Introduction

Dietary lipids are susceptible to oxidative processes, giving rise to the development of off-flavours and a decrease of the nutritional quality and safety. Autoxidation is the most important process leading to oxidative deterioration. It is based on the spontaneous reaction of atmospheric oxygen with lipids under mild conditions via a radical chain reaction [1].

The first analytical methods to assess lipid deterioration were empirical volumetric and UV/ViS spectroscopic methods. They were developed and standardized several decades ago and have only undergone minor changes since then [2]. In routine, the classical quality parameters such as peroxide value (PV), acid value, and anisidine value (AV) are still widely used. In the EU, these parameters play an important role in official control of oils and fats. For example, in Germany, for refined oils, the AV and two times the PV must not exceed a value of ten. The disadvantages of these parameters are the high consumption of solvents and other toxic chemicals and especially the poor specificity [2]. In a previous study, we found that some oil types contain natural oxidizing or reducing compounds which have significant influence on the result of the PV method [3]. Our results showed that the PV of black seed oil may be strongly increased due to the naturally occurring component thymoquinone, whereas the PV of olive oil may be decreased due to the naturally occurring component hydroxytyrosol. The AV method is the classical approach to determine aldehydes in edible oils. Aldehydes are generated as secondary oxidation products from the degradation of hydroperoxides and are, besides vinyl ketones, the major contributors to off-flavour associated with the rancidity of oils and fats [4, 5]. The AV method is based on the chemical reaction of the carbonyl group and p-anisidine leading to the formation of an intensively coloured Schiff base which is measured UV-spectroscopically at λ = 350 nm [6]. The AV is a sum parameter depending not only on the aldehyde concentration in the sample, but also to a large extent on the molecular structure of the aldehydes. For example, the reaction product of an aldehyde that possesses a double bond in the carbon chain conjugated with the carbonyl double bond exhibits a higher absorption at the measurement wavelength than derivatives of saturated aldehydes. As a consequence, unsaturated aldehydes give higher AV results than saturated aldehydes [2].

As an alternative to the AV, some authors propose the application of new fast instrumental techniques like IR or Raman spectrometry. Their results show good correlations with the classical AV [7, 8]. In the last years, the relevance of NMR as an analytical tool for the determination of oxidation products has strongly increased. NMR spectroscopy enables a rapid analysis that, in principle, requires no sample preparation. 1H NMR facilitates to determine the individual aldehyde species in the downfield region of the spectrum. Several studies have been published dealing with the classification and characterization of the aldehyde proton signals in oxidized edible oils for different oxidation conditions [9–12].

Since the integral of an 1H NMR signal is proportional to the number of protons, the molar concentration of the aldehyde groups can be determined by 1H NMR [13, 14]. Recently, Guillen and Goicoechea reported the use of the proton signal of non-deuterated chloroform occurring in deuterated chloroform for the quantification of the oxidation products [11].

In this work, we present a 1H NMR method for the quantification of aldehydes in edible oils. It is shown that the classical AV can be modelled as a linear combination of the normalized NMR integrals of the aldehydes. The suitability of the model is demonstrated by comparing the NMR-determined AVs with the conventionally measured AVs of several commercially available edible oils of different oil types.

Experimental

Samples and standards

Seventy-nine edible oil samples were provided by the “Chemisches und Veterinäruntersuchungsamt Karlsruhe”. The sample collection comprises the following oil types: nut oil (walnut oil, peanut oil, hazelnut oil, almond oil, pistachio oil) (29 %), sunflower oil (16 %), rapeseed oil (16 %), grape seed oils (10 %), thistle oil (9 %), corn oil (5 %), linseed oil (5 %), argan oil, plant oil (mixture of different oil types), soybean oil, allium ursinum oil, and sesame seed oil.

Hexanal, (E)-2-hexenal, 2,4-hexadienal were purchased from Merck (Darmstadt, Germany) and Sigma/Aldrich (Steinheim, Germany). The 2,4-hexadienal standard consisted of 88 % (E,E)-2,4-hexadienal and 12 % (Z,E)-2,4-hexadienal (ratio was determined by 1H NMR). Delios® V, an artificial triacylglyceride mainly consisting of capryl and capric acid, was donated from Cognis (Mannheim, Germany). Deuterated chloroform (CDCl3, 99.8 atom% D) was purchased from Roth (Karlsruhe, Germany). 4-Methoxyaniline (p-anisidine) from Merck (Darmstadt, Germany), acetic acid 100 %, and isooctane from Roth (Karlsruhe, Germany) which were used for the AV determination were of analytical grade and comply with the requirements of the international standard ISO 6885 [6].

NMR experiments/spectra acquisition and processing

The 1H NMR spectra were recorded on a Bruker 400 MHz spectrometer operating at 400.17 MHz equipped with a 5-mm SEI probe. The acquisition and processing parameters were as follows: acquisition time 7.97 s, spectral width 20.5396 ppm, time domain 128 k, relaxation delay 1 s, number of scans 128, size 128 k. The experiment was carried out at 300 K using a flip angle of 30°. All data were processed using Bruker’s TOPSPIN-NMR software (version 2.1, Bruker, Rheinstetten, Germany).

NMR method to quantify aldehydes in edible oils

The sample (250 mg) was dissolved in 0.6 ml of CDCl3 containing 0.1 % tetramethylsilane (TMS) as internal reference. The aldehyde group protons (CHO) of n-alkanals resonate as triplets at δ = 9.76 ppm, CHO of (E)-2-alkenals as doublets at δ = 9.51 ppm and CHO of (E,E)-2,4-alkadienals as doublets at δ = 9.53 ppm. In order to quantify the individual aldehyde species, the CHO signals were integrated as well as the proton signal of the glyceride methylene group (glyceride-CH 2) appearing at δ = 3.8–4.6 ppm. To determine the aldehyde amount in mmol/mol triacylglycerol (TAG), the integral of the glyceride-CH 2 signal was normalized to a value of 4,000 (one TAG possesses four Glyceride-CH 2 protons). The processing and integration for every spectrum was performed three times. For the calculation, the mean value was chosen.

p-Anisidine value (AV)

The AV of all samples was determined according to the international standard ISO 6885:2006 [6]. Description in brief: The test solution is prepared in isooctane. It is reacted with an acetic acid solution of p-anisidine, and the increase in absorbance at 350 nm is measured, and the AV is calculated.

Computations

Multiple linear regression and deming regression were done with Excel 2007 (Microsoft, Redmond, USA).

Results and discussion

The signals of the carbonyl protons of autoxidation-derived aldehydes in edible oils appear in the downfield region between 9.4 and 9.8 ppm (solvent: CDCl3). Depending on the oxidizing conditions and the oil type, different aldehydes are formed by the oxidation of edible oils like n-alkanals, short-chain n-alkanals, (E)-2-alkenals, (E,E)-2,4-alkadienals, (Z,E)-2,4-alkadienals, 4,5-epoxy-(E)-2-alkenal, 4-hydroxy-(E)-alkenal, and 4-hydroperoxy-(E)-alkenal [10, 11].

1H NMR analysis of 400 oil samples revealed that the aldehyde concentrations in commercial edible oils generally are low and that mainly the three aldehyde species n-alkanals, (E)-2-alkenals, and (E,E)-2,4-alkadienals are detected. The aldehyde proton region of a mixture of n-hexanal, (E)-2-hexenal, and (E,E)-2,4-hexadienal (and (Z,E)-2,4-hexadienal as an impurity) is shown in Fig. 1.

Fig. 1
figure 1

Carbonyl region of the 1H NMR spectrum of a mixture of n-hexanal (triplet), (Z,E)-2,4-hexadienal (doublet), (E,E)-2,4-hexadienal (doublet), and (E)-2-hexenal (doublet)

Full size image

To establish an equation that relates the AV to the normalized NMR integrals of aldehydes, the following assumptions were made:

  1. 1.

    The AV of a specific aldehyde species (i.e. n-alkanals, (E)-2-alkenals and (E,E)-2,4-alkadienals) is directly proportional to the corresponding normalized NMR integral.

  2. 2.

    The AV of a mixture represents a linear combination of the normalized NMR integrals of the aldehydes.

  3. 3.

    Aldehydes of the same species with different chain length exhibit the same coefficients (e.g., a hexanal = a nonanal)

N-hexanal, (E)-2-hexenal, and (E,E)-2,4-hexadienal were chosen as model aldehydes. For each compound, 6 different mixtures of the aldehyde standard and Delios V (an artificial triacylglyceride solely consisting of saturated middle chain fatty acyl residues) were prepared. These mixtures were analysed by both the novel 1H NMR method and the classical AV method. The data are given in Fig. 2.

Fig. 2
figure 2

Plot of the AV versus the NMR-determined amount fraction for the aldehyde standards

Full size image

For hexanal and (E)-2-hexenal, a model of the form

$$ {\text{AV}} = ax + \varepsilon $$
(1)

(a denotes the compound specific coefficient, x the normalized NMR integral, and \( \varepsilon \) the random error) was fitted to the respective data set by least squares regression [15] to obtain the estimates \( \hat{a}_{\text{alkanal}} \) und \( \hat{a}_{\text{alkenal}} \).

Since the (E,E)-2,4-hexadienal standard contains 12 % of its isomer (Z,E)-2,4-hexadienal as a contaminant, the following model was applied to the corresponding data set:

$$ \hbox{AV}=a_{\text{alkadienal}} x_{\text{alkadienal}} + a_{{(Z,E){\text{alkadienal}}}} x_{{(Z,E){\text{alkadienal}}}} + \varepsilon $$
(2)

Table 1 shows the calculated coefficients and the associated 95 % confidence intervals.

Table 1 Compound-specific coefficients to model the AV
Full size table

The coefficient \( \hat{a}_{{(Z,E){\text{alkadienal}}}} \) does not deviate significantly from zero. This can be explained by the relatively small proportion of (Z,E)-2,4-hexadienal combined with the low precision of the AV and the low precision of the integrals of the small NMR signals. Since (Z,E)-2,4-hexadienal is in general not present in commercial oil samples, \( \hat{a}_{{(Z,E){\text{alkadienal}}}} \) is not relevant for our model.

The coefficients of n-hexanal, (E)-2-hexenal, and (E,E)-2,4-hexadienal are considerably different. \( \hat{a}_{\text{alkenal}} \) exceeds \( \hat{a}_{\text{alkanal}} \) 12 times, and \( \hat{a}_{\text{alkadienal}} \) is even 63 times higher than \( \hat{a}_{\text{alkanal}} \). This means that (E)-2-hexenal contributes 12 times more to the AV than n-hexanal, and (E,E)-2,4-hexadienal contributes 63 times more to the AV than n-hexanal. These findings are in agreement with the results of Pardun [16]. He found that the absorption of the (E)-2-heptenal product is 12 times higher and the absorption of the (E,E)-2,4-decadienal product is 40 times higher than the absorption of the n-hexanal product.

Combining our results, we can express the AV as

$$ {\text{AV}} = 0.54x_{\text{alkanal}} + 6.2x_{\text{alkenal}} + 34x_{\text{alkadienal}} $$
(3)

where \( x_{\text{alkanal}} \), \( x_{\text{alkenal}} \), and \( x_{\text{alkadienal}} \) designate the 1H NMR-determined molar concentration of n-alkanals, (E)-2-alkenals, and (E,E)-2,4-alkadienals.

In order to verify the model, 79 commercial oil samples were measured by both methods (see Fig. 3). The modelled AVs were plotted versus the classical AVs, and a straight line was fitted to the data. Taking into account that both quantities are subject to error, Deming regression was applied [17]. Confidence intervals of 95 % were calculated for both the empirical and the modelled AV. If the confidence intervals overlap, then both AVs are considered to be not significantly different. (Note: This way of proceeding is in fact overly conservative but in our case more convenient.)

Fig. 3
figure 3

Plot of the AVs versus the NMR-determined AVs of 79 oil samples

Full size image

Eighty-six per cent of the oil samples with AVs above 0.9 showed overlapping confidence intervals. The aldehyde signals in the 1H NMR spectrum of oils with AVs below 0.9 were often too small for a precise integration.

In conclusion, it appears that the NMR method is suitable to reproduce the results of the classical AV method. Yet, in contrast to the latter, the NMR method has the crucial advantage of providing quantitative information on the content of individual aldehyde species.