Introduction

Tempeh is a traditional Indonesian food of high nutritional quality, obtained by fungal fermentation of dehulled, soaked, and cooked soybean grains (^{Starzyǹska-Janiszewska et al., 2014}). It is commercially available in different forms, such as fried, boiled, steamed or roasted. In general, the tempeh is obtained by fermentation of soybean grains with *Rhizopus oligosporus*, and it is an excellent protein source that has all the essential amino acids for human nutrition, in addition to dietary fibers, calcium, vitamin B12, folic acid, low sodium content, and unsaturated fats (^{United States Department of Agriculture, 2008}). Moreover, fermented soybean foods contain several health beneficial bioactive compounds (^{Murooka & Yamashita, 2008}) of which the most important are folate, vitamin B12, and isoflavones (^{Mo et al., 2013}).

The preparation process of soy tempeh includes soaking soybean grains, cooking their cotyledons and fermentating them by the inoculation with *Rhizopus oligosporus*; however, these processes are not standardized. Usually, soybean grains are soaked in water for 10 to 12 hours, at room temperature, in order to facilitate the removal of the hulls. Cotyledons are then cooked in boiling water, drained, cooled at 35 to 40°C, inoculated with microorganisms, and fermented at 37°C for 24 to 48 hours (^{Nout & Kiers, 2005}). *Rhizopus oligosporus* is the most often used fungus for fermentation of soybean tempeh, due to its rapid growth between 30 and 42°C (^{Shurtleff & Aoyagi, 2001}).

Fermentation is the major processing stage for obtaining tempeh, due to the changes that occur in the sensory characteristics and chemical composition of the food, mainly associated with the bioconversion of isoflavones (^{Chang et al., 2009}). The fermentation process depends on soybean tempeh processing conditions, such as time, temperature, pH, inoculum amount, and relative moisture content (^{Nout & Kiers, 2005}). In this stage, β-glycosidase hydrolyzes β-glycoside forms to aglycones, which are more bioactive and absorbable isoflavones in the intestine, compared to their glycosidic forms (^{Izumi et al., 2000}; ^{Liggins et al., 2000}). Isoflavones, and mainly the aglycones forms, have drawn attention because of their ability to reduce the risk of cardiovascular diseases (^{Zhuo et al., 2004}), to inhibit cancer cell growth (^{Sarkar & Li, 2003}; ^{Lund et al., 2004}), to alleviate the menopause symptoms (^{Messina & Hughes, 2003}), and to prevent diseases such as osteoporosis (^{Arjmandi et al., 1996}). The bioconversion of β-glucoside forms into aglycones (daidzein, genistein and glycitein) requires hydrolysis by β-glycosidase.

^{Nakajima et al. (2005)} concluded that the content of three isoflavones aglycones (daidzein, glycitein and genistein) in soybean tempeh had increased about two times after 24 hours of fermentation. During soybean tempeh processing, glycosides isoflavones are converted to their corresponding aglycones, increasing the concentrations of daidzein and genistein, and decreasing the ones of some isoflavones (^{Haron et al., 2009}). The bioconversion of glycosides isoflavones into aglycones occur during fermentation, with higher contents found after long fermentation times (^{Villares et al., 2011}). ^{Ferreira et al. (2011)} showed that the total glycoside forms (daidzin and genistin) in soybean cotyledons decrease, and that the malonyl forms, which react unsteadily to heat treatments, reduced greatly as a result of cooking. These authors also mention that aglycones forms (daidzein and genistein) increase after 24 hours of fermentation. ^{Bavia et al. (2012)} observed a 43% reduction in malonyl forms after cooking soybean grains, and a 50% increase in the aglycones levels after fermentation.

The objective of this study was to evaluate the effect of the processing conditions of soybean tempeh on the contents of β-glycosides isoflavones and on their bioconversion into aglycones.

Material and Methods

Soybean cultivar 'BRS 267' harvested in Londrina, PR, Brazil, in 2011/2012, was used to prepare the soybean tempehs. Lyophilized *Rhizopus microsporus* var. *oligosporus* was obtained from Intsoy (International Soybean Program, University of Illinois, EUA). Isoflavone standard solutions were prepared from 6''- O-acetylglucosides and 6''- O-malonylglucosides (Wako Pure Chemical Industries, Ltd., Osaka, Japan), and from β-glucosides and aglycones (Sigma-Aldrich Co., St. Louis, MO, USA). The other reagents were of analytical grade and obtained from different sources.

To evaluate the effects of tempeh processing variables on the bioconversion of β-glycoside isoflavones into aglycones, a central composite design (2³) was used, with 3 replicates of central point, totaling 17 random assays. The tempehs were prepared according to the traditional method described by ^{Wei (1991)}, with modifications. The following processing variables were investigated: X1 (soaking time, h), X_{2} (cooking time, min), and X_{3} (fermentation time, h). For each assay, 100 g of soybean grains (proportion 1:10, w/v) were soaked (X_{1}), and grains were manually dehulled. Cotyledons were cooked in boiling water (X_{2}), and then they were drained and cooled at 25°C. For the inoculation, 2 g of *R. oligosporus* were used per 100g of cooked and cooled cotyledons. After inoculation, the cotyledons were homogenized, packaged in perforated polypropylene bags, maintained at 37°C, and fermented (X_{3}) according to the experimental design. Table 1 shows the coded and real levels of the independent variables. These levels were established in a preliminary test. The produced tempehs were lyophilised in a freeze-dryer, model Alpha 2-4 LD plus (Martin Christ, Osterode am Harz, Germany) and stored at -26°C, for the extraction and quantification of isoflavone contents.

Coded variables: x_{1}, x_{2}, and x_{3}. Real variables: X_{1}, hydration time (h); X_{2}, cooking time (min); and X_{3}, fermentation time (h). Y_{1}, μmol of genistin g^{-1} of tempeh; Y_{2}, μmol of malonyldaidzin g^{-1} of tempeh; Y_{3}, μmol of malonylgenistin g^{-1} of tempeh; Y_{4}, μmol of daidzein g^{-1} of tempeh; and Y_{5}, μmol of genistein g^{-1} of tempeh

The response functions (Y) to different isoflavone forms in tempeh, Y_{1} (μmol genistin g^{-1}), Y_{2} (μmol malonyldaidzin g^{-1}), Y_{3} (μmol malonylgenistin g^{-1}), Y_{4} (μmol daidzein g^{-1}), and Y_{5} (μmol genistein g^{-1}), were determined according to the following mathematical model:

Y = β_{0} + β_{1}x_{1} +β_{2}x_{2} +β_{3}x_{3} +β_{11}x_{1}
^{2}+β_{22}x_{2}
^{2}+β_{33}x_{3}
^{2}+β_{12}x_{1}x_{2} + β_{13}x_{1}x_{3} + β_{23}x_{2}x_{3} + ε

in which: Y is the response function; x_{1}, x_{2}, and x_{3} are coded variables; β is the estimated coefficient of the response surface model; and ε is the pure error. The response functions were evaluated by regression analysis and analysis of variance. The equation model was fitted to experimental data to propose the model. Response surface graphs and desirability parameters were generated for each response function, in order to obtain a tempeh with the maximum content of aglycones. All analyses, desirability parameters, and response surfaces were performed with the software Statistica, version 8.0 (StatSoft, Inc.,Tulsa, OK, USA).

The lyophilised and milled tempehs were defatted with hexane, in a 1:10 ratio (w/v), for 1 hour at room temperature by continuous and rotary agitation, followed by vacuum filtration. The isoflavone extraction was performed in triplicate with 0.1 g of the defatted sample, using 2 mL of an extraction solution containing ultrapure water, acetone and ethanol (1:1:1, v/v/v) (^{Yoshiara et al., 2012}). The separation and quantification of isoflavones were performed in triplicate, using an ultraperformance liquid chromatography (UPLC) system, as described by ^{Handa et al. (2014)}. Since different isoflavone forms have different molecular masses, in order to improve the comparisons, the isoflavone concentrations were expressed in micromoles of isoflavones per grams of lyophilized and defatted tempeh.

Results and Discussion

Grains of 'BRS 267' soybean cultivar had the following isoflavone components quantified by UPLC: 1.48 μmol β-glucoside isoflavones g^{-1} (1.12 μmol genistin g^{-1}, and 0.36 μmol daidzin g^{-1}), 4.25 μmol malonylglucosides g^{-1} (3.04 μmol malonylgenistin g^{-1}, and 1.21 μmol malonyldaidzin g^{-1}), and only 0.33 μmol genistein g^{-1}. The aglycone forms were present in lower concentrations than the other isoflavone forms (^{Carrão-Panizzi et al., 2003}). However, this same cultivar, harvested at different times and with isoflavones quantification by high performance liquid chromatography (HPLC), showed the following contents (^{Silva et al., 2009}; ^{Bavia et al., 2012}): 2.16 μmol daidzin g^{-1}, 1.71 μmol genistin g^{-1}, 2.33 μmol malonyldaidzin g^{-1}, 1.70 μmol malonylgenistin g^{-1}, 0.24 μmol daidzein g^{-1}, and 0.80 μmol genistein g^{-1}. This differences in isoflavone contents can be attributed to the genetic variability and to different environmental conditions (^{Wang & Murphy, 1996}; ^{Carrão-Panizzi et al., 2003}).

The X_{1} variable (soaking time, h) showed a significant linear negative effect (β_{1}x_{1} = − 0.05x_{1}, p<0.05) on the Y_{1} response function (μmol of genistin per g of tempeh). The effects of the X_{2} (cooking time, min) and X_{3} (fermentation time, h) were not significant (p>0.05). Based only in the significant variable, the mathematical model Ŷ_{1} = 0.24 − 0.05x_{1} was developed, indicating that the genistin level in tempeh is independent of the cooking and fermentation times. The lack-of-fit of the model was not significant (at 95% confidence interval), and 77% of the experimental data were properly adjusted (R^{2}) to the model. The response surface (Figure 1 A) and the desirability parameters (Figure 1 B) showed a region where Y_{1} was maximum (Ŷ_{1} = 0.40 μmol g^{-1}), and obtained when the following conditions were used: x_{1} = − 1 or X_{1} = 6 h of soaking; x_{2} = 0.5 or X_{2} = 22.5 min of cooking; and x_{3} = − 1 or X_{3} = 18 h of fermentation with *Rhizopus oligosporus*. These processing conditions of soybean tempeh agreed with the assay 3, in which the experimentally obtained Y_{1} was 0.40 μmol of genistin per g of tempeh.

Genistin content in the tempeh was 36% lower (0.40 μmol g^{-1}) than in the grains (1.12 μmol g^{-1}), and daidzin content was neglectable (while its contents in the grains were 0.36 μmol g^{-1}).

For the Y_{2} response function (μmol of malonyldai-dzin per g of tempeh), the X_{2} (cooking time, min) and X_{3} (fermentation time, h) variables showed significant linear negative effects (β_{2}x_{2} = − 0.21x_{1} and β_{3}x_{3} = − 0.05x_{3}, respectively), and the interaction between X_{1} and X_{2} resulted in a significant linear positive effect (β_{12}x_{1}x_{2} = 0.07x_{1}x_{2}) (p<0.05). The other effects were not significant, and the following mathematical model was developed: Ŷ_{2} = 0.32 − 0.21x_{2} − 0.05x_{3} + 0.07x_{1}x_{2}. The lack-of-fit of the model was not significant (at 95%), and 95% of the experimental data were properly adjusted (R^{2}) to the model.

For the Y_{3} response function (μmol malonylgenistin per g of tempeh), the X_{1} (soaking time, h), X_{2} (cooking time, min), and X_{3} (fermentation time, h) variables showed significant linear negative effects (β_{1}x_{1} = − 0.11x_{1}, β_{2}x_{2} = − 0.46x_{2}and β_{3}x_{3} = − 0.09x_{3}, respectively). The interaction between X_{1} and X_{2} variables had a significant linear positive effect (β_{12}x_{1}x_{2} = 0.134x_{1}x_{2}) (p<0.05). The other effects were not significant, and the mathematical model Ŷ_{3} = 1.11 − 0.11x_{1} − 0.46x_{2} − 0.09x_{3} + 0.13x_{1}x_{2} was developed, indicating that low levels of the three variables (X_{1}, X_{2}, and X_{3}) are recommended, in order to prepare tempeh with the maximum malonyl content. The lack-of-fit of the model was not significant (at 95%), and 99% of the experimental data were properly adjusted (R^{2}) to the model. The response surface (Figure 2 A and C) and desirability parameters (Figure 2 B and D) suggested a region where Y_{2} (0.58 μmol malonyldaidzin g^{-1}) and Y_{3} (2.01 μmol malonylgenistin g^{-1}) were maximal, with the following conditions: x_{1} = − 1 or X_{1} = 6 h soaking time; x_{2} = − 1 or X_{2} = 15 cooking time; and x_{3} = 0 or X_{3} = 24 h fermentation time. These conditions coincided with the assay 1, in which Y_{2} (0.58 μmol malonyldaidzin g^{-1}) and Y_{3} (2.02 μmol malonylgenistin g^{-1}) were experimentally obtained. In the grains, the contents of malonylglucosides were 4.25 μmol g^{-1}, with 3.04 μmol malonylgenistin g^{-1}.

The X_{2} variable (cooking time, min) showed a significant linear negative effect (β_{2}x_{2} = − 0.01x_{2}) on the Y_{4} response function (μmol of daidzein per g of tempeh). The X_{1} and X_{3} variables showed a significant quadratic effect (β_{1}x_{1}
^{2} = 0.01x_{1}
^{2} and β_{3}x_{3}
^{2} = − 0.01x_{3}
^{2}, respectively). The interaction between X_{1} (soaking time, h) and X_{2} showed a significant linear positive effect (β_{12}x_{1}x_{2} = 0.01x_{1}x_{2}), and the interactions between X_{1} and X_{3} (fermentation time, h) and X_{2} and X_{3} showed significant linear negative effects (β_{13}x_{1}x_{3} = − 0.01x_{1}x_{3} and β_{23}x_{2}x_{3} = − 0.01x_{2}x_{3}, respectively) (p<0.05). The other effects were not significant, and the mathematical model Ŷ_{4} = 0.08 + 0.01x_{1}
^{2} − 0.01x_{2} − 0.01x_{3}
^{2} + 0.01x_{1}x_{2} − 0.01x_{1}x_{3} − 0.01x_{2}x_{3} was developed. The lack-of-fit of the model was not significant (at 95%), and 90% of the experimental data were properly adjusted (R^{2}) to the model. The response surface (Figure 3 A) and desirability parameters (Figure 3 B) suggested a region where Ŷ_{4} (Ŷ_{4} = 0.11 μmol g^{-1}) was maximal, when obtained by using the following conditions: x_{1} = − 1 or X_{1} = 6 h soaking time; x_{2} = − 1 or X_{2}= 15 min cooking time; and x_{3}= + 0.5 or X_{3} = 30 h fermentation time. This processing conditions did not differ from the assay 7, in which the experimentally obtained Y_{4} was 0.11 μmol of daidzein per g of tempeh (Table 1).Therefore, according to the equation Ŷ_{4}, in order to obtain the maximum daidzein content in the tempeh, it is recommended to use low levels of x_{1} (−1) or X_{1} (6 h of soaking); for x_{2} (−1) or X_{2} (15 min of cooking); for x_{3} (+ 0.5) or at least X_{3} (30 h of fermentation).

For the Y_{5} response function (μmol of genistein per g of tempeh), only the X_{3} variable (fermentation time, h) showed significant linear negative effect (β_{3}x_{3} = − 0.17x_{3}) (p<0.05). Based only in this significant variable, the mathematical model Ŷ_{5} = 1.07 − 0.17x_{3} was developed. The lack-of-fit of the model was not significant (at 95%), and 86% of the experimental data were properly adjusted (R^{2}) to the model. The response surface (Figure 3 C) and desirability parameters (Figure 3 D) suggested that the maximum genistein content (Ŷ_{5} = 1.35 μmol g^{-1}) was obtained when using the following conditions: x_{1} = +1 or X_{1} = 18 h of soaking time; x_{2} = 0 or X_{2}= 20 min of cooking time; and x_{3} = − 1 or X_{3} = 18 h of fermentation time. This estimated result of the maximum genistein content coincided with the assay 7 (Y_{5} = 1.35 μmol g^{-1}) (Table 1). Therefore, according to the equation Ŷ_{5}, in order to obtain a higher genistein content in tempeh, it is recommended to use the lower level condition (x_{3} = − 1 or 18 h fermentation time), which was independent of the X_{1} or X_{2} variables.

Based on the confidence interval (CI) of the assay 7 (CI = 1.03 to 1.49), and on the response functions of the assays 1, 5, 9, 10, 11, 13, 16, and 17 where x_{3} = − 1, all response functions were within the interval of the assay 7, confirming that X_{1} (soaking time) and X_{2} (cooking time) were independent variables to obtain tempehs with maximum genistein content. In the grains, 'BRS 267' showed only 0.33 μmol g^{-1}. For tempeh obtained in the assay 7, the aglycone levels (0.11 μmol daidzein g^{-1} and 1.35 μmol genistein g^{-1}) increased by 4.4 times, mainly due to the fermentation of cotyledons.

The processing variables for tempeh preparation changed the level, profile, and distribution of the different isoflavone forms (Table 1). However, when comparing tempeh with cooked soybean grains, tempeh showed a 57% reduction in the content of glucosides, and the aglycone levels increased 6.5 fold due to the action of β-glucosidase produced by the fungus during fermentation (^{Wang & Murphy, 1996}). The cooking of soybean grains does not change the profile or distribution of the isoflavones, but it causes a drastic reduction in malonylglucosides. The isoflavone malonyl forms are unstable at high temperatures and are converted into their respective β-glycosides forms, due to the cleavage of malonyl ester groups, to daidzin and genistin forms. ^{Kudou et al. (1991)} reported that the content of malonylglucosides increased, after 12 to 24 h of fermentation. ^{Bavia et al. (2012)} observed an increase of 78% in the content of isoflavone aglycones (from 27.22 to 82.72 mg 100g^{-1} of tempeh) in tempehs prepared from different soybean cultivars. The authors also observed an increase of 15.01 to 26.74 mg 100g^{-1} in tempeh by soaking for 17 h, cooking for 30 min, and fermenting for 26 h. In the present work, the fermentation for 24 h resulted in the conversion of glycosides isoflavones and in the increase of the concentration of the corresponding aglycones, which agrees with other studies (^{Nakajima et al., 2005}; ^{Haron et al., 2009}; ^{Ferreira et al., 2011}; ^{Villares et al., 2011}). However, in the process of fermentation, despite the nonsignificant losses of isoflavones, the β-glycosidase enzyme hydrolyzes β-glucosides forms to aglycones, which have high biological activity and are the isoflavones readily available to humans (^{Izumi et al., 2000}).

Therefore, the models established the following conditions to obtain tempeh with maximum content of different isoflavone forms, which would favor its classification as a functional food: x_{1} (− 1), for genistin; x_{1}, x_{2}, and x_{3} (− 1, − 1, and 0), for malonyl; x_{1}, x_{2}, and x_{3} (− 1), for daidzein; and x_{3} (− 1), for genistein. To achieve the maximum content of different isoflavone forms, including aglycones, it is recommended to use the minimal amount of time for the processing variables, which corresponded to the assay 1, as follows: 6 h of soaking, 15 min of cooking and 18 h of fermentation.

Conclusions

In the preparation of soybean tempeh, the processing variables including soaking, cooking, and fermentation times with

*Rhizopus oligosporus*change the level, profile, and distribution of the different isoflavone forms.The highest bioconversion of β-glycoside isoflavones into aglycones occurs when soybean grains are soaked for 6 h, cotyledons are cooked for 15 min, and the fermentation with

*R. oligosporus*lasts for 18 h.