Three production batches of DL were carried out in duplicate at pH_is of 6.5, 7.0, and 7.5. The three assayed pH_is are within the common pH range used in DL production. Samples were taken at 0, 20, 40, 80, and 150 min after the initiation of the preparation.

The DL preparation was performed according to Rodríguez et al. (14) by concentration of a solution of sucrose (20% w/v) in pasteurized milk with a protein content of 2.9%−3.0% and fat content below 0.1% in a 60-L industrial kettle. The pH_i of 6.5 is obtained solely from dissolution of sucrose in milk, and the higher pH_i values were adjusted by the addition of sodium bicarbonate. In all cases, pH was monitored with a pH meter equipped with a penetration electrode (Hanna HI8424, HANNA Instruments, RI, USA).

Colorimetric measurements of DL were performed in cylindrical white cells (28 mm diameter, 4 mm high). The CIELab color system was used to characterize the samples with the parameters L^*, a^*, b^*, and C^*. The measurements were performed in a Minolta CM 508d spectrophotometer (Minolta Co. Ltd., Osaka, Japan) with an illuminant D65, a 10° standard observation angle, and specular component excluded. Color differences were measured as dEab* = (dL*² + da*² + db*²)^1/2, using milk and water as blanks for DL and melanoidins, respectively.

The extraction and fractionation of melanoidins were mainly performed as described in Rodríguez et al. (14). Briefly, a water suspension of DL was extensively dialyzed against water to remove sucrose and other low molecular weight (MW) components using a membrane with a cut-off of 6–8 kDa (Spectra/Por, Spectrum Laboratories Inc., Rancho Dominguez, CA, USA). The macromolecular components of DL were then fractionated by centrifugation at 5,500 rpm for 20 min at 25°C, and the supernatant was collected and freeze-dried to give the soluble component (S). The resultant precipitate (macromolecular water insoluble fraction) was washed with water, centrifuged, and freeze-dried to give the insoluble material (I, Figure 1).

The soluble and insoluble percentage, %S and %I, respectively, of the macromolecular fraction of DL was calculated as the percent ratio of the mass of the S and I components relative to their sum. The average value was calculated from the results of three repetitions.

The I and S macromolecular fractions were suspended in 50 mM phosphate buffer at pH 7.0 to a final concentration of 20 mg/mL and treated with 40 UI/mL of Pronase E from Streptomyces griseus (4 UI/mg, Sigma-Aldrich, St. Louis, MI, USA) at 37°C for 24 h with magnetic stirring, and then centrifuged at 4,500 rpm for 30 min at 4°C. The supernatants obtained from the treatment of the I and S fractions were separated by centrifugation, freeze-dried, and labeled IHS and SH, respectively (Figure 1) (14).

The melanoidins from the IHS and SH fractions were isolated by gel filtration using a BioGel P2 column (2.6 × 50 cm, BioRad, Hercules, CA, USA). Aqueous acetic acid (0.2% v/v) at a flow rate of 0.6 mL/min was used as the mobile phase, and the eluent was continuously monitored with a Shimadzu RID-6A refractive index (RI) detector. Aliquots of IHS or SH (50 mg in 2 mL of mobile phase) were loaded onto the column, and the fractions were collected every 5 min. The absorbance at 280 and 420 nm of each fraction was recorded on a Pharmacia Ultrospec 1000 spectrophotometer (Pharmacia, Uppsala, Sweden) (14).

The chromatographic separation procedure was repeated three times for each sample to produce enough material for further characterization. Melanoidin fractions corresponding to a nominal MW range from 400 to 1,800 Da were pooled.

Elemental analyses of the samples (2 mg each) were performed to estimate the nitrogen/carbon ratio (N/C) and sulfur content using a Carlo Erba EA1108 elemental analyzer (Sabadell, Spain).

Selected melanoidin gel filtration fractions were analyzed by NMR. Samples were dissolved in D₂O containing a trace of sodium 2,2,3,3-d₄-3-(trimethylsilyl)propionate (TSP) used as the internal standard, and transferred to 5-mm tubes (NE-HL5-7, New Era Enterprises, Newark, NJ, USA). Spectra were recorded at 25°C on a Bruker AVANCE III 500 NMR spectrometer (Bruker Corp., Billerica, MA, USA) operating at ¹H and ¹³C frequencies of 500.13 and 125.76 MHz, respectively, and equipped with a z-gradient TXI probe. Water-suppressed ¹H spectra were recorded using a 30° pulse and presaturation during a 2-s repetition delay, accumulating a total of 256 scans. HSQC spectra were obtained with a gradient-enhanced pulse sequence (15), using 128 scans per slice and a repetition delay of 1 s. ¹H and ¹³C chemical shifts (δ_H and δ_C, respectively) are reported in ppm.

As described in Rodríguez et al. (14), methanol suspensions of selected melanoidin fractions (6 mg/mL) were stirred in an orbital shaker for 1 h and then centrifuged at 10,000 rpm for 20 min. The supernatant was collected, the solvent was evaporated under a nitrogen stream, and the residue was dissolved in aqueous methanol (5% v/v, 200 μL). These solutions (20 μL) were analyzed by HPLC using a Dionex Ultimate 3000 HPLC (Thermo Scientific, Waltham, MA, USA), equipped with a C18 Zorbax Eclipse column (250 × 4.6 mm, 5 μm particle size) and a DAD detector. A gradient of methanol and water at a flow rate of 1 mL/min was used, going from 5 to 35% methanol in 30 min, and up to 100% methanol at 42 min.

Colored compounds that were bound and not bound to protein were determined on the final product (processing time 2.5 h) according to Morales and van Boekel (16) with some modifications. Each sample was analyzed in triplicate.

For determination of colored compounds not bound to protein, DL (2.0 g) was mixed with aqueous TCA (24% w/v, 2.0 mL), homogenized, and centrifuged for 10 min (12,000 g, 25°C). The supernatant was removed and diluted as necessary to read the absorbance at 420 and 550 nm. The blank was a sample of the initial mixture of milk and sucrose, prior to the heat treatment (sample at time 0), and was submitted to the treatment described above.

The browning index was defined as BI = (Abs 420 – Abs 550 nm)/g, after corrections by dilution. The browning index in this case corresponds to colored compounds not bound to proteins (BI_nb).

Colored compounds bound to protein were released by proteolysis. Briefly, DL (2.0 g) was suspended in 50 mM phosphate buffer (2.0 mL, pH 7.0), homogenized, and treated with a solution of Pronasa E (2 mL of 1 % w/v in 50 mM phosphate buffer, pH 7.0) from S. griseus (Sigma Aldrich, ~4 UI/mg). The enzymatic treatment was carried out in capped tubes for 24 h in a water bath (37°C) with agitation (100 rpm). The reaction mixture was then centrifuged at room temperature (12,000 g, 10 min). The supernatant was diluted as necessary to read the absorbances at 420 and 550 nm, and similarly as described above, the browning index corresponding to the total colored compounds was calculated as BI_tot = (Abs 420 – Abs 550 nm)/g (bound and no bound to protein).

Colored compounds bound to protein were then calculated as BI_b = BI_tot – BI_nb, and the amount of colored compounds bound and not bound to protein were calculated as percentage: (BI_nb/BI_tot) × 100 and (BI_b/BI_tot) × 100, respectively.

Analysis of variance (ANOVA) with Tukey tests were carried out to determine significant differences between values (α < 0.05) using Infostat (version 2016, Grupo Infostat, Facultad de Ciencias Agrarias, Universidad Nacional de Cordoba, Argentina). Results were expressed as mean ± SD, and different letters indicate significant differences when α < 0.05.

The effect of pH_i on DL color development along the production process showed that higher pH favors an earlier development of darker colors, and thus, probably, the formation of the MR products (Supplementary Figure 1).

The effect of pH on the advancement, and the preferred mechanism of reaction of the MR has been extensively studied (17). Higher pH appears to favor the degradation of sugars, leading to reactive species that speed up the reaction, and the generation of MR products (12). Thus, it cannot be determined a priori whether the increase in color is due to a higher concentration of colored compounds or to structural differences of these compounds due to the different mechanisms favored by a specific pH.

As observed in Figure 2, the color and color parameters of DL were noticeably dependent on the pH_i, even for a relatively narrow pH range of 1.0 unit.

The highest pH_i produced lower luminosity (L^*) and a higher color development (dE^*ab) in the product compared with the initial mix of milk and sucrose.

As we reported previously (14), the semisolid DL dispersed in water can be fractionated in water-soluble (S) and water-insoluble (I) fractions, both colored, but it was observed that the fraction I contributed the most to the color of the product. This was evaluated for DL prepared at three different pH_i values, and it was found that the protein-bound colored products were the most abundant based on proteolysis and absorbance measurement (16) (Table 1).

The I fraction yield was evaluated for the different pH_i values. As expected, pH_i can influence protein interactions and also the elaboration time. In addition since salt removal can affect protein stability and cause aggregation, the I fraction yield was evaluated before and after dialysis.

The pH_i did not significantly affect (α ≤ 0.05) the proportion of the I fraction formed during DL elaboration. Indeed, the I fraction before and after dialysis went, approximately, from 10 to 30% when pHi increased (or decreased) from 6.5 to 7.5 (Supplementary Figure 2).

Furthermore, the I fraction before dialysis was almost insignificant when milk and sucrose were initially mixed. This was independent of pH_i even at production times of 60 min, despite a brown color that developed for the highest pH_i. After dialysis, the percentage of I fraction was higher for the initial mixture and the DL, suggesting that salt concentration plays an important role in macromolecule stability (data not shown).

These results show that the formation of the I fraction in DL takes place at an advanced elaboration time, independent of color development and pH_i. We can assume that the formation of the I fraction is due to the thermal treatment that induces protein associations or the formation of aggregates due to denaturation, disulfide bonds formation, gelling, and even precipitation (18), and where the MR can also contribute through crosslinking (19–21). Lack of a direct correlation between color development and protein insolubilization was reported in the case of ovalbumin (20), but it has not been described before for DL production.

The color of the different fractionation components of the DL produced at different pH_i values, vis the I fraction, the S fraction, and the insoluble fraction after enzymatic hydrolysis of the DL (IHS), are shown in Figure 3. The influence of pH_i can be seen in the remarkable differences between the water-soluble colored compounds (S) and the highly colored soluble fraction obtained after hydrolysis (HIS).

It can be observed that the color contribution of the insoluble I fraction (I) is released in a soluble form after hydrolysis (IHS), whereas the remaining insoluble residue remains colorless. This indicates that the main-colored components may be attributed to melanoproteins in I, which released water-soluble melanoidins after enzymatic hydrolysis (IHS fraction).

The color of lyophilized fractions was affected due to the incident light on the powder. To avoid this effect, the color was measured in the corresponding aqueous solutions to obtain the parameters da^*, db^*, dC^*, and dE^*ab. Higher da^*, db^*, dC^*, and dEab* were obtained for SH and IHS. As higher pHi of DL is produced, we can conclude that fraction solutions that contain the soluble melanoidins (SH and HIS) contain the main components responsible for the color differences between the three DL (Supplementary Figure 3).

Regarding the results above, we can conclude that melanoidins could be liberated by enzymatic proteolysis from the melanoproteins that compose the insoluble macromolecular colored components, and this colored solution correctly represents DL color.

Soluble melanoidins could be isolated and their MWs could be estimated by gel filtration. The use of an in-line RI detector permits the evaluation of the mass abundance of the different fractions since the detector response is independent of the nature of the solute for sufficiently diluted solutions. Figure 4 shows the chromatograms of the IHS fractions of the DLs prepared at different pH_i values.

The chromatogram obtained with RI detection reveals similarities between the IHS from DLs elaborated at pH_i values of 6.5 and 7.0 (Figure 4A). On the other hand, the IHS from DL elaborated at a pH_i of 7.5 presented a major peak at K_av = 0.85–0.90, representing a higher amount of compounds with MW ≥ 500 Da (K_av ≤ 0.6), and also compounds eluting with the dead volume (K_av = 0) corresponding to a nominal MW ≥ 1,800 Da. These compounds also had higher absorbance at 280 and 420 nm (Figures 4B,C, respectively), and may explain the more intense color of DL prepared at a higher pH_i.

For further studies and characterization, we selected the melanoidins fractions with K_av between 0.17 and 0.65, corresponding to nominal MW from 400 up to 1,800 Da, thus avoiding interference from amino acids or sugars in peaks II and III.

The color difference of the melanoidins fractions can be seen in Figure 5. The color parameters of the selected fractions correlate with those of the corresponding IHS and the original DL, and their dependence on pH_i (Supplementary Figure 4).

The nitrogen (N) and carbon (C) content of the different DL fractions and melanoidins were determined by elemental analysis, and the N/C ratios were calculated (Table 2).

The N/C ratio obtained for the I fraction and the melanoidins are similar to those found in the studies of the incorporation of sugar into caseins (N/C ratios from 0.23 to 0.25) (22). However, the N/C ratios determined for simple model systems diverge, with N contents of 6% in high MW melanoidins obtained in model systems of glycine with glucose, fructose, or HMF (23), and in glucose–glycine model systems (24). These differences between DL and the model systems are likely due to the different reactivity of macromolecules in the real systems with respect to the models.

The N/C ratio of the insoluble fraction and the melanoidins are very similar, and range from 0.54 to 0.66% (Table 2). Sulfur (S) was detected in the isolated melanoidins from DL produced at different pH_i values. S-containing compounds could originate by the MR in the presence of cysteine or cystine, giving rise to thiazolines or thiazolidines among other types of compounds. These are expected to be in higher concentration in melanoidins, explaining the presence of sulfur in these samples (25).

According to Hayashi and Namiki (13), the N/C ratio of melanoidins can increase when the fragmentation route is favored at basic pH. However, the N/C ratio for the fractions was very similar, regardless of the pH_i value. This similar N/C ratio can be due to the narrow pH_i range tested, close to neutrality. Therefore, the elemental composition would not explain the difference in the color of melanoidins, and so these differences may be better explained by structural differences such as the degree of unsaturation and aromaticity.

As recently reported by our group (14), the NMR spectra of melanoidins reveal their structural complexity. Despite isolation and purification, they include a mixture of products comprising initial to end products of the MR, many of which can exist as stereoisomers (26, 27). However, we can partially characterize the chemical groups associated with the potential products of the MR, allowing us to compare melanoidin fractions of DL elaborated at different pH_i through a preliminary structural characterization (Figures 6, 7 and Supplementary Figure 5). To do this, we compared the variations observed between the resonances from aromatic protons (δ_H 6.65–6.90, 7.12–7.50, and 8.45–8.65), those from protons bound to carbons bearing heteroatoms (δ_H 4.93, 4.85, and 4.48), and those from aliphatic protons (δ_H 0.5–2.5).

As shown in Figure 6, the ¹H NMR spectra of the melanoidins fractions from DL elaborated at different pH_i values have a similar profile, with only minor differences in the regions of interest. The ratios between the different chemical groups were measured using the spectral regions detailed above, using the area of the aliphatic protons as an arbitrary reference (Figure 7).

The ratio of aromatic to aliphatic protons increases as the pH_i in DL elaboration rises. This is consistent with the fact that heteroaromatic, including derivatives of pyrrole, furane, or pyridine, are typical intermediaries, and the final products of the MR and their formation are favored in basic environments (12, 28–30).

As we have recently reported (14), the formation of Amadori products from the reaction between milk proteins and sugars can be evidenced in the HSQC spectra of the melanoidin fractions. Indeed, the one-bond correlation between the proton resonance at δ_H 4.48 and the carbon signal at δ_C 100.6 corresponds to the anomeric center of the reducing end of β-d-glucopyranosyl residue of lactose after condensation with the ε-amine group of a lysine residue (Figure 8). Imine condensation is followed by the formation of the Amadori products and subsequent reactions where the galactose residue is preserved in the 1-deoxyosone and 3-deoxyosone routes, which affect only the glucose residue at the reducing end of lactose. On the other hand, the transformation of the Amadori product via the formation of a 4-deoxyosone leads to the loss of a galactose residue. As shown in Figures 7, 8, the signals of the anomeric proton are larger at lower pH_i and almost undetectable at higher pH_i, indicating that different routes could predominate at different DL production pH_i values. The hypothesis that the galactose residue in disaccharides can be lost or transformed is further supported by the proposals of Pischetsrieder et al. (21, 31), which suggest that at neutral to basic pH the Amadori product degradation is via the 4-deoxyosone route with galactose residue loss.

The HPLC chromatograms of the melanoidins fractions with a MW from 400 to 1,800 Da obtained from DLs elaborated at different pH_i with detection at 420 nm are shown in Figure 9.

While the chromatograms are relatively simple, some peaks are not completely resolved due to broadening. It has been proposed that this is due to the presence of polymeric compounds which lowers the resolution when this chromatographic system is employed (MeOH/H₂O gradient) (32).

The hydrophobic character of the melanoidins results in higher retention times relative to the noncolored aromatic compounds present in the mixture, indicating their higher hydrophobicity (14). These results were observed regardless of the pH_i used in DL preparation (Supplementary Figure 6). This behavior is compatible with the formation of structures with a higher degree of condensation and aromaticity as advanced MR products are formed. This observation is also consistent with the analysis of the ¹H NMR data of melanoidins.

The chromatographic profiles of melanoidins are simpler for products obtained from DL produced at the highest and lowest pH_i assayed, showing the presence of one or two main compounds. On the other hand, the melanoidins from DL produced at an intermediate pH_i are more complex. As previously discussed, this may reflect that at more acidic or basic pH_i values some reaction pathways are favored, whereas at pH_i 7.0 several pathways leading to melanoidin formation may occur with a similar probability.

The UV–Vis spectra of the compounds eluting in the major peaks annotated in Figure 9 for the melanoidins obtained from DL elaborated at pH_i 7.0 present absorbance bands in the UV and, in particular, the visible regions. Peak I presented two maxima in the visible region at λ 537 and 594 nm, peak II presents maxima at λ 562 and 592 nm, and peak III presents three maxima at λ 553, 592, and 599 nm (Supplementary Figure 7). Similar spectra were obtained for the corresponding peaks in the chromatograms of the melanoidins obtained from DL prepared at pH_i 6.5 and 7.5 (data not shown).

From the evaluation of the UV–Vis spectra it can be inferred that although the compounds may be structurally similar, they likely correspond to different degrees of advancement of the MR, and thus confer different color to the sample.