Delimitation of the study area and description of the Goat Production Systems (SPC)

In this study, we included the SPC of the Sistema Producto Especie Caprinos A.C. organization in the mountainous central region and the high plateau of Veracruz with mountain mesophyll forest, fir and pine forests, and xerophyll scrub as the dominant vegetation (Table 1) (^{García et al., 2008}; ^{Márquez and Márquez, 2009}).

Artisan cheese manufacture

The cheeses are made with milk from Alpine and Sannen goats milked manually. The milk is filtered, pasteurized (63 °C for 30 min) and cooled (37 °C). Commercial rennet (curdling force 1:10,000 or 110 IMCU mL^-1, Industrias Cuamex, México) is added at a proportion of 30 mL 100 L^-1 milk. The curds are cut 45 min later and molded in vinyl polychloride hoops and compacted with a stainless steel press (2 kg_f kg^-1 cheese for 7 h). The cheese is submerged in brine (28 %, pH 8) and stored at room temperature (18±2 °C) for 2 d. They are then inoculated by spraying Penicillum candidum (8x10⁷ UFC per dose) Choozit™ PC-VB (commercial brand Danisco, Dupont of Mexico) and stored for seven weeks in wooden or masonry cellars at 18±2 °C and 80-85 % RH.

Microbiological, chemical, color and texture analyses

The microbiological, chemical and instrumental data were analyzed with standard deviation, one-way ANOVA and comparison of means with the Tukey test (^{Delgado et al., 2011a}). The experimental design was completely randomized with four treatments (Q_NOC, Q_BMJ, Q_DHE, Q_LIG) and three or five replications per treatment of the chemical or color and texture variables.

Sensorial characterization of the aged cheeses and consumer study

The sensorial characterization data were analyzed with a three-way factorial ANOVA (product, judge, session) and interaction (product x judge) to determine the performance of the panel (^{Ryffel et al., 2008}). The preference values were analyzed with an ANOVA, in a factorial design with two factors (product and consumer) and later grouped into homogeneous classes by hierarchical ascendant classification (CJA-abbreviation of Spanish) with the Ward method (^{Schmidt et al., 2010}).

Typing aged cheeses using causal models

Partial least squares (PLS-path modeling) regression, coupled to a multiple factorial analysis (AFM) was applied to analyze the interactions between factors or aggregates, following the methodology of ^{Pagès and Tenenhaus (2001)}.

The aggregates (latent, or not observable, grouping variables) and their variables considered were agroecological conditions (CC: altitude, precipitation and temperature), goat production system (SIS; type of system and type of feeding), milk (protein, fat, SNG and titratable acidity), cheese chemical composition (FQ; protein, fat, moisture, ash, Aw and pH), instrumentals (NS; L^*, a^*, b^*, C^* and H°), sensorial (SEN; all sensorial attributes) and homogeneous classes of consumers.

The selection criteria were the following: 1) the milk aggregate was retained because of its effect on cheese yield, and 2) the rest of the aggregates, their variables were in function of probability (p≤0.05), in accord with the ANOVA models (^{Brunschwig et al., 2004}; ^{Salvador and Martínez, 2007}). After selecting the variables, AFM was applied to determine those that formed each causal model in function of its correlation (r)>0.70 (positive or negative) with principal factor one and two, and the variables of the aggregates CC and SIS were considered supplementary. Equations one to eight showed the connections between aggregates in the causal models:

where ξ, η, β and Z are exogenous and endogenous variables, path or route coefficients, and error term, respectively. The numbers in parentheses correspond to each aggregate and those on the side of the β coefficient indicate the connection between aggregates.

Validation of the proposed causal models was done in the following phases: 1) evaluation of the measurement model by correlating (r) the variable and the aggregate, considering that r>0.70 (positive or negative) corresponds to relevant variables; 2) convergent validity of each aggregate using the extracted analysis of variance (AVE); 3) evaluation of the structural model with the β coefficients, the variance (R²) explained by aggregate and r between aggregates; and 4) goodness of fit (GoF) index between the measurement model and the structural model (^{Tenenhaus et al., 2005}).

The ANOVA were performed with the software STATGRAPHIC PLUS^® version 5.2 (Statistical Graphics Corp., USA). The AFM, CJA and PLS-path modeling were done with the software XLSTAT, version 2009 (Addinsoft, New York, NY, USA).

Microbiological analyses of pasteurized milks and aged artisan cheeses

In the MA and CT counts, differences (p≤0.05) were observed. All the samples had MA below the value of 4.48 LogUFC m L^-1 (3x104 UFC m L^-1) permitted by NOM-091-SSA1-1994. Only the samples BMJ and DHE (3.36 and 4.63 LogUFC m L^-1) exceeded the maximum CT limit (1.04 LogUFC m L^-1 o 10 UFC m L^-1) permitted by ^{NOM-243-SSA1-2010}. Escherichia coli, S. aureus, Salmonella spp. and B. melitenses were not detected (Table 2). These results coincided with the elevated loads of MA (2 to 7.4 Log UFC m L^-1) and CT (2 to 7.54 Log UFC m L^-1) determined by ^{Gómez-Ruiz et al. (2012)} in goats’ milk in the semi-arid regions of San Luis Potosí, Mexico. In contrast, ^{Martínez et al. (2010)} found a low incidence (0.5 %) of B. melitenses in goats’ milk from the UPC of the municipalities of Perote and Jalancingo, Veracruz. The results of these authors are due to pasteurization (^{Kousta et al., 2010}). The high loads of CT are the result of deficiencies in hygiene in the production process and in post-pasteurization stages (^{Yamazi et al., 2013}).

Incidence of MA was found (p≤0.05) only in DHE cheese (4.61 Log UFC g^-1) and it exceeded the maximum value of 4.48 Log UFC g^-1 (3x104 UFC g^-1) permitted by ^{NOM-243-SSA1-2010} (Table 3). Absence of CT, E. coli, S. aureus and Salmonella spp., is attributed to the interaction of acid pH, Aw and competition with the lactic culture (^{Durán et al., 2010}; ^{Kousta et al., 2010}).

Physicochemical analyses of pasteurized milks and aged cheeses

The milk from Coatepec (NOC and BMJ) had the highest contents of SNG, protein and lactose (p≤0.05). NOC milk had high content (p≤0.05) of fat (5.82 %), which may be due to the presence of orange bagasse in the goats’ diet (^{Salvador and Martínez, 2007}). In contrast, LIG milk (Tatatila) showed low (p≤0.05) protein content (1.38 %), SNG (4.56 %) and lactose (2.71 %) (Table 4).

The fat and lactose contents in our study were higher than those reported by Part et al. (2007) (3.8 % fat and 4.1 % lactose) and similar to those observed by ^{Chacón-Villalobos and Pineda-Castro (2009)} (4.1 % fat and 4.3 % lactose). The contents of proteins and SNG are comparable to those found by ^{Soryal et al. (2004)} and ^{Salvador and Martínez (2007)} (2.9 % and 7.83 %, respectively).

All the samples, except the LIG sample, exhibited density, conductivity and cryoscopic point similar to those reported by ^{Park et al. (2007)} (1028 kg m^-3, 4.3 mS cm^-3 and -0.540 a -0.573 °C, respectively). Salt contents and titratable acidity were higher than those obtained by ^{Chacón-Villalobos and Pineda-Castro (2009)} (0.5 % salts and 1.5 g L^-1 acidity).

Only the LIG sample (-0.32 ºC) had a cryoscopic point outside the range (-0.540 to -0.573 °C) reported by ^{Park et al. (2007)}. This effect may be due to the proportion of water added to the milk (37.24 %), which causes modifications in density and salt and lactose contents. According to ^{Raynal-Ljutovac et al. (2005)} and ^{Inglingstad et al. (2014)}, diets with low quality forages and the expenditure of energy during grazing can modify the cryoscopic point, protein and lactose content and SNG of the milk.

The LIG and BMJ cheeses had high (p≤0.05) contents of fat (43.27 %) and protein (25.33 %). NOC cheese had low (p≤0.05) contents of moisture (29.14) and ash (2.70 %). The lowest (p≤0.05) values of Aw and pH were found in BMJ and DHE cheeses (Table 5).

The fat contents in our study were higher than those documented by ^{Salvador et al. (2014)} (24.926.9 %) in cheeses made with milk from goats fed orange bagasse and aged for 60 d. ^{Fresno and Álvarez (2012)} determined 36.42 % moisture, 51.10 % fat and 34.46 % protein in Spanish cheeses aged for 60 d. ^{Guizani et al (2006)} and ^{Delgado et al. (2011a)} found pH 4.95 to 4.8 in goat cheese aged for 30 and 60 d. Cheeses with a high content of fat was related to inclusion of orange bagasse in the goats’ diet and with low Aw and pH (^{Salvador et al., 2014}). Accumulation of lactic acid, from lactose, may have caused a descent in pH (^{Delgado et al., 2011a}).

The protein content is due in part to the goats’ diet and to the concentration of nutrients because of dehydration of the cheese during aging (^{Peláez et al., 2004}), while moisture and Aw depend on the added salt (^{Las Casas et al., 2008}).

Color and texture profile of aged artisan cheeses

NOC, DHE and LIG cheeses had higher (p≤0.05) values in a^* (-2.30 to -3.13), b^* (12.43 to 17.40), C^* (12.66 to 17.64) and H° (75.85-81.04) (Table 6), and thus these cheeses were in the yellow zone, while only the BMJ cheese exhibited higher luminosity (p≤0.05).

Increases in a^* and b^* are due to biochemical reactions, such as lipolysis and proteolysis during aging, to addition of vegetable oils (such as that in orange bagasse) and to diversified feeding of the goats (^{Lacerda de Medeiros et al., 2013}). The color results were similar to those reported by ^{Fresno and Álvarez (2012)} (a^*: -2.28, b^*: 11.89, C^*: 12.13 and H°: 87.42), while ^{Sert et al. (2014)} reported ranges of -2.9 to -2.3 (a^*) and 8.98 to 16.10 (b^*) in Tulum cheese made with pasteurized goats’ milThe high value of L^* of BMJ cheese could be due to the high moisture content (^{Salvador et al., 2014}), and its luminosity (L^*: 95.98) was similar to that of Ibores cheeses aged for 60 d (^{Delgado et al., 2011a}).

Hardness, breakability, cohesiveness, gumminess, and chewiness of NOC cheese were higher (p≤0.05). In contrast, the BMJ, DHE and LIG cheeses had similar gumminess and chewiness (p>0.05) (Table 7).

The hardness, breakability and cohesiveness of the cheeses of our study were similar to the cheeses aged for 60 and 90 d analyzed by ^{Fresno and Álvarez (2012)}: harness 120.00 to 165.73, breakability 61.58 to 74.71, and cohesiveness 0.11. Differences in hardness, cohesiveness, gumminess, elasticity and chewiness may be due to: 1) gradual loss of moisture during aging and rupture of the casein network, 2) the plastic texture of the cheeses caused by high pH, and 3) high content of fat due to the inclusion of a large amount of fiber and orange bagasse in the goats’ diet (^{Delgado et al., 2011a}; ^{Medeiros et al., 2013}; ^{Salvador et al., 2014}).

Sensorial characterization of the aged cheeses

According to the ANOVA, with interaction (product x judge) with three factors: 1) the panel detected differences (p≤0.05) among the cheeses (product factor), 2) agreement (p≤0.05) in the attributes DUR-T, AREN-T, O-MADE, O-CAB, SALA and A-CAB, 3) consistency in the results of all of the attributes among the sessions (p≤0.05), and 4) consensus in the placing the cheeses on an intensity scale of DUR-T, AREN-T, O-MADE, AREN-B and A-CAB.

NOC cheese had higher (p£0.05) intensity CAMA, O-CITRI, ACID and A-FERM.BMJ cheese was perceived (p£0.05) as DUR-T, AREN-T, O-CAB, O-MADEH, SALA, A-CAB, A-HONG and R-SEQU, LIG cheese was characterized (p£0.05) by the attributes O-LECH, O-FRU and A-FRU, while DHE cheese had intermediate intensities in all the attributes (Figure 1).

Figure 1 Sensorial profile of goat mature cheese. 

The CAMA attribute of the NOC cheese may have been caused by the addition of oils in the orange bagasse (^{Lacerda de Medeiros et al., 2013}). The attributes of DUR-T and AREN-T are related to the decrease in moisture during aging. The O-CABR attribute is due to the presence of carboxylic, decanoic, hexanoic and octanoic acids (capric, caproic and caprylic) generated during aging (^{Delgado et al., 2011b}). Octanoic and decanoic acids, 2-heptanone ketone and 2-heptanol alcohol are identified as responsible for the O-FRU, A-FRU, O-MADEH and A-HONG attributes, respectively (^{Poveda et al., 2008}). O-LECH is associated with the lactone (-dodecalactone (^{Hayaloglu et al., 2013}). The ACID attribute may have been caused by fermentation of the lactose, by compounds derived from lipolysis such as 3-methylbutanoic acid, and by orange bagasse (^{Salvador et al., 2014}). The SALA attribute could be due to the release of peptides by proteolysis, which decreases moisture and increases the salt concentration (^{Las Casas et al., 2008}). Thus, the cheeses made in semi-intensive SPC exhibited greater sensorial diversity of odors and aromas than those made in intensive SPCs. This coincides with observations of ^{Soryal et al. (2004)} of the diversity of odors and aromas of Domiati cheeses made from milk from grazing goats.

Consumer evaluation

The preference scores given to the cheeses were similar (p>0.05), oscillating between 6.7 and 7.1 and between “slightly pleasing” and “moderately pleasing”. ^{Ryffel et al. (2008)} and ^{Salvador et al. (2014)} reported preference scores of 6.3 to 6.8 for semi-aged cheeses and cheese from goats fed orange bagasse. The differences (p £ 0.05) in scores within the panel differentiated consumers into classes: class 1 (25), class 2 (18), class 3 (16), and class 4 (17).

Typing aged cheeses using causal models

Acidity, pH and Aw of causal model one (Figure 2) and the AREN-B variable of model two (Figure 3) were not considered in the explanation of the interrelationships because r<0.70. The AVE of the aggregates CA, SIS, Milk, FQ, INS and SEN in model one were 0.80, 0.51, 0.50, 0.58. 0.80 and 0.98 and in model two they were 0.76, 0.54, 0.99, 0.73, 0.83 and 0.78. This indicated that more than 50 % of the aggregated variance is explained with the retained variables (^{Espejel et al., 2014}). Both models showed that R² and β met the minimum values of 0.1 and 0.2, respectively, highlighting the explanatory power of the aggregates. The goodness of fit (GoF) values 0.66 and 0.68 show good fit of the proposed models (^{Tenenhaus et al., 2005}; ^{Espejel et al., 2014}). The β values of the first model (Figure 2) showed that the SIS-FQ route (r=-0.61, β₂₄ =-0.93) was more important than SIS-Milk (r=-0.54, β₂₃ =-0.09) and MIlk-FQ (r=-0.07, β₃₄ =-0.58). Therefore, semi-intensive SPC (-0.93) and diversified feeding with acorns, Kikuyo and native grass (-0.62) were associated with high ash (0.95) and moisture (1) contents in the cheeses, and the latter had a negative relationship to the other INS parameters (r=-0.87, β₄₆ =-0.87). This result agreed with ^{Chacón-Villalobos and Pineda (2009)}, who pointed out that high moisture content is associated with cheeses that exhibit high luminosity and low color values (b^* and C^*). ^{Fresno and Álvarez (2012)} determined that proteins decrease their mobility because of moisture loss during aging, which increases hardness, breakability, gumminess and chewiness.

Figure 2 Causal model one of aged goat cheese relative to principal component one of the multiple factorial analysis. 

Figure 3 Causal model two of aged goat cheese relative to principal component two of the multiple factorial analysis. 

Thus, the INS-Class 2 (r=0.57, β₆₇ =0.38) route indicated that 23.7% of the consumers preferred NOC cheese because of its yellow color (b^* and C^*), hardness, breakability, gumminess, chewiness and sensorial characteristics (route SEN-Class 2, r=-0.57, β₅₇ =-0.38) due to A-FERM (-0.98). Consumers of class 4 (22.4 %) preferred BMJ cheese because of its sensorial characteristic A-CAB (route SEN-Class, r=0.68, β₅₈ =0.24).

β gave evidence that route CA-SIS (r=0.48, β₁₂ =0.48) is more important than route CA- Milk (r=-0.40, β₁₃ =-0.65) (Figure 3). This effect is in agreement with the low negative correlations (-0.035, -0.082) between agroecological parameters and milk composition documented by ^{Echeverri and Fernando (2009)}.

β and r of the routes SIS-Milk (r=0.2, β₂₃ =0.53) and SIS-FQ (r=-0.37, β₂₄ =-0.53) showed low direct influence of the SIS aggregate in the milk aggregates and FQ. To the contrary, the route MilkFQ (r=0.72, β₃₄ =0.86) highlighted the influence of the milk in the physicochemical characteristics of the cheese.

The indirect SIS-Milk-FQ route showed the effect of the type of system and feeding on physicochemical aspects of the cheeses. In this respect, the type of dairy system has an influence in chemical characteristics of the cheeses (^{Bergamaschi et al., 2015}).

The aggregate FQ had a positive influence in the INS (r=0.88, β₄₆ =0.88) and SEN (r=0.98, β₄₅ =0.72) aggregates. The model showed that the semi-stabled SPC (-0.97), with diversified feed, such as reeds and King grass (-0.99), produce milk with high contents of SNG and protein, which increase protein content of the cheese.

The relationship of the protein content of the cheese (0.99) with the parameters (INS) L^* (0.95), a^*(0.89) and H° (0.97) is due to the proteolysis reactions, and darkening during aging can decrease L^* and increase a^* (^{Tejada et al., 2006}). The cheese with high contents of fat (-0.70) is related to the CAMA attribute (-0.81), and this tonality may have been due to the orange pulp and vegetable oils in the diet of the goats (^{Lacerda de Medeiros et al., 2013}; ^{Salvador et al., 2014}). Fat was also related to cohesivity (-0.81) and this could be due to the loss of fat globules, resulting from lipolysis, which decreases cheese plasticity.

The inverse relationship between cohesiveness (-0.81), sensorial attributes DUR-T (0.90), AREN-T (0.76) and protein content (0.99) could be due to the modification in cheese texture because of the union between the peptides and polypeptides derived from proteolysis during aging (^{Karami et al., 2008}).

Class 1 consumers (32.8 %) preferred INS cheeses (r=0.90, β₆₇ =0.53), with L^* and low SEN characteristics (r=0.90, β₅₇ =0.40), DUR-T (0.90), AREN-T (0.76), O-CABR (0.92), O-MADEH (0.99), A-HONG (0.99), SALA (0.92) and R-SEQUE (0.99). These characteristics are particular of BMJ cheese, in which the O-CAB attribute can have a greater influence on consumer preference (^{Ryffel et al., 2008}). Class 3 (21.1 %) preferred INS cheeses (r=-0.61, β₆₈ =-1.04), with cohesiveness, and SEN (r=-0.51, β₅₈ =-0.42) and the atributes CAMA (-0.81), O-LECH (-0.80), O-FERME (-0.91), O-CITRI (-0.87), O-FRU (-0.94), ACID (-0.73) and A-FRU (-0.94). These characteristics are peculiar to cheeses made with milk from goats fed a diversified diet (LIG) and orange bagasse (NOC).