Advertisement

Mineral equilibrium in commercial curd and predictive ability of near-infrared spectroscopy

Open ArchivePublished:January 20, 2021DOI:https://doi.org/10.3168/jds.2020-18712

      ABSTRACT

      Curd samples (n = 83) from 3 European dairy companies were analyzed for micellar and soluble mineral fractions content using inductively coupled plasma optical emission spectrometry as a gold standard method. The same curd samples were analyzed through 3 different near-infrared (NIR) instruments, and NIR spectra were merged with reference data. Prediction equations were developed using modified partial least squares analysis, and the accuracy of prediction was evaluated through leave-one-out cross validation. Overall, NIR spectroscopy was capable of predicting micellar and soluble mineral fractions in curd, but with differences among instruments. Fitting statistics showed that the visible NIR instrument in reflectance mode outperformed the NIR instrument in transmittance mode as well as the portable NIR instrument in reflectance mode. Prediction accuracies for most of the analyzed mineral fractions can be used for curd quality control in dairy companies and to aid in decision-making during the cheesemaking process.

      Key words

      INTRODUCTION

      Curd is a product of enzymatic coagulation or acidic precipitation of milk constituents. Cheese curd represents an important intermediate product in cheese manufacturing, allowing easier conservation and transport compared with raw milk. Mineral equilibrium in milk plays an important role in curd manufactured through enzymatic coagulation because Ca and P are directly involved in curd formation through the stabilization of casein micelles (
      • Holt C.
      Casein and casein micelle structures, functions and diversity in 20 species.
      ). Together with milk Mg, milk Ca is also involved in paracasein reticulum formation, acting as a bridge between casein micelles (
      • Lucey J.A.
      • Fox P.F.
      Importance of calcium and phosphate in cheese manufacture: A review.
      ;
      • Malacarne M.
      • Franceschi P.
      • Formaggioni P.
      • Sandri S.
      • Mariani P.
      • Summer A.
      Influence of micellar calcium and phosphorus on rennet coagulation properties of cows milk.
      ). On the other hand, mineral equilibrium in curd can have an effect on the determination of curd quality, the assessment of technological traits used for curd formation, and the understanding of the behavior of the curd (and minerals) in the following processing steps.
      Although total mineral quantification in dairy products is a common procedure involving an acidic digestion step followed by direct quantification through different induced couple plasma protocols, the distinction between micellar and soluble mineral fractions is not negligible (
      • Lante A.
      • Lomolino G.
      • Cagnin M.
      • Spettoli P.
      Content and characterisation of minerals in milk and in Crescenza and Squacquerone Italian fresh cheeses by ICP-OES.
      ;
      • Reykdal O.
      • Rabieh S.
      • Steingrimsdottir L.
      • Gunnlaugsdottir H.
      Minerals and trace elements in Icelandic dairy products and meat.
      ). Recently,
      • Franzoi M.
      • Niero G.
      • Penasa M.
      • Cassandro M.
      • De Marchi M.
      Technical note: Development and validation of a new method for the quantification of soluble and micellar calcium, magnesium, and potassium in milk.
      proposed a new method for the quantification of the 2 mineral fractions in milk based on a double step dilution of enzymatically coagulated samples. The method allows for the direct quantification of micellar and soluble fractions without the need of correction factors that could introduce a bias in estimated mineral amounts. Nevertheless, the method is still demanding in terms of trained personnel, costs, and time.
      Near-infrared (NIR) spectroscopy has been recognized as a methodology to overcome such problems because it is fast and cost-effective for the characterization of routine samples; it is already used in many dairy companies for the routine characterization of samples and as an internal check of final products (
      • De Marchi M.
      • Toffanin V.
      • Cassandro M.
      • Penasa M.
      Invited review: Mid-infrared spectroscopy as phenotyping tool for milk traits.
      ). To our knowledge, there is limited information about curd mineral quantification using NIR spectroscopy, in particular as it regards mineral fractions. The aims of the present study were to (1) quantify soluble and micellar mineral fractions of commercial curds, (2) assess the relationships between mineral fraction content and chemical composition, and (3) investigate the ability of NIR spectroscopy to predict the content of soluble and micellar fractions.

      MATERIALS AND METHODS

      Sample Collection

      Eighty-three commercial curd samples intended for mozzarella cheese production were collected between January and May of 2019 from 3 European dairy companies. Thirty-five samples were imported frozen from European countries, and 48 samples were produced in Italy (16 using imported milk from European countries and 32 using Italian milk). The latter samples were immediately frozen and stored at −20°C until chemical analysis. Analyses were conducted in the laboratory of the Department of Agronomy, Food, Natural resources, Animals and Environment (DAFNAE, Legnaro, PD, Italy) of the University of Padova (Padova, Italy).
      Samples were thawed at 4°C for 48 h, homogenized with a knife mill (Grindmix GM200; Retsch GmbH, Haan, Germany), and analyzed for total protein, soluble protein, humidity, fat, and ash content using FoodScan Lab calibrated with Foss Artificial Neutral Networks Dairy Calibration (Foss, Electric A/S, Hiller⊘d, Denmark). Successively, curd samples were divided in 2 aliquots: the first aliquot intended for mineral content analysis and the second aliquot intended for NIR spectra collection.

      Mineral Analysis

      Ultrapure water was produced with Arium 611 UV (Sartorius, Monza Brianza, Italy), and all chemicals, if the supplier is not mentioned, were purchased from Sigma-Aldrich (St. Louis, MO) at the highest available purity. Total, micellar, and soluble minerals were analyzed following the method proposed by
      • Franzoi M.
      • Niero G.
      • Penasa M.
      • Cassandro M.
      • De Marchi M.
      Technical note: Development and validation of a new method for the quantification of soluble and micellar calcium, magnesium, and potassium in milk.
      with necessary modifications. After homogenization, 10 g of curd was immediately frozen for quantification of total minerals, and 20 g was introduced in a tube with 10 mL of ultrapure water. The tube was incubated for 2 h at 37°C in a water bath. Samples were centrifuged at 5,000 × g for 20 min at 25°C, and 5 mL of the supernatant was collected and filtered with a 0.45-μm membrane (d1). Then, 5 mL of ultrapure water was added to the tube, vortexed for 20 s, and left for 1 h at 37°C in a water bath to equilibrate the soluble phase. Samples were centrifuged as described previously, and 5 mL of supernatant was collected and filtered with a 0.45-μm membrane (d2). To determine how and to what extent the use of ultrapure water could affect the pH, we determined the pH of the soluble phases before and after dilution. Differences in pH for tested samples were always between 0.0 and +0.2 (data not shown). All obtained fractions were stored frozen until the quantification procedure, which was performed using inductively coupled plasma optical emission spectrometry (ICP-OES) as described by
      • Visentin G.
      • Penasa M.
      • Gottardo P.
      • Cassandro M.
      • De Marchi M.
      Predictive ability of mid-infrared spectroscopy for major mineral composition and coagulation traits of bovine milk by using the uninformative variable selection algorithm.
      and
      • Franzoi M.
      • Niero G.
      • Penasa M.
      • Cassandro M.
      • De Marchi M.
      Technical note: Development and validation of a new method for the quantification of soluble and micellar calcium, magnesium, and potassium in milk.
      . Briefly, samples were digested by nitric acid in a microwave Milestone Start D apparatus (Milestone Srl, Sorisole, Italy) and analyzed through ICP-OES Spectro Arcos (Spectro Analytical Instruments GmbH, Kleve, Germany) using wavelengths 317.933 nm for Ca, 285.213 nm for Mg, 766.941 nm for K, 177.495 nm for P, and 589.592 nm for Na. Calibrations of instrument were performed using single element solutions (Inorganic Ventures, Christiansburg, VA) in the range from 0 to 25 mg/L. The final amount of soluble minerals (Mw, g) was calculated as follows:
      Mw = (D × Cd12)/(Cd1 − Cd2),


      where D is the volume of supernatant collected for d1 (5 mL), Cd1 is the concentration of mineral in the fraction d1 determined by ICP-OES, and Cd2 is the concentration of mineral in the fraction d2 determined by ICP-OES. Soluble mineral concentration in curd was determined as the ratio of Mw to the weight of curd (wt/wt). Micellar concentration was calculated as difference between total and soluble mineral concentration.
      Preliminary analysis showed that results for all traits were normally distributed. Outliers were defined as values deviating more than 3 standard deviations from the mean, and no outliers were detected according to this procedure. Pearson correlations (r) between mineral fractions content and chemical composition were assessed using the CORR procedure of SAS software ver. 9.4 (SAS Institute Inc., Cary, NC).

      Near-Infrared Spectroscopy Calibration Models

      Near-infrared spectra were obtained using 2 laboratory instruments and 1 portable device in the range of visible and NIR spectra as follows: (1) NIR region—FoodScan Lab, which operates in transmittance, collecting spectral variables every 2 nm for wavelengths between 850 nm and 1,050 nm (Lab-NIR); (2) visible and near-infrared region—NIRS DS2500 (Foss, Electric A/S), which operates in reflectance, recording spectral response every 0.5 nm between 400 nm and 2,500 nm (Lab-VIS-NIR); (3) NIR region—Scio Portable Device (VeriFood LTD, Herzliya, Israel), which operates in reflectance, recording spectral response on each wavelength between 740 and 1,070 nm (portable NIR); for the latter instrument, 5 scans were averaged to obtain the final spectrum.
      Spectra were merged with reference values, and calibrations were developed using WinISI software (Infrasoft International, Port Matilda, PA) and modified partial least squares regression analysis combined with scatter correction (NONE = no correction; DET = detrending; SNV = standard normal variate; SNV+D = standard normal variate + detrending; MSC = multiplicative scatter correction). Moreover, different mathematical treatments were applied (0,0,1,1; 1,10,10,1; 1,4,4,1; 1,8,8,1; 2,5,5,1; and 2,10,10,1; where the first digit is the derivative, the second is the range of wavelength in which the derivative is calculated, the third is the number of wavelength used for the first smoothing, and the fourth is the number of wavelengths used for the second smoothing;
      • Manuelian C.L.
      • Currò S.
      • Visentin G.
      • Penasa M.
      • Cassandro M.
      • Dellea C.
      • Bernardi M.
      • De Marchi M.
      Technical note: At-line prediction of mineral composition of fresh cheeses using near-infrared technologies.
      ). Near-infrared spectra were excluded from prediction models when predicted data deviated more than 3 standard deviations from the mean of reference data. For each calibration model, the standard error in leave-one-out cross validation was calculated, and the mathematical treatment providing the best performance was selected. Fitting statistics were the standard error of calibration, the coefficient of determination in calibration, the coefficient of determination in leave-one-out cross validation, and the ratio of performance to deviation in leave-one-out cross validation (RPD).

      RESULTS

      Curd Composition

      Total protein, soluble protein, humidity, fat, and ash of curd samples averaged 22.82, 2.08, 48.15, 27.16, and 3.44%, respectively, and soluble protein and ash had the greatest coefficient of variation (Table 1). With regard to the curd mineral composition, the micellar fractions of Ca, Mg, and P were more abundant compared with their soluble fractions. Potassium was more present in the soluble phase, and Na was found only in the soluble phase. All minerals and their fractions had a coefficient of variation ≥0.15; in particular, total Na, soluble P, micellar K, and micellar Mg had coefficients of variation of 1.37, 0.48, 0.45, and 0.36, respectively (Table 1).
      Table 1Mean, SD, CV, minimum, and maximum of predicted chemical composition and minerals content (n = 83)
      TraitMeanSDCVMinimumMaximum
      Chemical composition (%)
       Total protein22.821.980.0918.3426.21
       Soluble protein2.080.500.241.393.29
       Humidity48.153.560.0741.9556.23
       Fat27.161.920.0722.6832.08
       Ash3.440.900.261.965.01
      Mineral fraction (mg/g)
       Ca total6.951.430.213.8311.15
       Ca soluble1.380.380.270.732.00
       Ca micellar5.571.720.312.4310.42
       K total0.880.130.150.461.17
       K soluble0.610.100.160.360.87
       K micellar0.270.120.45<0.010.63
       Mg total0.280.050.200.160.42
       Mg soluble0.120.020.150.080.17
       Mg micellar0.160.060.360.050.32
       Na total0.891.221.370.105.31
       P total4.760.820.172.967.25
       P soluble0.560.260.480.190.73
       P micellar4.201.030.242.417.06

      Prediction Models

      Fitting statistics of calibration models for mineral fractions in curd are summarized in Table 2 for Lab-NIR, Table 3 for Lab-VIS-NIR, and Table 4 for portable NIR. Prediction models developed using spectra collected with Lab-NIR had RPD values between 1.25 (soluble Mg) and 3.36 (total Na). Total K and its fractions had RPD lower than 2.00. The RPD values of total Ca and Mg were greater than their micellar and soluble fraction counterparts. Only micellar P could be predicted slightly better than total P (RPD of 2.88 and 2.70, respectively; Table 2). The Lab-VIS-NIR performed well for different minerals. Soluble P had RPD of 3.16, and micellar Ca and P showed RPD of 4.12 and 3.53, respectively (Table 3). Considering total mineral content, Na and Ca had the best RPD (4.28 and 3.42, respectively). Total K, its fractions, and soluble Mg were poorly predicted with RPD lower than 2.00. In general, for portable NIR, RPD values were lower than those obtained from the other 2 instruments (Table 4).
      Table 2Calibration and cross validation statistics of prediction models for minerals content for near-infrared instrument (FoodScan Lab; Foss, Electric A/S, Hillerød, Denmark)
      SEC = standard error of calibration; R2 = coefficient of determination in calibration; SEcv = standard error in cross validation; R2cv = coefficient of determination in cross validation; RPD = ratio of prediction to deviation in cross validation.
      Mineral fraction (mg/g)Model
      Scatter correction and statistical treatment used for spectra calibration; MSC (multiplicative scatter correction); SNV (standard normal variate); SND (standard normal variate + detrending). The first digit is the derivative, the second is the range of wave-length in which the derivative is calculated, the third is the number of wavelength used for the first smoothing, and the fourth is the number of wavelengths used for the second smoothing.
      No.SECR2SEcvR2cvRPD
      Ca totalSND1441760.340.920.370.913.29
      Ca solubleSNV2551800.160.830.190.752.02
      Ca micellarSND0011780.370.950.500.903.16
      K totalSND0011750.070.650.100.391.29
      K solubleMSC2551750.050.700.060.541.48
      K micellarSND0011730.060.660.070.601.59
      Mg totalMSC110101760.010.910.020.903.22
      Mg solubleSNV2551800.010.600.010.361.25
      Mg micellarMSC2551770.010.910.020.872.78
      Na totalSNV2551790.250.950.330.913.36
      P totalMSC210101760.220.890.250.862.70
      P solubleSND2551780.090.880.110.822.39
      P micellarSND210101760.270.910.300.882.88
      1 SEC = standard error of calibration; R2 = coefficient of determination in calibration; SEcv = standard error in cross validation; R2cv = coefficient of determination in cross validation; RPD = ratio of prediction to deviation in cross validation.
      2 Scatter correction and statistical treatment used for spectra calibration; MSC (multiplicative scatter correction); SNV (standard normal variate); SND (standard normal variate + detrending). The first digit is the derivative, the second is the range of wave-length in which the derivative is calculated, the third is the number of wavelength used for the first smoothing, and the fourth is the number of wavelengths used for the second smoothing.
      Table 3Calibration and cross validation statistics of prediction models for minerals content for visible and near-infrared instrument (NIRS DS2500; Foss, Electric A/S, Hillerød, Denmark )
      SEC = standard error of calibration; R2 = coefficient of determination in calibration; SEcv = standard error in cross validation; R2cv = coefficient of determination in cross validation; RPD = ratio of prediction to deviation in cross validation.
      Mineral fraction (mg/g)Model
      Scatter correction and statistical treatment used for spectra calibration; NONE (no scatter correction); DET (detrending); MSC (multiplicative scatter correction); SNV (standard normal variate); SND (standard normal variate + detrending). The first digit is the derivative, the second is the range of wave-length in which the derivative is calculated, the third is the number of wavelength used for the first smoothing, and the fourth is the number of wavelengths used for the second smoothing.
      No.SECR2SEcvR2cvRPD
      Ca totalMSC1441780.320.930.350.913.42
      Ca solubleSNV1441780.090.940.160.812.32
      Ca micellarSND1441770.320.950.360.944.12
      K totalSNV110101770.050.840.100.481.39
      K solubleSND2551750.030.920.060.591.58
      K micellarNONE1441730.050.700.070.411.31
      Mg totalNONE1441780.010.920.020.893.04
      Mg solubleSND1441800.010.760.010.461.37
      Mg micellarNONE1441770.010.910.020.882.94
      Na totalMSC110101790.160.980.260.944.28
      P totalDET0011770.210.900.230.872.81
      P solubleSND210101780.030.990.080.903.16
      P micellarSNV210101750.210.940.240.923.53
      1 SEC = standard error of calibration; R2 = coefficient of determination in calibration; SEcv = standard error in cross validation; R2cv = coefficient of determination in cross validation; RPD = ratio of prediction to deviation in cross validation.
      2 Scatter correction and statistical treatment used for spectra calibration; NONE (no scatter correction); DET (detrending); MSC (multiplicative scatter correction); SNV (standard normal variate); SND (standard normal variate + detrending). The first digit is the derivative, the second is the range of wave-length in which the derivative is calculated, the third is the number of wavelength used for the first smoothing, and the fourth is the number of wavelengths used for the second smoothing.
      Table 4Calibration and cross validation statistics of prediction models for minerals content for near-infrared portable instrument (Scio Portable Device; VeriFood Ltd., Herzliya, Israel)
      SEC = standard error of calibration; R2 = coefficient of determination in calibration; SEcv = standard error in cross validation; R2cv = coefficient of determination in cross validation; RPD = ratio of prediction to deviation in cross validation.
      Mineral fraction (mg/g)Model
      Scatter correction and statistical treatment used for spectra calibration; NONE (no scatter correction); DET (detrending); MSC (multiplicative scatter correction); SNV (standard normal variate); SND (standard normal variate + detrending). The first digit is the derivative, the second is the range of wave-length in which the derivative is calculated, the third is the number of wavelength used for the first smoothing, and the fourth is the number of wavelengths used for the second smoothing.
      No.SECR2SEcvR2cvRPD
      Ca totalDET1881760.380.900.510.822.37
      Ca solubleNONE1441820.170.800.220.671.74
      Ca micellarDET1881760.390.930.550.862.73
      K totalDET2551740.090.360.110.131.08
      K solubleMSC210101720.050.710.060.461.37
      K micellarNONE0011740.080.550.090.471.39
      Mg totalDET110101780.020.880.020.802.25
      Mg solubleMSC1441800.010.690.010.461.38
      Mg micellarDET0011790.020.890.020.812.31
      Na totalSND1881760.390.880.490.802.26
      P totalSNV2551750.270.810.300.762.04
      P solubleNONE1441800.080.910.110.822.40
      P micellarDET110101760.370.830.410.792.19
      1 SEC = standard error of calibration; R2 = coefficient of determination in calibration; SEcv = standard error in cross validation; R2cv = coefficient of determination in cross validation; RPD = ratio of prediction to deviation in cross validation.
      2 Scatter correction and statistical treatment used for spectra calibration; NONE (no scatter correction); DET (detrending); MSC (multiplicative scatter correction); SNV (standard normal variate); SND (standard normal variate + detrending). The first digit is the derivative, the second is the range of wave-length in which the derivative is calculated, the third is the number of wavelength used for the first smoothing, and the fourth is the number of wavelengths used for the second smoothing.

      Pearson Correlations

      Total protein was positively associated with micellar fractions, having correlation from 0.43 with micellar K to 0.82 with micellar Ca (P < 0.001), and negatively associated with soluble Ca and P (r = −0.71; P < 0.001) and soluble K (r = −0.35; P < 0.01) (Table 5). Overall, soluble protein was less associated with minerals compared with total protein; in particular, soluble protein was moderately correlated with soluble P (r = −0.49: P < 0.001) and total Na (r = 0.33; P < 0.01), and weakly associated with other mineral fractions. Correlations of curd humidity with micellar fractions were moderately negative (from −0.67 to −0.40; P < 0.01), whereas correlations with soluble Ca (P < 0.001), K (P < 0.05), and P (P < 0.001) were positive (0.24 to 0.52; Table 5). Moderate to strong associations (r = 0.42 to 0.86, in absolute value; P < 0.01) were estimated between ash and micellar and soluble mineral fractions (except for soluble Mg; P > 0.05). Finally, the correlations between fat and mineral content, total K and curd composition, and soluble Mg and curd composition were weak and not statistically significant (P > 0.05; Table 5).
      Table 5Pearson correlations between curd chemical composition and major mineral fractions
      Mineral fraction (mg/g)Total protein (%)Soluble protein (%)Humidity (%)Fat (%)Ash (%)
      Ca total0.79
      P < 0.001.
      0.22
      P < 0.05
      −0.65
      P < 0.001.
      0.020.85
      P < 0.001.
      Ca soluble−0.71
      P < 0.001.
      −0.22
      P < 0.05
      0.45
      P < 0.001.
      0.11−0.71
      P < 0.001.
      Ca micellar0.82
      P < 0.001.
      0.24
      P < 0.05
      −0.64
      P < 0.001.
      −0.010.86
      P < 0.001.
      K total0.14−0.18−0.190.090.13
      K soluble−0.35
      P < 0.01
      −0.040.24
      P < 0.05
      −0.01−0.42
      P < 0.01
      K micellar0.43
      P < 0.001.
      −0.17−0.40
      P < 0.01
      0.110.48
      P < 0.001.
      Mg total0.79
      P < 0.001.
      0.15−0.65
      P < 0.001.
      0.040.82
      P < 0.001.
      Mg soluble−0.100.180.0020.06−0.10
      Mg micellar0.79
      P < 0.001.
      0.09−0.62
      P < 0.001.
      0.020.82
      P < 0.001.
      Na total0.38
      P < 0.01
      0.33
      P < 0.01
      −0.39
      P < 0.01
      0.070.68
      P < 0.001.
      P total0.78
      P < 0.001.
      0.15−0.67
      P < 0.001.
      0.070.83
      P < 0.001.
      P soluble−0.71
      P < 0.001.
      −0.49
      P < 0.001.
      0.52
      P < 0.001.
      0.04−0.74
      P < 0.001.
      P micellar0.81
      P < 0.001.
      0.24
      P < 0.05
      −0.67
      P < 0.001.
      0.050.85
      P < 0.001.
      * P < 0.05
      ** P < 0.01
      *** P < 0.001.
      Within each mineral, the soluble phase was negatively correlated with total and micellar fraction (Table 6), except for the association between soluble and total K (r = 0.48; P < 0.001), whereas the correlation between total and micellar fraction was positive, ranging from 0.72 for K to 0.99 for Ca (P < 0.001). Total and micellar Ca were positively correlated with total K, Mg, Na, and P, exhibiting correlations from 0.24 (P < 0.05) to 0.98 (P < 0.001), and with micellar K, Mg, and P (r = 0.57 to 0.99; P < 0.001), whereas associations of total and micellar Ca with soluble K and P were negative (r = −0.82 to −0.29; P < 0.01; Table 6). Correlations between soluble Ca and other mineral fractions were of opposite sign compared with those of total and micellar Ca with other fractions. Similarly, total and micellar K correlated positively with total and micellar Mg and P (r = 0.25 to 0.74; P < 0.05), and soluble K correlated negatively with total and micellar Mg and P (r = −0.48 to −0.27; P < 0.05). Moreover, soluble K correlated positively (P < 0.01) with soluble Mg (r = 0.72) and P (r = 0.37). Correlations between total Mg and other fractions mirrored those of micellar Mg with other fractions in terms of direction and magnitude (Table 6). Overall, moderate correlations were estimated between total Na and other mineral fractions ranging from −0.51 (P < 0.001) to 0.47 (P < 0.001).
      Table 6Pearson correlations between major mineral fractions (mg/g)
      TraitCa solubleCa micellarK totalK solubleK micellarMg totalMg solubleMg micellarNa totalP totalP solubleP micellar
      Ca total−0.71
      P < 0.001.
      0.99
      P < 0.001.
      0.32
      P < 0.01
      −0.29
      P < 0.01
      0.59
      P < 0.001.
      0.94
      P < 0.001.
      −0.070.92
      P < 0.001.
      0.43
      P < 0.001.
      0.98
      P < 0.001.
      −0.76
      P < 0.001.
      0.98
      P < 0.001.
      Ca soluble−0.81
      P < 0.001.
      0.100.60
      P < 0.001.
      −0.37
      P < 0.01
      −0.66
      P < 0.001.
      0.56
      P < 0.001.
      −0.81
      P < 0.001.
      −0.48
      P < 0.001.
      −0.69
      P < 0.001.
      0.88
      P < 0.001.
      −0.78
      P < 0.001.
      Ca micellar0.24
      P < 0.05
      −0.38
      P < 0.01
      0.57
      P < 0.001.
      0.93
      P < 0.001.
      −0.180.95
      P < 0.001.
      0.47
      P < 0.001.
      0.97
      P < 0.001.
      −0.82
      P < 0.001.
      0.99
      P < 0.001.
      K total0.48
      P < 0.001.
      0.72
      P < 0.001.
      0.48
      P < 0.001.
      0.40
      P < 0.01
      0.32
      P < 0.01
      −0.200.35
      P < 0.01
      0.080.25
      P < 0.05
      K soluble−0.27
      P < 0.05
      −0.27
      P < 0.05
      0.72
      P < 0.001.
      −0.48
      P < 0.001.
      −0.43
      P < 0.001.
      −0.28
      P < 0.05
      0.37
      P < 0.01
      −0.32
      P < 0.01
      K micellar0.73
      P < 0.001.
      −0.130.74
      P < 0.001.
      0.130.61
      P < 0.001.
      −0.210.53
      P < 0.001.
      Mg total0.020.95
      P < 0.001.
      0.37
      P < 0.01
      0.93
      P < 0.001.
      −0.65
      P < 0.001.
      0.91
      P < 0.001.
      Mg soluble−0.30
      P < 0.01
      −0.20−0.100.28
      P < 0.05
      −0.15
      Mg micellar0.41
      P < 0.01
      0.92
      P < 0.001.
      −0.71
      P < 0.001.
      0.92
      P < 0.001.
      Na total0.38
      P < 0.01
      −0.51
      P < 0.001.
      0.43
      P < 0.001.
      P total−0.71
      P < 0.001.
      0.98
      P < 0.001.
      P soluble−0.83
      P < 0.001.
      * P < 0.05
      ** P < 0.01
      *** P < 0.001.

      DISCUSSION

      Curd Composition

      The great variability of mineral content for the analyzed samples could be due to (1) the different geographical origin of samples, (2) different milk types and milk compositions and, (3) different manufacturing protocols across dairy companies. The chemical composition of curd samples was in accordance with results obtained in cheese and pasta filata by
      • Manuelian C.L.
      • Currò S.
      • Penasa M.
      • Cassandro M.
      • De Marchi M.
      Characterization of major and trace minerals, fatty acid composition, and cholesterol content of Protected Designation of Origin cheeses.
      . To our knowledge, this is the first study that dealt with the determination of mineral content in curd samples; indeed, the scientific literature reports only a few studies on mineral composition determined in different types of cheese (
      • Hassan A.
      • Johnson M.E.
      • Lucey J.A.
      Changes in the proportions of soluble and insoluble calcium during the ripening of Cheddar cheese.
      ;
      • Fox P.F.
      • Guinee T.P.
      • Cogan T.M.
      • McSweeney P.L.H.
      Salting of cheese curd.
      ). Calcium content in mozzarella cheese was 3.26 mg/g (
      • Manuelian C.L.
      • Currò S.
      • Penasa M.
      • Cassandro M.
      • De Marchi M.
      Characterization of major and trace minerals, fatty acid composition, and cholesterol content of Protected Designation of Origin cheeses.
      ) and 5.90 mg/g (
      • Fox P.F.
      • Guinee T.P.
      • Cogan T.M.
      • McSweeney P.L.H.
      Salting of cheese curd.
      ), and it was 6.34 mg/g (
      • Hassan A.
      • Johnson M.E.
      • Lucey J.A.
      Changes in the proportions of soluble and insoluble calcium during the ripening of Cheddar cheese.
      ) and 8.33 mg/g (
      • Lucey J.A.
      • Fox P.F.
      Importance of calcium and phosphate in cheese manufacture: A review.
      ) in Cheddar cheese.
      • Cichoscki A.J.
      • Valduga E.
      • Valduga A.T.
      • Tornadijo M.E.
      • Fresno J.M.
      Characterization of Prato cheese, a Brazilian semi-hard cow variety: Evolution of physico-chemical parameters and mineral composition during ripening.
      reported concentrations of major minerals in semihard Prato cheese maturated for a very short period (1 d) that were greater than those reported in the present study. Curd samples analyzed in our study were intended for mozzarella cheese production. We found greater content of minerals in curd compared with mozzarella cheese (
      • Manuelian C.L.
      • Currò S.
      • Penasa M.
      • Cassandro M.
      • De Marchi M.
      Characterization of major and trace minerals, fatty acid composition, and cholesterol content of Protected Designation of Origin cheeses.
      ), and this is likely attributable to the high temperature during manufacturing of pasta filata cheese that removes part of minerals through the whey.
      Regarding the differentiation between soluble and micellar fractions of major minerals, only few studies have characterized such traits, and all dealt with Cheddar cheese (
      • Lucey J.A.
      • Fox P.F.
      Importance of calcium and phosphate in cheese manufacture: A review.
      ;
      • Hassan A.
      • Johnson M.E.
      • Lucey J.A.
      Changes in the proportions of soluble and insoluble calcium during the ripening of Cheddar cheese.
      ). Previous studies reported a decrease of micellar Ca content, and consequently an increase of soluble fraction during ripening (
      • Hassan A.
      • Johnson M.E.
      • Lucey J.A.
      Changes in the proportions of soluble and insoluble calcium during the ripening of Cheddar cheese.
      ;
      • Lee M.R.
      • Lee W.J.
      The role of Ca equilibrium on the functional properties of cheese: A review.
      ).
      • Hassan A.
      • Johnson M.E.
      • Lucey J.A.
      Changes in the proportions of soluble and insoluble calcium during the ripening of Cheddar cheese.
      reported total Ca content of 8.33 mg/g and micellar Ca content of 6.44 mg/g at 1 d of ripening, in agreement with results of the current study. Further trials are needed to establish the equivalence and bias between the proposed method and the previously published protocols, and to assess its precision and effectiveness as a gold standard method for the assessment of NIR instruments performances (
      • Hassan A.
      • Johnson M.E.
      • Lucey J.A.
      Changes in the proportions of soluble and insoluble calcium during the ripening of Cheddar cheese.
      ).

      Calibration Models

      Near-infrared spectroscopy calibration models for major mineral contents in milk (
      • Costa A.
      • Visentin G.
      • De Marchi M.
      • Cassandro M.
      • Penasa M.
      Genetic relationships of lactose and freezing point with minerals and coagulation traits predicted from milk mid-infrared spectra in Holstein cows.
      ;
      • Visentin G.
      • Niero G.
      • Berry D.P.
      • Costa A.
      • Cassandro M.
      • De Marchi M.
      • Penasa M.
      Genetic (co)variances between milk mineral concentration and chemical composition in lactating Holstein-Friesian dairy cows.
      ) and cheese have been discussed in the literature (
      • Manuelian C.L.
      • Currò S.
      • Visentin G.
      • Penasa M.
      • Cassandro M.
      • Dellea C.
      • Bernardi M.
      • De Marchi M.
      Technical note: At-line prediction of mineral composition of fresh cheeses using near-infrared technologies.
      ), whereas no information on the prediction of soluble and micellar fractions in cheese or curd is currently available. As proposed by
      • Williams P.
      The RPD statistic: A tutorial note.
      , the accuracy of calibration models can be evaluated through RPD statistics. In particular, RPD values smaller than 2.0 are not recommended for industry application; values between 2.0 and 3.0 limit the opportunity of using prediction models only for screening purposes; values between 3.1 and 3.4 underline good prediction ability of the models and the opportunity of using them for quality control; values between 3.5 and 4.0 highlight good prediction models and the opportunity of using them for on-line control in food process operations; values greater than 4.00 are considered suitable for any application. Both in terms of RPD and R2, the Lab-VIS-NIR instrument showed the best performances. This is likely due to the broader wavelength spectrum and to the higher resolution of Lab-VIS-NIR (400–2,500 nm and 0.5 nm, respectively) compared with the wavelength spectra and resolutions of Lab-NIR (850–1,050 nm and 2 nm, respectively) and portable NIR (740–1,070 nm and 1 nm, respectively). Regarding NIR transmittance technology, prediction models showed RPD of 3.29, 1.29, 3.22, 3.36 and 2.70 for Ca, K, Mg, Na, and P, respectively, which are lower than those reported by
      • Manuelian C.L.
      • Currò S.
      • Visentin G.
      • Penasa M.
      • Cassandro M.
      • Dellea C.
      • Bernardi M.
      • De Marchi M.
      Technical note: At-line prediction of mineral composition of fresh cheeses using near-infrared technologies.
      , except for Na. Possible reasons to explain differences between our results and those of
      • Manuelian C.L.
      • Currò S.
      • Visentin G.
      • Penasa M.
      • Cassandro M.
      • Dellea C.
      • Bernardi M.
      • De Marchi M.
      Technical note: At-line prediction of mineral composition of fresh cheeses using near-infrared technologies.
      are the sample size and the variability of data.
      • Manuelian C.L.
      • Currò S.
      • Visentin G.
      • Penasa M.
      • Cassandro M.
      • Dellea C.
      • Bernardi M.
      • De Marchi M.
      Technical note: At-line prediction of mineral composition of fresh cheeses using near-infrared technologies.
      used 145 cheese samples (different types, from fresh to hard cheeses), which resulted in high standard deviations compared with our study. The RPD values of prediction models developed using spectra collected by Lab-VIS-NIR (which works in reflectance mode) ranged from 1.39 (total K) to 4.28 (total Na).
      • Lucas A.
      • Andueza D.
      • Rock E.
      • Martin B.
      Prediction of dry matter, fat, pH, vitamins, minerals, carotenoids, total antioxidant capacity, and colour in fresh and freeze-dried cheeses by visible-near-infrared reflectance spectroscopy.
      investigated the feasibility of predicting mineral content in cheese using NIR reflectance spectroscopy in a data set of 445 samples and reported greater RPD for Ca (4.56) and K (2.14), and lower RPD for Mg (2.33) compared with the present study. Regarding portable NIR, previous studies investigated only prediction of fat and moisture content of cheese, testing different mathematical treatments. We focused on major minerals obtaining RPD between 1.08 and 2.73, meaning that some calibration may be used for screening purposes. It is worth noting that portable NIR can predict micellar minerals with better accuracy compared with total minerals (
      • Wiedemair V.
      • Langore D.
      • Garsleitner R.
      • Dillinger K.
      • Huck C.
      Investigations into the Performance of a novel pocket-sized near-infrared spectrometer for cheese analysis.
      ).

      Correlations

      Considering that the matrix of the present study was curd, significant associations of total protein with micellar and total minerals were expected. Indeed, most of milk mineral is contained in the casein reticulum formed after milk coagulation; only a small fraction is lost in the soluble phase, especially during cheese manufacturing (
      • Fox P.F.
      • Guinee T.P.
      • Cogan T.M.
      • McSweeney P.L.H.
      Salting of cheese curd.
      ). This was also confirmed by the association between total proteins and micellar minerals (Table 5). In regards to humidity, a significant positive correlation with soluble mineral fractions was expected because part of the minerals was lost with the whey after reaching mineral equilibrium in the curd matrix. It is worthy to report that the loss of micellar minerals depends also on the pH values of the curd under whey during the cheesemaking process. The timing of this manufacturing step may have influenced the mineral composition at a single curd level, and may explain (at least partially) the variability of mineral composition across curds involved in the present study. Instead, total mineral fraction showed negative correlation with humidity because curd minerals were mostly composed by micellar minerals that were trapped in the curd matrix (
      • Fox P.F.
      • Guinee T.P.
      • Cogan T.M.
      • McSweeney P.L.H.
      Salting of cheese curd.
      ). The associations between ash, protein, and mineral contents were similar or higher than those reported by
      • Cichoscki A.J.
      • Valduga E.
      • Valduga A.T.
      • Tornadijo M.E.
      • Fresno J.M.
      Characterization of Prato cheese, a Brazilian semi-hard cow variety: Evolution of physico-chemical parameters and mineral composition during ripening.
      for 108 cheese samples. In fact, those authors reported a correlation of 0.75 (P < 0.001) between ash and Na, and 0.51, 0.80, and 0.39 (P < 0.001) between protein content and Ca, P, and Mg, respectively. In particular, curd ash content tended to increase when the soluble phase of minerals decreased (
      • Fox P.F.
      • Guinee T.P.
      • Cogan T.M.
      • McSweeney P.L.H.
      Salting of cheese curd.
      ). The strong correlation between total and micellar phases in all minerals was expected, considering that curd can be imagined as a concentration of milk solids, and micellar fractions of Ca, P, and Mg are almost totally found in casein micelles. All correlations between total Na and total phase of other minerals were stronger and positive compared with those reported by
      • Cichoscki A.J.
      • Valduga E.
      • Valduga A.T.
      • Tornadijo M.E.
      • Fresno J.M.
      Characterization of Prato cheese, a Brazilian semi-hard cow variety: Evolution of physico-chemical parameters and mineral composition during ripening.
      , who observed correlations of −0.34 and −0.35 (P < 0.01) between Na and Mg, and Na and Ca, respectively. The very strong relationship between micellar Ca and P (r = 0.99) was expected, considering the strict chemical relationship they have as constituents of the micellar calcium phosphate, which is an essential component of the casein micelle (
      • Holt C.
      Casein and casein micelle structures, functions and diversity in 20 species.
      ). The strong correlation between micellar phases of Mg, Ca, and P resembled those estimated by
      • Lucas A.
      • Andueza D.
      • Rock E.
      • Martin B.
      Prediction of dry matter, fat, pH, vitamins, minerals, carotenoids, total antioxidant capacity, and colour in fresh and freeze-dried cheeses by visible-near-infrared reflectance spectroscopy.
      for 445 samples of cheese. Moreover, a similar association between total Ca and total Mg (0.71; P < 0.001) was assessed by
      • Cichoscki A.J.
      • Valduga E.
      • Valduga A.T.
      • Tornadijo M.E.
      • Fresno J.M.
      Characterization of Prato cheese, a Brazilian semi-hard cow variety: Evolution of physico-chemical parameters and mineral composition during ripening.
      : Mg can be thought of as a competitor of Ca during the curd formation process because both Ca and Mg act as a bond between casein micelles (
      • Alexander T.G.
      • Ford T.F.
      Magnesium in the casein-containing colloid of milk.
      ;
      • Malacarne M.
      • Franceschi P.
      • Formaggioni P.
      • Sandri S.
      • Mariani P.
      • Summer A.
      Influence of micellar calcium and phosphorus on rennet coagulation properties of cows milk.
      ).

      Implications

      Calibrations described in the current study can be used for different purposes, from screening to process or quality control, with applications in field conditions. The portable device is cheaper, faster, and easier to be handled by operators compared with laboratory instruments, but it is useful only for screening purposes, whereas laboratory instruments are potentially useful to control any process that involves curd. According to the correlations estimated in the present study, the Italian dairy industry may be able to make decisions regarding curd management and handling from the knowledge of the concentration of a group of minerals or their micellar phase. In addition, the possibility to predict soluble and micellar mineral fractions could provide the industry a valuable instrument to rapidly evaluate the quality of curd according to an on-site or on-line approach.

      CONCLUSIONS

      The determination of mineral fractions is important to evaluate the quality of commercial frozen and fresh curds and for prompt decision-making at the industry level. The present study dealt with the characterization of total, micellar, and soluble contents of 5 major minerals (Ca, Mg, K, P, Na) predicted in curd intended for mozzarella manufacture. The accuracy of prediction of different NIR instruments was also discussed. Overall, micellar and total phase of the same mineral were strongly correlated; instead, the soluble phase was negatively associated with both micellar and total fractions, especially for Ca and P. Considering calibration models in terms of RPD, the laboratory instrument operating in visible and NIR regions had higher performances for all mineral fractions, except for micellar K, total Mg, and soluble Mg. In general, the portable NIR device showed lower precision compared with the laboratory instruments, and its application is likely limited to screening purposes.

      ACKNOWLEDGMENTS

      We gratefully acknowledge the “Innovazioni di processo e di prodotto nel settore caseario per un miglioramento della qualità, delle proprietà salutistiche, della shelf-life e della sostenibilità ambientale - LatTraZione 2020 - CDS000628” project funded by Sabelli (Ascoli Piceno, Italy). The authors have not stated any conflicts of interest.

      REFERENCES

        • Alexander T.G.
        • Ford T.F.
        Magnesium in the casein-containing colloid of milk.
        J. Dairy Sci. 1957; 40: 1273-1276
        • Cichoscki A.J.
        • Valduga E.
        • Valduga A.T.
        • Tornadijo M.E.
        • Fresno J.M.
        Characterization of Prato cheese, a Brazilian semi-hard cow variety: Evolution of physico-chemical parameters and mineral composition during ripening.
        Food Control. 2002; 13: 329-336
        • Costa A.
        • Visentin G.
        • De Marchi M.
        • Cassandro M.
        • Penasa M.
        Genetic relationships of lactose and freezing point with minerals and coagulation traits predicted from milk mid-infrared spectra in Holstein cows.
        J. Dairy Sci. 2019; 102 (31155264): 7217-7225
        • De Marchi M.
        • Toffanin V.
        • Cassandro M.
        • Penasa M.
        Invited review: Mid-infrared spectroscopy as phenotyping tool for milk traits.
        J. Dairy Sci. 2014; 97 (24440251): 1171-1186
        • Fox P.F.
        • Guinee T.P.
        • Cogan T.M.
        • McSweeney P.L.H.
        Salting of cheese curd.
        in: Fundamentals of Cheese Science. Springer, Boston, MA2017: 251-277
        • Franzoi M.
        • Niero G.
        • Penasa M.
        • Cassandro M.
        • De Marchi M.
        Technical note: Development and validation of a new method for the quantification of soluble and micellar calcium, magnesium, and potassium in milk.
        J. Dairy Sci. 2018; 101 (29274961): 1883-1888
        • Hassan A.
        • Johnson M.E.
        • Lucey J.A.
        Changes in the proportions of soluble and insoluble calcium during the ripening of Cheddar cheese.
        J. Dairy Sci. 2004; 87 (15259219): 854-862
        • Holt C.
        Casein and casein micelle structures, functions and diversity in 20 species.
        Int. Dairy J. 2016; 60: 2-13
        • Lante A.
        • Lomolino G.
        • Cagnin M.
        • Spettoli P.
        Content and characterisation of minerals in milk and in Crescenza and Squacquerone Italian fresh cheeses by ICP-OES.
        Food Control. 2006; 17: 229-233
        • Lee M.R.
        • Lee W.J.
        The role of Ca equilibrium on the functional properties of cheese: A review.
        Korean J. Food Sci. Anim. Resour. 2009; 29: 545-549
        • Lucas A.
        • Andueza D.
        • Rock E.
        • Martin B.
        Prediction of dry matter, fat, pH, vitamins, minerals, carotenoids, total antioxidant capacity, and colour in fresh and freeze-dried cheeses by visible-near-infrared reflectance spectroscopy.
        J. Agric. Food Chem. 2008; 56 (18646761): 6801-6808
        • Lucey J.A.
        • Fox P.F.
        Importance of calcium and phosphate in cheese manufacture: A review.
        J. Dairy Sci. 1993; 76: 1714-1724
        • Malacarne M.
        • Franceschi P.
        • Formaggioni P.
        • Sandri S.
        • Mariani P.
        • Summer A.
        Influence of micellar calcium and phosphorus on rennet coagulation properties of cows milk.
        J. Dairy Res. 2014; 81 (24345431): 129-136
        • Manuelian C.L.
        • Currò S.
        • Penasa M.
        • Cassandro M.
        • De Marchi M.
        Characterization of major and trace minerals, fatty acid composition, and cholesterol content of Protected Designation of Origin cheeses.
        J. Dairy Sci. 2017; 100 (28237598): 3384-3395
        • Manuelian C.L.
        • Currò S.
        • Visentin G.
        • Penasa M.
        • Cassandro M.
        • Dellea C.
        • Bernardi M.
        • De Marchi M.
        Technical note: At-line prediction of mineral composition of fresh cheeses using near-infrared technologies.
        J. Dairy Sci. 2017; 100 (28624277): 6084-6089
        • Reykdal O.
        • Rabieh S.
        • Steingrimsdottir L.
        • Gunnlaugsdottir H.
        Minerals and trace elements in Icelandic dairy products and meat.
        J. Food Compos. Anal. 2011; 24: 980-986
        • Visentin G.
        • Niero G.
        • Berry D.P.
        • Costa A.
        • Cassandro M.
        • De Marchi M.
        • Penasa M.
        Genetic (co)variances between milk mineral concentration and chemical composition in lactating Holstein-Friesian dairy cows.
        Animal. 2019; 13: 477-486
        • Visentin G.
        • Penasa M.
        • Gottardo P.
        • Cassandro M.
        • De Marchi M.
        Predictive ability of mid-infrared spectroscopy for major mineral composition and coagulation traits of bovine milk by using the uninformative variable selection algorithm.
        J. Dairy Sci. 2016; 99 (27522421): 8137-8145
        • Wiedemair V.
        • Langore D.
        • Garsleitner R.
        • Dillinger K.
        • Huck C.
        Investigations into the Performance of a novel pocket-sized near-infrared spectrometer for cheese analysis.
        Molecules. 2019; 24 (30682872): 428
        • Williams P.
        The RPD statistic: A tutorial note.
        NIR News. 2014; 25: 22-26