chem. 70 (1998) 3828. [152] V. Giarrocco, B.D. Quimby, M.S. Klee, Hewlett Packard Application Note 228-393 (1997) [153] H.G.J. Mol, Th. Hankemeier, U.A.Th. Brinkman, LC-GC Int. 12 (1999) 108. [154] D.K. Nguyen, A. Bruchet, P. Arpino, J. High Resolut. Chromatogr. 17 (1994) 153. [155] B.D. Quimby, J.J. Sullivan, Anal. Chem. 62 (1990) 1027. [156] P.L. Wylie, B.D. Quimby, J. High Resolut. Chromatogr. 12 (1989) 813. [157] B.D. Quimby, P.A. Larson, P.C. Dryden, Hewlett Packard Application Note 228-363 (1996) 1. [158] C. Webster, M. Cooke, J. High Resolut. Chromatogr. 18 (1995) 319. [159] A.L.P. Valente, P.C. Uden, Analyst 115 (1990) 525. [160] J. Dallüge, J. Beens, U.A.Th. Brinkman, J. Chromatogr. A 1000 (2003) 69. [161] P.J. Marriott, P. Haglund, R.C.Y. Ong, Clin. Chim. Acta 328 (2003) 1. [162] J.B. Phillips, J. Beens, J. Chromatogr. A 856 (1999) 331. [163] H.-J. de Geus, J. de Boer, U.A.Th. Brinkman, Chromatographia 55 (2002) 339. 190 12. References [164] P. Korytár, L.L.P. van Stee, P.E.G. Leonards, J. de Boer, U.A.Th. Brinkman, J. Chromatogr. A 994 (2003) 179. [165] P. Korytár, P.E.G. Leonards, J. de Boer, U.A.Th. Brinkman, J. Chromatogr. A 958 (2002) 203. [166] J. Dallüge, L.L.P. van Stee, X. Xu, J. Williams, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, J. Chromatogr. A 974 (2002) 169. [167] M. van Deursen, J. Beens, J. Reijenga, P. Lipman, C. Cramers, J. Blomberg, J. High Resolut. Chromatogr. 23 (2000) 507. [168] J. Beens, M. Adahchour, R.J.J. Vreuls, K. van Altena, U.A.Th. Brinkman, J. Chromatogr. A 919 (2001) 127. [169] J.J. Sullivan, B.D. Quimby, Anal. Chem. 62 (1990) 1034. [170] J. Dallüge, M. van Rijn, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, J. Chromatogr. A 965 (2002) 207. [171] R.A. Shellie, L.L. Xie, P.J. Marriott, J. Chromatogr. A 968 (2002) 161. [172] R. Firor, B.D. Quimby, Hewlett Packard Application Note 5988-2426EN (2001) [173] A. Amorelli, Y.D. Amos, C.P. Halsig, J.J. Kosman, R.J. Jonker, M. de Wind, J. Vrieling, Hydrocarb. Process. 71 (1992) 93. [174] H.J. Coleman, D.E. Hirsh, J.E. Dooley, Anal. Chem. 41 (1969) 800. [175] X. Ma, K. Sakanishi, T. Isoda, I. Mochida, Fuel 76 (1997) 329. [176] J. Beens, R. Tijssen, J. High Resolut. Chromatogr. 20 (1997) 131. [177] F.C.Y. Wang, F.P. Disanzo, F.C. McElroy, presented at the First International Symposium on Two- Dimensional Gas Chromatography, Volendam, the Netherlands, 2003. [178] C. Willey, M. Iwao, R.N. Castle, M.L. Lee, Anal. Chem. 53 (1981) 400. [179] J.S.S. Damsté, C.M. White, J.B. Green, J.W. De Leeuw, Energy Fuels 13 (1999) 728. [180] A.C. Belfroid, M. van Drunen, M.A. Beek, S.M. Schrap, C.A.M. van Gestel, B. van Hattum, Sci. Total. Environ. 222 (1998) 167. [181] A.J. Hendriks, J.L. Maas-Diepeveen, A. Noordsij, M.A. van der Gaag, Water Res. 28 (1994) 581. [182] H.J.M. Verhaar, F.J.M. Busser, J.L.M. Hermens, Environ. Sci. Technol. 29 (1995) 726. [183] F. Stuer-Lauridsen, Environ. Pollut. 136 (2005) 503. [184] J.N. Huckins, M.W. Tubergen, G.K. Manuweera, Chemosphere 20 (1990) 533. [185] J.N. Huckins, G.K. Manuweera, J.D. Petty, D. Mackay, J.A. Lebo, Environ. Sci. Technol. 27 (1993) 2489. [186] J.N. Huckins, J.D. Petty, H.F. Prest, C.E. Orazio, R.C. Clark, Report 4690, A guide for the use of semipermeable membrane devices (SPMDs) as samplers of waterborne hydrophobic organic contaminants., American Petroleum Institute, Washington, DC, 2002 [187] W.M.G.M. van Loon, F.G. Wijneker, M.E. Verwoerd, J.L.M. Hermens, Anal. Chem. 68 (1996) 2916. [188] K. Booij, H.M. Sleiderink, F. Smedes, Environ. Toxicol. Chem. 17 (1998) 1236. [189] F.W. McLafferty, R.H. Hertel, R.D. Villwock, Org. Mass. Spectrom. 9 (1974) 690. [190] C.F. Poole, I.D. Wilson (Eds.) J. Chromatogr. A 885 (2000) [191] J.J. Vreuls, A.J.H. Louter, U.A.Th. Brinkman, J. Chromatogr. A 856 (1999) 279. [192] F. Smedes, Int. J. Environ. Anal. Chem. 57 (1994) 215. [193] C.J. Roghair, J. Struijs, D. de Zwart, Report 607504 002, Measurement of toxic potency in fresh waters in the Netherlands; Part A: Methods, RIVM, Bilthoven, The Netherlands, 1997 [194] M. Collombon, R. Van de Kamp, J. Struijs, Report 607042 008, Procedure for extracting organic micro- pollutants from water samples to monitor toxicological stress, RIVM, Bilthoven, The Netherlands, 1997 [195] J. Struijs, R. Van de Kamp, E.A. Hogendoorn, Report 607602 001, Isolating organic micropollutants from water samples by means of XAD resins and supercritical fluid extraction, RIVM, Bilthoven, The Netherlands, 1998 [196] K. Grob, On-Column Injection in Capillary Gas Chromatography: basic technique, retention gaps, solvent effects. Hüthig, Heidelberg, 1987, Ch. B5, pp. 334-346. [197] C.F. Poole, S.K. Poole, Chromatography today. Elsevier, Amsterdam, 1991, Ch. 4.7, pp. 493-508. [198] H. Brunn, G. Rimkus, Ernahrungs-Umschau 44 (1997) 4. [199] A.M. Api, R.A. Ford, Toxicol. Lett. 111 (1999) 143. [200] M.S. Christian, R.M. Parker, A.M. Hoberman, R.M. Diener, A.M. Api, Toxicol. Lett. 111 (1999) 169. [201] R.A. Ford, Dtsch. Lebensm.-Rundsch. 94 (1998) 268. 191 12. References [202] G.G. Rimkus, Toxicol. Lett. 111 (1999) 37. [203] K. Bester, H. Hühnerfuss, W. Lange, G.G. Rimkus, N. Theobald, Water Res. 32 (1998) 1857. [204] E.M.J. Verbruggen, W.M.G.M. van Loon, M. Tonkes, P. Van Duijn, W. Seinen, J.L.M. Hermens, Environ. Sci. Technol. 33 (1999) 801. [205] R. Draisci, C. Marchiafava, L. Palleschi, E. Ferretti, A. Anastasio, P. Visciano, Ind. Aliment. 38 (1999) 374. [206] S. Franke, C. Meyer, N. Heinzel, R. Gatermann, H. Hühnerfuss, G. Rimkus, W.A. Konig, W. Francke, Chirality 11 (1999) 795. [207] S. Muller, P. Schmid, C. Schlatter, Chemosphere 33 (1996) 17. [208] G.G. Rimkus, M. Wolf, Chemosphere 33 (1996) 2033. [209] S.J. Klaine, G.P. Cobb, R.L. Dickerson, K.R. Dixon, R.J. Kendall, E.E. Smith, K.R. Solomon, Environ. Toxicol. Chem. 15 (1996) 21. [210] J.E.M. Beurskens, G.A.J. Mol, H.L. Barreveld, B. van Munster, H.J. Winkels, Environ. Toxicol. Chem. 12 (1993) 1549. [211] J. de Jonge, J.M. Brils, A.J. Hendriks, W.C. Ma, Aquat. Ecosys. Health Manag. 2 (1999) 9. [212] P.J. den Besten, C.A. Schmidt, M. Ohm, M.M. Ruys, J.W. van Berghem, C. van de Guchte, J. Aquat. Ecosyst. Health 4 (1995) 257. [213] A.J. Hendriks, J.L. Maas-Diepeveen, A. Noordsij, M.A. van der Gaag, Water Res. 28 (1994) 581. [214] J. Lahr, J.L. Maas-Diepeveen, S.C. Stuijfzand, P.E.G. Leonards, J.M. Druke, S. Lucker, A. Espeldoorn, L.C.M. Kerkum, L.L.P. van Stee, A.J. Hendriks, Water Res. 37 (2003) 1691. [215] Montana State University, Quantitative Structure Activity Relationships (Computer Programme). Montana, USA, 1989. [216] C.A. Staples, D.R. Peterson, T.F. Parkerton, W.J. Adams, Chemosphere 35 (1997) 667. [217] A.J. Hendriks, Aquat. Toxicol. 31 (1995) 1. [218] Aquasense, Report 94.0454, Statistical Analysis of Bioassay Results, AquaSense Consultants, Amsterdam, the Netherlands, 1994 [219] E.T.H.M. Peeters, A. Dewitte, A.A. Koelmans, J.A. van der Velden, P.J. den Besten, Environ. Toxicol. Chem. 20 (2001) 2883. [220] F. Fehsenfeld, J. Calvert, R. Fall, P. Goldan, A.B. Guenther, C.N. Hewitt, B. Lamb, S. Liu, M. Trainer, H. Westberg, P. Zimmerman, Global Biogeochem. Cycles 6 (1992) 389. [221] M.O. Andreae, P.J. Crutzen, Science 276 (1997) 1052. [222] A. Limbeck, H. Puxbaum, Atmos. Environ. 33 (1999) 1847. [223] Z. Krivácsy, A. Gelencsér, G.E. Mészáros, A. Molnár, A. Hoffer, T. Mészáros, Z. Sárvári, D. Temesi, B. Varga, U. Baltensperger, S. Nyeki, E. Weingartner, J. Atmos. Chem. 39 (2001) 235. [224] C.D. O'Dowd, P. Aalto, K. Hämeri, M. Kulmala, T. Hoffmann, Nature 416 (2002) 497. [225] M.F. Mohamed, D. Kang, V.P. Aneja, Chemosphere 47 (2002) 863. [226] WMO, Report 37, Scientific assessment of ozone depletion: 1994, Global Ozone Research and Monitoring Project, World Meteorological Organization, Geneva, Switzerland, 1994 [227] IPCC 2001: Climate Change 2001: The Scientific Basis, Cambridge University Press, Cambridge, UK, 2001 [228] D. Helmig, W. Pollock, J. Greenberg, P.R. Zimmermann, J. Geophys. Res. 101 (1996) 697. [229] B. Rappenglück, P. Fabian, P. Kalabokas, L.G. Viras, I.C. Ziomas, Atmos. Environ. 32 (1998) 2103. [230] A. Wedel, K.P. Müller, M. Ratte, J. Rudolph, J. Atmos. Chem. 31 (1998) 73. [231] A.C. Lewis, N. Carslaw, P.J. Marriott, R.M. Kinghorn, P. Morrison, A.L. Lee, K.D. Bartle, M.J. Pilling, Nature 405 (2000) 778. [232] J.B. Phillips, J.Z. Xu, J. Chromatogr. A 703 (1995) 327. [233] G. Schomburg, J. Chromatogr. A 703 (1995) 309. [234] J. Beens, R. Tijssen, J. Blomberg, J. Chromatogr. A 822 (1998) 233. [235] R.M. Kinghorn, P.J. Marriott, J. High Resolut. Chromatogr. 21 (1998) 32. [236] W. Bertsch, J. High Resolut. Chromatogr. 22 (1999) 647. [237] W. Bertsch, J. High Resolut. Chromatogr. 23 (2000) 167. [238] E.B. Ledford, Recent advances in GC×GC, presented at the 23rd International Symposium on Capillary Chromatography, Riva del Garda, Italy, 2000 192 12. References [239] C.A. Bruckner, B.J. Prazen, R.E. Synovec, Anal. Chem. 70 (1998) 2796. [240] C.G. Fraga, B.J. Prazen, R.E. Synovec, J. High Resolut. Chromatogr. 23 (2000) 215. [241] K.J. Johnson, B.J. Prazen, R.K. Olund, R.E. Synovec, J. Sep. Sci. 25 (2002) 297. [242] X. Xu, J. Williams, C. Plass-Dülmer, H. Berresheim, G. Salisbury, L. Lange, J. Lelieveld, Atmos. Chem. Phys. Discuss. 3 (2003) 1477. [243] J. Lelieveld, H. Berresheim, S. Borrmann, P.J. Crutzen, F.J. Dentener, H. Fischer, J. Feichter, P.J. Flatau, J. Heland, R. Holzinger, R. Korrmann, M.G. Lawrence, Z. Levin, K.M. Markowicz, N. Mihalopoulos, A. Minikin, V. Ramanathan, M. De Reus, G.J. Roelofs, H.A. Scheeren, J. Sciare, H. Schlager, M. Schultz, P. Siegmund, B. Steil, E.G. Stephanou, P. Stier, M. Traub, C. Warneke, J. Williams, H. Ziereis, Science 298 (2002) 794. [244] J. Beens, J. Blomberg, P.J. Schoenmakers, J. High Resolut. Chromatogr. 23 (2000) 182. [245] E.B. Ledford, C. Billesbach, J. High Resolut. Chromatogr. 23 (2000) 202. [246] I.G. Zenkevich, J. High Resolut. Chromatogr. 21 (1998) 565. [247] P. Ciccioli, A. Cecinato, E. Brancaleoni, M. Frattoni, A. Liberti, J. High Resolut. Chromatogr. 15 (1992) 75. [248] P. Ciccioli, E. Brancaleoni, A. Cecinato, R. Sparapani, M. Frattoni, J. Chromatogr. B 643 (1993) 55. [249] P. Ciccioli, A. Cecinato, E. Brancaleoni, M. Frattoni, F. Bruner, M. Maione, Int. J. Environ. Anal. Chem. 62 (1996) 245. [250] C.G. Fraga, B.J. Prazen, R.E. Synovec, Anal. Chem. 73 (2001) 5833. [251] J. Beens, H. Boelens, R. Tijssen, J. Blomberg, J. High Resolut. Chromatogr. 21 (1998) 47. [252] J.P. Greenberg, D. Helmig, P.R. Zimmermann, J. Geophys. Res. 101 (1996) 581. [253] A.R. Katritzky, E.S. Ignatchenko, R.A. Barcock, V.S. Lobanov, M. Karelson, Anal. Chem. 66 (1994) 1799. [254] C.C. Lutes, P.H. Kariher, Report EPA-600/R-96-128, Evaluation of emissions from the open burning of land-cleaning debris, U.S. Environmental Protection Agency, Research Triangle Park, NC, USA, 1996 [255] H.R. Friedli, E. Atlas, V.R. Stroud, L. Giovanni, T. Campos, L.F. Radke, Global Biogeochem. Cycles 15 (2001) 435. [256] G. Salisbury, J. Williams, R. Holzinger, V. Gros, N. Mihalopoulos, M. Vrekoussis, R. Sarda-Estéve, H. Berresheim, R. von Kuhlmann, M. Lawrence, J. Lelieveld, Atmos. Chem. Phys. Discuss. 3 (2003) 911. [257] M.O. Andreae, P. Merlet, Global Biogeochem. Cycles 15 (2001) 955. [258] C. Yin, T. Lin, S. Liu, L. Wang, W. Guo, R. Fu, Z. Pan, J. Chin. Chem. Soc. 48 (2001) 739. [259] J. Hayes, E.W. Pitzer, J. High Resolut. Chromatogr. 8 (1985) 230. [260] Haagen-Smit Laboratory, Report Tech. Rep. SOP NO. MLD 118, Revision NO. I.I, Procedure for the detailed hydrocarbon analysis of gasolines by single column high efficiency (capillary) column gas chromatography, Air Resource Board, California Environmental Protection Agency, 1997. [261] R.J. Laub, J.H. Purnell, J. High Resolut. Chromatogr. 11 (1988) 649. [262] A.J. Lubeck, D.L. Sutton, J. High Resolut. Chromatogr. 7 (1984) 542. [263] J. Bermejo, C.G. Blanco, M.A. Diez, M.D. Guillen, J. High Resolut. Chromatogr. 10 (1987) 461. [264] E.H. Andrade, J.G. Maia, M.d.G. Zoghbi, J. Food Comp. Anal. 13 (2000) 27. [265] Flavornet: Kovats Retention Indices, http://flavornet.org/f_kovats.html [266] HSE, Methods for the Determination of Hazardous Substances, 88, Volatile organic compounds in air. HSE, Suffolk, UK, 1997. [267] C.E. Rostad, W.E. Pereira, J. High Resolut. Chromatogr. 9 (1986) 328. [268] Agilent Solvent Retention Data, http://www.chem.agilent.com/cag/cabu/pdf/b-0292.pdf [269] F. David, F. Scanlan, P. Sandra, Retention time locking in flavor analysis, presented at the 23rd International Symposium on Capillary Chromatography, Riva del Garda, Italy, 2000. [270] M.S. Madruga, D.S. Mottram, J. Braz. Chem. Soc. 9 (1998) 261. [271] L. Weber, J. High Resolut. Chromatogr. 9 (1986) 446. [272] S.E. Reichenbach, X. Tian, C. Cordero, Q. Tao, J. Chromatogr. A 1226 (2012) 140. [273] Z.D. Zeng, H. Hugel, P. Marriott, Anal. Bioanal. Chem. 401 (2011) 2373. [274] J. Vial, H. Nocairi, P. Sassiat, S. Mallipatu, G. Cognon, D. Thiébaut, B. Teillet, D.N. Rutledge, J. Chromatogr. A 1216 (2009) 2866. [275] K.M. Pierce, J.C. Hoggard, R.E. Mohler, R.E. Synovec, J. Chromatogr. A 1184 (2008) 341. 193 12. References [276] K.M. Pierce, B. Kehimkar, L.C. Marney, J.C. Hoggard, R.E. Synovec, J. Chromatogr. A 1255 (2012) 3. [277] O. Amador-Muñoz, P.J. Marriott, J. Chromatogr. A 1184 (2008) 323. [278] S. Castillo, I. Mattila, J. Miettinen, M. Oresic, T. Hyötyläinen, Anal. Chem. 83 (2011) 3058. [279] J.C. Hoggard, R.E. Synovec, in L. Ramos (Editor), Comprehensive Two Dimensional Gas Chromatography. Elsevier, Amsterdam, 2009, pp. 107-122. [280] S.E. Reichenbach, in L. Ramos (Editor), Comprehensive Two Dimensional Gas Chromatography. Elsevier, Amsterdam, 2009, pp. 77-106. [281] M. Adahchour, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, Trends Anal. Chem. 25 (2006) 438. [282] G. Vivó-Truyols, J.R. Torres-Lapasió, A.M. Van Nederkassel, Y. Vander Heyden, D.L. Massart, J. Chromatogr. A 1096 (2005) 133. [283] P. Nikitas, A. Pappa-Louisi, A. Papageorgiou, J. Chromatogr. A 912 (2001) 13. [284] J.P. Foley, J. Chromatogr. A 384 (1987) 301. [285] A.N. Papas, T.P. Tougas, Anal. Chem. 62 (1990) 234. [286] A.P. de la Mata, K.D. Nizio, J.J. Harynuk, J. Chromatogr. A 1255 (2012) 190. [287] C.J. Xu, Y.Z. Liang, F.T. Chau, Talanta 68 (2005) 108. [288] S.E. Stein, J. Am. Soc. Mass. Spectrom. 10 (1999) 770. [289] V.G. van Mispelaar, Chromametrics, Ph.D. Thesis, University of Amsterdam, The Netherlands, 2005, http://dare.uva.nl/record/160951 [290] J. Dallüge, R.J.J. Vreuls, J. Beens, U.A.Th. Brinkman, J. Sep. Sci. 25 (2002) 201. [291] S. Peters, G. Vivó-Truyols, P.J. Marriott, P.J. Schoenmakers, J. Chromatogr. A 1156 (2007) 14. [292] G. Vivó-Truyols, Anal. Chem. 84 (2012) 2622. [293] M. Adahchour, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, Trends Anal. Chem. 25 (2006) 821. [294] M. Adahchour, J. Beens, U.A.Th. Brinkman, J. Chromatogr. A 1186 (2008) 67. [295] M.F. Almstetter, P.J. Oefner, K. Dettmer, Anal. Bioanal. Chem. 402 (2012) 1993. [296] J.W. Wong, K. Zhang, D.G. Hayward, C. Kai-Meng, Methods Mol. Biol. 747 (2011) 131. [297] B.T. Weldegergis, A.M. Crouch, T. Górecki, A. de Villiers, Anal. Chim. Acta 701 (2011) 98. [298] J. Vestner, S. Malherbe, T.M. Du, H.H. Nieuwoudt, A. Mostafa, T. Górecki, A.G. Tredoux, A. de Villiers, J. Agric. Food Chem. 59 (2011) 12732. [299] A.L. Robinson, P.K. Boss, H. Heymann, P.S. Solomon, R.D. Trengove, J. Agric. Food Chem. 59 (2011) 3273. [300] A.L. Robinson, P.K. Boss, H. Heymann, P.S. Solomon, R.D. Trengove, J. Chromatogr. A 1218 (2011) 504. [301] M.P. Pedroso, E.C. Ferreira, L.W. Hantao, Bogusz S Jr, F. Augusto, J. Sep. Sci. 34 (2011) 1547. [302] A. Papetti, Curr. Opin. Biotechnol. 23 (2011) 168. [303] B. Mitrevski, B. Veleska, E. Engel, P. Wynne, S.M. Song, P.J. Marriott, Forensic Sci. Int. 209 (2011) 11. [304] S.D. Johanningsmeier, R.F. McFeeters, J. Food. Sci. 76 (2011) C168-C177. [305] J. Ha, D. Seo, D. Shin, Talanta 85 (2011) 252. [306] J.Y. Gardner, D.E. Brillhart, M.M. Benjamin, L.G. Dixon, L.M. Mitchell, J.M. Dimandja, J. Sep. Sci. 34 (2011) 176. [307] S.T. Chin, G.T. Eyres, P.J. Marriott, J. Chromatogr. A 1218 (2011) 7487. [308] M.C. Waldhier, M.F. Almstetter, N. Nurnberger, M.A. Gruber, K. Dettmer, P.J. Oefner, J. Chromatogr. A 1218 (2011) 4537. [309] A.G. Scarlett, R. Clough, C. West, C.A. Lewis, A.M. Booth, S.J. Rowland, Environ. Sci. Technol. 45 (2011) 6160. [310] M. Pena-Abaurrea, A. Covaci, L. Ramos, J. Chromatogr. A 1218 (2011) 6995. [311] M. Menendez-Carreno, H. Steenbergen, H.G. Janssen, Anal. Bioanal. Chem. 402 (2012) 2023. [312] M.M. Koek, F.M. van der Kloet, R. Kleemann, T. Kooistra, E.R. Verheij, Th. Hankemeier, Metabolomics 7 (2011) 1. [313] I. Bondia-Pons, E. Nordlund, I. Mattila, K. Katina, A.M. Aura, M. Kolehmainen, M. Oresic, H. Mykkanen, K. Poutanen, Nutr. J. 10 (2011) 116. [314] M. Zapadlo, J. Krupcik, T. Kovalczuk, P. Majek, I. Spanik, D.W. Armstrong, P. Sandra, J. Chromatogr. A 1218 (2011) 746. 194 12. References [315] V. Vasilieva, K.E. Scherr, E. Edelmann, M. Hasinger, A.P. Loibner, J. Biotechnol. 157 (2011) 460. [316] J. Ruiz-Jimenez, J. Parshintsev, T. Laitinen, K. Hartonen, M.L. Riekkola, T. Petaja, M. Kulmala, J. Environ. Monit. 13 (2011) 2994. [317] M.J. Gomez, S. Herrera, D. Sole, E. Garcia-Calvo, A.R. Fernandez-Alba, Anal. Chem. 83 (2011) 2638. [318] J. de Vos, R. Dixon, G. Vermeulen, P. Gorst-Allman, J. Cochran, E.R. Rohwer, J.F. Focant, Chemosphere 82 (2011) 1230. [319] R. van der Westhuizen, H. Potgieter, N. Prinsloo, A. de Villiers, P. Sandra, J. Chromatogr. A 1218 (2011) 3173. [320] T. Sfetsas, C. Michailof, A. Lappas, Q. Li, B. Kneale, J. Chromatogr. A 1218 (2011) 3317. [321] S.J. Rowland, C.E. West, D. Jones, A.G. Scarlett, R.A. Frank, L.M. Hewitt, Environ. Sci. Technol. 45 (2011) 9806. [322] S.J. Rowland, C.E. West, A.G. Scarlett, D. Jones, Rapid Commun. Mass Spectrom. 25 (2011) 1741. [323] S.J. Rowland, C.E. West, A.G. Scarlett, D. Jones, R.A. Frank, Rapid Commun. Mass Spectrom. 25 (2011) 1198. [324] S.P. Pyl, C.M. Schietekat, K.M. Van Geem, M.F. Reyniers, J. Vercammen, J. Beens, G.B. Marin, J. Chromatogr. A 1218 (2011) 3217. [325] H. Parastar, J.R. Radovic, M. Jalali-Heravi, S. Diez, J.M. Bayona, R. Tauler, Anal. Chem. 83 (2011) 9289. [326] B. Omais, M. Courtiade, N. Charon, D. Thiébaut, A. Quignard, M.C. Hennion, J. Chromatogr. A 1218 (2011) 3233. [327] M.E. Machado, F.C. Fontanive, J.V. de Oliveira, E.B. Caramao, C.A. Zini, Anal. Bioanal. Chem. 401 (2011) 2433. [328] B.S. Mitrevski, P. Wilairat, P.J. Marriott, J. Chromatogr. A 1217 (2010) 127. [329] J. Ratel, E. Engel, J. Chromatogr. A 1216 (2009) 7889. [330] A. Kohl, J. Cochran, D.M. Cropek, J. Chromatogr. A 1217 (2010) 550. [331] E.M. Humston, J.D. Knowles, A. McShea, R.E. Synovec, J. Chromatogr. A 1217 (2010) 1963. [332] K.A. Kouremenos, J. Pitt, P.J. Marriott, J. Chromatogr. A 1217 (2010) 104. [333] T. Laitinen, S.H. Martin, J. Parshintsev, T. Hyötyläinen, K. Hartonen, M.L. Riekkola, M. Kulmala, J.L. Pavon, J. Chromatogr. A 1217 (2010) 151. [334] V. Matamoros, E. Jover, J.M. Bayona, Anal. Chem. 82 (2009) 699. [335] E. Hoh, S.J. Lehotay, K.C. Pangallo, K. Mastovska, H.L. Ngo, C.M. Reddy, W. Vetter, J. Agric. Food Chem. 57 (2009) 2653. [336] S. O'Hagan, W.B. Dunn, J.D. Knowles, D. Broadhurst, R. Williams, J.J. Ashworth, M. Cameron, D.B. Kell, Anal. Chem. 79 (2007) 464. [337] L. Mondello, M. Herrero, T. Kumm, P. Dugo, H. Cortes, G. Dugo, Anal. Chem. 80 (2008) 5418. [338] S.E. Reichenbach, Anal. Chem. 81 (2009) 5099. [339] http://www.shimadzu.com/an/2d-chrom/gc_cs.html [340] P.Q. Tranchida, G. Purcaro, C. Fanali, P. Dugo, G. Dugo, L. Mondello, J. Chromatogr. A 1217 (2010) 4160. [341] P.Q. Tranchida, S. Salivo, I. Bonaccorsi, A. Rotondo, P. Dugo, L. Mondello, J. Chromatogr. A 1313 (2013) 194. [342] P.Q. Tranchida, G. Purcaro, L. Conte, P. Dugo, G. Dugo, L. Mondello, J. Chromatogr. A 1216 (2009) 7301. [343] P.Q. Tranchida, M. Zoccali, I. Bonaccorsi, P. Dugo, L. Mondello, G. Dugo, J. Chromatogr. A 1305 (2013) 276. [344] P.Q. Tranchida, G. Purcaro, A. Visco, L. Conte, P. Dugo, P. Dawes, L. Mondello, J. Chromatogr. A 1218 (2011) 3140. [345] P.Q. Tranchida, F.A. Franchina, P. Dugo, L. Mondello, J. Chromatogr. A 1359 (2014) 271. [346] P.Q. Tranchida, F.A. Franchina, P. Dugo, L. Mondello, J. Chromatogr. A 1372 (2014) 236. [347] P.Q. Tranchida, M. Zoccali, F.A. Franchina, A. Cotroneo, P. Dugo, L. Mondello, J. Chromatogr. A 1314 (2013) 216. [348] R. Costa, C. Fanali, G. Pennazza, L. Tedone, L. Dugo, M. Santonico, D. Sciarrone, F. Cacciola, L. Cucchiarini, M. Dachà, L. Mondello, LWT-Food Sci. Technol. 60 (2015) 905. 195 12. References [349] G. Dugo, F.A. Franchina, M.R. Scandinaro, I. Bonaccorsi, N. Cicero, P.Q. Tranchida, L. Mondello, Food Chem. 142 (2014) 262. [350] P.Q. Tranchida, F.A. Franchina, M. Zoccali, S. Pantó, D. Sciarrone, P. Dugo, L. Mondello, J. Chromatogr. A 1278 (2013) 153. [351] D. Cavagnino, G. Zilioli, S. Trestianu, Poster entitled: "Instrumental Solutions For Quantitative Target Analysis In Comprehensive Two-Dimensional Gas Chromatography", 2004, http://unicam.hu/index.php?option=com_phocadownload&view=category&download=486:ga zkromatograf-anyagok&id=65:gazkromatografok&start=200&Itemid=38 [352] L.L.P. van Stee, U.A.Th. Brinkman, J. Chromatogr. A 1218 (2011) 7878. [353] M.K. van der Lee, G. van der Weg, W.A. Traag, H.G.J. Mol, J. Chromatogr. A 1186 (2008) 325. [354] B.J. Prazen, R.E. Synovec, B.R. Kowalski, Anal. Chem. 70 (1998) 218. [355] V.G. van Mispelaar, A.C. Tas, A.K. Smilde, P.J. Schoenmakers, A.C. van Asten, J. Chromatogr. A 1019 (2003) 15. [356] K.M. Pierce, L.F. Wood, B.W. Wright, R.E. Synovec, Anal. Chem. 77 (2005) 7735. [357] C.G. Johnson, G.S. Frysinger, R.K. Nelson, R.B. Gaines, N. Ohkouchi, C.M. Reddy, T.I. Eglinton, Mar. Chem. 83 (2003) 5. [358] R. Bro, Chemom. Intell. Lab. Syst. 38 (1997) 149. [359] N. Faber, R. Bro, P.K. Hopke, Chemom. Intell. Lab. Syst. 65 (2003) 119. [360] A.K. Smilde, R. Bro, P. Geladi, Multi-way analysis with applications in the chemical sciences. John Wiley & Sons, Chichester, Great Britain, 2004. [361] G.M. Escandar, A.C. Olivieri, N. Faber, H.C. Goicoechea, A. Muñoz de la Peña, R.J. Poppi, Trends Anal. Chem. 26 (2007) 752. [362] G. Tomasi, R. Bro, Comp. Stat. Data Anal. 50 (2006) 1700. [363] C.G. Fraga, B.J. Prazen, R.E. Synovec, Anal. Chem. 72 (2000) 4154. [364] L. Xie, P.J. Marriott, M. Adams, Anal. Chim. Acta 500 (2003) 211. [365] A.E. Sinha, C.G. Fraga, B.J. Prazen, R.E. Synovec, J. Chromatogr. A 1027 (2004) 269. [366] A.E. Sinha, J.L. Hope, B.J. Prazen, C.G. Fraga, E.J. Nilsson, R.E. Synovec, J. Chromatogr. A 1056 (2004) 145. [367] J.C. Hoggard, R.E. Synovec, Anal. Chem. 79 (2007) 1611. [368] J.C. Hoggard, R.E. Synovec, Anal. Chem. 80 (2008) 6677. [369] J.C. Hoggard, W.C. Siegler, R.E. Synovec, J. Chemom. 23 (2009) 421. [370] E.M. Humston, K.M. Dombek, B.P. Tu, E.T. Young, R.E. Synovec, Anal. Bioanal. Chem. 401 (2011) 2387. [371] J.C. Hoggard, J.H. Wahl, R.E. Synovec, G.M. Mong, C.G. Fraga, Anal. Chem. 82 (2010) 689. [372] H.A.L. Kiers, J.M.F. Ten Berge, R. Bro, J. Chemom. 13 (1999) 275. [373] R. Bro, C.A. Andersson, H.A.L. Kiers, J. Chemom. 13 (1999) 295. [374] Z. Zeng, J. Li, H.M. Hugel, G. Xu, P.J. Marriott, Trends Anal. Chem. 53 (2014) 150. [375] T. Skov, J.C. Hoggard, R. Bro, R.E. Synovec, J. Chromatogr. A 1216 (2009) 4020. [376] L.A. de Godoy, L.W. Hantao, M.P. Pedroso, R.J. Poppi, F. Augusto, Anal. Chim. Acta 699 (2011) 120. [377] H. Parastar, R. Tauler, Anal. Chem. 86 (2014) 286. [378] S. Peters, H.G. Janssen, G. Vivó-Truyols, Anal. Chim. Acta 799 (2013) 29. [379] H. Parastar, M. Jalali-Heravi, R. Tauler, Chemom. Intell. Lab. Syst. 117 (2012) 80. [380] J.A. Arancibia, P.C. Damiani, G.M. Escandar, G.A. Ibañez, A.C. Olivieri, J. Chromatogr. B 910 (2012) 22. [381] S.E. Reichenbach, M. Ni, V. Kottapalli, A. Visvanathan, Chemom. Intell. Lab. Syst. 71 (2004) 107. [382] X. Li, Z. Xu, X. Lu, X. Yang, P. Yin, H. Kong, Y. Yu, G. Xu, Anal. Chim. Acta 633 (2009) 257. [383] S.E. Reichenbach, V. Kottapalli, M. Ni, A. Visvanathan, J. Chromatogr. A 1071 (2005) 263. [384] S.E. Reichenbach, M. Ni, D. Zhang, J. Ledford, J. Chromatogr. A 985 (2003) 47. [385] G. Vivó-Truyols, H.G. Janssen, J. Chromatogr. A 1217 (2010) 1375. [386] I. Latha, S.E. Reichenbach, Q. Tao, J. Chromatogr. A 1218 (2011) 6792. [387] C. Cordero, E. Liberto, C. Bicchi, P. Rubiolo, S.E. Reichenbach, X. Tian, Q. Tao, J. Chromatogr. Sci. 48 (2010) 251. 196 12. References [388] C. Cordero, E. Liberto, C. Bicchi, P. Rubiolo, P. Schieberle, S.E. Reichenbach, Q. Tao, J. Chromatogr. A 1217 (2010) 5848. [389] T.G. Santos, K. Fukuda, M.J. Kato, A. Sartorato, M.C.T. Duarte, A.L.T.G. Ruiz, J.E. de Carvalho, F. Augusto, F.A. Marques, B.H. Sales Maia, Microchem. J. 115 (2014) 113. [390] L.F. Oliveira, S.C.G.N. Braga, P.R. Filgueiras, F. Augusto, R.J. Poppi, Talanta 129 (2014) 303. [391] C. Cordero, C. Cagliero, E. Liberto, L. Nicolotti, P. Rubiolo, B. Sgorbini, C. Bicchi, J. Chromatogr. A 1318 (2013) 1. [392] G. Purcaro, C. Cordero, E. Liberto, C. Bicchi, L.S. Conte, J. Chromatogr. A 1334 (2014) 101. [393] S.E. Reichenbach, X. Tian, Q. Tao, E.B. Ledford, Jr., Z. Wu, O. Fiehn, Talanta 83 (2011) 1279. [394] C.H.V. Fidelis, P.T.B. Sampaio, P.M. Krainovic, F. Augusto, L.E.S. Barata, Microchem. J. 109 (2013) 73. [395] L.W. Hantao, B.R. Toledo, F.A. de Lima Ribeiro, M. Pizetta, C.G. Pierozzi, E.L. Furtado, F. Augusto, Talanta 116 (2013) 1079. [396] J. Omar, M. Olivares, J.M. Amigo, N. Etxebarria, Talanta 121 (2014) 273. [397] S. Hashimoto, Y. Takazawa, A. Fushimi, K. Tanabe, Y. Shibata, T. Ieda, N. Ochiai, H. Kanda, T. Ohura, Q. Tao, S.E. Reichenbach, J. Chromatogr. A 1218 (2011) 3799. [398] Y. Zushi, S. Hashimoto, M. Tamada, S. Masunaga, Y. Kanai, K. Tanabe, J. Chromatogr. A 1338 (2014) 117. [399] M.S. Alam, C.E. West, A.G. Scarlett, S.J. Rowland, R.M. Harrison, Atmos. Environ. 65 (2013) 101. [400] M.R. Djokic, T. Dijkmans, G. Yildiz, W. Prins, K.M. Van Geem, J. Chromatogr. A 1257 (2012) 131. [401] G.D. Wardlaw, J.S. Arey, C.M. Reddy, R.K. Nelson, G.T. Ventura, D.L. Valentine, Environ. Sci. Technol. 42 (2008) 7166. [402] N.G.S. Mogollon, F.A.d.L. Ribeiro, M.M. Lopez, L.W. Hantao, R.J. Poppi, F. Augusto, Anal. Chim. Acta 796 (2013) 130. [403] B.J. Prazen, C.A. Bruckner, R.E. Synovec, B.R. Kowalski, J. Microcol. Sep. 11 (1999) 97. [404] R.E. Mohler, K.M. Dombek, J.C. Hoggard, E.T. Young, R.E. Synovec, Anal. Chem. 78 (2006) 2700. [405] J.L. Hope, B.J. Prazen, E.J. Nilsson, M.E. Lidstrom, R.E. Synovec, Talanta 65 (2005) 380. [406] G. Guiochon, Chromatogr. Rev. 8 (1966) 1. [407] S. Vezzani, G. Castello, D. Pierani, J. Chromatogr. A 811 (1998) 85. [408] G. Castello, S. Vezzani, P. Moretti, J. Chromatogr. A 742 (1996) 151. [409] H. Snijders, H.G. Janssen, C. Cramers, J. Chromatogr. A 718 (1995) 339. [410] W. Welthagen, J. Schnelle-Kreis, R. Zimmermann, J. Chromatogr. A 1019 (2003) 233. [411] J. Beens, H.G. Janssen, M. Adahchour, U.A.Th. Brinkman, J. Chromatogr. A 1086 (2005) 141. [412] L.M. Blumberg, J. High Resolut. Chromatogr. 20 (1997) 597. [413] P.J. Schoenmakers, P. Marriott, J. Beens, LC-GC Eur. 16 (2003) 335. [414] J.C. Giddings, J. High Resolut. Chromatogr. 10 (1987) 319. [415] P. Korytár, C. Danielsson, P.E.G. Leonards, P. Haglund, J. de Boer, U.A.Th. Brinkman, J. Chromatogr. A 1038 (2004) 189. [416] P. Korytár, P.E.G. Leonards, J. de Boer, U.A.Th. Brinkman, J. Chromatogr. A 1086 (2005) 29. [417] D. Ryan, P. Morrison, P. Marriott, J. Chromatogr. A 1071 (2005) 47. [418] R. Ong, P. Marriott, P. Morrison, P. Haglund, J. Chromatogr. A 962 (2002) 135. [419] P. Haglund, M. Harju, R. Ong, P. Marriott, J. Microcol. Sep. 13 (2001) 306. [420] S. Peters, G. Vivó-Truyols, P.J. Marriott, P.J. Schoenmakers, J. Chromatogr. A 1146 (2007) 232. [421] T. Dutriez, M. Courtiade, D. Thiébaut, H. Dulot, F. Bertoncini, J. Vial, M.-C. Hennion, J. Chromatogr. A 1216 (2009) 2905. [422] M.R. Schure, J. Microcol. Sep. 9 (1997) 169. [423] J.C. Giddings, Anal. Chem. 32 (1960) 1707. 197 Samenvatting Dit proefschrift beschrijft de ontwikkeling en praktische toepassing van verscheidene multidimensionele en gekoppelde technieken—alle gebaseerd op het gebruik van gaschromatografie (GC) als scheidingsmethode, en met de identificatie van onbekende verbindingen in complexe monsters als een belangrijk oogmerk. Bij het uitwerken van het programma nam—naast atoomemissie (AED)- en massaspectrometrische (MS) detectie—de efficiënte verwerking van de (vele) verkregen gegevens en het gebruik van tweedimensionele GC (GC×GC) een centrale plaats in. Hoofdstuk 2 beschrijft in het kort de ontwikkeling van de AED en verschillende soorten toepassingen (bijv. water, grond, lucht, voedsel en petrochemie) waarin de unieke eigenschappen van AED gekoppeld aan GC naar voren komen. In een (commerciële) detector wordt een helium plasma gebruikt, waarin de moleculen van de in een monster aanwezige verbindingen worden geatomiseerd, en de atomen vervolgens aangeslagen. Bij het terugvallen naar de grondtoestand wordt een voor ieder element karakteristieke straling uitgezonden; dit geeft AED detectie een zeer grote selectiviteit en de mogelijkheid om stofonafhankelijk te ijken. De belangrijkste conclusie is dat GC–AED uitermate geschikt is voor selectief screenen en het identificeren van onbekende verbindingen in complexe monsters. In Hoofdstuk 3 wordt de opzet van GC met simultane AED en MS (in dit proefschrift dikwijls vluchttijd MS, ToF-MS) detectie beschreven. De lengtes van de koppelingscapillairen en de plaatsing in de verhitte zones zijn zodanig berekend dat een eluerende piek tegelijkertijd door de AED en de MS gedetecteerd wordt gedurende het hele temperatuurprogramma. De analysestrategie wordt gedemonstreerd aan de hand van niet-doelstof analyse van rioolwater. Eén van de conclusies die uit de bestudering van de chloor-, totaal ion- (TIC) en geëxtraheerde ion chromatogrammen (EIC) kan worden getrokken is dat de exacte retentietijd-correlatie zoals verkregen met de gebruikte opstelling, essentieel is om de gebruikte strategie toe te passen. In Hoofdstuk 4 is de correlatie van AED en MS data toegepast op tweedimensionele GC data. In dit geval zijn de data opgenomen op twee aparte GC×GC systemen en gecorreleerd door middel van grafische manipulatie van de chromatogrammen. Een van de belangrijkste zaken die gedemonstreerd wordt is dat de AED, ondanks de lage acquisitiefrequentie, toch gebruikt kan worden voor detectie in GC×GC. Aan de hand van de analyse van een hoogzwavelige ruwe petrochemische olie wordt tevens aangetoond dat de grafische correlatiemethode uitstekende praktische resultaten oplevert. Een interessant voorbeeld was de detectie van een gehele klasse van N- Samenvatting houdende verbindingen, dimethylcarbazolen, waarbij de zeer precieze correlatie van de MS en AED gegevens van doorslaggevend belang bleek te zijn. In de onderzoeken beschreven in de Hoofdstukken 5–7 zijn verschillende van de eerder gebruikte technieken toegepast op de analyse van riool-, oppervlakte- en poriewater monsters. De aanleiding voor deze studies waren toxiciteitsonderzoeken door Rijkswaterstaat (voorheen onder de naam RIZA) in oppervlaktewater, een belangrijke bron voor drinkwater. De totale toxiciteit in bioassays met bijvoorbeeld Daphnia magna kon niet worden verklaard door de eigenschappen en concentraties van de in routineanalyes gevonden contaminanten. Eén hypothese is dat de toxiciteit wordt veroorzaakt door lipofiele verbindingen die kunnen bioaccumuleren in vetweefsel van organismen. Vanwege hun lage wateroplosbaarheid zijn deze stoffen echter meestal in zeer lage concentraties aanwezig in watermonsters en worden daarom niet gedetecteerd tijdens routineanalyses. De detectie en identificatie van bovengenoemde stoffen is in de huidige studies mogelijk gemaakt met behulp van biomimetische extractie. Ter bemonstering wordt een houder die gevuld is met, of bestaat uit, een vetachtig hydrofoob materiaal in water gehangen. Analoog aan wat gebeurt bij de bioaccumulatie in organismen zullen de hydrofobe verbindingen nu veel sterker vanuit het water naar de vette fase migreren dan meer hydrofiele analieten. De bemonsteringstijd varieert in de regel van een aantal dagen tot verscheidene weken; in veldstudies ontstaat dan een evenwicht in de concentratieverdeling tussen het water en de hydrofobe fase. In het laboratorium kan dit worden nagebootst als het volume van de hydrofobe fase veel kleiner is dan het watervolume. Naast de aanzienlijke verhoging van de concentraties van lipofiele verbindingen is de afwezigheid van biotransformatie een belangrijk voordeel. Diverse biomimetische samplers zijn geëvalueerd, zowel in het laboratorium als in het veld, en de resultaten zijn ook vergeleken met conventionele ‘volledige’ vaste-fase extractie (SPE). In conventionele extracten van poriewater uit een grote veldstudie zijn met GC– AED/MS onder andere ftalaten, zwaardere koolwaterstoffen, polycyclische geurstoffen en polycyclische aromatische koolwaterstoffen (PAKs) aangetoond; de berekende bijdrage in de toxiciteit was echter verwaarloosbaar. Een andere veldstudie in rioolwater influent en effluent liet zien dat conventionele SPE en biomimetische extractie complementair zijn. Naast de semi-polaire verbindingen die met SPE werden gevonden, werden in het biomimetische extract een gehele reeks lipofiele geurstoffen, PAKs, ftalaten en andere klassen van verbindingen geïdentificeerd. Met GC–AED/MS werden er meer dan 20 gehalogeneerde verbindingen gedetecteerd. Een veldstudie in oppervlaktewater, waarin ongeveer 150 verbindingen werden geïdentificeerd, toonde 200 Samenvatting dezelfde complementariteit: veel lipofiele verbindingen zoals bijvoorbeeld PAKs worden alleen gevonden in de biomimetische extracten. Hoofdstuk 8 beschrijft de analyse van vluchtige organische verbindingen in boven Kreta genomen luchtmonsters. Veldmetingen werden uitgevoerd met GC×GC–FID (vlam ionisatie detector); van dezelfde monsters werd ook een deel opgevangen op vaste-fase kolommetjes, die later in het laboratorium aan GC×GC–MS analyse werden onderworpen ten behoeve van identificatie. Aan de hand van MS bibliotheek data en retentie indices zijn 250 verbindingen geïdentificeerd, die daarna op basis van de FID data werden gekwantificeerd. Dit type luchtmonsters blijkt zeer complex te zijn, en het gebruik van GC×GC is absoluut noodzakelijk om tot betrouwbare resultaten te komen. Hoofdstuk 9 beschrijft verscheidene methoden voor geautomatiseerde piekherkenning in GC×GC—een herkenningsmethode die vooral nodig is bij de analyse van (grote) series complexe monsters, met name wanneer het opsporen van onbekende verbindingen een rol speelt. Er bestaat een duidelijk onderscheid tussen methoden gebaseerd op de gebruikelijke 1D-integratie, en methoden die de data behandelen als echte 2D data. De principes van de diverse typen methode worden besproken, evenals de te volgen experimentele procedures en de eisen die moeten worden gesteld aan de input en output van de data. Bij dit alles speelt ook het soort analyse waaraan wordt gewerkt—bijv. doelstofanalyse of algemene screening—een duidelijke rol. Een aantal relevante toepassingen wordt kort besproken. De belangrijkste conclusie is dat voor algemeen gebruik alleen procedures waarvoor commerciële software beschikbaar is—zoals de watershed en de ChromaTOF methoden—aanbevolen kunnen worden. In Hoofdstuk 10 wordt een nieuwe, in-huis ontworpen methode voor piekherkenning beschreven. Een van de stappen in alle op 1D-analyse gebaseerde methoden is het bij elkaar voegen van de verschillende ‘peaklets’ die van één enkel analiet afkomstig zijn. De juiste vaststelling van de grenzen waarbinnen aan elkaar gerelateerde ‘peaklets’ kunnen voorkomen, kan dit proces nauwkeuriger laten verlopen. Op basis van eenvoudige thermodynamische vergelijkingen kan de tweedimensionele piekvorm—die varieert afhankelijk van de retentietijden—binnen relatief nauwe grenzen berekend worden. Vervolgens worden ‘peaklets’ binnen deze grenzen bij elkaar gevoegd op basis van MS bibliotheek identificatie. De methode, die wordt toegelicht aan de hand van voorbeelden, blijkt in staat een overzichtelijk en betrouwbaar beeld te geven van de— soms grote aantallen—verbindingen aanwezig in complexe monsters. Hoofdstuk 11 behandelt een onderzoek naar de scheidingscapaciteit van vijf verschillende GC×GC-kolomcombinaties; zowel de scheiding van (een groot aantal) afzonderlijke componenten als die van klassen van verbindingen is bekeken. Voor de studie werden zo’n 150 analieten, behorend tot negen verschillende stofklassen, 201 Samenvatting uitgezocht. Bij een dergelijk groot aantal stofklassen blijkt ook in GC×GC geen enkele kolomcombinatie complete groepsscheidingen te kunnen bewerkstelligen. Anderzijds leveren de analyses wel een duidelijk beeld op van het gebruik van de 2D scheidingsruimte. Dit maakt het mogelijk snel en efficiënt een voorspelling te doen over de bij specifieke kolomcombinaties te verwachten kwaliteit van een scheiding. 202 List of publications [1] G.W. Somsen, L.L.P. van Stee, C. Gooijer, U.A.Th. Brinkman, N.H. Velthorst, T. Visser, Anal. Chim. Acta 290 (1994) 269. Isomer and congener identification of chlorinated pyrenes by narrow-bore liquid chromatography-Fourier transform infrared spectrometry. [2] J.G.M. Bessems, J.M. te Koppele, P.A. Van Dijk, L.L.P. van Stee, J.N.M. Commandeur, N.P.E. Vermeulen, Xenobiotica 26 (1996) 647. Rat liver microsomal cytochrome P450-dependent oxidation of 3,5-disubstituted analogues of paracetamol. [3] J.G.M. Bessems, L.L.P. van Stee, J.N.M. Commandeur, E.J. Groot, N.P.E. Vermeulen, Toxicol. Vitro 11 (1997) 9. Cytotoxicity of paracetamol and 3,5- dihalogenated analogues: Role of cytochrome P-450 and formation of GSH conjugates and protein adducts. [4] L.L.P. van Stee, P.E.G. Leonards, R.J.J. Vreuls, U.A.Th. Brinkman, Analyst 124 (1999) 1547. Identification of non-target compounds using gas chromatography with simultaneous atomic emission and mass spectrometric detection (GC-AED/MS): Analysis of municipal wastewater. [5] J. Dallüge, L.L.P. van Stee, X. Xu, J. Williams, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, J. Chromatogr. A 974 (2002) 169. Unravelling the composition of very complex samples by comprehensive gas chromatography coupled to time- of-flight mass spectrometry: Cigarette smoke. [6] L.L.P. van Stee, U.A.Th. Brinkman, H. Bagheri, Trends Anal. Chem. 21 (2002) 618. Gas chromatography with atomic emission detection: a powerful technique. [7] L.L.P. van Stee, P.E.G. Leonards, W.M.G.M. van Loon, A.J. Hendriks, J.L. Maas, J. Struijs, U.A.Th. Brinkman, Water Res. 36 (2002) 4455. Use of semi- permeable membrane devices and solid-phase extraction for the wide-range screening of microcontaminants in surface water by GC-AED/MS. [8] M. Adahchour, L.L.P. van Stee, J. Beens, R.J.J. Vreuls, M.A. Batenburg, U.A.Th. Brinkman, J. Chromatogr. A 1019 (2003) 157. Comprehensive two- dimensional gas chromatography with time-of-flight mass spectrometric detection for the trace analysis of flavour compounds in food. [9] P. Korytár, L.L.P. van Stee, P.E.G. Leonards, J. de Boer, U.A.Th. Brinkman, J. Chromatogr. A 994 (2003) 179. Attempt to unravel the composition of toxaphene by comprehensive two-dimensional gas chromatography with selective detection. [10] J. Lahr, J.L. Maas-Diepeveen, S.C. Stuijfzand, P.E.G. Leonards, J.M. Druke, S. Lucker, A. Espeldoorn, L.C.M. Kerkum, L.L.P. van Stee, A.J. Hendriks, Water List of publications Res. 37 (2003) 1691. Responses in sediment bioassays used in the Netherlands: can observed toxicity be explained by routinely monitored priority pollutants? [11] L.L.P. van Stee, J. Beens, R.J.J. Vreuls, U.A.Th. Brinkman, J. Chromatogr. A 1019 (2003) 89. Comprehensive two-dimensional gas chromatography with atomic emission detection and correlation with mass spectrometric detection: principles and application in petrochemical analysis. [12] X. Xu, L.L.P. van Stee, J. Williams, J. Beens, M. Adahchour, R.J.J. Vreuls, U.A.Th. Brinkman, J. Lelieveld, Atmos. Chem. Phys. Discuss. 3 (2003) 1139. Comprehensive two-dimensional gas chromatography (GC×GC) measurements of volatile organic compounds in the atmosphere. [13] M.M. Koek, B. Muilwijk, L.L.P. van Stee, Th. Hankemeier, J. Chromatogr. A 1186 (2008) 420. Higher mass loadability in comprehensive two-dimensional gas chromatography-mass spectrometry for improved analytical performance in metabolomics analysis. [14] L.L.P. van Stee, U.A.Th. Brinkman, J. Chromatogr. A 1186 (2008) 109. Developments in the application of gas chromatography with atomic emission (plus mass spectrometric) detection. [15] L.L.P. van Stee, U.A.Th. Brinkman, J. Chromatogr. A 1218 (2011) 7878. Peak clustering in two-dimensional gas chromatography with mass spectrometric detection based on theoretical calculation of two-dimensional peak shapes: The 2DAid approach. [16] P.J. Boogaard, K.O. Goyak, R.W. Biles, L.L.P. van Stee, M.S. Miller, M.J. Miller, Regul. Toxicol. Pharmacol. 63 (2012) 69. Comparative toxicokinetics of low-viscosity mineral oil in Fischer 344 rats, Sprague-Dawley rats, and humans—implications for an Acceptable Daily Intake (ADI). [17] L.L.P. van Stee, U.A.Th. Brinkman, Trends Anal. Chem. 83 (2016) 1. Peak detection methods for GC×GC: an overview. 204 Dankwoord Allereerst wil ik Udo bedanken. Jij bent de constante factor geweest gedurende het promotietraject, en ondanks je altijd drukke programma (nu nog steeds) heb je altijd de tijd genomen om manuscripten te corrigeren, mee te denken in de koers van de onderzoeken, en de resultaten kort en bondig op papier te zetten. Meerdere malen zijn dus ook flinke stukken manuscript waar vele weken, zo niet langer, uitzoekwerk in zat, gesneuveld. Dit was soms even slikken, maar uiteindelijk waren we het over het eindresultaat altijd eens. René, onder jouw begeleiding ben ik begonnen, en jouw hulp, enthousiasme, theoretische en praktische kennis heb ik enorm gewaardeerd; helaas namen andersoortige werkzaamheden steeds meer van jouw tijd in beslag. Pim, jij nam de directe begeleiding over en je was bedreven in het opzetten van samenwerkings- projecten, en dankzij jou hebben we kunnen deelnemen aan het grote waterscreeningsproject. Helaas kwam aan onze samenwerking na enige tijd een eind omdat je van werkgever veranderde. Hans-Gerd, via verschillende directe en indirecte lijnen kreeg je periodieke updates van mijn onderzoek en gezien onze gedeelde interessegebieden zou je een waardige copromotor zijn. Nu, door de loop der tijd, zijn de rollen tussen jou en Udo omgedraaid. Bedankt voor de levendige discussies en dat je mijn promotor wilt zijn. Bij deze ook dank voor de bijdrage van de leden van de leescommissie, Govert, Michel, Jens, Stefan en Jacob. Daarnaast dank aan de staf van AAC/ACAS Freek, Cees, Hubertus, Henk, Nel, Gerd, Herman, Gisèle, Dik, Pim (postuum) voor jullie inzet om een geweldige groep onderzoekers bij elkaar te brengen en te faciliteren. Naast de gestaag wisselende AIO- kern waren dit vele tijdelijke studenten en buitenlandse gasten uit o.a. Duitsland, Japan, Iran, Finland, China en Spanje. Een mooie enthousiaste club die bij de afdelingsuitjes (Eva bedankt) een aardig restaurant kon vullen. Het was een mooie tijd die ik niet zal vergeten. Daarnaast veel dank aan de vakmensen van de elektronische en mechanische werkplaatsen: Dick, Klaas, Roald, Dirk, Lex, Joost. Van complexe reparaties, selectie van materialen voor speciale omstandigheden, tot de ambitieuze bouw van een plasma- MS interface, een van de projecten die we helaas niet hebben kunnen afronden. Dank aan Michel en Sjoerd voor de AEDs die ik mocht kannibaliseren voor onderdelen. Dank aan Ralf en het Leco-team voor de hulp als we er zelf niet meer uitkwamen met de Dankwoord Pegasus. En, Henk, Hadil en Marc, het is niet in het boekje gekomen maar bedankt voor de samenwerking in het OWWA project. Special thanks to Jonathan and Xiaobin for your cooperation in the air analysis project. Those were long days, but well worth it. Dank aan Jaap, Hannie, Jan, Joost voor de samenwerking in het waterproject. Peter, bedankt voor de leuke samenwerking in het toxafenen project: naast het GC×GC werk was het een mooie inspiratie om lekker te programmeren. Jan bedankt voor jouw kennisoverdracht en wijze lessen over GC×GC. Jens, aan onze samenwerking denk ik met heel veel plezier terug. Van de eerste-jaars practica tot het in leven brengen en houden van allerlei MSsen en modulatoren; vroeg naar huis zat er vaak niet in. Je bracht mij o.a. de kneepjes van de Pegasus GC×GC bij, en je was fervent c’t lezer en computer freak. Eén van onze verzinsels—het ‘on-line’ monitoring systeem met webcam-Gotcha-foto’s op de ftp-server—betekende wel meerdere nachtelijke tripjes naar de VU om de boel weer snel op te starten. Herman, je was een waardige opvolger van Ben als massa-spectrometrist die velen van ons heeft geholpen met jouw inzichten. Daarnaast liep het puzzelen aan de massa spectra vaak uit op aangename existentiële discussies. Joost, onder andere electronicus, geduldige uitzoeker en ‘die hard’ computeraar; van het samen beheren van het ‘netwerk binnen het netwerk’ en oplossen van andere computerzaken heb ik erg genoten. Voor de speciale vriendschap: Jens, Isabel, Herman, Reyer, Joost, Mohamed, Maria en Eva. Verder dank aan alle andere AIOs, externe AIOs, ex-AIOs, mensen uit het bedrijfsleven, gasten, studenten, stagiairs, Lunteren- en Riva-gangers enz. die ik in deze mooie tijd heb leren kennen (op volgorde van achternaam): Lawrence, Arjen, Habib, Max, Marco, Jan, Tjipke, Peter, Jan, Arjen, Alois, Jeroen, Gisele, Rico, Evtim, Wil, Carmen, Conchi, Henk-Jan, Leonie, Petra, Thomas, Yan, Joshi, Junko, Monique, Ariadne, Edwin, Gerard (postuum), Johannes, Kick, Niels, Eric, Ralf, Minna, Dik, Richard, Richard, Maud, Sander, Sjaak, Peter, Hans, Maria, Marleen, Koos, Olaf, Thomas, Arjan, Gregor, Philip, Diego, Wim, Valentijn, Frank, Harm, Rahma, Eva, Silvia, Lourdes, Eva, Martijn, Edwin, Patrick, Johan, Huib, Reza, Soheila, Mirka, Lineke, Philip, Aike, Rob, Natasja, Arjan. Kortom, dank aan iedereen die ik genoemd heb, en ook degenen die ik vergeten ben te noemen, voor de mooie inspirende tijd en samenwerking. En als laatste, maar niet als minste: dank aan Aletta, Tobias en Lara voor jullie geduld en steun. 206