Effect of temperature on sugarcane ethanol fermentation: Kinetic modeling and validation under very-high-gravity fermentation conditions

Accepted Manuscript Title: Effect of temperature on sugarcane ethanol fermentation: Kinetic modeling and validation under very-high-gravity fermentation conditions Author: Elmer Ccopa Rivera Celina K. Yamakawa Marcelo B.W. Saad Daniel I.P. Atala Wesley B. Ambrosio Antonio Bonomi Jonas Nolasco Junior Carlos E.V. Rossell PII: DOI: Reference:

S1369-703X(16)30340-0 http://dx.doi.org/doi:10.1016/j.bej.2016.12.002 BEJ 6609

To appear in:

Biochemical Engineering Journal

Received date: Revised date: Accepted date:

19-7-2016 2-12-2016 4-12-2016

Please cite this article as: Elmer Ccopa Rivera, Celina K.Yamakawa, Marcelo B.W.Saad, Daniel I.P.Atala, Wesley B.Ambrosio, Antonio Bonomi, Jonas Nolasco Junior, Carlos E.V.Rossell, Effect of temperature on sugarcane ethanol fermentation: Kinetic modeling and validation under very-high-gravity fermentation conditions, Biochemical Engineering Journal http://dx.doi.org/10.1016/j.bej.2016.12.002 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Effect of temperature on sugarcane ethanol fermentation: Kinetic modeling and validation under very-high-gravity fermentation conditions Elmer Ccopa Rivera*a,b, Celina K. Yamakawaa, Marcelo B. W. Saad a, Daniel I.P. Atalac, Wesley B. Ambrosio c, Antonio Bonomia,b, Jonas Nolasco Juniora, Carlos E.V. Rossella a

Laboratório Nacional de Ciência e Tecnologia do Bioetanol (CTBE), Centro Nacional de

Pesquisa em Energia e Materiais (CNPEM), Caixa Postal 6192, CEP 13083-970, Campinas, São Paulo, Brazil b

School of Chemical Engineering, State University of Campinas, P.O. Box 6066, 13081-970,

Campinas, São Paulo, Brazil c

British Petroleum - Biofuels, CEP 04516-000, São Paulo, Brazil

* Corresponding author Highlights  Ethanol fermentation by Saccharomyces cerevisiae is conducted in a wide temperature range.  A mechanistic kinetic model is developed to predict reaction rates.  An optimization-based procedure was proposed to estimate kinetic parameters as a function of temperature.  The applicability of the kinetic model is validated for VHG ethanol fermentation.  Conditions required to produce ethanol with a higher yield and productivity are investigated.

Abstract In this work, a mechanistic model is developed to simulate the effect of temperature on Saccharomyces cerevisiae growth and ethanol production of batch fermentations. A wide temperature range is used to estimate the temperature-dependent kinetic parameters of the reaction kinetics. Because multi-parameter estimation problems are complex, an optimization-

1

based procedure is used to determine the optimum parameter values. The calculated reaction rates are used to construct a mechanistic fed-batch model. Experimental data from several cycles of very-high-gravity (VHG) ethanol fermentation from sugarcane are used to validate the model. Acceptable predictions are achieved in terms of the residual standard deviation (RSD). In addition, a suitable fermentation temperature profile, nutrient supplementation and micro-aeration during cell treatment are essential factors to obtain a yield of up to 90%, with a productivity of 10.2 g/L.h and an ethanol concentration of 120 g/L. Keywords: Mathematical modeling; Parameter estimation; VHG ethanol fermentation; Temperature; Sugarcane. 1. Introduction Currently, ethanol fuel from sugarcane is used on a large scale as hydrous ethanol in vehicles powered by ethanol or gasoline, also referred to as flex-fuel vehicles. Additionally, this fuel is used as an anhydrous ethanol blend with gasoline. The ethanol industry in Brazil processes approximately 630 million tons of sugarcane to produce 35.4 million tons of sugar, 28.5 million m3 of ethanol, and 32,300 GWh of electric power annually [1] and employs approximately 4.5 million people [2]. As a result, this industry continues to present challenges regarding the increasing concern about environmental impacts, primarily climate change effects and the reduced dependence on fossil fuel resources. The ethanol industry predominantly uses fed-batch fermentation with cell recycling; 7080% of distilleries utilize this mode of operation to produce ethanol [3]. In this configuration, 99.5% of the cells are reused in sequential fermentation (intensive recycling). The high cell density inside the bioreactors contributes to reducing the fermentation time to 6-11 h and to increasing the ethanol yield to 90-92%. The final ethanol concentration varies between 8 and 11 °GL [4]. However, this current fermentation technology is limited in processing substrates at a concentration of up to 200 g/L TSAI (total sugars as invert in which the original sucrose is equivalent to 0.95 of the reducing sugars formed (glucose and fructose) and glucose and 2

fructose originally present), representing the potential of obtaining an ethanol concentration of at least 11 °GL. The increased amount of sugars and ethanol in conventional fermentation causes a decrease in the cell maintenance rate, thus reducing the percentage of cell viability and stuck fermentation [5]. One method to improve the current ethanol fermentation profitability is through process intensification. Very-high-gravity (VHG) technology is one type of improvement process aimed at obtaining high ethanol concentrations of up to 15 °GL from moderately high sugar concentrations (> 250 g/L) [6]. Furthermore, VHG technology enables reduction of the process water requirements, thus reducing the related distillation operational cost, vinasse generation and its treatment cost, resulting in significant energy savings [7]. Over the course of VHG fermentation, a cell encounters significant stress induced by osmotic pressure, which leads to variations in the fermentation kinetics. Furthermore, the level of ethanol in the late stage of the process depends strongly on the nutritional conditions for cell maintenance [8]. Another common source of kinetic fluctuation could be related to the variability of feedstock quality, which is caused by the quality of molasses (related to the level of exhaustion and sulfitation in the sugar production) and the quality of sugarcane juice, which may vary during the sugarcane season, depending on the variety, harvest period, climate conditions and juice extraction procedure [9]. Rossell et al. [10] recently developed a novel continuous VHG ethanol fermentation by enhancing the engineering design and operation process. Ethanol fermentation in multistage bioreactors with a well-defined temperature profile was considered in this technology. Higher temperatures were considered in the initial stages to maximize conversion, and lower temperatures were considered in the later stages to minimize the inhibition and cellular damage from the high ethanol concentration. In addition, a distributed carbon source feeding was considered in the first and second stages to avoid inhibition by high sugar concentrations. The process also included intracellular detoxification through a second centrifugation followed by a 3

cell reactivation stage to promote membrane recovery and enzymatic restoration. The primary goal of this technology was to maintain active and viable cells for an entire harvest season, which is a critical condition to provide operating stability and a high rate of sugar conversion to ethanol. The primary challenge in the VHG fermentation process is to determine the effect of temperature as well as high substrate and ethanol concentrations on cell growth kinetics, which clearly affect the performance of the fermentation process. Although different empirical studies were developed for improving VHG processes, they lack a systematic model-based approach. A realistic kinetic model is required for the design, optimization and control of the process. Temperature is a crucial operating parameter due to its influence on the conversion of sugars to ethanol, which is an exothermic reaction, i.e., heat is released. In the case of conventional fermentation, a temperature control is required, often in the range of 32 to 35°C [4]. For VHG fermentation, a quantitative understanding of the effect of temperature on substrate consumption and ethanol production rates must be investigated to define the most suitable operating conditions using the definition of optimum temperature profiles. Previous work [11] has addressed the effect of temperature on ethanol production from sugarcane by S. cerevisiae, resulting in the formulation of a mechanistic model, which includes terms for the cell, substrate and ethanol inhibitions. The results reported that the maximum specific growth rate increases as the temperature increases. After approximately 37°C, the maximum specific growth rate begins to decrease. The maximum ethanol and cell concentrations, i.e., the concentrations at which cell growth ceases, are inversely related to the fermentation temperature. This model was validated during batch fermentation under conventional fermentation conditions. This study investigated the effect of temperature on ethanol production from sugarcane. For this purpose, a mechanistic fermentation model was developed, with kinetic parameters determined using a model-based optimization algorithm. The model was validated through 4

experiments, including fed-batch fermentation with cell recycling, operating under VHG conditions. The proposed methodology drives the systematic development of an industrially reliable mathematical model for VHG fermentation. 2. Material and Methods 2.1. Microorganism This study used an unclassified Saccharomyces cerevisiae strain cultivated in the Bioprocess Development Laboratory at CTBE, provided by the Faculty of Food Engineering / State University of Campinas and originally obtained from the Santa Adélia sugarcane mill. The stock culture was maintained in YPD (10 g/L yeast extract, 20 g/L peptone and 20 g/L dextrose) at – 80°C with 30% (v/v) glycerol. 2.2. Batch fermentations A stock culture was activated in liquid YPD medium at 33°C under agitation (250 rpm) for 24 h on an Innova 44 orbital shaker (New Brunswick, NJ, USA). Then, a sample was streaked onto YPD agar plates, incubated at 33°C for 48 h, and stored at 5°C. The inoculum for all fermentations was prepared by transferring three loops from agar plates to a new semisynthetic liquid medium. The medium consisted of 2.30 g/L urea, 6.60 g/L K2SO4, 3.0 g/L KH2PO4, 0.50 g/L MgSO4.7H2O, 1.0 g/L CaCl2.2H20, 5.0 g/L yeast extract, 25.44 mg/L trace elements (as detailed in Basso [12]), 3.0 ppm thiamine and 80.0 g/L TSAI. Then, the cells were incubated at 33°C under agitation (250 rpm) for 12 h on the Innova 44 orbital shaker. Commercial crystal sugar was used as the source of TSAI. After the inoculation period was complete, the cells were recovered using a Beckman Avanti J-26 XP centrifuge (JLA-16.250 rotor, 5,509 × g, 10°C, 15 min). The cells were diluted with sterilized potable water, and this suspension was inoculated in a 7.5 L Bioflo 115 bioreactor (New Brunswick, NJ, USA). The initial cell concentration was approximately 1 g/L. The propagation was conducted in fed-batch mode at 33°C and aerated with the carbon source limited to 18 g TSAI/L.h to minimize ethanol production through the Crabtree effect [13]. The agitation and airflow were controlled by 5

dissolved oxygen (DO) control in cascade mode, with the dissolved O2 concentration maintained above 25% air saturation. The propagation medium was the same as that used for the inoculum, with 180 g TSAI/L. After all sugars were consumed, to recover cells, the fermented medium was centrifuged using a Beckman Avanti J-26 XP centrifuge (JLA-9.100 rotor, 5,509 × g, 10°C, 15 min). The cells were re-suspended with sterilized potable water and refrigerated at 5°C until inoculation for further fermentations. The fermentation medium for the cell propagation was similar to the semi-synthetic medium described above. The salt and nutrient requirements were varied according to the initial cell density. All batch fermentations were performed in a 3 L Bioflo 115 bioreactor (New Brunswick, NJ, USA) under agitation at 200 rpm with a working volume of 2 L. The experiments were performed using the initial cell and substrate concentrations and at the temperatures shown in Table 1. 2.3. VHG fed-batch fermentation with cell recycling The inoculum and cell propagation were conducted according to the aforementioned batch fermentation. The inoculum medium contained 5.0 g/L urea, 1.1 g/L (NH4)2 HPO4 (DAP), 1.0 g/ L MgSO4.7H2O, 5.0 ppm ZnSO4.7H2O, 3.0 ppm thiamine, 5.0 g/L yeast extract and 80.0 g/L TSAI. The propagation medium contained 5.0 g/L urea, 1.1 g/L DAP, 1.0 g/L MgSO4.7H2O, 5.0 ppm ZnSO4.7H2O and 150.0 g/L TSAI. The TSAI source consisted of 79% (w/w) from sugarcane juice and 21% (w/w) from sugarcane molasses. The expected final concentration was 400 g TSAI/L. Physicochemical treatment, referred to as clarification, was performed to reduce the salt concentration, which influences the fermentation performance due to the elevated osmotic pressure. Substrate clarification was performed with the addition of 85% of 0.4 mL/L phosphoric acid and lime at a concentration of 8 °Be until pH 6.4 was reached, followed by heating at 95°C and the addition of 4 ppm non-ionic polymer [14]. Several minutes of rest were required for flake formation and decantation. These conditions allowed the drag of soluble impurities. Thus, a clarified substrate with high quality, minimal turbidity and low calcium concentration was produced. After clarification was complete, 2-L 6

Erlenmeyer flasks containing the clarified substrate required for each fermentation were sterilized at 121°C for 30 min. The VHG ethanol fermentation was based on the patent WO2014078924A1 of CNPEM/CTBE [10]. A schematic representation of the process is shown in Figure 1. Fed-batch fermentations were performed sequentially with cell recycling under six different conditions. Under each condition, five fermentation cycles were performed. All stages shown in Figure 1 are performed in each cycle. Under the first condition, referred to as the reference condition, the fermentations were performed in the 3 L Bioflo 115 bioreactor (New Brunswick, NJ, USA), with a working volume of 2 L at 200 rpm and temperature controlled over the range of 28-34°C. The oxidation-reduction potential (ORP), pH, capacitance and DO were monitored. A capacitance system (Aber Instruments, UK) was correlated with viable cell concentrations. The fed-batch started with an initial cell suspension volume of 0.9 L at a concentration of approximately 80 g/L (dry basis), followed by substrate feeding for 5 h. The temperature was set according to the ethanol concentration. The temperature started at 34°C and ended at 28°C until all sugar was consumed, producing an ethanol concentration of approximately 120 g/L. The fermentation time for each cycle was 11.5 h. A micro-aeration rate of 0.2 vvm was required when the ethanol concentration reached 108 g/L. After fermentation was complete, the cells were recovered using a Beckman Avanti J-26 XP centrifuge (JLA-9.1000 rotor, 13,261 × g, 20°C, 20 min). The cells were re-suspended with 0.9 L of sterilized potable water, and then, they were transferred back to the bioreactor vessel. Acid treatment was performed on the cells by adding sulfuric acid (2 M) through a silicon septum until a pH of 2.5 was reached. The acid treatment conditions were 30°C, 0.2 LPM of air (positive pressure to avoid contamination) and 600 rpm for 30 min. This cell suspension was centrifuged as previously described. The cells were then re-suspended with cooled reactivation medium that contained carbon, nitrogen and phosphate sources. This cell suspension was transferred back to the bioreactor vessel for the cellular reactivation stage. After this stage was complete, a new fed-batch fermentation was 7

performed. In the reactivation medium, sugarcane substrate and DAP were used to supply carbon, phosphorus and nitrogen for membrane restoration and cellular activity recovery. The cell concentration ranged from 80 to 100 g/L (dry basis) in a medium with 100 g/L TSAI and 3.1 g/L DAP. The operating parameters of the bioreactor with a working volume of 0.9 L were 600 rpm, 33°C, and 2 L/min of air, which produced an environment with an ORP over the range of + 50-100 mV. The residence time was 60 min. To determine the robustness, stability and reproducibility of the VHG ethanol fermentation described above, changes to certain key parameters, such as the DAP concentration, temperature during fermentation, micro-aeration rate and acid treatment, were investigated. Table 2 summarizes the conditions studied. 2.4. Analytical methods All sucrose, glucose, fructose, and ethanol measurements from batch fermentations were performed via high-performance liquid chromatography (HPLC). Details are provided elsewhere [15]. Measurements of the same components in the VHG fed-batch fermentations were also taken via HPLC. However, HPLC was applied only for the first and last samples during fermentation. Measurements of the remaining samples, the feeding substrate and the fermented medium concentration were conducted using mid-infrared (MIR) spectroscopy. The technique aimed to use a rapid and accurate method for analyzing VHG ethanol fermentation samples instead of using HPLC. The cellular viability and budding were analyzed by staining the cells with methylene blue [16] and counted using a Neubauer chamber in an optical microscope (Eclipse CI-S; Nikon, Japan) for samples at the end of the fermentations. The dry cell concentration was determined gravimetrically after centrifuging, washing two times with Milli-Q water and drying at 80°C until a constant weight was obtained in an analytical balance. 2.5. MIR analytical procedures Calibration curves were developed for ethanol and sugars using mid-infrared spectra and HPLC as primary reference analyses. VHG fermentation samples were analyzed in a 8

platinum diamond attenuated total reflectance (ATR platinum T Diamond 1 Refl #263C3643) single reflection cell, mounted in a mid-infrared Bruker Alpha Fourier transform infrared (FTIR) spectrometer (Bruker Optics Inc., USA). Previously, a number of samples were used to calibrate the spectrometer using HPLC as the primary analyses. The ATR-MIR spectra were recorded using OPUS software provided by Bruker Optics. The spectrum for each sample was obtained by taking the average of 16 scans (resolution of 4 cm−1, between 5000 and 550 cm−1), with a scanner velocity of 7.5 kHz (background of 16 scans), and using 100 µL samples at 40°C without cells (centrifuged sample). Air was used as the reference background spectra, and the ATR diamond surface was cleaned with ethanol (95% v/v) before each sample was analyzed to minimize or avoid contamination between samples. A calibration model (multivariate regression) was applied using a partial least squares (PLS) regression, with full cross validation of spectroscopic data and the corresponding concentration values from the HPLC. The PLS calibration requires the choice of suitable frequency ranges, optimum data preprocessing, and an optimum number of factors. Thus, the OPUS Quant Analyzer was used. This software employed an optimization tool to select among eleven mathematical methods for preprocessing the spectral data (e.g., vector normalization and second derivative). In addition, the software evaluated the optimal spectrum region to construct the calibration curve. After this analysis was complete, one method was selected for calibration study, and the optimum number of terms in the PLS calibration models was indicated by the lowest number of factors (PLS terms or rank), yielding the minimum value of the root mean square error of cross validation (RMSECV). The outliers were removed based on F-values obtained from the OPUS software calculator. Statistics calculated for the calibrations included the coefficient of determination in cross validation (R2), the RMSECV, the bias (systematic averaged deviation between the true and predicted values), the number of PLS factors (rank), and the residual prediction deviation (RPD). Glucose and fructose were analyzed using only one calibration curve by reducing sugars (glucose plus fructose). The statistical results for the PLS calibrations 9

developed using the VHG fermentation samples are shown in Table 3. The obtained calibration curves can be used to quantitatively determine these fermentation compounds in VHG fermentation samples with high accuracy (R2 > 99, RPD > 5, RMSECV (%) < 5). RPD values greater than 5 are considered appropriate for quality control and routine analysis [17]. 2.6. Determination of the fermentation performance parameters The fermentative yield was calculated using Eq. (l), which is based on the by-products method initially proposed by Finguerut [18]: Yield 

100 0.54  Aacetic Alactic 1.0032  ( X   G  S residual )  0.504  0.525 P P P

(1)

where P is the ethanol produced, X is the cell produced, G is the glycerol produced, Sresidual is the TSAI residual, Aacetic is the acetic acid produced and Alactic is the lactic acid produced. The definition of yield based on the by-products method was chosen because it is less sensitive to variations in the specific weight of the fermented medium (which requires an accurate measurement) compared with the conventional method. This method, currently used in ethanol plants in Brazil, consists of measuring the formation of the main by-products (cells, acetic acid, lactic acid and CO2) as well as unconverted sugars related to ethanol formation. The stoichiometric relationship among the acids and CO2 was based on empirical data from the ethanol plants. The volumetric productivity was calculated according to Eq. (2):

Productivity 

(2)

P V .t

where P represents the ethanol produced in g, V is the final volume of the fermented medium in L and t is the total fermentation time in h. 3. Modeling and parameter estimation First, the temperature-dependent kinetic model for ethanol fermentation was developed with experimental data from batch experiments. The model was expanded to simulate fed10

batch ethanol fermentation operated under VHG conditions and validated against experimental observations. 3.1. Kinetic and theoretical aspects of batch fermentation A mechanistic model that describes the kinetics of the concentrations of cells - X (g/L), substrate - S (g/L), and ethanol - P (g/L), in terms of three ordinary differential equations, i.e., Eqs. (3-5), was applied to the ethanol fermentation process [11]:

dX  rx dt

(3)

dS  rs dt

(4)

dP  rp dt

(5)

where rx, rs and rp represent the cell growth, substrate consumption and ethanol production reaction rates, respectively (g/(L.h)). During VHG ethanol fermentation, the cell is exposed to stress barriers, including osmotic stress due to high substrate concentration, ethanol stress at the end of fermentation and high cell densities inside the bioreactor. To simulate these phenomena in terms of the physiological state of the cell, substrate saturation and inhibition terms were considered in the cell growth rate, rx, in Eq. (6) (first and second terms, respectively). The cell responses to ethanol inhibition and inhibition due to high cell concentrations were also incorporated into the model (third and fourth terms in Eq. (6), respectively).

 S P   rx  μmax exp( Ki S) 1  Ks  S  Pmax 

n

m

 X   1   X X max  

In the above equation, µmax denotes the maximum specific growth rate (h-1), Ks denotes the substrate saturation constant (g/L), Ki denotes the substrate inhibition parameter (g/L), Xmax denotes the cell concentration when cell growth ceases (g/L), Pmax denotes the ethanol 11

(6)

concentration when cell growth ceases (g/L) and m and n denote parameters related to cellular and product inhibitions, respectively. Cells meet their energy requirements for growth and maintenance by producing ethanol. Maintenance is the energy required to support cell function, other than that required for cell synthesis. For instance, repairing active cell material involves a protein synthesis system, among others. The ethanol production rate, rp, is consistent with the growth and non-growth associated product model proposed by Luedeking and Piret [19] and is represented in Eq. (7):

rp  Υ p/xrx  mp X

(7)

where Yp/x denotes the Luedeking–Piret growth-associated constant (g/g) and mp denotes the Luedeking–Piret non-growth-associated constant (g/(g.h)). Similar to the ethanol production rate, the Luedeking–Piret equation was proposed for the substrate consumption rate, rs, as shown in Eq. (8), by considering that the substrate consumed is not related to cell growth but is also used for other processes requiring energy:

rs  rx / Yx  ms X

(8)

where Yx and ms are the cell yield (g/g) and maintenance parameter (g/(g.h)), respectively. The kinetic model is formed using Eqs. (3-8), which contain eleven kinetic parameters. The model should include or be able to estimate these parameters for estimating rx, rs and rp. The reaction rates of the cell are affected by different factors, and one of the most important of these is temperature. 3.2. Modeling of the temperature influence In this study, the temperature-dependent parameters are modeled using Eqs. (9) and (10). As demonstrated in previous studies, these equations have been effectively used to simulate the effects of temperature on cell growth in sugarcane juice, molasses [9] and bagasse hydrolysates [20]:

12

Parameter (T)  A exp B/T   C exp D/T 

(9)

Parameter (T)  A exp B/T 

(10)

where T is the temperature (°C) at which the kinetic parameters are evaluated, A and C are preexponential factors and B and D are exponential factors. A main feature of the proposed temperature-dependent kinetic model is the behavior of X, S, and P when the temperature changes. Given that the same parameters have been modeled as being temperature dependent, the kinetic rates change when the temperature is varied. This approach enables the simulation of an adequate temperature profile for improving the fermentation performance. For instance, in VHG fermentation, a gradual reduction in temperature during fermentation is required, which alleviates the inhibition effects while maintaining sufficient rates of cell division and ethanol production [21]. 3.3. Parameter estimation procedure An accurate estimation of model parameters is a major issue in the development of mechanistic models. Here, a five-step procedure is proposed to estimate the kinetic parameters for the current model. In Step 1, an identifiability analysis of the kinetic model was performed, which was achieved by the calculation of a meta-model based on Plackett-Burman (PB) design [22], capable of providing all the required sensitivity information. For this analysis, the influence of kinetic parameters (µmax, Xmax, Pmax, Yx, Yp/x, Ks, Ki, ms, m, n, mp) on responses X, S and P (data at a single timepoint) was evaluated using a PB design for eleven factors. In the PB design, the simulations involve systematically varying all kinetic parameters within ±10% of nominal values reported elsewhere [23]. The simulation aims to determine the subset of model parameters with the largest effect on the model prediction, which are modeled as temperaturedependent.

13

The simulation results were evaluated using the Pareto chart, which indicates each of the estimated effects of the parameters on X, S and P in decreasing order of magnitude. The size of each bar is relative to the standardized effect, calculated as the estimated effect divided by its standard error, which is equivalent to estimate a t-statistic for each effect. In addition, a horizontal line (critical t-value) represents the effect's limit of significance. Bars extending beyond that line relate to effects that are statistically significant at the 95% significance level. The parameter estimation largely depends on the initial guesses of the parameter values due to the nonlinear nature of the kinetic model, which contains multiple parameters. Thus, in Step 2 the initial guesses of the temperature-dependent parameters to be estimated are obtained from a previous study [9,11,20,23,24]. The remaining parameters were fixed at values given by Atala et al. [24]. This information is incorporated into the model along with the initial conditions for X, S and P. Hence, in Step 3, the kinetic model written in FORTRAN is integrated using the routine LSODE [25] to obtain the time-course concentrations of X, S and P. Next, in Step 4, an objective function J() is defined as the least square error given by Eq. (11): 3

J() =

Np

 i 1 j 1

(yexpi (t j )  y simi (t j ))2

(11)

2 yexp i max

where  is a vector containing all the temperature-dependent parameters, yexpi (t j ) is the ith measured concentration, i.e., the measured concentrations of the cell, substrate and ethanol at sampling time j, ysimi (t j ) and yexpi max are the concentrations computed by the kinetic model and maximum measured concentration, respectively, and Np is the number of sampling points. In Step 5, a nonlinear optimization problem (NP) is formulated, as shown in Eq. (12), which minimizes the objective function J():

14

Min J() Subject to: NP:

(12) Eqs. (3-8)

lb ≤  ≤ ub where lb and ub are the lower- and upper-bound vectors for the temperature-dependent parameter vector , respectively. In this study, a hybrid optimization algorithm based on the genetic algorithm (GA) and the Quasi-Newton method (QN) [11] is adopted to solve Eq. (12). The hybrid algorithm comprises a thorough exploration of the search space (lb ≤  ≤ ub) and finding the neighborhood of the global optimum through the GA. Then, the hybrid algorithm, starting from the best estimate of the GA (best), uses gradient information from the QN method to accelerate the convergence toward the global optimum, which involves determining the optimum values for the parameters (optimum) that produce the best fit between the measured concentrations of X, S and P and their corresponding computed concentrations from the kinetic model while minimizing J(). The execution of Steps 1-5 generated a set of optimal values for the temperaturedependent parameters for each temperature considered (24, 27, 30, 33 and 36°C). Then, the set of optimal values for each parameter was adjusted to Eqs. (9) or (10). This adjustment yielded the best constants A, B, C and D for each temperature-dependent parameter. Once the kinetic model was established, the constants were fine-tuned using the GA such that J was minimized. The nonlinear optimization problem (Eq. (12)) was implemented in a FORTRAN routine on a 3.6 GHz Intel(R) Core(TM) i7-4790 CPU. 3.4. Mass balance equations for predicting VHG fed-batch fermentation After the kinetic parameters were estimated, the kinetic rates rx, rs and rp that incorporate these parameters were used in a mechanistic fed-batch model with the mass 15

balance equations described in Eqs. (13)-(15). Thus, the kinetic study provides a basis to allow for the dynamic simulation of the VHG ethanol fermentation process through a model that is sensitive to inhibition factors and changes in temperature:

dX F  rx  X dt V

(13)

dS F  (S f  S)  rs dt V

(14)

dP F  rp  P dt V

(15)

where F denotes the feed flow rate (L/h), V denotes the working volume (L) of the bioreactor, and Sf denotes the total sugar concentration (g/L) in the feed. The reaction rates rx, rs and rp are the same as those obtained from the previous batch experiments. During model simulation, it was assumed that the sugars are fed at a fixed F, and the measured V is an input data at each integration time during the simulation. 4. Results and Discussion 4.1. Identifiability analysis using the PB design The parameter estimation procedure considered iterative steps with an evaluation of the model reliability at several points. These evaluations included the identifiability analysis of the kinetic model using the calculation of a meta-model based on the Plackett-Burman (PB) statistical design. Figures 2A, 2B and 2C show the Pareto charts obtained from PB design for responses X, S and P, respectively. According to the estimated effects, Pmax, Yx and Yp/x were considered statistically significant for all responses. In addition, max was considered statistically significant for S, and its absolute standard effect on X is close to the cut-off line for significant effects. The parameters mp and ms appeared relevant only for P with very small effects (nine times smaller) compared to those for Yx and Yp/x.

16

In this study, the statistically significant parameters µmax, Pmax, Yp/x and Yx were modeled as functions of the fermentation temperature. In addition, the parameters Xmax and ki were modeled as a function of temperature based on previous studies [11,20,24]. The remaining parameters were fixed as follows [24]: Ks = 4.1 g/L, mp = 0.1 g/(g.h), ms = 0.2 g/(g.h), m = 1.0 and n = 1.5. 4.2. Batch fermentation and kinetic study results A series of batch fermentation experiments was performed using temperatures of 24, 27, 30, 33 and 36°C and different initial concentrations of X and S, as shown in Table 1. The ethanol productivity for all fermentations typically resulted in high values due to the rapid conversion of sugars into ethanol and other co-products. The fermentation conditions at 24°C with a low substrate concentration were unfavorable for conversion to ethanol; this resulted in a lower productivity and yield. In contrast, a temperature of 33°C led to a higher productivity and yield, as expected [26]. The initial conditions and temperatures of the batch experiments were used to perform the previously described parameter estimation procedure. A set of optimal values of the temperature-dependent parameters for each temperature considered was generated. Figure 3 shows the behavior of the optimal values of the kinetic parameters at 24, 27, 30, 33 and 36°C (solid symbols). Then, the constants A and B in the Arrhenius-type Eq. (9) were adjusted for the temperature-dependent kinetic parameters Xmax, Pmax, Yp/x and Yx and the constants A, B, C and D in Eq. (10) for the parameters µmax and ki. As in previous studies [9,11,20], Eq. (10) was used instead of the Arrhenius-type equation to accurately describe the nonlinear influence of temperature on some kinetic parameters. The optimal values of the constants and the correlation coefficient (R2) for each adjustment are shown in Table 4. Figure 3 shows the fit using Eqs. (9) and (10) (solid curves). These data clearly demonstrate the influence of temperature on the kinetic parameters. As µmax increases from 24 to 33°C, multiplication and metabolism begin to be inhibited at higher temperatures. Xmax and Pmax decreased by mean 17

values of approximately 1.2 and 1.4 g/L, respectively, for each 3°C decrease in temperature, suggesting that tolerance to ethanol and cell concentrations increases at low temperatures. In addition, Yp/x and Ki exhibited lower values at higher temperatures, whereas the cell yield Yx decreases at lower temperatures, which could lead to significant changes in the substrate consumption rate. Previous studies of ethanol fermentation from sugarcane juice, molasses [9,27] and bagasse hydrolysates [20] as substrates have shown that the yields are temperature dependent; temperatures above 34 °C begin to favor glycerol formation. Ethanol plants have higher losses at higher temperatures due to the ethanol dragged with CO2. The computed profiles (from the model described by Eqs. (3-10)) of the cell, substrate and ethanol concentrations at 27, 30 and 33°C and with different initial concentrations are shown in Figures 4A, B and C, respectively. The measured concentrations used for the parameter estimation are also shown in these figures for comparison. The results illustrate that the model effectively tracks the desired trajectory of the measured data. A more rigorous assessment of the goodness-of-fit between the measured and computed data can be conducted using residual standard deviation (RSD), as suggested by Cleran et al. [28]. Recent studies [29,30] have adopted RSD as a measure of the prediction accuracy of proposed models. RSD is written as a percentage of the average of the measured data, yexp, defined by Eq. (16):

 1   Np RSD(%)  

Np

0.5

 (yexp(t j )  ysim (t j ))2   j 1  yexp



(16)

 100

where yexpi (t j ) represents the ith measured concentration, i.e., the measured concentrations of the cell, substrate and ethanol at sampling time j, ysimi (t j ) are the concentrations computed by the model and Np is the number of sampling points.

18

Figure 5A illustrates that the model predictions were particularly accurate, as determined by the RSD (%). These results demonstrate that the model can be used to predict the dynamic behavior under the current batch conditions. Therefore, the kinetic model that incorporates inhibition terms and temperature-dependence functions can be used to predict the concentrations of X, S and P at changing reaction temperatures, as well as with different initial concentrations of X and S. This model can handle temperature fluctuations between 24 and 36°C. These kinetic results were further used to predict the dynamic behavior of X, S and P in VHG ethanol fermentation. 4.3. Prediction of VHG fed-batch experiments with cell recycling VHG fed-batch fermentations with cell recycling were performed sequentially under six different conditions, as described in Table 2. Five fermentation cycles were performed under each condition. All stages described in Figure 1 are conducted for each cycle. Figure 6 shows the average fermentation performance for each condition in terms of yield, productivity and final ethanol concentration. Condition 1 (the reference condition, described in detail in Section 2.3) obtained a yield of 84%, a productivity of 10.2 g/L.h and an ethanol concentration of 120 g/L (average values for five fed-batch fermentations). After condition 1 was tested, a second set of five fed-batch fermentations were conducted (condition 2) with the aim of improving the fermentation performance by varying the DAP concentration in the reactivation stage to a value above the reference condition (the other operating conditions remained the same as the reference conditions). A slightly better fermentation performance than the reference condition was obtained, with a yield of 87%, a productivity of 10.3 g/L.h and an ethanol concentration of 122 g/L. For condition 3, the temperature at the final stage during the exhaustion of sugar conversion was slightly above the final reference temperature. As shown in Figure 6, this condition was the most efficient process in terms of fermentation yield, with a value of 90%. However, the productivity of 10.2 g/L.h and ethanol concentration of 120 g/L were lower than 19

those of condition 2. In condition 4, in which the micro-aeration rate was reduced while maintaining the same temperature profile of condition 2, the fermentation performance was the poorest of the examined conditions, with a yield of 84%, a productivity of 9.8 g/L.h and an ethanol concentration of 115 g/L. An attempt was made to remedy this poor fermentation performance situation in condition 5, which used the same operating parameters as those in condition 3. However, despite a slight improvement (a yield of 84%, a productivity of 9.9 g/L.h and an ethanol concentration of 117 g/L), the fermentation performance of condition 3 could not be reproduced. Thus, a micro-aeration rate of 0.2 vvm at the end of the fermentation should be maintained to avoid a decrease in the fermentation performance. Finally, in condition 6, nitric acid was used instead of sulfuric acid. This condition provided a yield of 88%, a productivity of 10.4 g/L.h and an ethanol concentration of 124 g/L. Hence, this condition was the most favorable in terms of productivity and ethanol concentration. These results demonstrate that all average productivities were approximately 10 g/L.h while maintaining yields between 84% and 90%. These productivity values are high compared to the typical industrial value of 5.5 g/L.h in which the yield is 89% working in continuous mode. The unexpected low value of yield is related to the residual sugar at the end of fermentation, indicating that more time is required for its consumption. The fed-batch model can describe the dynamics of cell growth, substrate consumption and ethanol production during VHG fermentation under varying conditions (Table 2). These results are shown in Figure 7. The corresponding assessment using RSD (%) is shown in Figure 5B. This proposed model could provide a basis for the design and operation of VHG ethanol fermentation at an industrial scale. The model assumed that the ethanol production rate, rp, is consistent with the growth and non-growth associated product, as described by Eq. (7). However, above the Pmax value in VHG fermentation, rp is consistent only with the non-growth associated product. Pmax is a temperature-dependent parameter; therefore, it has different values for different temperatures, 20

as shown in Figure 3. Alfenore et al. [31] also observed these two phases (growth and nongrowth associated ethanol production) during VHG ethanol fermentation. The phase in which ethanol production was disconnected from cell growth was observed when the ethanol concentration in the medium reached values of approximately 100 g/L. A nutrient feeding and aeration strategy was used for higher ethanol concentrations to avoid loss of cell membrane integrity because ethanol acts as a chaotropic solute in the aqueous phase of the membrane system and induces changes in its structure and fluidization of lipid bilayers [5]. In this work, the application of micro-aeration when the ethanol concentration was approximately equal to Pmax, as well as with appropriate temperature and nutrient supplementation, was essential to overcome the inhibitory and detrimental effect of high ethanol concentrations. These operating conditions, along with the process configuration shown in Figure 1 (where a second centrifugation and a cell treatment step were incorporated, in contrast to the conventional process), render ethanol fermentation feasible under VHG conditions. 5. Conclusions This study demonstrated how a systematic model-based approach can be applied to address kinetic model development. The resulting model can predict cell growth and ethanol production from sugarcane under different VHG conditions. An advanced temperaturedependent parameter estimation procedure was used to guarantee the accuracy of the proposed model. The influence of temperature on fermentation kinetic behavior was also explicitly demonstrated. The VHG fermentation results demonstrated that an appropriate temperature profile, micro-aeration around the Pmax value and nutrient supplementation during cell treatment produce yields of up to 90%, with an ethanol productivity of 10.2 g/L.h. Acknowledgements The authors acknowledge British Petroleum Biofuels (BP), the Brazilian Bioethanol Science and Technology Laboratory / Brazilian Center of Research in Energy and Materials 21

(CTBE/CNPEM) and FAPESP (process number 2016/01785-0) for their financial support, assistance regarding the use of facilities, sharing expertise and the opportunity to develop this work. References [1]

EPE, Brazilian energy balance 2015 year 2014. Empresa de Pesquisa Energética (EPE) (in Portuguese), Rio de Janeiro, 292p., 2015.

[2]

R. Zeidan, C. Boechat, A. Fleury, Developing a sustainability credit score system, J. Bus. Ethics 127 (2015) 283-296.

[3]

S. Brethauer, C.E. Wyman, Review: continuous hydrolysis and fermentation for cellulosic ethanol production, Bioresour. Technol. 101 (2010) 4862-4874.

[4]

L.C. Basso, H.V. de Amorim, A.J. de Oliveira, M.L. Lopes, Yeast selection for fuel ethanol production in Brazil, FEMS Yeast Res. 8 (2008) 1155-1163.

[5]

J.A. Cray, A. Stevenson, P. Ball, S.B. Bankar, E.C. Eleutherio, T.C. Ezeji, R.S. Singhal, J.M. Thevelein, D.J. Timson, J.E. Hallsworth, Chaotropicity: a key factor in product tolerance of biofuel-producing microorganisms, Curr. Opin. Biotechnol. 33 (2015) 228259.

[6]

C.J. Cassan, J. Riess, F. Jolibert, P. Taillandier, Optimization of very high gravity fermentation process for ethanol production from industrial sugar beet syrup, Biomass Bioenergy 70 (2014) 165-173.

[7]

P. Puligundla, D. Smogrovicova, V.S. Obulam, S. Ko, Very high gravity (VHG) ethanolic brewing and fermentation: a research update, J. Ind. Microbiol. Biotechnol. 38 (2011) 1133–1144.

[8]

L. Benbadis, M. Cot, M. Rigoulet, J. Francois, Isolation of two cell populations from yeast during high-level alcoholic fermentation that resemble quiescent and nonquiescent cells from the stationary phase on glucose, FEMS Yeast Res. 9 (2009) 1172–1186.

22

[9]

E.C. Rivera, A.C. Costa, R.R. Andrade, D.I.P. Atala, F. Maugeri Filho, R. Maciel Filho, Development of adaptive modeling techniques to describe the temperature-dependent kinetics of biotechnological processes, Biochem. Eng. J. 36 (2007) 157-166.

[10] C.E.V. Rossell, J. Nolasco Junior, C.K. Yamakawa, Processo e equipamento para fermentação continua multiestágio com recuperação, reativação, e reciclo de fermento para obtenção de vinhos com alto teor alcoólico. Patent WO2014078924A1, 2012 [11] E.C. Rivera, A.C. Costa, D.I.P. Atala, F. Maugeri, M.R.W. Maciel, R. Maciel Filho, Evaluation of optimization techniques for parameter estimation: application to ethanol fermentation considering the effect of temperature, Proc. Biochem. 41 (2006) 1682-1687. [12] T.O. Basso, Improvement of Alcoholic Fermentation in Saccharomyces cerevisiae by Evolutionary Engineering. Ph. D. Thesis. University of São Paulo, Brazil (in Portuguese), 2011. [13] J.P. Barford, R.J. Hall, An examination of the crabtree effect in Saccharomyces cerevisiae: the role of respiratory adaptation, J. Gen. Microbiol. 114 (1979) 267–275. [14] COPERSUCAR (Cooperative of Sugarcane, Sugar and Ethanol Producers of the State of Sao Paulo), Fermentation, Technical Report (in Portuguese), São Paulo, Brazil, 1987. [15] E.C. Rivera, C.K. Yamakawa, M.H. Garcia, V.C. Geraldo, C.E.V. Rossell, R. Maciel Filho, A. Bonomi, A procedure for estimation of fermentation kinetic parameters in fedbatch bioethanol production process with cell recycle, Chem. Eng. Trans. 32 (2013) 1369-1374. [16] S.S. Lee, F.M. Robinson, H.Y. Wong, Rapid determination of yeast viability, Biotechnol. Bioeng. Symp. 11 (1981) 641-649. [17] Y. Huang, J. Carragher, D. Cozzolino, Measurement of fructose, glucose, maltose and sucrose in barley malt using attenuated total reflectance mid-infrared spectroscopy, Food Anal. Methods 9 (2015) 1079–1085.

23

[18] J.I. Finguerut, Research Project in public policy (PPPP) ethanol, I Technological Workshop on Obtaining Ethanol (in Portuguese). University of São Paulo, Lorena, SP, 2006. [19] R. Luedeking, E.L. Piret, A kinetic study of the lactic acid fermentation. Batch process at controlled pH, Biotechnol. Bioeng. 1 (1959) 393–412. [20] R.R. Andrade, F. Maugeri Filho, R. Maciel Filho, A.C. Costa, Kinetics of ethanol production from sugarcane bagasse enzymatic hydrolysate concentrated with molasses under cell recycle, Bioresour. Technol. 130 (2013) 351-359. [21] M.J. Torija, N. Rozès, M. Poblet, J.M. Guillamón, A. Mas, Effects of fermentation temperature on the strain population of Saccharomyces cerevisiae, Int. J. Food Microbiol. 80 (2003) 47–53. [22] R.R. Plackett, J.P. Burman, The design of optimum multifactorial experiments, Biometrika. 33 (1946) 305-325. [23] R.R. Andrade, E.C. Rivera, A.C. Costa, D.I.P. Atala, F. Maugeri Filho, R. Maciel Filho, Estimation of temperature dependent parameters of a batch alcoholic fermentation process, Appl. Biochem. Biotechnol. 136–140 (2007) 753-763. [24] D.I.P. Atala, A.C. Costa, R. Maciel Filho, F. Maugeri Filho, Kinetics of ethanol fermentation with high biomass concentration considering the effect of temperature, Appl. Biochem. Biotechnol. 91–93 (2001) 353-365. [25] K. Radhakrishnan, A. Hindmarsh, Description and use of LSODE, the Livermore solver for differential equations, NASA reference publication 132, 1993. [26] S.R. Andrietta, Modeling, Simulation and Control of Industrial Scale Processes for Continuous Alcoholic Fermentation. Ph.D. Thesis. State University of Campinas Brazil (in Portuguese), 1994.

24

[27] M. Phisalaphong, N. Srirattana, W. Tanthapanichakoon, Mathematical model to investigate temperature effect of kinetic parameters of ethanol fermentation, Biochem. Eng. J. 28 (2006) 36-43. [28] Y. Cleran, J. Thibault, A. Cheruy, G. Corrieu, Comparison of prediction performances between models obtained by the group method of data handling and neural networks for the alcoholic fermentation rate in enology, J. Ferment. Bioeng. 71 (1991) 356-362. [29] D. Farias, R.R. Andrade, F. Maugeri Filho, Kinetic modeling of ethanol production by Scheffersomyces stipitis from Xylose, Appl. Biochem. Biotechnol. 172 (2014) 361-379. [30] G.H.S.F. Ponce, J. Moreira Neto, S.S. De Jesus, J.C.C. Miranda, R. Maciel Filho, R.R. Andrade, M.R.Wolf Maciel, Sugarcane molasses fermentation with in situ gas stripping using low and moderate sugar concentrations for ethanol production: Experimental data and modeling, Biochem. Eng. J. 110 (2016) 152-161. [31] S. Alfenore, X. Cameleyre, L. Benbadis, C. Bideaux, J.L. Uribelarrea, G. Goma, C. Molina-Jouve, S.E. Guillouet, Aeration strategy: a need for very high ethanol performance in Saccharomyces cerevisiae fed-batch process, Appl. Microbiol. Biotechnol. 63 (2004) 537-542.

25

Figure Captions

Figure 1. VHG ethanol fermentation flowchart. Figure 2. Pareto chart of the effects for the concentration of: A) cell, B) substrate and C) ethanol (at a 95% significance level) Figure 3. Optimal values of the kinetic parameters at 24, 27, 30, 33 and 36°C (solid symbols) and their corresponding fit using Eqs. (9) and (10) (solid curves). Figure 4. Experimental data from the batch fermentations (Cells (■); Substrate (▲) and Ethanol (●)) compared with simulated concentration time curves (): A) 27°C, B) 30°C and Initial Condition IC-2 in Table 1 and C) 33°C. Figure 5. RSD (%) to evaluate the goodness-of-fit of the (A) kinetic model and (B) fed-batch model against VHG fermentations with cell recycling. Figure 6. Performance parameters of fermentation. Figure 7. Experimental data from the VHG fed-batch fermentations with cell recycling (Cells (■); Substrate (▲) and Ethanol (●)) compared with simulated concentration time curves () under Condition A) 1, B) 2, C) 3, D) 4, E) 5 and F) 6 in Table 2. The fermentation time for each cycle was 11.5 h.

26

Table 1. Temperature and initial conditions of the batch experiments.

Temperature

Initial condition (IC)

Fermentation

(°C)

performance Cell

Substrate

Ethanol

(g/L)

(g/L)

(g/L)

24 (IC-1)

0.88

23.33

0.0

0.77

78.55

24 (IC-2)

17.17

291.10

1.47

4.96

91.11

27

5.13

161.44

0.82

5.97

87.31

30 (IC-1)

5.38

154.93

0.77

7.29

96.70

30 (IC-2)

4.93

161.11

0.81

7.09

90.81

30 (IC-3)

18.05

277.65

8.22

6.39

91.28

33

4.88

164.80

0.68

8.40

90.15

36

5.25

164.36

1.33

6.63

78.91

27

Productivity Yield (g/L.h)

(%)

Table 2. Process conditions for VHG fed-batch ethanol fermentation.

Condition

Parameter assessed

Description of conditions

1

Reference



Acid treatment: sulfuric acid.



Reactivation: 100 g/L TSAI, 3.1 g/L DAP, micro-aeration rate 0.2 vvm, 33°C, 60 min.



Fermentation: Temperature beginning at 34°C and ending at 28°C. Temperature of 28°C and micro-aeration rate of 0.2 vvm were set when the ethanol concentration reached 108 g/L.

2

DAP

DAP concentration is increased to 5.6 g/L.

3

Fermentation temperature

Beginning at 34°C and ending at 30°C. This last temperature was set when the ethanol concentration reached 108 g/L.

4

Fermentation temperature

Beginning at 34°C and ending at 30°C. The

and micro-aeration

micro-aeration rate is reduced by half when ethanol concentration reached 108 g/L.

5

Fermentation temperature

Returned to the same condition as that used in 3.

and micro-aeration 6

Acid treatment

Nitric acid instead of sulfuric acid.

28

Table 3. Descriptive statistics of calibration for the measurement of VHG fermentation samples analyzed using attenuated total reflectance mid-infrared spectroscopy.

Glucose + Ethanol

Sucrose Fructose

Range (g/ L) Spectra

Preprocessing

0-129.7

0-108.2

0-77.0

487

70

355

Multiplicative

Vector

No spectral

scattering

normalization

data

correction

(SNV)

preprocessing

Frequency

1739.5-1488.7; 1636.2-976.9 -1

regions (cm )

1476.0-860.6 1174.6-854.5

R2

99.30

99.74

99.58

RMSECV (%)

2.64

1.28

0.863

Bias

0.000469

0.0279

-0.00368

RPD

12

19.5

15.5

PLS terms

7

8

10

R2: coefficient of determination in cross validation; RMSECV: root mean square error of cross validation; RPD: standard deviation / standard error of cross validation; PLS terms: optimal number of terms used to develop the PLS models.

29

Table 4. Optimal values of the constants in Eqs. (9) and (10) for temperature-dependent kinetic parameters.

Parameter

Equation

R2

A

B

C

D

max

(9)

0.98

-0.92273

45.20986

1.26669

38.29429

Xmax

(10)

0.99

57.89562

4.89193

-

-

Pmax

(10)

0.98

102.12366

3.58256

-

-

Yp/x

(10)

0.99

18.27807

-30.6689

-

-

Yx

(10)

0.96

0.01965

36.15008

-

-

ki

(9)

0.99

1.29176

15.68355

-1.28591

15.76083

30

Figure 1

31

Figure 2

(A)

Standardized Effect Estimate (Absolute value)

26.63

4.50 -4.28

p=0.05

1.75 -1.59 -1.43 -1.15 0.70 0.36 - 0.35 - 0.35

Umax Yx Xmax Pmax Pmax Yp/x Yx ms max Ypx

Ks n

Ki Ki

mx m

Kms Xnmax mp mp

(B) 5.52 - 5.38 5.28

-3.11

p=0.05

2.51 1.53

Umax Yx Xmax Pmax Pmax Yp/x Yx n max Ypx

Ksi K

0.97 Kis m

- 0.25 -0.18 - 0.13 -0.08 mx Ks

n Xm max m

57.8956.98

p=0.05

mp mp

(C)

6.60 -5.64 3.12 -1.20 -1.19 0.90 - 0.77 - 0.70 0.15

Yx Xmax Yp/x Pmax mp m Umax Yxs

PYpx Ks KKii mx n max Ks max m

Parameter

32

m n Xmp max

Figure 3

10.0

0.5 0.41

0.38

8.4

Yp/x (g/g) p/x

µmax (h-1)

0.4

0.43

0.31

0.3 0.2

0.18

0.1 0 24

27

30

33

36

5.9

5.1

3.6

39

21

24

27

30

33

36

39

0.1

73 71

0.088

71.0

70

Yx (g/g) x

Xmax (g/L)

5.2

7.8

7.2

6.6

2.0 21

69.4

68

68.1 67.1

67

0.089

0.076

0.075 0.066

0.064

0.059

0.052

66.3

0.054

0.04

65 21

24

27

30

33

36

21

39

24

0.0058

122 120 117

116.6 115.1

115

113.8

112

27 30 33 36 Temperature (oC)

39

0.0052

118.6

Ki (g/L) i

Pmax max (g/L)

6.8

112.8

0.0046

0.0043

0.0039

0.0046

0.0048

0.0039

0.0033

0.0033

0.0027

110 21

24

27 30 33 36 o Temperature ( C)

39

21

33

24

27 30 33 36 o Temperature ( C)

39

Figure 4

18

13

Cells (g/L)

Cell (g/L)

7

11 7 4

3

5

144

8 Time (h)

10

13

108 72 36 0 80 0

5

10

13 Ethanol (g/L)

64

8 Time (h)

48 32 16 0 3

5

8 Time (h)

10

13

3

144

144

108 72 36

72

17

6 Time (h)

8

10

2

4

6 Time (h)

8

10

2

4

6 Time (h)

8

10

36

68 34

4

72

85 51

2

108

0 90 0

54 36 18 0

0

0

6

0 180 0

0

3

10

0 180

Substrate (g/L)

4

Substrate (g/L)

Cells (g/L)

11

0 180 0 Substrate (g/L)

16

(C)

14

14

Ethanol (g/L)

18

(B)

Ethanol (g/L)

(A)

0

2

4

34

7 Time (h)

9

11

0

Figure 5

(A)

Cell

18.0

Substrate

Ethanol

RSD(%)

14.4 10.8 7.2 3.6 0.0 24

RSD(%)

(B)

25.2 21.6 18.0 14.4 10.8 7.2 3.6 0.0 1

27 30 33 o Temperature ( C)

Cell

Substrate

2

3 4 Condition

35

36

Ethanol

5

6

Yield (%), Ethanol concentration (g/L) 100

1 2

36

3 4 Condition 5

88

10.3 122

10.4 124

9.9 117

9.8 115

10.2 120

Ethanol concentration

84

84

90

87

125

Yield 10.2 120

150

84

Productivity

6

11

9

75 8

50 7

25 5

0 4

Productivity (g/L.h)

Figure 6

12

Cycle 2 4 5 0 Cycle 5 110 15 20 25 Cycle 30 353 40 Cycle 45 50 55Cycle 60 65 70

(D)

(F)

Cycle 2 4 5 0 Cycle 5 110 15 20 25 Cycle 30 353 40 Cycle 45 50 55Cycle 60 65 70

37

Concentrations of X, S and P (g/L)

210 184 158 131 105 79 53 26 0

Cycle 2 4 5 0 Cycle 5 110 15 20 25 Cycle 30 353 40 Cycle 45 50 55Cycle 60 65 70

210 184 158 131 105 79 53 26 0

Concentrations of X, S and P (g/L)

210 184 158 131 105 79 53 26 0

(B)

210 184 158 131 105 79 53 26 0

Concentrations of X, S and P (g/L)

(E)

Concentrations of X, S and P (g/L)

(C)

210 184 158 131 105 79 53 26 0

Concentrations of X, S and P (g/L)

(A)

Concentrations of X, S and P (g/L)

Figure 7

210 184 158 131 105 79 53 26 0

Cycle 2 4 5 0 Cycle 5 110 15 20 25 Cycle 30 353 40 Cycle 45 50 55Cycle 60 65 70

Cycle 2 4 0 Cycle 5 110 15 20 25 Cycle 30 353 40 Cycle 45 50 55Cycle 60 565 70

Cycle 2 4 0 Cycle 5 110 15 20 25 Cycle 30 353 40 Cycle 45 50 55Cycle 60 565 70

Highlights  Ethanol fermentation by S. cerevisiae is conducted over a wide temperature range.  A mechanistic kinetic model is developed to predict reaction rates.  A methodology was proposed to estimate temperature-dependent kinetic parameters.  The applicability of the kinetic model is validated for VHG ethanol fermentation.  Conditions to produce ethanol with a higher yield and productivity are studied.

38