2.1 Establishment of fingerprint atlas
2.1.1 Establishment of shared maximum absorption intensity of goat milk
Fingerprints are based on the inherent quality characteristics of foods and can reflect the internal characteristics of the food detected by modern instruments such as spectroscopy and chromatography. They are integral, systematic, characteristic, and stable (Fang et al. 2006).
As a fast and non-destructive detection technique, NIRS is often used for qualitative and quantitative analysis of chemical components, with a wavelength range of 780 ~ 2526nm. Frequency multiplication and combined frequency absorption of H-containing groups (CH, OH, NH, and SH) take charge to generate the region. It has the advantages of simplicity, speed, and non-destructiveness, but exists poor anti-interference, low sensitivity, and requires a large number of samples. During the experiments, these methods can be combined to take advantage of their strengths and circumvent their weaknesses to obtain more accurate experimental data (Chavan et al. 2017).
Figure 1 presented the spectra of goat milk from 11 different batches. The absorption bands of goat milk NIR spectra resembled those found by Andueza et al. (2013). The 6 maximum absorption intensities identified as the main common absorption maximum in the figure were shared by 11 batches of samples. The corresponding NIR wave values of the common maximum absorption intensity were 10270.95, 8612.31, 8334.77, 6914.02, 5539.54, and 5202.52 cm− 1, which meant that there were 6 groups of similar major substances in these 11 different batches.
Observing the obtained spectra (Fig. 1), the peaks at 8612.31, 8334.77 cm− 1 are from the C-H stretch and the second overtone, which may be related to fat and fatty acids in goat milk (Temizkan et al. 2020; Nunez-Sanchez et al. 2016); whereas that at 5202.52 cm− 1 is associated with the stretching and bending of the O-H groups of water in the samples (Vasafi et al. 2021). Also, the combination of bending of the O-H group and stretching of the C-O group in lactose leads to the emergence of wave peak 5539.54 cm− 1 (Mohamed et al. 2021).
2.1.2 Establishment of shared maximum absorption intensity of infant goat milk formulas
The spectra of 11 batches of samples from three types of Stage 1, 2, and 3 infants goat milk formulas of Y company showed the 9 maximum absorption intensities shared by these samples and were the main common maximum absorption intensities (Fig. 2). The corresponding NIR wave values of the common maximum absorption intensity were 8262.08, 6795.08, 6352.36, 5790.64, 5684.91, 5162.87, 4753.17, 4336.86, and 4257.56 cm − 1, which indicated that there were 9 groups of samples with similar main substances in these 11 different batches.
Analyzing these received spectra (Fig. 2), secondary overtones of C-H stretching vibrations in saturated fat structures cause the band at 8262.08 cm− 1 (Huang et al. 2016). The first overtone of the NH stretching and the first overtone of the OH stretching result in peaks of 6795.08 and 6352.36 cm− 1, respectively (Henn et al. 2017). The bands at 5790.64, 5684.91 cm− 1 could be attributed to the first overtone of the CH, CH2, and CH3 vibration while the band at 5612.87 cm− 1 is related to the secondary overtones of OH stretching and CH stretching (Kang et al. 2006; Ferreira et al. 2014). The band at 4753.17 cm− 1 is connected to the N − H symmetric stretching vibration (Wu et al. 2009). The combination of CH stretching and CH2 deformation determines the band at 4336.86 cm− 1. The band at 4257.56 cm− 1 depends on CH2 stretching and =CH2 bending of the C༝C alkene group (Hourant et al. 2000).
Goat milk, an opaque medium, has multiple light dispersion effects because of opaque. Diffuse reflection adopted in the experiment minimizes signal interference from surface reflections, thus obtaining sufficient spectral information (Melfsen et al. 2012). In addition, although NIR is widely used in adulteration identification of bovine milk, goat milk, camel milk, or their milk products, the corresponding wavelength range set by NIR is short, and the sample common peaks measured are few. Kasemsumran et al. (2007) used NIR to quantify the feasibility of milk adulteration. The spectral range opted was 1100 to 2500 nm, and the obtained samples had two common peaks. By contrast, the NIRS of the experiment has a wide scanning range and more common peaks are obtained. Therefore, the identification of more organic groups can improve the accuracy of subsequent identification of adulteration.
2.1.3 Fingerprint methodology verification results
It can be seen from Table 1 that the absorbance value corresponding to the characteristic common maximum absorption intensity in goat milk was calculated, the range of relative standard deviation (RSD) of the stability test was 0.11%~0.55%; the range of RSD of the precision test was 0.46%~0.94%; the RSD of the reproducibility of the test ranges from 2.95–4.62%; the RSD values of both precision and reproducibility were less than 5%, the stability RSD value is less than 1%, which proved that the reproducibility, stability, and precision of this NIR fingerprinting method in goat milk were validated.
Table 1
Method validation of near-infrared spectroscopy fingerprint of goat milk
Wavenumber (cm− 1) | Reproducibility | Stability | Precision |
Mean value | RSDα/% | Mean value | RSDα/% | Mean value | RSDα/% |
10270.95 | 0.47 | 3.49 | 0.51 | 0.45 | 0.50 | 0.46 |
8612.31 | 0.64 | 4.46 | 0.69 | 0.55 | 0.68 | 0.63 |
8334.77 | 0.67 | 4.20 | 0.72 | 0.42 | 0.71 | 0.54 |
6914.02 | 1.70 | 4.48 | 1.75 | 0.11 | 1.77 | 0.47 |
5539.54 | 1.28 | 4.62 | 1.45 | 0.52 | 1.43 | 0.94 |
5202.52 | 1.90 | 2.95 | 1.91 | 0.20 | 1.95 | 0.50 |
α RSD, relative standard deviation |
The results of the validation experiments of the NIRS of infant goat milk formulas were shown in Table 2. For the reproducibility test, RSDs of the absorbances of the nine features ranged from 0.33–1.43%. For the stability test, the RSD ranged from 0.39–2.71%. The RSD for the precision test ranged from 0.19–1.11%. All of the RSD values were below 5.0%. Therefore, the reproducibility, stability, and precision of this NIR fingerprinting method in infant goat milk formulas were validated.
Table 2
Method validation of near-infrared spectroscopy fingerprint of infant goat milk formulas
Wavenumber (cm− 1) | Reproducibility | Stability | Precision |
Mean value | RSDα/% | Mean value | RSDα/% | Mean value | RSDα/% |
8262.08 | 0.37 | 0.33 | 0.37 | 0.39 | 0.37 | 1.11 |
6795.08 | 0.60 | 0.69 | 0.63 | 1.79 | 0.61 | 0.34 |
6352.36 | 0.58 | 0.79 | 0.60 | 1.74 | 0.58 | 0.56 |
5790.64 | 0.66 | 0.85 | 0.67 | 0.96 | 0.66 | 0.53 |
5684.91 | 0.62 | 0.85 | 0.64 | 1.11 | 0.62 | 0.96 |
5162.87 | 0.68 | 0.93 | 0.74 | 1.49 | 0.68 | 0.49 |
4753.17 | 0.91 | 1.10 | 0.94 | 2.71 | 0.90 | 0.42 |
4336.86 | 1.07 | 1.43 | 1.10 | 1.56 | 1.07 | 0.48 |
4257.56 | 1.06 | 1.40 | 1.09 | 1.61 | 1.06 | 0.19 |
α RSD, relative standard deviation |
2.2 Detection Of Sample Adulteration Based On Fingerprint Atlas
PCA is a multivariate statistical analysis method that extracts a few comprehensive indicators from multiple indicators that have correlations. It can simplify multiple variables into a few comprehensive variables through orthogonal transformation, then explain the prominent relationship among the elements in the original variables. The process of extracting principal components is to filter out the principal component factors through matrix calculation after deriving the principal components (Cozzolino et al. 2019).
2.2.1 PCA of the difference in infant goat milk formulas at different stages
Figure 3 was obtained by combining the NIRS and PCA of infant goat milk formulas with three different stages. There existed two principal components (PC1 and PC2). Principal component 1 contained 76.7% of sample information, and principal component 2 contained 13.2% of sample information. PC1 and PC2 explained 89.9% of sample information together. The main location of PCA score points for Stage 1, 2, and 3 of infant goat milk formulas samples were respectively in the third quadrant, the first quadrant, and the fourth quadrant, which can be explained by the differences in their main nutritional ingredients (Table 3). The results of the PCA scores map proved that the use of NIRS technology to distinguish and identify infant goat milk formulas with three different objects was applicable.
Table 3
The main chemical composition of Stage 1, 2, and 3 infant goat milk formulas of Y brandα
Stage No. | Dry matter (g/100g, means ± SD, n = 11) |
Protein | Fat | Carbohydrate | Moisture | Ash |
Stage 1 | 13.10 ± 0.20c | 26.05 ± 0.25a | 55.03 ± 0.13a | 2.69 ± 0.11b | 3.13 ± 0.05c |
Stage 2 | 18.35 ± 0.10b | 20.93 ± 1.25c | 53.25 ± 1.27b | 2.92 ± 0.06a | 4.58 ± 0.15b |
Stage 3 | 19.06 ± 0.27a | 21.94 ± 0.51b | 51.62 ± 0.37c | 2.92 ± 0.18a | 4.94 ± 0.11a |
α Different superscript letters within a column denote statistically significant differences according to Duncan’s test (P < 0.05) |
2.2.2 PCA of goat milk blended with bovine milk
The PCA score chart of principal component regression analysis was shown in Fig. 4. Two principal components (PC1 and PC2) were in the figure. PC1 represented 74.8% of sample information and PC2 contained 21.6% of sample information. PC1 and PC2 jointly explained 96.3% of sample information together.
It was clear from Fig. 4 that goat milk samples were separated from adulterated samples on the PCA score chart, indicating that the coupling of PCA with the NIRS fingerprinting method can effectively identify adulterated goat milk from pure goat milk.
The main chemical composition of goat milk and bovine milk were shown in Table 4. Although there was no significant difference in protein content between goat milk and bovine milk (P > 0.05), the structure of the casein micelles, various biologically active peptides, and non-protein nitrogen compounds such as amino acids, nucleotides, and nucleosides are different between the two (Teixeira et al. 2022). Furthermore, the levels and proportion of αS1-casein, αS2-casein, and β-casein + κ-casein varied in goat milk and bovine milk (Ceballios et al. 2009). Samples with subtle differences can be differentiated using NIRS fingerprint technology combined with PCA according to their ingredients and the proportion relationships among ingredients because the main characteristics of fingerprint technology lie in its “integrity” and “fuzziness” (Chang et al. 2020). It was also easy to obtain that the fat content of goat milk and bovine milk was significantly different (P < 0.05) from Table 4. The difference in the structure and content of organic matter in two types of milk is the reason why PCA combined with the NIRS fingerprinting method can well distinguish goat milk and adulterated goat milk mixed with different proportions of bovine milk.
Table 4
Comparison of main chemical composition of between goat milk and bovine milkα
Type of milk | Dry matter (g/100g, means ± SD, n = 11) |
Protein | Fat | Lactose | Moisture | Ash |
Goat milk | 3.56 ± 0.04a | 4.07 ± 0.11a | 4.58 ± 0.01a | 87.01 ± 0.08b | 0.70 ± 0.01a |
Bovine milk | 3.04 ± 0.16a | 3.26 ± 0.10b | 4.41 ± 0.13a | 88.37 ± 0.18a | 0.71 ± 0.03a |
α Different superscript letters within a column denote statistically significant differences according to Duncan’s test (P < 0.05) |
2.2.3 Principal component analysis of desalted goat whey powder mixed with desalted bovine whey powder
The NIRS fingerprints of desalted goat whey powder mixed with desalted bovine whey powder in different proportions to apply the established NIRS fingerprinting method. The correlations of two principal components (PC1 and PC2) with desalted goat whey powder blended with different proportions of desalted bovine whey powder were shown in Fig. 5.
The cumulative variance contribution rate was 99.8%. The first component PC1 accounted for 99.5% of the total variance while the second component PC2 contained 0.3% of the total variance. It can be seen from Fig. 5 that the PCA score points of Y brand Stage 2 infant goat milk formulas samples were mainly located on the right half of the score map, while the PCA score points of adulterated milk powder samples were located on the left side of the score map. They were perfectly separated. The main component results showed that the use of NIRS technology can validly identify the adulteration of infant goat milk formulas blending desalted bovine whey powder.
It is known the main chemical composition in desalted goat whey powder and desalted bovine whey powder from Table 5. Among them, the protein content of desalted goat whey powder and desalted bovine whey powder was not significantly different (P > 0.05). However, Whey powder is composed of β-lactoglobulin (β-LG), α-lactoglobulin (α-LA), serum albumin (SA), lactoferrin, immunoglobulin, and other components (Kerasioti et al. 2019). Bovine whey powder is made up of 50–63% β-LG, 20% α-LA, 6–8% SA, and 1% immunoglobulin G (lg G) while goat whey powder consists of 40% β-LG, 30% α-LA, 10% SA, and 10% immunoglobulin GH (lg G-H) composition (Saxton et al. 2021; Zhao et al. 2020). Furthermore, the fat and lactose content of the two kinds of desalted whey powder was significantly different (P < 0.05). Differences in the content of various types of protein and other organic components lead to NIRS fingerprints can well identify adulteration of varying proportions of desalted bovine whey powder in infant goat milk powder.
Table 5
Comparison of main chemical composition of between desalted goat whey powder and desalted bovine whey powderα
Type of whey powder | Dry matter (g/100g, means ± SD, n = 3) |
Protein | Fat | Lactose | Moisture | Ash |
Desalted goat whey powder | 13.96 ± 0.15a | 0.77 ± 0.02b | 83.94 ± 0.02a | 1.51 ± 0.14b | 0.95 ± 0.05a |
Desalted bovine whey powder | 13.10 ± 0.17a | 1.30 ± 0.17a | 83.00 ± 0.01b | 2.00 ± 0.00a | 0.93 ± 0.12a |
α Different superscript letters within a column denote statistically significant differences according to Duncan’s test (P < 0.05) |