CHEMICAL AND SENSORY CHARACTERISATION OF SWEET WINES OBTAINED BY DIFFERENT TECHNIQUES

Little is known about the chemical and sensory characteristics of natural sweet wines obtained by different grape dehydration processes. The main aim of this work is to characterise several natural sweet wines, in order to understand the influence of grape dehydration on the chemical and sensory profile of those wines. First, conventional oenological parameters and low molecular weight phenolic compounds have been determined. Next, sensory descriptive analysis was performed on individual samples based on citation frequencies for aroma attributes and conventional intensity scores for taste and mouth-feel properties. Low molecular weight phenolic compounds and acidity were found in a lower concentration in most wines from off-vine dried grapes. Late harvest wine presented higher amounts of phenolics. White wines showed higher sensory and chemical acidity. Most wines obtained from off-vine and on-vine grape dehydration presented common notes of dry fruits and raisins as aroma properties. Chemical and sensory analyses performed in this study were able to define sweet wines. No significant differences have been found in chemical and sensory profiles of sweet wines according to dehydration processes of the grapes. Late harvest wine and white wines were differentiated from other wines through chemical and sensory analysis


INTRODUCTION
The first wines obtained by human beings were sweet wines due to the climatic conditions where grapes were harvested and processed, which did not allow to finish fermentation. In the Mediterranean basin, the climatic conditions were favorable to leave the grapes in the vine to overripe, or to collect the clusters and allow them to dry in the sun and wind. Throughout history, it has been important to preserve the sweetness in wine, and sweet wines were highly prized in ancient Rome and in the Middle Ages (Scienza, 2013). Nowadays, sweet wines constitute a very small percentage of the world wine trade; however, there is a growing interest in high quality sweet wines. The wine industry tries to diversify the wine market by producing wines with distinctive characteristics and creating new products of high quality. In this regard, the production of sweet wines could be a good way to achieve this goal.
In the production of natural sweet wines (without addition of ethanol), the sugar content can be increased by different techniques. The most common techniques are (Reboredo-Rodríguez et al., 2015): • On-vine: late harvest wines, noble rot wines (Sauternes, Tokaj), icewines (the grapes are allowed to freeze on the vine in the winter frost season, once harvested they are pressed and the must is concentrated because the water is retained as ice). Due to the difficult climatic conditions needed to produce freezing in certain regions, freezing grapes by technical means (freezing chamber, liquid nitrogen) is used to produce non-authentic (artificial) icewines. The International Organisation of Vine and Wine (OIV) set definition and prescriptions for authentic icewine. Harvesting and pressing should be performed at a recommended temperature lower or equal to -7 °C. The potential alcohol strength by volume for musts cannot be increased and should be as a minimum 15% volume (corresponding 110 °Oechsle or 25.3 °Brix). The minimal alcoholic strength acquired should be 5.5% by volume. The maximum limit of volatile acidity should be (2.1 g/L expressed in acetic acid). All grapes used in icewine should come from the same region (OIV, 2018). Hermann (2014) has developed a method for icewine authentication.
• Off-vine: Post harvest dehydration is carried out when grapes are shrinking and losing water. There are several methods such us sun drying (Malaga wines), drying in closed environments (Supurao, Passito, Tostado de Ribeiro), where grapes are dried in a chamber under artificially controlled conditions of temperature, airflow and humidity.
During grape dehydration, in addition to sugar concentration, phenolic and aromatic compounds are concentrated or new compounds are produced (Moreno et al., 2008). Phenolic compounds are abundant in grapes and play an important role in wine quality. Among the non-coloured phenolic compounds, phenolic acids, flavanols and flavonols represent an important group of secondary metabolites in grapes that play an essential role in certain sensory characteristics of wines such as astringency and bitterness (Fischer and Noble, 1994;Vidal et al., 2004). The phenolic profile of the wine varies according to grape variety and degree of ripening, cultural practices, vineyard location and processing and storage techniques (Ramos et al., 1999;Fang et al., 2008). For example, in ice wines, some authors have determined that phenolic acids and flavanols are the major phenolic compounds (Tian et al., 2009a;Tang et al., 2013). Phenolic composition has been studied in icewines from various regions or cultivars (Li et al., 2016;Luo et al., 2017). Kilmartin et al. (2007) observed the impact of freeze concentration and harvest time upon polyphenol composition and concentration. The variations of phenolic compounds in the making of sweet wines has been determined as a function of environmental conditions employed for the dehydration process and of the rate of water loss (Serratosa et al., 2011;Figueiredo-González et al., 2013).
Moreover, aroma is one of the most important quality factors that determines the characteristics of products. The aroma in wine consists of hundred compounds derived from grapes. Due to grape dehydration, either on vine or once harvested, changes in volatile composition of grapes are produced, which means also changes in aromatic profiles of obtained sweet wines (D'Onofrio, 2013;Reboredo-Rodríguez et al., 2015). The mechanisms by which flavours appear in sweet wines are not very clear. However, some studies have pointed out that variety, origin, addition and fermentation management (yeast metabolism in a high sugar concentration medium) influence sensory characteristics (Bellincontro et al., 2004;Budic-Leto et al., 2010;González-Álvarez et al., 2013). The effect of yeast strain in aroma compounds has been also studied (Synos et al., 2015). Crandles et al. (2015) observed that yeast effects depended upon cultivar and vintage. Changes in aroma profile throughout on-vine over-ripening and freezing processes have been also studied (Bowen et al., 2016;Lan et al., 2016). Sensory analysis techniques have been applied to wines during the last decades and have become a common practice to obtain objective characterisation and discrimination of products as well as a tool for quality control (Sáenz-Navajas et al., 2011).
Few studies have been published about the effect of techniques of grapes dehydration on the chemical composition and sensory characteristics of obtained sweet wines. Therefore, the main aim of this work is the chemical and sensorial characterisation of different sweet wines obtained by different dehydration techniques, in order to compare and distinguish them, and thus to understand the influence of dehydration processes of grapes. In a previous work, a general picture of the grouping of samples (sweet wines under study) in a multidimensional scaling space derived from a sensory sorting task was presented (Rosáenz-Oroz et al., 2012). Thus, a larger scale study was carried out in order to obtain accurate descriptions of those sweet wines and differences among them.

Wine Samples
Ten sweet wines were selected in order to check out different origins, vintages, varieties and processing techniques. Table I shows codes and general information of studied wines. Winemaking conditions of selected wines are detailed below according to the commercial information provided by the wineries. RMV and RMV2: grapes were harvested at optimal moment of ripeness (October) and were then frozen in a freezer chamber at -18 ºC. Grapes at a temperature of -9 ºC were pressed in January. Must with 35 ºBrix was fermented in a stainless steel tank at low temperature by selected yeasts.
RMV1: grapes were left on vines until February when cold weather conditions and frosts in vineyard took place. Grapes were collected and pressed immediately in vineyard at a temperature of -4 ºC. Must with 32 ºBrix was fermented in stainless steel tank at low temperature by selected yeasts.
Ojuel: after the harvest in October, clusters were hung in a high place with natural ventilation, where they would dry until January. The sweet must extracted from those bunches fermented in an oak barrel in the course of a year.
Blue Nun: after grapes that were being picked at -7 °C were pressed and then the juice slowly fermented.
Malizia: after picking grapes at -8º C they were pressed immediately in vineyard, fermentation took place in oak barrels, followed by aging in American oak barrels for six months.
No Phone: dried grapes on vines were harvested in December and then were pressed. Obtained juice fermented in oak barrel.
Verdil de Gel: grapes were harvested with 13% Probable Alcohol Degree in October. They were introduced in a cold chamber at -10 ºC for 15 days. Frozen grapes were pressed and the obtained must fermented.
Vi de glass (Riesling and Gewürztraminer): these wines were made with a prototype fed by two freezing chambers which worked with liquid nitrogen at -15 °C. The must fermented in stainless steel tanks for 2.5 months.

Analysis of conventional oenological parameters in wines
The determination of reducing sugars, ethanol content, pH, titratable and volatile acidities were determined by Infrared Spectrometry with Fourier Transformation with a WineScan ™ FT 120 (FOSS®, Barcelona, Spain), which was previously calibrated with the official OIV methods (OIV, 2016). The scanning interval set from 930 to 5011/cm. Table II shows the obtained slope and intercept adjustments. Total Polyphenol Index (TPI) was estimated as absorbance at 280 nm multiplied by 100 (Ribéreau Gayon et al., 2006). L-lactic and L-malic acids were determined by enzymatic methods according to the official methods of AOAC (AOAC, 2016). The analyses were performed in triplicate.
The ESI source was operated in negative mode recorded in the range of m/z 120 and 1500. The optimised conditions of the ESI source were as follows: capillary voltage, 3.5 kV; ESI source temperature, 180 °C; desolvation temperature, 200 °C; cone gas flow, 9 L/min; the nebulizer gas was set at 3 bar and 25 °C. The identity assignation of compounds was carried out by comparison of their retention time (t R ), MS and MS/MS spectra with those of their respective commercially available standards. Compounds with no available standards were identified by their MS/MS spectrum compared to the retention time and spectral mass from the literature (Schwarz et al., 2009;Sáenz-Navajas et al., 2010a). The quantification was expressed in mg/L of caffeic acid, catechin and quercetin (Sigma-Aldrich) for phenolic acids, flavanols and flavonols, respectively. The analyses were performed in triplicate. Table III shows the phenolic compounds identified by this method in wine samples.

Sensory analysis
Sensory analysis methodology applied in this work was developed by other authors (Sáenz-Navajas et al., 2011). A total of 29 students, wine consumers and members of the University of La Rioja (16 female and 13 male from 21 to 57 years old, average age 29 years) were recruited based on their interests and availability to conduct a session of one hour per week for 8 weeks. All judges were interested in the world of wine, with experience in similar studies of sensory analysis.
Panellists attended eight descriptive sensory training sessions (about 1 hour per session) over a two-month period, during which panellists worked in subgroups led by the same leader and following the same guidelines. They were given a list of 113 terms (including 94 specific terms and 19 more general terms) obtained from the literature (Campo et al., 2008;Sáenz-Navajas et al., 2011). During the training, different reference standards representative of aroma, taste and astringency were presented. Aroma standards were mainly natural products (fruits, juices, spices, vegetables, among others) prepared at the beginning of each session, or odorants supplied by Sentosphère (Paris, France), Firmenich (Geneva, Switzerland) and "Le Nez du Vin "(Jean Lenoir, Provence, France). For taste and astringency training, solutions with different concentrations of sugar (0-200 g/L) for sweetness, tartaric acid (0-2 g/L) for acidity, quinine sulfate (0-10 mg/L) for bitterness, and potassium and aluminum sulphate (0-5 g/L) for astringency were presented to the panel. These solutions aid to recognition and discrimination between different oral sensations. The training period comprised of two phases: a general phase and a specific phase of the type of product. During the general training phase (six sessions), the panellists became familiar with the attributes of the aroma and with intensity rating of sweetness, acidity, bitterness, astringency, aromatic intensity and global intensity as well as persistence. Then, they had to evaluate three to five different wines by describing their odour properties (orthonasally) by choosing up to five descriptors from the list and by rating tastes and astringency on a 10point scale (0 = "absence," 1 = "very low" and 9 = "very high"), global intensity on a nine-point scale (1 = "very low" and 9 = "very high") and global persistence on a nine-point scale (1 = "very short" and 9 = "very long"). The session ended with a discussion during which the panel leader compared the aroma descriptors and the taste intensity scores given by panellists to describe each wine. The wines selected for this training phase had intense and easily recognisable flavour and astringency properties, including red, white and rosé wines of various varieties and origins.
The specific training phase consisted of two sessions during which the panellists became familiar with the type of samples of the study. During this phase, panellists described the aromatic properties by choosing up to five terms from the list provided, and rated the intensity of sweetness, acidity, bitterness, astringency, aromatic intensity and global intensity, as well as persistence of commercial sweet wines.
In the evaluation of the wines, each panellist took part in a 45-minute session. 15 mL wine samples were presented in dark wine glasses approved by ISO (ISO, 1977) and coded with three-digit random numbers and arranged in random order. Samples were served at room temperature and covered with plastic Petri dishes. Panellists evaluated wines in separate individual booths. They first had to evaluate each wine orthonasally and rating the aromatic intensity and then describe the aroma by choosing a maximum of five attributes from the list of aromatic terms according to frequency citation method (Campo et al., 2008). Then, they were asked to rate sweetness, acidity, bitterness, astringency, global intensity and persistence of each sample using the abovementioned structured scales for each wine. Mineral water and unsalted crackers were provided for cleansing palate and rinsing between each sample. The sessions were held in a ventilated and airconditioned sensory analysis room (about 20 °C).

Analysis of panel performance
In order to assess panel performance, an average reproducibility index (R i ) was calculated for each panellist according to the method proposed by Campo et al. (2008). A value of R i greater than 0.2 was established to consider the judge's data. As for the attributes evaluated in mouth and aromatic intensity, a principal component analysis (PCA) was performed for each attribute. From the panellist's projections in the loading plot of the PCA, the agreement (grouped projections) and disagreement (spread projections) of panellists were evaluated. For each attribute, panellists' data that were spread were not considered. According to the R i and the PCA projections, the data of 22 panellists were considered for the statistical treatment of the data.

Product characterisation
Citation frequency of aromatic terms was calculated. Only the descriptors cited by a minimum of five judges (15% of the panel) in at least one of the wines were considered for subsequent statistical analysis. From these terms, those with citation frequencies higher than those expected by chance were chosen according to the chi-square distribution. Thus, 13 significant terms were considered (p < 0.05).
A one-way ANOVA (Analysis of Variance) was carried out on chemical analysis. A two-way ANOVA with judges as a block factor was performed on scores derived from in-mouth attributes and aromatic intensity. Significant differences were determined through Tukey test (p < 0.05).
Data derived from chemical analyses and descriptive sensory analysis were analysed using Principal Component Analysis (PCA) and Correspondence Analysis (CA). PCA was performed on the average ratings over the judges for in-mouth attributes sweetness, acidity, bitterness, astringency, global intensity and persistence and for each wine (correlation matrix) while CA was performed on the contingency table containing the mean citation frequencies of significant aroma descriptors. In order to choose the number of principal components that should be retained, dimensions with eigenvalue higher than the average eigenvalue (Kaiser's law) were used for both PCA and CA spaces.
Finally, a hierarchical cluster analysis (HCA) with the Ward criteria was applied to the factorial coordinates of the wines in the spaces defined by the PCA and CA according to these parameters. The clusters identified from the tree diagram were consolidated by aggregation around their mobile centres. In both cases, the terms that best characterised each of the clusters were identified by their probability p < 0.05. (Morineau et al., 1995).
All analyses were performed with IBM SPSS Statistics program (Version 20.0, SPSS Inc., Illinois, USA).

Chemical Characterisation
Average values of conventional oenological parameters and low molecular weight phenolic compounds determined in wine samples are shown in Tables IV and V. The calculated ANOVA indicates that there are significant differences between wines (p <0.05) for each analysed variable.
In order to simplify the results presentation and their discussion, phenolic compounds have been grouped into four different categories: acids and derivatives, hydroxycinnamic acids, flavanols and flavonols, as shown in Table III.

TABLE V
Phenolic compounds concentrations (mg/L) in wine samples

Concentrações de compostos fenólicos (mg/L) em amostras de vinho
There was a negative correlation between pH and titratable acidity, as expected (F = -0.779, p = 0.088) and positive correlation between hydroxybenzoic acids and derivatives (F = 0.681, p = 0.030) and flavanols (F = 0.669, p = 0.035). Volatile acidity is positively correlated with the concentration of sugars (F = 0.802, p = 0.005), which is logical in production of sweet wines (response to yeast osmotic stress), and negatively with flavonols (F = -0.748, p = 0.013). Ethanol content is inversely correlated with sugar content (F = -0.683, p = 0.030) and malic acid (F = -0.681, p = 0.030). This fact is due to a lower uptake of sugars in fermentation, malolactic fermentation carried out by some wines or a less fruit ripening state, since less mature grapes present higher concentration of malic acid and a lower probable alcoholic content. There is no correlation between the TPI and phenolic compounds determined, indicating that other groups of phenolic compounds are involved in the determination of TPI.
In order to assess similarities in the studied samples, a PCA was calculated with parameter values in Tables IV and V. Figure 1 shows the two-dimensional plot with the projection of samples and variables. The first dimension accounted for 35.5% of the total variance and it was mainly characterised positively by the concentration of phenolic acids and derivatives (89%), ethanol (79%) and flavanols (71%) and   In performed hierarchical cluster analysis (HCA), 9 components of the PCA were considered, accumulating 100% of the original variance. Table VI shows the yielded clusters and the parameters that best characterised each cluster. Figure 1 also shows the different groups. For each group, the wine closest to the centre of gravity was identified as the most typical example of the cluster.  The concentrations of different phenolic compounds determined in wine samples were within usual ranges found in white and rosé wines, in which there has not been a maceration with the skins as in the case of target wines. (Pereira et al., 2010;Gonçalves et al., 2012). Cluster 1 was characterised by a greater amount of hydroxybenzoic acids and derivatives, hydroxycinnamic acids and flavonols. It consisted of two wines in which the grape has been harvested in an advanced ripening stage, a late-harvest wine (NOPHO) and a wine from grapes frozen by frost in February (RMV1), which resulted in dehydration of the grapes on-vine. The phenolic composition of wines is affected by the ripening stage of the grapes (Ribéreau Gayon et al., 2006). Some studies have shown that there is an accumulation of polyphenolic compounds until overripening or late harvest in some grape varieties (Ivanova et al., 2011), and a concentration of those compounds is also produced by dehydration on-vine (become raisins) and accumulation in seeds (Rolle et al., 2009). Other authors (Tian et al., 2009a) observed that the contents of hydroxybenzoic acids, hydroxycinnamic acids and flavan-3-ols increased as grape harvest time was delayed to produce icewines in Vidal variety. This cluster is also characterised by a high alcohol content and a lower amount of sugars and titratable acidity, showing a higher fermentation yield with more mature grapes (decrease in acidity).
Cluster 2 was characterised by higher values of volatile acidity and sugars. High volatile acidity was typical in sweet wines production, since yeasts produce higher amounts of acetic acid as a result of the process of maintaining redox balance in response to the osmotic stress imposed by high sugar levels (Nurgel et al., 2004). In any case, the concentrations found in studied sweet wines were below the sensory threshold for acetic acid in icewines, 3.185 g/L (Cliff and Pickering, 2006) and even below that th threshold established for table wines, 1.3 g/L (Etievant, 1991). This cluster included two icewines (MAL and BLUE), a supurao wine (dehydration off-vine, OJU) and a wine obtained from grapes frozen in cold chamber (RMV). The first two wines did not have a complete fermentation while the last two wines have experienced a higher concentration of sugars reaching a similar alcoholic degree. This cluster presented intermediate values of phenolic compounds between the other clusters. The icewines of this cluster did not accumulated by dehydration on-vine as many phenolic compounds as icewines of cluster 1, which may be due to environmental, genotypic (variety) or technological (grape pressing) factors. Some studies have observed an increase in the concentration of phenolic compounds in Tempranillo and Merlot grapes (such as hydroxybenzoic acids, esters of hydroxycinnamic acids, flavanols and flavonols) by dehydration or drying inside a chamber under controlled conditions of temperature and humidity (Marquez et al., 2012), which may be similar to the drying process of grapes in supurao winemaking.
Cluster 3 comprised of four artificial icewines made from grapes frozen in freezing chamber. It is characterised by a high acidity and low pH, due to an earlier ripening stage in harvest time, without degradation of acids by overripening. In addition, three of the wines of this group (GLASSRIES, GLASSGEW and VERD) were made from white varieties, and therefore, were intended to achieve higher acidity values. Wines belonging to this cluster presented lower values of hydroxybenzoic acids, so the extraction and concentration of these compounds by freezing have been lower than in the cases of wines of the other two clusters. This fact may also be due to other factors such as variety (white versus red) and grape pressing (diffusion of phenolic compounds from the skin and seeds to the pulp).The presence and concentration of phenolic acids, flavan-3-ols and flavonols could differentiate between methods of artificial and natural icewines, as other authors revealed in Vidal variety (Tian et al., 2009b).
The low values of lactic acid indicated that wines of this group have not carried out malolactic fermentation, whereas wines belonging to the other two clusters presented quantities of lactic acid that indicated that they have undergone part of malolactic fermentation.
The chemical composition analysed in studied sweet wines had not been significantly different according to the different types of dehydration undergone by grapes (on vine vs. off vine, drying or freezing). So that, more research is needed about the mechanisms and factors that influence the chemical composition during dehydration, apart from the influence of the variety, origin and technology.

In-mouth attributes
According to two way ANOVA with judges as a block factor. The judge effect was significant on all attributes (p < 0.001). This effect is commonly found in sensory analysis and can be explained by interindividual differences (Sáenz-Navajas et al., 2011). The wine effect was significant for bitterness (F = 13.47, p < 0.001), acidity (F = 7.413, p < 0.001), sweetness (F = 13.70, p < 0.001), and persistence (F = 5.709, p = 0.004), while there were no differences in astringency (F = 3.308, p = 0.103) and global intensity (F = 1.406, p = 0.186). This indicates that these attributes are not useful to characterise and discriminate within this group of wines. Therefore, they were not taken into account for PCA analysis. Figure 2 shows the two-dimensional PCA plane showing the distribution of the samples on the first two dimensions, as well as the projection of variables. The first two components accumulated more than 90% of the total variance.
The first PC was mainly correlated with sweetness (87%) in positive way and bitterness (89%) in negative way, while the second component correlated positively with acidity (75%). The wines located to the right of the plane are the sweetest and least bitter, while those located at the top are the most acidic. Pearson's correlation analysis showed that sweetness was inversely correlated with acidity (F = -0.698, p = 0.025) and bitterness (F = -0.667, p = 0.035), and persistence was positevely correlated with bitterness (F = 0.745; p = 0.013). This fact agrees with that found in literature, which establishes a suppression of bitterness and acidity by sugar (sweetness) (Jackson, 2009a).  Figure 2 show the different clusters found by the HCA, together with the descriptors or attributes that best characterise the groups (p < 0.05).
The first cluster formed by a late harvest wine (NOPHO) stood out for its bitterness, which is related to higher concentrations of hydroxybenzoic acids, hydroxycinnamic acids and flavanols, as previosuly mentioned. In addition, the higher alcohol content in this wine could cause an increase in perceived bitterness (Fischer and Noble, 1994). The wine was located opposite to sweetness, which agrees with less sugar content.  Cluster 2 was characterised by a low perceived acidity (p < 0.05) and sweetness (p = 0.112), located in the lower right part of the PCA plane. It was constituted by supurao wine (OJU) and an artificial ice wine (RMV), which presented a high sugar concentration which could also mask the perceived sensation of acidity in mouth (Jackson, 2009a).
The third cluster formed by natural icewines (RMV1 and MAL) and an artificial icewine (RMV2) was determined by persistence (p < 0.05) and sweetness significant at 90% (p = 0.092). Those wines presented appreciable amounts of sugar and phenolic compounds (hydroxybenzoic and hydroxycinnamic acids and flavanols), which contribute to sweet and persistent sensations, respectively (Jackson, 2009a;Sáenz-Navajas et al., 2010b). A higher glycerol production takes place in fermentations with highe sugar levels due to osmotic stress that yeasts undergo. Glycerol may contribute to volume in the mouth, in addition to sweetness (Lubbers et al., 2001).
The last cluster consisted of three artificial icewines (GLASSGEW, GLASSRIES and VERD) and one natural icewine (BLUE), all made with white grape varieties. Therefore, this group was highlighted by their perceived acidity in mouth and their low persistence, which might be due to the lower pH and lower concentration values of phenolic compounds that these white wines presented. The greater acidity could contribute to taste quality by providing a fresh and lively character required in this type of wines,and balancing sweetness.

Aromatic attributes
According to two way ANOVA, the effect of wine was significant at the 1% level (F = 8.492; p < 0.001) for the attribute aromatic intensity. Wines with the highest intensity were MAL, RMV2 and NOPHO, which were made with red grapes or mixture of red and white grapes.
Discriminant terms can explain sensory differences found among samples. These terms were identified from the χ 2 test performed on frequencies data. 13 terms with a probability of 95% were selected: citrus fruits, lemon, cherry, dry fruits, prune, dry fig, banana/English candy, muscat, menthol/fresh, woody, alcohol, sulfur and glue.
In the correspondence analysis (CA) performed on citation frequencies of discriminant terms, the first eight factors were retained and they explained total variance. The projection of the samples and the aromatic terms on the two-dimensional CA plot is shown in Figure 3. The first factor, explained 32% of the original variance and was mainly defined by the attributes of banana/English candy (82%) and cherry (67%), which were opposed to dry fruits (48%), prune and dry fig (35%). The second dimension, which accounts for almost 19% of the original variability, opposed lemon (57%) and menthol/fresh (23%) attributes to sulfur and glue (37%) attributes.

Projeção de descritores de aroma e vinhos no espaço de análise de correspondência (dimensões 1 e 2).
Hierarchical Cluster Analysis (HCA) revealed that the five-partition option was the most appropriate. Table  VIII shows aromatic terms that best characterise samples in each cluster. Figure 3 also shows the different clusters. Cluster 1 was formed by NOPHO, a late harvest wine made from Tempranillo, Viura and Cagazal varieties. It was characterised by alcoholic notes, which agrees with its alcoholic level, and balsamic notes of menthol/fresh and lemon, which makes it a unique wine, as has been also seen in taste and in-mouth description. These aroma nuances could come from a greater presence of terpene varietal aromatic compounds (Setkova et al., 2007) that have been extracted into wine. Cluster 2 also consisted of a singular wine (RMV2, made with frozen grapes in a cold chamber) which stands out for banana/English candy and cherry notes. These descriptors are characteristic of esters (isoamyl acetate presents banana odour), alcohols and aldehydes that are formed in fermentation by yeast activity or by chemical esterification (López de Lerma et al., 2012). It is possible that the precursors of these compounds remained at high levels in Tempranillo red grapes as they were harvested at their optimum ripening stage and then they were concentrated by freezing and pressing. Technological factors such as yeast strain and fermentation conditions could also influenced the resulting aroma.
Cluster 3 consisted of natural ice wines (RMV1 and MAL), supurao wine and an artificial icewine (VERD). Their descriptors agreed with those found in the literature for sweet dessert wines from the Mediterranean area (dry fruits, nuts, raisins) (Genovese et al., 2007;Budic-Leto et al., 2010). Other authors (Reboredo-Rodríguez et al., 2015) have revealed changes in the aromatic profile and volatile composition of grapes for sweet wines production by bifferent dehydration techniques (on vine, off vine in controlled chambers, under the sun, standard conditions in closed environment, freezing), with different varieties and origins, and also have indicated common notes of raisins, figs, caramel, nuts, dry fruits, together with varietals nuances. However, more scientific research on the mechanisms of changes in aromatic profile change is still necessary, due to the enzymatic metabolism of grapes during the dehydration process. Also yeast metabolism during fermentation should be taken into account for the appearance of compounds responsible for these aromas. Other authors have suggested that grapes harvested before and the dried under controlled conditions undergo changes in flavour consistent with those occurring during fruit overripening (Moreno et al., 2008). Although VERD is a white wine made from frozen grapes in cold chamber, without a possible overripening, it presented these typical aromas of dry fuits and raisins, a fact that has also been observed by other authors in Riesling variety when compared to natural ice wines (Clary et al., 2006).
In the fourth cluster, ice wines from white grapes, both natural (BLUE) and artificial (GLASSGEW and GLASSRIES) were grouped, and typical aromas of Riesling and Gewürztraminer varieties predominated (Jackson, 2009b). Overripening notes did not emphasize in its aromatic profile.
The last cluster was formed by RMV (artificial icewine) and it was characterised by sulfur and glue off odours. The increased amounts of sulfur dioxide added to inhibit bacteria due to high sugar concentration in wine have caused the predominance of sulfur notes. As regards glue descriptor, it has been indicated above that a high production of volatile acidity (ethyl acetate, which presents glue aroma) by easts is due to osmotic stress in a medium with large amounts of sugars . Although this wine had the highest value of volatile acidity (1.19 g/L) and was below sensory threshold of acetic acid in icewines (3.1 g/L), it is possible that the quantity of ethyl acetate present in wine (not quantitatively determined) was above sensory threshold in ice and table wines (0.198 and 0.160 g/L, respectively) (Cliff and Pickering, 2006).
Through orthonasal aroma description, white wines and late harvest wine have been classified, but no differences have been observed between the other sweet wines from different dehydration processes (supurao, natural or artificial icewines). This finding agrees with a preliminary study, where white wines and late harvest wine were also separated from the other wines by a sorting method of sensory analysis (Rosáenz-Oroz et al., 2012).

CONCLUSIONS
Chemical analyses of conventional oenological parameters using FTIR technique and low molecular weight phenolic compounds using the UPLC-MS technique were suitable in order to characterise sweet natural wines presented in this study. Also, descriptive sensory analysis of in-mouth attributes and aromatic descriptors by citation frequency were adequate for organoleptic characterisation of target natural sweet wines. In view of these results, no significant differences have been established in chemical composition among different dehydration techniques of grapes from which sweet wines were obtained. Nevertheles, most of artificial ice wines (off-vine dried grapes) presented lower concentrations of phenolics, so they could be differentiated from wines obtained from on-vine dried grapes. Through sensory analysis of in-mouth attributes and evaluated aromatic descriptors, no significant differences among different processes of grapes dehydration have been observed. Aromatic descriptors of dry fruits and prune were observed in most wines from on-vine and off-vine dehydration. However, sensory analysis was able to differentiate late harvest wine, and white wines, without discrimination of dehydration techniques in this last group. More studies will be needed on factors affecting sensory parameters during dehydration.