An artificial neural network for membrane-bound catechol-O-methyltransferase biosynthesis with Pichia pastoris methanol-induced cultures

The structural and functional characterization of a MP depends on the production of
a sufficient amount of active protein, meaning a regular supply of milligram quantities
of the target enzyme 1]. Therefore, to fulfill this requirement, in this work and for the first time the
biosynthesis of MBCOMT by P. pastoris bioreactor cultures is reported. Initially, in order to select the most appropriated
P. pastoris strain for MBCOMT biosynthesis, trials at a small-scale in baffled shake-flasks were
carried out. Then, a three-stage bioprocess for the biosynthesis of the target protein
by P. pastoris bioreactor cultures was implemented and the lengths of the glycerol fed-batch and
the methanol induction phases were optimized.

Moreover, after selecting a set of independent variables associated with the methanol
induction phase that greatly influence the levels of the MBCOMT, ANN modeling was
carried out in order to maximize the biological activity of the target protein. The
massic and volumetric productivities were not incorporated as an output since the
values of those parameters are in strictly dependence on MBCOMT biological activity
18]. Also, the biomass levels were evaluated in all assays performed in this work but
were not considered in the optimization and validation procedures as an output, since
higher biomass levels not always lead to higher mass productivities of the target
protein.

Small-scale MBCOMT biosynthesis in P. pastoris

Membrane-bound catechol-O-methyltransferase biosynthesis was initially carried out in shake-flasks containing
BMGH medium using a Mut
⁺
(X33) and a Mut
^S
(KM71H) P. pastoris strains 21]. Sometimes, an increase in the number of the heterologous gene can possibly lead
to an increase in transcription and translation rate of the desired gene 22]. In fact, although opposite results had already been published, there are several
examples including the mouse epidermal growth factor or miniproinsulin in which higher
target gene copy numbers lead to higher titers for P. pastoris bioprocesses driven by AOX1 promoter 22]. Therefore, upon the transformation procedure with the target recombinant plasmid,
clones from both strains in study were isolated from plates containing high zeocin
concentrations (2 mg/mL). Following the isolation of these clones from both strains,
it was determined the target gene copy number that was integrated in each strain.
Therefore, using the method previously reported by Nordén and collaborators 23] that takes advantage of the fact that part of the plasmid pPICZ ?, namely the AOX1
TT region is incorporated in the P. pastoris genome together with the gene to be expressed. In particular, for the X33 strain,
the primer efficiencies were 1.88 and 1.87, respectively for the AOX1 TT and AOX2
PROM primer pairs. Similarly, for the KM71H strain, the primer efficiencies were 1.91
and 1.94, respectively, for the AOX1 TT and AOX2 PROM primer pairs. Finally, according
the equation described in the “Methods”, the target gene copy number introduced in
each recombinant strain was determined and it was found that X33-PICZ?-MBCOMT had
nine copies of the target plasmid while the KM71H-PICZ?-MBCOMT had ten copies. In
fact, Nordén and coworkers 23] reported with the aquaporins that colonies isolated from 0.5 mg/mL zeocin could harbor
from 4 to 15 plasmids while from 1 mg/mL, as many as 17 heterologous DNA sequences
can be incorporated. Therefore, although the isolation of clones from plates containing
higher antibiotic concentrations doesn’t exclude completely the occurrence of false
positives, the values here reported (9 and 10 copies for the X33 and KM71H strains,
respectively) are in the same order of magnitude. Then, small-scale fermentation trials
were carried out using 0.5% (v/v) methanol and higher biomass levels were detected
for the X33 strain (OD
₆₀₀
 = 7.5) when compared with those obtained for the KM71H strain (OD
₆₀₀
 = 1.8). Similarly, the target enzyme recovered from the X33 strain presented higher
biological activity (60.25 nmol/h/mg) in comparison to KM71H cells (25.77 nmol/h/mg
of protein) 21]. On the other hand, when the methanol concentration is lowered from 1 to 0.25% (v/v),
similar values for MBCOMT biological activity were obtained for the X33 (61.73 nmol/h/mg
of protein) and the KM71H (60.62 nmol/h/mg of protein) strains 21]. Specifically, we believe that the observed differences in these two strains concerning
their performance in MBCOMT biosynthesis seem to be associated with the methanol concentration
used for induction and not for example with the target gene copy number inserted in
the genome since it is similar.

The value previously reported 21] with both P. pastoris strains for MBCOMT biological activity is higher than those previously reported by
our research group using Brevibacillus choshinensis as the expression system (48.07 nmol/h/mg of protein) 24]. In general, for intracellular expression, it has been reported that it is preferable
use Mut
^S
over Mut
⁺P. pastoris strains because of increased specific yield of heterologous protein 25]. However, as previously reported by Maurer and collaborators, the volumetric productivity
QP is the most plausible target for optimization in fed-batch processes 26]. Therefore, since the main aim of this work was to maximize MBCOMT expression irrespective
the biomass levels, P. pastoris Mut
⁺
X33 was chosen for further bioreactor trials since regardless the methanol concentration
used, the expression levels of the target protein were the highest obtained and they
didn’t significantly change when different methanol concentrations are applied.

MBCOMT biosynthesis from methanol-induced Pichia pastoris bioreactor cultures

Membrane-bound catechol-O-methyltransferase biosynthesis was carried out in mini-bioreactors (working volume
0.25 L) in modified basal salts medium (BSM) containing 4.35 mL/L trace metal solution
(SMT) 27] and the pH was adjusted to 4.7 in order to minimize precipitation and, consequently,
undesired operational problems such as starvation of nutrients and optical densities
measurement interferences 14]. P. pastoris cultivations in bioreactor were initiated with a glycerol batch phase (30 g/L glycerol)
that ends when glycerol was depleted, indicated by a sharp increase in the dissolved
oxygen (DO) 14]. After this stage, a fed-batch growth on glycerol [50% (v/v) at 18.54 mL/L/H] during
different periods was employed, followed by the methanol induction phase where P. pastoris was cultivated on a methanol fed-batch mode. In order to promote the derepression
of the AOX promoter prior to induction, 1 h before starting the induction phase, methanol
was added to the reaction vessel at the flow-rate later employed in the methanol fed-batch
phase.

Preliminary trials were carried out in order to analyze the optimal period of the
glycerol fed-batch phase as well as the optimal duration of the methanol induction
phase that maximize MBCOMT expression. Therefore, keeping constant the methanol flow-rate
(3.6 mL/L/H) in the induction phase, assays with 3, 5 or 7 h glycerol fed-batch phase
were carried out. Methanol induction phase was maintained during 60 h and samples
were collected with an interval of 2 h until 12 h and then every 12 h to follow the
MBCOMT expression profile. As depicted in Fig. 1, the highest MBCOMT biological activity levels were detected when a 3 h period was
applied in the glycerol fed-batch phase. In addition, concerning to the methanol induction
phase, MBCOMT achieved a maximum expression of 121.0 nmol/h/mg of protein at 12 h
of induction, what led us to assume a 3 h glycerol fed-batch period and a 12 h induction
period for further experiments. In fact, a shorter induction period can be greatly
advantageous over other previously reported strategies 27], 28] where induction usually takes more than 48 h, being more time-consuming and laborious.
Moreover, the shorter induction period allows terminating the fermentation before
a decrease in the cell’s physiological activity is observed 29].

Fig. 1. Typical time profile of MBCOMT specific activity (nmol/h/mg of protein) obtained by
P. pastoris bioreactor cultures using different periods of the glycerol fed-batch phase with
a methanol constant feed flow-rate at 3.6 mL/L/H (each value represents the mean of
three independent samples).

Following these findings, we evaluated if the expression of the target protein was
significantly affected by the methanol constant flow-rate as well as the addition
of the chemical chaperone DMSO that has been described to increase the expression
levels of some MP 11]–13], 30], 31]. Therefore, keeping constant the operational parameters previously optimized, distinct
assays were performed: with different methanol constant flow rates at 2, 3.6 and 5.2 mL/L/H
while others were performed maintaining the methanol flow-rate at 3.6 mL/L/H and changing
the DMSO concentration [2.5, 5 and 7.5% (v/v)] in the culture according to what previously
described 11]–13], 30]. As demonstrated in Fig. 2a, for the lowest methanol constant flow-rate (2 mL/L/H), a highest MBCOMT expression
level of 158 nmol/h/mg were obtained, contrasting with 120 and 107 nmol/h/mg for 3.6
and 5.2 mL/L/H, respectively. Also, the methanol and the biomass levels at distinct
stages of the induction phase were quantified in these assays, as depicted in Fig.
2b and Table 1, respectively. In general, for the different methanol flow-rates applied, the methanol
levels increase from 0 to 6 h and then they decrease until the end of the induction
phase. At the early stage of the induction phase, methanol doesn’t seem to be consumed
in a large extent since P. pastoris cells may be going through a transition period where they stop consuming glycerol
and start to oxidize methanol. Nevertheless, it is possible to observe that for methanol
constant-flow rates of 3.6 and 5.2 mL/L/H, the concentration of methanol in the culture
broth is higher (near 10 and 12.5 g/L respectively) at 6 h of induction when compared
with the lowest flow-rate employed (1 g/L). Therefore, it is feasible to assume that
using a lower flow rate (2 mL/L/H) allows the establishment of an appropriated balance
between activation of the AOX promoter and, consequently, production of the target
enzyme and accumulation of methanol in the culture medium that can be responsible
for the undesired toxicity, as it may be happening for 3.6 and 5.2 mL/L/H 14]. Moreover, an optimal ratio of methanol to cell concentration should be applied 32], otherwise high methanol feeding rates stress the cell machinery and negatively affect
the process performance 32], 33].

Fig. 2. a Analysis of different methanol flow-rates (without the addition of DMSO) and different
DMSO concentrations (keeping constant the methanol flow-rate at 3.6 mL/L/H) on MBCOMT
specific activity (nmol/h/mg of protein) obtained by P. pastoris bioreactor cultures. b Time course analysis of the methanol levels in the above mentioned assays measured
by HPLC-RID. In both experiments, a 3-h period of the glycerol fed-batch was applied
and induction was conducted during 12 h (each value represents the mean of three independent
samples).

On the other hand, when different DMSO concentrations were added to the P. pastoris cultures, the highest MBCOMT biosynthesis of 216 nmol/h/mg was detected for 5% (v/v),
which represents an increase of 1.8-fold when compared with the control (without DMSO).
Again, the methanol levels were also quantified in these trials and interestingly,
its time course profile with the addition of 5% (v/v) DMSO conducted with 3.6 mL/L/H
of methanol resembles the profile previously obtained for the 2 mL/L/H methanol flow
rate and not the 3.6 mL/L/H. Following these results, it is reasonable to think that
adjusting the DMSO concentration to the cell needs, the methanol is more efficiently
used what, in a last analysis, leads to an increase in the biosynthesis of the target
protein.

The addition of 5% (v/v) DMSO proved to have a positive effect on the expression of
this particular MP, has been demonstrated previously for G protein-coupled receptors
by other authors 11]–13], 30], 31]. Although the mechanism by which DMSO increases MP expression is not yet fully understood,
Murata and collaborators showed that DMSO induces membrane proliferation through the
increase of the phospholipid content within Saccharomyces cerevisiae cells 34]. On the other hand, it was also reported that DMSO possess antioxidant properties,
preventing protein oxidation (increase in protein carbonyl content and decrease in
free thiol content) in rat brain homogenates induced by ferrous chloride/hydrogen
peroxide 35]. Therefore, it is likely that the benefits of using DMSO on the expression of membrane
proteins can be associated with the induction of membrane proliferation or with the
reduction of protein oxidation or a combination of both. Moreover, despite the optimal
temperature for growth and production of proteins in P. pastoris is 30°C 14], some authors claim that working at lower temperatures (until 20 to 25°C) may improve
the target protein biosynthesis 36], lower cell lysis 37] and decrease the proteolytic activity 38]. Therefore, in this work, the temperature was also included as an independent process
parameter to optimize MBCOMT biosynthesis from P. pastoris and the ranges (20, 25 and 30°C) were selected according to what has been reported
in the literature 14], 37].

According to the results reported in this section and the synergy observed between
methanol flow rate and DMSO concentration in the culture broth, the most appropriated
ranges of the independent variables selected for performing the experimental design
were defined, as shown in Table 2. Finally, a summary of the optimized conditions for the expression of MBCOMT from
P. pastoris bioreactor methanol-induced cultures is presented in Fig. 3 where the ranges of the independent variables selected for the ANN modeling are presented
as well as the major experimental conditions selected.

Table 2. Coded levels used for temperature, methanol constant feed flow-rate and DMSO in the
CCD

Fig. 3. Structure of the optimized four-stage bioprocess implemented in this work for recombinant
MBCOMT biosynthesis by P. pastoris bioreactor cultures.

Experimental design and artificial neural network modeling

A set of 17 experiments defined by CCD for optimization of the induction phase for
maximizing MBCOMT biosynthesis in P. pastoris culture are listed in Tables 2 and 3. In general, lower MBCOMT biological activity levels were detected when the input
variables defined in CCD were at the lowest levels. Specifically, MBCOMT biosynthesis
is maximized at higher methanol constant-flow rate concentrations and when the concentration
of DMSO added is higher. On the other hand, an increased in the induction temperature
coupled to an increase in the other input variables also lead to an increase in biologically
active MBCOMT expression. According to the ANN modeling results in calibration dataset
(DoE experiments) (Table 3), the predicted maximum for MBCOMT specific activity (384.8 nmol/h/mg of protein)
was achieved at 30°C, 2.9 mL/L/H methanol constant flow-rate and with the addition
of 6% (v/v) DMSO. In general, as previously demonstrated for others MP 11]–13], 30], 31], the addition of DMSO to the culture proved to have a positive effect on MBCOMT expression
since over the model optimization the maximum target protein specific activity is
achieved at higher DMSO concentrations. In addition, the output seems to be maximized
when the methanol constant flow-rate and the induction temperature are close to the
upper values defined in the CCD. This can be explained by the increase in the biomass
levels (data not shown) caused by the increase in the temperature and, since there
is more methanol that is being oxidized by the AOX promoter, the supply of inducer
needs to be higher in order to maintain induction. An ANN model was developed in order
to optimize the induction phase for maximizing MBCOMT biosynthesis from P. pastoris bioreactor cultures. The model was calibrated with the experiments defined in Table 3.

Table 3. List of experiments performed for MBCOMT biosynthesis from P. pastoris bioreactor methanol-induced cultures based on CCD and ANN modeling

Modeling of the methanol induction phase using artificial neural network

The ANN model was applied for the optimization of the induction phase for MBCOMT biosynthesis
in P. pastoris bioreactor cultures using a stepwise process until the maximum MBCOMT biological
activity was achieved. Four iterations were required to achieve the maximum MBCOMT
specific activity (384.8 nmol/h/mg of protein) under the optimal conditions [30°C,
2.9 mL/L/H methanol constant flow-rate and 6% (v/v) DMSO] and to validate the model
with new experiments. In the end, an improvement of 1.53-fold over the best conditions
performed in the DoE step (see experiment 15, Table 3) was achieved while an improvement of 6.4-fold over the small-scale biosynthesis
in baffled shake-flasks was achieved.

The obtained ANN model is mostly unbiased because the slope and R
²
of the fitting between the measured and predicted output were close to 1 (0.9064 and
0.97161, respectively) (see Fig. 4). In Fig. 5 are depicted the contour plots obtained from the ANN model for two combinations between
the three operational conditions in study. The modeling results showed that the MBCOMT
specific activity is sensitive to the operational conditions. The ANN parameters for
the final validation model are presented in Additional file 1.

Fig. 4. ANN modeling results of MBCOMT specific activity for the last optimization steps.
Blue circle, red circles and green triangles represent data from the CCD, outliers and from model optimization.

Fig. 5. Contour plots of MBCOMT specific activity as function of induction temperature, methanol
constant flow-rate and DMSO concentration: a modeling results of MBCOMT specific activity as function of DMSO concentration and
methanol constant flow-rate for the last optimization step. b Modeling results of MBCOMT specific activity as function of induction temperature
and methanol constant flow-rate for the last optimization step.

Bioprocess monitoring at the optimal conditions estimated by the ANN model

At the optimal conditions estimated by the ANN model [30°C, 2.9 mL/L/H methanol constant
flow-rate and 6% (v/v) DMSO], the carbon source levels as well as the biomass levels
and the number of viable/depolarized cells were analyzed, as depicted in Fig. 6. In what concerns to the P. pastoris growth, a marked increase in OD
₆₀₀
was detected between the end of the batch phase and the fed-batch growth of glycerol
and it keeps increasing until the end of the induction phase with a value near 123
units of OD
₆₀₀
. The methanol and glycerol levels were quantified using a HPLC with refractive index
detection and it was verified that the glycerol concentration also increases during
the fed-batch glycerol phase, despite the higher accumulation of biomass during this
stage. On the other hand, a low consumption of methanol was verified between the second
and the third hours of the glycerol fed-batch phase since we consider that the consumption
of glycerol is preferred over the methanol. On the other hand, at the end of the induction
phase, almost no methanol was detected since P. pastoris cells are oxidizing it all, what can be indicating that the AOX promoter is highly
active. Finally, the flow cytometry analysis led us to conclude that the changes introduced
at the second hour of the glycerol fed-batch phase (namely the shift to the induction
temperature, the addition of DMSO and the initiation of the methanol flow-rate) did
not altered significantly the number of viable cells (94.8% compared to 95.4%) in
culture. Furthermore, at the end of the induction phase, approximately 90% of viable
cells were obtained, a value that is acceptable and is in accordance with P. pastoris bioprocesses that include the AOX promoter with a 12 h induction period 39].

Fig. 6. Time course analysis of biomass levels, carbon sources concentrations and number of
healthy P. pastoris cells at different stages of the optimal point estimated by the ANN model [30°C,
2.9 mL/L/H methanol constant flow-rate and 6% (v/v) DMSO]. a Biomass levels measured by spectrophotometric determination at 600 nm and methanol
and glycerol levels measurements by HPLC with RID; (each value represents the mean
of three independent samples). b Dot plots of green fluorescence of cells (BOX, x-axis) plotted against red fluorescence
(PI, y-axis) obtained with cell samples taken at different stages of the optimum point
retrieved from the ANN modelling. Three main subpopulations of cells can be distinguished
corresponding to: healthy cells, no staining; cells with depolarized membranes, stained
with BOX; and cells with permeabilized membranes, stained with PI. A total of 10,000
events were collected for these analysis. The variation on the percentage of healthy
cells at different stages of the bioprocess is depicted in the graph. Each experiment was conducted in duplicate.

To our best knowledge, this is the first systematic study where the interaction between
two commonly studied operational parameters (induction temperature and methanol flow
rate) and the addition of chemical chaperones (specifically the DMSO) are successfully
reported to optimize MP expression by P. pastoris bioprocesses using ANN modeling.