3.1. ORAC Values and Antioxidant Status of CLO Products
ORAC determinations were first performed in order to explore the peroxyl radical-scavenging antioxidant capacity of each CLO product. These measurements revealed that Product 4 had ‘between-batch’ mean ± SEM values of 91.4 ± 19.5 mmol. trolox equivalents/kg (n
= 5 samples, each from separate batches), whereas those of n
= 5 different batches of the non-fermented natural Products 1, 2, and 3 were found to be 4.9 ± 1.0, 46.2 ± 4.6, and 5.9 ± 1.1 mmol. trolox equivalents/kg, respectively. The literature ORAC value data available for antioxidant-rich virgin olive oil range from only 1.8–6.2 mmol. trolox equivalents/kg (with positively-correlating total phenolic levels of 50–254 mg/kg), whereas those for the refined olive and peanut oil products were only 1.0–1.6 mmol. trolox equivalents/kg (phenolic contents only 1–10 mg/kg) [31
]. These data provided a high level of evidence for Product 4’s powerful peroxidative resistance. However, it should also be noted that one of the other products tested (Product 2) contained added rosemary extract and natural mixed tocopherols, and these additions may account for its relatively high mean ORAC value. However, Product 3 also contained an unspecified concentration of cholesterol, along with natural levels of vitamins A and E. Similarly, Product 1 was a cold-pressed virgin CLO product, which was supplemented with small amounts of added rosemary extract and natural vitamin E. The substantial difference that was observed between the much elevated mean ORAC value of Product 4 and that of the other products combined was found to be very highly statistically significant (p
= 7.53 x 10-5
, two-sample t-test). Principally, Product 3 contained alpha-tocopherol (α-TOH) added as its acetate ester, which fails to offer UFAs protection against peroxidative damage unless α-TOH itself is liberated therefrom via any hydrolysis reactions occurring.
Furthermore, a range of antioxidant and potential antioxidant species, including flavonoids, flavonones, total phenolics, tocopherols, carotenoids, tannins, anthocyanins, and retinol, was determined in four or five separate batches each of Products 1–4. Table 1
shows these results.
The total flavonoids, flavanones, phenolics, tannins, and anthocyanins were undetectable, i.e. below the limits of quantification, in Products 1–3 (Table 1
); however, Product 4 was found to have substantive contents of flavanones, phenolics, and tannins. Although all tested products had similar concentrations of chlorophyll A, that of chlorophyll B was much greater in Product 4. Moreover, the total carotenoid contents of Products 1–3 were only ca. 40% of that present in Product 4 (p
< 0.01). Therefore, higher levels of these antioxidants and chlorophyll B observed in Product 4 appear to arise from the fermentation process that was employed during its commercial production.
However, no significant differences were found between the mean α-TOH contents of each of the four CLOs. Moreover, for retinol, the only significant difference found was that between Products 2 and 3, with the latter being almost two-fold higher. One further noteworthy observation is that the ‘between-batch’ variation of some of the analytes monitored in Table 1
was higher for Product 4 than it was for those of Products 1–3. For example, the intra-product variance for the Product 4 α-TOH concentrations determined was significantly greater than those of Products 1–3 (p
, Bartlett’s test), but this was not the case for the retinol levels. This wider variance in Product 4 levels observed for some of these analytes might also be attributable to its fermented route of production.
The results acquired confirmed that although Product 4 contained peroxidatively-significant, near-millimolar concentrations of at least some of these agents (predominantly tyramine and 2-phenylethylamine), such amines were completely undetectable in all analysed batches of Products 1–3, as expected (Table 2
Further experiments conducted involved the monitoring of biogenic amines in Product 4 samples both before and after exposure to a 90 min. heating episode at 180 °C, according to our TSE protocol (Section 2.3
). The data acquired clearly demonstrated the complete removal of these agents from this CLO product when heated in this manner.
3.2. Investigation of an Unusual Broad 1H NMR Resonance in 1H NMR Spectral Profiles of a CLO Product: Assignment to Collagen/Collagen Hydrolysate Products and Ammonia
Evidence for the identities of agents that are responsible for the high ORAC values and, hence, peroxidative resistivity of Product 4 was provided by high-field 1
H NMR analysis experiments (600 MHz). Indeed, an examination of the 6.40–10.00 ppm regions of spectra acquired on n
= 6 unheated batches revealed a very broad resonance centered at ca. δ
= 6.40 to 9.48 ppm (highly variable maximal intensity, centralised chemical shift values of 6.87, 7.40, 7.80, 7.95, 8.43, and 9.18 ppm were typically observed), and that generally spanned chemical shift ranges of ≥ 2.5 ppm (Δv1/2
≥ 250 Hz) This unusual signal was not at all present in corresponding spectra of Products 1-3 (Figure 1
a). Micro-extraction of the sample with the 1
H NMR profile shown in this spectrum with 2
O removed this resonance from the 1
H NMR profiles (Figure 1
b), an observation that might be consistent with the greater solubility of the agent(s) giving rise to it in aqueous systems. Moreover, following an equilibration period of 1.0 h at ambient temperature, this signal was also completely removed from the spectra that were acquired following its treatment with a very small C2
-miscible volume of 2
O (Figure 1
c). This signal was also removed with a further 1:5 (v/v) dilution of analysis solutions with C2
(the red arrow in Figure 1
c indicates this very broad 2
O-extinguishable resonance). Additional experiments involved monitoring the intensity of this broad resonance as a function of TSE heating time at 180 °C (Section 2.3
), and the 1
H NMR data acquired demonstrated that it was eliminated from spectra with increasing sampling time-point, a substantial decrease in its intensity occurring within a 10–20 min. heating duration. Full details of these investigations will be reported in a second, follow-up publication in view of this observation’s novelty, and also the expansive number of further experiments performed to fully establish the identity/identities of this broad Product 4 resonance. However, in summary, the results from these studies revealed that this 1
H NMR resonance is predominantly ascribable to an admixture of protein, predominantly collagen and collagen hydrolysate peptide linkage-CO-NH
- protons, together with lower levels of the fermentation product ammonia, the latter as C2
-solubilised ammonium ions. Broad amide-/peptide-CO-NH
- function resonances have highly variable chemical shift values (δ = 5–9 ppm) [32
], and these are further complicated by their macromolecular nature when present in proteins/polypeptides.
The high frequency chemical shift values that are associated with this deuterium-exchangeable resonance precludes assignments to aliphatic alcohol-OH,
, and carboxylic acid-COOH
functions (limited to the δ = 0.5–5.0, 0.5-4.5, and 10.0–13.2 ppm ranges, respectively), although it is within the range for the phenolic-OH
H NMR signals (δ = 4–8 ppm) [32
Intact collagen, along with the higher molecular mass fractions of its hydrolytic degradation products, have very short T2
values under our experimental conditions due to their high molecular masses (up to 300 kDa for intact collagen). However, this is also the case for ammonia’s 1
H nuclei in view of chemical exchange phenomena and associated 14
N nucleus couplings. Indeed, the application of the macromolecule-filtering CPMG pulse sequence to unheated Product 4 solutions in this solvent gave rise to the complete removal of this signal from the 1
H NMR profiles that were acquired. Further resonances that were removed from these spectra with this pulse sequence were a series of much sharper α-CH
amino acid proton resonances that were located within the 4.2–4.7 ppm range [33
], which presumably also arise from collagenous sources, e.g. its lower molecular mass hydrolytic degradation products. However, this resonance-dampening effect of the CPMG pulse sequence was also observed on a very broad resonance (centered at δ= 9.2 ppm, but spanning the entire 7.1–12.0 ppm range) that was found in the spectra of C2
sequentially treated with (1) NH3
and (2) glacial acetic acid to generate NH4+
therefrom (data not shown).
The total protein, collagen, and ammonia contents of a number of different batches of this fermented CLO product were found to be up to 1.6% (w/w), 1.5% (w/w), and 80 mg/kg, with the latter presumably largely present as ammonium ions, possibly as C2
-soluble ion-pair complexes with free FA carboxylate functions. Indeed, Product 4 has higher concentrations of free FAs in view of the fermentation process that is involved in its commercial preparation [22
]. Water, which might aid the solubility, emulsification, and/or dispersion of collagenous degradation products in Product 4, was found to be present at levels ranging from 0.3 to 1.0 % (w/w) (mean content ca.
0.5% (w/w)). The total collagen concentrations were determined as free hydroxyproline in hydrolysates of CLO sample extracts, using an HPLC method, as described in Section S14.2.2 of the Supplementary Materials
. The total protein content of Products 1–3 was reported as none detectable, i.e. below the specified reporting limit of 0.10% (w/w).
The electronic integration of this broad signal gave an estimated single proton-equivalent concentration as high as 0.44 mol./kg. However, this level was 0.11 mol./kg. for ammonium ion with 4 1H NMR-equivalent proton contributors. These values, which are much higher than those of phenols, flavanones, tannins, and biogenic amines determined in Product 4, are, however, more similar to its estimated ORAC values. Notwithstanding, the above estimated ammonia concentration is markedly greater than that determined by the non-NMR method (ca. only 5 mmol./kg) and, therefore, it appears that the broad signal observed is predominantly attributable to collagenous sources, notably a variable molecular mass range of its hydrolytic degradation products, along with smaller amounts of the intact or virtually intact protein.
The collagen content of Product 4 was also investigated by a SDS-PAGE gel electrophoresis analytical strategy, and the results from these experiments revealed that it had a highly variable molecular mass range, i.e. from 5 to 270 kDa, an observation that was consistent with a high proportion of it being in the form of hydrolytic fragmentation products. Such gelatin-type hydrolysis products presumably arise from the fermentation of collagen substrates during its production. Linear regression analysis confirmed that there was a very strong positive linear relationship between the total FA-normalised overall intensity of this broad resonance and the (% w/w) protein content of this fermented CLO product (r = 0.99, p = 1.45 × 10-4; n = 5).
Similarly, there was also a strong positive correlation between Product 4’s ORAC value and its protein contents (r = 0.91, p = 0.031; n = 5). No significant positive correlations between ORAC values and total phenolics, flavonones, α-TOH and tannins were found. However, there were between ORAC values and the biogenic amine tyramine (r = 0.90, p = 3.56 × 10-2); this observation is consistent with its fermentational source. Moreover, the correlation observed between Product 4 ORAC values and 2-phenylethylamine concentrations was close to statistical significance (p = 0.073). However, that between this amine and tyramine was statistically significant (r = 0.92, p = 2.66 × 10-2), as might be expected from any between-batch differences in fermentation length or conditions during product preparation.
Although the full identity/identities of this broad Product 4 resonance remains complex and is undoubtedly multicomponent, it might be concluded that it is very unlikely to arise from phenols in view of their relatively low concentration. A further possibility is that it partially arises from phenolic-OH
function-containing flavonones and/or polymeric tannins. Notwithstanding, although the final added level of the DTBHQ antioxidant was 6.58 mmol./kg (equivalent to 13.16 mmol./kg phenolic-OH
group equivalents) in CLO/C2
analyte samples, this agent is clearly not also contributing towards this very broad signal, since equivalent concentrations were added to all of the CLO samples investigated here, and it was also not visible in the spectra acquired on Products 1–3. Moreover, this final laboratory-added DTBHQ level is similar to or higher than that of the combined phenol, flavanone, tannin, and tocopherol levels found in Product 4 (Table 1
Product 4 spectra also contained relatively low intensity resonances that are ascribable to C2
- and lipid-soluble low-molecular-mass aromatic compounds located within the δ = 6.70–7.36 ppm range, which are much more 1
H NMR-observable following the removal of the broad, predominantly collagenous peptide-CO-NH
- signal; these are attributable to biogenic amine antioxidants. Indeed, the multiplet resonances centered at δ = 7.23 ppm are assignable to the combined aromatic protons of 2-phenylethylamine in this solvent [35
], which are present in this product at levels that are readily 1
H NMR-detectable (up to 0.85 mmol./kg, Table 2
). Moreover, two clear p
-substituted aromatic ring doublet signals that are located at δ = 6.74 and 6.99 ppm are assignable to tyramine. These assignments were confirmed via the acquisition of 1
H NMR spectra on the aqueous (2
O) and C2
H extracts of this product (Supplementary Materials Section S17 and Figure S1
Interestingly, the spectra acquired also demonstrated that singlet resonances ascribable to both DTBHQ and its corresponding benzoquinone oxidation product were also clearly visible, an example of which is shown in Figure 1
3.3. H NMR Analysis and Time-dependent Monitoring of Secondary Aldehydic LOPs in Thermally-stressed CLOs
H NMR analysis demonstrated the thermally-induced production of aldehydic LOPs in all CLOs evaluated, and Figure 2
shows partial 1
H NMR profiles demonstrating the time-dependent production of aldehyde-CH
O function signals assignable to a range of these cytotoxic/genotoxic agents when Products 1–4 were exposed to TSEs for periods of 0, 30, and 90 min. (assignments for these signals were ratified by one- and two-dimensional 1
H COSY and TOCSY spectroscopic analyses of each heated oil [38
], in addition to the standard addition ‘spiking’ experiments using calibrated authentic standard solutions of these aldehydes in C2
). These spectra also contained resonances that were assignable to cis
- and trans,trans
-CHPDs (conjugated diene olefinic multiplet proton resonances located within the 5.40-6.60 and 5.40-6.30 ppm spectral regions, respectively [38
], along with broad -OOH
function signals that were located at δ = 8.40–8.85 ppm, which represent both CHPDs and HPMs. These primary LOPs remained detectable in spectra that were acquired at and after the 90 min. TSE time-point. Moreover, relatively low concentrations of these aldehydes and their CHPD precursors were also detectable in two of the unheated CLO products tested, notably low-molecular-mass n
-alkanals and cis,trans
-alka-2,4-dienals that are present in the 1
H NMR profiles of Products 2 and 3.
However, it should also be noted that, at the later time-points (≥ 60 min.), one or more additional doublet resonances evolved within the trans,trans
-alka-2,4-dienal aldehyde classification ISB (δ = 9.51–9.55 ppm). This signal heterogeneity was especially the case for Products 2 and 3, but less so for Product 4 (Figure 2
). In principle, these further signals could arise from differing molecular mass homologues of this class of aldehyde, e.g. a chemical shift distinction between trans,trans
-hepta-2,4-dienal (molecular mass 110.15 Da) that arises from the peroxidation of O-3 FA sources [41
] (of total content 21–27 molar %, Table S1
). Specifically, this aldehyde is derived from the fragmentation of the C14- and C16-hydroperoxides of EPA and DHA acylglycerols, respectively. Moreover, trans,trans
-deca-2,4-dienal (molecular mass 152.53 Da) arises from the peroxidation of linoleoylglycerols (ca.
≤ 2–3% (w/w) only). The partial resolution of two sub-classes of these 1
H signals is conceivable at an operating frequency of 600 MHz.
presents the plots of mean ± SEM total FA-normalised concentrations of aldehydes for all nine classes of them. These results show that the pre-fermented Product 4 generates significantly lower levels of these secondary LOPs than those found in Products 1–3 when exposed to TSEs. For example, at the 30 min. TSE time-point, the highest levels of trans
-alka-2,4-dienals, and n
-alkanals formed in Products 1–3 were ca.
1.3, 2.8, and 0.9 mmol./mol. FA, respectively, but only ca.
0.3, 0.6, and 0.25 mmol./mol. lipid, respectively, were found in Product 4. Only a very limited amount of these LOPs were formed in Product 4 at the 10 and 20 min. sampling time points.
Univariate ANCOVA of the experimental data acquired (Equation 1) revealed very highly significant differences between (1) the CLO products investigated (p < 10-9), and (2) heating time-points (p < 10-9) for all of the aldehydic LOP-CHO function ISBs monitored, except for the ‘between-time-point’ effect for n-alkanals, which had a p value of 9.12 x 10-9. Moreover, the CLO product x sampling time-point interaction effect was also very highly significant for all determined aldehyde ISBs (p < 10-9 for trans-2-alkenals; 2.63 × 10-7 for trans,trans-alka-2,4-dienals; 1.78 × 10-8 for 4,5-epoxy-trans-2-alkenals; 3.66 × 10-3 for cis,trans-alka-2,4-dienals; 1.95 × 10-6 for n- alkanals; 2.72 × 10-8 for 4-oxo-n-alkanals; 7.18 × 10-8 for low-molecular-mass n-alkanals; and 2.44 × 10-8 and 1.16 × 10-6 for the unidentified aldehydes U1 and U2, respectively). The mean square estimates for the ‘between-replicates’ components of variance were found not to be statistically significant for any of the ISBs evaluated (p = 0.067–0.17), as expected in view of the high level of reproducibility of the acquired results.
(a) displays a heatmap diagram showing the time-dependent generation of all nine classes of aldehydes in all four CLO products exposed to the above TSEs (mean levels of n
= 3 replicates are plotted for each time-point). Agglomerative hierarchal clustering (AHC) provided valuable molecular information regarding the primary sources of aldehydic LOPs and the mechanism of their generation. For example, trans
-2-alkenals and n
-alkanals derived from both CHPD and hydroperoxymonoene (HPM) precursors, and 4,5-epoxy-trans
-2-alkenals from the oxidation of relatively prevalent trans, trans
The U1 and U2 high-frequency ISB region -CH
O function multiplet resonances may arise from further aldehydes/aldehyde classifications that are not generally encountered in the 1
H NMR profiles of thermally-stressed vegetable-derived culinary oils [26
], e.g. those derived from the peroxidation of O-3 FAs. These include 2,4,7-decatrienal and 3,6-nonadienal from their 9-hydroperoxides, and 3-hexenal from their 12- and 13-hydroperoxides [40
]. However, a further unsaturated aldehyde, which is also generated from O-3 FA sources, is cis, trans
-nona-2,6-dienal (molecular mass 138.21 Da) [40
]. An additional possibility is cytotoxic 4,5-dihydroxy-2,3-decenal [41
]. The intensity of the U2 signal was approximately six-fold greater than that of U1 in all thermally-stressed CLO samples. Notwithstanding, for Products 2 and 4, this U2 resonance was only marginally detectable at all heating time-points (Figure 3
Moreover, malondialdehyde (MDA), which is also selectively derived from the thermally-induced oxidative degeneration of O-3 FAs, has been reported to have a -CH
O function resonance located at δ
= 9.72 ppm [42
]. However, this assignment was only tentative. In C2
solution, the intramolecularly H-bonded cis
-enolic form of this dialdehyde appears to be the predominant species present [43
]. Notably, formic acid, which is a known degradation product of this aldehyde [44
], was detectable in three of the Products tested (Figure 1
). Similarly, if present in thermally-oxidised CLO products, acetaldehyde, which is another MDA degradation product, will feature in and, hence, contribute towards the total intensity of the low-molecular-mass (LMM) aldehyde ISB (δ
= 9.79–9.80 ppm).
The major determinant of the very highly significant CLO product × time-point ANCOVA interaction effect was the differing responses in secondary aldehydic LOP generation occurring at each time-point for all secondary aldehydic LOPs monitored. For example, Product 4 exhibited a significantly longer lag period for aldehyde generation than all of the other products tested, and this observation might indeed arise from higher levels of chain-breaking antioxidants and/or aldehyde-trapping products present therein in view of the fermentation episode employed for its production (Section 2.1
). Such antioxidant enrichment of this product will presumably involve the leaching of such PUFA peroxidation-inhibiting agents, for example, collagenous agents, ammonia, aromatic amines, such as phenylethylamine, phenols, flavonones, and tannins from the cod liver tissue matrix into the separating liquid oil medium during this process.
Notwithstanding, it should also be considered that the relative evaporative losses of the differing aldehydic LOPs generated, each containing a range of differing boiling-point (b.pt) homologues, may also be partially responsible for the non-additive responses observed [26
ANCOVA analysis of the observed ‘between-CLO products’ differences demonstrated that the magnitudes of aldehyde generation were in the increasing orders of product 4 < product 1 ≈ product 2 ≈ product 3 (trans
-2-alkenals); 4 < 1 ≈ 3 ≈ 2 (trans,trans
-alka-2,4-dienals); 4 < 1 ≈ 2 ≈ 3 (4,5-epoxy-trans
-2-alkenals); 4 < 1 < 3 ≈ 2 (cis,trans
-alka-2,4-dienals); 4 ≈ 1 < 2 ≈ 3 (n
-alkanals); 4 < 1 ≈ 2 ≈ 3 (4-oxo-n
-alkanals); 4 < 1 ≈ 2 ≈ 3 (U1); 4 < 1 ≈ 2 ≈ 3 (low-molecular-mass n
-alkanals); and, 4 < 1 ≈ 2 ≈ 3 (U2). Table S2
provides the probability (p
) values of all statistically significant differences found in these post-hoc
ANCOVA analyses. Figure S2b
shows plots of the LSM values of the FA-normalised concentrations of each aldehyde classification monitored for each product tested (i.e. those corrected for highly significant differences between TSE time-points).
Minor resonances that were ascribable to the biogenic amines 2-phenylethylamine and tyramine were found to be completely eliminated from the 1
H NMR spectral profiles of Product 4 when exposed to TSEs of ≥ 10–20 min., and these data indicate their peroxidative consumption during these heating treatments. Section S17 of the Supplementary Materials
provides plausible explanations for and a discussion of this.
3.5. Chemometric Principal Component Analysis (PCA) of the Aldehydic LOP Concentration Patterns Observed at the Extreme (90 min.) TSE Heating Time-point
Finally, principal component analysis (PCA) was performed on the 1
H NMR 9–10 ppm aldehydic function dataset region for each of the CLO products exposed to TSEs at 180 °C, but this model only featured samples that were thermally-stressed at the maximal 90 min. sampling time-point. This form of multivariate analysis revealed that, for the nine detectable aldehyde classifications monitored, there were two clear PCs isolable, with the first and second accounting for 90 and 9%, respectively, of the total variance. The first of these (PC1) was loaded with eight strongly-correlated total acylglycerol FA-normalised aldehyde levels (all aldehydes, except that responsible for the U2 signal, loadings scores 0.96–0.99), whilst the second orthogonal (uncorrelated) PC was loaded with only one of these (U2, loading score 0.75). A clear distinction in the patterns of these secondary aldehydic LOP levels between Products 4 and the remaining three CLOs was clearly notable in a corresponding PCA scores plot (Figure 4
), and this demonstrated that Product 4 had the highest centroidal score values for both PC1 and PC2. Indeed, the 95% confidence ellipses (CEs) computed demonstrated clear discrimination between Product 4 and all other investigated products.
However, 95% CEs for Products 1 and 2, and 1 and 3, overlapped, although that for Product 1 to only a minor extent. However, there were also clear differences in these aldehydic LOP product patterns between CLO Products 2 and 3, with the former PC having lower and higher centroidal PC1 and PC2 score values, respectively, than the latter. With the exception of the unidentified U2 aldehyde, all of these strongly PC1-loading aldehydes loaded negligibly on PC2 (loading scores of only −0.23 to 0.26). Loading scores values of < −0.40 or > 0.40 were considered to be statistically significant.
Although the relative orthogonality of the U2 aldehyde classification-loaded PC (PC2) is not simply explicable, this secondary LOP might arise from a differing peroxidation mechanism, or alternatively might be generated from the chemical modification or oxidation of an alternative aldehydic LOP classification during TSE exposure.
A preliminary receiver operating characteristic (ROC) curve analysis was performed in order to evaluate their levels of classification success from the patterns of aldehydic LOP concentrations monitored at the 90 min. TSE time-point in view of the PCA distinction that was achieved between the fermented and unfermented CLO products. The ROC curve arising therefrom (i.e. a plot of true positive vs. false positive rates) had area under the receiver-operating characteristic curve (AUROC) values 0.923 rising to 0.974 for models featuring from two to eight aldehyde resonance ISB intensities. However, these values were only statistically significantly greater than the null 0.50 value with eight or nine variables considered in view of the limited sample sizes available (n = 3 replicates per product). Therefore, this preliminary analysis provided evidence that this multivariate analysis approach might successfully discriminate between the fermented from unfermented CLO products on the basis of the patterns and concentrations of thermally-inducible aldehydic LOPs generated at the 90 min. TSE time-point.