The aroma of a cheese is an important factor when it is acquired. The typical flavor of each variety of cheese is the result of a complex balance between volatile and non-volatile compounds originated during the ripening process from milk fats, proteins and carbohydrates. (Pillonel et al., 2003; Fox and Wallace, 1997).

The importance of volatile compounds is due to their correlation with the flavor, which depends on the ripening, time, cheese technology, seasonality, etc. The volatile fractions of some Iberian Peninsula cheeses such as Manchego (Gómez-Ruiz et al., 2002), Roncal (Izco and Torre, 2000), Zamorano (Barron et al., 2005a, Fernández-García et al., 2004a), Serra da Estrela (Dahl et al., 2000; Tavaria et al., 2004; Tavaria et al. 2006) have been studied as a response to the growing interest in the characterization of traditional products protected by a Denomination of Origin.

Canonical biplot (Vicente-Villardón, 1992), when it is oriented to the discrimination between groups or MANOVA-biplot (Gabriel, 1971; Galindo, 1986), when the aim is to study the variables responsible for the discrimination, are two effective methods. The main advantage to the biplot version of the technique is that it is possible not only for establishing the differences between groups but also to characterize the variables responsible for them.

Classic multivariate techniques, such as Principal Component Analysis and Canonical Correlation have been applied, among other areas, in fresh fruit (King et al., 2012), in the characterization of the sensory properties and texture of French cheeses (Antoniou et al., 2000), and to study the effects of mixtures of fats in the texture and sensory properties of cheeses (Lobato-Calleros et al., 1997). The Canonical biplot method has been used in the conservation of historical buildings and monuments (Varas et al., 2005), in civil engineering (Iñigo et al., 2005; Iñigo et al., 2013) and, in the case of foods, in the study of margarines (Rui Alves and Beatriz Oliveira, 2003), but no reference has been found regarding its application to the characterization of cheeses using volatile compounds as variables.

Taking into account that the volatile composition of cheese shows significant changes during the ripening period (Fernández-García et al., 2004a; Fernández-García et al., 2004b; Innocente et al., 2013), cheeses ripened from 0 to 6 months have been analyzed to obtain the characteristic profile of the samples throughout the ripening stage.

Here we used the Canonical biplot method to study which volatiles from the cheeses studied were most affected by each of the factors explored: the type of milk used to elaborate the cheeses (cow, sheep, goat) and the seasonality of the milk (winter, summer).

The application of individual ANOVAs to the variables analyzed, (Table 2), shows that the variables isopropanol, 2-butanone, 2-butanol, propanol, 3-methyl-1-butanol, propanal, 3-methylbutanal and butanol did not have discriminatory power since they gave a result of p>0.05.

The results of the Canonical biplot are shown in Table 3, where it may be seen that the variance of the three main axes accounts for 89.175% of the variance explained. The global contrast based on Wilk's Lambda test has a value of 2.5647, with p<0.01, indicating that, globally, differences are present between the 6 groups of populations of cheeses to be compared (WC), (SC), (WSh), (SSh), (WG), (SG).

On the factorial planes (Figure 1a) Axis 1 vs. Axis 2; b) (Figure 1b), Axis 1 vs. Axis 3; c) (Figure 1c), Axis 2 vs. Axis 3), the labels with poor, acceptable and good quality of representation appear respectively in grey, normal font and in bold. The interpretation of the labels depends on a series of measurements such as the quality of representation for the different planes (variance absorption of the planes, the goodness of the projections of the measurements on the variables for the dimensions selected, etc.) (Gabriel, 1971; Gabriel and Odoroff, 1990; Galindo, 1986).

In general, the quality of representation of the projected groups and variables on the first three planes is good, with the exception of that of SC, which is acceptable, and butane, which is poor. Upon analyzing the first two factorial planes (Figure 1a, b) it may be seen that the group of cheeses from winter cow's milk (WC) is clearly separated from the rest of the groups. Analyzing the variables responsible for this separation, it may be seen that this difference is due to the fact that the samples take higher values in ethanol, ethyl acetate, acetaldehyde and 2-butanol, together with lower values in 2-heptanone, 2-pentanone, isopropanol, 2-pentanol, acetone and 2-heptanol. This result is in agreement with previous works indicating that ethanol is present at higher concentrations in cow's cheese than in sheep's cheese (Molina et al., 1999), while acetaldehyde is not one of the more important aldehydes in sheep's cheese (Barron, et al., 2005b; Fernández-García et al., 2004a, b) and it could not be found in goat's cheese (Castillo et al., 2007).

There is a clear difference between the WC and WG winter cheeses and the rest of the cheeses made in summer. This difference is hardly noticeable in the sheep cheeses (WS and SSh). The sheep cheeses (Figure 1a) were characterized by higher contents of methyl-ketones (2-pentanone, 2-heptanone and 2-nonanone), confirming earlier results indicating that methyl-ketones are compounds that appear in greater amounts and proportion in sheep cheeses (Barron et al., 2005a).

Regarding goat cheeses, these were characterized (Figure 1b) by 3-methyl-1- butanal and 3-methyl-1- butanol. The presence of this alcohol is a consequence of the reduction in the aldehyde and it has previously been reported that it is present at a higher level in goat's milk cheese than in cow and sheep's milk cheeses (Molina et al., 1999).

Regarding the variables, propanal clearly separates the winter from the summer cheeses (p<0.05) since it takes the lowest values in winter in comparison with the higher values found for summer. When the mean values of the six groups of cheeses are projected onto the propanal variable, it can be observed that winter milk cheeses WC, WG, WSh show lower values than their summer equivalents (SC, SG, SSh). The same results have been obtained in Zamorano cheeses by Fernández-García et al., (2004a), who found the highest values of propanal in summer cheeses, with lower values in autumn and winter. This compound, as all the aldehydes, appears in very low concentration since it is an intermediate and unstable compound which is usually reduced to alcohols.

The same difference, although less pronounced, can be seen in the variables 2-pentanol, acetone, 2-heptanol and isopropanol (Figure 1a). These results are in agreement with those reported previously. Thus, Fernández-García et al. (2004a, b), found lower values of propanal in winter cheeses while 2-pentanol and 2-heptanol were significantly lower in autumn cheeses and no significant differences were observed for 2-propanol. In Figure 1b it may be seen that the parameters ethanol, ethyl acetate 2-butanol, acetaldehyde and propanol are detectors of winter cow's cheese samples owing to their higher mean values than the rest. The parameters 3 methyl 1butanol, 3 methyl butanal separate the goat (WG, SG) and sheep samples (WSh, SSh), (Figure 1b, Axis 1-Axis 3). The parameter 2-butanone separates goat's cheese since it was not possible to detect it in a large number of the samples of this cheese. In Figure 1c, Axis 1 vs. Axis 3, the parameter 3 methyl butanal is responsible for the statistically significant differences between the winter and summer goat's cheeses, together with acetone and 2-heptanol, which also show differences between winter and summer with the rest of the samples.

With respect to the seasonality of the cheeses, a comparative analysis between the same cheese made at different times of the year (W and S) afforded the following results: the WC group vs. SC differs (p<0.05) in all the variables, but mainly in ethanol, acetaldehyde, 2-pentanone, 2-heptanone and 2-butanol. On comparing WG and SG, these differ (p<0.05) in most of the variables, but mainly in the variables 3-methyl butanal (Figure 1c) and acetone, 2-heptanol and 2-pentanol. (Figure 1a, b). Fedele et al., 2005, found that goat winter milk had smaller amounts of ketones and more alcohols than summer milk. However, they did not find significant variations in the volatile compounds of goat cheeses from one season to another. The WSh group vs. SSh only differed (p<0.05) in the variables isopropanol, 2-pentanol, acetone, 2-heptanol and propanal. 2-propanone has been found to be higher in winter-autumn sheep cheeses, while heptanol, isopropanol and 2-pentanol are less abundant in these cheeses (Fernández-García et al., 2004a, b).

These results are due to the fact that the concentration and composition of various milk components differ according to season. Differences in the amount and composition of milk fat have been reported (Johansen et al., 2002, Nudda et al., 2005) owing to variations in feed factors. During the first months of the year the forage consumed tended to be mainly hay (98% of dry matter) with a small contribution from silages, less than 30% of the ration. From spring to summer the contribution of silage, first cuts, green forages etc., increased even if the animals remained in the pens. This type of diet reduces the dry matter intake. Afterwards, from summer to winter the diets were standardized and were composed of 40% hay and a 30% silages. When summer and winter diets are compared, although the composition was similar, the ration changes, in fact, because the animals tend to eat more in winter than in summer due to the number of daylight hours and temperatures. Indeed, the rumen metabolism is also affected because of the contributions of melatonin in the blood that affect both the ruminal bacteria populations and rumen parasympathetic regulation. Therefore, there are modifications in food intake, food composition, rumen activity, ruminal flora and circadian rhythms that affect cheese composition.

Thus, the fat contents (% by weight) of the winter cheeses analyzed in this work had higher values than those of the summer cheeses (González-Martín et al., 2011). Moreover, important properties such as the protein:fat ratio and the casein:whey protein ratio are seen to be affected by time of year (Barron et al., 2001; Sitzia et al., 2015; Addis et al., 2015). Since volatile compounds are formed through the degradation of amino acids, leading to amines, aldehydes, alcohols, acids and sulfur compounds, and the breakdown of fatty acids, which produces esters, methyl ketones and secondary alcohols, the above changes in milk composition will affect the volatility profile.

The present results show that the volatile chemical compounds used as variables in the Canonical biplot method are suitable for establishing relationships between the significant variables (graphically and quantitatively) and the six groups of cheese: Winter Cow, Summer Cow, Winter Sheep, Summer Sheep, Winter Goat, Summer Goat. Winter cow's milk cheeses group (WC) is clearly separated from the rest of the groups because these cheeses have higher values of ethanol, ethyl acetate, acetaldehyde and 2-butanol, together with lower values of 2-heptanone, 2-pentanone, isopropanol, 2-pentanol, acetone and 2-heptanol. There is a clear difference between the WC and WG winter cheeses and the rest of the cheeses made in summer but this difference is hardly noticeable in the sheep cheeses. The variables propanal together with 2-pentanol, acetone, 2-propanol and isopropanol, although to a lesser extent, clearly separate winter and summer cheeses owing to their lower mean values in winter. Goat's and sheep's milk cheeses were separated by 3-methyl-1-butanol and 3-methyl-1-butanal compounds. This last compound together with acetone and 2-heptatone allow for the differentiation between winter and summer goat's milk cheeses. All this indicates that the Canonical biplot is a suitable method for explaining this type of population, in which many different factors are involved.