Chemical Engineering Science: X 8 (2020) 100085

Contents lists available at ScienceDirect

Chemical Engineering Science: X

journal homepage: www.elsevier.com/locate/cesx

Mathematical modelling for the optimization of cellulase production

using glycerol for cell growth and cellulose as the inducer substrate
Lucas Gelain a,b,⇑, Luuk van der Wielen a,c, Walter M. van Gulik a, José Geraldo da Cruz Pradella d,
Aline Carvalho da Costa b
a
Delft University of Technology, Department of Biotechnology, Van der Maasweg 9, 2629HZ Delft, the Netherlands
b
University of Campinas, School of Chemical Engineering, Av. Albert Einstein, 500, Campinas, Brazil
c
University of Limerick, Bernal Institute, V94 T9PX Limerick, Ireland
d
Federal University of São Paulo, Institute of Science and Technology, Av. Cesare Mansueto Giulio Lattes, 1201 S. J. Campos, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Cellulase production can be divided into two steps: growth stage; followed by an induction stage. To
develop a mathematical model for the optimization of this strategy, two sets of experiments were per-
formed in batch mode for parameter estimation. One set of experiments was performed to evaluate
Keywords: the influence of glycerol regarding cell growth (initial concentrations of 5, 10, 15 and 20 g/L). The other
Mathematical modelling set of experiments considered the induction stage using cellulose as the substrate (initial concentrations
Trichoderma harzianum of 5, 10, 20, 30 and 40 g/L). Two feeding strategies were simulated to maximize cellulase production using
Cellulose
glycerol to maintain a high cell concentration. The first simulation used a discrete feed and the second
Glycerol
Bioprocess
used a continuous feed of cellulose. The mathematical model proposed allows maintaining a high cell
Optimization of cellulase production concentration and the addition of optimal small amounts of the inducer substrate to prevent inhibition
of enzyme production.
Ó 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).

1. Introduction performance. Delabona et al. (2016) showed that a two-stage pro-

cess consisting of growth of T. harzianum on glycerol, followed by
In the biotechnological production of biofuels, such as ethanol induction with sugarcane bagasse pretreated led to an important
and other chemicals using lignocellulosic materials, hydrolysis is increase in productivity and cellulase activity. The authors sug-
one of the most important steps of the process. The enzymes used gested that the increase in production was due to a greater number
in the hydrolysis step (cellulase) can be produced by filamentous of active tips of mycelia as well as long hyphae, which increased
fungi of the genus Trichoderma, which is well adapted for biopro- protein secretion capacity. Additionally, glycerol is reported being
cesses (Strakowska et al., 2014). a ‘‘neutral” carbon source (Ilmén et al., 1997), thus could prevent
The importance of cellulase production goes beyond its use in catabolite repression to occur.
lignocellulose hydrolysis. Recent market reports show that cellu- Studies on decreasing cellulase production costs are important
lase has increasingly been used in many industrial applications, since productivity and enzymatic activity in the growth cultures
such as coffee processing, winemaking, fruit juice production, are in general low. To achieve this goal, mathematical modelling
paper and pulping as well as laundry detergents and the produc- becomes an important tool as mathematical models can be used
tion of cleaning and washing agents (Jayasekara and Ratnayake, to develop optimal strategies. Therefore, the development of math-
2019). These authors also cite applications in agriculture and med- ematical models for the two-stage process described in Delabona
ical area. et al. (2016) could be an interesting approach aiming at the maxi-
The production of cellulase has been determined as non-growth mization of the cellulase productivity in fed-batch mode since the
associated (Gelain et al., 2015). Thus, separation of the process into model can be used to suggest optimal feeding strategies.
two-stage, allowing the optimization of the growth stage and the In this work, a mathematical model for cellulase production
production stage separately may result in increased process using glycerol for cell growth and cellulose as the inducer substrate
is proposed and used to simulate optimal feeding policies in a fed-
batch cellulase production process. The experiments were per-
⇑ Corresponding author. formed in batch mode using different initial concentrations of
E-mail address: [email protected] (L. Gelain).

https://doi.org/10.1016/j.cesx.2020.100085
2590-1400/Ó 2020 The Author(s). Published by Elsevier Ltd.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
L. Gelain, L. van der Wielen, W.M. van Gulik et al. Chemical Engineering Science: X 8 (2020) 100085

the substrates and then the data were used to estimate the param- Zhang et al. (2009). The cellulose and mycelium concentrations
eters for the mathematical model. After that, the mathematical for the experiments using cellulose were determined according
model was adapted for fed-batch mode and used to simulate opti- to Ahamed and Vermette (2009). The enzymes are present in the
mal feeding policies for the production of cellulase. The simula- supernatant of the samples and can be separated from the cell
tions consider the consumption of glycerol for cell growth and and residual concentration of cellulose by centrifugation. More
the consumption of cellulose for cellulase production. The strate- details of these analytical procedures are described in Gelain
gies proposed allow maintaining a high cell concentration using et al. (2015).
a ‘‘neutral” substrate and the addition of optimal small amounts For glycerol assays, 10 mL of culture broth was withdrawn and
of the inducer substrate to prevent inhibition of enzyme centrifuged (3000x g for 20 min). The supernatant was used to
production. measure glycerol concentration and the pellet was dried (70 °C)
until constant weight for the determination of cell concentration.
Glycerol was measured using the column Aminex HPX-87H
2. Materials and methods
300x7.8 mm (BIO-RAD), a flow rate of 0.6 mL/min, isocratic condi-
tions, and H2SO4 as the eluent for 30 min. The equipment was the
2.1. Microorganism
Agilent 1260 Infinity with an infrared detector.
The wild strain Trichoderma harzianum P49P11 was used in this
study. The strain was isolated from the Amazon forest (Delabona 2.4. Mathematical methods
et al., 2012). It was grown on potato dextrose agar at 29 °C for
5 days and then used for inoculum preparation. Parameter estimation and simulations were performed using
Matlab R2013b. The differential equations were solved by the
2.2. Culture conditions ode8 function, the objective function was minimized by the fmin-
con function using the interior-point algorithm, and the interp1
The culture conditions were prepared according to Gelain et al. was used for interpolation. The simulations of the equations were
(2015). The spore suspension of T. harzianum was transferred to a 2 performed using Simulink (Matlab). The optimization of feeding
L shake flask containing per litre: glucose, 10 g; peptone, 1 g; strategies was performed according to Becerra (2004). The experi-
Tween 80, 1 mL; saline solution, 50 mL. After 60 h of cultivation ments with initial glycerol concentrations of 5, 10 and 20 g/L and
at 29 °C and 200 rpm in a shaker (New Brunswick Scientific inno- initial cellulose concentrations of 10, 20 and 30 g/L were employed
va44), 10% (v/v) was transferred to a 3 L bioreactor (New Bruns- for parameter estimation. The experiment with 15 g/L of glycerol
wick Scientific BioFlo 115) containing per litre: glycerol, 5, 10, was used for validation of the mathematical model using glycerol
15, or 20 g; or cellulose (Celufloc 200TM, Celuflok Ind. Com., Brazil), as the substrate and the experiments with 5 and 40 g/L of cellulose
5, 10, 20, 30 or 40 g; peptone, 1 g; Tween 80, 1 mL; saline solution, were used for extrapolation analysis of the mathematical model
50 mL. The solution of Mandels was used (Mandels and Reese, using cellulose as the substrate. The repeated batch was used to
1957), in g/L: KH2PO4, 20; (NH4)2SO4, 14; urea, 3; MgSO4
7H2O, test the prediction capacity of the model using cellulose as the sub-
3; CaCl2, 3; FeSO4
7H2O, 0.05; ZnSO4
7H2O, 0.014; MnSO4
H2O, strate. The objective function described in Andrade et al. (2013)
0.016; CoCl2, 0.02. Batch mode experiments were performed in was used for parameter estimation (Eq. (1)). The mathematical
duplicate with a working volume of 1.9 L. The inocula used for model and simulation platforms used in this project are available
the experiments using glycerol were prepared separately from in Gelain (2020).
the inocula used for cellulose conditions. Additionally, the dupli- " #
X
np
ðX n  X en Þ2 ðSn  Sen Þ2 ðPn  Pen Þ2
cates were carried out by using different inocula. For parameter f ¼ þ þ ð1Þ
estimation, an average of the cell concentration from the inocula n¼1 X em 2 Sem 2 Pem 2
was used that resulted in an initial cell concentration of
0.4 ± 0.02 g/L for the experiments using cellulose, and 0.5 ± 0.03 where, X en , Sen and Pen are the experimental data, X n , Sn and P n are
g/L for the experiments using glycerol. The temperature was con- the concentrations provided by the model, X em , Sem and Pem are
trolled at 29 °C and the pH was controlled at 5.0 ± 0.5 by the addi- the maximum measured concentrations of cell, substrate and pro-
tion of an aqueous solution of NH4OH (1:3) and 0.4 M H2SO4. The duct, respectively, and np is the number of samples.
stirring speed was kept between 200 and 300 rpm, and the airflow
between 0.48 and 0.7 vvm to prevent dissolved oxygen to drop 3. Results and discussion
below 30%. The value of 30% of dissolved oxygen was used as a
standard value for all experiments to guarantee an excess of dis- 3.1. Mathematical modelling
solved oxygen present in the growth medium. Furthermore,
1 mL/L of polypropylene glycol antifoaming agent (P2000, Dow The substrate consumption rate was considered dependent on
Chemical, Brazil) was added. One experiment was performed start- cell growth rate and proportional to mycelium concentration (C X )
ing in batch mode with 15 g/L of glycerol, and after 24 h, a mass of (Eqs. (2) and (3)). Eq. (2) describes the consumption of glycerol
cellulose was added resulting in 20 g/L inside the bioreactor. This (C G ) for cell growth and Eq. (3) describes the consumption of cellu-
experiment was named repeated batch. The media were sterilized lose (C C ) for cell growth and has also a second term that describes
at 121 °C for 30 min. the consumption of cellulose by active cells (C Xact ). Active cells are
an attempt at cell segregation, wherein this case, they are consid-
2.3. Analytical procedures ered as being the part of cells responsible for synthesizing
enzymes. This term considers that part of the cells is being acti-
The samples were taken at 8, 12, 24, 32, 48, 72 and 96 h for the vated due to the presence of an inducer (cellulose). These equa-
experiments in batch mode for the visualization of the profiles of tions consider a dilution rate (D), which represents the feeding of
cell, substrate and enzymes. Cellulase activity was determined glycerol. Eq. (3) has also a term corresponding to continuous feed-
using the filter paper activity assay (Ghose, 1987). Reducing sugars ing of cellulose through an inflow rate, u (g/h). It was considered
were measured by the DNS method (Miller, 1959). The method for that there was no dilution effect inside the bioreactor when
the estimation of beta-glucosidase activity was adapted from cellulose (solid material) was added. For the batch condition, the
2
L. Gelain, L. van der Wielen, W.M. van Gulik et al. Chemical Engineering Science: X 8 (2020) 100085

dilution rate (D) is equal to zero as well as the inflow rate of cellu- Table 2
lose (u). Parameters are displayed in Tables 1 and 2. Parameters for the mathematical model using cellulose as the substrate.

  lXmC Maximum specific cell growth rate (h1) 0.48

dC G dC X  
¼ a0 C X þ D C G;f  C G ð2Þ kC Monod constant for cell growth (g/L) 6
dt dt g
C X mC Maximum cell concentration (g/L) 12
    lXmdC Maximum specific rate for cell death (h1) 0.095
dC C dC X dC Xact u kCd Monod constant for cell death (g/L) 0.44
¼ a CX  b C Xact þ  DC C ð3Þ qFm Maximum specific rate for cellulase production (FPU/g (of 23.5
dt dt g dt g V
CXact) h)
CF m Maximum cellulase activity (FPU/L) 2 513
where, a0 is the constant of glycerol consumption, a is the constant
kiF Inhibition constant for cellulase production (g/L)2 1.91
of cellulose consumption for cells, b is the constant of cellulose con- kdF Deactivation constant for cellulase (h1) 0.002
sumption for active cells, C G;f is the glycerol concentration from the qBm Maximum specific rate for beta-glucosidase production (U/ 63.42
feed, and V is the volume. g (of CXact) h)
CBm Maximum beta-glucosidase activity (U/L) 5 013
Eq. (4) describes the cell growth rate depending on the sub-
kiB Inhibition constant for beta-glucosidase production (g/L)2 3.97
strate (C S = cellulose (C C ) or glycerol (C G )) according to the Monod kdB Deactivation constant for beta-glucosidase (h1) 0.0011
equation. The cell growth rate has an inhibition term dependent on lem Maximum specific growth rate for active cells (g (of CXact) / 0.25
cell concentration according to the logistic equation for population g h)
growth (Fujikawa et al., 2004). It was assumed that there was a kCe Monod constant for active cell growth (g/L) 2.84
C Xact m Maximum active cell concentration (g/L) 5.76
control of cell growth based on cell concentration where C X mS is
S iF Concentration of cellulose that inhibits cellulase 10
considered the maximum cell concentration allowed by the envi- production (g/L)
ronment. Eq. (5) represents the cell growth for the conditions using S iB Concentration of cellulose that inhibits beta-glucosidase 8
glycerol and Eq. (6) describes the cell growth for the conditions production (g/L)
using cellulose as the substrate. Eq. (6) does not consider the dilu- kda Deactivation constant for active cells (h1) 0.38
a Inhibition control for cellulase 0/0.15
tion rate because it was only used to describe cell growth in batch
b Inhibition control for beta-glucosidase 0/0.15
conditions. a Constant of cellulose consumption for cells (g (of CC) L/ g2 0.069
     (of CX))
dC X CS CX
¼ lXmS 1 C X  DC X ð4Þ b Constant of cellulose consumption for active cells (g (of CC) 0.21
dt g C S þ kS C X mS L/ g2 (of CXact))
u Inflow rate of cellulose (g/h)
    V Volume (L)
dC X  CG CX  
¼ l 1  C X  C X lXmdG þ D ð5Þ CX Cell concentration (g/L)
dt G XmG
C G þ kG C X mG C Xact Active cell concentration (g/L)
CC Cellulose concentration (g/L)
      CF Cellulase activity (FPU/L)
dC X  CC CX CC
 ¼ lXmC 1 C X  lXmdC CX ð6Þ CB Beta-glucosidase activity (U/L)
dt C C C þ kC C X mC C C þ kCd

where, lXmS , lXmG , lXmC are the maximum specific cell growth rates,
kS , kG , kC are the Monod constants for cell growth, C X mS , C X mG , C X mC
are the maximum cell concentrations, lXmdG , lXmdC are the maxi-     
dC Xact CC C Xact
mum specific rates for cell death and kCd is the Monod constant ¼ lem 1 C X  DC Xact ð7Þ
dt g C C þ kCe C Xact m
for cell death for cellulose conditions. The Monod constant for cell
death for glycerol conditions was considered zero by the parameter   
dC Xact CC C Xact
estimation. Subscripts S, C and G correspond to substrate, cellulose ¼ lem 1 C X  C Xact ðkda þ DÞ ð8Þ
and glycerol, respectively. dt C C þ kCe C Xact m
Velkovska et al. (1997) proposed that the cellulase production where, lem is the maximum specific growth rate for active cells, kCe
by Trichoderma reesei Rut C30 using cellulose was not associated is the Monod constant for active cell growth, C Xact m is the maximum
with cell growth. They developed a mathematical model with cell active cell concentration and kda is the deactivation constant for
segregation where first the formation of a primary mycelium active cells.
responsible for a high substrate consumption rate occurs, followed Cellulase (C F ) and beta-glucosidase (C B ) production rates have
by the formation of a secondary mycelium. This secondary myce- an inhibition term dependent on enzymatic activity, and a second,
lium was considered responsible for cellulase synthesis. This segre- dependent on cellulose concentration (C C ) (Equations (9) and (10),
gation was also considered in this project and the cells responsible respectively). The enzyme production rates are proportional to
for enzyme production were named active cells. Eq. (7) describes active cell concentration. Parameters a and b represent logical con-
the active cell growth rate and Eq. (8) includes the deactivation trols regarding the inhibition influence by cellulose according to
rate (kda ). another parameter, SiF and SiB , respectively, and when cellulose
concentration is above these values (SiF and SiB ), a and b are equal
Table 1 to 0.15, adding an inhibition effect on the enzyme production rate.
Parameters for the mathematical model using glycerol as the substrate. Otherwise, a and b are equal to zero. SiF and SiB are cellulose con-
centrations that inhibit the production of cellulase and beta-
lXmG Maximum specific cell growth rate (h1) 0.23
glucosidase, respectively. The values of a and b were manually
kG Monod constant for cell growth (g/L) 2
adjusted according to the residual value of the objective function
C X mG Maximum cell concentration (g/L) 25.8
lXmdG Maximum specific rate for cell death (h1) 0.04
(Eq. (1)). They were the only parameters kept constant during
a0 Constant of glycerol consumption (g (of CG) L/ g2 (of CX)) 0.38 parameter estimation. Values between 1 and 0 were tested.
D Dilution rate (h1) 0.0006   !
C G;f Glycerol concentration in the feed (g/L) 1 200 dC F CF 1
¼ qFm 1  C Xact  C F ðkdF þ DÞ ð9Þ
CX Cell concentration (g/L) dt CF m aC C 2 =kiF þ 1
CG Glycerol concentration (g/L)

3
L. Gelain, L. van der Wielen, W.M. van Gulik et al. Chemical Engineering Science: X 8 (2020) 100085

  !
dC B CB 1 The fit of the model proposed for cell growth followed the pro-
¼ qBm 1  2
C Xact  C B ðkdB þ DÞ ð10Þ files of the experimental data. Glycerol consumption was fast for
dt C Bm bC C =kiB þ 1
all the conditions and the model described the profiles based on
where, qFm is the maximum specific rate for cellulase production, the experimental data available. Nevertheless, more samples dur-
ing the variation period in glycerol concentration would have con-
C F m is the maximum cellulase activity, kiF is the inhibition constant
tributed to improving the fit and the visualization of the profiles of
for cellulase production, kdF is the deactivation constant for cellu-
glycerol consumption. The data from 15 g/L condition was not used
lase, qBm is the maximum specific rate for beta-glucosidase produc-
for parameter estimation and the simulation of this condition pre-
tion, C Bm is the maximum beta-glucosidase activity, kiB is the
dicted the profiles of cell and glycerol concentrations. The param-
inhibition constant for beta-glucosidase production and kdB is the
eter estimation considered the average values from the duplicates
deactivation constant for beta-glucosidase.
of the experimental data and the model does not include the influ-
Eq. (11) describes the variation in the volume (V).
ence of the experimental errors. Parameters are shown in Table 1.
dV
¼ DV ð11Þ
dt 3.3. Results for mathematical modelling using cellulose in batch mode

Assays with initial concentrations of cellulose of 10, 20 and

3.2. Results for mathematical modelling using glycerol in batch mode 30 g/L were used for parameter estimation. Eq. (3) was used for cel-
lulose consumption, Eq. (6) for cell growth, Eq. (8) for active cell
The experiments varying the initial concentrations of glycerol concentration and Equations (9) and (10) for enzyme production.
(5, 10 and 20 g/L) in the batch mode were employed for parameter The initial conditions for parameter estimation were 0.4 g/L of cell,
estimation for the growth stage. Eqs. (2) and (5) were used. The ini- 10, 20 and 30 g/L of cellulose and the enzymatic activities of cellu-
tial conditions for parameter estimation were 0.5 g/L of cell and 5, lase and beta-glucosidase were considered zero. Fig. 2A, 2B and 2C
10 and 20 g/L of glycerol. The experiment using 15 g/L of glycerol show the fit of the model for cell growth, substrate consumption
was used to validate the model. The results of the fit (continuous and cellulase production using 10, 20 and 30 g/L of cellulose,
lines) and the experimental data are shown in Fig. 1. Error bars cor- respectively, and Fig. 2D shows the fit of beta-glucosidase activity
respond to the sample standard deviation. for those three conditions. The fit of the model proposed for cell
growth, cellulose consumption and enzyme production followed
the profiles of the experimental data. Parameters are shown in
Table 2.
According to parameter estimation using cellulose, the maxi-
mum concentration of cells and active cells allowed in the bioreac-
tor are 12 g/L and 5.76 g/L, respectively. The maximum cellulase
and beta-glucosidase activities are 2 513 FPU/L and 5 013 U/L,
respectively. According to parameter estimation using glycerol
(Table 1), the maximum concentration of cells allowed in the
bioreactor is 25.8 g/L. The use of glycerol with cellulose could allow
the increase in the cell concentration compared to experiments
only using cellulose, creating the possibility of also increasing the
concentration of active cells, which could provide an increase in
enzyme synthesis.

3.4. Prediction capacity of the mathematical model

For extrapolation of the mathematical model using cellulose as

the substrate, the 5 and 40 g/L conditions were used. These exper-
iments were not included in parameter estimation. The simula-
tions are shown in Fig. 3A and 3B for cell growth, cellulose
consumption, and cellulase production. Fig. 3C shows the beta-
glucosidase simulation. Simulation for the 40 g/L condition indi-
cates a good fit for cell concentration, substrate consumption and
beta-glucosidase production. For the 5 g/L condition, the model
could only predict well the substrate consumption and beta-
glucosidase activity. Cellulase production was overestimated, but
follows the same profile of the experimental data.
A simulation was also performed to predict the production of
enzymes in the repeated batch, which started with glycerol for
the batch phase (15 g/L), followed by one feeding of cellulose after
24 h. This experiment was not included in parameter estimation.
The cellulose feeding resulted in a concentration of 20 g/L inside
the bioreactor. Only cellulose consumption was considered for
the simulation. Assuming that the cellulose consumed was mainly
for cellulase synthesis, Eq. (3) was used to represent the substrate
consumption rate considering the cellulose consumption for cell
Fig. 1. Fit of the model (continuous lines) for cell concentration (A) and glycerol
growth (aðdC X =dtÞg C X ) and u equal to zero. The results of the sim-
concentration (B). ulation are presented in Fig. 3D only after the feeding of cellulose.
4
L. Gelain, L. van der Wielen, W.M. van Gulik et al. Chemical Engineering Science: X 8 (2020) 100085

Fig. 2. Fit of the model for the assays using 10 (A), 20 (B) and 30 g/L of cellulose (C), ( ) cellulose, (j) cells, ( ) cellulase activity, ( ) active cell simulation. Fit of the model
for beta-glucosidase activity (D).

Fig. 3. Extrapolation of the mathematical model for the assays using 5 (A) and 40 g/L of cellulose (B) and simulation of beta-glucosidase activity (C). Fit of the model for the
repeated batch (D) starting with glycerol (15 g/L, data not presented) for cell growth and then cellulose for cellulase induction (20 g/L), ( ) cellulose, (j) cells, ( ) cellulase
activity, ( ) active cell simulation.

5
L. Gelain, L. van der Wielen, W.M. van Gulik et al. Chemical Engineering Science: X 8 (2020) 100085

The batch phase using glycerol is not presented. It can be seen that optimization of the manipulated variable every 8 h, not every inte-
the model could represent well the production of cellulase, cell gration step (integration step = 1 h).
growth and cellulose consumption. The first simulation (Fig. 4B, 4C and 4E) starts with 10 g/L of cel-
Fig. 3D, shows the induction of cellulase production after an ini- lulose, 1 g/L of glycerol and 0.4 g/L of cells. It considers a constant
tial concentration of glycerol. The induction stage started with 20 g/ dilution rate of 0.0006 h1 from a glycerol solution of 1 200 g/L to
L of cellulose and 6.6 g/L of cells. It can be observed that the cell con- keep the cell concentration close to 7 g/L. A discrete feed of cellu-
centration did not increase due to the presence of cellulose. There- lose was optimized to provide the maximum cellulase activity at
fore, cellulose presence only influenced the production of cellulase, 96 h. The cellulase activity at 96 h was the objective function to
and this influence was represented in the model by active cells. be maximized. Eq. (2) was used for glycerol consumption, Eq. (3)
for cellulose consumption (aðdC X =dtÞg C X and u ¼ 0), Eq. (5) for cell
growth, Eq. (8) for active cell growth and Eqs. (9) and (10) for
3.5. Simulation of strategies for cellulase maximization enzyme production. Glycerol consumption was considered for
the cell growth rate, and cellulose consumption was only consid-
The mathematical model was developed to simulate optimal ered for enzyme production. The maximum concentration of cellu-
feeding concentrations of the inducer substrate (cellulose) to pro- lose allowed inside the bioreactor was 10 g/L. The volume
vide the highest cellulase activity. The model was used to simulate increased from 0.9 to 0.95 L.
one condition where there is a continuous feeding of glycerol and a The second simulation (Fig. 4B, 4D and 4F) starts with 1 g/L of
discrete feeding of cellulose and one condition where there is a cellulose and follows the same considerations described earlier,
continuous addition of cellulose and glycerol from the feed. The but in this case, it considers a continuous feed of cellulose (u)
glycerol feeding is to keep the cell at the desired concentration, described by Eq. (3) (aðdC X =dtÞg C X ¼ 0). The cellulase activity at
and cellulose consumption is only for enzyme production. In these
47.5 h was the objective function to be maximized. The simulations
simulations, cellulose consumption does not influence cell growth
using the cellulase activity at 47.5 and 96 h as the objective func-
rate because glycerol was considered as the carbon source more
tions provide similar maximum cellulase activities, however, at
easily available for cell growth. Cellulose was only considered to
47.5 h, the simulation provides higher activity for beta-glucosidase.
be used to produce active cells.
Fig. 4A summarizes the possible interactions that the model
According to Ilmén et al. (1997), glycerol is considered a ‘‘neu-
provides to simulate strategies using glycerol for cell growth and
tral” substrate, meaning that it might not repress or induce cellu-
cellulose for active cell growth. The active cells are responsible
lase synthesis. Thus, it was considered that the presence of
for producing enzymes. Cellulose concentration influences the pro-
glycerol as an initial substrate concentration or as a continuous
duction of enzymes according to the inhibition parameters. The
inflow rate, would not interfere with the cellulase production rate
simulations of enzyme production are described in Fig. 4B. Accord-
and the parameters estimated in the batch will still be suitable for
ing to the simulation for the optimization of cellulase production,
fed-batch mode using cellulose as the inducer.
the discrete and continuous feeding generated the same enzymatic
For the optimization of cellulose feeding, first, it was considered
activity. Therefore, if a continuous feeding device is not available,
a discretization of the feed. For this purpose, in Simulink (Matlab),
discrete feeding could be an alternative to provide similar produc-
the integrator block had the ‘‘External reset: rising” option active,
tion of the target enzymes.
which enables the reset of the integrator when the feed increases
Fig. 4C and 4D present the cell, active cell and glycerol concen-
due to an addition of a mass of cellulose. In every feed, the integra-
trations for the discrete and continuous feeding strategies, respec-
tor block considers the cellulose inside the bioreactor plus the
tively. Cell growth provides the same profile since the cell
mass of cellulose added and uses this value as the initial substrate
concentration was considered dependent on the glycerol concen-
concentration to initiate the integration again. Eq. (3) was used to
tration for both strategies with the same dilution rate. The slight
optimize the cellulose feeding using the discretization approach
differences regarding the active cell concentration have resulted
considering the cellulose consumption for cell growth
from the cellulose feeding. Fig. 4E and 4F show cellulose concentra-
(aðdC X =dtÞg C X ) and u equal to zero.
tions and feeding concentrations. For the discrete feeding strategy
Another possibility of optimizing cellulose feeding is consider- (Fig. 4E), the bars represent the concentrations of cellulose inside
ing an inflow rate of cellulose. For this purpose, Eq. (3) was used the bioreactor considering the current volume. For the continuous
considering the cellulose consumption for cell growth equal to feeding strategy (Fig. 4F), the optimization algorithm provided an
zero. This equation allows the generation of an optimal feeding optimal profile of cellulose feeding (g/h).
profile considering u (g/h) as the manipulated variable. The strategies proposed here are examples to demonstrate
An algorithm described in Becerra (2004) was used for the opti- some possibilities that can be exploited in future works. If the
mization of the continuous feed of cellulose. The method considers experiments show that a continuous addition of glycerol prevents
the manipulated variable as parameters and performs parameter the consumption of cellulose and consequently the induction of
estimation using the Simulink (Matlab). The manipulated variable enzymes, then glycerol can only be used until the achievement of
is represented by a matrix containing two columns. The first col- the desired cell concentration. For this purpose, the dilution rate
umn is the time changing according to a fixed integration step can be calculated to provide the desired cell concentration at
and the second corresponds to parameters (manipulated variable) steady-state using Eqs.s (2) and (5). First, the dilution rate is iso-
that change according to an objective function. lated from Eq. (2) (Eq. (12)) and also from Eq. (5) (Eq. (13)). The
The discrete feeding was optimized following an adapted algo- desired cell concentration is assigned in both equations and a con-
rithm from Becerra (2004), where is this case, the optimization centration of glycerol should be found that satisfies the equality
occurs at one value of the manipulated variable at a time, keeping between D1 and D2 .
the remaining values static in the matrix. Once this first value of   
the manipulated variable is optimized based on the objective func- a0 lXmG CG CX
D1 ¼ h  i 1 CX 2 ð12Þ
tion, the algorithm moves to the next one, and after building up an C G;f  C G þ a0 C X 2 C G þ kG C X mG
optimal profile, the algorithm repeats the single optimization at a
time until satisfying a condition. This alteration was an attempt to   
CG CX
optimize a range of values of the manipulated variable instead of D2 ¼ lXmG 1  lXmdG ð13Þ
C G þ kG C X mG
optimizing all of them for every integration step. For example,
6
L. Gelain, L. van der Wielen, W.M. van Gulik et al. Chemical Engineering Science: X 8 (2020) 100085

Fig. 4. Diagram of the simulation strategy (A), cell growth depends on glycerol concentration, active cell growth depends on cell and cellulose concentrations, cellulase and
beta-glucosidase production depend on the active cell and cellulose concentrations. Simulated strategies using a continuous feed of glycerol with a discrete or continuous
feed of cellulose, cellulase and beta-glucosidase activities (B), cell, active cell and glycerol concentrations using a discrete feed of cellulose (C) and a continuous feed of
cellulose (D), cellulose concentration and a discrete feed of cellulose (concentration of the feed inside the bioreactor) (E) and a continuous feed of cellulose (F).

The advantages of the model designed in this project include batch experiments using 10 and 20 g/L of cellulose. Perhaps, the
the possibility of optimizing the feed of the inducer substrate using repeated batch strategy provoked inhibition of enzyme production.
a discrete or continuous approach. To date, it was not found in the In the repeated batch, there was only one feed corresponding to
literature works proposing mathematical models for the optimiza- 20 g/L of cellulose inside the bioreactor, but a better strategy could
tion of cellulase production using a discrete feeding. The inhibition be a small continuous or periodic addition of the inducer substrate
parameters of the enzyme production equations can easily be to prevent inhibition of enzyme production. This strategy was rep-
changed according to new experiments to describe strains less resented by the simulations using a continuous and discrete feed of
repressed. The feed of glycerol can be a constant value or follow cellulose, which indicates the obtaining of higher productivity of
a specific inflow rate dependent on other components such as cell cellulase (FPU/L h) than the batch and repeated batch analysed in
concentration and time. this project.
The cellulase production rate of 75 FPU/L h is considered a
3.6. Analysis of enzyme production desired productivity that can be used in industrial processes
(Himmel et al., 1999). Thus, the mathematical model developed
Table 3 shows a comparison of productivities between the in this work can be adapted for other microorganisms and inducer
batch conditions using 10 and 20 g/L of cellulose, the repeated substrates (such as sugarcane bagasse) aiming at the development
batch and the results of the simulations described earlier. The of new strategies to keep increasing the productivity of cellulase
repeated batch results indicated similar productivity than the until the achievement of the desired value.
7
L. Gelain, L. van der Wielen, W.M. van Gulik et al. Chemical Engineering Science: X 8 (2020) 100085

Table 3
Cellulase production rate (DC F =Dt) and beta-glucosidase production rate (DC B =Dt) from 0 to 72 h. Specific cellulase production rate (DC F =DtC X m ) and specific beta-glucosidase
production rate (DC B =DtC X m ) from 0 to 72 h considering the maximum cell concentration (C X m ).

Batch R. batcha Batch (model) Simulation

10 g/L 20 g/L 10 g/L 20 g/L Repeated batch Continuous feeding Discrete feeding
DC F =Dt(FPU/L h) 12.3 11.3 13.4 11.3 10.7 16.3 22.5 22.5
DC B =Dt(U/L h) 26.3 29.9 26.3 26.3 27.1 29.7 34.1 35.9
DC F =DtC X m (FPU/g h) 2.1 1.4 1.9 1.8 1.3 1.9 2.7 2.7
DC B =DtC X m (U/g h) 4.4 3.6 3.8 4.2 3.4 3.5 4 4.3
a
Repeated batch, starting with 15 g/L of glycerol and fed once with cellulose at 24 h (20 g/L).

4. Conclusions

A mathematical model for glycerol and cellulose conditions was

References
developed and presented a good fit for the majority of the experi-
mental data. Although extrapolation analysis indicated limitations, Ahamed, A., Vermette, P., 2009. Effect of culture medium composition on
the model can still predict the profiles of the experimental data Trichoderma reesei’s morphology and cellulase production. Bioresour. Technol.
using cellulose for conditions out of the range of concentrations 100 (23), 5979–5987. https://doi.org/10.1016/j.biortech.2009.02.070.
Andrade, R.R., Filho, F.M., Filho, R.M., Costa, A.C., 2013. Kinetics of ethanol
used for parameter estimation. Analysis of the repeated batch production from sugarcane bagasse enzymatic hydrolysate concentrated with
has indicated that the mathematical model can predict the profiles molasses under cell recycle. Bioresour. Technol. 130, 351–359. https://doi.org/
of experiments at first using glycerol for cell growth then cellulose 10.1016/j.biortech.2012.12.045.
Becerra, V., 2004. Solving optimal control problems with state constraints using
for enzyme production. Simulations of strategies were presented nonlinear programming and simulation tools. IEEE Trans. Edu. 47 (3), 377–384.
as possibilities that can be exploited in future works. The model https://doi.org/10.1109/TE.2004.825925.
and strategies were developed as tools to be used for cellulase Delabona, P.S., Farinas, C.S., da Silva, M.R., Azzoni, S.F., Pradella, J.G., 2012. Use of a
new Trichoderma harzianum strain isolated from the Amazon rainforest with
maximization and to be adapted for less repressed strains and
pretreated sugar cane bagasse for on-site cellulase production. Bioresour.
other substrates. Technol. 107, 517–521. https://doi.org/10.1016/j.biortech.2011.12.048.
Delabona, P.S., Lima, D.J., Robl, D., Rabelo, S.C., Farinas, C.S., Pradella, J.G., 2016.
Enhanced cellulase production by Trichoderma harzianum by cultivation on
CRediT authorship contribution statement glycerol followed by induction on cellulosic substrates. J. Ind. Microbiol.
Biotechnol. 43 (5), 617–626. https://doi.org/10.1007/s10295-016-1744-8.
Lucas Gelain: Software, Investigation, Writing - original draft, Fujikawa, H., Kai, A., Morozumi, S., 2004. A new logistic model for Escherichia coli
growth at constant and dynamic temperatures. Food Microbiol. 21 (5), 501–
Writing - review & editing. Luuk Wielen: Visualization, Conceptu- 509. https://doi.org/10.1016/j.fm.2004.01.007.
alization. Walter M. Gulik: Validation, Conceptualization. José Gelain, L., Pradella, J.G.C., Costa, A.C., 2015. Mathematical modeling of enzyme
Geraldo Cruz Pradella: Methodology, Resources, Conceptualiza- production using Trichoderma harzianum P49P11 and sugarcane bagasse as
carbon source. Bioresour. Technol. 198, 101–107. https://doi.org/10.1016/j.
tion. Aline Carvalho Costa: Supervision, Conceptualization, Writ-
biortech.2015.08.148.
ing - review & editing. Gelain, L., 2020. Mathematical modelling for the optimization of cellulase
production. Mendeley Data v2. https://doi.org/10.17632/shd3wcczsr.2.
Ghose, T.K., 1987. Measurement of cellulase activities. Pure Appl. Chem. 59, 257–
Declaration of Competing Interest
268. https://doi.org/10.1351/pac198759020257.
Himmel, M.E., Ruth, M.F., Wyman, C.E., 1999. Cellulase for commodity products
The authors declare that they have no known competing finan- from cellulosic biomass. Curr. Opin. Biotechnol. 10, 358–364. https://doi.org/
cial interests or personal relationships that could have appeared 10.1016/S0958-1669(99)80065-2.
Ilmén, M., Saloheimo, A., Onnela, M.L., Penttilä, M.E., 1997. Regulation of cellulase
to influence the work reported in this paper. gene expression in the filamentous fungus Trichoderma reesei. Appl. Environ.
Microbiol. 63, 1298–1306.
Acknowledgements Jayasekara, S., Ratnayake, R., 2019. Microbial Cellulases: An Overview and
Applications [Online First]. IntechOpen. https://doi.org/10.5772/
intechopen.84531.
This project was supported by the São Paulo Research Foundation Mandels, M., Reese, E.T., 1957. Induction of cellulase in Trichoderma viride as
(FAPESP), process number 2014/22537-9, the Brazilian National influenced by carbon sources and metals. J. Bacterial. 73 (2), 269–278. https://
doi.org/10.1128/JB.73.2.269-278.1957.
Council for Scientific and Technological Development (CNPq), pro- Miller, G.L., 1959. Use of dinitrosalicylic acid reagent for determination of reducing
cess number 142478/2014-8, Brazilian Biorenewables National sugar. Anal. Chem. 31 (3), 426–428. https://doi.org/10.1021/ac60147a030.
Laboratory (LNBR), the University of Campinas (UNICAMP) and Strakowska, J., Błaszczyk, L., Chełkowski, J., 2014. The significance of cellulolytic
enzymes produced by Trichoderma in opportunistic lifestyle of this fungus. J.
Delft University of Technology (TU Delft). The authors would like
Basic Microbiol. 54 (S1). https://doi.org/10.1002/jobm.201300821.
to thank Deise Juliana da Silva Lima for assistance in the prepara- Velkovska, S., Marten, M.R., Ollis, D.F., 1997. Kinetic model for batch cellulase
tion of the experiments. production by Trichoderma reesei RUT C30. J. Biotechnol. 54 (2), 83–94. https://
doi.org/10.1016/S0168-1656(97)01669-6.
Gelain, L., 2020. Mathematical modelling for the optimization of
Zhang, Y. H., Hong, J., Ye, X., 2009. Cellulase assays, in J. Mielenz (Ed) Biofuels.
cellulase production. Mendeley Data, v2, http://dx.doi.org/ Methods in Molecular Biology (Methods and Protocols), Humana Press, Totowa,
10.17632/shd3wcczsr.2. N. J., pp. 213-231. https://doi.org/10.1007/978-1-60761-214-8_14.