The β-casein ( CSN2 ) A2 allelic variant alters milk protein profile and slightly worsens coagulation properties in Holstein cows

Milk proteins genetic variants have long attracted interest as they are associated with important issues relating to milk composition and technological properties. An important debate has recently opened at an international level on the role of β-casein (β-CN) A1 and A2 polymorphisms, toward human health. For this reason, a lot of efforts has been put into the promotion of A2 milk by companies producing and selling A1-free milk, leading the farmers and breeders to switch toward A2 milk production without paying attention on the potential effect of the processability of milk into cheese. The aim of the present work was to evaluate the effects of β-CN, specifically the A1 and A2 allelic variants, on the detailed milk protein profile and cheese-making traits in individual milk samples of 1,133 Holstein Frie-sian cows. The protein fractions were measured with reversed-phase (RP)-HPLC (expressed in g/L and % N), and the cheese-making traits, namely milk coagulation properties, cheese yield, and curd nutrient recoveries assessed at the individual level, with a nano-scale cheese-making procedure. The β-CN ( CSN2 ), κ-CN ( CSN3 ), and β-lactoglobulin ( LGB ) genetic variants were first identified through RP-HPLC and then confirmed through genotyping. Estimates of the effects of protein genotypes were obtained using a mixed inheritance model that considered, besides the standard nuisance variables (i.e., days in milk, parity, and herd-date), the milk protein genes located on chromosome 6 ( CSN2 , CSN3 ) and on chromosome 11 ( LGB ), and the polygenic background of the animals. Milk protein genes ( CSN2 , CSN3 , and LGB ) explained an important part of the additive genetic variance in the traits evaluated. The β-CN A1A1 was associated with a significantly lower production of whey proteins, particularly of β-lactoglobulin (−8.2 and −6.8% for g/L and % N, respectively) and α-lactalbumin (−4.7 and −4.4% for g/L and % N, respectively), and a higher production of β-CN (6.8 and 6.1% for g/L and % N, respectively) with respect to the A2A2 genotype. Regarding milk cheese-making ability, the A2A2 genotype showed the worst performance compared with the other genotypes, particularly with respect to the BA1, with a higher rennet coagulation time (7.1 and 28.6% compared with A1A1 and BA1, respectively) and a lower curd firm-ness at 30 min. Changes in milk protein composition through an increase in the frequency of the A2 allele in the production process could lead to a worsening of the coagulation and curd firming traits.


INTRODUCTION
Milk protein composition is an important factor in the nutritional and technological characteristics of milk.The quantities and proportions of milk caseins and whey proteins, in particular, have been identified as playing a major role in milk coagulation and curd firming processes (Caroli et al., 2009;Bittante et al., 2012), which are of particular economic importance in the Italian dairy sector, as most of the country's milk is destined for high-quality cheese production.Milk protein polymorphisms are the result of SNP, nucleotide insertion or deletion, or post-translational modification (Caroli et al., 2009).These have been extensively studied in recent decades as they have a significant effect on milk yield, the protein profile, and functional properties, being responsible not only for changes in the quantities and proportions of the protein components of milk (Amalfitano et al., 2020) but also affecting

The β-casein (CSN2) A2 allelic variant alters milk protein profile and slightly worsens coagulation properties in Holstein cows
milk coagulation properties (MCP; Jõudu et al., 2008;Caroli et al., 2009;Cipolat-Gotet et al., 2018).For example, different studies have found that the genetic polymorphisms of κ-CN have a substantial effect on the milk coagulation process, with the B variant producing a higher amount of κ-CN and smaller casein micelles, reducing curd firming time and increasing whey expulsion compared with the A variant (Bittante et al., 2012;Gambra et al., 2013).
One of the most important milk protein fractions is β-CN, which accounts for almost one-third of the total milk protein content (Huppertz, 2013).It can be found in different genetic variants, although the most common are A1 and A2 (Guantario et al., 2020).The variant A2, which is the ancestral variant shared with other wild and domestic species related to cattle, differs from A1 only in the substitution in position 67 of a Pro (A2) with a His (A1) residue (Summer et al., 2020).In recent years, the scientific community has started to pay more attention to β-CN polymorphisms due to possible effects of the A1 variant on human health, related to its production of β-casomorphin-7 during protein digestion.Indeed, Truswell (2005) thoroughly reviewed the claims related to the hypothesis that β-CN A1 could be a risk factor for type 1 diabetes and coronary heart disease and concluded that there is no substantial evidence for stating that the consumption of A1 milk can be associated with these adverse effects on human health.Nevertheless, Pal et al. (2015) concluded that the release of β-casomorphin-7 had effects on gastrointestinal motility and proinflammatory and immunomodulatory outcomes, in both in vitro models and in vivo animal experiments.
Several studies have extensively assessed the effects of milk protein genetic variants on the composition and protein profile of milk (Gustavsson et al., 2014;Amalfitano et al., 2020) and its coagulation properties (Bonfatti et al., 2011;Vallas et al., 2012;Amalfitano et al., 2019).In this context, the studies carried out by Amalfitano et al. (2019) and Bonfatti et al. (2011), considered Brown Swiss and Simmental breeds respectively, having a shift in allele distribution toward A2, which might affect the estimates of the effects of the alleles on MCP.Moreover, only a few studies considered the additive genetic background of the animals (Bonfatti et al., 2011).
Although the potential nutritional and biological differences between A1A1 and A2A2 milk are still being debated (Summer et al., 2020), some countries are seeing a rapidly increasing trend in the consumption of A2A2 milk (Mencarini et al., 2013;Nguyen et al., 2015), driving farmers and breeding associations to increase the frequency of the β-CN A2 allele in dairy cattle populations without paying attention on the potential effect of this change on milk technological characteristics.To the best of our knowledge, no studies have focused on the effects of β-CN A1 and A2 alleles on protein profile and specific curd firming, cheese yield, and curd recovery traits at the individual level.Therefore, the aim of this work was to evaluate, for the first time, the effects of β-CN, and in particular the A1 and A2 alleles, on (1) milk yield and composition, (2) milk detailed protein profile expressed both quantitatively (g/L) and qualitatively (% N), and (3) 18 technological traits, including MCP, curd-firming traits (CF), cheese yields (CY), and curd nutrient recoveries (REC), in a population of 1,133 Holstein cows, using a mixed inheritance model taking into account β-CN, κ-CN, and β-LG protein genotypes (encoded by CSN2, CSN3, and LGB genes, respectively) together with the additive polygenic background of the animals.

Field Data and Sample Collection
The present study was part of the AGER project FARM-INN (Farm-level interventions supporting dairy industry innovation; AGER 2 Project under Grant no.2017-1130).Individual milk samples were collected from 1,133 Holstein Friesian cows reared in 5 different herds in northern Italy.Milk samples were collected once from all animals during the evening milking between May 2019 and February 2021 (2 or more sampling days per herd, depending on the farm size; 21 herd-date combinations in total).The cows in all herds were housed in freestalls and fed TMR based mainly on corn silage, sorghum silage, and concentrates.Diets were formulated to meet nutritional requirements as recommended by the NRC (2001).Drinking water was available in automatic water bowls, and the cows were milked twice a day.Three of the herds were located in the production area of Grana Padano Protected Designation of Origin hard cheese.Animals with clinical signs of disease or under medical treatment were not sampled.
After collection, one subsample of milk (50 mL) from each cow (with bronopol preservative added) was sent to the Milk Quality Laboratory of the Veneto Region Breeders' Association (ARAV, Padua, Italy) for composition analysis.One subsample (100 mL) per cow was taken to the Cheese-Making Laboratory of the Department of Agronomy, Food, Natural Resources, Animals and Environment (DAFNAE) of the University of Padua (Legnaro, Padua, Italy) for analyses of milk technological traits.An additional 2-mL aliquot was taken and stored at −80°C within 3 h of sampling for analysis of the milk protein fractions.Milk composition and technological traits were measured within 48 h of sampling, whereas milk protein fractions were analyzed within 4 mo.Pedigree information was supplied by the Italian Holstein-Friesian and Jersey Cattle Breeders Association (ANAFIJ, Cremona, Italy) and included 6 generations of known ancestors for each cow sampled.The pedigree file included 5,773 animals.

Analysis of Milk Composition and Cheese-Making Traits
Milk composition (fat, protein, casein, lactose, and urea contents) was measured with a Milkoscan FT6000 infrared analyzer (Foss Electric A/S).Two mechanical lactodynamographs (Formagraph, Foss Electric A/S) were used to measure the following traditional MCP in duplicate as proposed by Stocco et al. (2017): (1) rennet coagulation time (RCT; min), the time from rennet addition to the start of coagulation; (2) curd firming time (k 20 ; min), the time to reach a curd firmness of 20 mm; (3) curd firmness at 30, 45, and 60 min after rennet addition (a 30 , a 45 , a 60 ; mm).In addition, the width of the coagulation was recorded by the instruments every 15 s for 60 min, for a total of 240 measurements per milk sample.These 240 measurements were used in the modeling equation proposed by Bittante et al. (2013) to estimate the following curd firming and syneresis parameters (CF t ): RCT eq (min) is RCT estimated using the individual curd firming equations; k CF (%min) is the curd firming instant rate constant; k SR (%min) is the curd syneresis instant rate constant; CF max (mm) is the maximum curd firmness reached within 45 min; and t max is the time needed to reach CF max .Repeated measures were averaged before the statistical analysis.
Cheese yield and milk nutrient recoveries were calculated according to the 9-Milca method proposed by Cipolat-Gotet et al. (2016), and the procedure was carried out in duplicate.Briefly, it consisted in adding 200 µL of rennet solution (Hansen Standard 215 with 80 ± 5% chymosin and 20 ± 5% pepsin; Pacovis Amrein AG) diluted to 1.2% (wt/vol) with fresh distilled water to 9 mL of milk heated to 35°C in a glass tube.The cheesemaking procedure was carried out as follows: after an initial incubation step of 30 min at 35°C, the first cut was made with a stainless steel spatula; this was followed by a curd-cooking phase of 30 min at 55°C, during which a second manual cut was made; after the heating phase, the curd was separated from the whey for 30 min at room temperature, applying gentle pressure to the curd to better drain the whey.Lastly, whey composition (fat, protein, lactose, and TS) was measured with an FT2 infrared spectrophotometer (Milkoscan FT2; Foss Electric A/S).This procedure enabled us to obtain 7 cheese-making traits.The first 3; namely, curd weight (CY CURD ), curd dry matter (CY SOLIDS ), and water retained in the curd (CY WATER ), were expressed as percentages of the total milk processed.The other 4 traits, REC PROTEIN , REC FAT , REC ENERGY , and REC SOLIDS (%), representing milk nutrient recoveries in the curd, were measured by computing the difference in weight and composition between the milk and the whey, according to Cipolat-Gotet et al. (2013).Repeated measures were averaged before statistical analysis.

Protein Fraction Identification
The contents of κ-CN, α S1 -CN, α S2 -CN, and β-CN, β-LG, α-LA, and lactoferrin (LF) in the individual milk samples were measured using the validated RP-HPLC method proposed by Maurmayr et al. (2013), which also enabled discrimination of the β-CN, κ-CN, and β-LG genotypes.Cows with the following combinations of genetic variants were identified: (1) AA, AB, and BB κ-CN; (2) A1A2, A1A1, A2A2, BA1, BA2, and I/H2 β-CN (which cannot be distinguished due to equal elution times); and (3) AA, AB, and BB β-LG.We were also able to quantify the proportions of glycosylated and nonglycosylated κ-CN.Total κ-CN was calculated by summing the 2 peaks.A detailed description of the method can be found in Maurmayr et al. (2013).Briefly, the HPLC equipment was an Agilent 1260 Series chromatograph (Agilent Technologies) coupled with an Agilent 1260 Series G1311B quaternary pump.A C:8 reversed-phase analytical column (Aeris WidePore XB-C8, Phenomenex) was used to separate the protein fractions.All protein fractions were then expressed in 2 ways, as reported in detail by Amalfitano et al. (2020).Briefly, quantitative measures (g/L) were calculated by multiplying the proportion of each protein fraction obtained from the HPLC analysis by the milk casein contents obtained by FTIR, and qualitative measures of the protein fractions were calculated as their percentages of the total nitrogen content (%N), assuming to be 15.67% the N content of milk proteins and 46.67% of that of urea.

Genotyping
To validate the HPLC identification of the β-CN (CSN2) and β-LG (also known as PAEP, progestogenassociated endometrial protein) genotypes, we outsourced to LGC Biosearch Technologies (LGC group) the genotyping of informative SNPs in all samples: for allele calling using automatic allele calling option and viewed using the SNPviewer software (LGC Biosearch Technologies).

Statistical Analysis
Quality control of the data on all the investigated traits was performed to identify outliers, and any falling outside the interval of the mean ± 3 standard deviations were excluded from the analyses.Hardy-Weinberg equilibrium of the CSN2, CSN3, and LGB loci was assessed by chi-squared test in the genetics package of the R software version 4.1.2(https: / / www .r-project .org).
The effects of protein genotypes on the phenotypic variation of milk protein fractions (expressed as both g/L and %N), and for cheese-making traits (traditional MCP, CF, CY, and REC) were estimated using the following single-trait animal model in the lme4qtl package of the R software (Ziyatdinov et al., 2018): where y ijklm is a measure for a trait; is the additive genetic variance and A is the additive relationship matrix between individuals (Wright, 1922; the pedigree file contained data on 5,773 animals); and e ijklm is a random residual assumed to follow a normal distribution with e N ijklm e ~, , 0 2 σ ( ) where σ e 2 is the re- sidual variance.
To evaluate the specific contribution of CSN2, CSN3, and LGB genotypes in explaining the total genetic variance for the investigated traits, a mixed inheritance model was fitted, first by adding one genotype at a time as a fixed effect in M1, and then with the inclusion of all genotypes together (M2).Therefore, the second model was: where y ijklm is a measure of a trait; CSN2 k is the fixed effect of the kth class of CSN2 genotype (6 levels: A1A2, A1A1, A2A2, BA1, BA2, other); CSN3 l is the fixed effect of the lth class of CSN3 genotype (3 levels: AA, AB, BB); LGB m is the fixed effect of the mth class of LGB genotype (3 levels: AA, AB, BB).All the other terms were as previously defined.
The effects of the different milk protein genotypes on the traits evaluated were compared using the lsmeans package in R (Lenth, 2016) and Tukey's test (P < 0.05).A tendency was also considered if 0.05 ≤ P < 0.10.

Allele and Genotype Frequencies
Distributions of the allele and genotype frequencies of CSN2, CSN3, and LGB are reported in Table 1.The most frequent CSN2 genotype was the heterozygote A1A2 (0.43), followed by A2A2 (0.33) and A1A1 (0.14).The "other" genotype category included animals for which complete identification of the genotype was not possible, and accounted for 0.06 of the frequency distribution.The A allele of CSN3 was the predominant one, leading to a distribution shift toward the AA genotype (0.49, 0.44, and 0.07 for AA, AB, and BB, respectively).An approximately even frequency distribution was found between the A and B alleles of LGB (0.54 and 0.46, respectively).
No extreme deviation from Hardy-Weinberg equilibrium was detected (P > 0.001, data not shown) for the computed CSN2, CSN3, and LGB genotype frequencies.

Descriptive Statistics
Descriptive statistics for milk yield, composition, and quantitative (g/L) and qualitative (% N) protein profiles are reported in previous work carried out on the same herds and cows (Pegolo et al., 2021).The predominant casein fractions were α S1 -CN and β-CN, with averages of 9.04 g/L (26.38%N) and 8.99 g/L (26.30%N), respectively, whereas α S2 -CN (3.47 g/L, 10.12% N) and κ-CN (5.25 g/L, 15.30% N) were less represented.The predominant whey protein was β-LG, with a mean value of 3.46 g/L (10.05%N), followed by α-LA (1.00 g/L, 2.92% N) and LF (0.16 g/L, 0.48% N).Overall, the mean values for both the quantitative (g/L) and qualitative (% N) measures were similar to those found by Amalfitano et al. (2020) and Maurmayr et al. (2018), with the exception of total caseins and β-CN, which were lower than those reported by Amalfitano et.al (2020) (28.36 g/L and 78.49% N, 10.69 g/L and 29.69% N, respectively), and α S2 -CN, which was instead slightly higher (2.78 g/L and 7.2% N).These differences could be due to several factors, such as differences in population size and breed, but also analytical method (Bonfatti et al., 2011;Gebreyesus et al., 2016).
Table 3 displays the descriptive statistics for cheesemaking traits.Regarding coagulation properties, we obtained an average RCT of 22.74 ± 7.42 min, curd firmness at 20 mm of 7.87 ± 3.95 min, and curd firmness at 30 min of 20.22 ± 14.40 mm.Mean values of the modeled CF parameters were: 23.04 min for RCT eq calculated from all 240 data points, 46.04 mm for CF p , 8.69%/min for k CF (curd-firming instant rate constant) and 0.69%/min for k SR (syneresis instant rate constant).The average value of CF max was 34.36 mm, and was obtained at 50.01 min after rennet addition (t max ).
Regarding cheese yields and curd nutrient recoveries, the mean value of CY CURD was 20.54%, CY WATER 6.21%, and CY SOLIDS 13.83%.The average values for REC PROTEIN , REC FAT , REC SOLIDS, and REC ENERGY were 79.10, 79.76, 47.61, and 63.10, respectively.These re-

Variance Components of Milk Composition and Protein Profile
Estimates of variance components for milk yield, composition and detailed protein profile obtained from the animal model (including or without the effects of the CSN3, CSN2, and LGB genotypes) are reported in Table 4.
The contribution of herd/date in explaining the proportion of phenotypic variation was particularly relevant in the case of urea, which is well known to be affected by diet (Schiavon et al., 2015).In agreement with Schopen et al. (2009) and Bonfatti et al. (2011), the herd/date variance in milk protein fractions was very low, confirming the negligible influence of farming and feeding conditions on milk protein composition.
The estimated animal additive genetic variance decreased significantly when the effects of the 3 major milk protein genes were taken into consideration, a pattern that was particularly evident for milk protein fractions (Table 4).When the protein fractions were expressed as g/L, the caseins in which the milk protein genes explained most of the total variance were glycosylated κ-CN (71%) and κ-CN (59%), followed by α S1 -CN (29%) and β-CN (22%).The most significant differences in genetic variation in the whey proteins were for total whey proteins (71%) and for β-LG (76%).
The contribution of the milk protein genes in explaining the genetic variances was even higher when the protein fractions were expressed as percentages of total N, and this was particularly noticeable for the most abundant protein fractions (the proportion of genetic variance explained ranged from 55 to 78% for α S1 -CN, β-CN, and β-LG).The protein fractions expressed qualitatively (% N) better highlighted the trend in the variation in the proportions of the single protein fractions out of the total protein content, whereas the quantitative data (g/L) did not appear to be as informative.
Furthermore, in our study milk protein genes explained over 62% of the genetic variance in α S1 -CN, in agreement with Bonfatti et al. (2011;84%), and in contrast with Schopen et al. (2009) who reported that the CSN2, CSN3, and LGB genotypes had no effect on the polygenic variance in α S1 -CN.This difference may be due to the way in which the protein fractions are expressed, whether as percentages of the total casein content, as in the present work and Bonfatti et al. (2011), or as percentages of the total protein fractions, as in Schopen et al. (2009).
Looking at the proportions of additive polygenic variance explained by the single-locus genotypes on individual proteins, we observed that CSN2 was the only one explaining the genetic variance in α S2 -CN, in agreement with Schopen et al. (2009).This finding might be explained either by the fact that in the tight 250-kb cluster in Chromosome 6 CSN2 and CSN1S2 (α S2 -CN) are in a considerable linkage disequilibrium (Huang et al., 2012), or by a possible homology among the promoter region of all the Ca 2+ -sensitive casein genes [CSNS1 (α S1 -CN), CSNS2, and CSN2; Groenen et al., 1993], and therefore, high correlations in their expression among these caseins might be expected.
Moreover, CSN2 contributed to the genetic control of both total κ-CN and its glycosylated fraction, even though, as expected, the major part of this variability was explained by the κ-CN genotype itself.In agreement with Bobe et al. (1999), both κ-CN and β-CN, together with β-LG genotypes had an effect on the α S1 -CN genetic variance.With regard to the whey proteins, β-LG was, as expected, significantly genetically controlled by the LGB gene, as previously reported by Bobe et al. (1999) and Schopen et al. (2009).Finally, it is interesting to note that the CSN2 gene explained most of the genetic variance in α-LA and LF and further studied will be needed for unravelling the molecular pathways of this contribution.
Overall, in agreement with Bonfatti et al. (2011) and Schopen et al. (2009), we found that milk protein genes explained a considerable part of the genetic variance in the milk proteins.However, there is still a residual genetic variance controlled by other loci elsewhere in the genome that might contribute to regulation of the relative proportions and characteristics of milk proteins (Gustavsson et al., 2014;Dadousis et al., 2017).

Variance Components for Milk Cheese-Making Traits
The estimates of variance components for MCP and cheese-making-related traits (including or without the effects of κ-CN, β-CN, and β-LG genotypes) are reported in Table 5.Most of the genetic variance in these traits was explained by CSN3; CSN2 described the greater portion of variance in both t max (27%) and k cf (35%).With regard to CY, CSN2 together with CSN3 contributed to explaining a moderate portion of the genetic variance in CY CURD (6 and 11%, respectively) and CY SOLIDS (5 and 10%, respectively).The LGB genotype did not seem to play a significant role in explaining the genetic variance in the traditional MCP, CF, and CY traits, but made a limited contribution to CY SOLIDS (4%).On the other hand, among the REC traits, it was responsible for explaining, together with CSN3, most of the genetic variance in REC PROTEIN (86%), and a small portion in REC ENERGY (5%).

β-CN A1 and A2 Variants Exert Different Effects on Milk Protein Fractions and Coagulation Traits
The results of the ANOVA for milk yield, composition, and protein fractions are reported in Table 6, whereas the ANOVA results for traditional MCP, CF, CY, and REC traits are reported in Table 7.
As expected, DIM and parity were confirmed as important sources of variation in the traits evaluated (Amalfitano et al., 2020;Pegolo et al., 2021), supporting the importance of including them in the statistical model to better assess the effects of the CSN2, CSN3, and LGB genotypes.
Regarding protein composition, we found that the β-CN variants had significant effects on milk proteins by altering the quantities and relative proportions of them.Interestingly, CSN2 did not have any significant effect on total caseins, whether measured quantitatively or qualitatively, but significantly affected single casein fractions (P < 0.001 for α S1 -CN, α S2 -CN, and β-CN; P < 0.01 for κ-CN-gly), with the exception of κ-CN, only in the case of g/L (P < 0.05 when expressed as percentage of total N).Moreover, it had a significant effect on total whey proteins (P < 0.001), β-LG (P < 0.001), α-LA (P < 0.001), and LF (P < 0.05 for g/L, P < 0.01 for %N).
Figure 1 shows the trend in the expression of the single protein fractions that were significantly affected by the 6 CSN2 genotypes identified, both quantitatively (Figure 1A) and qualitatively (Figure 1B).As far as the casein fractions are concerned, the A2A2 genotype, together with BA2 and "other" genotype were the ones that produce the most α S2 -casein.Conversely, BA1 and A1A1 produced the lowest concentration of α S2 -CN.No significant differences between A1A1 and A2A2 were found in respect to the production of α S1 -CN, which were significantly higher than in the BA1 genotype.Moreover, the production of β-CN was significantly reduced in samples with the A2A2 genotype, and was in the highest concentrations with the BA1 genotype.Because there were appreciable differences in the quantities and proportions of the single casein fractions, but not of the total caseins, a possible hypothesis is that when the expression of one CN gene is downregulated the expression of others is upregulated in compensation (Leroux et al., 2003).
The β-CN genotype A2A2 was responsible for the highest production of TP.Looking at the different proportions of the single protein fractions, however, we found that it was associated with higher whey protein concentrations, and, more specifically, with higher β-LG and α-LA, and lower β-CN concentrations.The opposite trend was observed in the case of the β-CN A1A1 genotype, and even more clearly in the case of the BA1 genotype, with lower TP, β-LG and α-LA concentrations and higher β-CN concentrations.
The only whey protein that exhibited an opposite pattern to the one previously described was LF, which is an iron-binding protein with antimicrobial activities that seems to also play a role in the development of the microbiota and immune system of newborn babies (Giansanti et al., 2016).The highest concentration was RCT = rennet coagulation time; k 20 = curd firming rate as the time to a curd firmness of 20 mm; a 30 = curd firmness at 30 min from rennet addition; RCT eq = rennet coagulation time estimated using the equation; CF P = asymptotic potential curd firmness; k CF = curd firming instant rate constant; k SR = syneresis instant rate constant; CF max = maximum curd firmness achieved within 45 min; t max = time at achievement of CF max ; %CY CURD = weight of fresh curd as percentage of weight of milk processed; %CY SOLIDS = weight of curd solids as percentage of weight of milk processed; %CY WATER = weight of water curd as percentage of weight of milk processed; REC PROTEIN = protein of the curd as percentage of the protein of the milk processed; REC FAT = fat of the curd as percentage of the fat of the milk processed; REC SOLIDS = solids of the curd as percentage of the solids of the milk processed; REC ENERGY = energy of the curd as percentage of energy of the milk processed.
observed with the homozygote A1A1 (Figure 1A and  B).Although the overall concentration was low, an interesting line of further study would be to explore in greater depth their putative relationship and pathway regulation.
Changes in the relative proportions of the protein fractions are directly linked to changes in MCP (Guinee, 2003).As such, an important role is played by α S1 -CN, which various authors have considered a "molecular detergent" due to its ability to reduce the amount of either β-CN or κ-CN (Farrell et al., 2006).This pattern was particularly noticeable in the case of the β-CN A2A2 genotype and its effects on α S1 -CN and β-CN production (Figure 1A and B).The β-CN A1A1 genotype was associated with the lowest amount of α S2 -CN (Figure 1A and B).Among the casein fractions, α S2 -CN seems to have little influence on the efficiency of the cheese-making process (Cipolat-Gotet et al., 2018) or even to negatively affect milk protein recovery in the curd (Amalfitano et al., 2019).
Moving on to MCP (Table 7), CSN2 significantly affected RCT and a 30 (P < 0.01).Among the CF, CY and REC traits, CSN2 significantly affected RCT eq , t max , k CF , k SR (P < 0.001), and CY WATER (P < 0.05).More specifically, A2A2 milk exhibited higher RCT, RCT eq , t max , and CY WATER , but lower a 30 , k CF , and k SR compared with the other genetic variants (Figure 2).Regarding RCT, even if the only significant difference was found between β-CN A2A2 and BA1 it is noteworthy that the average RCT for A1A1 milk was 20.8 min, almost 2 min shorter than A2A2 (22.4 min).Tendencies have been observed between A1A1 and A2A2 genotypes in respect to RCT eq and k CF (P < 0.10).To better observe the specific behavior of A1 and A2 alleles on MCP we combined the CF traits in a graphic form to build a modeling curve to see the specific behavior of the A1 and A2 alleles on milk technological properties (Figure 3A).This approach allowed us to better capture the slightly impairment of the coagulation performances of A2A2 milk in respect to the A1A1 genotype.These results agree with those of Jensen et al. (2012), who found that the worst MCPs were observed in milk with the A2A2 genotype.Previous studies have found a positive association between the CSN2 B allele and coagulation and cheese-making abilities (di Stasio and Mariani, 2000;Caroli et al., 2009).It is worth noting that the CSN2 B and A1 variants share the same mutation at position 67 of the amino acid chain (histidine instead of proline) in respect to the A2 variant, which is considered the original β-CN (Pal et al., 2015).Further studies are needed to ascertain whether this mutation may have a favorable role for the cheese making abilities of milk.It is worth noting that milk payments system in the dairy sectors that produce hard cheese with EU Protected Designation of Origin label often take into consideration in their payments criteria coagulation and curd firming properties, which are strongly influenced by the amounts, proportion and genetic polymorphisms of protein fractions (Bittante et al., 2012).Aside from CY WATER , no CSN2 significant effect was found for CY and REC traits, which in our work seems to be mostly related to the effect of CSN3.Conversely, Cipolat-Gotet et al. ( 2018) evidenced that β-CN was the protein fraction with the strongest effects on CY CURD and was positively associated with REC PROTEIN and REC SOLIDS .Moreover, κ-CN also exhibited a linear and positive association with all the %CY traits in agreement with our findings.However, in that work authors did not consider the effect related to protein polymorphisms and processed the samples using a laboratory cheese-making procedure that consisted in 1,500 mL/sample instead of the 9 mL used in our analysis.

Effects of Other Milk Protein Genotypes
CSN3 played an important role in regulating the variability in the milk protein fractions (P < 0.01), both in terms of g/L and % N, with the exception of LF (Table 6).Not surprisingly, CSN3 had a significant effect on almost all the MCP, CF, CY, and REC traits (P < 0.001), with the exception of CY WATER and REC PROTEIN .As expected, the κ-CN BB was the one that presented the best coagulating abilities, with a significantly lower RCT and a significantly higher a 30 , in respect to the homozygote AA (data not shown).Moreover, Figure 3B confirms that κ-CN BB displayed the most positive pattern of curd firming over time.The importance of CSN3 for MCP and CF traits was also consistent with  previous findings (Dadousis et al., 2016), and confirms the strong influence of κ-CN on milk technological properties.
The LGB affected the α S1 -CN concentration (P < 0.01), the β-CN proportion (P < 0.001), and the total whey proteins and β-LG concentrations and proportions.Previous studies have pointed to LGB as a candidate gene controlling the variability in α-CN, β-CN, and κ-CN, and in β-LG and α-LA (Heck et al., 2009;Berry et al., 2011).Furthermore, LGB also had a significant effect on RCT (P < 0.05), RCT eq and REC PROTEIN (P < 0.001), with β-LG BB showing increased RCT in respect to the AA genotype (data not shown), as previously reported by Bonfatti et al. (2010) and Cecchinato et al. (2012).The role of the LGB genotypes on MCP requires further investigation because previous works reported contradictory results.For example, Bonfatti et al. (2010) found a positive association between allele A and β-LG content and relative proportion and favorable effects on MPC in respect to allele B. On the other hand, Kübarsepp et al. (2005) found a positive association between β-LG BB and RCT, whereas Boland and Hill (2001) showed that selecting for the B variant of β-LG increased the milk casein and cheese yield per kilogram of milk protein.

CONCLUSIONS
The present study found that the CSN2 genotypes exert an effect on the MCP and cheese yield-related traits.In particular, the BA1 genotype was associated with the best cheese-making abilities.However, given the low frequency of this genotype, this finding warrants additional investigation.No significant differences were found between the cheese-making abilities of β-CN A1 and A2 milk, except for k SR even if a tendency was observed toward a slight worsening of coagulation properties for A2A2 milk.We conclude from these results that, if it can be demonstrated that A2A2 genotype is associated with higher digestibility and better gut physiology, the dairy industry should favor this genotype in milk destined for fresh consumption.However, it is inadvisable to use milk containing the A2 alleles for cheese production, as these are associated with a worsening of milk technological characteristics, and as a consequence, a less efficient cheese-making process.
DIM i is the fixed effect of DIM class i (12 classes of 30-d intervals, except for class 12 which included cows with DIM > 330 d); Parity j is the fixed effect of parity j of the cow (j = 1: first parity, j = 2: second parity, j = 3: third parity, j = 4: fourth and later parities); Herd/date k is the random effect of herd/date k (k = 1,…, 21) assumed to follow a normal distribution with Herd/date k /date variance and I is the identity matrix; animal l is the random additive genetic effect of animal l (l = 1,…, 5,733) assumed to follow a normal distribution with animal l

2
The model includes polygenic effects of animals.σ A 2 = additive genetic variance, polygenic effects of animals and CSN2, CSN3 and LGB genotypes.Percentage of genetic variance explained by milk
Bisutti et al.: A2 MILK AND CHEESE PRODUCTION

Table 2 .
Overall, average values for milk yield and composition were consistent with a

Table 2 .
Bisutti et al.:A2 MILK AND CHEESE PRODUCTION Descriptive statistics for milk protein composition expressed both quantitatively (g/L) and qualitatively (% of total milk N)

Table 6 .
Bisutti et al.:A2 MILK AND CHEESE PRODUCTION Results from ANOVA (F-value and significance) for milk yield, composition, and protein fractions expressed quantitatively (g/L) and qualitatively (% of total milk N)

Table 7 .
Bisutti et al.:A2 MILK AND CHEESE PRODUCTION Results from ANOVA (F-value and significance) for milk coagulation properties (MCP), cheese yield (CY) and curd nutrient recoveries (REC) traits max = maximum curd firmness achieved within 45 min; t max = time at achievement of CF max ; CY CURD