Y. lipolytica ATCC 18942 was obtained from the National Institute for Quality Control in Health, Oswaldo Cruz Foundation collection (INCQS 40149, Fiocruz, Rio de Janeiro, Brazil). Cells were grown in Yeast Malt Broth medium (glucose 10 g∙L⁻¹, yeast extract 3 g∙L⁻¹, malt extract 3 g∙L⁻¹ and peptone 5 g∙L⁻¹) at 28˚C under stirring (200 rpm) during 48 h. For the bioconversion experiments, cells were separated by centrifugation (3000 ´ g) and re-suspended in phosphate buffer (50 mM) at 20 g∙L⁻¹. R-(+)-limonene 97% (Sigma) was added and the mixture was incubated under stirring at 200 rpm. Temperature, pH, nutrient addition and limonene concentration were the factors selected for experimental design and their values varied through the experiments.

A fractional factorial design [24] was carried out for screening significant factors for limonene bioconversion to perillic acid. Process temperature, process pH, initial R-(+)-limonene concentration and nutrient addition were selected at the levels shown in Table 1. The experiments were performed in duplicate and perillic acid concentration, measured at 24 h and 48 h, was the response variable.

For the bioconversion optimization, a central composite design [24] was carried out having initial R-(+)-limonene concentration and pH as variables at the levels shown in Table 2. Nutrients were not added and the temperature was kept at 25˚C. Experiments were performed in non-simultaneous triplicate and aliquots of the bioconversion medium were collected at 24 and 48 h for perillic acid quantification. Two responses (dependent variables) were considered for the statistical analysis: perillic acid concentration and molar bioconversion yield (which considers perillic acid production related to the initial limonene concentration). For each response at each time (24 h and 48 h), the prediction model was calculated such as:

*Nutrients solution: 3 g∙L⁻¹ yeast extract, 3 g∙L⁻¹ malt extract, 5 g∙L⁻¹ peptone, 10 g∙L⁻¹ glucose.

where ŷ is the predicted response, b₀ is the intercept, b_Lim is the linear coefficient of the limonene concentration, x_Lim is the codified value of the limonene concentration, b_pH is the linear coefficient of the pH, x_pH is the codified value of the pH, b_LimpH is the coefficient for the interaction between limonene concentration and pH, is the quadratic coefficient of the limonene concentration, is the quadratic coefficient of pH value. Design settings and calculations were performed in JMP Statistical Discovery Software version 8.

The conditions optimized by central composite design, pH 6.9 and R-(+)-limo- nene concentration of 0.16% (v/v) were applied in a bioconversion experiment where a fresh equal R-(+)-limonene dose was added after 24 h. Aliquots were collected at 24 h and 48 h for perillic acid concentration measurement. Experiment was performed in duplicate.

The high-purity limonene optimized process conditions were applied to the bioconversion of the orange essential oil (from Tropfruit Nordeste S. A, Estância, Brazil). This substrate was added to the yeast medium in two portions (2 × 0.16% v/v, pH 6.9) at the beginning and after 24 h of bioconversion. Overall yield was taken after 48 h reaction.

Cell mass concentration was determined by optical density at 600 nm (OD₆₀₀). One OD₆₀₀ unit corresponded to a dry cell concentration of 0.444 g∙L⁻¹. Glucose concentration was measured using an YSI 2700 Select Biochemistry Analyzer (Yellow Springs). Perillic acid was determined by gas chromatography (GC) using (S)-(-)-perillic acid 95% from Sigma-Aldrich as standard. A sample of the bioconversion supernatant was cleared of debris by centrifuging 15 min at 15000 ´ g and a 1.00 mL supernatant aliquot was transferred to a 1.7 mL microcentrifuge tube (tube 1). Perillic acid was precipitated by adding 100 µl of HCl 0.6 M solution and the resulting suspension was homogenized and centrifuged at 15000 ´ g during 15 minutes. The supernatant was transferred to another 1.7 mL microcentrifuge tube and after repeating the precipitation procedure, the supernatant was discarded. The precipitates from both tubes were dissolved with ethyl acetate and transferred to a 2.00 mL volumetric flask that was filled up with the same solvent. After homogenization the perillic acid solution was analyzed using a GC instrument from Agilent Technologies model 7890 equipped with a 7683 injector and autosampler, split/splitless injector and flame ionization detector (Santa Clara, USA). A J & W Scientific HP-Innowax column with a bonded polyethylene glycol phase, 30 m length, 250 µm internal diameter and 0.250 µm film thickness, was used. All analyses were performed using a temperature program, constant flow, and split injection. The chromatographic conditions were as follows: oven temperature initiated at 50˚C, temperature increasing to 250˚C at 20˚C min⁻¹, held for 5 min; the run time was 15 min. The injector temperature was set at 280˚C, and detector temperature set at 300˚C. Helium was the carrier gas at a flow rate of 1 mL∙min⁻¹. Injection volume was 1.0 µl with the split ratio set at 25:1.

Limonene content in the crude orange essential oil was determined by GC using R-(+)-limonene 97% from Sigma as standard. GC-column was a DB-5 30 m length, 250 µm internal diameter and 0.250 µm film thickness. The chromatographic conditions were as follows: the oven temperature was kept at 70˚C for 5 min and then increased at 4˚C min⁻¹ up to 170˚C; the run time was 30 min. The injector was set at 250˚C and the detector at 300˚C. Helium was the carrier gas at a flow rate of 0.5 mL∙min⁻¹. Injection volume was 1.0 µl with the split ratio set at 50:1.

Previous results from our laboratory showed the yeast cell mass and limonene concentrations, pH, temperature and nutrient addition as relevant variables to be assessed for limonene bioconversion to perillic acid, as well as their range values [21] . Cell mass concentration, assayed to sub-levels 5.0 - 12.5 - 20.0 g∙L⁻¹ in a preliminary fractional factorial design experiment, showed a preponderant influence on the response, impairing the assessment of the remaining variables (data not shown). Hence, the screening was designed by fixing the cell mass concentration at 20 g∙L⁻¹ and varying the other factors as shown in Table 1.

The statistical analysis (Figure 1) showed that the most relevant factors were the limonene concentration followed by the pH. In the range assayed, the temperature was of low significance at 24 h and non-significant at 48 h. Therefore,

for the subsequent experiments, the temperature was fixed at 25˚C. The addition of a nutrient solution (containing macro-micronutrients, and growth factors that would be required for cell maintenance and/or enzymatic activity) was not significant for the perillic acid production, even though glucose, when present, was fully metabolized (data not shown). This result indicated that the bioconversion could be carried out without appending extra nutrients to the medium.

Process optimization with the two selected variables, limonene concentration and pH, was performed by central composite design according to Table 2. Figure 2 shows the effects of the factors on perillic acid concentration and on the bioconversion molar yield at 24 h and 48 h. The linear and the quadratic effects of limonene concentration were significant for perillic acid concentration at both time intervals. When considering the yield as the response variable, the linear effect of limonene concentration was significant at 24 h while the quadratic effect was significant at 48 h. PH effects in the range assayed were not significant for both responses in this experiment. Predictive models were generated for each response (product concentration and molar bioconversion yield) as shown in Table 3. The profiles displayed in Figure 3 and Figure 4 showed that the optimum conditions were different according to the type of response evaluated thereof. The overall maximum predicted perillic acid concentration (0.868 ± 0.076 g∙L⁻¹) was reached with 0.44% (v/v) initial limonene concentration and pH 6.94 at 48 h (Figure 4(a)). Nevertheless, this value was only 23% higher than the maximum predicted at 24 h (0.706 ± 0.062 g∙L⁻¹) for 0.40% (v/v) limonene concentration and pH 6.96 (Figure 3(a)). The predicted bioconversion yields for these conditions were 20.4% ± 2.6% and 18.4% ± 2.2%, respectively. On the other

hand, the maximum predicted bioconversion yields were 25.7% ± 2.4% (0.428 ± 0.069 g∙L⁻¹ perillic acid) at 24 h (Figure 3(b)) and 24.3% ± 2.3% (0.681 ± 0.067 g∙L⁻¹ perillic acid) at 48 h (Figure 4(b)) at much lower limonene concentrations, 0.16% (pH 6.9) and 0.28% (pH 6.76), respectively. The fact that the higher limonene concentrations resulted in lower bioconversion yields could be assigned to the limonene toxicity towards the yeast cells as well as to its volatility. These effects had already been described during previous limonene bio-oxidation studies

[21] . The pH was not significant for the measured responses at 24 and 48 h. Nevertheless, this parameter was included in the predictive model.

The statistical modeling indicated that the use of 0.16% limonene allowed the maximum molar yield within 24 h bioconversion suggesting that this initial limonene concentration coupled to a fresh limonene addition after 24 h might increase perillic acid concentration while keeping the bioconversion yield. To assess this possibility, a new test was conducted using 0.16% limonene at pH 6.9 followed by an equal limonene addition after 24 h. The product concentration reached 0.368 g∙L⁻¹ perillic acid (23.1% yield) after 24 h, within the model prediction interval between 0.358 g∙L⁻¹ and 0.496 g∙L⁻¹. After the second limonene

addition, the bioconversion process resumed reaching 0.793 g∙L⁻¹ of perillic acid (24.2% yield) within 48 h of the total bioconversion process. The two steps limonene addition resulted advantageous either comparing with the intermittent six-portions limonene feeding used in the previous study (0.5% total over 48 h; perillic acid concentration increasing by 50%) or to the total single limonene addition at the beginning of the experiment. Furthermore, the present study improved the reaction molar yield by 1.5 fold in comparison to the previous maximum yield of 16.7% [21] .

The optimized conditions found to the bioconversion of high-purity limonene with Yarrowia lipolytica were replicated by using orange essential oil containing 89.1% limonene. Within 48 h bioconversion, an accumulation of 0.872 g∙L⁻¹ of perillic acid (29.7% molar yield) was achieved; these values being higher than the observed for the use of high-purity limonene. The yield increase could be due to better substrate dispersion, as visually observed. In fact, crude orange oil pre- sents a large variety of components such as oxidized derivatives of other mono- terpenes and sesquiterpenes (data not shown) [25] that, nevertheless in minor amounts, would increase the availability of the limonene molecule to the yeast cells.