A novel strategy to construct multi-strain starter cultures: an insight to evolve from natural to directed fermentation

Abstract Some biotechnological strategies have succeeded in the attempt to imitate natural fermentation, and bioprocesses have been efficiently designed when the product is the result of a unique biological reaction. However, when the process requires more than one biological reaction, few bioprocesses have been successfully designed because the available tools to construct multi-strain starter cultures are not yet well defined. In this work, a novel experimental strategy to construct multi-strain starter cultures with selected native microorganisms from natural fermentation is proposed. The strategy analyses, selects, and defines the number and proportion of each strain that should form a starter culture to be used in directed fermentations. It was applied to evolve natural fermentation to directed fermentation in distilled agave production. The results showed that a starter culture integrated by Kluyveromyces marxianus, Clavispora lusitaniae, and Kluyveromyces marxianus var. drosophilarum in proportions of 35, 32, and 33%, respectively, allows obtaining fermented agave juice containing a 2.1% alcohol yield and a distilled product with a broad profile of aromatic compounds. Hence, the results show, for the first time, a tool that addresses the technical challenge for multi-strain starter culture construction, offering the possibility of preserving the typicity and genuineness of the original traditional product.


Introduction
Microorganisms are found almost anywhere on earth due to billions of years of evolution and adaptation to their environment, giving rise to different ecosystems.In each ecosystem, natural biological processes of raw native material transformation are realized.The role of biological process in ecosystems is not totally clear, but their contribution to biogeochemical cycles to maintain equilibrium is essential [1][2][3] .In some ecosystems, the result of natural fermentation is products appreciated by their flavor, taste, or nutritional value, forming part of the diet of the local population [4][5][6] .Some products have been commercialized beyond their frontiers, but others have not because their shelf life is of a few hours under normal environmental conditions or of some days under refrigerated conditions because the composition changes due to the dynamic metabolism of the native strains [7] .This phenomenon implies that native strains that live in an ecosystem present a high level of interaction, where the activity of each strain has an advantage, disadvantage, or neutrality for one or more of the other strains, thus changing both the active population of the native strains and the substrate composition with time [2,6,[8][9][10] .
Several studies developed in wine-producing areas have shown that the interactions that evolved between microbial consortium members strongly modulate the final sensory properties of wine, and specifically, some synergistic interactions lead to the synthesis and accumulation of metabolites that improve wine flavor and taste complexity compared with wines obtained using only a selected Saccharomyces cerevisiae strain [11][12][13][14] .Moreover, the use of multi-strain synergistic cultures constructed with selected strains isolated from natural fermentation from a vineyard ecosystem could allow for a product with unique oenological traits to be obtained, hence the vineyard microbiome to the fore as a new insight into mimicking natural fermentation [11,14,15] .Unfortunately, most microbial consortia studies in natural fermentations fall significantly short of providing this quantitative knowledge due to the high number of microorganisms present, which makes it difficult to scrutinize.The closest thing has been population dynamic studies based on ARISA (Automated Ribosomal Intergenic Spacer Analysis) method [16] , and FCM (flow cytometry) technique, which monitor microbial community structures throughout fermentation [17,18] .Other works have been focused on studying synthetic microbial consortia composed of a subset of culturable native strains that simulate the natural fermentation in a tractable model system of reduced complexity which makes it easier to study interspecific interactions from different aspects, such as inoculum ratio, the timing of inoculation of strains, cell-cell contact, and production of inhibitory metabolites, to decipher the mechanisms underlying yeastyeast interactions and selecting the best mixture and inoculation technique to develop directed fermentation [13,15,19,20] .However, and despite these efforts, the overall interactions among wine yeast species in a fermentation modulated by multiple species remain unclear.Therefore, further efforts are needed to understand both how the modality affects the magnitude of such interactions and how each species contributes to fermentation, so there are still major challenges to developing tools that provide quantitative knowledge related to microorganism interactions and construct multistrain starter cultures [2,6,14,21] .
In previous works, we used an experimental mixture design named centroid simplex to select an optimum set from a pool of candidate microorganisms to form a starter culture for prickly pear wine production [22,23] .The results showed that a mixed culture leads to the production of a fermented product that contains alcohol and a broader range of desirable aromatic products, which led to the conclusion that a mixed culture can be used to obtain highquality fermented beverages.However, the centroid simplex design is impractical to apply when the number of strains (p) to study is high because the number of assays required (N) increases exponentially (N ¼ 2 p À1) [24] .Therefore, an alternative experimental design that allows the study of a relatively high number of strains with a reduced number of assays (screening design) is desirable to construct a synthetic microbial consortium prior to mixture optimization.
In this work, a novel experimental mixture design, named fractional mixture design which is a screening design, is proposed as a new tool for analyzing a higher number of variables or strains while using a fraction [N ¼ (pÂ(p À 1)) þ1] of the effort of a centroid simplex design in terms of experimental runs and resources, with the objective of discriminating variables and reducing their number to allow the use of a centroid simplex design to optimize the mixture that will be used as a starter culture for directed fermentations.These sequential two stages experimental strategy was applied to evaluate the native microbiota present in a multi-strain culture collected from a previously successful natural fermentative process, which will be used to subsequently inoculate other fermentations and develop directed processes with the native microbiota amplified.
This idea was born due to the gradual loss of efficiency in the original native starter culture that is associated with the lack of knowledge of what strains are present and prevents the development of culture conservation protocols, in addition to answering questions such as which of the strains present are essential and which are detrimental with relation to the quality final product, and what is the optimal mixture for the fermentation process.The goal of this study was to characterize the starter culture for distilled agave production as the first stage of the design directed fermentation process and to show a way to evolve natural fermentation that can be applied in all processes that use natural fermentation to produce traditional fermented food and beverages.

Raw material
Mature agave 'heads' or 'piñas' (plants without leaves) were harvested from eight-year-old (on average) of the species Agave Salmiana Otto ex Salm growing in San Luis de la Paz Gto.M exico (21 41 0 N, 21 04 0 S, 100 12 0 E, 100 45 0 W).Agave piñas were chopped and cooked in an autoclave using steam at 112 C (7.54 Psi) for 12 hours.The cooked agave was pressed to obtain agave sirup containing 140 g/L (on average) of fermentable sugars, which was mixed with a multistrain culture collected from a previous successful natural fermentation process in a 10:1 ratio between agave sirup volume and inoculum weight in a wet base that contains 25% (±2) of moisture.Along fermentation, the process variables were not controlled.

Microorganism isolation
Samples from fermented products were taken at different times and inoculated on a set of Petri dishes containing nutrient agar (Difco Lab.Sparks, MD.USA) for bacteria, a set containing potato dextrose agars (Difco Lab.Sparks, MD.USA) for yeast and a set containing agave sirup plus bacteriological agar.In all cases, classic plate-casting processes were used for inoculation according to Stanier et al. [25] .The inoculated Petri dishes were incubated at 28 C, 32 C, and 37 C for 24, 48, and 72 hours, respectively.Samples from colonies with different morphologies were taken and inoculated on new Petri dishes containing fresh nutrient, potato dextrose, or agave sirup agar and incubated under the same temperature and time conditions described above.The procedure was repeated until pure cultures were obtained.

Microscopy analysis and biochemical test
To ensure that the isolated colonies were pure cultures, a microscopic (Model DMRAX2, Leica Microsystems GmbH, Wetzlar, Germany) analysis based on morphological studies and Gram staining was performed.In addition, each isolated pure culture was analyzed by biochemical tests based on the API biochemical cards 20E, 20NE, 20 C and 32 C to increase the certainty that the isolates corresponded to different strains according to the results of the biochemical reactions.

Microorganism propagation
Each isolated pure culture was cultured in potato dextrose (for yeast) and nutritive (for bacteria) agar slants at 28 C for 24 h.Biomass samples taken from the slants were transferred to a 10 mL tube containing 15 mL of potato dextrose and nutrient broth and incubated for 24 h on a rotary shaker (Model 4520, Forma Scientific, Marietta, OH) at 28 C and 100 rpm for culture propagation.A 5 mL sample of each culture was loaded in a Neubauer chamber, covered with a slip, and placed under the microscope for cell counting.

Screening native strains
To discriminate against unnecessary strains, directed fermentations using pure cultures as starter inoculum was developed.Each assay was performed by mixing 1 mL from each pure culture containing 1 Â 10 6 UFC with 100 mL of agave sirup in Erlenmeyer flasks and incubated for 24, 48, and 72 h on a rotary shaker at 28 C and 50 rpm.The fermented products were sensory evaluated and divided into two groups: one formed by fermented products with pleasant oenological sensory traits or positive group (e.g., odor or aroma to alcohol, herbal, floral, or fruity) and the other formed by products with unpleasant sensory traits or negative group (e.g., odor or aroma to rancid, leather, damp, rotten, acid, spicy, etc).With the positive group strains, the sequential two stages experimental strategy was performed (Figure 1).

Fractional mixture design
These designs are expressed using the notation [N ¼ (pÂ(p À 1)) þ1], where "p" is the number of variables and "N" describes the size of the experimental scheme to construct.The experimental scheme construction is as follows: the scheme must contain as many blocks as variables to study.Each block, except the last (centroid), must contain as many assays as components to be studied, and each variable must be evaluated the same number of times.The last block contains only the mixture that includes all variables.The assays that contain each block, except the last (centroid), are defined by aleatory form by their selection from a full simplex-centroid design or performing cyclical permutations in each block, beginning in the first cell of each block with the value described in the block, and ending with this value in the last cell of the column "p" for the same block (Supplementary table).
As in the screening study, the mixture assays for both fractional and simplex-centroid designs were evaluated by mixing 1 mL of inoculum containing 1 Â 10 6 UFC (in the proportions of each culture according to experimental design) with 100 mL of agave sirup and incubating for 24, 48 and 72 h on a rotary shaker at 28 C and 50 rpm.The fermented samples were evaluated as the percentage of alcohol present in the fermented extract according to Pilone [26] , the profile of the synthesized alcohols was analyzed by gas chromatography, and the profile of the volatile components was analyzed by gas chromatography and mass spectrometry.The results were used to construct a canonical polynomial model, which was analyzed by statistical methods for a probability distribution value (PV) of 0.05 and solved by the Levenberg-Marquardt method [27] .

Sample preparation for gas chromatography and mass spectrometry (GC-MS) analysis
The fermented product was distilled and treated by solidphase microextraction according to Peña-Alvarez et al. [28]

Successful natural fermentaƟon
Figure 1.Sequential two stages experimental strategy to construct multistrains starter cultures.
using a polydimethylsiloxane-divinylbenzene fiber (PDM/ DVB, Supelco, Bellefonte, Pa., U.S.A.).Samples of 8 mL were collected in a vial and incubated at 28 C for 50 min.After this time, the microfiber (PDM/DVB) was placed inside the vial, and the sample was incubated further at 40 C for 40 min to allow volatile compound adsorption.Finally, the sample was injected into the GC-MS system.

Gas chromatography and mass spectrometry (GC-MS) analysis
The samples were injected into a gas chromatograph coupled with a mass spectrometer (GC-MS) (Clarus 500MS, Perkin-Elmer Inc. Wellesley, Mass., U.S.A.).The operation conditions for GC analysis and the programming for MS analysis were carried out according to Navarrete-Bolaños et al. [29] .The signals of the resulting chromatogram were tentatively identified by comparing their mass spectra with those of the MS database in the Natl.Inst. of Standards and Technologies (NIST2002) library.

Sensory analysis
Development of a new beverage involves factors beyond technical ones, such as market factors that consider the needs of and acceptance by the consumer.For acceptance, we used a sensorial analysis based on descriptive tests evaluated on a ten-point scale and carried out by ten trained panelists from the sensory analysis laboratory of the Mexico Nacional Technological in Celaya.The test scores were used to establish visual, olfactory, gustatory, and aftertaste agave distilled descriptors.We also performed a consumer survey for discrimination (to determine whether a difference exists between the agave distilled products) and preference (to identify liking or acceptability).

Data and statistical analysis
The results from the experimental designs were used to mathematical model construction via least squares method followed of the analysis of variance using a statistical framework for a confidence interval of 95% (a ¼ 0.05).For sensory analysis, the descriptors were classified using the comparison of geometric means, and the data were statistically analyzed using multivariate techniques.The statistical analysis was performed using Statgraphics Centurion software (Statgraphics Technologies, Inc. Plains, VA.USA).

Mathematical mixture models
Mixture models differ from usual polynomial models because of the constraints 0 x i 1:0 and R p i¼1 x i ¼ 1:0 Â Ã : Hence, these models estimate one less term by removing the intercept, and the order of the model is given with respect to the numbers of variables in the term and not the exponent of the variable, resulting in the so-called canonical polynomials, i.e., where each b i is the expected response for the pure mixture x j¼0 , j 6 ¼ I, which represent the linear blending portion, and b ij the quadratic blending also referred to as synergism (positive term sign) or antagonism (negative term sign) as the blending is non-linear, is often necessary when the linear relationship is not sufficient [24] .

Strain identification
Once the strain was selected, molecular identification was performed, which included ribosomal sequence analysis of the D1/D2 domains of the 28S rRNA gene and of the ITS region (ITS1: 5 0 -TCCGTAGGTGAACCTGCGG-3 0 ) and ITS2: 5 0 -GCTGCGTTCTTCATCGATGC-3 0 , intergenic spaces and which includes the 5.8S rRNA gene).The method is based on genomic DNA extraction from selected strains, according to the protocol proposed by Ausubel et al. [30] using an Invitrogen TM

Results
Microbial consortia analysis in "natural" fermentation.Based on standard microbiology techniques, microscopy analysis and biochemical tests, twenty-two (X 1 , X 2 , X 3 , X 4 , … , X 22 ) pure cultures were isolated from samples "naturally" fermented.Nine (X 1 , X 2 , X 8 , X 10 , X 11 , X14, X 15 , X 16 , and X 18 ) presented characteristics of yeast, and thirteen (X 3 , X 4 , X 5 , X 6 , X 7 , X 9 , X 12 , X 13 , X 17 , X 19 , X 20 , X 21 , and X 22 ) presented characteristics of bacteria.Each pure culture was propagated and inoculated into agave sirup, and the fermented product was analyzed.The results showed that only the fermented extracts inoculated with strains X 1 , X 2 , X 8 , X 10 , X 11 , X 14 , X 15 , X 16 , and X 18 showed positive results, i.e., they contained alcohol or pleasant aromatic notes like herbal, floral, or fruity.With these strains, a fractional mixture design was constructed (Table 1) and performed, and the data were analyzed.The results (last column of Table 1) were fitted to a quadratic model constructed using the least squares method, generating the following mathematical expression: þ 5:42X 14 X 15 À 6:55X 14 X 16 À 2:59X 14 X 18 À 6:21X 15 X 16 À 9:08X 15 X 18 þ 6:63X 16 X 18 The model was analyzed by statistical tools based on analysis of variance (Table 2), which showed, for a probability distribution value (P-V) of 0.05, that the model exhibited a statistically significant relationship between alcoholic fermentation yield (Y OH ) and the microbial culture (variables) used as the starter inoculum.
In the ANOVA table, the R-Squared statistic indicates that the model as fitted explains 73.8715% of the variability in Y OH .The adjusted R-squared statistic, which is more suitable for comparing models with different numbers of independent variables, is 32.8123%.The standard error of the estimate shows the standard deviation of the residuals to be 0.278751.The mean absolute error (MAE) of 0.129278 is the average value of the residuals.Since the P-value is near 5.0%, there is an indication of serial autocorrelation in the residuals at the 5.0% significance level.
Based on these preliminary results, a second mixture design defined as the simplex centroid for five variables [X 1 , X 2 , X 14 , X 16 , and X 18 ] was constructed and performed (Table 3).
The results (last column of Table 3) were analyzed using a combination of both descriptive and inductive statistics.First, the results were fitted to a quadratic model constructed using the least squares method, generating the following mathematical expression: The model was analyzed by statistical tools based on analysis of variance (Table 4), which showed, for a probability distribution value (PV) of 0.05, that the model is not a statistically significant relationship between Y OH and the microbial culture at the 95.0%confidence level.The R-Squared statistic indicates that the model as fitted explains 49.8629% of the variability in Y OH .The adjusted Rsquared statistic, which is more suitable for comparing models with different numbers of independent variables, is 5.99289%.The standard error of the estimate shows the standard deviation of the residuals to be 0.251602.The mean absolute error (MAE) of 0.136846 is the average value of the residuals.Since the p-value is greater than 5.0%, there is no indication of serial autocorrelation in the residuals at the 5.0% significance level.
Based on the above, the results were also fitted to a special cubic model, obtaining the following mathematical expression Again, the model was analyzed by statistical tools based on analysis of variance (Table 5), which showed, for a probability distribution value (PV) of 0.05, that the model is not a statistically significant relationship between Y OH and the components at the 95.0%confidence level.
However, the R-Squared statistic indicates that the model as fitted explains 79.2364% of the variability in Y OH .The adjusted R-squared statistic, which is more suitable for comparing models with different numbers of independent variables, is 0.0%.The standard error of the estimate shows the standard deviation of the residuals to be 0.264405.The mean absolute error (MAE) of 0.0707307 is the average value of the residuals.
In this model, the coefficients of the linear terms were very similar, indicating that these microorganisms alone would achieve similar yields in the fermentation process.The nonlinear terms for the X 1 X 2 , X 1 X 14 , X 2 X 14 , X 1 X 2 X 16 , X 1 X 14 X 16 and X 2 X 16 X 18 interactions had negative coefficients of regression, therefore there were antagonistic effects among these strains.The rest of the interactions showed synergistic effects, and the interaction among the X 1 , X 16 , and X 18 strains had the highest value of the coefficient of regression and therefore was the most significant in the process yield.The numerical solution of the model based on the Levenberg-Marquardt method and the restrictions [0 x i 1:0 and R p i¼1 x i ¼ 1:0] of the system showed the following results: X 1 ¼ 0.35, X 2 ¼ 0.00, X 14 ¼ 0.00, X 16 ¼ 0.32, and X 18 ¼ 0.33.These results indicate that a mixture of strains X 1 , X 16 , and X 18 in proportions of 35, 32, and 33%, respectively, allows us to obtain a fermented product containing an alcohol yield of 1.9.Considering that the inoculum must contain 1 Â 10 6 CFU/mL, then the multistrain starter culture must be formed by 350,000 CFU/mL of X 1 , 320,000 CFU/mL of X 16 , and 330,000 CFU/mL of X 18 .
Additionally, using inductive statistical analysis on the special cubic model, one can construct the response surface graphics, which show the existence of synergic effects among microorganisms X 1 , X 16 and X 18 (Figure 2a) and antagonistic effects between microorganisms X 2 and X 14 with X 1 , among X 2 with the mixture X 1 and X 16 , and among X 16 with the mixture X 1 and X 18 (Figure 2b-d).These observations lead to the exclusion of X 2 and X 14 from the optimum mixed culture for alcoholic fermentation and allow us to establish that the optimum composition of the mixed culture that maximizes the alcoholic fermentation yield corresponds to the values obtained from the analytical solution.Verification assays using a mixture of X 1 , X 16 , and X 18 as the starter inoculum in proportions of 35%, 32%, and 33%, respectively, on agave sirup containing 120 g/L fermentable sugars gave a 2.1% (average) alcohol yield, which compares favorably with the predicted yield value (1.9%).

GC-MS analysis
Directed fermentations using the constructed mixed culture and pure cultures of strains X 1 , X 16 , and X 18 as the starter inoculum were performed, and the products were analyzed by GC-MS to determine the profile and abundance of the main volatile components.The resulting chromatograms showed differences in the volatile compound profiles among the distilled products (Figure 3a-d), with the distilled product obtained using the mixed culture having the highest profile and abundance of the main volatile components, seemingly as a result of the individual strain contributions and confirming the synergic effect among the strains as resulted by the mathematical model.The importance of each component present in the distilled products was defined based on previously published sensory analysis reports (Table 6) [31][32][33][34] .According to these results, the volatile compounds found in all distilled products are included in the group reported for the main agave distilled products: tequila and mezcal.

Sensory analysis
The distilled products could be clearly differentiated by sensory analysis, highlighting the quality of the distilled product obtained using the mixed culture.According to the panelists' appreciation, it was described as: in style: spicy, herbal, and complex; in aroma: smoked meats and nuts, cream, and herbal stewed vegetables and beans; in flavor and taste: intricate roasted tropical fruit, smoked peppers, and grass; in smoothness: silky Finish; as the bottom line: A marvelously complex and stylish distilled product with unique distinctive and pleasurable traits.

Strains identification
In addition to defining the optimum culture, the identity of the microorganisms that must be present in the starter culture was achieved via the blast algorithm of the National Center of Biotechnology Information (NCBI).The results show that the microorganisms that integrate the optimum consortium are X 1 : Kluyveromyces marxianus, X 16 : Clavispora lusitaniae, and X 18 : Kluyveromyces marxianus var.drosophilarum.Therefore, the starter inoculum in the alcoholic fermentation process to obtain mezcal must be formed by Kluyveromyces marxianus, Clavispora lusitaniae, and Kluyveromyces marxianus var.drosophilarum, in proportions of 35%, 32%, and 33%, respectively.These results are according to previous studies that have shown that the Kluyveromyces species yeasts are high ethanol yield and tolerance, and can synthesize pleasant aromatic compounds that increase the quality of the fermented food and beverages, besides other biotechnological advantages over classical fermentative yeasts such as the Saccharomyces species [35] .Respecting to Clavispora lusitaniae strain, it is an environmentally ubiquitous yeast, with no known specific ecological niche, which has been isolated from different traditional fermented food and beverages, from normal animal microbiota, as well as from clinical samples collected from patients with oral lesions [36][37][38] .Related to agave processing, P erez-Brito et al. [39] isolated C. lusitaniae in different stages of the processing agave fourcroydes to produce distillates, and Verdugo-Valdez et al. [40] isolated C. lusitaniae in the alcoholic fermentations of two separate musts obtained from Agave salmiana processing.So, the results of this study help to confirm that C. lusitaniae is part of the native microbiota of the spontaneous fermentation of agave sirup, but also shows that C. lusitaniae can be a promising microorganism to be considered to perform the alcoholic fermentation on agave sirups.
Process production of agave distilled using a multi-strain starter culture.Pure cultures of K. marxianus, C. lusitaniae, and K. marxianus var.drosophilarum were cultured in potato dextrose agar slants at 28 C for 24 h.Biomass samples were taken from each culture and transferred to a 250 mL Erlenmeyer flask containing 100 mL of potato dextrose broth and incubated for 24 h on a rotary shaker at 28 C and 100 rpm for cultures propagation.Aliquots from each flask were collected and mixed in proportions of 35%, 32%, and 33%, respectively to construct the multi-strain starter culture and used to inoculate agave sirup containing 120 g/L of fermentable sugars in a volume: volume ratio of 1:10 (inoculum: sirup).The alcoholic fermentation process was developed for 72 h at 33 C, and then distilled.The main product is a liquid clear-crystalline containing 55% (v/v) of alcohol on average and a broad profile of volatile compounds that together provide a pleasant aroma and flavor sensation.

Discussion
For years, technical procedures for producing traditional fermented food and beverage have continuously evolved, from the discovery of spontaneous fermentations to the industrial application of pure selected starter cultures.In recent years, the use of multi-strain starter cultures has received great attention because it improves wine flavor and taste complexity, and moreover, when used native strains allow for a product with unique oenological traits associated with vineyards [12][13][14][15] .
Therefore, a tool to construct synergistic multi-strain starter cultures that include both the number of strains and the proportion of each strain in the mixture was, until now, the unattained cornerstone required to evolve from the use of one specific strain to multi-trains as a starter culture for fermentative processes [2,6,14,21] .Now, based on the results obtained in this work, the novel methodology and fractional mixture design proposed and supported by the construction, development, and data analysis emerges as a viable alternative for constructing a multi-strain culture and the fractional mixture design as an important statistical contribution to the exploration of the effects of several variables on a response when the number is high, and a classical mixture design cannot be applied.
The validity of the protocol was carried out in agave spirit production, obtaining an alcoholic product with a broad profile of desirable volatile compounds that enhance the quality of the product, which favorably compared with the product obtained using only one specific strain as a starter culture, confirming that a mixture of strains provides a better-quality product in comparison with that obtained using a pure culture.Hence, the results achieved show, for the first time, a resolution of the technical challenge for constructing multi-strain cultures that act synergistically and allow obtain fermented products closest to those obtained from successful spontaneous fermentations.
The native microbial consortia constructed confirmed that yeasts of the Kluyveromyces species are important to increase the quality of fermented alcoholic beverages and showed C. lusitaniae as a promising yeast to participate in alcoholic fermentation, at least, in agave sirups to increase the quality of distilled products.
In respect to developed methodologies, this is similar to the response surface methodology which is probably one of the most widely used optimization methods in recent years.Both begin with a screening design (a factorial design for parameters or a fractional mixture design for mixture components) prior to response-surface design (e.g., a composite central design for parameters or centroid mixture design for mixture components).Hence, this novel methodology could be of great importance in the industry.
In respect to fractional mixture design, this is a novel mixture design that allows reliable results to be obtained and saves time and resources by reducing the number of experiments when the mixture component number is high.In addition, the fractional mixture design has the advantage of reducing the number of experimental assays as the number of mixture components increases related to simplex centroid design.With up to three variables (p ¼ 3), the experimental scheme for the fractional mixture design [N ¼ (pÂ(p À 1)) þ 1] is equal to an experimental scheme for the simplex centroid design [N ¼ 2p À 1].However, for four variables, the number of assays for the fractional mixture design is two less assays than for the simplex centroid design, for five variables it is ten assays less, and so on.
Therefore, the experimental strategy developed in this work has shown that it could become a powerful tool for solving numerous problems related to microbial ecology to study the relationship between the components of an ecosystem and productivity, and to construct multi-strain cultures with selected native microorganisms, which may be used as a model to impel the production of traditional fermented foods and beverages around the world and to obtain products with standardized quality.

Figure 3 .
Figure 3. Relative volatile compounds abundance found on the agave distilled obtained from directed fermentations using a mixed culture of K. marxianus, C. lusitaniae, and K.marxianus var.drosophilarium and using pure cultures.

Figure 2 .
Figure 2. Contours of estimated response surface used to find the optimum consortium composition.

Table 1 .
Fractional mixture design for nine pure cultures or nine variables and output function of each assay.

Table 2 .
Analysis of variance for the quadratic model constructed for fractional mixture design results.

Table 3 .
Experimental centroid mixture design for five variables or microorganisms and output function of each assay.

Table 4 .
Analysis of variance for the quadratic model constructed for centroid mixture design results.

Table 5 .
Analysis of variance for the special cubic model constructed for fractional mixture design results.

Table 6 .
Identified volatile compounds and abundance in the fermented products obtained from directed fermentations using a mixed culture of X1, X16 and X18, and the pure cultures of X1, X16, and X18.