The affinity of milk fat globule membrane fragments and buttermilk proteins to hydroxyapatite

Buttermilk differs from skim milk by the presence of milk fat globule membrane (MFGM) fragments that are released during cream churning. MFGM is rich in health-promoting components, such as phospholipids and membrane proteins, but these compounds have a negative impact on buttermilk techno-functional properties in dairy applications. The isolation of MFGM from buttermilk improved its functionality while also recovering the MFGM bioactive components. Hydroxyapatite (HA) can be used to extract MFGM by adsorption via charged site interactions. However, the affinity of HA to MFGM or the main buttermilk proteins (casein micelles (CM), β-lactoglobulin (β-lg) and α-lactalbumin (α-lac)) is not known. The influence of important physicochemical parameters such as pH and temperature on these interactions is also unclear. For each buttermilk component, a quartz crystal microbalance diffusion analysis was performed to determine the maximum adsorption time and the attached mass density on HA-coated gold sensors. The influence of pH, ionic strength (IS), and temperature (T) on the affinity of each buttermilk component for HA particles was assessed using a 3-levels and 3-factors Box-Behnken design. The absorption rate was highest for the CM, followed by β-lg and α-lac, and then by the MFGM. Nevertheless, the final maximal attached mass densities to the HA were similar for the MFGM and CM, and 2.5 times higher than for β-lg and α-lac. This difference can be explained by the higher number of binding sites found in CM and their heavier mass. The model obtained by the Box-Behnken design plan showed that the adsorption of the CM changed with T, pH and IS. These results suggest that the techno-functional properties of buttermilk may be restored by specifically extracting MFGM with HA. Experiments are ongoing to determine conditions for fractionating MFGM directly from buttermilk.


INTRODUCTION
Milk fat globule membrane (MFGM) fragments are released in buttermilk when fat globules are disrupted during cream-churning (Dewettinck et al., 2008).MFGM fragments are comprised of membrane proteins and phospholipids (PLs), such as glycerophospholipids and sphingolipids.In buttermilk, the PL content is about 6 times higher than that of skim milk and is consequently considered a valuable source of MFGM (Calvo et al., 2020).MFGM PLs have several specific healthy, nutritional, and techno-functional properties (Avalli & Contarini, 2005), since they are antimicrobial (Clare et al., 2008), nutraceutical (Spitsberg, 2005), bactericidal (Hancock et al., 2002), cholesterol-lowering (Rueda, 2014), and anticarcinogenic (Zanabria et al., 2014).However, MFGM impairs the use of buttermilk in dairy products such as cheese, as it causes moisture retention and off-flavors (Turcot et al., 2001).For these reasons, minimal amounts of buttermilk are used for human consumption and it is mostly used as animal feed (Lambert et al., 2016).Extraction of MFGM from buttermilk has been widely studied, to recover both this health-promoting fraction enriched with MFGM proteins and PLs, and a delipidated buttermilk fraction with a composition and techno-functional properties comparable to skim milk.However, fractionating MFGM from buttermilk has proven difficult on a larger scale.Indeed, commercial-scale membrane processing commonly used in the dairy industry, such as ultrafiltration and microfiltration, are not selective enough to separate the casein micelles from the similarly sized MFGM fragments in buttermilk (Holzmüller & Kulozik, 2016).A new approach to separate the MFGM is therefore necessary.This will allow obtaining, on the one hand, the MFGM, of significant health interest, and on the other hand, a delipidated buttermilk with restored good improved techno-functional properties, usable as skim milk.This will contribute to valorizing the hundred thousand tons of buttermilk generated by the Canadian dairy industry each year, which have been underutilized until now (Statistique Canada, 2021).
Recent work has shown that MFGM fragments can interact with hydroxyapatite (HA) (Iung et al., 2023).HAs are part of the apatites, a group of phosphate minerals composed of different species of calcite phosphates (Fihri et al., 2017).They can be fluorinated, chlorinated, or hydroxylated, such as fluorapatite (Ca 5 [PO 4 ] 3 F), chlorapatite (Ca 5 [PO 4 ] 3 Cl), or hydroxyapatite (Ca 5 [PO 4 ] 3 [OH]) (Hsu et al., 2020;Iqbal et al., 2013).HA have a hexagonal crystallographic structure composed of tunnels or pores occupied by calcium or hydroxyl ions (Bee & Hamid, 2020).These ions produce charged sites when HA is in an aqueous environment.These sites are called C and P and are heterogeneously distributed on the surface of HA.The dissolution of hydroxyl ions and the exposure of a calcium ion gives the C-sites a positive charge and the P-sites negatively charged phosphate residues.These charges can interact with various compounds like proteins, through electrostatic interactions.The P-sites can interact with positively charged protein functional groups, such as amino groups, and C-sites with negatively charged protein functional groups, like phosphoserine groups (Fihri et al., 2017).The selective adsorption of proteins onto HA particles could be an attractive alternative to separate MFGM from other constituents of buttermilk.Tercinier et al. (17) have demonstrated that other proteins present in buttermilk, namely the caseins, β-lactoglobulin (β-lg), and α-lactalbumin (α-lac), can also interact with HA, with caseins adsorbing preferentially (Tercinier et al., 2013).Other studies have shown that the interactions between HA and proteins depend on the physicochemical characteristics of the solution, such as pH, ionic strength, and temperature (Sharpe et al., 1997;Tercinier, Ye, Singh, et al., 2014).Indeed, variations in pH and salt composition changed the HA surface charge and had an impact on the adsorption of proteins, such as lysozyme (Ma et al., 2016), bovine serum albumin (Yin et al., 2002a), human serum proteins (Mohsen-Nia et al., 2012), and lactoferrin (Iafisco et al., 2011).For instance, higher NaCl concentration increased adsorption of bovine serum albumin, while increasing pH reduces it (Yin et al., 2002a).Other works have reported that temperature affects the affinity of proteins and their diffusion rate on the HA surface.For example, Mohsen-Nia et al. (2012) observed that an increase in temperature enhanced the adsorption of human serum proteins to HA.These results strongly suggest that it may be possible to modulate the adsorption of MFGM on HA compared with other proteins by changing the environmental conditions.Nevertheless, to our knowledge, the preferential reactivity of the main buttermilk components toward HA has not been studied, nor have the parameters that could influence their adsorption.
The aim of this study was to evaluate the different affinity of proteins and MFGM with HA and the physical parameters which control their interaction.A quartz crystal microbalance with dissipation (QCM-D) was used to compare the adsorption of casein micelles, β-lactoglobulin (β-Lg), α-lactalbumin (α-La) proteins and MFGM fragments on HA particles.The QCM-D was used to provide qualitative and quantitative information about the adsorption kinetics and the quantity of proteins or MFGM fragments adsorbed onto HA coated on QCM-D gold sensors.Moreover, a Box-Behnken design was used to determine the effect of the physicochemical conditions of pH, ionic strength, and temperature on the adsorption of MFGM fragments and the main buttermilk proteins on HA.

Casein micelles extraction
The casein micelles were extracted from skim milk (Natrel).The skim milk was ultracentrifuged (Optima Iung et al.: Affinity of MFGM to hydroxyapatite XE-90, Beckman Coulter, Brea, CA) for 1h at 75,000 x g at 4°C.The casein micelle pellets were washed and redispersed at a protein concentration adjusted to 4 g/L using an Ultra-Turrax homogenizer (model T 25, IKA, Wilmington, SE) equipped with an S 25 N-18G spindle (IKA) for 1 min at 12 300 x g.Concentration was chosen according to Iung et al. (2023) (Iung et al., 2023).Then the colloidal CM suspension was stirred overnight at 10°C before use.

MFGM extraction
The MFGM was obtained from pasteurized (85°C for 30 s) cream.After a maturation at 10°C overnight, the cream was churned at 12°C by a pilot plant scale butter churn (Qualtech Equipment, Quebec City, Canada) to obtain buttermilk (BM).MFGM isolates were prepared following the method outlined by Corredig and Dalgleish (Corredig & Dalgleish, 1998).In summary, BM was treated with 2% (wt/vol) sodium citrate and incubated overnight at 4°C to facilitate the dissociation of casein micelles.Subsequently, the BM was centrifuged at 100,000 x g for 50 min at 15°C using an ultracentrifuge (Beckman Coulter, Optima XE-90, USA).The resulting pellet containing the MFGM, was collected, freeze-dried (Labconco, Lyph lock 4.5, Kansas City, MO, USA), and stored at −20°C.The MFGM isolate composition is provided in Table 1.

QCM-D analysis
For QCM-D measurements, a qCell T instrument equipped with a 10MHz AT-cut gold-coated quartz crystal with gold electrode sensors from 3T analytic (Tuttlingen, Germany) was used.For each experiment, a sensor with a freshly prepared HA coating was used.The HA coating was prepared with a HA suspension at a concentration of 5 g/L in distilled water.The HA suspension was sonicated for 15 min and left to stand for 30 min to ensure an optimal dispersion of the particles.The spin coating was prepared by depositing 40 µL of HA suspension on the top of the QCM sensor and spinning for 3 min at 5,000 x g at 20°C with a swing bucket centrifuge (Centrifuge 5804R from Eppendorf AG).This spin-coating step was repeated twice.The quantity of HA deposited on the sensor was around 360 µg (obtained by precise mass measurements of the sensor before and after spin coating), and the volume of the cell was 30 µL (given by operating instructions).Accordingly, the concentration of HA was 12 g/L.After that, the HA-coated sensor was dried under a UV lamp for 10 min and was placed under a nitrogen stream to remove the excess HA.Before and after experimentation, the sensors were washed with, in order, a 1.0 M HCl solution, distilled (DI) water, and ethanol.Then, the sensors were dried under a nitrogen stream before being placed under a UV lamp to remove all impurities.Before each use, all sensors were inspected for the absence of particles at their surface using a light microscope (model BX50, Olympus, Richmond Hill, ON, Canada) with a 40x objective (model BX50, Olympus).
The QCM-D method and analysis conditions were the same for all experiments.The flow rate was 100 µL/min, and the temperature was set at 21°C.Before the coating, the sensor was put in the QCM-D chamber, and calibration and a blank measurement were taken with Milli-Q water.After the coating, the sensor was mounted in the QCM-D cell, and a DI water flux was passed over the surface for 20 min to obtain a stable frequency signal.Then, the solution was switched with the analyte solution individually (MFGM, casein micelles, β-lactoglobulin, and α-lactalbumin) at 4 g/L until signal stabilization.Then, the sensor was again rinsed by flowing DI water over it to remove all unattached proteins from its surface.Finally, Q-tools software (3T analytic) was used to extract and analyze the data to calculate the adsorption rate (µg/m 2 /min).The attached mass density of the protein or MFGM adsorbed on the HA-coated sensor was then calculated according to the Sauerbrey equation: where n is the overtone number (8), c is the mass sensitivity constant (4.3 ng.cm −2 .Hz −1 ), Δf n is the overtone variation (5-50MHz), and Δm is the mass change per unit area of the surface adsorbate (ng/cm 2 ).

Box-Behnken experimental design
A 3-factors and 3-levels Box-Behnken design (BBD) was used to study the effects of different physicochemical parameters on protein and MFGM adsorption on HA.The 3 levels were low, medium, and high, coded as −1, 0, and +1, respectively.The 3 parameters selected were pH (5, 6.5, and 8), ionic strength (IS) (50, 100, and 150 mM NaCl) and temperature (T) (10, 20, and 30°C) as given in Table 2.The pH was selected to stay above the isoelectric points of the proteins.The ionic strength was chosen to frame the ionic strength of milk who was around 80mM (Augustin & Clarke, 1991), and the temperature was kept moderate to prevent excessive protein denaturation.All the proteins or the MFGM extract were redispersed overnight at 10°C at a concentration of 4 g/L.The HA concentration was 80 g/L, and the mixed HA-protein or HA-MFGM suspension were allowed to adsorb for 1 h under constant stirring.Then, the HA particles were recovered by centrifugation (200 x g during 5min), and the unattached proteins in the supernatant were measured.The BBD was analyzed using Design Expert 8.0.6.1 (Free trial, StatEase, Minneapolis, MN).
To evaluate the influence of these 3 parameters, 15 experiments for each protein and the MFGM extract were performed according to the experimental design matrix in Table 3.The effect of these 3 independent variables on the protein or the MFGM adsorption can be described by the following second-order polynomial equation: where, Y is the predicted response variable (% of unattached proteins), β 0 is the model constant, β i is the linear effect parameter, β ii is the quadratic effect of input factor X i , β ij is the first order interaction coefficients between input factors X i and X j and ε is the error of the model.

Determination of protein concentration
The protein concentration in the supernatants of all samples was measured with the Dumas combustion method (Rapid Micro N Cube Nitrogen analyzer, Elementar Analysensysteme GmbH, Langenselbold, Germany) using a nitrogen-to-protein conversion factor of 6.38.Before analysis, all the samples were dried in a vacuum drying oven at 30°C for 24h to concentrate them to be above the detection limit of the apparatus.

Determination of PL concentration in the MFGM isolate
A modified Folch procedure (Folch et al., 1957) was used to extract the PLs from the MFGM samples.In a test tube, 2 mL of chloroform and methanol (2:1) were mixed with 1 mL of sample, vortexed for 1 min, and centrifuged at 2,270 x g for 10 min.Then, 1 mL of the chloroform layer was removed with a low-volume syringe (Hamilton, Reno, NV) and put in a separate test tube.Finally, the excess chloroform was evaporated at room temperature under a stream of nitrogen.To determine PL concentrations, Stewart's protocol was used (Stewart, 1980).After preparation of an ammonium ferro-thiocyanate reaction solution (27.03 g ferric chloride hexahydrate and 30.4 g ammonium thiocyanate in deionized distilled water for a total volume of 1 L), the dried extract was dissolved in equal volumes (2 mL) of the reaction solution and chloroform.Then, the mixture was vortexed for 3 min and centrifuged at 2,800 x g for 5 min, and the lower chloroform phase was kept.The optical density of this phase was read at a wavelength of 488 nm with a spectrophotometer (Multiskan Spectrum model 1500, Thermolabsystem, Finland).The PL concentration was calculated as described by Iung et al. (2023).

Statistical analysis.
All experiments were performed in triplicates.The data were analyzed using a multifactor ANOVA and t-test using the SAS software (SAS University Edition, SAS Institute Inc., Cary, NC) to compare all data of the QCM-D analysis.Evaluations were based on a significance level of P < 0.05.For BBX, the statistics and repetitions were included in the model and the significance level is P < 0.05.

Adsorption of proteins and MFGM on HA-coated sensors by QCM-D
The affinity of each sample to HA was analyzed with a QCM-D with HA-coated sensors.For CM, the adsorption curve first showed a rapid increase in the adsorbed mass (2 min with 20.40 mg/m 2 ), followed by a slow increase (Figure 1 A).Confocal imaging was used to understand this 2-step adsorption behavior.This imaging revealed that the CM were adsorbed on the surface of the HA, which promoted particle aggregation Hydroxyapatite has been reported to induce the dissociation of CM, exposing their reactive phosphoserine sites and facilitating association with the C-site of HA (Tercinier, Ye, Anema, et al., 2014).β-lactoglobulin and α-lac showed similar adsorption profile curves characterized by a rapid increase between 0 to 4 min before reaching an equilibrium around 4 min (Figure 1 A).Moreover, the maximum amount adsorbed was almost the same (11.98 mg/m 2 and 14.49 mg/m 2 , respectively).The absence of any significant difference between these proteins suggests that they share the same affinity for HA (P > 0.05).As shown in the confocal images (Figure 2 A), the HA surface was covered by protein in both cases.Furthermore, an absence of HA aggregation was observed.This indicates that β-lg and α-lac only interacted with HA and not with themselves.These results demonstrate that unlike CM, β-lg and α-lac did not adsorb in a multilayer on the HA, but instead as a monolayer, which agrees with Tercinier et al. (17).
For the MFGM, a different adsorption pattern with 2 distinct phases was noticed (Figure 1).Initially adsorption was similar to CM, characterized by an augmentation of the adsorbed MFGM material until a first plateau was obtained at 420 min with 12.22 mg/ m 2 of MFGM attached (Figure 1 B1).Then, increased adsorbed material was observed with a second plateau (734 min at 21.73 mg/m 2 ) (Figure 1 B2).For the first phase, as observed on the confocal images (Figure 2 B1), proteins (in green) and PLs (in red) were initially adsorbed on the HA surface causing aggregation of the HA particles by promoting interactions between juxtaposing particles.Furthermore, this aggregation between several HA particles created interparticle cavities, trapping supplemental MFGM material, as reported in our previous work (Iung et al., 2023).For MFGM, a 2-step adsorption profile was observed, similar to CM.Following the first adsorption phase, the quantity adsorbed sharply increased at around 550 min ((Figure 1 B2), which could be the result of a reorganization of the adsorbed MFGM proteins at the surface, and the internalization of MFGM PLs inside HA particles, as can be seen in the confocal images (Figure 1 B2).As previously reported (Iung et al., 2023), this progressive displacement of PLs inside porous HA particles decreased the steric hindrance on the surface of the HA, which in turn allowed more MFGM to adsorb to this newly freed HA surface.Finally, the second plateau of the QCM-D adsorption curve corresponded to the maximum adsorbed equilibrium, corresponding to the saturation of HA with MFGM proteins and PLs at their surface and with PLs internally trapped in the HA pores.
The amount of protein adsorbed was different between whey proteins and the CM (Figure 1 A).There are major differences in molecular weights between whey proteins and CM, with α-lac and β-lg having molecular weights of 14.8 and 15.2 kDa respectively, and CM colloidal structures being between 100 and 300 nm (Wang et al., 2021).Size did not appear to drive these differences in affinity toward HA.Earlier studies by Tercinier et al. (2014) (17) reported higher adsorption of CM to HA compared with β-g and α-lac.Instead, these differences could be explained by their composition.Indeed, CM possess several phosphoserine groups forming clusters which have stronger interaction with HA's C-sites than the carboxyl groups found on the side chains of the amino residues of whey proteins (Juriaanse et al., 1981).This also explains why CM adsorbs faster to HA than the whey proteins (about 2 min for CM and 4 min for the whey proteins).However, in contrast to previous work we found a difference in the affinity of whey proteins for HA, with α-lac having greater affinity than β-lg.This difference among the whey proteins might be explained by the smaller size of α-lac (1.8 nm) compared with β-lg (2.4 to 3.19 nm for the monomer or dimer forms existing at neutral pH, respectively), allowing denser adsorption and closer α-lac packing on the HA surface (Aymard et al., 1996;Spöttel et al., 2021).The high sensitivity of QCM-D may also cause this discrepancy, as it allowed continuous monitoring of the adsorption of protein to the nearest nanogram, compared with the less sensitive batch adsorption method used by Tercinier et al. (2014) (17).Nevertheless, the difference between β-lg and α-lac affinity toward HA was negligible.
Similarly, CM adsorbed faster than MFGM (700 min for MFGM versus 2 min for CM), even though they were similar in terms of quantity adsorbed (P > 0.05) (Figure 1).It has been suggested in the literature that CM adsorbed to HA could be dissociated before or during adsorption (Tercinier, Ye, Anema, et al., 2014).Indeed, since HA can chelate ions, they can bind the micellar calcium, leading to the dissociation of the CM.In this way, the HA could better access the phosphoserine residues of free caseins in solution and in the partly dissociated CM.As MFGM fragments have large sizes (0.1-4 µm) and irregular shapes, their adsorption would be slower and less likely to occur as a result of these multiple steric constraints (Danthine et al., 2000).Furthermore, the presence of surface membrane proteins, particularly glycoproteins, could exacerbate the steric hindrance (Vanderghem et al., 2010).Following the internalization of the PLs after adsorption, these steric constraints would decrease, allowing the slow adsorption of more MFGM (Iung et al., 2023).That could be the reason why there was a difference in adsorption rate between MFGMs and the other major buttermilk proteins, with the final quantity adsorbed greater than for whey proteins but similar to CM.Overall, the preferential adsorption rate for the different buttermilk components tested on HA was in increasing order: CM > α-lac ; β-lg > MFGM.
Overall, these results show differences in the binding rate and affinity of the main buttermilk proteins and MFGM fragments, with CM having the highest affinity to HA, adsorbing faster than the other proteins.
Effect of pH, ionic strength (IS) and temperature (T) on the buttermilk components adsorption on HA.

Model determination
A BBD was used to determine the effects of varying physicochemical parameters (pH, IS, and T) on the affinity of the main buttermilk proteins and MFGM fragments for HA (Table 3).The BBD statistical model was first validated.The adjusted and predicted R 2 and the p-value were calculated to choose a model with the highest R 2 value according to Sahu et al. (2018).R 2 indicates the quality of a simple linear regression and is a measure of similarity of the model and the experimental data.For MFGM, β-lg, and α-lac, no statistical differences were found for any of the physicochemical parameters tested.MFGM and β-lg were completely adsorbed, whereas most α-lac was adsorbed (91%) to the HA particles.However, for the CM, the model showed that some of the environmental conditions tested had an impact on HA adsorption (Table 5).As shown in Table 4, a quadratic model had the higher R 2 and p-value CM adsorption on HA.Therefore, it was chosen for further statistical analysis.

Fitting of the second-order polynomial equation and statistical analysis
A second-order polynomial equation with interaction terms was used to express the empirical relationship between input variables and the experimental results, as given below.% = 13,45 + 3,59 x A +1,59 x B + 0.2503 x C + 2,42 x AB + 1,51 x AC + 1,84 x BC + 14,17 x A 2 -1,63 x B 2 + 2,11 x C 2 (3) The equation represents the percent (%) of unadsorbed CM following incubation with HA.ANOVA (ANOVA) was first performed to verify the model's accuracy.The results are reported in Table 5.The model and the associated terms were statistically significant (F-value of 262.88 and P < 0.0001) confirming that the equation obtained satisfactorily represents the real relationship between the input parameters and the responses since the regression equation explained the variation in responses (Zhang & Zheng, 2009).

Iung et al.: Affinity of MFGM to hydroxyapatite
Furthermore, a lack-of-fit test was performed (Table 5) to determine whether the chosen model fitted the data and predicted them properly (Fox & Weisberg, 2018).In this case, the test was not significant, confirming that the chosen model was adapted to the data and that the predictions were deemed accurate.
Finally, as shown in Table 6, the R 2 was high at 0.9979, and close to the adjusted R 2 of 0.9941, showing that the quadratic model explained 99.79% of the total variance and that just 0.21% was not explained.Since both results were close to a value of 1.0, we obtained a reasonable agreement between the experimental and predicted values.Moreover, the coefficient of variation (CV) was less than 10%, so the model was highly reliable and reproducible (Sahu et al., 2018).Lastly, as shown in Table 6, this model's adequate precision (Adeq precision) was 47.1724, which indicates an adequate signal and demonstrates that it was possible to navigate the design space with this model.The Adeq precision allowed us to measure the signal-to-noise ratio, since a ratio greater than 4 is desirable.

Effect of ionic strength, pH and temperature on CM adsorption on HA
As shown in Table 5, the pH, IS and T significantly influenced the percent of non-adsorbed CM on HA.Furthermore, there were also interactions between T and pH, T and IS, and pH and IS.These results show that variations in these parameters and interactions between them influenced the adsorption of CM to HA.
Figure 3 shows the response surface composed of the percent of non-adsorbed CM on the HA in relation to the different significant statistical interactions mentioned above.For each graph, the influence of 2 of the 3 parameters is shown, with the third parameter being fixed at its median value for ease of understanding. Figure 3A represents the significant interaction between temperature (T) and ionic strength (IS).The percent of unbound CM first decreased between 10°C and 20°C, before increasing between 20°C and 30°C.Between 10 to 30°C, a decrease in the percent of unbound CM was observed as the IS rose from 50 to 100 mM whereas above 100 mM, the opposite effect was observed.Figure 3B shows the effect of T and pH on the percent of unadsorbed CM.For T, a parabolic effect was again observed with the percent of unbound CM decreasing from 10°C and 20°C, then increasing from 20°C to 30°C.As the pH increased, the amount of unbound CM increased from acidic to neutral pH (pH 5 to 7) and then remained stable from neutral to basic (pH 7 to 8).The same effects for the interaction of pH and IS on the percent of unbound CM were observed, but the magnitude of the response was much lower than for the other interactions with smaller changes in unbound CM observed (Figure 3C).These variations can be explained by different mechanisms.
First, these results may be explained by the impact of pH and IS on HA surface charge.It has been reported that increasing IS by adding NaCl increases the zeta potential of HA from negative to positive between 0 and 100 mM (Anema, 2009;Yin et al., 2002b).This would explain the higher amount of adsorbed CM between 0 and 100 mM.A decrease in repulsion between the negative groups of the CM (mostly the phosphoserine groups) and the negatively charged P-sites of HA has been reported before (Tercinier, Ye, Singh, et al., 2014).However, an increase in IS also affects protein adsorption.At low IS, electrostatic repulsions between protein molecules or their negatively charged groups limit the quantity of proteins that can be bound to HA particles (Tercinier, Ye, Singh, et al., 2014).As the IS  increases, the charges on the protein side chains will be protected, reducing the electrostatic repulsion between the adsorbed proteins (Kandori et al., 2004).This may result in denser packing of the adsorbed proteins, explaining the reduction of unbound CM observed from 0 to 0.1 M (Nakanishi et al., 2001).Nevertheless, high salt concentrations (generally >0.1 M) can modify protein conformation and solubility (Zhang et al., 2004).This is the "salting out effect," which can reduce protein adsorption by restricting access and visibility of charged sites (Bazinet et al., 2001).Second, the zeta potential of HA has been reported to increase inversely with pH (Kawasaki, 1991;Tercinier, Ye, Singh, et al., 2014;Wassell et al., 1995).According to the literature (Tercinier, Ye, Singh, et al., 2014;Wassell et al., 1995), the zero-charge point of HA is around pH 6.4 to 8.0.The zeta potential would therefore be positive above this point and negative below.A decrease in pH would result in protonation of HA phosphate ions, increasing the zeta potential (Zhu et al., 2007).In addition, electrostatic repulsions between negatively charged protein groups and HA p-sites would favor protein adsorption (Zhu et al., 2007).This would explain the decrease of unbound CM while decreasing the pH from pH 7.0 to 5.0 despite being close to the isoelectric point (pH 4.6) (Holt & De Kruif, 2003).In fact, the net charge of the proteins decreases as the pH approaches the isoelectric point and potentially reduces HA binding.On the other hand, it reduces electrostatic repulsions and protonates the amine groups of phosphoserines, enabling them to be denser and to bind to P-sites.However, self-aggregation of CM (with free CM or CM binding on HA) will likely happen due to the decrease in CM solubility, resulting in a larger amount of CM being bound to HA (HadjSadok et al., 2008).From pH 7.0 to 8.0, the decrease in unbound CMs could be due to an increase in CM negative charge, favoring interaction with HA C-sites.However, static repulsions between the negatively charged proteins and with the HA P-sites, will likely reduce the density of adsorption.This would provide some explanation as to why, despite an increase in the net charge of CMs, the quantity of unbound CM remains stable.Moreover, the pK a s of the phosphoserines residues of the caseins are around 6.4 to 7.2 (Baumy et al., 1989).The protonation of phosphoserine residues in this pH range could lead to an increase in electrostatic repulsions and reduced binding of CM to HA particles (Tercinier, Ye, Singh, et al., 2014), as well as their packing density on the HA surface.Liu et al. (2013) (48) observed that temperature affects CM average size.These authors found that CM size changed with temperature and a polydispersity was observed, with the CM average size being at its lowest at 20°C and highest at 30°C.At 10°C, the CM average size was closer to the size at 30°C than at 20°C.These results suggest that the CM steric hindrance was probably lowest at 20°C, allowing more CM to adsorb on the surface of the HA.Furthermore, these size variations can be explained by micelle structural reorganization effects linked to the inhibition of hydrophobic bonds at low temperatures, also affecting the adsorption of CM on HA.In fact, around 10°C, β-casein is well-known to dissociate from the CM, along with the micellar calcium phosphate in the serum phase (McMahon & Oommen, 2008;Post et al., 2012), and this could lead to a decrease in CM size at lower temperature in water (Beliciu & Moraru, 2009).On the other hand, at 30°C mineralisation of the CM is observed through the presence of more colloidal CaPO 4 (Croguennec et al., 2008).Furthermore, at 30°C, an increase in CM mineralization would inhibit their active phosphoserine sites.The interaction of pH with IS could therefore be due to the known demineralization of the CM micelles via the addition of NaCl, resulting in a decrease in pH (Huppertz & Fox, 2006).Finally, the interaction of IS  20), on the amount of unadsorbed casein micelles after incubation with a suspension of HA particles for 60 min.The solution of CM was at 4g/L and the suspension of HA particles at 80g/L.To separate the HA and CM adsorbed to it from unadsorbed CM, the suspension was centrifuged at 200 x g during 5min.To obtain the amount of non-adsorbed CM, the supernatant was analyzed using the Dumas combustion method (with a conversion factor of 6.38) after the sample had been concentrated by vacuum drying at 30°C for 24 h.and T could play a role in the release of ionic calcium from the CM by increasing temperature and decreasing CM adsorption (Figure 3 A) (Huppertz & Fox, 2006).
These results imply that the adsorption of CM could be modulated as a function of different physical parameters without affecting the adsorption of MFGM, and this could be used to separate these 2 entities.

CONCLUSION
A QCM-D was used to characterize the adsorption rate and affinity between different buttermilk proteins and MFGM fragments for HA.Our results showed a higher affinity for the CM than the MFGM.Furthermore, the amount of adsorbed material was maximal for the MFGM and the CM followed by the whey proteins.We determined the influence of different physicochemical parameters on the affinity of the individual proteins against HA, finding that temperature, pH and IS affect the adsorption of CM to HA but not the other buttermilk components tested.The study's findings reveal that, despite CM exhibiting faster adsorption kinetics than MFGM, manipulating physicochemical parameters enables the adjustment of CM adsorption on HA.In contrast, MFGM adsorption remains unaltered under the same range of physicochemical conditions.Therefore, it can be inferred that in a mixture, it would be potentially feasible to immobilize our MFGM without affecting the CM by carefully adjusting the physicochemical parameters used for their adsorption.The buttermilk, after removing its lipid content, can be used as skim milk which has excellent techno-functional properties.Additionally, the MFGM extract can be utilized as a bioactive compound in various fields, including the production of food products for infant nutrition.
Iung et al.: Affinity of MFGM to hydroxyapatiteTable2. Box-Behnken experimental design parameters tested and their respective levels used to determine the impact of the environmental conditions on the adsorption of milk proteins and MFGM These results suggest that CM adsorbed not only to the surface of the HA particles but also to the CM already adsorbed onto HA.Tercinier et al.  (2013) (17)  demonstrated that the CM were affected by the heterogeneous surface of the HA particles and their adsorption was multi-layered.A multi-layered adsorption indicates that the CM interacted directly with HA but also with other CM already adsorbed at the HA surface.Eventually, this aggregation between several HA particles creates interparticle cavities which can trap supplemental CM material(Iung et al., 2023).

Figure 1 .Figure 2 .
Figure 1.Determination of buttermilk main proteins (A) and milk fat globule membrane fragments (B) attached mass density on the HAcoated gold sensor by means of quartz crystal microbalance diffusion at 21°C and at a flux of 100µL/min, without changing pH and IS.The solutions of proteins and MFGM were at 4g/L and were each individually run over a new sensor.One and 2 indicate the different saturation plateaus of the MFGM adsorption on HA.
Iung et al.: Affinity of MFGM to hydroxyapatite Table 3. Coded factor levels for Box-Behken design matrix for the three variables (pH, T and IS) temperature, and IS = Ionic strength.

Figure 3 .
Figure3.Response surface plots showing the effects of: (A) temperature (T) and ionic strength (IS) with pH set to the coded value 0 (6.5), (B) T and pH with IS set to the coded value 0 (100mM), (C) pH and IS with T set to the coded value 0 (20), on the amount of unadsorbed casein micelles after incubation with a suspension of HA particles for 60 min.The solution of CM was at 4g/L and the suspension of HA particles at 80g/L.To separate the HA and CM adsorbed to it from unadsorbed CM, the suspension was centrifuged at 200 x g during 5min.To obtain the amount of non-adsorbed CM, the supernatant was analyzed using the Dumas combustion method (with a conversion factor of 6.38) after the sample had been concentrated by vacuum drying at 30°C for 24 h.

Table 4 .
Box Behnken design statistical model summary for the impact of temperature, pH, and ionic strength on the interaction between the casein micelles and hydroxyapatite

Table 5 .
Iung et al.:Affinity of MFGM to hydroxyapatite ANOVA results of the quadratic regression model for the impact of physicochemical parameters influence on the CM adsorption to HA