Comprehensive two-dimensional GC after HPLC preseparation for the characterization of aromatic hydrocarbons of mineral oil origin in contaminated sunflower oil



Many foods are contaminated with mineral oil at concentrations in the order of or above 10 mg/kg. Often, these mineral oils are of technical grade and contain around 20–30% aromatic hydrocarbons, as previously shown by an on-line HPLC-GC-flame ionization detection method for determining in foods the sum of the mineral oil aromatic hydrocarbons (ranging from 1 to at least 5 aromatic rings). Now, a comprehensive two-dimensional GC (GC×GC) method was added for characterizing these aromatics by ring number and degree of alkylation. In contrast to the polycyclic aromatic hydrocarbons originating from pyrolysis, mineral oil aromatic hydrocarbons are highly alkylated and form extremely complex mixtures. Through MS and addition of standards, sectors in the GC×GC plots were allocated to 1 to 5 ring aromatics in order to estimate the relative abundance of each group. The quantitative composition is approximated by integration of 2nd dimension chromatograms laid as a grid over the three-dimensional hump of unresolved hydrocarbons of the GC×GC-flame ionization detection plot. The procedure is applied to Ukrainian sunflower oils contaminated with mineral oil.

1 Introduction

During winter/spring 2007/2008, in the Ukraine nearly 100 000 tons of sunflower oil were contaminated with often more than 1000 mg/kg of mineral oil. The source and the identity of the mineral oil product have not been officially confirmed, but it is commonly assumed that a base oil for manufacturing lubricating oils was added as a fraud. In April/May 2008, the European Food Safety Authority (EFSA) advised the EU Commission about the risk to human health 1. It concluded that “exposure to such oil, although undesirable, would not be a public health concern.” This evaluation was based on the analytical data then available, and since this data was restricted to saturated hydrocarbons, the EFSA formed its opinion by comparing the amounts and composition of these with the acceptable daily intakes of the Joint FAO/WHO Expert Committee on Food Additives (JECFA) 2.

The assumption that the added mineral oil consisted of highly refined white paraffin oil was not really convincing: for a fraud, a cheaper mineral oil would be used. Technical oils contain aromatic hydrocarbons at concentrations of 10–40% 3, and these are probably more relevant for the risk assessment than the saturated hydrocarbons. In the meantime, an on-line HPLC-GC-flame ionization detection (FID) method has been developed for the determination of the sum of all mineral oil aromatic hydrocarbons (MOAH; ranging from 1 to at least 5 aromatic rings) in edible oils or food extracts 4. Together with the analysis of the mineral oil saturated hydrocarbons (MOSH), this enabled the characterization of the mineral oil quality in a food. It turned out that the mineral oil contaminating the Ukrainian sunflower oil contained 17–37% MOAH 5. The highest MOAH concentration measured in refined sunflower oil was 1800 mg/kg.

The MOAH differ from the widely analyzed polycyclic aromatic hydrocarbons (PAHs) formed during combustion or heat treatment found in smoked or roasted foods: while the PAHs consist of a limited number of primarily non-alkylated compounds and can be analyzed individually, e.g. using mass spectrometry (MS) with selected ion monitoring 6, the MOAH are highly alkylated and form broad humps of unresolved components, similar to the branched and cyclic MOSH 7.

The technical constraints for the MOAH analysis resemble those of the MOSH analysis. The response of specific detectors, like the mass spectrometer 8 or spectroscopic HPLC detectors, cannot be calibrated, since aromatics of differing alkylation perform spectroscopically different and in MS there are no fragments common to groups of components with an intensity enabling calibration for groups. The only system providing virtually the same response for all aromatic hydrocarbons is FID. Since it is not selective, MOAH are analyzed as a sum of all or as groups of given types.

A complex on-line HPLC-HPLC-GC-FID method enabling MOAH analysis in foods by ring number was described by Moret et al.9, 10. The hydrocarbons were isolated from edible oil or food extracts by a large HPLC silica gel column. The eluent of this fraction was evaporated in a miniaturized chamber working on the principle of concurrent evaporation into a vacuum 11. The aromatic hydrocarbons were further preseparated by number of rings on an amino HPLC column and resulting fractions analyzed by GC-FID. This method revealed the presence of mineral polycyclic aromatics (at least two rings) in the order of 10 mg/kg in various foods.

The more recent on-line HPLC-GC-FID method 4 was intended for routine analysis and does not separate the MOAH into groups. The edible oil or crude food extract is injected into a 25 cm×2 mm id silica gel HPLC column, which retains the lipids in the upper part and separates the MOAH from the MOSH in the lower. To achieve complete separation of the MOSH and the MOAH, a silica gel with a high retention power for the MOAH (Lichrospher 60) was selected. A steep gradient with dichloromethane produced a narrow fraction of the MOAH ranging from the highly alkylated benzenes up to perylene. As this preseparation provided no selectivity for removing certain olefins present as food components, such as squalene, its isomerization products formed during raffination of edible oils, sterenes (dehydroxylation products from sterols) and carotenoids, the polarity of these was enhanced by epoxidation to delay their elution beyond that of the MOAH.

For risk evaluation, the concentration of the sum of all MOAH might not be satisfactory. The method of Moret et al. used HPLC to group the MOAH by ring number. In the meantime, comprehensive two-dimensional GC (GC×GC) was established 12, 13, which is an attractive alternative because of its simplicity, the enhanced sensitivity resulting from focusing by modulation and the additional information provided on the extent of alkylation. Mineral oil analysis is the most widely described application of GC×GC, though mainly on products like gasoline, heavy naphtha or kerosene with a lower molecular mass than the food contaminants most frequently found.

GC×GC does not enable a complete separation of MOSH and MOAH, since naphthenic hydrocarbons (cyclic saturates) tend to be coeluted with highly alkylated aromatic components. This prompted Vendeuvre et al. 14 to use an “olefin trap”, a silver-modified silica gel precolumn used in GC mode, to complete the separation. Such olefin traps are not suitable for high temperature GC, however.

Edam et al. 15, 16 described the problem as a dilemma: phenyl polysiloxanes as GC stationary phases provide group-type separation of both aromatics and naphthenics, but by separating naphthenics from paraffins they also cause the naphthenics to be partly coeluted with aromatics. Edam et al. solved the problem by an off-line LC preseparation on an amino-derivatized silica gel. A similar route was chosen for this work.

Two aspects rendered the task particularly demanding. Firstly, the typically high molecular mass of the mineral oils contaminating foods renders the separation of the naphthenes and the highly alkylated MOAH more difficult, both in LC and GC. Second, highly refined paraffin oils may contain little aromatics, and the MOAH have to be separated from the MOSH to the extent that the tail of the large MOSH peak corresponds to less than 1% in the MOAH fraction. Such increased demands can be met by HPLC preseparation.

The method described in this paper is tied up to that for determining the sum of the MOAH 4 and aims at characterizing the composition of the MOAH: the HPLC fraction of the MOAH is further separated by GC×GC, with MS for identification and FID for routine analysis and quantitation. Presently, it is performed in off-line mode, but it could easily be converted on-line. Because of the complexity of the mixture, it is not possible to identify single components, but for a risk evaluation, considered to be the next step in assessing the relevance of this food contamination, identification and toxicological evaluation of single components is not promising anyway: there are far too many components and they are not available as pure substances. It is assumed that testing will occur with mixtures and that the mixtures need to be characterized in more general terms, such as by number of aromatic rings and molecular mass (extent of alkylation). Accordingly, GC×GC plots are interpreted by specification of regions in which signals are summed up.

2 Experimental

2.1 Chemicals and solutions

Hexane from Brenntag (Schweizerhall AG, Basel, Switzerland) was redistilled after purification through a silica gel column (450 g silica gel activated at 400°C/15 h for 10 L hexane). Dichloromethane was from J.T. Baker (Deventer, Holland). Silica gel 60, 0.063–0.200 mm and biphenyl (BP) were from Merck (Darmstadt, Germany). Benzene (p.A.), 1,1,2-trichloroethane, n-dodecane (C12), n-tetradecane (C14), n-hexadecane (C16), hexyl benzene (6B), nonyl benzene (9B), perylene (Per), 1,3,5-tri-tert. butyl benzene (TBB), 5-alpha-cholestane (Cho), phenanthrene, 1-chloro hexane, 1-chloro octadecane, aluminum chloride and PS-255 (a dimethyl polysiloxane) were from Fluka/Sigma-Aldrich (Buchs, Switzerland). SOP 50, a symmetric 50% phenyl methyl polysiloxane, was a gift from W. Blum (Novartis, Basel, Switzerland). The contaminated Ukrainian sunflower oils were the same as those described in 5.

The stock solutions of internal standards contained 100 mg of the components in 10 mL 1,1,2-trichloroethane: mixture 1, 6B, 9B and BP; mixture 2, C12, C14 and C16; mixture 3, Per and Cho. The internal standard solution contained 100 μL mixture 1, 300 μL mixture 2 and 500 μL mixture 3 in 10 mL 1,1,2-trichloroethane.

Mixtures of hexyl and octadecyl benzenes were obtained by Friedel Kraft reaction of benzene with the corresponding alkyl chlorides: to 5 mmol benzene and 0.1 mmol aluminum chloride cooled in an ice bath, 5 mmol 1-chloro alkane was added in portions, monitoring the onset of the reaction by the hydrogen chloride leaving the condenser. The product was picked up in hexane and washed with water. Octadecyl phenanthrene was obtained in the same way, starting from a 10% solution of phenanthrene in benzene.

2.2 Sample preparation

Mineral oils were diluted in two steps: 500 mg was filled up to 10 mL hexane and 100 μL of the resulting solution diluted to 10 mL hexane together with 100 μL internal standard solution. Of edible oils and fats, 2 g were weighed in and filled up to 10 mL with hexane together with 20 μL internal standard solution. Isolation of the mineral oil from the foods and preseparation of MOSH and MOAH involved an instrument for automated on-line HPLC-GC-FID from Thermo Scientific (Milano, Italy) consisting of a TriPlus autosampler with a 100 μL syringe, a Phoenix 40 dual syringe pump with three switching valves, a microUVIS 20 UV detector and a TRACE gas chromatograph equipped with on-column injector, early vapor exit and a switching valve for the transfer.

Of the samples, 50–90 μL was injected onto a 25 cm×2 mm id HPLC column packed with Lichrospher Si 60, 5 μm (Grom, Rottenburg-Hailfingen, Germany) and chromatographed at 300 μL/min with a gradient starting with hexane (0.1 min) and reaching 30% dichloromethane after 1.5 min. The column was backflushed 6 min after injection with dichloromethane at 500 μL/min for 9 min, then reconditioned in normal flow direction with hexane at 500 μL/min for 10 min and at 300 μL/min up to the subsequent injection.

From HPLC, the MOSH were eluted from 2.0 to 3.5 min, the MOAH from 3.75 to 5.5 min. Instead of transferring them to GC for the determination of the sum of the MOSH and the MOAH 4, the fraction was collected in an 1.5 mL autosampler vial from the dismounted transfer line and reconcentrated to 20–100 μL (dilution with MTBE). Before use, the autosampler vials were cleaned by heating for 5 h at 350–400°C.

2.3 GC×GC

GC×GC involved a TRACE gas chromatograph from Thermo Scientific equipped with a PTV injector containing an “on-column liner”, a cryogenic jet modulator and FID. It was coupled to a quadrupole mass spectrometer DSQ II (Thermo Scientific). Ten microliter was injected on-column onto a GC system composed of a 1 m×0.53 mm id deactivated precolumn followed by a 20 m×0.25 mm id 1st dimension separation column coated with a 0.12 μm film of PS-255 and a 1.5 m×0.15 mm id 2nd dimension column coated with a 0.075 μm film of SOP-50. For GC×GC-MS, helium was used as carrier gas at 2.0 mL/min constant flow; the oven was programmed from 60°C (1.5 min) at 5 /min to 350°C (2 min). The MS transfer line was kept at 340°C. Temperature program of the PTV injector: 60°C (1 min), 0.2°/s to 350°C (10 min); flow rate through the split outlet, 15 mL/min. For GC-FID, hydrogen was used at 1.5 mL/min constant flow; the temperature programs of the GC oven and the PTV injector remained the same. The modulation time was 4 s. The scan range comprised 89–450 amu and the scan rate was at 11 111 amu/s, resulting in 12.6 scans/s. Data was collected after a solvent delay of 2.5 min. FID chromatograms were recorded at 100 Hz.

3 Results

3.1 Concept of the method

As an introductory illustration of the subject, Fig. 1 shows standard GC-FID chromatograms and, below, the corresponding GC×GC-FID contour plots for two mineral oil products. The chromatographic system was the same for both, i.e. consisted of the combined non-polar first dimension column and the small second dimension column coated with the 50% phenyl methyl polysiloxane. For the GC×GC-FID plot, the effluent from the 1st dimension column was recollected into small portions by cryofocusing and released into the small 2nd dimension column every 4 s (modulation). The 2nd dimension separation is shown in the y-axis, which covers a retention time of 4 s. Peaks now stand vertical to the plain and the height is expressed in the darkness of the spot. First dimension retention times almost exactly correspond to those in the upper chromatograms, since the same chromatographic system was used. From the polar 2nd dimension column, the paraffins were eluted first (at the bottom), whereas the aromatic hydrocarbons were the more strongly retained the more aromatic rings they include and formed signals further up.

Figure 1.

Normal GC-FID (no modulation; upper chromatograms) and GC×GC-FID contour plots of two mineral oil products. Numbers, n-alkanes of given carbon number.

The left GC-FID chromatogram and GC×GC-FID plot are from a brownish, crude mineral oil fraction used as batching oil for rendering jute or sisal fibers more flexible before spinning (used for jute and sisal bags). In the GC-FID chromatogram, a hump of unresolved components is observed, topped by signals from n-alkanes. The GC×GC plot confirms that the majority of the material (some 70%, as determined by on-line HPLC-GC-FID 4) consists of paraffins. The aromatics form many identifiable spots (peaks), but also zones of high complexity with fused signals, particularly at the higher elution temperatures.

The right chromatograms are from a higher molecular weight crude mineral oil fraction for manufacturing products like lubricating oil. The aromatic hydrocarbons made up 53% 4. For these, the higher molecular mass primarily means higher alkylation (rather than more rings), which implies a larger number of isomers. In fact, in the GC×GC plot most of the aromatics are represented as a gray continuum without resolution, i.e. as a hump also in the 2nd dimension GC. The main task of the work described below was the extraction of information from such plots.

In this work, the hydrocarbons as shown in the GC-FID chromatograms are further separated in two steps: HPLC separates MOSH and MOAH, and the MOAH are analyzed by GC×GC. The MOAH can be characterized by ring type (number of rings, arrangement of the rings and possibly the presence of heteroatoms, such as sulfur) as well as by the number of carbon atoms in the alkyl groups. The aromatics of a given ring type and differing in alkylation were localized in the GC×GC plot using MS. Owing to the complexity and overlapping, particularly for the high molecular mass oils often observed as food contaminants, the interpretation was limited to the number of aromatic rings: lines were established to separate the zones comprising MOAH of 1–5 rings. Since no software enabling the integration of three dimensional humps was available, the quantitative composition was approximated using a grid of 2nd dimension chromatograms. The lines separating by ring number were projected into the 2nd dimension humps to integrate the fractions of given ring number. These fractions were added up and expressed as percent of the total of the MOAH. This method was applied to characterize the MOAH in contaminated Ukrainian sunflower oil.

3.2 Identification of bands of given ring systems

In the GC×GC plots, the classes of aromatics of a given ring type form more or less horizontal bands with increasing number of carbon atoms in the alkyl groups. These bands were traced by MS in EI-scan mode, using the MOAH fraction of a mixture of the two oils shown in Fig. 1 covering a broad range of molecular masses. For a given ring type, the alkylated components were extracted by the molecular mass. For some masses, the resulting plots included other compounds having the same molecular mass or forming a fragment of this mass. These were identified by their spectra and chromatographic retention behavior and removed from the plot using picture processing software (Adobe Photoshop®). The intensity of the extracted signals was adjusted to be similar using this same software.

The extraction of MOAH of given number of carbon atoms in the alkyl group resulted in a large number of GC×GC-MS plots. In Fig. 2, such plots were superposed to specify the positions of the alkylated species of a given ring system. In the upper left plot, some of the benzenes and phenanthrenes/anthracenes are shown, the latter not being separated. Phenanthrene (added as standard) is visible as a single signal and specified as non-alkylated by “0”. Two signals for methyl phenanthrene/anthracene (“1”) are visible, a larger number for ethyl- and dimethyl derivatives (“2”), and a complex group of C3 species. As the groups of signals grew broader and started to overlap, only some of the larger phenanthrenes/anthracenes are shown (the C5, C8, C12 and C17 alkylated species). A line was drawn through the centers of the spots. These lines are shown also in the other plots of this figure. For the benzenes, the C7 species were the first being present in significant amounts in the batching oil. There were benzenes with at least 30 carbon atoms in the side chain, but they were difficult to isolate from other components.

Figure 2.

GC×GC-MS plots of extracted ions representing selected alkylated species of the most important aromatics, processed by Photoshop to remove compounds having the same mass but not corresponding to the components of interest and to present all spots at similar intensity. Numbers indicate carbon atoms in the alkyl groups. B, Benzene; N, naphthalene; Fluo, fluorene; BT, benzothiophene; DBT, dibenzothiophene; BDBT, benzo dibenzothiophene; Phe, phenanthrene; An, anthracene; Flu, fluoranthene; Py, pyrene; Chry, chrysene; BPy, benzopyrene.

The other plots of Fig. 2 show the bands of other aromatic components, with the line interconnecting the centers shown for all aromatics investigated in the four plots. It is reminded that the intensity of the signals was adjusted, i.e. the chromatogram did not look as the sum of all signals shown.

Figure 3 shows the real GC×GC-FID plot of the mixture from which the extracted signals shown in Fig. 2 were derived and highlights the standards added for confirming the interpretation. Most components were commercially available or gifts. Benzene and phenanthrene were alkylated with 1-chloro hexane and/or 1-chloro octadecane by Friedel Kraft reaction, which resulted in numerous isomers (marked by bars), many of which did not seem to be identical with those found in mineral oil.

Figure 3.

GC×GC-FID plot of the mixture of oils shown in Fig. 1 and used for the identifications shown in Fig. 2, with the substances added for confirmation being labeled. Abbreviations: C6-B, hexyl benzenes; di-C6-B, dihexyl benzenes; 9B, n-nonyl benzene; C18-B, octadecyl benzenes; Et-N, ethyl naphthalene; DMe-N, dimethyl naphthalene; Et-Phe, ethyl phenanthrene; TMeAN, tetramethyl anthracene; C18-Phe, octadecyl phenanthrenes; Cho, cholestane; BaPy, benzo(a)pyrene; for other abbreviations, see Fig. 2.

3.3 Need for HPLC preseparation

Figure 4 compares the MOAH and the MOSH of the above mixture after preseparation by HPLC. The MOSH fraction (lower plot) shows the band of the paraffins, with the large signals of the n-alkanes and the incompletely resolved iso-alkanes in-between. The n-alkanes formed broadened signals and reached to higher 2nd dimension retention times as a result of overloading the 2nd dimension column. The naphthenes (cyclic saturates) are more strongly retained in the 2nd dimension, as visible for the weak signals slightly above the series of paraffins, but particularly for the steranes and hopanes. The window of the steranes and hopanes is shown at a lower attenuation. The 5-alpha-cholestane (Cho) was added as verification standard for the end of the MOSH fraction in HPLC 4. These 4- and 5-ring saturated hydrocarbons are coeluted with the highly alkylated two- and three-ring aromatics (compare with upper chromatogram) and explain why the HPLC preseparation is a prerequisite for the quantification of the MOAH. The position of the bicyclic sesquiterpanes 17 is marked by a line.

Figure 4.

GC×GC-FID plots of the MOSH and the MOAH in the mixture used for Figs. 2 and 3. For abbreviations, see Fig. 3; Sesqui, bicyclic sesquiterpanes; TAS, triaromatic steranes 1.

3.4 Allocation of MOAH by ring number

A complete resolution of the MOAH by type of aromatic ring system was out of reach. MS enabled the distinction of most of these classes (see above), but calibration of the response for a quantitative determination is virtually impossible. Since a detailed analysis was assumed to rapidly produce more data than could be managed, FID was preferred and the MOAH were just grouped by ring number (Fig. 5). Lines were introduced, separating these zones on the basis of the above identifications. Separations are quite complete in the early eluted part, but there is some overlapping for the high molecular mass MOAH.

Figure 5.

GC×GC-FID plot of the MOAH in the mixture used for identifications. Dotted lines show the centers of the bands of given ring systems, the bold lines separate the MOAH by ring number. The compositional analysis described below is based on these lines.

As shown in Fig. 2, the number of carbon atoms in the alkyl groups can be identified by MS, which enables to specify the range of alkylations for a given ring type in a mineral oil product. This needs to be done separately for each ring system, since the 1st dimension GC does not separate by number of carbon atoms: for instance, as shown in Fig. 2, above the C22-benzenes (C28) are the C10 phenanthrenes/anthracenes (C24) or the C1-chrysenes (C19).

3.5 Overloading of the 2nd dimension column

Figure 6 advises against a potential problem for such compositional analysis: shifts in 2nd dimension retention time resulting from overloading. The chromatograms are from a crude Ukrainian sunflower oil contaminated with 20 000 mg/kg MOAH (though apparently from a process enriching the mineral oil 5). At the upper left, two chromatograms are overlaid. The first shows the n-alkanes of a standard mixture reaching up to C44 and forming a more or less straight band (see “overlaid paraffins”; dotted line at the bottom). The second was obtained by injecting 25 μg isolated MOAH to which a smaller amount of the same n-alkanes was added. Now the n-alkanes (highlighted by the dashed line) show increased retention in the most charged zone and form an arc. The double arrow points out the shift in retention time. The same shift towards higher retention in the second dimension GC is observed for the benzenes and the 2-ring components (bold dotted lines): they are no longer in their sectors of the plot. Only a small shift was observed for perylene (Per), added as a verification standard 4.

Figure 6.

Overloading of the 2nd dimension GC: GC×GC-FID plots of a sunflower oil contaminated with 20 000 mg/kg MOAH. Upper left: overlay of 25 μg MOAH with a mixture of n-alkanes (interconnected by dotted line at bottom). In the heavily loaded zone, the n-alkanes added to the MOAH show increased retention (arc with dashed line), as do the benzenes and 2-ring components (bold dotted lines). For adequate grouping of the components by ring number, overloading must be avoided (bottom right; 2.5 μg MOAH). Per, 6B, TBB, 9B and BP, internal standards for quantitation and verification of the cuts of the HPLC preseparation 4, as well as for the control of the retention times in the GC×GC plots.

The overloading effect rapidly decreased when less sample material was injected: with 8 μg total MOAH, the deviation for the benzenes was small, and it became negligible with 2.5 μg (bottom right chromatogram). This limited capacity of the 2nd dimension column restricts the detection limit for the less abundant classes of MOAH, particularly the 5-ring components (including the benzopyrenes): to lower the detection limit, HPLC preseparation must be modified such that the bulk of the benzenes and the 2-ring MOAH are removed (see below).

3.6 Quantitative determination

The software used did not enable comprehensive integration of a three-dimensional hump of unresolved material with introduction of the lines separating by ring number. As an approximation, a grid of 2nd dimension chromatograms with 5 min intervals in the 1st dimension was used to assess the composition of the hump at these given points. In these 2nd dimension chromatograms (shown at the right in Fig. 7), the retention times corresponding to the lines separating the MOAH by ring number according to Fig. 5 were manually introduced as drop lines for integrating the various sections. The early peak in the chromatogram at 40 min and the late peak in those at 50 and 55 min were disregarded.

Figure 7.

Left, GC×GC-FID plot of the MOAH in the crude mineral oil fraction shown at the right of Fig. 1. Vertical lines show a grid with 5 min intervals at which 2nd dimension chromatograms were extracted. These are shown at the right and specified by their 1st dimension retention time (min). Forced drop lines separate by ring number according to the lines in the GC×GC plot.

Assuming that the distributions in the 2nd dimension chromatograms selected as a grid were representative for their sections of the plot, the areas determined for aromatics of a given ring number were expressed as percent of the total area of these chromatograms. The concentrations of the MOAH of given ring number was calculated from these percentages and the total MOAH concentration obtained by HPLC-GC-FID 4.

This approach presupposes a fairly homogeneous hump of unresolved components: sharp signals on top of the hump cannot be treated in this way, as their importance would be overestimated when falling into the integrated chromatogram or they would be neglected when eluted next to it. Sharp peaks are mostly to be neglected: MOAH do not commonly form sharp peaks of significant size, and in the MOAH analysis in food, sharp peaks are usually food components. Mostly, sharp peaks fall between the grid. Otherwise, either an adjacent 2nd dimension chromatogram is chosen or the signals are subtracted.

Table 1 shows the quantitative composition of the MOAH in this mineral oil fraction (containing 53% MOAH 4). Benzenes, 2-ring and 3-ring components made up 25–30% of the MOAH, 4-ring aromatics 18% and 5-ring components about 3%. Being a crude mineral oil fraction, this sample might have contained a particularly high concentration of aromatics and a particularly high proportion of large ring systems.

Table 1. Composition of the aromatics by ring number in the crude mineral oil fraction shown in Fig. 7, as obtained by integration of the 2nd dimension chromatograms shown in Fig. 7 specified by their 1st dimension retention time (min)
Time1 ring2 rings3 rings4 rings5 rings
(min)(% referring to aromatics)

3.7 Patterns of aromatics

Figure 8 presents GC×GC-FID plots characterizing various types of aromatics isolated by HPLC and is intended to support the interpretation of plots obtained in practice. The lubricating oil at the upper left contained 26% MOAH 4. Most of these consisted of alkylated benzenes and 2-ring compounds, i.e. the heavier aromatics seem to be largely removed. The plot at the lower left is from a distillate aromatic extract oil used as an extender oil for rubber articles, such as tires and handles 3. Obtained by extraction from crude oil fractions, it contains what has been removed from oils used for lubricating oil. The fraction of the aromatic hydrocarbons made up 85% of the oil; the heavy aromatics are enriched, the 3-ring components representing more than half of the oil.

Figure 8.

GC×GC-FID plots of aromatics showing the variability of mixtures.

At the upper right, an oil extracted from a handle of a craftsman's tool is shown, which made up 8.9% of the rubber mass. The pattern strongly differs from those shown at the left: there are virtually no benzenes and 2-ring components; most of the material is in the fraction of the 4-ring aromatics and is little alkylated. It is closer to the aromatics from the tar of a double chamber wood furnace, which are even less alkylated (as typical for PAHs formed at high temperature).

3.8 MOAH in contaminated sunflower oils

Figure 9 shows the Ukrainian sunflower oil with the perhaps highest contamination by MOAH reported so far: 1800 mg/kg (next to 3100 mg/kg MOSH) 5. The oil was refined, but not marketed. Integration of the MOAH fractions was performed in the same way as in Fig. 7. Owing to the small amount injected (10 out of 100 μL) and the high attenuation, interferences by the sunflower oil were small: apart from the overloaded squalene there are just some sterenes, sterol dehydroxylation products from raffination. The 5 min grid of 2nd dimension chromatograms happened to fall beside the individual peaks, i.e. these chromatograms did not need to be moved or corrected.

Figure 9.

MOAH fraction of a refined heavily contaminated sunflower oil. Integration as shown in Fig. 7.

As shown in the lower part of Table 2, the percentages of the 4- and 5-ring MOAH were not much lower than in the crude mineral oil fraction of Fig. 7 and Table 1 (10 and 1.2%, respectively, compared to 18 and 2.8%), which is indicative for a rather crude fraction. The concentrations in the oil were above 500 mg/kg for the 1- and 2-ring aromatics and still reached 186 and 22 mg/kg for the 4- and 5-ring components, respectively (upper part of Table 2).

Table 2. Concentrations of MOAH in a refined sunflower oil by number of aromatic rings and relative composition in percentages; data referring to the 2nd dimension chromatograms shown in Fig. 9
Time1 ring2 rings3 rings4 rings5 rings
(min)(mg/kg sunflower oil)
Time1 ring2 rings3 rings4 rings5 rings
(min)(% referring to total MOAH)

Figure 10 shows the MOAH of four other, less contaminated Ukrainian sunflower oils (90–600 mg/kg MOAH), numbered as in Table 3. Oils 2 and 6 were crude and contained volatile constituents, such as kaurene, which are absent in the lower plots of the refined oils 5 and 7 (the latter being the major constituent of a margarine). Conversely, raffination left behind sterenes. The margarine also contained carotenoids, presumably from palm oil.

Figure 10.

MOAH of crude and refined contaminated Ukrainian sunflower oils. Oils numbered as in Table 3, with the total MOAH concentration as given in 5.

Table 3. Composition of the MOAH in seven contaminated Ukrainian sunflower oils in terms of concentrations in the oil and percent referring to the MOAH
Oil1 ring2 rings3 rings4 rings5 ringsSum
 (mg/kg sunflower oil)
  1. a

    Refined (r) as well as crude (c) oils, some of which are shown in Fig. 10. Oil 1 corresponds to that in Table 2.

Oil1 ring2 rings3 rings4 rings5 rings 
 (% of the MOAH)

The MOAH in the upper two plots have clearly different compositions. Those of the two at the left might be the same, taking into consideration that some of the more volatile MOAH were removed during deodoration. Also, the two contaminants in the right plots might be the same. However, for tracing back the mineral oils, a more detailed analysis with MS, as described in 18, would be more appropriate.

Table 3 summarizes the MOAH compositions of seven contaminated Ukrainian sunflower oils by numbers of aromatic rings. All these oils did not reach the food market. Oil 1, shown in Fig. 9 and Table 2, not only had the highest concentration of MOAH, but also contained mineral oil with a higher percentage of 4- and 5-ring aromatics than all others, suggesting that the mineral oil was less refined or even crude. Oil 2 contained MOAH with 4.6% 4-ring and 0.1% 5-ring components, corresponding to 28 and 0.61 mg/kg, respectively, in the sunflower oil. The data on the 5-ring aromatics of oils 3–7 is from small tails of the 2nd dimension chromatograms and merely indicates upper possible concentrations. The detection limit of the method was around 0.2 mg/kg oil.

The detection limit of 0.2 mg/kg for 5-ring MOAH, including the alkylated benzopyrenes, might be unsatisfactory compared to their relevance (for instance, benzo[a]pyrene is considered critical at a 100 times lower concentration; EU Regulation 1881/2006). Figure 11 shows the result of an attempt to decrease this detection limit by an HPLC preseparation focusing on the heavy aromatics for an oil containing 370 mg/kg MOAH (oil 3 in Table 3): the window normally used was split. With the benzenes and the 2-ring aromatics as well as the squalene and the sterenes, the first fraction comprised the bulk of the material, which enabled to inject a five times larger aliquot of the later eluted heavy MOAH. The internal standards Per and BP were in the late fraction, the other three (6B, TBB and 9B) in the early one.

Figure 11.

Attempt to decrease the detection limit for the 4- and 5-ring MOAH by splitting the HPLC fraction of the MOAH and increasing the sample aliquot for the heavy components.

As shown in the lower chromatogram, most material in the late fraction was again eluted in the region of the 1- and 2-ring MOAH. For the 5-ring fraction, the integration of the 2nd dimension chromatograms yielded an area corresponding to 50 μg/kg oil, but most of this material could have been from background interference. Enrichment of the sample by collecting and combining five of these HPLC fractions increased the response of the background contamination to a similar extent as that of the MOAH, suggesting that the detection limit achievable by this method was largely determined by the background.

4 Discussion and conclusions

Mineral oil might be the largest contaminant in our body 19 and deserve more attention than it generally receives. If analyzed, the focus is on the saturated hydrocarbons (MOSH). The aromatics are, however, likely to be more toxic, and more often than not the mineral oils found in foods are of technical quality – not the “white paraffin oil” sometimes termed “food-grade.”

MOAH differ from the PAHs commonly analyzed by a high degree of alkylation. The non-alkylated mother compounds are almost absent, which renders the determination of the priority PAHs misleading: it underestimates the presence of MOAH by orders of magnitude.

Previous analysis of the contaminated Ukrainian sunflower oils from early 2008 has shown that the mineral oil typically contained 20–30% aromatic components 5. For a sunflower oil containing 300 mg/kg MOSH, considered to be of no health risk by the EFSA in May 2008 1, this means the presence of around 100 mg/kg MOAH. Using the average composition of the samples investigated (Table 3), roughly a third of each of the MOAH consisted of benzenes, 2-ring components (naphthalenes and benzothiophenes) and heavier aromatics. The 5% 4-ring components would correspond to 5 mg/kg in the sunflower oil. Assuming the mineral oil with the highest content of heavy aromatics detected (sample 1 in Table 3), a sunflower oil contaminated with 300 mg/kg MOSH would have contained 174 mg/kg MOAH (37% of the mineral oil 5), of which 2.1 mg/kg were 5-ring components, including the largely alkylated benzopyrenes. This is about 1000 times more than benzo[a]pyrene is tolerated in edible oil.

There is a need for a toxicological evaluation of the MOAH. It is well possible that most of the alkylated aromatics are less toxic than the parent compounds, because the sites critical for epoxidation are blocked by alkyl groups. However, since concentrations in food are around two orders of magnitude higher, the MOAH represent a health risk even if merely 1% of their components are as toxic as the PAHs.

To prepare the data needed for risk assessment, the analysts should know the questions of the toxicologists. These are unlikely to be interested in a list of hundreds or thousands of individual constituents. We assumed that they will perform tests with mixtures and that they need a more general characterization to find those which contain the MOAH typically found in food. This is what the described method was heading for.