Ultra high efficiency/low pressure supercritical fluid chromatography (UHE/LP‐SFC) for triglyceride analysis: Identification, quantification, and classification of vegetable oils

Abstract

Triglycerides of vegetable oils have been extensively studied. Non‐aqueous reversed‐phase liquid chromatography and silver‐ion chromatography are most frequently used to achieve their separation. In previous works, we presented the use of supercritical fluid chromatography with long columns (75 cm) packed with fused‐core particles to provide ultra‐high‐performance separations, with a low‐toxicity fluid (carbon dioxide) compared to the usual liquid‐phase methods. In the present paper, we describe the quantification of triglycerides with supercritical fluid chromatography and evaporative light‐scattering detection. Thanks to the isocratic elution mode, this quantification can be simplified, assuming (a) identical response coefficients for compounds having a close structure, (as only triglycerides are quantified), and (b) constancy of the response coefficient along the analysis (no elution gradient). Therefore, the relative concentrations of triglycerides were easily assessed. Only one calibration curve for one reference compound (in this case triolein) was required. The resulting relative concentrations are in good accordance with the numerous publications available. Relative quantification with UV detection at 210 nm is also proposed, facilitated by the very low UV absorption of carbon dioxide and with a calibration curve taking account of the variation of UV response according to double bond number. Nineteen vegetable oils are compared. The identification of triglycerides was carried out based on previous knowledge of these oils, but also with the help of a Goiffon retention diagram, based on the relationship between the logarithm of retention factor and the total double bond number. Finally, cluster analyses were computed, based on evaporative light‐scattering detection or UV quantification data. They allow a quick comparison of the triglyceride content between the oils, in the goal to exchange one by the other for certain applications, or to compare a new oil to well‐known ones.

1. INTRODUCTION

Triglycerides (or triacylglycerols, TG) are generally separated either by non‐aqueous reversed‐phase liquid chromatography (NARP‐LC) with C18‐bonded silica stationary phases, or by silver‐ion (or argentation) liquid chromatography (Ag‐HPLC), on silica phases impregnated with silver nitrate. They generally result in different separations as NARP‐LC retains the analytes according to the equivalent carbon number (ECN, defined by the difference between the total number of carbons and twice the number of double bonds in carbon chains), while silver‐ion chromatography retains the analytes according to the double‐bond number. These two methods are well suited for these hydrophobic compounds, because the mobile phases contain no water, that is, are composed only of pure organic solvents that dissolve the lipids well. However, beyond the use of toxic organic solvents, both methods display other drawbacks. First, elution gradients are required, resulting in an additional equilibration time. Second, in Ag‐HPLC the stability of the silver coating over several analyses is poor, so the accurate comparison of many samples is difficult. Whatever the HPLC method, there are always unresolved pairs of peaks. This is mostly the case for TG having the same ECN for NARP‐LC or the same unsaturation number for Ag‐HPLC.

To identify and quantify triglycerides in vegetable oils, various detectors coupled to chromatographic methods can be employed. ¹ , ² , ³ The UV absorbance of TG is rather weak, requiring measurement at low wavelengths to achieve a measurable response. Unfortunately, the strong UV absorption of acetone or methylene chloride, which often compose mobile phases in NARP‐LC, hinder UV detection. ⁴ Evaporative light‐scattering detector (ELSD), Corona charged‐aerosol detector (CAD), and mass spectrometry (MS) (with atmospheric pressure chemical ionization – APCI, but also occasionally electrospray ionization – ESI) are most commonly used. Unfortunately, with these detectors, the response coefficients vary depending on mobile phase composition, ² , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² while a mobile phase gradient is always necessary to ensure the elution of all TG in liquid chromatography. As a result, to quantitate the analytes, it is necessary to carry out multi‐calibration, ⁶ , ⁷ , ⁹ , ¹⁰ , ¹¹ or post‐column addition, ¹³ or post‐column gradient compensation, ⁶ because of the continuous change of the mobile phase composition in gradient elution conditions. Despite relevant and accurate results, these approaches are rather time‐ or solvent‐consuming.

TG separation can also be achieved with supercritical fluid chromatography (SFC) with mobile phases composed of carbon dioxide and a co‐solvent (or modifier). The first reports employed silver‐ion SFC, ¹⁴ , ¹⁵ , ¹⁶ possibly coupled to mass spectrometry. ¹⁷ Later, reversed‐phase SFC analysis on C18‐bonded phases was also demonstrated. ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² The relationships between the chemical structures and the retention behavior, in regards of modifier nature and percentage, were extensively studied, as well as the effects of temperature and backpressure changes. ¹⁶ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ While retention rules observed in NARP‐LC on C18 phases also apply in SFC, some specific behaviors were reported, such as the reversal of the retention order between TG differing by one palmitic (P) or one oleic (O) chain. As described in Table S1 (supplementary information), P chains have 16 carbon atoms and no double bond, while O chains have 18 carbon atoms and 1 double bond. This particular point resolves most of the difficulties encountered in NARP‐LC for pairs of peaks such as PLO/OLO. This explains the generally higher performance obtained in SFC compared to one‐dimension LC, based on improved selectivity together with high peak capacity.

Moreover, carbon dioxide‐based mobile phases have very low viscosity. For example, the viscosity of pressurized CO₂‐solvent 90:10 mixtures (typical composition in the analysis of TG), is about 1/10^th of the viscosity of liquids. This allows for high flow rates (typically 3 to 5 mL/min) and/or long column lengths to achieve higher efficiency or to combine selectivities of different stationary phases. Overall, higher SFC separation performance is observed, without any dramatic increase in the analysis duration compared to LC methods.

The detectors used in LC and SFC are identical (ELSD, CAD, MS). SFC however has the advantage of the facilitated use of UV at low wavelengths, because of the weak UV absorption of CO₂. The molar extinction coefficient obviously depends on the chemical structure of analytes, but, as opposed to ELSD or MS, the UV detector response does not depend on the mobile phase conditions. ¹⁷ The advantages and the drawbacks for ELSD and CAD are identical in SFC to those observed in LC. ²³ , ²⁴ The main drawback for ELSD detection is the change in the coefficient response when working in gradient conditions.

Previously, ²⁵ we developed an SFC method using isocratic elution conditions, which is avoiding this major quantification issue. Isocratic elution in SFC is facilitated by the high solubility of TG in CO₂‐solvent mixtures, whereas the lower solubility of TG in liquid mobile phases requires a gradient elution mode in LC methods. This SFC approach was based on several coupled columns (for total length of 75 cm) filled with sub‐3 µm superficially porous particles, which provides a high efficiency but avoids reaching prohibitive inlet pressure. The method was called Ultra‐High Efficiency/Low Pressure‐Supercritical fluid Chromatography (UHE/LP‐SFC), as opposed to ultra‐high‐performance liquid chromatography (UHPLC) that is associated to high pressure drops to achieve high efficiency. The use of superficially porous particles was also reported in isocratic NARP‐LC to determine the botanical origin of oils with low‐wavelength (205 nm) UV detection, ³ or with small column length and ESI‐MS detection, ¹² or by coupling four C18 columns (60 cm) for a fine separation of menhaden oil TG. ²⁶

Besides, without the use of MS detection, identification of TG requires the cross‐correlation of data. Firstly, the knowledge acquired from previous studies on classical vegetable oils, which were largely described in the literature, is useful to annotate the major peaks. Secondly, a classical retention diagram, called Goiffon diagram, ²⁷ , ²⁸ provides additional suggestions to identify minor compounds that would not have been described previously, or to annotate the chromatograms of previously undescribed oils. This diagram is based on the identical energy transfer, i.e. retention variation, which is related to an identical structural modification between two analytes. For instance, increasing the number of methylene groups causes a regular variation in retention. The same effect can be observed when varying the number of double bonds. It is then possible to draw straight lines by plotting the logarithm of retention factor (log k) as a function of the total number of double bonds for TG families. For instance, in the group (PPP, PPL, PLL, LLL), where the total number of double bonds (DB) is 0, 2, 4, and 6, a plot of log k versus DB is a straight line. Drawing such plots for the previously identified TG is helpful in identifying more TG.

In the present paper, 19 vegetable oils are analyzed with UHE/LP‐SFC‐UV‐ELSD. Goiffon diagrams are provided to propose possible identification of the observed TG. Quantification is achieved with both ELSD and UV. Finally, the lipid content of vegetable oils is compared, in the goal to classify the oils having close TG composition. Hierarchical cluster analysis (HCA), based on the relative abundance of TG in each oil, provides a simple and visual classification.

2. MATERIAL AND METHODS

2.1. Chemicals and samples

Carbon dioxide of industrial grade 99.5% was provided by Messer (Puteaux, France). The solvents used for the chromatographic analysis and oil dilution were HPLC‐grade methanol (MeOH), acetonitrile (ACN), and dichloromethane provided by VWR, Fontenay‐sous‐Bois, France).

Vegetable oils were kindly provided by Olvea (Saint‐Leonard, France): Almond, Apricot‐kernel, Argan, Avocado, Corn, Cotton, Linseed, Macadamia, Olive, Peanut, Pistachio, Rapeseed, Safflower, Sesame, Sunflower, Walnut, Wheat germ; by LVMH Recherche (Saint‐Jean‐de‐Braye, France): Kniphofia Uvaria seeds; and obtained from Colombia: Kahai nut (Caryodendron orinocense nut). The oils were dissolved in methanol‐dichloromethane 50:50 mixture with a 1/5 dilution factor.

2.2. Instrument and operating conditions

The instrument was a Jasco system (Tokyo, Japan) equipped with a Gilson UV 151 detector (provided by Waters, Guyancourt, France) and a Sedex 85 ELSD detector (provided by Sedere, Alfortville, France), as previously described. ²⁵

The separation method and the sample preparation are described in the previous paper describing the development and optimization of ultra‐high efficiency low‐pressure supercritical fluid chromatography (UHE/LP‐SFC). ²⁵ Briefly, five columns were connected in series, four Kinetex C18 (150 × 4.6 mm, 2.6 µm superficially porous particles, provided by Phenomenex, Le Pecq, France) and one Accucore C18 (150 × 4.6 mm, 2.6 µm superficially porous particles, provided by Thermo‐electron, les Ulis, France). This combination of stationary phases was previously optimized to provide both high efficiency and excellent selectivity, as the two stationary phases are complementary in resolving critical pairs of peaks. The mobile phase was CO₂/ACN/MeOH 88/10.8/1.2 v/v/v, flow rate 1.8 mL/min; column oven temperature 17°C; back‐pressure regulator 10 MPa, 60°C. The UV detection wavelength was set at 210 nm. The ELSD conditions were as follows: nebulizer gas: N₂, 3 bars; drift temperature 40°C. The injection volume was 1 µL.

2.3. ELSD quantification

As described in the introduction, the ELSD response should be mostly the same for the TG examined in the present study, because they have comparable size and volatility, and because they were all eluted with the same mobile phase composition (isocratic elution). For ELSD, the relationship between m_i (mass of analyte injected) and A_ELS (peak area measured with the ELS detector) is described by the simple equation:

$${\mathrm{A}}_{\mathrm{ELS}}={\mathrm{a}}_{\mathrm{ELS}}\times {\mathrm{m}}_{\mathrm{i}}^{\mathrm{b}}$$ (1)

where a_ELS is the ELSD response factor and b is a constant. Thus, a linear relationship is obtained as follows:

$$\log {\mathrm{A}}_{\mathrm{ELS}}=\mathrm{b}\times \log {\mathrm{m}}_i+\log {\mathrm{a}}_{\mathrm{ELS}}$$ (2)

where log a_ELS is the response for an injected mass equal to 1 (in that case, log m is equal to zero and log A_ELS = log a_ELS).

A calibration curve was prepared with OOO standard as follows. Five solutions were prepared at concentrations of 342.75, 152.33, 101.55, 67.70 and 45.14 mg/ml. Each solution was analyzed in triplicate by SFC‐ELSD (full data in Table S2, supplementary information), and the average peak areas were used to plot the calibration curve (Figure 1) using equation 2. The b slope is equal to 0.8912 and log (a) is equal to 2.5342. The graph of normalized residuals is provided in Figure S1 (supplementary information) and shows that the residuals are well scattered and not dependent on concentration.

FIGURE 1.

Calibration curve for SFC‐ELSD quantification: logarithm of peak area obtained with OOO standard solutions, versus the logarithm of OOO concentration. Full data are presented in Table S2 (supplementary information) and the graph of standardized residuals in Figure S1

Previous studies showed that, in SFC conditions with the ELSD Sedex 85 instrument used in these experiments, the b coefficient does not change for different TG, meaning that log a_ELS values are similar and could be eliminated for ratio calculation. ²⁹

Combining two identical equations (1) for two different TG, we can calculate the mass ratio for one TG (m₁, A₁) relative to the main peak (m_max, A_max).

$$\frac{m_1}{m_{max}}={\left(\frac{A_1}{A_{max}}\right)}^{\frac{1}{b}}$$ (3)

Finally, the results were expressed as relative concentrations, so that the sum of the masses was 100%:

$$C_i\%=\frac{\frac{m_1}{m_{\max }}}{\sum \frac{m_i}{m_{\max }}}\ast 100$$ (4)

Full results for the 19 oils are presented in Table S4 (supplementary information).

2.4. UV quantification

The UV absorbance of TG at 210 nm results from (a) the ester functions linking the hydrocarbon chains to the glycerol head, which should be the same to all TG, and (b) the double bonds in the hydrocarbon chains, which would differ between the TG. Judging from this observation, the TG having different chain length and identical double bonds show close UV absorbance, while TG with different numbers of double bonds have significantly different absorbance. UV response coefficient were thus calculated based on published data ³⁰ for 12 TG having different chain lengths and different total numbers of double‐bonds (data in Table S3, and Table S1 to identify TG).

The linear relationship between C_i (concentration) and A_UV (peak area measured with the UV detector) is described by the simple equation:

$${\mathrm{A}}_{UV}={\mathrm{a}}_{\mathrm{UV}}\times {\mathrm{C}}_{\mathrm{i}}$$ (5)

where a_UV is the response factor. The a_UV values for 12 TG obtained from the literature are presented in Table S3. From this table, a calibration curve was plotted of the a_UV response factor vs. the total number of double bonds (DB) (Figure 2). Thus, the a_UV response factor was assessed for any TG with known DB value, based on the best fitting curve:

$$\begin{array}{ccc}\hfill {\mathrm{a}}_{\mathrm{UV}}& =& -0.0154\times \mathrm{D}{\mathrm{B}}^4+0.259\times \mathrm{D}{\mathrm{B}}^3-0.9783\times \mathrm{D}{\mathrm{B}}^2\hfill \\ & & +1.5899\times \mathrm{DB}+0.2562\hfill \end{array}$$ (6)

FIGURE 2.

UV response coefficient (A) versus the total number of double bonds in 12 TG (based on data from ref. ³⁰ presented in Table S3). UV response coefficient for any TG was calculated from this calibration curve. The graph of standardized residuals is available in Figure S2

The graph of normalized residuals is provided as Figure S2 (supplementary information) and shows that the residuals are well scattered and not dependent on a double‐bond number.

Then, the C_i concentration of any identified compound (ie, knowing the DB value) was back‐calculated from the peak area (A_UV) and a_UV values from Equation ( (5).

Finally, relative concentrations were calculated as C_i% so that the sum of the concentrations was 100%:

$$C_i\%=\frac{C_i}{\sum C_i}\ast 100$$ (7)

Full results for the 19 oils are presented in Table S5 (supplementary information).

2.5. Data analysis

Hierarchical cluster analyses (HCA) were performed using XLStat 19.03.44850 software (Addinsoft, New York, NY). HCA were computed on the data in Tables S4 and S5 with Ward aggregation method and Euclidean distances. Because the data in these tables are already relative concentrations, no normalization of data was done prior to calculation.

3. RESULTS AND DISCUSSION

3.1. Quantification

Nineteen vegetable oils were selected and analyzed with the SFC‐UV‐ELSD method. Quantification of all the peaks was completed first (identified and unidentified peaks), so the relative quantities could later support peak identification, based on previous knowledge on the oil composition. As detailed in the experimental section, the concentration ratio (UV) and the mass ratio (ELSD) were calculated as the ratio between the area of the studied peak and the area of the main peak (for each vegetable oil). The relative amounts, expressed as percentage values, are presented in Table S4 for ELSD data and Table S5 for UV data. Only the peaks with relative amount larger than 0.1% are reported and the major components (relative amount > 10%) are highlighted with an asterisk.

3.2. Identification

The identification of numerous TG is facilitated when knowing the main compounds contained in vegetable oils, i.e. by examining the compositions previously determined with HPLC or SFC analyses. ² , ⁵ , ⁶ , ⁷ , ⁸ , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ³⁰ For the peaks presented in Tables S4 and S5, chromatographic diagrams can also be used. Figure 3 shows the Goiffon diagram (log k vs. DB) obtained from the retention data measured in the present analysis conditions. ²⁵ This diagram was constructed from the analyses of Argan, Rapeseed, Macadamia, Peanut and Linseed oils, because of their well‐known and varied TG composition. Argan oil is the simplest one, mainly composed of P, L, O and S chains. Rapeseed oil contains Ln chains and critical pairs of TG like LLLn/OLnLn, LLL/OLLn, and OLL/OOLn. Linseed oil contains a great amount of Ln chains leading to the presence of many TG containing two Ln chains. Avocado and Macadamia oils both have many TG with P and Po chains (16 carbon atoms, with respectively 0 and 1 double bond). Finally, Peanut oil contains longer chains of 20 (arachidic and gadoleic), 22 (behenic), and 24 carbon atoms.

FIGURE 3.

Goiffon diagram of the logarithm of retention factor vs. the total number of double bonds, for the TG observed in the 19 vegetable oils. Analytical conditions: see details in experimental part. For the sake of clarity, not all useful lines have been drawn in this figure

As expected, by using C18‐bonded stationary phases, this diagram shows that retention decreases when the double bond number increases, with straight lines connecting compounds of one family. For instance, PPP, PPO, POO and OOO are aligned; PPP, PPL, PLL, and LLL are also aligned, etc. The four compounds in each line are called homologous series.

One major difference from NARP‐LC is that the slope of the line joining PPP, PPO, POO, and OOO is positive in SFC, while it is negative in NARP‐LC. ²⁶ The same is observed for the line joining PoPoPo, PoPoL, PoLL, and LLL. It means that the solubility of Palmitate chain (P, C16:0) in the subcritical mobile phase is higher than the one of Oleate (O, C18:1), and the solubility of Palmitoleate (Po, C16:1) is higher than the one of Linoleate (L, C18:2), whereas the reverse is observed for the mobile phases used in NARP‐LC. As mentioned in the introduction, this different retention and solubility between NARP‐LC and SFC partly explains the improvement of resolution obtained in SFC, which is not only related to efficiency performance.

With the Goiffon diagram and based on simple chromatographic rules, one can more easily address the identification of an unknown peak. However, as reported in previous studies, ²⁰ , ²¹ , ²² some co‐elutions remain between critical pairs of TG. For instance, SSL (C18:0/C18:0/C18:2) and ALP (C20:0/C18:2/C16:0) are co‐eluted. These two TG have the same carbon number, partition number, total double bond number, and the same number of double bond by chain, that is, the same number of chains having no double bond (two for SSL and ALP). Fortunately, all remaining co‐elutions concern minor compounds. Despite a satisfactory separation, several other difficult identifications concern TG having identical partition number with the same total carbon number and the same double bond number such as LLL and OLLn (found in Rapeseed oil), PoLL and PLLn (found in Pistachio oil), PoOL and PLL, or PoOO and POL (both found in Macadamia oil). Identifying such peaks can be unsuccessful from a simple examination of the chromatogram. Consequently, the use of chromatographic behavior models to achieve identification is an attempt based on objective data. Obviously, to ensure the absolute identification, especially for minor compounds, numerous works showed that the use of MS was required. ² , ⁴ , ⁶ , ⁷ , ⁸ , ¹² The trivial names indicated in Tables S4 and S5 are thus proposed identification, which will require confirmation with further SFC‐MS experiments.

3.3. Comparison of oils based on TG composition

Hierarchical cluster analyses (HCA) were computed, based on the data in Table S4 (ELSD) and Table S5 (UV). The purpose was to define clusters of oils that contain similar distribution of TG. HCA is an unsupervised classification method allowing to group objects (in this case, the 19 vegetable oils) in terms of similarity. Similarity was evaluated by the measurement of Euclidean distance between pairs of vegetable oils defined by their coordinates (the proportions of 50 TG with relative amounts > 0.1%). The result is visualized as a dendrogram (Figure 4), providing a simple and visual representation of the samples. In the dendrogram, the oils linked by a short horizontal line and placed close‐by on the vertical axis are similar, i.e. they have a similar TG composition. Conversely, the oils linked by a long horizontal line are dissimilar, i.e. they have different TG compositions. This allows for an intuitive interpretation of the data, which is much easier than comparing a data table.

FIGURE 4.

Hierarchical cluster analysis from (A) ELSD data (in Table S4) and (B) UV 210 nm (in Table S5) to classify the 19 vegetable oils. Identical colors of the oil name indicate belonging to the same cluster, defined by the cutting lines (interrupted grey lines)

Figure 4A shows the HCA based on ELSD quantification data (Table S4). First, to understand how the level of dissimilarity can be related to the TG composition, let us take one example with the three following samples: Apricot kernel, Almond and Rapeseed oils. As shown in Figure 5, for these three vegetable oils, the main peaks are the same (OOL, OLL and OOO), but also minor peaks are the same (eg, LLL and SOL). However, Rapeseed oil contains additional TG, mainly due to the presence of Linolenic acid (Ln, C18:3), as in OLLn (5.3%), and OOLn (12%), and in several minor peaks as PLLn (0.2%), OLnLn (0.7%), and POLn (0.9%). Consequently, it is understandable that these three oils are located in the same cluster of the dendrogram, due to the major peaks they have in common. However, Rapeseed oil is slightly dissimilar from Almond and Apricot kernel, due to the presence of minor TG containing Ln chains. More significant differences will naturally lead to larger dissimilarity levels between the oils, appearing as longer horizontal lines in the dendrogram.

FIGURE 5.

ELSD chromatograms of Almond (red lower trace), Apricot Kernel (blue medium trace) and Rapeseed (green upper trace), belonging to the same HCA cluster. Apricot kernel and Rapeseed traces were shifted (dotted blue lines) to facilitate comparison of the overlaid chromatograms

The cutting of the dendrogram serves to define clusters of similar samples. First, a simple cutting divides the oils into two major groups: Avocado, Macadamia, Olive, Linen, Apricot kernel, Almond, Rapeseed, Peanut, Sesame, Pistachio, Argan (top of the dendrogram); Kahai, Kniphofia, Walnut, Safflower, Sunflower, Corn, Cotton, Wheat germ (bottom of the dendrogram). The origin of the large dissimilarity observed between the two groups can be found by examining Table S4: the major TG (with concentrations typically above 10%) that are present in the top group are OLL, OOL, POO, and OOO, while the bottom group is characterized by large concentrations of LLL, PLL, and OLL.

This simple clustering is however insufficient, judging from the diversity in TG composition observed among the samples. The appropriate cutting level is best evaluated by comparing the chromatograms to decide on the desired similarity. Lowering the dissimilarity level below 1000 (see dotted line in Figure 4A) reveals 8 clusters: (a) Avocado and Macadamia; (b) Olive; (c) Linen; (d) Apricot kernel, Almond, and Rapeseed; (e) Peanut, Sesame, Pistachio, and Argan; (f) Kahai nut, Knifophia uvaria seeds, Walnut, and Safflower; (g) Sunflower and Corn; (h) Cotton and Wheat germ.

Figure 4B shows another HCA dendrogram of vegetable oils on the basis of the UV responses (Table S5). First, it is important to note that the figure is closely resembling the ELSD dendrogram (Figure 4A). Indeed, the oils are grouped in the same clusters as before, with only minor changes observed: (a) Pistachio is now grouped with Apricot kernel, Almond, and Rapeseed ; (b) Sunflower and Corn are no longer a separate cluster but are now grouped with Kniphofia, Kahai nut, Walnut, and Safflower. The similar classifications obtained are proving the relevance of the two different quantification methods employed with ELSD and UV. Because of this similarity and for the sake of clarity, the following discussion is based solely on the ELSD dendrogram (Figure 4A).

Peanut, Sesame, Pistachio, and Argan display very close TG profile and are included in one cluster, showing that these four oils are more or less identical. However, Figure 6, Table S4, and Table S5 show that Sesame and Argan display higher amounts of minor TG containing stearic chains (C18:0) with SLL, SOL, and SOO TG, a higher percentage of LLL and PLL, but also lower amounts of OOL. Argan and Pistachio have higher relative amounts of POO and OOO and lower amounts of OLL compared to Sesame and Peanut.

FIGURE 6.

ELSD chromatograms of Pistachio (lower green trace), Argan (second blue trace), Peanut (third red trace) and Sesame (upper orange trace). The chromatographic traces were shifted (see interrupted grey lines) to facilitate comparison of the overlaid chromatograms

Figure 7 displays the chromatograms of Macadamia and Avocado, located in the same cluster. They share comparable proportions of PoPO and OOO, but Macadamia that is well known to have a high amount of C16:1 (Po) chain ⁵ , ³¹ has more PoPL and PoOO, while Avocado has more POL and POO. ⁷ Olive, in a different cluster but not too far from these two, contains comparable amounts of OOL and POO, but much larger quantities of OOO. ⁶ In Figure 7, the chromatogram of Avocado oil is also illustrating the impressive separation performance achieved by UHE/LP‐SFC in comparison to previous methods using NARP‐LC, ⁷ as several critical pairs are well resolved, which remained unresolved with previous LC methods. This separation performance was also reported in our previous paper ²⁵ with the chromatogram of Rapeseed oil, which is one of the most complicated oil in terms of mixtures of C18‐chains TG (S, O, L and Ln).

FIGURE 7.

ELSD chromatograms of Macadamia (lower blue trace) and Avocado (upper red trace)

In the top section of the dendrogram, Linen is singled out due to its abundant content of Ln chains, typically in LnLnLn, LLnLn, PLnLn, LLLn, and OLnLn.

The proposed classification also allows to compare “new” vegetable oils to well‐known ones. For instance, Knifophia Uvaria belongs to the family of Asphodelaceae, coming from Africa. This oil can be used as anti‐aging ingredient in cosmetic formulations. ³² Kahai nut is produced in Colombia and could also be an interesting oil for cosmetic applications. Figure 8 displays the chromatograms of Kniphofia Uvaria, Kahai nut, and Safflower, which are included in the same cluster (Figure 4). These three oils, together with Walnut oil, mainly contain LLL, PLL, and OLL, with lower amounts of POL, OOL, and OOO. For Safflower, the percentage of LLL represents 63% of the total TGs, whereas for the two others, LLL is also the main TG but in lower proportions (46 and 37%, see Table S4). Corn and Sunflower also contain a lot of L chains, but the major TG is OLL rather than LLL, explaining the larger dissimilarity from the previous oils, placing them in a different cluster. Finally, Cotton and Wheat germ also contain a lot of L chains, the major compound being PLL. These last three clusters are rather close to each other due of the great amount of L chains in their compositions.

FIGURE 8.

ELSD chromatograms of Knifophia Uvaria (lower blue trace), Kahai nut (medium red trace), and Safflower (green upper trace). Safflower and Kahai nut were shifted (dotted blue lines) to facilitate comparison of the overlaid chromatograms

Finally, it is interesting to compare the chromatograms obtained with UV and ELSD detection to understand the minor differences observed between the dendrograms in Figure 4, but also to illustrate the complementarity of the two detectors. Figure S3 shows the chromatograms of Walnut oil. A slight retention shift appears between the two chromatograms because the UV detector was located before the back‐pressure regulator and the ESDL after. As expected, the peak height of the main compounds is improved with UV for TG having Ln chains, thus containing more double‐bonds. The relative responses of PLL vs. OLL and POL vs. OOL are very close between UV and ELSD. Those two pairs of compounds differ by one double bond on the Oleate chain. On another hand, the LLnLn, PLLn and OLLn peaks are higher with UV, relatively to LLL. It probably results from significant UV absorbance of the three double bonds in Ln chains.

In Figure S3, the asterisks indicate the additional minor compounds detected with UV, i.e. at least 14 compounds. The presence of Mo (C17:1) chains was reported elsewhere, ⁷ but the calculation of the partition number of either MoLnLn (PN = 39) or MoLLn (PN = 41) does not clearly discriminate these TG. Other studies reported the presence of very small amounts of Po (C16:1) in Walnut. ³⁰ PoLnLn (PN = 38), PoLLn (PN = 40) and PoLL (PN = 42), could correspond to some of the additional peaks (as suggested in Table 2). In that case, mass spectrometry would undoubtedly be useful to achieve more precise identification.

The higher sensitivity of UV with regards to ELSD and the response changes can also be observed on Figure S4, by comparing the UV and the ELSD profiles of Pistachio oil. Again, many additional minor compounds (<0.5%) marked with an asterisk are observed with UV detection. In this chromatogram, the number of TG peaks observed with UV is around 30. Some new peaks observed at the beginning of the TG elution zone, before 30 min, could have Mo (C17:1) chains, ⁵ and also M (C14:0) ones, in combination with Ln chains (eg, MLnLn; MLLn). ³³ Using retention rules as described above, some compounds observed with UV detection could also have Po (C16:1) chains. These TGs elute just before TGs having the same total chain length and the same double bond number, for instance PoLL/PLLn; PoOL/PLL; PoOO/POL.

Finally, on both Figures S3 and S4, it is interesting to note that the band dispersion in ELSD was not much larger than in UV. This is related to the composition of the mobile phase (small proportion of methanol in CO₂) and to the total introduction of the column effluent in the ELSD. The small decrease in chromatographic performances for ELSD, due to classical dispersion effects into the open cell, seems rather limited, inducing some loss in resolution only for the heavier, late‐eluting TG.

4. CONCLUSIONS

The use of supercritical fluids with superficially porous particles dramatically improves the chromatographic performance to resolve TG in vegetable oils. In addition, the isocratic elution mode favors simple relative quantification of the compounds by using two detectors in series.

UV and ELSD quantification of the relative TG proportions yielded very similar HCA classification of the 19 vegetable oils, with only minor differences between them. This is validating the two different approaches for quantifying TG with UV and ELSD. Basically, the 19 oils could be divided into 8 clusters showing different similarity levels between the oils. In addition, the sensitivity of UV at low wavelength (210 nm) is undoubtedly a great advantage in SFC vs. HPLC, as it revealed minor compounds having additional double bonds (Ln and Po chains). However, these minor compounds may have limited interest in the purpose of comparing oil compositions, either to substitute one oil by another for specific application, or to examine product adulteration or oil origin. ³⁴ , ³⁵

Finally, comparison to literature data and the use of retention diagrams together provided possible identification of all major peaks observed. For minor peaks, where some uncertainty remains, hyphenation to mass spectrometry will be necessary, and will be the topic of future works.

Supporting information

Supporting information

Lesellier E, Latos A, West C. Ultra high efficiency/low pressure supercritical fluid chromatography (UHE/LP‐SFC) for triglyceride analysis: Identification, quantification, and classification of vegetable oils. Anal Sci Adv. 2021;2:33–42. 10.1002/ansa.202000156

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.