Introduction

Research in viticulture mostly relies on measuring yield and grape composition to assess which cultural practices convey an improvement in vine performance and, as a consequence, could be worth implementing in field. This approach, though useful, does not allow a completely satisfactory evaluation, since researchers cannot assess to which extent the effects observed are transferred to wine composition, i.e.: to final product quality.
In order to overcome that limitation, some authors introduce small-scale fermentations in their experiments to obtain a more complete evaluation, which is widely recognized as a step-forward in research applicability. However, despite its relevance, little attention has been paid to evaluate to which extent reducing grape volume in small-scale winemaking, affects fermentation dynamics, wine composition and reproducibility. On the one side, some authors such as \cite{1999}, Dallas et al (2003) and \cite{Gonz_lez_Manzano_2004} González-Manzano et al. (2003) published results on grape seed extraction in model wine solutions, whereas other authors such as \cite{Lopes_2002} and Rossouw et al. (2012) studied yeast performance as influenced by commercial and small-scaled tanks. \cite{Santesteban_2013} also studied the changes occurring in grapes after frozing the samples to check the lack of repeatability between fresch and frozen grapes.  On the other side, as direct comparison of small-scale to commercial winemaking, \cite{Casalta_2013} compared the aromatic compounds of Chardonnay, and only three experiments used different fermentation volumes in red varieties. \cite{SCHMID_2009} compared three wine volumes (20, 50 and 300 kg) of a blend with Cabernet Sauvignon and Cabernet Franc), in an experiment that was focused on evaluating the suitability of frozen must, to finally report that winemaking outcomes were comparable among the three volumes compared. The same research team, Schmid and Jiranek (2011), compared fresh, frozen and blast-frozen grapes fermentation using two different volumes (80 and 500 kg), concluding that wines were similar in winetasting. Last, Sampaio et al. (2007) compared Pinot noir in very small-volume (3.5 kg) to a commercial fermentation (4540 kg), and observed that it was possible to effectively control oxidation and spoilage at this volume, although significant differences were observed in wine composition between both scales.\cite{Gordillo_2010}
Therefore, the existing research in this field is scarce. Thus, taking into account that small-scale winemaking conditions vary between experiments, as regional or winemaker preferences and protocol modifications may affect any stage of winemaking (i.e., yeast inoculation, cap management regime and malolactic fermentation), there is a clear need of learning how winemaking conditions and, particularly, tank size, affect the composition of the wines obtained. Moreover, the above mentioned research did not pay attention to repeatability, which is particularly relevant since, if high variability showed up, the interest of small-scale winemaking would be limited, as far as an additional source of variation could interfere with data analysis.
The aim of this work is to evaluate how repeatable and reproducible small-scale winemaking is. The differential aspect of this research is that 4 replicates are used with 4 different volumes in 2 distinct red varieties, and that the small-scale fermentation protocols used mirrored typical winemaking techniques commonly used in small wineries producing premium red wines worldwide.

Materials and Methods

Experimental design
This research was conducted in the experimental winery of the Enology Faculty in Tarragona, Spain, using grapes from the Faculty experimental vineyards (41º8'54'' N, 1º11'54''E, Altitude: 50m). Vineyards are located near the coast in the Designation of Origin Tarragona (Spain), which has a Mediterranean climate, soils are typically fertile and dense, and were managed according to standard practices in the region. Grapes from two distinct varieties were used, Tempranillo (TE) and Cabernet Sauvignon (CS), the former showing large berries and low-to-medium phenolic potential, and the latter small sized berries and high phenolic content. For both varieties, four different small-scale volumes (10, 25, 50 and 100 L) were compared. All vessels have a ratio height/diameter ranging between 1.4-1.5. All the tanks were made of stainless-steel, with a rubber gasket that helped to keep the lid tight. For each variety and tank volume, four replicates were vinified. Additionally, a commercial-sized large fermentation was performed in a 2500 L stainless steel tank.
Winemaking procedure
Grapes were handpicked at full ripeness into 20 kg boxes, and stored at 21ºC in a cold room before crushing. Grapes were de-stemmed and crushed individually for each tank volume using a Bücher Delta E2. Tanks were filled one by one 3/4s full in order to ensure an upper appropriate fermentation cap management. Room temperature during fermentation was kept at 23ºC; 40 mg L-1 sulfur dioxide was added to the must. All tanks were inoculated with 0.2 g L-1 ICV GRE yeast (Selection Inter Rhône, Lallemand®). The pomace was gently hand punched down twice a day until alcoholic fermentation was accomplished. During the tumultuous stage, must density and temperature were both measured daily, controlling sugar consumption and avoiding extremely high temperatures (higher than 28ºC) during the winemaking process. The pomace was pressed once fermentation was completely exhausted (reducing sugars < 2 g L-1), which took 8 days in TE and 12 in CS. Free run wines were then obtained by using a cone-shaped funnel (Lacor inox 18/20-diameter 22cm) to separate the pomace from the wine. Press wine was obtained using a 40L Hydropress with a capacity of juice yield of up to 20-25 liters per pressing, depending on variety and ripeness of fruit (http://www.vigopresses.co.uk). After pressing, the juice was settled overnight and racked to the same tank to promote clarity. Potassium metabisulphite was added (Winy Sepsa Enartis) to reach 20mg L-1  of Sulphur dioxide to prevent microbial spoilage. Wines were stabilized at 4ºC for 2 months, followed by racking before bottling in December, and kept at 4ºC for further storage. Finished wines were bottled without fining or filtering. The wines did not undergo malolactic fermentation to avoid unwanted apparent malolactic deviations, and no oak or ageing was made.
Grape analysis
All grape batches were analyzed before they were introduced in each tank. 100-berries from each variety were used to determine the sugar level, acidity and pH; another 300 berries were used to analyze phenolic maturity. Sugar content was determined using a handheld portable refractometer (Model 102/112/102bp). Titratable acidity (TA, g/L) was measured by titration with sodium hydroxide, and pH measured using a pH-meter (Crison Micro CM 2201). The modified Glories method, consisting of berry samples macerated at pH 3.6 instead of pH 3.2 \cite{Nadal_2010} was used to analyze phenolic maturity. Berries were blended (Oster Blender Classic 3 Model 4655) and macerated in an agitator Edmund Bühler GmBH SM-30, to determine total anthocyanins (TAnt) (Ribéreau-Gayon et al. 2000) and tannins (Ribéreau-Gayon et Stonestreet, 1965). 
Wine analysis
ABV, pH, TA, TAnt and tannins within the tank-size was analyzed. Anthocyanin contents were determined following the methodology detailed in Valls et al. (2009) through HPLC using a Hewlett Packard Liquid Chromatograph (Waters Corporation, Mildford, MA, USA) equipped with a Zorbax Eclipse Plus C18 Column (150x2.1mm 3.5µm) and a Zorbax Eclipse Plus-C18 Precolumn (12.5x4.6mm 5µm). Injection volume was 5 µL; elution was performed with a mobile phase A of water HPLC-grade (0.2% trifluoroacetic acid) and a mobile phase B using methanol (0.2% trifluoroacetic acid). Column temperature was set at 50 ºC. The HPLC was coupled to a Diode Array Detector (DAD). Quantifications were performed using the DAD detector, and identifications were made considering the TOF (time of flight). Mass spectrometry (MS) detector was used to assist for the identification. Content of free anthocyanins was determined using a calibration curve (based on peak area, y= 0.7968x+7.5756, R2=0.9774), which was established using malvidin 3-glucoside standard solutions submitted to the same procedure. The anthocyanidin-3-monoglucosides and respective acetylated and coumaroylated glycosides were identified on the basis of their UV-vis spectra and retention times (Table 1). The anthocyanidins were identified by HPLC by comparison with internal standards. Calibration curves were obtained by injecting standards with different concentrations of malvidin 3-glucoside (Extrasynthese, Genay, France). The range of the linear calibration curves was 0.1 to 1.0 mg/L for the lower (R2>0.996), 0.1 to 5.0 mg/L for intermediate (R2>0.987), and 10.0 to 200.0 mg/L for the higher concentration compounds (R2 >0.987). Unknown concentrations were determined from the regression equations and the results were expressed on mg of malvidin 3-glucoside. Repeatability of HPLC analysis gave a coefficient of variation < 7%.
Wine procyanidins were analyzed through RRLC (Rapid Resolution Liquid Chromatography) using a Zorbax Eclipse XDB-C18 50x30 1.8 µm. (SFF-C038), RRLC in-line filter, 4.6mm, 0.2µm. HPLC Injection volume was 1.4 µL, flux: 0.7 mL/min, Mobile phase A: water (0.1% formic acid), Mobile phase B: methanol (0.1% formic acid), Column temperature: 30 ºC. A Diode Array Detector (DAD) was used for the quantification and the TOFMS (Time of Flight Mass Spectrometer) was used for the identification. Table 2 shows the retention time and m/z for each compound.
Data analysis
The effect of tank size on wine composition was evaluated through one-way ANOVA, and, when P<0.05, Tukey post-hoc test was used. The comparison of small-scale wines to commercial-size tank was done using Principal Component Analysis (PCA), considering 2500 L tank as a supplementary individual, i.e. not including it to calculate the principal components (PC) but evaluating its performance. Statistical analyses were performed using R (R Core Team, 2014, Foundation for Statistical Computing, http://www.R-project.org/), using FactoMineR (Husson et al. 2016) and “factoextra” (Kassambara and Mundt, 2016) packages for PC calculation and graphical representation respectively.

Results

Grape composition
Grape composition before fermentation was very similar for all tank sizes (Table 3), and low variability occurred between the tanks of the same size (CV <5%). This fact is essential to guarantee that the differences eventually observed in wine composition were not due to differences in grape composition, but associated to the winemaking process.