Quantifying second generation ethanol inhibition: Design of Experiments approach and kinetic model development

Bioresource Technology 179 (2015) 219–226

Contents lists available at ScienceDirect

Bioresource Technology

journal homepage: www.elsevier .com/locate /bior tech

Quantifying second generation ethanol inhibition: Designof Experiments approach and kinetic model development

http://dx.doi.org/10.1016/j.biortech.2014.11.0870960-8524/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author at: Department of Chemical and Biological Engineering,South Dakota School of Mines and Technology, 501 East Saint Joseph Street, RapidCity, SD 57701, United States. Tel.: +1 (605) 394 1239; fax: +1 (605) 394 1232.

E-mail address: [email protected] (P.C. Gilcrease).

Steven J. Schneiderman a, Roger W. Johnson b, Todd J. Menkhaus a, Patrick C. Gilcrease a,⇑a Department of Chemical and Biological Engineering, South Dakota School of Mines and Technology, Rapid City, SD 57701, United Statesb Department of Mathematics and Computer Science, South Dakota School of Mines and Technology, Rapid City, SD 57701, United States

h i g h l i g h t s

� Novel DoE methodology was used to guide kinetic model development.� High levels of ethanol masked the effects of other inhibitors.� Removal of ethanol from the DoE design enabled identification of other effects.� A simple kinetic model accounting for DoE-identified inhibition was developed.� DoE was also used to identify significant effects on HMF and furfural reduction.

a r t i c l e i n f o

Article history:Received 8 September 2014Received in revised form 21 November 2014Accepted 22 November 2014Available online 4 December 2014

Keywords:Ethanol fermentationSoftwoodsFermentation inhibitorsDesign of ExperimentsKinetic modeling

a b s t r a c t

While softwoods represent a potential feedstock for second generation ethanol production, compoundspresent in their hydrolysates can inhibit fermentation. In this study, a novel Design of Experiments(DoE) approach was used to identify significant inhibitory effects on Saccharomyces cerevisiae D5A forthe purpose of guiding kinetic model development. Although acetic acid, furfural and 5-hydroxymethylfurfural (HMF) were present at potentially inhibitory levels, initial factorial experiments only identifiedethanol as a significant rate inhibitor. It was hypothesized that high ethanol levels masked the effectsof other inhibitors, and a subsequent factorial design without ethanol found significant effects for allother compounds. When these non-ethanol effects were accounted for in the kinetic model, �R2 was sig-nificantly improved over an ethanol-inhibition only model (�R2 = 0.80 vs. 0.76). In conclusion, when eth-anol masking effects are removed, DoE is a valuable tool to identify significant non-ethanol inhibitors andguide kinetic model development.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Ethanol production from lignocellulosic material such as woodor agricultural wastes is becoming increasingly important due tothe environmental effects, extensive land use, and limited poten-tial of first generation biofuels. Although second generation biofu-els offer the potential for increased quantities of moreenvironmentally friendly fuels, the recalcitrant nature of biomassmakes conversion more challenging. To break the complex struc-ture of lignocellulose and produce fermentable sugars, a harsh pre-treatment step is often required before enzymatic hydrolysis.

Several compounds created during pretreatment are inhibitory toyeasts, including organic acids, aldehydes, and phenolics. Inhibi-tors slow cell growth and can negatively affect ethanol productionrates and yields. If the hydrolysate is then concentrated (a poten-tial strategy to improve ethanol titers), individual inhibitor concen-trations also change. For example, recent results show thatevaporation can remove 100% of furfural, 10.8% of acetic acid and8.9% of 5-hydroxymethyl furfural (HMF) from a pine hydrolysateat pH 5.0 (Gurram and Menkhaus, 2013); however, the 3.4-foldreduction in total volume still resulted in a 3.1-fold increase in ace-tic acid and HMF concentrations. Low pH evaporation can removeadditional acetic acid, but higher pH adjustment costs are thenincurred, and overall acetic acid concentrations still increase (Coxet al., 1993). Similarly, membrane separations such as reverseosmosis or nanofiltration can be used to selectively concentratesugars while removing inhibitory compounds, but membrane foul-

220 S.J. Schneiderman et al. / Bioresource Technology 179 (2015) 219–226

ing can become problematic (Gautam and Menkhaus, 2014;Leberknight and Menkhaus, 2013). It is also possible to use specificoperations that target the removal of inhibitory compounds frombiomass slurries, such as polyelectrolyte flocculation/adsorption,but these can add complexity and cost to the overall process(Burke et al., 2011; Carter et al., 2011a,b). Because inhibitorremoval from concentrated hydrolysates is costly, it is importantto establish which inhibitors are most detrimental, if they interactsynergistically, and the upper limits for hydrolysate concentrationand subsequent fermentation.

Organic acids like acetic acid affect Saccharomyces cerevisiae bydiffusing through the plasma membrane and lowering the intracel-lular pH, which must be neutral for optimal cell function. Low ace-tic acid concentrations can actually stimulate ethanol productionrates and yields (Palmqvist et al., 1999; Palmqvist and Hahn-Hägerdal, 2000); however, higher concentrations are inhibitory,slowing growth and ethanol production. Acetic acid tolerance inS. cerevisiae is strain dependent; for one strain, concentrations aslow as 6 g/l reduced ethanol production by 74% (Phowchindaet al., 1995), while higher concentrations (up to 9 g/l) improvedethanol productivity for another strain (Palmqvist et al., 1999).The medium pH also plays a role in inhibition, as only the un-dis-sociated acid diffuses through the cell membrane, lowering theintracellular pH (Palmqvist and Hahn-Hägerdal, 2000).

Furfural and HMF inhibit glycolytic enzymes and furfural mayalso affect cell membrane integrity. These compounds cause a lagin cell growth until they are metabolized by the organism(Banerjee et al., 1981; Palmqvist and Hahn-Hägerdal, 2000). Furfu-ral inhibition is also strain dependent, with concentrations as lowas 2 g/l reported as inhibitory (Boyer et al., 1992). Larger inoculatend to decrease aldehyde inhibition due to aldehyde metabolismat faster rates. In batch experiments with 2 g/l furfural, inhibitionwas not observed with a relatively high cell concentration(3 gDW/l), but was significant for a cell concentration of0.2 gDW/l (Palmqvist et al., 1999). HMF is not as well studied; itis reported to be less inhibitory than furfural, but is metabolizedmore slowly (Taherzadeh et al., 2000). For one S. cerevisiae strain,HMF concentrations of 3 g/l were reported to be inhibitory(Keating et al., 2006), while another strain maintained 50% of itsinherent ethanol production capacity at HMF concentrations upto 8 g/l (Clark and Mackie, 1984).

The effects of high sugars and ethanol concentrations on S. cere-visiae are also important. Glucose becomes inhibitory at concentra-tions above 100 g/l through osmotic stress effects (Pratt et al.,2003; Shuler and Kargi, 2002), but high gravity concentrations(up to 330 g/l) are still fermentable by industrial strains (Pereiraet al., 2010). As sugars are converted to ethanol, osmotic effectsare replaced with ethanol inhibition. Ethanol decreases membranefluidity, leading to increased proton flux and a lower intracellularpH (Ma and Liu, 2010). Ethanol tolerance has also been shown tobe strain dependent; in high gravity batch experiments, laboratorystrain CEN.PK 113-7D was limited to a final ethanol titer of 130 g/lwhile an industrial strain (PE-2) was able to produce 147 g/l(Pereira et al., 2010).

When multiple inhibitors are present in a fermentation system,the net inhibition may be greater than the additive effects of theindividual compounds (synergistic effect). Acetic acid and furfuralare reported to interact synergistically to inhibit the cell yield, eth-anol yield, and specific growth rate in S. cerevisiae even at low con-centration combinations (0.5 g/l furfural and 5 g/l acetic acid)(Palmqvist et al., 1999). Furfural and HMF can also synergisticallyinhibit S. cerevisiae, completely stopping growth at levels of30 mmol/l each (2.9 g/l Furfural and 3.8 g/l HMF) (Liu et al.,2004). These synergisms become increasingly important in ligno-cellulose hydrolysates which may contain all of these compoundsat inhibitory levels. Because inhibition is strain dependent, syner-

gism reported for one S. cerevisiae strain at one set of concentra-tions may not be observed in another strain; therefore, it isimportant that a kinetic model be strain and concentration specificto accurately predict fermentation performance.

To accurately quantify inhibition in a kinetic model, each signif-icant effect needs to be accounted for. Using the traditional onefactor at a time (OFAT) approach, trial and error is required, eitherexperimentally by varying one inhibitor at a time (some of whichmay not be important), or mathematically by adding and deletingmodel terms to find a best fit to experimental data. In either case,some experiments or kinetic model guesses will be unnecessary orwrong, making the process time consuming and inefficient. Inaddition, many OFAT experiments are required to check for syner-gistic effects.

An alternative to OFAT is the Design of Experiments (DoE)approach. DoE is a rigorous statistical method used to determinethe significant effects of multiple variables on a given systemresponse. In contrast to OFAT, DoE determines significance by run-ning carefully designed experiments in which all possible effectsare present and varied from run to run. By simultaneously check-ing for all possible main effects and synergisms, DoE can reducethe number of experimental runs and amount of trial and errorrequired. DoE methodology has been previously utilized in fermen-tation experiments for growth medium design, inhibitor identifica-tion, fermentation optimization, and identification of variablesimportant to fermentor operation (Graves et al., 2007; Laluceet al., 2009; Palmqvist et al., 1999; Pereira et al., 2010; Unreanand Nguyen, 2012). Specifically relevant to our study is anotherDoE fermentation study (Palmqvist et al., 1999) in which combina-tions of acetic acid, furfural and p-hydroxybenzoic acid were variedfor their effect on different S. cerevisiae strains. DoE methodologyidentified several main effects as well as a synergistic effectbetween acetic acid and furfural; a comparable study with OFATmethodology would have required additional runs and may nothave identified this important synergism.

Our current study uses DoE methodology to identify which lig-nocellulose hydrolysate inhibitors and/or combinations of inhibi-tors significantly impact S. cerevisiae D5A fermentation.Significant DoE responses will then be used to guide developmentof a strain-specific kinetic model that can predict hydrolysate fer-mentation performance under stress from multiple inhibitors. Themajor softwood hydrolysate inhibitors (acetic acid, furfural andHMF), were studied for their effect on our organism. Glucose andethanol were also included to account for possible substrate andproduct inhibition, both of which are important in high gravity fer-mentations. Predicting continuous fermentation with biocatalystrecycle performance was of special interest for follow-on studies;high cell density continuous cultures have been shown to improveethanol productivity substantially and quickly metabolize alde-hydes. To capture ethanol effects present in continuous fermenta-tion systems at steady state, starting ethanol concentrations of 17–50 g/l (DoE #1) and 25–75 g/l (DoE #2) were screened in the initialfactorial designs. Unique to this work is the simultaneous DoEscreening of acetic acid, aldehydes, glucose and ethanol to guiderigorous kinetic model development.

2. Methods

2.1. Experimental design

Design-Expert� 8 software (Stat-Ease) was used to design facto-rial experiments to study the inhibitory effects of glucose, ethanol,acetic acid, furfural and HMF on S. cerevisiae D5A; the specific ratesof cell growth (l), glucose consumption (dS0/dt), and ethanol pro-duction (dP0/dt) were quantified as responses. Later, the overall fer-

S.J. Schneiderman et al. / Bioresource Technology 179 (2015) 219–226 221

mentation rate as measured by mass loss due to CO2 evolution(dCO2/dt) was used as the primary rate response. In addition, celland product yields (Yx/s and Yp/s) were quantified. Initially, a reso-lution V, 2(5 � 1) factorial was selected to quantify main effectsand 2-factor interactions; in a resolution V design, 3-factor andhigher interactions are assumed to be negligible, reducing thenumber of experimental runs to half of a full factorial design; forthis design, 16 runs were required for each experiment. Inhibitorconcentration ranges were based on inhibitory values reported inthe literature and those measured in a dilute-acid pretreated pinehydrolysate (Gurram et al., 2011; Maiorella et al., 1983; Pienkosand Zhang, 2009). The experiment was run at high (OD600 nm = 8)and low (OD600 nm = 2) starting cell densities. Later a full factorial(24) DoE with no starting ethanol was completed at low cell den-sity (OD600 nm = 2.4) with the other inhibitor levels unchanged.The full factorial design allowed for identification of main effects,2-factor interactions, and any possible 3-factor interactions with16 required runs. In total, 48 runs were completed at the concen-tration combinations listed in Table 1. In addition a one factor ata time (OFAT) experiment with only one inhibitor present at a timewas completed. In this experiment, fermentations contained 50 g/lglucose plus one other inhibitor at the high DoE level. For the glu-cose OFAT run, no other inhibitors were added, and the initial glu-cose concentration was 150 g/l. In addition, a control run of 50 g/lglucose (without any other inhibitors) was completed.

2.2. Organism, growth medium, and culture conditions

Starter cultures for the experiments were prepared from a fro-zen culture of S. cerevisiae D5A (provided by the National Renew-able Energy Laboratory (NREL), Golden, CO.). This strain wasselected for tolerance to inhibitors present in dilute-acid pretreat-ment hydrolysates, and is the reference organism for standardNREL protocols. The starter medium (chemicals from Fisher Scien-tific unless noted otherwise) consisted of glucose (50 g/l), peptone(20 g/l), and yeast extract (10 g/l – Marcor) adjusted to pH 5 withsulfuric acid prior to autoclaving at 121 �C for 15 min. DoE fermen-tation medium was adapted from a very high gravity recipedesigned to provide a cheap source of nutrients that could be usedindustrially (Pereira et al., 2010). Yeast extract was substituted forcorn steep liquor based on nitrogen content for ease of preparationand use. The base high gravity medium consisted of: Yeast extract(11.8 g/l), Urea (2.3 g/l), MgSO4�7H2O (3.8 g/l) and CuSO4�5H2O(.03 g/l). To validate the defined medium as a surrogate hydroly-sate, appropriate levels of acetic acid, furfural (Sigma Aldrich)and HMF (Sigma Aldrich) were added to the base recipe and fer-mented side-by-side with a nutrient supplemented pine hydroly-sate. Fermentation results were very similar, confirming that areasonable hydrolysate surrogate could be made by adding thesethree main inhibitors.

Table 1High/Low Inhibitor Levels – Inhibitor concentrations used in the factorial designs(DoE # 1–3) and OFAT experiment Individual OFAT fermentations included 50 g/lglucose plus one additional inhibitor at the concentration listed, except for theglucose only fermentation, which was run at 150 g/l glucose. A control at 50 g/lglucose (no inhibitors) was also included.

2-Level factorial low–high [g/l]

Component DoE #1 DoE #2 DoE #3 OFAT

Glucose 50–150 50–150 50–150 150Ethanol 17–50 25–75 0 75Acetic acid 2–6 2–6 2–6 6Furfural 1–3 1–3 1–3 3HMF 1–3 1–3 1–3 3

2.3. Starter and experimental fermentations

Starter cultures were grown aerobically at 30 �C and 150 RPMagitation in a bench-top shaker. The cells were centrifuged (7000RPM for 10 min) and washed with 0.9% NaCl buffer followed byre-centrifugation and re-suspension in NaCl buffer to a final vol-ume of approximately 100 ml.

Serum bottles (100 ml working volume) containing the modi-fied high gravity medium were supplemented with the five poten-tial inhibitors (glucose, ethanol, acetic acid, furfural and HMF)according to the concentration levels in Table 1. The bottles wereadjusted to pH 5, diluted with water to the final 100 ml volume,and sealed with rubber septa and metal crimp rings to preventthe loss of volatile components during autoclaving (15 min at121 �C). After cooling, the bottles were vented with 23G, 100 longsterile needles with sterile gauze covering the luer lock fitting. Fiveml of the concentrated starter culture was used to inoculate each ofthe 16 experimental serum bottles; cultures were then incubatedat 30 �C (for better ethanol productivity) and 200 RPM.

Three ml samples were taken throughout the fermentation bysyringe (with flame sterilized needles) to quantify optical den-sity, pH, and the concentration of each inhibitor. Glucose, etha-nol, acetic acid, and aldehydes were quantified via HPLC(Beckman System Gold HPLC equipped with Aminex HPX-87Hcolumn and Jasco RI detector). Optical densities were read at600 nm after diluting with sterile medium to obtain raw read-ings 60.5 (Beckman DU 640). For the OFAT and third DoE exper-iment, bottles were weighed at regular intervals to quantify CO2

production. The amount of CO2 evolved was used as the primaryresponse to determine significant effects along with yields calcu-lated from initial and final HPLC and cell dry weight sampling.Dry weights were obtained by centrifuging 10 ml of fermentationbroth, washing with 10 ml 0.9% NaCl, re-centrifuging, suspendingthe cell pellet in DI water and drying in a pre-weighed tin at65 �C.

2.4. Response calculation and analysis

For DoE experiment #1, responses for specific growth rate (l),specific substrate utilization rate (dS0/dt), and specific product for-mation rate (dP0/dt) were regressed from the approximately linearrange of data following the reduced growth or lag phase (if pres-ent). For DoE experiment #2, the first five hourly data points wereregressed. In both cases, the specific growth rate was determinedfrom the slope of a natural log OD vs. time plot, and specific glu-cose and ethanol rates were calculated by dividing the slope ofthe concentration vs. time plot by the average optical density overthe regression range. Individual yields for experiment #1 were cal-culated by fitting each separate run to the Monod equation usingACSLX (Aegis Technologies) software. The software minimizesthe overall model error through an iterative process of numericalintegration and parameter optimization. This approach allowedfor using best-fit values for Yx/s and Yp/s over the entire run asresponses as opposed to relying only on the initial and final datapoints. For DoE experiment #3 and the OFAT experiment, CO2 pro-duction rates were determined by regressing the approximatelylinear range of the data between 1 and 6 h, and yields were calcu-lated from initial and final samples.

Responses were analyzed with Design Expert 8� (Stat-Ease)for significant main effects and interactions. Half-normal andPareto charts were used to identify significant effects that wereconfirmed through analysis of variance (ANOVA). Diagnosticssuch as normal and Box-Cox plots were used to confirm thatthe modeled effects were statistically valid. In some cases, theraw data was transformed as suggested by the Box-Cox analysisand the responses re-analyzed by the above procedure. Least

222 S.J. Schneiderman et al. / Bioresource Technology 179 (2015) 219–226

squared difference (LSD) analysis was used to confirm the 2-fac-tor interactions; if an error bar overlap was observed, the inter-action was deemed insignificant and removed from thesignificant effects list prior to reanalysis.

Table 2DoE #1 ANOVA – ANOVA results for DoE #1 detailing responses, response range,transform used (if any), significant terms identified (and their effect +/�), p-valuesand standardized effect.

DoE #1 – High cell density (ODstart = 8)

Response Source p-Value Standardizedeffect

Specific growth rate (l) Ethanol (�) 0.0294 �0.037Range: 0.03–0.19 h�1

Transform: inverse

Specific glucose utilizationrate (dS0/dt)

Ethanol (�) <0.0001 �0.57

Range: 0.11–1.18 g/l/h/ODTransform: SQRT

Specific ethanol productionrate (dP0/dt)

Ethanol (�) 0.0001 �0.17

Range: 0.01–0.38 g/l/h/ODTransform: none

Cell yield (Yx/s) Glucose (�) 0.0295 �0.023Range: 0.04–0.11 OD/gTransform: none

Product yield (Yp/s) None – –

2.5. Kinetic model development

Based on the results of the DoE analysis, kinetic models weredeveloped to quantify the effects of significant inhibitors. Forsimplicity, exponential inhibition terms were used to accountfor growth inhibition with one constant per inhibitory term(Nagatani et al., 1968). Differential mass balance equations wereused to account for changes in substrate, cell and product con-centrations using the growth rate and yield coefficients. Best-fitmodel parameter values were derived from the experimentaldata using ACSLX (Aegis Technologies) software. The Nelder-Mead and Quasi-Newton optimization routines (provided as partof the program) were used to fit the data in this study. Threeseparate kinetic models based on the Monod equation were usedto quantify specific growth rate inhibition (Eqs. 1–3):

l ¼ lmax � S � exp ð�k1 � PÞðKsþ SÞ ð1Þ

l ¼ lmax � S � exp ð�k1 � PÞ � exp ð�k2 � A0 � k3 � F0 � k4 � H0ÞðKsþ SÞ ð2Þ

l ¼ lmax � S � exp ð�k1 � PÞ � exp ½�k2 � ðA0 þ F0 þ H0Þ�ðKsþ SÞ ð3Þ

For all three growth rate models, the maximum specificgrowth rate (lmax) and half-velocity constant (Ks) were set asconstant after fitting the unmodified Monod Equation to a fer-mentation run with the same media as all experiments here withlow levels of glucose and no other inhibitors. The regressedparameters (lmax = 0.30 h�1 and Ks = 0.70 g/l) fall within therange of reported values for S. cerevisiae. All models quantify eth-anol inhibition with an exponential inhibition term containingthe ethanol inhibition constant (k1) and ethanol concentration(P). Eqs. (2 and 3) also quantify inhibition from initial concentra-tions of acetic acid (A0), furfural (F0) and HMF (H0). In Eq. (2) eachnon-ethanol inhibitor has its own inhibition constant while Eq.(3) lumps the inhibitors together with one inhibition constant(k2). The specific growth rate is used to predict concentrationsthroughout batch fermentation using the following differentialequations for cell mass in OD units (X), glucose in g/l (S), and eth-anol in g/l (P) (Eqs. 4–6):

dXdt¼ l � X ð4Þ

dSdt¼ �l � X

Yx=sð5Þ

dPdt¼ l � X � Yp=s

Yx=sð6Þ

Rates of glucose consumption and ethanol production arerelated to the absolute growth rate (dX/dt) using yield coefficientsfor the amount of cell mass produced per glucose consumed (Yx/s)and ethanol produced per glucose consumed (Yp/s).

To compare the fit of the three inhibition models, a non-linear,adjusted R2 value (�R2) was calculated for the predicted vs. actualdata using the following set of equations for regression sum ofsquares (RSS), total sum of squares (TSS), R2, and �R2 (Eqs. 7–10).�R2 corrects for the automatic increase in R2 as additional termsare added to the model:

RSS ¼XðYexp � YpredÞ2 ð7Þ

TSS ¼XðYexp � �YpredÞ

2 ð8Þ

R2 ¼ TSS� RSSTSS

ð9Þ

�R2 ¼ R2 � ð1� R2Þ pn� p� 1

ð10Þ

In Eq. (7), RSS is calculated from experimental values (Yexp) andpredicted values (Ypred). TSS is calculated from Yexp and an averageof all experimental values �Yexp. In Eq. (10), �R2 is calculated from R2,the number of regressors (p), and the number of samples (n).

3. Results and discussion

3.1. Significant effects analysis

Significant effects on the responses (l, dS0/dt, dP0/dt, Yx/s andYp/s) for Experiment #1 are reported in Table 2. In the table the signon significant effects (+/�) corresponds to a positive or negativeeffect on that specific rate/yield. P-values are reported fromANOVA analysis in which a probability of less than 5% (p = 0.05)indicates significance. In addition, the magnitude of each signifi-cant inhibition effect can be seen in the Standardized Effect col-umn. Standardized effect represents the difference between theaverage response at high factor level and average response atlow factor level. In cases where a transform was applied to thedata, standardized effect is reported from the non-transformeddata to provide a meaningful magnitude in the same units as theresponse.

Ethanol was the only significant factor identified by DoE analy-sis to affect rate; in the half-normal analysis it stood out from allother factors, especially in the specific glucose reduction and spe-cific ethanol production responses. This is not surprising as ethanolconcentrations (17–50 g/l) were much higher than the aldehydeand acetic acid concentrations. In addition ethanol affects multiplecell systems by damaging the cell membrane; this impairs nutrientand ionic species transport and requires additional maintenanceenergy to repair. Glucose was identified as negatively affecting cellyield, but did not show up as a significant rate effect. Previous inhi-bition studies have identified acetic acid, furfural, and HMF as sig-nificant inhibitors; as such, we were somewhat surprised when

Table 3DoE #2 ANOVA – ANOVA results for DoE #2 detailing responses, response range,transform used (if any), significant terms identified (and their effect +/�), p-valuesand standardized effect.

DoE #2 – low cell density (ODstart = 2)

Measured response Source p-Value Standardizedeffect

Specific growth rate (l) Ethanol (�) 0.0060 �0.052Range: 0–0.14 h�1

Transform: inverse SQRT

Specific glucose utilization rate(dS0/dt)

Ethanol (�) <0.0001 �0.94

Range: 0–1.4 g/l/h/ODTransform: SQRT

Specific ethanol productionrate (dP0/dt)

Ethanol (�) 0.0145 �0.29

Range: 0–0.94 g/l/h/ODTransform: SQRT

S.J. Schneiderman et al. / Bioresource Technology 179 (2015) 219–226 223

significant effects were not observed for these non-ethanol factors.A possible problem identified in experiment #1 was the fast alde-hyde reduction associated with higher cell densities. On average,the furfural concentration had declined by 88% and HMF by 26%by the second sample point (after 2.8 h). Because furfural wasalmost completely reduced before the time interval used to regressspecific rates, the experiment may not have accurately capturedthe rate effects of furfural/HMF or their synergisms. A similar prob-lem was encountered by Palmqvist, who observed furfural inhibi-tion at low cell density (0.2 gDW/l) but not at high cell density(3 gDW/l) (Palmqvist et al., 1999). To better capture the effects ofshort-lived aldehyde inhibitors, DoE experiment #2 was carriedout at a lower cell density (OD = 2) with more frequent initial sam-ple times. At a lower cell density, aldehyde reduction rates wereslower, allowing for better quantification of slopes while the alde-hyde concentrations were relatively unchanged. To ensure ratesrepresented the initial effect of these inhibitors, only the first 5 hof fermentation samples were used to regress rates. Because sev-eral of the runs did not have observable growth or substrate con-sumption during this period, yields were not quantified in thisexperiment. ANOVA results from the second DoE experiment areshown in Table 3; again ethanol was the only significant inhibitordetected. This result argues against the hypothesis that high celldensities masked the effects of non-ethanol inhibitors by removingaldehydes quickly. Either S. cerevisiae D5A is exceptionally inhibitortolerant or something else is causing significant non-ethanol inhib-itors to be missed. One potential explanation is that the differencebetween low and high concentration levels for the other inhibitors

Fig. 1. DoE Response Plots – Plots of specific growth, substrate reduction and ethanol proconcentrations.

was not large enough for an observable difference in rate to occur.To evaluate non-ethanol inhibition effects over a larger concentra-tion range, responses from the first DoE experiment were plottedvs. the sum of acetic acid, furfural and HMF, shown in Fig. 1.

A negative trend in specific rates was observed as the sum ofthese components increased, even though in many cases furfuraland HMF were significantly reduced before the time period fromwhich these responses were regressed; however, these trends wereless discernable for high ethanol concentration runs. The samegeneral trend was observed in Experiment #2, but because of thelower cell density, several responses had very small absolute rates(close to the experimental error of OD and HPLC) making specificrates more variable and the trend less clear. These results suggestthat because ethanol is present in such a large quantity at the highDoE levels (from 50 to 75 g/l), its effect may be so large that itmasks the effects of other inhibitors. The DoE conclusion that eth-anol is the only significant factor may be misleading in that othersignificant factors are present, but their effects are small comparedto the effect of high ethanol concentrations. A similar hypothesiswas given for a chemostat study on byproduct inhibition(Maiorella et al., 1983); feed glucose concentrations were reducedto 20 g/l to prevent ethanol from masking the other inhibitoreffects. To test this idea, a one factor at a time (OFAT) experimentwas carried out; six separate runs were completed in which eachrun (except the control run) had one potential inhibitor at the highDoE level. A control with 50 g/l glucose and no other inhibitors wasalso included. Yields were quantified using initial and final sam-ples, and the fermentation rate was measured by weighing theserum bottles at regular intervals. Fig. 2 shows the results in termsof % change from the control (100%). Control responses were0.77 gCO2/hr, Yx/s = 0.11 OD/g and Yp/s = 0.42 g/g.

The OFAT experiment showed that significant effects can beobserved when each potential inhibitor is present by itself. In par-ticular, furfural and HMF exhibited strong negative effects on theCO2 production rate. All compounds reduced cell yield with glu-cose giving the largest response (44% reduction), consistent withearlier DoE results. Ethanol and glucose affected product yield less,reducing Yp/s by 15% and 5%, respectively. Acetic acid and HMFincreased Yp/s slightly while furfural caused a slight reduction. Thismay help explain why no significant effects were detected for Yp/s

in DoE experiment #1, because all compounds are present in eachrun and positive effects may cancel out negative effects. The effectof ethanol on rate was by far the largest negative value, reducingCO2 evolution by 71%. Overall, the experiment indicated that sig-nificant effects besides ethanol are indeed present. Because of this,the DoE experiment was repeated at low starting cell density(OD = 2.4) without ethanol, but keeping other inhibitors at the

duction rates from DoE #1 plotted vs. the sum of initial acetic acid, furfural and HMF

Fig. 2. OFAT Experiment Results – OFAT results showing % response obtained compared to the control run with no inhibitors.

224 S.J. Schneiderman et al. / Bioresource Technology 179 (2015) 219–226

same levels. Rather than reducing runs to a 1/2 factorial level (8runs), 16 runs were again completed (full factorial), allowing forquantification of any 3-factor interactions that might be present.Fermentation rate was quantified by mass loss similarly to theOFAT experiment; using mass loss decreased sampling time andits associated error. Table 4 shows the results of this experiment.

Several significant effects not captured in the previous experi-ments were identified when ethanol was not present. In particular,fermentation rate was most affected by furfural and HMF, consis-tent with the OFAT experiment. A slight positive effect on fermen-tation rate from acetic acid was observed when ethanol was notpresent. Cell yield was negatively affected by all four componentsbut glucose provided the largest decrease, consistent with DoEexperiment #1 and the OFAT experiment. No significant effectswere seen for Yp/s, consistent with DoE #1. In addition, no two-fac-tor or higher interactions were observed, indicating that syner-gisms do not appear to be significant for the concentrationranges studied in this work.

3.2. Kinetic model development

In the initial Design of Experiments analysis, ethanol was iden-tified as the only significant inhibitor that affected all rateresponses; therefore, we started with a kinetic model where theonly inhibition effect on rate was ethanol. The extended Monodequation (Eq. (1)), along with Eqs. (4–6) was fit to the data usingACSLX software. Cell yield was quantified by the model equation

Table 4DoE #3 ANOVA – ANOVA results for DoE #3 detailing responses, response range,transform used (if any), significant terms identified (and their effect +/�), p-valuesand standardized effect.

DoE #3 – No EtOH (ODstart = 2.4)

Measured response Source p-Value Standardized effect

Fermentation rate(gC02/hr)

Furfural (�)HMF (�)Glucose (�)Acetic acid (+)Model

<0.0001<0.00010.00160.0273<0.0001

�0.094�0.091�0.0300.018

Range: 0.27–0.51 g/hTransform: none

Cell yield (Yx/s) Glucose (�)Acetic acid (�)Furfural (�)HMF (�)Model

<0.0001<0.00010.00060.0022<0.0001

�0.022�8.3E-3�6.6E-3�5.5E-3

Range: 0.03–0.07 gDW/gTransform: none

Product yield (Yp/s) None – –

provided by DoE #1 based on glucose being the only negative sig-nificant effect (Yx/s = 0.088–2.267E-4 * S(t = 0) where S(t = 0) repre-sents the initial glucose concentration). Though DoE #3 identifiedother significant yield effects besides glucose, glucose was by farthe largest in magnitude. Since none of the DoE experiments iden-tified a negative effect on product yield, Yp/s was allowed to varywith the other kinetic parameters in ACSLX to find the single bestvalue that fit all of the DoE #1 data; only the DoE#1 high cell den-sity data was used because it was collected over a much longerperiod (30 h vs. 5 h in DoE experiment #2), and included multiplepoints for cell mass, glucose and ethanol (as opposed to DoE #3). Inaddition to the ethanol-only kinetic model, a more complex modelin which rate was inhibited by individual ethanol, acetic acid, fur-fural and HMF terms was also fit to the data (Eq. (2)). Again, cellyield was quantified with the DoE based equation and productyield allowed to vary in ACSLX. Although DoE #3 did not identifyacetic acid as a negative effect on fermentation rate, other OFATexperiments (not shown), literature, and Fig. 1 suggest that aceticacid negatively affects growth rates and cell yield at levels at orbelow 5 g/l (Maiorella et al., 1983); therefore, acetic acid wasincluded in the inhibition model to account for its negative effecton growth even though it appears to have a slight positive effecton ethanol productivity.

As another simplification to the model, initial (t = 0) inhibitorconcentrations were used in the exponential inhibition terms foracetic acid, furfural, and HMF, similar to the approach of Boyer inthe development of a furfural inhibition model (Boyer et al.,1992). In contrast, the ethanol inhibition term in each model uses

Table 5Kinetic Parameters and Model Fits – Summary of kinetic model parameters and �R2

values for cell mass (X), substrate (S), and ethanol (P).

Parameter EtOH only Eq. (1) EtOH and A, F,H Eq. (2)

EtOH & sum(A, F, H)Eq. (3)

lmax [hr�1] 0.30 0.30 0.30Ks [g/l] 0.70 0.70 0.70k1 [l/g] .058 0.040 0.043k2 [l/g] – 0.120 0.085k3 [l/g] – 0.092 –k4 [l/g] – 0.033 –Yp/s [g/g] 0.37 0.40 0.33

X � �R2 0.39 0.51 0.48

S � �R2 0.96 0.97 0.98

P � �R2 0.92 0.92 0.93

Average �R2 0.76 0.80 0.79

Fig. 3. Kinetic Model Fits – Kinetic model fits for Eqs. (1 and 3) models to selected runs from the DoE #1 dataset (Left Y-axis: Glucose/EtOH concentration [g/l], Right Y-axis:Optical Density (OD600 nm), X-axis: Time [hr]).

S.J. Schneiderman et al. / Bioresource Technology 179 (2015) 219–226 225

a real-time concentration as predicted by Eq. (6). Accounting forthe transient behavior of inhibitor concentration (and its effecton inhibition) is of interest in future work, but was beyond thescope of this paper. Table 5 shows the parameters regressed forthe models along with �R2 values for all runs for each variable (X– cell mass, S – glucose, P – ethanol) and an average �R2 value forall variables.

The model described by Eq. (2) improved the average �R2 valuefrom 0.76 to 0.80 over the ethanol-only model (Eq. (1)). Ethanoland glucose concentrations were predicted very well with �R2 val-ues >0.90 for both models. Cell mass model predictions had lower

�R2 values, partly due to the nature of the �R2 calculation (responsesin which a larger change in concentration is present automaticallyhave higher �R2 values), but also due to the presence of a growthlag phase not included in our model. A simpler model was alsofit to the data in which the non-ethanol inhibitors were repre-sented by a lump sum term with one inhibition constant (Eq.(3)). An average �R2 of 0.79 was obtained with the Eq. (3) model,suggesting that the additional complexity of the Eq. (2) model(with 3 separate inhibition constants) may not be justified.

Comparing the inhibition parameters obtained here with litera-ture parameters proved difficult because few experiments have

226 S.J. Schneiderman et al. / Bioresource Technology 179 (2015) 219–226

tried to develop inhibition models for multiple non-ethanol inhib-itors (none for ethanol + acetic acid + furfural + HMF), and the com-bined inhibition effect term (A + F + H) is unique to this study. Inaddition, no kinetic models were found specific to our strain (S.cerevisiae D5A). However, exponential ethanol inhibition termsare widely used and multiple literature sources were found forcomparison. A wide range of values (from 0.016 to 0.127 l/g) havebeen reported for ethanol inhibition (Boyer et al., 1992; Damianoet al., 1985; Jin et al., 1981; Nagatani et al., 1968); values obtainedfrom our experimental fits (0.040–0.058 l/g) fall within this range.Averaging 8 values reported in the referenced literature gave anaverage k1 value of 0.070 l/g, which is higher than the valuesobtained in our model fits. However, Boyer found that k1 decreasedat higher cell densities with a value of 0.056 l/g obtained with aninoculum of 5 gDW/l and 0.034 l/g obtained at 9 gDW/l (Boyeret al., 1992). Our experiment was run with an inoculum of 8 OD,which corresponds to 5.6 gDW/l; interpolation of Boyer’s datawould predict a k1 of 0.053 which is closer to our k1 values of0.040 and 0.043 l/g obtained with Eqs. (2 and 3) fits, respectively.Our Eq. (1) fit gave a k1 of 0.058 l/g, but in this case we hypothesizek1 is accounting for the combined inhibitory effects of ethanol andthe other compounds and is therefore higher.

Fig. 3 depicts the model fits for four runs that describe the limitsof the inhibition space studied (i.e. ‘‘Low to High [Ethanol + Glu-cose]’’ and ‘‘Low to High [Acetic Acid + Furfural + HMF]’’). Whileboth models fit the all low (Run 7) and all high (Run 11) inhibitionextremes similarly, model differences are observed when only oneof the inhibition axes is at a high level (Runs 14 and 12). In thesecases, the two term model (Eq. (3)) fits the data significantly betterthan the ethanol-only model (Eq. (1)). Adding the second term todistinguish aldehyde and acetic acid inhibition from ethanol inhi-bition allows for a better fit when one or the other inhibitor levelsdominate. Conversely, when all inhibitors are present at high levels(Run 11), it is not so important to distinguish individual inhibitorcontributions in the kinetic model. The fact that Run 12 (high glu-cose + ethanol but low acetic acid and HMF) is improved by addingthe combined inhibitor term suggests that the ethanol term in Eq.(1) is actually capturing the effect of ethanol plus the other inhib-itors; when levels of these other inhibitors are reduced the etha-nol-only model does not predict ethanol inhibition as well. The 2term model (Eq. (3)) allows for an improved fit over the entirerange of inhibitor concentrations studied without becoming need-lessly complex as with the Eq. (2) model.

4. Conclusions

A DoE approach was used to identify S. cerevisiae D5A inhibitionin the presence of glucose, ethanol, acetic acid, furfural, and HMF forthe purpose of guiding kinetic model development. Results indicatethat high ethanol concentrations can mask the significant effects ofother inhibitors in factorial experiments; kinetic modeling verifiedthe importance of these non-ethanol effects, as the overall �R2

improved from 0.76 to 0.80 when they were included in the model.In conclusion, when ethanol masking effects are removed, this DoEapproach can be used to identify significant non-ethanol inhibitorsand guide development of a working kinetic model.

Acknowledgements

This work was supported by a USDA NIFA, AFRI CompetitiveGrant #2010-65504-20372. A pure culture of Saccharomyces cerevi-siae D5A was kindly provided by the National Renewable EnergyLaboratory, Golden, CO. We also wish to thank Stat-Ease Inc., fortheir technical support and guidance in using Design Expert� soft-ware and Design of Experiments methodology.

References

Banerjee, N., Bhatnagar, R., Viswanathan, L., 1981. Inhibition of glycolysis by furfuralin Saccharomyces cerevisiae. Eur. J. Appl. Microbiol. Biotechnol. 11, 226–228.

Boyer, L.J., Vega, J.L., Klasson, K.T., Clausen, E.C., Gaddy, J.L., 1992. The effects offurfural on ethanol production by Saccharomyces cerevisiae in batch culture.Biomass Bioenergy 3, 41–48.

Burke, D.R., Anderson, J., Gilcrease, P.C., Menkhaus, T.J., 2011. Enhanced solid–liquidclarification of lignocellulosic slurries using polyelectrolyte flocculating agents.Biomass Bioenergy 35, 391–401.

Carter, B., Gilcrease, P.C., Menkhaus, T.J., 2011a. Removal and recovery of furfural, 5-hydroxymethylfurfural, and acetic acid from aqueous solutions using a solublepolyelectrolyte. Biotechnol. Bioeng. 108, 2046–2052.

Carter, B., Squillace, P., Gilcrease, P.C., Menkhaus, T.J., 2011b. Detoxification of alignocellulosic biomass slurry by soluble polyelectrolyte adsorption forimproved fermentation efficiency. Biotechnol. Bioeng. 108, 2053–2060.

Clark, T.A., Mackie, K.L., 1984. Fermentation inhibitors in wood hydrolysates derivedfrom the softwood Pinus radiata. J. Chem. Technol. Biotechnol. 34, 101–110.

Cox, M.G.S., Mohabir, K., Hoekstra, R.G., 1993. The volatilisation and condensation ofacetic acid during cane juice evaporation. Proc. Annu. Cong. S. Afr. SugarTechnol. Assoc. 67, 148–154.

Damiano, D., Shin, C.-S., Ju, N-h., Wang, S., 1985. Performance, kinetics, andsubstrate utilization in a continuous yeast fermentation with cell recycle byultrafiltration membranes. Appl. Microbiol. Biotechnol. 21, 69–77.

Gautam, A., Menkhaus, T.J., 2014. Performance evaluation and fouling analysis forreverse osmosis and nanofiltration membranes during processing oflignocellulosic biomass hydrolysate. J. Membr. Sci. 451, 252–265.

Graves, T., Narendranath, N., Dawson, K., Power, R., 2007. Interaction effects of lacticacid and acetic acid at different temperatures on ethanol production bySaccharomyces cerevisiae in corn mash. Appl. Microbiol. Biotechnol. 73, 1190–1196.

Gurram, R.N., Menkhaus, T.J., 2013. Analysis and characterization of heat transferfouling during evaporation of a lignocellulosic biomass process stream. Ind. Eng.Chem. Res. 52, 11111–11121.

Gurram, R.N., Datta, S., Lin, Y.J., Snyder, S.W., Menkhaus, T.J., 2011. Removal ofenzymatic and fermentation inhibitory compounds from biomass slurries forenhanced biorefinery process efficiencies. Bioresour. Technol. 102, 7850–7859.

Jin, C.K., Chiang, H.L., Wang, S.S., 1981. Steady state analysis of the enhancement inethanol productivity of a continuous fermentation process employing aprotein–phospholipid complex as a protecting agent. Enzyme Microb.Technol. 3, 249–257.

Keating, J.D., Panganiban, C., Mansfield, S.D., 2006. Tolerance and adaptation ofethanologenic yeasts to lignocellulosic inhibitory compounds. Biotechnol.Bioeng. 93, 1196–1206.

Laluce, C., Tognolli, J., de Oliveira, K., Souza, C., Morais, M., 2009. Optimization oftemperature, sugar concentration, and inoculum size to maximize ethanolproduction without significant decrease in yeast cell viability. Appl. Microbiol.Biotechnol. 83, 627–637.

Leberknight, J., Menkhaus, T.J., 2013. Membrane separations for solid–liquidclarification within lignocellulosic biorefining processes. Biotechnol. Prog. 29,1246–1254.

Liu, Z.L., Slininger, P.J., Dien, B.S., Berhow, M.A., Kurtzman, C.P., Gorsich, S.W., 2004.Adaptive response of yeasts to furfural and 5-hydroxymethylfurfural and newchemical evidence for HMF conversion to 2,5-bis-hydroxymethylfuran. J. Ind.Microbiol. Biotechnol. 31, 345–352.

Ma, M., Liu, Z.L., 2010. Mechanisms of ethanol tolerance in Saccharomyces cerevisiae.Appl. Microbiol. Biotechnol. 87, 829–845.

Maiorella, B., Blanch, H.W., Wilke, C.R., 1983. By-product inhibition effects onethanolic fermentation by Saccharomyces cerevisiae. Biotechnol. Bioeng. 25,103–121.

Nagatani, M., Shoda, M., Aiba, S., 1968. Kinetics of product inhibition in alcoholfermentation. I. Batch experiments. J. Ferment. Technol. 46, 241.

Palmqvist, E., Hahn-Hägerdal, B., 2000. Fermentation of lignocellulosic hydrolysates.II: inhibitors and mechanisms of inhibition. Bioresour. Technol. 74, 25–33.

Palmqvist, E., Grage, H., Meinander, N.Q., Hahn-Hägerdal, B., 1999. Main andinteraction effects of acetic acid, furfural, and p-hydroxybenzoic acid on growthand ethanol productivity of yeasts. Biotechnol. Bioeng. 63, 46–55.

Pereira, F.B., Guimarães, P.M.R., Teixeira, J.A., Domingues, L., 2010. Optimization of low-cost medium for very high gravity ethanol fermentations by Saccharomycescerevisiae using statistical experimental designs. Bioresour. Technol. 101, 7856–7863.

Phowchinda, O., Délia-Dupuy, M.L., Strehaiano, P., 1995. Effects of acetic acid ongrowth and fermentative activity of Saccharomyces cerevisiae. Biotechnol. Lett.17, 237–242.

Pienkos, P., Zhang, M., 2009. Role of pretreatment and conditioning processes ontoxicity of lignocellulosic biomass hydrolysates. Cellulose 16, 743–762.

Pratt, P.L., Bryce, J.H., Stewart, G.G., 2003. The effects of osmotic pressure andethanol on yeast viability and morphology. J. Inst. Brew. 109, 218–228.

Shuler, M.L., Kargi, F., 2002. Bioprocess engineering: basic concepts. Prentice Hall,Upper Saddle River, NJ.

Taherzadeh, M.J., Gustafsson, L., Niklasson, C., Lidén, G., 2000. Physiological effectsof 5-hydroxymethylfurfural on Saccharomyces cerevisiae. Appl. Microbiol.Biotechnol. 53, 701–708.

Unrean, P., Nguyen, N.H.A., 2012. Rational optimization of culture conditions for themost efficient ethanol production in Scheffersomyces stipitis using Design ofExperiments. Biotechnol. Prog. 28, 1119–1125.