The detection of goat milk adulteration with cow milk using a combination of voltammetric fingerprints and chemometrics analysis

In this study, a novel analytical approach was developed for detecting and predicting adulteration of goat milk with cow milk using a combination of voltammetric fingerprints and chemometrics analysis. The fresh milk samples were obtained from local farmers and analyzed using cyclic voltammetry technique using a glassy carbon electrode as the working electrode and KClO4 as the supporting electrolyte. The voltammetric fingerprint was obtained from both milk samples and showed an anodic peak between a potential range of 0.40–0.75 V versus Ag/AgCl. This anodic peak is mainly attributed to several electroactive species contained in both milk samples. The current intensities at the potential range of 0 to + 1 V versus Ag/AgCl were further selected due to the majority of electroactive components in the milk samples having their oxidation potential in this potential range. The current intensities were further pre-treated using maximum normalization and submitted to the chemometric tools for multivariate analysis. Orthogonal partial least square-discriminant analysis provided clear discrimination between goat and cow milk. Meanwhile, the prediction of goat milk adulteration with cow milk was achieved using partial least squares regression analysis. This multivariate analysis enabled a satisfactory discrimination and successful model to predict the percentage of cow milk as adulterants in goat milk samples. The demonstrated results revealed that a combination of voltammetric fingerprints and chemometrics tools might offer a low-cost, simple, and rapid analysis which might be possible as a promising method to be developed further for the detection of adulterants.


Introduction
Milk is a complex liquid containing macronutrients, including water, lipids, lactose, and protein, with several microconstituents, such as minerals, vitamins, and enzymes (Goulding et al. 2020). Milk is obtained from the secretion of healthy ruminant species, for example, cows and goats, which play a significant role due to their nutrition and health protection for human lives. Recently, goat milk has been receiving much attention due to its specific composition and being considered a high-quality raw material, especially for people with specific needs, such as infants and elderly people (Park 2017). Therefore, goat milk is considered more relevant for the human diet due to its unique characteristics such as high digestibility, distinct alkalinity, greater buffering capacity, medium-chain fatty acids, high content of Fe, Zn, Mg, Ca, stronger antimicrobial activity, better immunological and antibacterial activity, and high levels of amino acids such as valine, glycine, and histidine (Ceballos et al. 2009). In addition, goat milk has been a functional food for those suffering from various allergies who need better nutrition absorption and a high buffering capacity for ulcer treatment (Fangmeier et al. 2019).
It has been studied previously that the low level of α s1casein protein might be a key reason for goat milk forming a softer curd, thus leading to higher digestibility than cow milk (Saikia et al. 2022). Furthermore, the content of certain nutrients in goat milk (e.g., cis polyunsaturated fatty acids (PUFA), eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), isoflavones, and other micronutrients) was also reported to be higher than cow milk (Stergiadis et al. 2019). Hence, goat milk can be an ideal substitute for cow milk and may lead to a high demand for people to consume more goat milk. Due to the high demand and value of goat milk, adulteration with cow milk might be a serious issue that reduces the quality of the product for more profit and milk consumption.
Up to now, several types of adulterants have been added to the goat for unethical economic profit, i.e., soy milk, urea, or cow milk, while the latter is the major and most difficult content to recognize in its mixture (Li et al. 2017). The adulteration of goat milk with cow milk has become a common practice due to its greater abundance and lowered price than goat milk (dos Santos Pereira et al. 2021). This type of adulteration would generate serious problems for consumer health, especially for those who have lactose intolerants and cow milk allergies, leading to several allergenic disorders (Haenlein 2004). Cow milk contains more than 20 allergenic proteins with the most allergenic effects caused by α s1casein and β-lactoglobulin protein (Claeys et al. 2013). In addition, direct identification of goat milk adulteration with cow milk is challenging since cow milk physically resembles to goat milk (Coitinho et al. 2017). Therefore, the detection of goat milk adulterated with cow milk in dairy products is imperative to ensure food safety and consumer confidence.
When electrochemical techniques are combined with appropriate multivariate statistical methods, it could be an intriguing method to be employed for detecting and quantifying adulterations in dairy products, especially in milk (Tazi et al. 2018;Nikolaou et al. 2020;Minetto et al. 2022). One technique in chemometrics analysis that has been intensively used in identifying food adulterants in food products is a partial least squares-discriminant analysis and orthogonal partial least square-discriminant analysis (PLS-DA and OPLS-DA) (Shi et al. 2018;Texeira et al. 2021). This type of chemometric analysis could be combined the voltammetric techniques to investigate and discriminates the adulteration in food products (Tsopelas et al. 2018;Perez-Rafols et al. 2022;Azcarate et al. 2022). Furthermore, until now, this type of analysis has been extensively utilized to form unique electrochemical fingerprints by investigating the redox processes of its active components through voltammetry scanning in several complex systems such as plants (Fu et al. 2018;Lu et al 2022a, b;Gandhi and Amreen 2022;Yu et al. 2022;Melucci et al. 2022;Xiao et al. 2023), fruits (Cen et al. 2022), opiate drugs (Ortiz et al. 2022), minerals (Stojanov et al. 2022), petrol (Suppajariyawat et al. 2022), and red wine (Thomas et al. 2022). However, to the best of our knowledge, the utilization of combined voltammetric fingerprints with chemometric analysis, in this case, OPLS-DA for the detection of goat milk adulterants with cow milk has not yet been reported.
In this study, a quantitative chemometric method was developed based on combining voltammetric fingerprinting with chemometrics for detecting cow milk as an adulterant in goat milk. The goat milk was obtained from the Etawa goat, one type of livestock widely bred in Indonesia. Meanwhile, a cyclic voltammetry technique was conducted to obtain voltammetric fingerprints using a glassy carbon electrode in the sample of Etawa goat milk. In addition, OPLS-DA was employed to distinguish the sample of goat milk from cow milk as adulterants, while partial least squares regression analysis (PLSR) was used to predict the percentage of adulteration in goat milk.

Chemicals
All chemicals were used in analytical grade and supplied commercially without further purification. Potassium perchlorate (99.99%, trace metal basis) was obtained from Sigma-Aldrich (Darmstadt, Germany), and deionized water was used in the whole experiments.

Instrumentation
All electrochemical measurements were performed using PalmSens Emstat 3 (ES316U669) connected to the personal computer via a USB port and equipped with PS Trace 5.9 software. Glassy carbon electrode (GCE, 3 mm diameter), Ag/AgCl electrode, and platinum wire were used as working, reference, and auxiliary electrodes, respectively. Before the measurements were carried out on each milk sample and in-between measurements, the working electrode was polished with alumina slurry (0.3 μm) and rinsed with deionized water. Voltammogram was obtained from each measurement with a potential window from −1000 to + 1000 mV versus Ag/AgCl, a scan rate of 20 mV s −1 , and a potential step of 0.05 mV. All electrochemical measurements were carried out at 28 °C.
The chemical content of goat and cow's milk, such as fatty acids, was analyzed using capillary gas chromatography equipped with a flame ionization detector (FID), using a Supelco SPTM -2560 column purchased from Sigma-Aldrich (Germany). High-performance liquid chromatography (HPLC) with a refractive index detector (RI) was used for lactose analysis, while HPLC-PDA was used for vitamin A analysis. Thermo Radial iCAP 6000 Series ICP-OES (Thermo Scientific, Waltham, USA) was used for calcium analysis, and all amino acids were analyzed using a Waters ACQUITY Ultra High-Performance LC 143 system (Waters, Milford, MA, USA) equipped with a photodiode array detector, except for a few amino acids, such as tryptophan, cysteine, and methionine. An HPLC system (Waters, Milford, MA, USA) equipped with a photodiode array detector was used for tryptophan analysis. In contrast, a Nexera HPLC system (Shimadzu Corporation, Kyoto, Japan) connected to an LC-MS-8060 (Shimadzu) was used to determine cystine and methionine.

Sample preparation
Six samples of two types of fresh milk (3 samples of goat milk and 3 samples of cow milk) were obtained from several farmers in Bogor region, West Java, Indonesia (i.e., Ciawi, Kebon Pedes, Tajur Halang, and Tanah Baru). The milk sample was obtained in the early morning. These samples matched nearly to almost all commercially available goat and cow milk in the Indonesian market. The farmers also guaranteed the authenticity of all samples. All samples were stored in the refrigerator (4 o C) until the measurements were performed one day after the milk was taken from the cattle.
For quantification of goat milk adulteration, cow milk was considered as the adulterant milk. The adulterated samples of goat milk were prepared by mixing goat milk with cow milk in binary mixtures with concentrations of adulterant (cow milk) ranging from 0 to 100%, as described in Table 1.
All samples were prepared in a total volume of 4 mL, and each series was placed in a glass vial with a clear label.

The composition analysis of goat milk and cow milk
The goat and cow milk analyzed its several contents, i.e., saturated fat, protein content, water content, lactose, vitamin A, and calcium to obtain information about the differences in compositions between the two samples. The composition of fatty acids was determined using a capillary gas chromatographic method following ISO 16958:2015 (ISO 2015) with a Supelco SP™ -2560 column, fixed injector temperature at 255 º C, nitrogen gas carrier, and FID detector. Each of protein and water content was determined using a titrimetric dan gravimetric method based on Indonesian National Standard (SNI 01-2891-1992) (National Standard Agency 1992). Lactose was determined according to the AOAC 980.13.2005 method (AOAC 2005a) by a high-performance liquid chromatography (HPLC) using a carbohydrate column, 80% acetonitrile as mobile phase, and refractive index detector. Vitamin A was determined by HPLC according to AOAC 2001.13.2011 method (AOAC 2011a) using RP-18 column, gradient composition of aquabidest and methanol as mobile phase with PDA detector. Calcium was analyzed using inductively coupled plasma-optical emission spectrometry (ICP-OES) following AOAC 2011a, b.14 method (AOAC 2011b). Amino acids except tryptophan, cystine, and methionine were quantified using an ultra-performance liquid chromatography-photodiode array (UPLC-PDA) following Waters (2012) using AccQ.Tag Ultra C18 column at 49 ºC and aquabidest as mobile phase. Tryptophan was determined according to AOAC 988.15.2005 method (AOAC 2005b) using a HPLC-PDA with a RP-18 column with mixture of sodium acetate and methanol as the mobile phase in isocratic elution. Cystine and methionine were determined using a liquid chromatography-tandem mass spectrometry  1 3 with 0.1% formic acid in acetonitrile and 100 mM ammonium formate as mobile phase.

Voltammetric analysis of the milk sample
Analysis of voltammetric fingerprints was carried out by mixing 4 mL of milk sample with 4 mL of electrolyte solution (0.1 M KClO 4 in deionized water) and stirring until obtaining a homogenous solution. Then, the sample was transferred to the electrochemical cell and all electrodes were connected to the Emstat3 potentiostat (ES316U669) to perform voltammetric measurements. Each sample was measured with the voltammetric technique in four replicates in random order.

Statistical analysis
Multivariate analysis, including unsupervised principal component analysis (PCA) and supervised orthogonal partial least square-discriminant analysis (OPLS-DA), was performed using Metaboanalyst 5.0 (http:// www. metab oanal yst. ca/) for classification and discrimination of milk samples. Then, partial least square regression (PLSR) analysis was performed using The Unscrambler X 10.4 software (CAMO, Trondheim, Norwegia) to predict the percentage of goat milk adulteration with cow milk. Before multivariate analysis was carried out, maximum normalization as data pre-treatment was applied to all data sets of cyclic voltammograms.

Voltammetric fingerprints of goat milk and cow milk
In this work, the cyclic voltammetry (CV) technique was employed directly on the fresh milk samples and with a pretreatment process using ethanol and centrifugation prior to CV measurement. The pre-treatment process does not show any significant difference between goat and cow milk in terms of its voltammetric fingerprints. Thus, for further measurement, the CV was directly applied to the fresh milk without pre-treatment. However, this is different from the results reported by Nikolaou et al. (2020), who obtained distinct voltammetric fingerprints through a pre-treatment process followed by centrifugation of the fresh and reconstituted milk samples before voltammetric analysis was carried out.
Cyclic voltammetry was carried out to obtain information on electroactive compounds in the fresh goat and cow milk samples. This technique was chosen to investigate all electrochemically active substances in milk due to its better reproducibility and more discrimination power than the differential pulse voltammetry (DPV) technique (Rosello et al. 2020). During potential scanning in the CV technique, we could obtain information about oxidation (forward scan) and reduction (backward scan) processes from all electroactive species contained in the fresh milk sample. The potential window between − 1.0 and + 1.0 V vs Ag/AgCl was chosen for CV investigations since at potential either below − 1.0 V or higher than 1.0 V, the current intensity dramatically increased without any discernible peak. Furthermore, the limitation of potential above 1.0 V due to the water oxidation process and lower than − 1.0 V caused by H + reduction makes other redox reaction processes harder to see. In addition, it is also expected that a greater quantity of electrochemical information from the milk sample can be obtained from this potential window, thus increasing the possibility of accurate discrimination provided by chemometrics analysis.
As shown in Fig. 1, representative voltammograms obtained from the fresh sample of goat and cow milk are almost similar, with a slight difference between the potential range of 0.4-0.75 V versus Ag/AgCl. This similarity could be attributed to proteins in milk, such as globulins, albumins, and caseins which might be bound to the electroactive species and adsorbed onto the electrode surface (Stebler et al. 1990). From the obtained voltammogram, as shown in Fig. 1, there is a small anodic peak at 0.40-0.75 V versus Ag/AgCl for cow milk. However, it is not feasible to link these peaks to individual molecules due to the complexity of the milk matrix, as shown in Table 2, which contains numerous concurrent oxidation-reduction processes (Noyhouzer et al. 2009). In addition, minerals such as calcium, / sugars (sucrose), and vitamins are also among the compound groups that may be involved in milk discrimination due to Fig. 1 Representative cyclic voltammogram of goat milk and cow milk measured in the potential range of − 1 V to + 1 V versus Ag/ AgCl their electroactivity, as depicted in Table 2 (Bougrini et al. 2014;Lovander et al. 2018;Gursoy et al. 2020).
It has been previously reported that there are five main electroactive amino acids, including histidine, methionine, tyrosine, tryptophan, and cysteine, in fresh milk (Xu et al. 2005;Palecek et al. 2015). All these electroactive amino acids were detected in the fresh sample of goat and cow milk as shown in Table 3. It is typically believed that amino acids will initially bind to the electrode surface through their carboxyl group, allowing electron transfer between the electrode and the electroactive part of amino acids. The side chains of amino acids commonly contain these electroactive amino acid components. For instance, the electroactivity of tryptophan, tyrosine, cysteine, histidine, and methionine are attributed to indole-, phenol-, thiol-, imidazole-, and sulfur-containing groups, respectively, which are found in the side chain of these amino acids (Moulaee and Neri 2021). Therefore, this work aims not to identify the electroactive species responsible for the obtained voltammogram but to evaluate the cumulative contribution of each component that can differentiate between cow milk and goat milk with the assistance of chemometrics analysis as will be explained in the next section.

Multivariate analysis for discrimination of goat milk with cow milk
As previously discussed, discrimination of goat and cow milk using cyclic voltammetry technique can be utilized as a basis for adulteration determination and with aid from multivariate analysis to interpret the obtained data analysis. This work used principal component analysis (PCA) and orthogonal partial least square-discriminant analysis (OPLS-DA) as multivariate analysis to differentiate between goat milk and cow milk based on its voltammetric fingerprints. The obtained voltammetric fingerprints from each fresh milk were then exported into a data matrix according to its current intensity derived from the different applied potential. Finally, multivariate analysis, i.e., PCA and OPLS-DA was further applied to the dataset from each voltammogram to check the classification and combination of goat and cow milk.
To improve the obtained discriminant model from multivariate analysis, we utilized several data pre-processing and pre-treatment or combination of both processes, as shown in Table 4. All these processes were applied to three raw datasets of voltammetric fingerprints (complete data on cyclic voltammogram, difference of current intensities between forward and backward scan, and current at the positive potential) before the multivariate analysis was carried out. In this work, we evaluated several pre-processing and pre-treatment methods that are commonly employed in voltammetric data, such as auto-scaling (Tsopelas et al. 2018;Rosello et al. 2020), mean centering (Baldo et al. 2019), a combination between mean-centering with auto-scaling (Oliveri et al. 2009;Bueno et al. 2014;Nikolaou et al. 2020), a combination of first-order derivative (15 points in a filter, 2nd order polynomial) with mean centering (Pigani et al. 2009), and maximum normalization (Table 4). Thus, the selected pre-processing and pre-treatment method, which provides the best sample discrimination, was maximum normalization. The selection was based on the highest PCA plot scores, coefficient of determination (R 2 ), and coefficient of correlation (Q 2 ) for the OPLS-DA model.
In this work, initially, we have employed the current difference between forward and backward scans in the potential range from − 1 to + 1 V and combined them with PCA and OPLS-DA as chemometric tools to differentiate goat milk and cow milk. However, we did not obtain good discrimination between goat and cow milk using these two techniques due to a lot of overlapping data. Therefore, we have limited the data of current intensities from the potential range of 0 to + 1 V and back to 0 V and obtained an excellent model to discriminate between goat milk and cow milk, as shown in Table 4. It is expected that several electroactive amino acids such as tryptophan, cysteine, histidine, and tyrosine contained in goat and cow milk (Motshakeri et al. 2022) could be oxidized and provide the data about their current densities that can be further processed by multivariate analysis such as PCA. The final data matrix to be processed by PCA analysis comprises 1148 data points consisting of 28 rows of milk (both goat and cow milk) and 41 columns of its current intensities obtained in the potential range of 0 to + 1 V. It extracted a model from this unsupervised analysis (PCA) whose first two principal components (PC) accounted for 97.8% of the variance (PC-1: 73.2% and PC-2: 24.6%). Figure 2a shows the score plot of the first 2 PCs provides discrimination of the projection of milk samples concerning their redox behavior. However, the resulting figure still does not show a clear grouping separation between the two classes of milk due to there being still overlapping data between them. Therefore, additional analysis is still needed to obtain a clear grouping separation between these two processed PCs to provide better visualization for projected samples for goat and cow milk. In order to obtain a better separation model between two PCs, the supervised OPLS-DA was employed by assuming its class previously defined and its resulting score plot is shown in Fig. 2b. As baseline parameters, the coefficient of determination (R 2 ) and the coefficient of correlation (Q 2 ) indicate the model's explanatory and predictive capacities, respectively, and their values should be higher than 0.5 (Dou et al. 2022). As R 2 and Q 2 approach 1, the results become increasingly accurate to the predicted model (Wang et al. 2019). The obtained R 2 X (cum), R 2 Y (cum), and Q 2 (cum) from the OPLS-DA model are 0.993; 0.933; and 0.881, respectively. These results indicated predictive effects and a satisfactory explanation for the classification and discrimination of goat and cow milk obtained from the OPLS-DA model. The OPLS-DA model was cross-validated using a 100 times permutation test (P < 0.01), as shown in Fig. 2c. It can be deduced that the developed model was not over-fitting (Worley and Powers 2013) because R 2 and Q 2 in the original OPLS-DA models are higher than those of the R 2 and Q 2 values from random permutation models, as shown in Fig. 2c. Therefore, these results demonstrate that goat milk and cow milk could be classified into two groups through supervised OPLS-DA analysis. In addition, the developed OPLS-DA models also show high goodness of fit and predictability.

Prediction of the adulteration percentage of goat milk with cow milk
Partial least square regression (PLSR) analysis was employed to predict and quantify the adulteration percentage of goat milk with cow milk. Before PLSR analysis was carried out, several pre-processing and pre-treatment processes or a combination of both processes were applied to three raw datasets of voltammetric fingerprints (complete data on cyclic voltammogram, difference of current intensities between forward and backward scan, and current intensities at the potential range of 0 to + 1 V) as described in Table 5. These pre-treatment processes were intended to enhance the performance of the generated PLSR model. According to Table 5, the selected pre-treatment process (maximum normalization) was successfully applied to the datasets obtained from current intensities at the potential range of 0 to + 1 V and back to 0 V. The selected parameters were based on the value of the coefficient of determination (R 2 ), root-meansquare error of calibration (RMSEC), and cross-validation (RMESCV). In this study, the PLSR model was generated from the training set consisting of 8 samples with 4 replicates, involving 8 adulteration levels including blank (0% adulteration level). The resulting PLSR model can predict the adulterant percentage of goat milk with cow milk, as shown in Fig. 3. This was revealed with the value of R 2 = 0.9773, RMSEC = 0.0462 strengthened with crossvalidation of the PLSR model as the value of R 2 = 0.9508, RMSECV = 0.0685, and bias model = 0 as shown in Table 6. A good model of PLSR can be deduced if the value of R 2 approaches 1 while the value of RMSEC, RMESCV, and the model bias is close to 0 (Zhao et al. 2018). In addition, Fig. 3 also shows a linear plot of reference (percentage of goat milk spiked with cow milk) versus predicted (percentage of cow milk concentration) was obtained up to 100% adulteration. This result shows that a combination of voltammetric fingerprints with PLSR could effectively predict the percentage of goat milk adulterants with cow milk.

Conclusions
A preliminary study was conducted to evaluate the applicability of the voltammetric technique with the chemometric method to predict cow milk adulterations in goat milk samples quantitatively. The voltammetric technique involves the direct measurements of milk samples by performing cyclic voltammetry technique using a glassy carbon electrode as a working electrode with KClO 4 as an electrolyte, collecting voltammetric fingerprints of the milk samples, data preprocessing and pre-treatment, and followed by data analysis by chemometric tools. Principal component analysis (PCA), orthogonal partial least squares-discriminant analysis (OPLS-DA), and partial least square regression were the chosen chemometric tools to investigate the adulteration of goat milk with cow milk. The results show that PCA analysis still could not discriminate between goat and cow milk. A good classification and discrimination of goat milk with cow milk were obtained using OPLS-DA analysis with the calculated value of R 2 X (cum) = 0.993, R 2 Y (cum) = 0.933, and Q 2 = 0.881, which shows good discrimination. Meanwhile, PLSR analysis is shown a satisfactory prediction model to Table 5 Optimization of pre-processing and pre-treatment of PLSR data sets These results were expressed in terms of coefficient of determination (R 2 ), root mean square error of calibration (RMSEC), root mean square error of cross-validation (RMSECV) for PLSR

Raw data
Pre-processing Pre-treatment Parameter  . 3 Correlation between the actual value and the predicted value of adulterers in goat milk, obtained from the PLSR model predict the percentage of cow milk adulteration in goat milk with the coefficient of determination (R 2 = 0.9773), rootmean-square error of calibration (RMSEC = 0.0462), and cross-validation (RMESCV = 0.0685), and bias model close to 0. Overall, the results of this preliminary study revealed that the combination of voltammetric technique with chemometric analysis could be a promising tool for quantifying cow milk adulteration in goat milk samples.