An efficient approach for lipase-catalyzed synthesis of retinyl laurate nutraceutical by combining ultrasound assistance and artificial neural network optimization

Article abstract of DOI:10.3390/molecules22111972

Although retinol is an important nutrient, retinol is highly sensitive to oxidation. At present, some ester forms of retinol are generally used in nutritional supplements because of its stability and bioavailability. However, such esters are commonly synthesized by chemical procedures which are harmful to the environment. Thus, this study utilized a green method using lipase as a catalyst with sonication assistance to produce a retinol derivative named retinyl laurate. Moreover, the process was optimized by an artificial neural network (ANN). First, a three-level-four-factor central composite design (CCD) was employed to design 27 experiments, which the highest relative conversion was 82.64%. Further, the optimal architecture of the CCD-employing ANN was developed, including the learning Levenberg-Marquardt algorithm, the transfer function (hyperbolic tangent), iterations (10,000), and the nodes of the hidden layer (6). The best performance of the ANN was evaluated by the root mean squared error (RMSE) and the coefficient of determination (R²) from predicting and observed data, which displayed a good data-fitting property. Finally, the process performed with optimal parameters actually obtained a relative conversion of 88.31% without long-term reactions, and the lipase showed great reusability for biosynthesis. Thus, this study utilizes green technology to efficiently produce retinyl laurate, and the bioprocess is well established by ANN-mediated modeling and optimization.

Full text of DOI:10.3390/molecules22111972

molecules
Article
An Efficient Approach for Lipase-Catalyzed Synthesis
of Retinyl Laurate Nutraceutical by Combining
Ultrasound Assistance and Artificial Neural
Network Optimization
Shang-Ming Huang 1, Hsin-Ju Li 2, Yung-Chuan Liu 2, Chia-Hung Kuo 3,
*
ID
and Chwen-Jen Shieh 1,
*
1
Biotechnology Center, National Chung Hsing University, 250 Kuokuang Road, Taichung 40227, Taiwan;
zxzxmj2323@hotmail.com
Department of Chemical Engineering, National Chung Hsing University, 250 Kuo-Kuang Road,
Taichung 40227, Taiwan; judyiaa@hotmail.com (H.-J.L.); ycliu@dragon.nchu.edu.tw (Y.-C.L.)
Department of Seafood Science, National Kaohsiung Marine University, 142 Haijhuan Road, Nanzih District,
Kaohsiung 81143, Taiwan
2
3
*
Correspondence: chkuo@webmail.nkmu.edu.tw (C.-H.K.); cjshieh@nchu.edu.tw (C.-J.S.);
Tel.: +886-7-361-7141 (ext. 3646) (C.-H.K.); +886-4-2284-0450 (ext. 5121) (C.-J.S.);
Fax: +886-7-364-0634 (C.-H.K.); +886-4-2286-1905 (C.-J.S.)
Received: 20 October 2017; Accepted: 13 November 2017; Published: 15 November 2017
Abstract: Although retinol is an important nutrient, retinol is highly sensitive to oxidation. At present,
some ester forms of retinol are generally used in nutritional supplements because of its stability and
bioavailability. However, such esters are commonly synthesized by chemical procedures which are
harmful to the environment. Thus, this study utilized a green method using lipase as a catalyst with
sonication assistance to produce a retinol derivative named retinyl laurate. Moreover, the process was
optimized by an artificial neural network (ANN). First, a three-level-four-factor central composite
design (CCD) was employed to design 27 experiments, which the highest relative conversion
was 82.64%. Further, the optimal architecture of the CCD-employing ANN was developed,
including the learning Levenberg-Marquardt algorithm, the transfer function (hyperbolic tangent),
iterations (10,000), and the nodes of the hidden layer (6). The best performance of the ANN
was evaluated by the root mean squared error (RMSE) and the coefficient of determination (R2)
from predicting and observed data, which displayed a good data-fitting property. Finally, the process
performed with optimal parameters actually obtained a relative conversion of 88.31% without
long-term reactions, and the lipase showed great reusability for biosynthesis. Thus, this study utilizes
green technology to efficiently produce retinyl laurate, and the bioprocess is well established by
ANN-mediated modeling and optimization.
Keywords: retinol; lipase; retinyl laurate; sonication; artificial neural network (ANN); central
composite design
1. Introduction
Retinol is an animal form of vitamin A that has many functions, such as the maintenance of
vision, UV filters, improvement of skin aging, promotion of bone growth, maintenance of reproductive
function, and can also strengthen the immune system to prevent infections [1]. Retinol and retinyl
esters are major compounds of vitamin A in the body, and retinol is unstable and prone to oxidation,
while the retinyl ester is not. When the retinol molecule is esterified with a long-chain fatty
acid, retinyl ester is formed. Major fatty acids within the body, such as palmitic acid, oleic acid,
Molecules 2017, 22, 1972; doi:10.3390/molecules22111972
www.mdpi.com/journal/molecules

Molecules 2017, 22, 1972
2 of 15
stearic acid, and linoleic acid are involved in the esterification reaction for the synthesis of retinyl
ester [ ]. Moreover, the retinyl esters with long chain fatty acid only form in the body of animals [ ].
2
2
Comparing with retinol obtained from foodstuffs, long-chain retinyl esters display highly lipophilic,
thermal stability, and multi-processable characteristics. Considering the importance of vitamin A in
human nutrition, it is also of interest to develop a convenient and economic technique to synthesize a
stable form of retinol derivatives, such as retinyl esters.
Retinyl esters can be synthesized by chemical methods or enzymatic catalysis.
In 2004, the bioactive retinyl ascorbates were synthesized by Abdulmajed and Heard using a
chemical technique, and this method requires several steps [3]. First, the retinyl chloride was
synthesized by chemical reaction of phosphorous trichloride with all-trans-retinoic acid in the
dry benzene. Thereafter, ascorbic acid was added to react with retinyl chloride in the presence of
potassium carbonate, and the mixture was subjected to thin-layer chromatography and column
chromatography to obtain retinyl ascorbates (purity of >98.5%). Moreover, retinyl retinoate
and many retinyl polyhydroxybenzoates have also been reported that they could be chemically
synthesized [4,5]. A study showed that retinyl retinoate could be synthesized by performing the
chemical reaction using retinol with retinoic acid in the solvent system containing a condensing
agent, such as N,N-carbonyl diimidazole (CDI), and a catalyst, such as N,N-dimethylaminopyridine
(DMAP) [
reaction of reactants [
4
]. Retinyl polyhydroxybenzoates can be chemically synthesized by the base-catalyzed
]. As mentioned above, chemical synthesis of retinyl esters is a complex
5
procedure, which is usually accompanied by increased production of byproducts, and the purification
of retinyl esters requires several steps, which are often time-consuming. Most notably, these procedures
are harmful to the environment because of using a large amount of organic solvent and alkaline
as catalysts.
In contrast to chemical-based methods, the biosynthesis of producing natural products has
been carried out in vitro by using purified enzymes, many of which have also been synthesized
by organic chemists and been applied in various research fields [6]. The biocatalysis seems to
be an attractive alternative to produce retinyl esters. Maugard et al. have used immobilized
Novozym®435 to catalyze retinol and L-methyl lactate to form retinyl L-lactate, and a high yield of
90% was obtained within 24 h [
using Novozym®435 to synthesize several water-soluble retinol derivatives, including retinyl adipate,
retinyl palmitate, and retinyl succinate [ ], there still remains some retinol derivatives, such as retinyl
7]. Although this research group also developed an enzymatic method
8
laurate, not synthesized by the biocatalytic methods. In such methods, enzymatic reaction products
can be easily purified by filtration of the biocatalyst and evaporation of the solvent under the condition
of reduced pressure, which is simpler and more environmentally friendly, as well as more economical,
than that using chemical methods as mentioned above.
Many reaction conditions, including pH, kind of solvents, the quantity of water, and the method of
enzyme preparation, have been suggested as the important factors on the efficiency of lipase-catalyzed
reactions [9]. Considering enzyme preparation, Yin et al. have compared the catalyzing efficiency of free
and immobilized lipases on retinyl palmitate synthesis, and the results showed that the immobilized
form was better than the free form of the activity of catalysis [10]. Additionally, immobilized lipases
have many advantages in the process of catalyzed reaction such as reusability and stability [11], so they
may be the best candidate to catalyze the synthesis of retinyl esters in the industry [7,8,12].
In recent years, ultrasound technique has gained increased attention because it is advantageous
for the acceleration of enzymatic reactions [13–17]. To the best of our knowledge, mechanical effects of
ultrasound can lead to a disruption of the phase boundary between compounds, an improvement of
mass transfer and a decrease in the activation energy of reactions, which could result in an accelerated
rate of enzyme-catalyzed reaction [14,16]. Therefore, ultrasound-assisted immobilized lipase catalysis
might be an excellent strategy for industrial biosynthesis. It has been reported that esterification
efficiency of immobilized lipase on the biosynthesis of phenethyl ester [13], 40-acetoxyresveratrol [14],

Molecules 2017, 22, 1972
3 of 15
ethyl butyrate [15] and D-isoascorbyl palmitate [18], can be enhanced by ultrasound assistance.
However, the effect of ultrasound on the lipase-catalyzed synthesis of retinyl esters is still unknown.
Design of experiments is a set of techniques that involve investigating the effect of different
variables on targeted outcomes in a controlled experiment. Response surface methodology (RSM) is one
of the well-established mathematical and statistical methods for designing and building experimental
models, evaluating the relative significance of each independent variable, as well as determining
the optimum operating condition for the predicted responses. Within the types of RSM design,
central composite design (CCD) consists of a factorial design with center points (the corners at +1 of the
cube) and increases “star” points to estimate curvature, which allows the evaluation of a second-order
polynomial equation [19]. Like CCD, Box-Behnken design (BBD) requires only three levels to run
an experiment. It is a special three-level design of each factor that does not contain any points at
the vertices of the experiment region. Now, RSM is a useful tool for understanding the interaction
among various parameters, which has been applied successfully for optimizing parameters in various
processes [20]. Recently, an artificial neural network (ANN) has been developed for modeling the
technological process as an alternative to RSM system. ANN can easily establish the nonlinear
relationship between independent and dependent variables without requiring prior knowledge of
the correlation between targeted responses [21], which has made ANNs a powerful tool having
higher accuracy and efficiency in the flexible fitting of experimental data, prediction, and modeling of
biochemical processes, when compared to other data collection methods, such as RSM [22,23].
Taken together, retinol is a pivotal nutrient that is an important participant in various biochemical
reactions; however, retinol is unstable and highly sensitive to oxidization by oxygen molecules.
At present, some ester forms of retinol are generally used in nutritional supplements because of its
bioavailability and stability. However, such retinyl esters are commonly synthesized by chemical
procedures using a large amount of organic solvents and alkaline catalysts, easily leading to the
formation of byproducts which are harmful to the environment. In this study, the aim was to develop an
efficient green process with ultrasound assistance for the synthesis of retinyl laurate (stable long-chain
retinyl esters), using the immobilized lipase as a biocatalyst to reduce the need for chemical solvents.
In addition, we have focused especially on the optimization of reaction conditions, including reaction
time, reaction temperature, enzyme concentration, and the molar ratio of reactants, modeled by CCD
and an ANN, a well-known modeling/optimizing methodology for non-linear multivariate processes.
2. Results and Discussion
2.1. Effect of Ultrasound
Ultrasonic assistance is increasingly becoming a quite useful method in chemical
engineering-related jobs, which utilizes the energy, enhancing heat and mass transfer, for the various
purposes of sample preparation. For example, Kumar et al. [24] and Yu et al. [25] investigated the
performance of ultrasound on lipase-catalyzed production of fatty acid methyl esters (FAME) from
soybean, which the optimum condition was finally established to obtain a high FAME yield without
long-term reaction time, suggesting that ultrasonic assistance is a faster and efficient method for the
production of biodiesel. Additionally, Vishwanath et al. indicated that the use of ultrasonic irradiations
was in both enhancing the rate of reaction, as well as in shifting the equilibrium and resulting in
a higher product yield of isopropyl esters from palm fatty acid [26]. Moreover, in our experience,
the ultrasound could increase the efficiency of processes such as extraction of naturally-occurring
materials, ester biosynthesis, and nutraceutical encapsulation. Thus, in this study, the catalyzing
efficiency of lipase on retinyl laurate synthesis was initially examined under ultrasonic or traditional
shaking (without ultrasound) conditions (Figure 1).

Molecules 2017, 22, 1972
4 of 15
Figure 1. The reaction diagram of ultrasound-assisted retinyl laurate synthesis catalyzed by
Novozym®435.
2.2. Preliminary Test
Initially, quantification of the ultrasound in enhancing the synthesis process was carried out
respectively by experiments in the presence and absence of ultrasonic assistance. Our results showed
that a higher relative conversion of retinyl laurate was obtained in the ultrasonic treatment group than
that in traditional shaking group without sonication (ultrasonic power of 0 W) (Figure 2), suggesting
that ultrasound assistance could significantly enhance the efficiency of the lipase-catalyzed reaction.
In addition, an increase in relative conversion of retinyl laurate was observed when the ultrasonic
power was increased in our experiment. In accordance with our findings, previous reports have
suggested that ultrasound could improve immobilized lipase-catalyzed ethyl butyrate synthesis [15],
and also promoted an acceleration of lipase-catalyzed acetylation [14]. In this preliminary test
(under the condition of 1:10 molar ratio), although there was no significant difference between the
relative conversion of retinyl laurate and ultrasonic treatment group (90–150 W), the relative conversion
of retinyl laurate (50% conversion; data not shown) by ultrasonic power of 120 W is still better than
that using ultrasonic power of 90 W (35% conversion; data not shown) under the condition of molar
ratio of 1:1. Similar to our study, Batistella et al. [27] indicated that a high yield (~90% conversion)
of lipase-catalyzed transesterification of soybean oil could be obtained at the mild irradiation power
◦
supply (~100 W), temperature (60 C), and short reaction time (4 h). Moreover, some studies indicated
that ultrasonic power could not be necessary, as a parameter, for the optimization of ultrasound-assisted
enzymatic synthesis [24–28]. Thus, we considered that the following experiments should only be
conducted by treatment with the ultrasonic power of 120 W for further process optimization.
Figure 2. Effects of ultrasonic treatment on the lipase-catalyzed synthesis of retinyl laurate.
The lipase-catalyzed synthesis of retinyl laurate was conducted under traditional shaking of 100 rpm
(ultrasonic power of 0 W) or indicated intensities of ultrasonic treatment (90–150 W).
It has been reported that many reaction conditions could control the efficiency of enzymatic
synthesis [
amount might have been shown to be the crucial reaction parameters in retinyl ester synthesis catalyzed
by immobilized Candida sp. lipase [ 10 12]. Therefore, the present study further investigated the
9,29]. Notably, the molar ratio of substrates, reaction temperature, reaction time, and enzyme
7
, ,

Molecules 2017, 22, 1972
5 of 15
effects of these key factors on Novozym®435-catalyzed synthesis of retinyl laurate with sonication
assistance. The effect of reaction time on Novozym®435-catalyzed biosynthesis of retinyl laurate
under the experimental conditions of indicated molar ratios of substrates, a reaction temperature
of 40 ◦C, enzyme amount of 50 mg and ultrasonic power of 120 W, which was shown in Figure 3A.
The relative conversion of retinyl laurate was increased to 83% at 5 h when the molar ratio of retinyl
acetate to lauric acid was set as 1:10. In such condition, the relative conversion of retinyl laurate
was nearly constant when the reaction time was over 6 h. The effect of reaction temperature on
Novozym®435-catalyzed biosynthesis of retinyl laurate under the experimental conditions of substrate
molar ratio of (1:10), the reaction time of 3 h, enzyme amount of 50 mg and ultrasonic power of 120 W,
which was shown in Figure 3B. According to the data, the relative conversion of retinyl laurate did not
◦
◦
significantly vary at different reaction temperatures from 40 C to 60 C. Moreover, one study indicated
that reaction temperature above 50 ◦C could also cause inactivation of the lipase [30], which might
attenuate reaction rate. Therefore, a constant relative conversion of retinyl laurate was observed in
this experiment. Figure 3C indicates the effect of catalytic amount of Novozym®435 on biosynthesis
of retinyl laurate under the experi◦mental conditions of substrate molar ratio (1:10), reaction time
of 3 h, reaction temperature of 40 C and ultrasonic power of 120 W. The results showed that the
relative conversion of retinyl laurate might be positively related to the amount of enzyme and the
substrate molar ratio within reaction conditions, and the relative conversion reached the highest levels
(about 80%) when the enzyme amount was used over 50 mg. As compared to other research [10],
our present study indicated that Novozym®435-catalyzed biosynthesis of retinyl laurate by sonication
assistance could obtain a higher conversion without long-term reactions. Moreover, in order to model
the ultrasound-assisted lipase-catalyzed retinyl laurate synthesis, reaction temperature (40–60 ◦C),
enzyme amount (10–50 mg), reaction time (2–6 h) and molar ratio of retinyl acetate to lauric acid
(1:1 to 1:10) were further employed in the three-level-four-factor CCD, and the optimum reaction
conditions for retinyl laurate synthesis were finally established by ANN operation.
Figure 3. Individual effects of selected experimental parameters on relative conversion of retinyl
laurate. (
(Novozyme®435) (
acetate (mmole) to lauric acid (mmole).
A
–
C
) Effects of reaction time and the molar ratio (
A); temperature (B); and enzyme amount
C
) on relative conversion of retinyl laurate. Molar ratio means the ratio of retinyl

Molecules 2017, 22, 1972
6 of 15
2.3. Artificial Neural Network (ANN)
ANN is generally known as a system with the ability to map a set of input variables into a set
of outcomes without knowing the complex relationship between these parameters [31]. It has been
successfully applied to investigate the possible interactions of reaction parameters and to optimize
various valuable ester synthesis by lipase [32]. In this study, the data generated from the CCD
(Table 1) were employed in an ANN model for the optimization of enzymatic synthesis of retinyl
laurate. The various architectures for ANN models were developed using Neural Power software
(CPC-X Software, version 2.5, Carnegie, PA, USA), and the best set of parameters was determined
based on the maximization of R2 value and minimization of the RMSE value [33]. The proposed
ANN consisted of three layers in the present work, including an input layer with four neurons
(reaction time, reaction temperature, enzyme amount, and substrate molar ratio), a hidden layer
with several neurons, and an output layer containing one output neuron (relative conversion).
According to Figure 4, the hidden layer of ANN with six neurons can generate the best performance
of the network, which exhibited lowest and stable RMSE levels and highest and stable R2 levels.
Therefore, a 4-6-1 topology of the ANN was developed (Figure 5). The transfer function (Figure 4A,B),
the iteration (Figure 4C,D) and the learning algorithm (Figure 4E,F) of the proposed ANN were also
determined statistically based on the R2 and RMSE values. Finally, the best ANN model with a
4-6-1 topology in this study was determined to be a multilayer feed-forward connection trained
by Levenberg-Marquardt (LM) algorithm using the hyperbolic tangent (Tanh) transfer function,
and the iteration was set to be 10,000 to avoid overtraining and to decrease training time in this
model. The learning process in this study was acquired with the values of RMSE = 0.223 and
R2 = 0.999 (Figure 6).
Table 1. Central composite design (CCD) experiments and observed data.
Experimental Parameters
a
Relative Conversion (%) b
Treatment No.
X1
X2
X3
X4
1
2
3
4
5
6
7
8
9
2
4
4
4
4
6
2
6
4
4
4
4
2
2
2
2
4
6
6
6
6
4
4
4
4
4
4
40
40
40
40
40
40
60
60
60
60
60
60
50
50
50
50
50
50
50
50
50
50
50
50
50
50
50
30
10
50
30
30
30
30
30
10
30
50
30
10
30
50
30
50
30
30
10
50
30
50
30
10
30
10
3
3
3
1
5
3
3
3
3
5
3
1
3
5
3
1
1
5
1
3
3
3
5
3
1
3
5
33.16 ± 1.21
20.97 ± 1.17
61.99 ± 3.33
30.27 ± 2.67
56.69 ± 3.56
54.56 ± 1.93
47.59 ± 1.14
73.19 ± 0.74
50.55 ± 4.50
79.36 ± 0.94
80.85 ± 0.57
42.87 ± 1.41
21.04±1.50
53.10 ± 1.83
59.30 ± 4.06
28.63 ± 1.16
47.27 ± 0.42
77.47 ± 2.49
40.33±1.72
44.71 ± 1.32
78.69 ± 0.31
61.06±1.82
82.64 ± 1.94
59.80 ± 2.46
14.52 ± 1.85
61.30 ± 1.27
36.35 ± 3.77
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
a
The treatments were run in a random order; b Data are expressed as mean ± SD (n = 3).

Molecules 2017, 22, 1972
7 of 15
Figure 4. Effects of the transfer function, iteration, learning algorithm and number of hidden
neurons on the performance of ANN. ( ) Effects of sigmoid or hyperbolic tangent (Tanh)
transfer functions employed in training sets on the performance of ANN; ( ) Effect of
iterations (100–100,000) on the performance of ANN; ( ) Effects of learning algorithm and
A,B
C,D
E,F
number of hidden neurons on the performance of ANN. IBP, increment back propagation;
BBP, batch backpropagation; QP, quick propagation; LM, Levenberg-Marquardt algorithm; GA, genetic
algorithm. ANN performance was evaluated by the coefficient of determination (R2) and the root mean
square error (RMSE).

Molecules 2017, 22, 1972
8 of 15
Figure 5. The optimal architecture of multilayer feed-forward neural network. The network contains
four inputs (reaction temperature, reaction time, enzyme amount, and molar ratio of retinyl acetate to
lauric acid), one hidden layer with six nodes and one output (relative conversion of retinyl laurate).
Figure 6. The performance of constructed ANN on data fitting. The ANN performance was evaluated
by the coefficient of determination (R2) and root mean square error (RMSE), respectively. Residual
values: ANN-predicted values minus experimental values.
According to Figure 7A, both enzyme amount and molar ratio exhibited great effects on the
relative conversion with a relative importance around 36% and 29%, respectively. These results are
consistent with the response surface plot shown in Figure 7B. In accordance with previous reports,
the enzyme amount is also determined to be the most important factor in Novozym®435-catalyzed
reaction [9,29].

Molecules 2017, 22, 1972
9 of 15
(A)
(B)
Figure 7. Influence of experimental variables on the lipase-catalyzed synthesis of retinyl laurate.
) Importance of experimental variables in retinyl laurate synthesis. The analysis was done by Neural
Power software (CPC-X Software, USA); and ( ) the response surface plot of the ANN showing the
(A
B
relationships between the relative conversion of retinyl laurate and reaction parameters (molar ratio
and enzyme amount).
In order to validate and test the predicted ability of ANN model, a completely new set of three
experiments was carried out within the experimental range, which did not belong to the datasets
of the design. The experimental and predicted values of the responses for ANN model are given in
Table 2. The performance of newly-constructed ANN model was statistically measured based on the
values of R2 and RMSE. The RMSE value for ANN was found to be 3.380, and the R2 value was found
to be 0.992. From the results, it was observed that the agreement of predictive ability of ANN and
experiment was found to be great. The previous report also indicated that ANN displayed a great
data-predicting ability in the lipase-catalyzed synthesis of palm-based wax ester [34]. Additionally,
the result showed that R2 value calculated between the actual and ANN-predicting responses were
equal to 1 [34], suggesting that the ANN is really a good tool to model and optimize the process of
enzymatic synthesis.

Molecules 2017, 22, 1972
10 of 15
Table 2. Validation of ANN-modeling retinyl laurate synthesis.
Independent Variable Relative Conversion (%)
Run
X1
X2
X3
X4
Experimental a
ANN-Predicted
1
2
3
4.5
2.25
3.25
46
54
54
20
40
15
4.5
1.5
4.5
60.98 ± 3.55
50.37 ± 2.29
54.91 ± 3.97
R2
58.75
46.70
50.93
0.992
3.380
RMSE
a
Data are expressed as mean ± SD (n = 3).
As shown in Table 3, the optimization for the lipase-catalyzed synthesis of retinyl laurate
determined by Neural Power software ◦was operated by the following conditions of the reaction
time of 4.4 h, a reaction temperature of 58 C, enzyme amount of 50 mg and substrate molar ratio of 1:5.
Finally, the response performed within the optimum experimental conditions could actually obtain a
relative conversion of 88.31% related to the ANN-predicted value of 84.8%, indicating that the ANN
model could effectively help predict and optimize the process/parameters of ultrasound-assisted
lipase biosynthesis by Novozym®435 to obtain a high yield of retinyl laurate.
Table 3. An optimal trial obtained from the ANN model.
Optimal Condition
Relative Conversion (%)
Experimental a
ANN-Predicted
88.31 ± 0.30 84.80
X1
X2
X3
X4
4.4
58
50
5
a
Data are expressed as mean ± SD (n = 3).
2.4. Enzyme Reusability
The enzyme reusability has been a matter of great concern when using immobilized lipases
as practical catalysts in industry. Therefore, the enzyme reusability of immobilized lipase for
biosynthesis was also investigated under the optimized experimental process by determining the
relative conversion of retinyl laurate (Figure 8). The immobilized lipase was recovered from the
medium after the experimental reaction, followed by reusing the lipase in the next batch [33]. After five
times of reuse for immobilized lipase used in this study, more than 80% of the relative conversion
of retinyl laurate was still obtained. Therefore, the catalytic activity did not decrease markedly after
reusing five times, indicating that an immobilized lipase, such as Novozym®435, could be effectively
applied for retinyl laurate synthesis. According to numerous studies [35–37] and our experience,
we determine that the lipase activity for reuse might depend on some factors, such as the solubility of
the ester, reactants, usage of solvents, reaction temperature/time, and the process/method of reuse.
For example, the esterification of 2-ethyl hexanol showed that after 10 cycles the enzyme retained
90% of its activity, however, the system consisting of ascorbic acid, palmitic acid, Novozym®435 and
tert-butanol showed that a reduction in enzyme activity was accompanied by a reduction in reaction
conversion [38]. Additionally, when we work with the immobilized lipases as catalysts for acidolysis or
esterification reactions, we often wash the used enzyme as soon as possible to minimize the decrease in
the enzyme activity, and the enzyme after use is washed with a suitable buffer/solvent of the reaction
and the washed mother liquor is checked for residual substrate or product, requiring the number of
washings to be optimized.

Molecules 2017, 22, 1972
11 of 15
Figure 8. Reusability of Novozyme®435 on the synthesis of retinyl laurate. Reusability of
Novozyme®435 was evalua◦ted under optimized experimental conditions (reaction time of 4.4 h,
a reaction temperature of 58 C, an enzyme amount of 50 mg, and a molar ratio of 1:5).
3. Materials and Methods
3.1. Materials
Immobilized lipase Novozym®435 was purchased from Novo Nordisk Bioindustrials Inc.
(Copenhagen, Denmark). Retinyl acetate ( 99%) and lauric acid (99%) were purchased from
≥
Sigma-Aldrich (St. Louis, MO, USA). Other chemicals and reagents used were of analytical
reagent grade.
3.2. Lipase-Catalyzed Retinyl Laurate Synthesis by Ultrasound Assistance
All reagents were dehydrated by molecular sieves (4 Å) for 24 h before use. Briefly, Novozym®435
was added into 1 mL hexane in a glass tube. Then, retinyl acetate and lauric acid were added to
the mixture. The glass tube was sealed and placed in a temperature-controlled 40 kHz ultrasonic
bath (Delta DC150H, New Taipei, Taiwan). Thereafter, the lipase-catalyzed retinyl laurate synthesis
was carried out under various conditions of ultrasonic power (0–150 W), reaction time (2–6 h),
◦
reaction temperature (40–60 C), enzyme amount (10–50 mg), and molar ratio of retinyl acetate
to lauric acid (1:1 to 1:10).
At the end of the reaction, the liquid samples from reaction mixture were further withdrawn for
determination of retinyl laurate by an instrument using high-performance liquid chromatography
(HPLC). First, the sample was diluted and injected (20
µL) into the HPLC equipped with an ultraviolet
(UV) detector (Hitachi L-7400; Tokyo, Japan) and a Thermo C18 column (250 mm
×
4.6 mm, Agilent,
Waltham, MA, USA). In gradient elution, all separations were car−ri1ed out with a mobile phase of 0.1%
acetic acid and methanol, and the flow rate was set at 1.0 mL
·
min
[29]. Moreover, the retinyl laurate
was absolutely detected under the condition of long-wavelength UV light at 364 nm as illustrated in
Figure 9. The relative conversion was defined as (mmol of retinyl laurate production per mmol of
initial retinyl acetate) × 100%.

Molecules 2017, 22, 1972
12 of 15
Figure 9. HPLC chromatogram of retinyl laurate. (A,B) show the chromatograms obtained before and
after the lipase-catalyzed reaction, respectively.
3.3. Central Composite Design
A three-level-four-factor CCD for 27 experimental runs was employed in the present study.
To avoid bias, the 27 runs were performed in a totally random order. The variables and their
◦
levels selected for the study of retinyl laurate biosynthesis were: reaction temperature (40–60 C),
enzyme amount (10–50 mg), reaction time (2–6 h), and the molar ratio of retinyl acetate to lauric acid,
which were coded as shown in Table 4. Moreover, Table 1 shows the independent factors (Xi) and
their levels, experimental design, and the observed data. Each experimental point was carried out in
triplicate. Statistical parameters [33] performed in this study including coefficient of determination
(R2) and root mean square error (RMSE) were respectively calculated by Equations (1) and (2):
ꢀ
ꢁ
2
n
Ypre − Yexp
∑
i = 1
R2 = 1 −
(1)
ꢀ
ꢁ
2
n
Ym − Yexp
∑
i = 1
where Ypre is the predicted yield of retinyl laurate (by ANN), Yexp is the observed yield of retinyl
laurate, and Ym is the average yield of retinyl laurate; n is the number of experiments (n = 27 for CCD
test and n = 3 for validation test).
s
ꢀ
ꢁ
2
n
Ypre − Yexp
∑
i = 1
RMSE =
(2)
n
where Ypre and Yexp are the predicted and experimental yields of retinyl laurate, respectively; and n is
the number of experiments (n = 27 for CCD test and n = 3 for validation test).

Molecules 2017, 22, 1972
13 of 15
Table 4. Coding of experimental parameters and related levels.
Coded Levels
Parameters
Reaction time (h)
Reaction temperature ( C)
Enzyme amount (mg)
Molar ratio a
Symbol
−1
0
1
X1
X2
X3
X4
2
40
10
1
4
50
30
3
6
60
50
5
◦
a
Molar ratio means the ratio of retinyl acetate (mmole) to lauric acid (mmole).
4. Conclusions
The present study demonstrated, for the first time, an ultrasonic system efficiently assisting with
the synthesis of retinyl laurate catalyzed by immobilized lipase (Novozym®435). Moreover, a CCD with
an ANN was operated to investigate the optimization of enzymatic synthesis. Finally, the ANN models
for retinyl laurate synthesis was built, and the ANN showed good data fitting and prediction abilities.
The optimal condition for ultrasound-assisted synthesis was performed by a reaction time of 4.4 h,
◦
a reaction temperature of 58 C, an enzyme amount of 50 mg, and a substrate molar ratio of 1:5.
Finally, the relative conversion of retinyl laurate of 88.31% could be actually obtained under these
optimized conditions. This study successfully performed an ultrasound-assisted process by combining
ANN optimization for the synthesis of retinyl laurate biocatalysis using the lipase as a catalyst to
reduce the use of chemical solvents. Moreover, as compared to the biosynthetic process, our study
could obtain a higher relative conversion of retinyl laurate in a shorter time. Our current evidence
provides useful information for implications: (1) ultrasound-assisted biocatalysis can be an efficient
strategy for producing retinyl laurate; and (2) the methodological approach using an ANN showed a
great ability for predicting and optimizing the biosynthesis. Finally, it is worth emphasizing that a
nutraceutical like retinol is sensitive to oxygen, which might lead to the limited practical application of
chemoprevention; however, this study provides important information that the sonication-assisted
biosynthesis optimized by the ANN methodology is an efficient strategy for producing targeted ester
structures of nutraceuticals with enhanced stability against oxidation by air, which could be well used
as supplements to promote human health.
Acknowledgments: The authors are very grateful to the Ministry of Science and Technology of Taiwan, ROC for
supporting this research (Grants No. 104-2221-E-005-061-MY3 and 103-2221-E-005-087).
Author Contributions: C.-J.S. and Y.-C.L. conceived and designed the experiments; H.-J.L. and S.-M.H. performed
the experiments; S.-M.H. and C.-H.K. analyzed the data; S.-M.H., C.-H.K., and C.-J.S. wrote the paper.
Conflicts of Interest: There is no any conflict of interest regarding the publication of this article.
References
1.
2.
3.
4.
Hammerling, U. Retinol as electron carrier in redox signaling, a new frontier in vitamin A research.
Hepatobiliary Surg. Nutr. 2016, 5, 15–28. [PubMed]
O’Byrne, S.M.; Blaner, W.S. Retinol and retinyl esters: Biochemistry and physiology. J. Lipid Res. 2013, 54,
1731–1743. [CrossRef] [PubMed]
Abdulmajed, K.; Heard, C.M. Topical delivery of retinyl ascorbate co-drug. 1. Synthesis, penetration into
and permeation across human skin. Int. J. Pharm. 2004, 280, 113–124. [CrossRef] [PubMed]
Kim, H.; Kim, B.; Kim, H.; Um, S.; Lee, J.; Ryoo, H.; Jung, H. Synthesis and in vitro biological activity of
retinyl retinoate, a novel hybrid retinoid derivative. Bioorg. Med. Chem. 2008, 16, 6387–6393. [CrossRef]
[PubMed]
5.
Kim, S.; Kim, Y.; Kong, Y.; Kim, H.; Kang, J. Synthesis and in vitro biological activity of retinyl
polyhydroxybenzoates, novel hybrid retinoid derivatives. Bioorg. Med. Chem. Lett. 2009, 19, 508–512.
[CrossRef] [PubMed]

Molecules 2017, 22, 1972
14 of 15
6.
7.
8.
9.
Roberts, A.A.; Ryan, K.S.; Moore, B.S.; Gulder, T.A. Total (bio)synthesis: Strategies of nature and of chemists.
Top. Curr. Chem. 2010, 297, 149–203. [PubMed]
Maugard, T.; Tudella, J.; Legoy, M.D. Study of vitamin ester synthesis by lipase-catalyzed transesterification
in organic media. Biotechnol. Prog. 2000, 16, 358–362. [CrossRef] [PubMed]
Maugard, T.; Rejasse, B.; Legoy, M.D. Synthesis of water-soluble retinol derivatives by enzymatic method.
Biotechnol. Prog. 2002, 18, 424–428. [CrossRef] [PubMed]
Adlercreutz, P. Immobilisation and application of lipases in organic media. Chem. Soc. Rev. 2013, 42,
6406–6436. [CrossRef] [PubMed]
10. Yin, C.H.; Liu, T.; Tan, T.W. Synthesis of vitamin A esters by immobilized Candida sp. lipase in organic media.
Chin. J. Chem. Eng. 2006, 14, 81–86. [CrossRef]
11. DiCosimo, R.; McAuliffe, J.; Poulose, A.J.; Bohlmann, G. Industrial use of immobilized enzymes.
Chem. Soc. Rev. 2013, 42, 6437–6474. [CrossRef] [PubMed]
12. Zhu, K.; Wang, J.Q.; Wang, Y.-H.; Liu, H.; Han, P.-F.; Wei, P. Synthesis of retinyl palmitate catalyzed by
Candida sp. 99–125 lipase immobilized on fiber-like SBA-15 mesoporous material. J. Nanosci. Nanotechnol.
2011, 11, 7593–7602.
13. Chen, H.C.; Chen, J.H.; Chang, C.; Shieh, C.J. Optimization of ultrasound-accelerated synthesis of enzymatic
caffeic acid phenethyl ester by response surface methodology. Ultrason. Sonochem. 2011, 18, 455–459.
[CrossRef] [PubMed]
14. Kuo, C.H.; Hsiao, F.W.; Chen, J.H.; Hsieh, C.W.; Liu, Y.C.; Shieh, C.J. Kinetic aspects of ultrasound-accelerated
lipase catalyzed acetylation and optimal synthesis of 4’-acetoxyresveratrol. Ultrason. Sonochem. 2013, 20,
546–552. [CrossRef] [PubMed]
15. Paludo, N.; Alves, J.S.; Altmann, C.; Ayub, M.A.; Fernandez-Lafuente, R.; Rodrigues, R.C. The combined
use of ultrasound and molecular sieves improves the synthesis of ethyl butyrate catalyzed by immobilized
Thermomyces lanuginosus lipase. Ultrason. Sonochem. 2015, 22, 89–94. [CrossRef] [PubMed]
16. Prajapat, A.L.; Subhedar, P.B.; Gogate, P.R. Ultrasound assisted enzymatic depolymerization of aqueous guar
gum solution. Ultrason. Sonochem. 2016, 29, 84–92. [CrossRef] [PubMed]
17. Sutar, R.S.; Rathod, V.K. Ultrasound assisted enzyme catalyzed degradation of Cetirizine dihydrochloride.
Ultrason. Sonochem. 2015, 24, 80–86. [CrossRef] [PubMed]
18. Cui, F.J.; Zhao, H.X.; Sun, W.J.; Wei, Z.; Yu, S.L.; Zhou, Q.; Dong, Y. Ultrasound-assisted lipase-catalyzed
synthesis of D-isoascorbyl palmitate: Process optimization and Kinetic evaluation. Chem. Cent. J. 2013, 7,
180–190. [CrossRef] [PubMed]
19. Allen, T.T.; Bernshteyn, M.A.; Kabiri-Bamoradian, K. Constructing meta-models for computer experiments.
J. Qual. Technol. 2003, 35, 264–274.
20. Wang, F.; Chen, L.; Jiang, S.; He, J.; Zhang, X.; Peng, J.; Xu, Q.; Li, R. Optimization of methazolamide-loaded
solid lipid nanoparticles for ophthalmic delivery using Box-Behnken design. J. Liposome Res. 2014, 24,
171–181. [CrossRef] [PubMed]
21. Sha, W.; Edwards, K.L. The use of artificial neural networks in materials science based research. Mater. Des.
2007, 28, 1747–1752. [CrossRef]
22. Maran, J.P.; Priya, B. Comparison of response surface methodology and artificial neural network approach
towards efficient ultrasound-assisted biodiesel production from muskmelon oil. Ultrason. Sonochem. 2015
23, 192–200. [CrossRef] [PubMed]
,
23. Nelofer, R.; Ramanan, R.N.; Rahman, R.N.; Basri, M.; Ariff, A.B. Comparison of the estimation capabilities
of response surface methodology and artificial neural network for the optimization of recombinant lipase
production by E. coli BL21. J. Ind. Microbiol. Biotechnol. 2012, 39, 243–254. [CrossRef] [PubMed]
24. Kumar, G.; Kumar, D.; Johari, R.; Singh, C.P. Enzymatic transesterification of Jatropha curcas oil assisted by
ultrasonication. Ultrason. Sonochem. 2011, 18, 923–927. [CrossRef] [PubMed]
25. Yu, D.; Tian, L.; Wu, H.; Wang, S.; Wang, Y.; Ma, D.; Fang, X. Ultrasonic irradiation with vibration for
biodiesel production from soybean oil by Novozym 435. Process Biochem. 2010, 45, 519–525. [CrossRef]
26. Deshmane, V.G.; Gogate, P.R.; Pandit, A.B. Ultrasound assisted synthesis of isopropyl esters from palm fatty
acid distillate. Ultrason. Sonochem. 2009, 16, 345–350. [CrossRef] [PubMed]
27. Batistella, L.; Lerin, L.A.; Brugnerotto, P.; Danielli, A.J.; Trentin, C.M.; Popiolski, A.; Treichel, H.; Oliveira, J.V.;
de Oliveira, D. Ultrasound-assisted lipase-catalyzed transesterification of soybean oil in organic solvent
system. Ultrason. Sonochem. 2012, 19, 452–458. [CrossRef] [PubMed]

Molecules 2017, 22, 1972
15 of 15
28. Lerin, L.A.; Feiten, M.C.; Richetti, A.; Toniazzo, G.; Treichel, H.; Mazutt, M.A.; Oliveira, J.V.; Oestreicher, E.G.;
de Oliveira, D. Enzymatic synthesis of ascorbyl palmitate in ultrasound-assisted system: Process optimization
and kinetic evaluation. Ultrason. Sonochem. 2011, 18, 988–996. [CrossRef] [PubMed]
29. Aouf, C.; Durand, E.; Lecomte, J.; Figuerola-Espinoza, M.C.; Dubreucq, E.; Fulcrand, H.; Villeneuve, P.
The use of lipases as biocatalysts for the epoxidation of fatty acids and phenolic compounds. Green Chem.
2014, 16, 1740–1754. [CrossRef]
30. Orellana-Coca, C.; Camocho, S.; Adlercreutz, D.; Mattiasson, B.; Hatti-Kaul, R. Chemo-enzymatic epoxidation
of linoleic acid: Parameters influencing the reaction. Eur. J. Lipid Sci. Technol. 2005, 107, 864–870. [CrossRef]
31. Tao, Y.; Wu, D.; Zhang, Q.A.; Sun, D.W. Ultrasound-assisted extraction of phenolics from wine lees: Modeling,
optimization and stability of extracts during storage. Ultrason. Sonochem. 2014, 21, 706–715. [CrossRef]
[PubMed]
32. Guo, T.; Su, D.; Huang, Y.; Wang, Y.; Li, Y.H. Ultrasound-assisted aqueous two-phase system for extraction
and enrichment of Zanthoxylum armatum lignans. Molecules 2015, 20, 15273–15286. [CrossRef] [PubMed]
33. Huang, K.C.; Li, Y.; Kuo, C.H.; Twu, Y.K.; Shieh, C.J. Highly efficient synthesis of an emerging lipophilic
antioxidant: 2-ethylhexyl ferulate. Molecules 2016, 21, 478. [CrossRef] [PubMed]
34. Basri, M.; Rahman, R.N.; Ebrahimpour, A.; Salleh, A.B.; Gunawan, E.R.; Rahman, M.B. Comparison of
estimation capabilities of response surface methodology (RSM) with artificial neural network (ANN)
in lipase-catalyzed synthesis of palm-based wax ester. BMC Biotechnol. 2007, 7, 53–67. [CrossRef] [PubMed]
35. Friedrich, J.L.R.; Pena, F.P.; Garcia-Galan, C.; Fernandez-Lafuente, R.; Ayub, M.A.Z.; Rodrigues, R.C. Effect of
immobilization protocol on optimal conditions of ethyl butyrate synthesis catalyzed by lipase B from
Candida antarctica. J. Chem. Tecnol. Biotechnol. 2013, 88, 1089–1095. [CrossRef]
36. Martins, A.B.; Graebin, N.G.; Lorenzoni, A.S.G.; Fernandez-Lafuente, R.; Ayub, M.A.Z.; Rodrigues, R.C.
Rapid and high yields of synthesis of butyl acetate catalyzed by Novozym 435: Reaction optimization by
response surface methodology. Process Biochem. 2011, 46, 2311–2316. [CrossRef]
37. Xun, E.N.; Lv, X.L.; Kang, W.; Wang, J.X.; Zhang, H.; Wang, L.; Wang, Z. Immobilization of Pseudomonas
fluorescens lipase onto magnetic nanoparticles for resolution of 2-octanol. Appl. Biochem. Biotechnol. 2012, 168,
697–707. [CrossRef] [PubMed]
38. Lerin, L.; Ceni, G.; Richett, A.; Kubiak, G.; Vladimir Oliveira, J.; Toniazzo, G.; Treichel, H.; Oestreicher, E.G.;
Oliveira, D. Successive cycles of utilization of novozym 435 in three different reaction systems. Braz. J.
Chem. Eng. 2011, 28, 181–188. [CrossRef]
Sample Availability: Samples of retinyl laurate are available from the authors.
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).