Triacylglycerol Analysis in Human Milk and Other Mammalian Species: Small- Scale Sample Preparation, Characterization and Statistical Classification Using HPLC-ELSD Profiles

In this work, a method for the separation of triacylglycerols (TAGs) present in human milk and other mammalian species by reversed-phase high performance liquid chromatography using a core-shell particle packed column with UV and evaporative lightscattering detectors is described. Under optimal conditions, a mobile phase containing acetonitrile/n-pentanol at 10 oC gave an excellent resolution between more than 50 TAG peaks. A small-scale method for fat extraction in these milks (particularly of interest for human milk samples) using minimal amounts of sample and reagents was also developed. The proposed extraction protocol and the traditional method were compared, giving similar results, with respect to the total fat and relative TAG contents. Finally, a statistical study based on linear discriminant analysis on the TAG composition of different types of milks (human, cow, sheep and goat) was carried out to differentiate the samples according to its mammalian origin.


INTRODUCTION
According to the World Health Organization, breastfeeding is the recommended way of providing to young infants the nutrients for a healthy growth and development.
This recommendation is based on knowledge that breast milk from healthy and well-fed mothers, provides sufficient energy and proper profile of nutrients to support normal growth and development of term infants, without any additional foods through the first 4 to 6 months of life.
1 Human milk constitutes a very complex fluid, which contains carbohydrates and salts in solution, caseins in colloidal dispersion, cells and cellular debris, and lipids mostly in emulsified globules. 2 Since lipids are the main energy source in human milk, contributing in 40-55% to the total energy intake, its compositional and physiological aspects have been of interest and research in the last decades and recently reviewed. 3 According to Koletzko et al., 4 the average amount of fat contained in human milk is ca. 3.8-3.9 g/100 mL, but this value varies widely. Thus, human milk is a dynamic system whose lipid composition is influenced by factors such as mother's diet, 5 stage of lactation, 6,7 phase of the feeding 2 and breast. 2,8 However, any difference has been observed with regard to frequency of breastfeedings 9,10 and time of day. 11,12 In spite of the good features in human milk, research to find valuable alternatives, especially when breastfeeding is not possible or may not be advisable, constitutes a high priority. To be nutritionally adequate, any substitute should have the same nutritional characteristics as breast milk. In addition, it should be hypoallergenic and palatable.
this sense, commercial infant formulas, usually based on mammalian milks such as cow, buffalo, donkey, sheep, camel, and goat milk, may represent an alternative to fulfill the nutritional needs of newborns. 14 However, these milks are different from human milk in terms of chemical composition (e.g. protein and fat contents), which may cause nutritional and immunological problems.
15 Regarding fat content, limited studies have been done so far to systematically compare the lipid composition in different mammalian milks. 14 Triacylglycerols (TAGs) represent 98% of total lipid fraction, and despite the changes in human milk composition, some TAGs, such as lauric acid-oleic acid-linoleic acid (LaOL), myristic acid-oleic acid-linoleic acid (MOL), can be considered as markers of the mature human milk. 16 Moreover, fatty acids (FAs) represent 90% of these TAGs, that is, 88% of total lipid, 2 and a balance ratio between ω-3 and ω-6 FAs in human milk is important to ensure the healthy growth of infants. 14 Although the analysis of FAs in human milk is much easier than TAGs evaluation, milk FAs are secreted, consumed and hydrolyzed as TAGs in globules. 2 For this reason, it is of great importance to achieve a reliable TAGs determination. These latter gradients led to low resolution of TAGs peaks, and the first-mentioned gave long analysis time (90 min).
In order to improve chromatographic performance in terms of throughput and/or resolution, particularly when numerous complex food extracts have to be analyzed, recent advances in LC instrumentation could be beneficial. 33 In this context several analytical strategies related to column technology have been developed in HPLC, including the use of monolithic supports, packed columns with sub-2 µm particles operating at ultra-high pressure (UHPLC) or with core-shell or fused-core particles. These latter core-shell particle columns are capable of maintaining high efficiencies at increasing flow rates with the subsequent reduction of analysis time. Also, these columns operate comfortably within the pressure limits of conventional LC instruments, rivaling the performance obtained with sub-2 columns on UHPLC instruments. However, the use of these columns in conventional LC systems for TAGs separation in human milk samples has not been reported to date.
Multivariate data analysis can be used to obtain more information on major, minor, and trace components in foods. 34 Within these statistical tools, linear discriminant analysis (LDA) is probably one of the best known methods and it has been successfully used for the identification/differentiation of several foods, such as dairy products, oils, wines and others. 36,37 In this work, the development of analytical conditions for the extraction and RP-HPLC separation of milk fat TAGs is described. For this purpose, a small-scale fat extraction protocol (with reduced consumption of reagents and processing time, in consistency with the recent trends in green chemistry) and in combination with the use of a fused-core HPLC column is presented and compared. In addition, on the basis of TAG profiles of milk samples from different mammalian species (human, bovine, caprine and ovine), a LDA model is applied to differentiate these matrices according to its species origin. probability values of F in and F out , 0.05 and 0.10, respectively, were adopted.

RESULTS AND DISCUSSION
Optimization of the Separation of TAGs. The initial separation conditions were adapted from a previous work for TAGs separation in vegetables oils. 39 The optimization study was performed using human milk samples. Using a flow rate of 1.5 mL min -1 and a column temperature of 10 ºC, a similar optimization study of mobile phase, including several binary mixtures of ACN with different alcohols (2-propanol, n-butanol and npentanol) at 70:30 ratio, was conducted. Using 2-propanol, poor resolution and relative long analysis time (ca. 50 min) were obtained. Using n-butanol and n-pentanol, analysis time and resolution improved significantly, although most peaks still overlapped. Despite this overlapping, ACN/n-pentanol mixture gave shorter analysis time with respect to nbutanol and therefore, n-pentanol was selected for the following studies.
Next, the influence of the content of n-pentanol in the 20-30% range on TAGs separation was studied. As shown in Figure S1 (see Supporting Information), resolution was improved with decreasing of n-pentanol content. Broad peaks with separation times higher than 45 min were obtained with less than 20% n-pentanol. Thus, an 80:20 (v/v) ACN/n-pentanol mixture was selected. This mobile phase also provided a satisfactory response in ELSD. Then, the effect of column temperature on TAGs separation and detection response of UV ( Figure S2, left) and ELSD ( Figure S2, right) detectors was performed under isocratic conditions. As it can be seen, the TAG resolution increased Next, the influence of flow rate on TAGs separation was also studied. A flow rate decrease from 1.5 mL min -1 to 1.0 mL min -1 led to an improvement in the resolution at expense of an increase of analysis time. Lower flow rates (< 1.0 mL min -1 ) provided a significant peak broadening (peak widths ranged between 10.2 and 80.3 s and asymmetry factors comprised between 0.85-1.60) at very long separation time (> 100 min). In addition, a decrease of the ELSD response with increasing flow rate was found, which was consistent with findings reported by several authors. 40,42 As a result, a flow rate of 1.0 mL min -1 was selected for further studies. Under these conditions, the late-eluting compounds were barely distinguished from baseline. In order to decrease the retention time and improve detection (peak shape), a solvent gradient step was established. Thus, the n-pentanol content was increased from 20% to 40% in 20 min after the first 45 min of isocratic elution with 80:20 ACN/n-pentanol. Figure 1 shows the ELSD chromatogram obtained under these conditions, where satisfactory peak shapes for highly retained TAGs were obtained.
The performance of the developed RPLC method was compared to those of previously reported for analysis of TAGs in human milk. Most of these studies were usually conducted using conventional microparticulate columns packed with 5-µm silica particles, 23,27,28 however, it should be mentioned that data of efficiency or other analytical parameters were not provided. In any case, an estimation of the number of resolved peaks of some of these works was done. These values were comprised between 17 and 22, 23,27,28 which were significantly lower than those obtained in this work (35). It should be noted that in spite of the satisfactory resolution achieved in this work, some cases of co-elution of TAGs were observed (see Table 1 and  Table 3, the peak areas were well-fitted by power model equations in the mass range studied (0.5-100 µg). The limits of detection (LOD) and limits of quantification (LOQ) were also estimated for signal-to-noise ratios of 3 and 10, respectively. The results obtained for homogeneous TAG standards were: LOD (9.2-13.1) ng and LOQ (30.6-43.3) ng (see Table 3). Then, the RRF calculated for pure homogeneous TAG standards in relation to OOO were obtained, giving values close to unity, which allows a quantitative estimation of TAGs on the basis of the percentage peak area. The relative content of each TAG (expressed as mean ± SD) obtained is given in Table 4. As shown in Table 4, no significant differences (p > 0.05, Student's t-test) were observed between results obtained with both methods for more than 85% of the TAGs. From this study, it can be derived that Method I provided satisfactory results, both in the total fat content as in the quantitative analysis of representative TAGs. Moreover, the Method I has several advantages over the Method II. Table 5  As shown in Table 4, from 69 TAG molecular species identified, the six major TAGs found in lipid fraction in both protocols were: POO, POL, LaPO + MMO, MOO, PSO (see Table 1 for abbreviations) with 20, 10, 7, 6 and 5%, respectively. is also important to remark that the molecular weight distribution of TAGs (expressed as CN:ND groups 44 ) in human milk reported in this work (see Figure S3) is in agreement with the findings of studies previously reported.

44,49
Taking into account the advantages described above for the Method I, it was extended to the other milk mammalian species (cow, sheep and goat). Thus, the total fat content varied within the ranges 0.023-0.046, 0.046-0.057 and 0.034-0.076 g mL -1 for cow, sheep and goat milks, respectively. These milk samples were also analyzed using the developed LC method with an excellent resolution/elution time ratio compared to those previously reported 14,16,23,27,28 ( Figure 2). Similar TAG profiles (particularly for the main TAG peaks) were found than those reported in literature for these samples. 14 From these profiles, 22 common peaks, which could be easily integrated, were selected for the four mammalian species, and used for statistical treatment.

Classification of Mammalian Milks Using TAG Profiles with LDA Model.
First, to reduce the variability associated to total amount of TAGs recovered from milk samples, and to minimize the sources of variance also affecting the sum of the areas of all the peaks, normalized rather than absolute peak area were used. In order to normalize the variables, the area of each peak was divided by each one of the areas of the other 21 peaks; in this way, and taking into account that each pair of peaks should be consider only  Table 6. As shown in Figure   3a, a large resolution between the human milk from the other mammalian milks was achieved along the first discriminant function (df). As deduced from  Figure 3b, along the third df, it is not possible to distinguish between sheep and human milk, but both were markedly differentiated from cow and goat milks. As shown in Table 6, the third df was mainly constituted with the peak area ratios     goat (c). Chromatographic conditions as in Figure 1.     a For abbreviations see Table 1.
b RRF values are given in relation to OOO. b Peak identification number, TAG information and abbreviations as indicated in Table 1.