Biomass Conversion and Biorefinery

https://doi.org/10.1007/s13399-023-04043-w

ORIGINAL ARTICLE

Optimization of biomass and polyhydroxyalkanoate production

by Cupriavidus necator using response surface methodology
and genetic algorithm optimized artificial neural network
Pema Lhamo1 · Biswanath Mahanty1 · Shishir Kumar Behera2

Received: 12 January 2023 / Revised: 28 February 2023 / Accepted: 7 March 2023

© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2023

Abstract
High polyhydroxyalkanoate (PHA) yield from selected substrates associated with minimal accumulation of residual biomass
can improve the process economy. In this study, different carbon (glucose, sucrose, glycerol, and acetic acid) and nitrogen
­(NH4Cl and urea) sources were screened for PHA production by Cupriavidus necator. The effects of incubation time, nitro-
gen, and phosphate concentration on biomass growth and PHA production were co-optimized through response surface
methodology (RSM) and genetic algorithm-optimized artificial neural network (GA-ANN). Sucrose and urea were found
to offer significantly better (p <0.001) biomass (1.468 ± 0.007 g ­l-1) and PHA (0.924 ± 0.02 g ­l-1) yield when compared
with other carbon and nitrogen sources. Though the performance of both the models remains similar for biomass (R2 =
0.97–0.98), GA-ANN (with six neurones in a hidden layer) seems exceptionally better in predicting PHA yield (R2 = 0.97)
when compared to the polynomial model (R2 = 0.92). The maximum PHA concentration of 2.69 g ­l-1 was predicted by the
ANN model at an incubation time of 62.80 h with 2.0 g l­ -1 of nitrogen and 4.0 g l­ -1 of phosphate concentration. The multi-
composite desirability using the GA-ANN model projected a better polymer-to-biomass ratio compared to the polynomial
model. The inclusion of a cost-benefit analysis framework may be warranted before recommending the optimal conditions
obtained through multivariate regression and GA-ANN models.

Keywords Polyhydroxyalkanoates · Central composite design · Multi-objective optimization · Composite desirability ·

Artificial neural networks

1 Introduction waste [1] warrant the exploration of biopolymers derived

from renewable resources as a sustainable and environment-
Plastics derived from fossil fuels have offered a comfort- friendly alternative [2]. Polyhydroxyalkanoates (PHAs),
able lifestyle and have driven the global economy for dec- biodegradable thermoplastic polymers accumulated by a
ades. However, serious environmental impacts associated diverse group of microorganisms, can be very similar to
with improper management and expensive recycling of the conventional petrogenic polypropylene in terms of molecu-
lar weight, stiffness, and melting point [3]. Though it has not
* Biswanath Mahanty been extensively commercialized, PHAs possess immense
bmahanty@karunya.edu potential in healthcare, packaging, tissue engineering, and
Pema Lhamo agriculture (Table S1). Being nontoxic and carbon neu-
pemalhamo20@karunya.edu.in tral, PHAs are promising biopolymers aligned with United
Shishir Kumar Behera Nations (UN) Sustainable Development Goals (SDGs) for
shishir.kb@vit.ac.in sustainable consumption and production (SDG 12) and
1
conservation and sustainable use of the oceans, seas, and
Department of Biotechnology, Karunya Institute marine resources (SDG 14) [4]. PHA-based products can be
of Technology and Sciences, Karunya Nagar, Coimbatore,
Tamil Nadu 641 114, India composted, mineralized, and reabsorbed by plants serving
2 as feedstock for PHA production, closing the loop of mate-
Industrial Ecology Research Group, School of Chemical
Engineering, Vellore Institute of Technology, Vellore, rial consumption cycles and fulfilling “circular economy”
Tamil Nadu 632 014, India model [5, 6].

13
Vol.:(0123456789)
 Biomass Conversion and Biorefinery

Mixed microbial culture utilizing volatile fatty acids of independent variables [20, 21]. Though the RSM-based
(VFAs) can produce copolymer (HB: 45–70 mol%, HV: model does not provide any mechanistic view of the process,
30–55 mol%) with decreased glass transition, melting tem- the prediction of the optimal factor setting and optimization
perature, crystallinity, and average molecular weight [7, 8]. in a nonlinear system is quite effective [22]. On the other
Among others, Cupriavidus necator has been known to be hand, artificial neural network (ANN) is increasingly being
an efficient polymer producer accumulating PHAs up to adopted to model complex nonlinear processes without the
90% of its dry cell weight using simple sugars (glucose and need for explicit mathematical formalization, unlike RSM.
fructose) and VFAs [9, 10]. Having the versatility of both Though an efficient ANN model requires larger data set in
autotrophic and heterotrophic growth, C. necator can utilize general, a smaller dataset statistically well distributed in
a wide range of organic substrates [9]. The polymer yield, input dimension (e.g., in the case of RSM) can effectively
composition, and material properties can differ based on be utilized [23]. ANN has been adopted to model and opti-
carbon and nitrogen sources [11]. mize poly-β-hydroxybutyrate (PHB) production in microbial
Exorbitantly high cost of traditional substrates has led to culture as a function of media composition [24]. The exten-
an exploration of agro-industrial wastes, such as lignocel- sion of RSM design into ANN modeling has been adopted
lulosic materials, biodiesel waste, sucrose-rich molasses, in different bioprocess systems, offering better predictability
and vinasse from the sugar industry as cheap renewable raw [25, 26]. However, the choice of ANN topology (number of
materials which can be used for PHA production with appro- neurons in the hidden layer, initialized weight, and bias) in
priate pre-treatment (Table S2) [12]. However, the nutrient most of the relevant studies remains heuristic [27]. Genetic
composition (i.e., C:N:P molar ratio) of the waste may not algorithm can be adopted to fine-tune the ANN topology
be optimal as low C:N or C:P ratio can have a negative influ- and cross-validation–based data splitting for training, test-
ence on the quantity and quality of the accumulated polymer ing, and validation subsets [28, 29]. Additionally, different
[13]. Therefore, it is crucial to select the appropriate sub- responses from the same process are separately mapped
strate feed and understand and optimize the defined media using independent networks [30], which may not be con-
composition (Table S3) for the overproduction of biopoly- venient for a co-optimization problem.
mers, before switching over to agro-industrial waste [14]. This work aims at selecting appropriate carbon (glucose,
An increase in biopolymer production alone may not be sucrose, glycerol, and acetic acid) and nitrogen ­(NH4Cl and
sufficient to improve the process economy as residual bio- urea) sources for the growth of Cupriavidus necator and PHA
mass growth, PHA accumulation, and substrate consumption production using factorial design. The biopolymer production
are highly interdependent [15]. Several studies have reported with selected substrates using nitrogen and phosphate concen-
the optimization of culture medium for improvement in bio- tration and incubation time as predictors was independently
mass and PHA production [16]. Though the biopolymer yield modeled adopting central composite design (CCD)–based
improves with the biomass concentration [17], the increased RSM methodology and genetic algorithm optimized ANN
cost of cell disruption and dissolution of residual biomass (GA-ANN). Subsequently, GA-driven co-optimization of the
might offset the benefit if the biopolymer recovery is not sub- process, i.e., maximized PHA accumulation while minimizing
stantially improved. For such uncertain and competing objec- biomass growth, was projected from a composite desirability
tives, no single solution can be offered [18] and a trade-off function in both the modeling frameworks. The optimal sub-
can be formalized using a composite desirability function for strate concentration evaluated from this study can be extended
those responses under multi-objective optimization [19]. to identify appropriate agro-industrial wastes for cost-effective
Statistical experimental design-based response surface PHA production. To the author’s best knowledge, this is the
methodology (RSM) has been widely explored to model first report on the efficient utilization of sucrose by Cupri-
measured response(s) as a linear, quadratic, or cubic function avidus necator for PHA production.

Table 1  Coded levels of different variables influencing PHA produc- 2 Material and methods
tion included in the CCD
Variables a Coded levels 2.1 Microorganism and culture conditions
−1.682 −1 0 1 1.682
Cupriavidus necator KCTC 22469, purchased from Korean
Time (h) 4 21 47 72 90 Collection for Type Cultures, South Korea, was maintained
Urea (g l­-1) 0.10 0.48 1.05 1.60 2.00 and subcultured in nutrient broth medium at 30°C. The
Phosphate (g ­l-1) 0.10 0.89 2.05 3.20 4.00 PHA production studies were carried out in mineral salt
a
The initial concentration of carbon (sucrose) was maintained at 30 g medium (MSM) containing (g l­-1) ­NaH2PO4, 4; ­Na2HPO4,
l­ -1 across the design 4.6; ­K2SO4, 0.45; ­Mg2SO4, 0.39; ­CaCl2, 0.062; and 5 ml ­L-1

13
Biomass Conversion and Biorefinery

Table 2  CCD for three variables Runa Time (h) Urea (g l­-1) Phosphate (g ­l-1) PHA (g ­l-1) Biomass (g l­-1)
with the observed and predicted
responses Observed Predicted Observed Predicted

1 21.43 1.61 0.89 0.47 0.27 0.48 0.46

2 4.00 1.05 2.05 0.05 0.08 0.47 0.44
3 21.43 0.48 0.89 0.71 0.71 0.48 0.54
4 47 1.05 2.05 1.64 1.55 0.83 0.85
5 47 0.10 2.05 0.57 0.68 0.77 0.72
6 47 1.05 2.05 1.64 1.55 0.83 0.85
7 47 1.05 2.05 1.61 1.55 0.87 0.85
8 47 1.05 2.05 1.63 1.55 0.85 0.85
9 47 1.05 2.05 1.42 1.55 0.84 0.85
10 21.43 0.48 3.20 1.1 0.05 0.51 0.54
11 72.56 0.48 3.20 0.88 0.77 0.8 0.81
12 72.56 1.61 0.89 0.46 0.28 0.98 0.98
13 47 1.05 4.00 1.34 1.41 0.82 0.82
14 47 2.00 2.05 1.05 1.28 0.83 0.86
15 47 1.05 2.05 1.43 1.55 0.87 0.85
16 72.56 0.48 0.89 0.98 0.72 1.07 1.06
17 21.43 1.61 3.20 1.25 1.20 0.48 0.79
18 90 1.05 2.05 0.38 0.69 1.1 1.11
19 72.56 1.61 3.20 2.1 1.92 1.08 1.06
20 47 1.05 0.10 0.32 0.59 0.76 0.75
a
Run no. 10 and run no. 17 are excluded from polynomial model for PHA and biomass, respectively

of trace element solution containing (g ­l-1) ­ZnSO4.7H2O, PHA production. All pairwise multiple comparisons
0.1; ­MnCl2.4H2O, 0.03; ­H3BO3, 0.03; ­CoCl2.6H2O, 0.20; between the groups were performed by adapting the
­CuCl2.2H2O, 0.01; N ­ iCl2.6H2O, 0.02; and N
­ a2MoO4.2H2O, Holm-Sidak method. A significance level of α = 0.05
0.03. The carbon and nitrogen sources (as per the experi- was used throughout the analysis. The statistical analysis
mental design) were added to the MSM from a filtered steri- module of SigmaPlot (version 12.0) (Systat Software Inc.,
lized stock solution. San Jose, California, USA) was used for the analysis.

2.2 Selection of carbon and nitrogen sources

2.3 Experimental design for modeling
The effects of various carbon (i.e., glucose, sucrose, glyc-
erol, and acetic acid) and nitrogen (i.e., urea and N ­ H4Cl) To model biomass growth and PHA production as a func-
sources on PHA production were evaluated using batch tion of incubation time, urea, and phosphate concentra-
studies. Each experimental set containing MSM (50 ml) tion, a CCD-based RSM was adopted. The initial concen-
supplemented with 30 g ­l-1 of carbon (filter sterilized) and tration of sucrose was maintained at 30 g l­-1 across the
0.5 g l­-1 of nitrogen, adjusted to pH 7.0, was inoculated design to offer different C/N ratios in the design space.
with overnight grown culture at a targeted biomass con- Five coded levels (−α, −1, 0, +1, +α) for each independ-
centration of 1.1 g ­l-1. The experimental sets were incu- ent variable and their corresponding values are shown
bated for 48 h at 30°C in a shaker incubator (110 rpm). in Table 1. The design contains 20 experimental runs
Following the measurement of biomass growth (λmax = including eight factorial points, six axial points, and six
600 nm), the culture was centrifuged at 4800 × g for 10 center-point replicates [31]. The CCD and associated
min. The culture supernatant and pellet were used for the response values are shown in Table 2. The experimental
measurement of residual substrate concentration and PHA results of CCD were fitted with a second-order polyno-
production, respectively. mial model (Eq. 1).
A two-way analysis of variance (ANOVA) was used 3 2 ∑
3 3
to detect the effects of carbon and nitrogen sources, (1)
∑ ∑ ∑
Y = 𝛽i0 + 𝛽i Xi + 𝛽ij Xi Xj + 𝛽ii Xi2
together with their interactions on biomass growth and i=1 i=1 j=i+1 i=1

13
 Biomass Conversion and Biorefinery

Fig. 1  Effect of different carbon sources on a biomass, b PHA production, and nitrogen sources on c PHA production. Different letters indicate
the statistical difference between the groups (p < 0.050). Error bars denote the standard deviation of duplicate measurements

13
Biomass Conversion and Biorefinery

Table 3  Two-way ANOVA Source of variation PHA Biomass

of carbon and nitrogen source
selection for biomass and PHA DF SS MS F P DF SS MS F P
production
C 3 1.686 0.562 493.960 <0.001 3 0.044 0.015 46.882 <0.001
N 1 0.006 0.006 5.498 0.047 1 0.001 0.001 4.194 0.075
C×N 3 0.027 0.009 7.824 0.009 3 0.002 0.001 1.966 0.198
Residual 8 0.009 0.001 8 0.002 0.001
Total 15 1.728 0.115 15 0.049 0.003

where Y represents any of the responses (PHA, biomass) optimized using genetic algorithm. The L-M backpropagation
and β0, βi, βij, and βii are the regression coefficients for inter- algorithm (trainlm) was used to train the network with a manual
cept, linear effects, two-way interactions, and square terms, sample division into training, validation, and test sets.
respectively.
2.5.1 Genetic algorithm–based optimization of ANN
2.4 Quadratic model selection and statistical architecture
analysis
The number of HL neuron initialized weight and bias values
The response obtained from the different batch experimen- were optimized by a genetic algorithm [32]. For a given
tal runs was fitted to a quadratic model and checked for number of predictors (­ Ni), response (­ No), and HL neurons
the significance of linear and nonlinear model terms in ­(Nh), the initialized population consists of a real-valued vec-
ANOVA. Following the removal of the outlier (based on tor ai ∈ [−1, 1] of length (l) as in Eq. (2).
studentized residual), a stepwise regression was performed
(2)
( ) ( ) ( )
while adding or removing the predictors at each step (criti- l = Nh + Ni × Nh + Nh × No + Nh + No
cal p values set at 0.05 and 0.10, respectively) until the
The first Nh genes are to be decoded in binary reflecting
final model is achieved. The accuracy of the final model
if a hidden neuron is included or not. The second and third
was determined by the coefficients of determination (R2),
term in the RHS of the equation represents the number of
and adjusted R2 (adj-R2) values, while the significance of
weights from IL to the HL and from the HL to the OL. The
the terms was evaluated using ANOVA. The model selec-
fourth term accounts for the number of bias values to the
tion and ANOVA analysis were performed using the step-
HL and OL neurons.
wiselm function in Matlab® ver. 9.5.
The decoding of HL neuron selection [33] is based on Eq. (3).
{
2.5 Artificial neural network model 1 if ai > 0
yi =
0 otherwise (3)

ANN model was developed to predict biomass growth and PHA

where ­yi refers to the ith hidden neuron which is utilized or not.
production from the experimental data set. The adopted ANN
All bias and weights pertinent to an unused HL neuron are also
architecture comprises one input layer (IL) containing three pre-
disregarded from architecture. No transformation for the weight
dictors, one hidden layer (HL), and one output layer (OL) with
and bias values was required. A population size of 500 and a
two responses, i.e., biomass and PHA. However, the number
maximum generation of 200 were adopted in GA optimization,
of neurons in HL and the initialized weight and bias (bj) were
retaining other parameters at their default settings in Matlab®

Table 4  ANOVA for the Model terms PHA Biomass

regression model of PHA and
biomass production SS DF Mean SS F p value SS DF Mean SS F p value

Total 5.898 18 0.328 0.686 18 0.038

Model 5.416 8 0.677 14.050 <0.001 0.674 8 0.084 70.925 <0.001
Linear 1.721 3 0.574 11.902 0.001 0.590 3 0.197 165.590 <0.001
Nonlinear 3.696 5 0.739 15.339 <0.001 0.084 5 0.017 14.124 <0.001
Residual 0.482 10 0.048 0.012 10 0.001
Lack of fit 0.426 5 0.085 7.600 0.022 0.010 5 0.002 6.481 0.031
Pure error 0.056 5 0.011 0.002 5 0.000

13
 Biomass Conversion and Biorefinery

13
 Biomass Conversion and Biorefinery

◂Fig. 2  Three-dimensional surface plots (left panels) and two-dimen- where subscript i ∈ {P, X} indicates either PHA (P) or bio-
sional contour plots (right panels) showing the binary effect of a, d mass (X) as the response. The Yi, Li, and Ui are modeled
urea concentration-time, b, e urea-phosphate concentrations, and c, f
phosphate concentration-time on microbial PHA production using the
and minimum and maximum acceptable responses attainable
reduced quadratic model. Variables not projected in surface/contour by changing the predictors. The shape of the desirability
plots are on hold at their respective center-point reference level, i.e., function depends on the value of “s.” The overall (global)
phosphate 2.05 g l­-1; urea 1.05 g ­l-1; and time 47 h desirability D is the weighted geometric mean of all the indi-
vidual desirability functions as in Eq. (7).
2022b. The average normalized square error (ANSE) for both the ��
responses (i.e., biomass and PHA) on the entire dataset was set as w
(7)
∑
D = wi di i
an objective function for minimization [32], as shown in Eq. (4).
n � n �
where wi is the weight factor for the responses indicating
their importance to the overall desirability [35].
⎛∑ �2 ∑ �2 ⎞
⎜ yX,i − ŷ X yP,i − ŷ P ⎟
i=1 i=1
ANSE = ⎜ n
�2
+ n ⎟∕2
�2 ⎟ (4)
2.7 Analytical methods
⎜ ∑� ∑�
⎜ tX,i − ̂tX tP,i − ̂tP ⎟
⎝ i=1 i=1 ⎠

where yX,i, yP,i are predicted and tX,i, tP,i are target responses of 2.7.1 Estimation of biomass and substrate concentration
the ith sample for biomass and PHA, respectively. The
ŷ X , ŷ P , ̂tX , ̂tP are sample averaged predicted and target The biomass concentration in experimental sets at the end
responses for biomass and PHA, respectively. of the incubation period was estimated from optical density
measurements of resuspended biomass (λmax = 600 nm) and
2.6 Multi‑objective optimization a calibration constructed in a priori between dry cell weight
and optical density measurement. The estimation of residual
The individual desirability [34] for PHA production (maximi- sucrose, urea, and phosphate concentration in the culture
zation) and biomass growth (minimization) in the multi-objec- supernatant was measured using the dinitrosalicylic acid
tive optimization are shown in Eqs. (5) and (6), respectively. (DNS) method [36], diacetyl monoxime [37], and ammo-
nium molybdate-ascorbic acid method [38], respectively.
⎧
⎪ 0 if YP < LP
⎪ � �S 2.7.2 PHA extraction and quantitation
dP = ⎨ YP −LP
UP −LP
if LP ≤ YP ≤ UP (5)
⎪
⎪ 1 if YP > UP
⎩ The biomass collected after PHA accumulation was centri-
fuged at 4800 × g for 10 min and dried at 60°C in a hot air
oven. The dried pellet obtained was suspended in chloroform
⎧ 1 if YX < LX
⎪� �S (25 ml), sonicated for 45 min at 50°C, and left to dissolve for
dX = ⎨ UX −YX
UX −LX
if LX ≤ YX ≤ UX (6) 5 days at 30°C in a shaker at 120 rpm [39]. The solution was
⎪
0 if Y > U then filtered to remove all undissolved material.
⎩ X X

Fig. 3  Optimized ANN archi-

tecture a performance during
training and b comparison of
the experimental and predicted
responses

13
 Biomass Conversion and Biorefinery

Table 5  Composite desirability for PHA and biomass production adopting different response weights
Model Weightage Variables Projected response Desirability D
-1 -1 -1 -1
PHA X T (h) N (g ­l ) P (g ­l ) PHA (g ­l ) X (g ­l ) P/X PHA X

Polynomial model 1.00 0.00 61.87 2.00 4.00 2.40 1.10 2.18 1.00 0.14 1.00
0.75 0.25 55.42 1.72 4.00 2.24 1.02 2.19 0.93 0.21 0.64
0.50 0.50 35.74 1.26 2.42 1.51 0.79 1.91 0.63 0.39 0.49
0.25 0.75 20.32 0.88 0.99 0.73 0.52 1.40 0.30 0.59 0.51
GA-ANN model 1.00 0.00 62.80 2.00 4.00 2.69 1.01 2.69 1.00 0.18 1.00
0.75 0.25 47.00 2.00 3.76 2.52 0.77 3.27 0.93 0.37 0.74
0.50 0.50 38.98 2.00 3.61 2.19 0.65 3.37 0.81 0.47 0.62
0.25 0.75 28.31 2.00 3.30 1.41 0.48 2.94 0.52 0.60 0.58

1
The extracted PHAs in the chloroform layer were quantified H proton NMR analysis was used to determine the com-
using a crotonic acid assay [40]. Briefly, 100 μl of the organic layer position of PHA produced by C. necator. The PHA samples
was air-dried to which 1 ml of concentrated sulfuric acid was added were dissolved in deuterated chloroform ­(CDCl3), and the
and kept in a water bath at 100°C for 10 min. The solution was then spectra was recorded against tetramethyl silane as an inter-
cooled and mixed properly and spectrophotometrically measured nal standard using a Bruker BioSpin GmbH 300 Mhz NMR
(λmax = 235 nm) against a sulfuric acid blank. Crotonic acid assay spectrometer. The chemical shifts were reported in ppm.
of PHB purchased from Sigma-Aldrich was used as standard.

2.7.3 Polymer characterization 3 Results and discussion

The IR spectrum of air-dried biopolymer was analyzed with 3.1 Selection of carbon and nitrogen sources
Bruker VERTEX 80v (Bruker, Germany) Fourier transformation
infrared (FTIR) spectrophotometer. The analysis was performed The effect of different carbon and nitrogen sources on
using a sample-laden KBr disk, and the spectrum was recorded in biomass growth and PHA production is shown in Fig. 1.
the range 4000–400 c­ m−1 with a resolution of 2 ­cm−1 measured It is evident that biomass accumulation across the carbon
in transmittance mode. sources (except between glucose and glycerol) is significantly

Fig. 4  Fourier transform infra-

red (FTIR) spectrum of biopoly-
mer extracted from Cupriavidus
necator

13
Biomass Conversion and Biorefinery

different (p <0.001), where it varied between 0.499 ± 0.006 g the difference between urea and ­NH4Cl was only discernible
­l-1 (acetic acid) and 1.468 ± 0.036 g l­ -1 (sucrose) after 48 h of when glucose was used as the carbon source.
incubation (Table S4). The amount of accumulated polymer The choice of an “ideal substrate” for biomass growth
from different carbon sources followed in the order of sucrose and PHA production significantly varies from one microor-
> glucose > glycerol > acetic acid, irrespective of the iden- ganism to another. For example, glucose has been shown to
tity of nitrogen sources. The difference in the mean biomass produce higher biomass and PHA yield for Bacillus megate-
accumulation between urea and N ­ H4Cl was not statistically rium, when compared to the other substrates (e.g., glycerol,
significant (p = 0.075) and can be attributed to random sam- sucrose, acetate) [41]. However, in the case of Ralstonia
pling variability (Table 3). Though carbon (p <0.001) and eutropha, such substrate selection makes only a nominal
nitrogen (p = 0.047) sources had a significant influence on difference [42]. Moreover, the preferred substrate for bio-
PHA production, multiple comparison tests revealed that mass growth may not necessarily be the best choice for

Fig. 5  1H NMR spectra of extracted PHA from Cupriavidus necator

13
 Biomass Conversion and Biorefinery

Table 6  PHA concentration and intracellular PHA content in pre- and post-optimization scenarios reported in the literature
Microbial strain Feedstock Optimized parameters PHA conc. PHA content Ref.
(g ­l-1) (% CDW)
Before After Before After

Thermus thermophilus Sodium gluconate Time, C/N ratio, and phosphate concentration 0.33 0.47 25.78 41.23 [53]
Cupriavidus necator Glucose Agitation speed, glucose, and urea concentra- 6.80 11.83 75.98 92.13 [63]
tion
Burkholderia cepacia Waste glycerol Aeration rate, agitation speed, and cultivation 1.18 3.60 51.98 63.94 [14]
period.
Ralstonia eutropha Fructose Carbon sources 1.40 6.75 43.08 50.37 [42]
Cupriavidus necator Crude glycerin Crude glycerin and nitrogen concentration 0.11 2.81 73.33 64.75 [46]
Bacillus cereus Molasses Molasses, ammonium sulfate, and initial pH 1.22 1.42 35.16 40.23 [64]
Burkholderia sp. Kraft lignin Incubation period, pH, and carbon concentra- 0.02 0.04 23.96 32.26 [65]
tion
Cupriavidus necator Crude glycerol Glycerol, ­(NH4)2SO4, and trace element con- 22.94 24.98 50.01 54.01 [66]
centration
Azohydromonas lata Sucrose Sucrose, urea, and trace elements concentration 3.85 5.95 47.53 48.57 [62]
Halomonas daqingensis Algal biodiesel waste residues Algal biodiesel waste residues, NaCl, and pH 0.21 0.24 68.71 65.19 [67]
Ralstonia eutropha Glucose Glucose, ammonium chloride, furfural, 2.10 2.10 60.00 61.76 [68]
hydroxymethylfurfural, and acetate
Bacillus aryabhattai Sweet sorghum juice Total sugar and urea concentration 2.30 2.43 35.49 39.07 [69]
*
Optimization refers to conditions for increasing PHA, if otherwise not stated

PHA production. For example, glucose and sucrose though volatile fatty acids) has been shown to accumulate about
resulted in equivalent biomass growth in Burkholderia 1.66 g l­-1 and 1.02 g l­-1 PHA [47, 48]. Under optimal
sacchari, the PHA production was at least one-fold higher culture conditions, PHAs accumulation by B. megaterium
with the former substrate than that in the latter one [43]. strain Ti3 has also been very low (0.61 g l­ -1) after 48 h
In this study, the low biomass growth and PHA production of incubation [49]. Exception to studies in fed-batch and
from acetic acid (when compared to carbohydrates) may continuous reactor, a high PHA yield is rather difficult in
be attributed to a change in intracellular pH and metabolic lab-scale batch experiments [50].
disturbance from the dissociation of the acid at a high con- The ANOVA analysis of the full quadratic model used
centration [44]. A reasonably good growth (6.843 ± 0.213 g to fit the biomass and PHA accumulation data from CCD
­l-1) and PHA production (4.440 ± 0.118 g ­l-1) were observed experiments suggested a significant lack of fit for both the
with C. necator when grown at moderate (1 g ­l-1) acetate responses (Table S5). Observations with unusually high
concentration [45]. studentized residuals, i.e., run no. 10 (for PHA) and run
no. 17 (for biomass), were dropped from step-wise regres-
3.2 Optimization of PHA production using sion analysis. Reduced quadratic model equations (in coded
a quadratic model form) after the elimination of insignificant terms (p >
0.05), for PHA and biomass, are shown in Eqs. (8) and (9),
The biomass and PHA concentration across all the twenty respectively.
experimental runs (Table 2) varied between 0.47–1.10 g YPHA = 1.553 + 0.182X1 + 0.179X2 + 0.245X3
­l-1 and 0.05–2.10 g ­l-1, respectively. The maximum bio-
mass and PHA production were observed in the experi- + 0.177X1 X3 + 0.397X2 X3 − 0.413X21 (8)
mental run no. 18 (urea: 1.05 g l­-1, phosphate: 2.05 g − 0.202X22 − 0.195X23

­l -1) and 19 (urea: 1.61 g l­-1, phosphate: 3.20 g l­-1) after

90 h and 72 h, respectively. Though biomass produc- YX = 0.848 + 0.199X1 + 0.041X2 + 0.020X3
tion was very similar, PHA accumulation was found to − 0.063X1 X3 + 0.0814X2 X3 − 0.025X21 (9)
be significantly different between the two experimental
− 0.019X22 − 0.023X23
conditions. A similar PHA concentration (2.81 g l­ -1) has
been reported from C. necator utilizing crude glycerol The fitness of the polynomial models for PHA produc-
and ammonium sulfate [46]. C. necator culture utilizing tion (R2 = 0.92, adj-R2 = 0.85) and biomass growth (R2 =
either defined (glucose) or complex media (waste-derived 0.98, adj-R2 = 0.97) suggested a good agreement between

13
Biomass Conversion and Biorefinery

experimental and predicted values. The model fitness and its optimal initialized weight and biases for the optimal ANN
coefficient (Table 4) were evaluated using the significance of architecture are shown in Table S6. The training of the
linear and nonlinear terms and their lack of fit assessment (α optimized network converges at 4 epochs (Fig. 3a), as net-
= 0.01). All the linear, interactions (except between time and work performance (MSE) gets stabilized. The performance
phosphate), and square terms were significant (p < 0.05) in of training (R2X: 0.98, R2P: 0.98), validation (R2X:0.98,
the PHA accumulation model (Eq. 8). R2P:0.96), testing (R2X:0.91, R2P:0.90), and entire data set
The coefficient (in absolute values) for the linear term remains satisfactory (Fig. 3b). The ANN model suggested a
of phosphate on PHA production (Eq. 8) is about 1.5 times maximum PHA yield of 2.69 g l­-1 at 62.80 h of incubation
higher than those of urea and incubation time, suggesting with initial nitrogen and phosphate concentrations 2.0 g ­l-1
its strong influence [51]. Negative coefficients of quadratic and 4.0 g ­l-1, respectively. Though the maximum PHA yield
terms for urea and phosphate suggest an inhibitory effect of predicted by the ANN model is higher than that of the poly-
those variables on PHA accumulation at a higher level. This nomial model, the optimal condition (i.e., time, nitrogen,
is in agreement with the nutrient-limiting condition required and phosphate concentration) for maximum PHA recovery
for PHA production [46, 52]. in both the models had been very similar.
The interaction effect of the predictors on biomass
growth remains three to four times smaller than in PHA 3.4 Multi‑objective optimization using polynomial
production. The negative coefficient of the square terms model and ANN
suggested that the impact of changing incubation time,
nitrogen, or phosphate concentration would get diminished Though PHA accumulation in microbial culture is closely
(low sensitivity) at their higher level. The significant influ- linked with the biomass growth, the relative proportion of
ence of time, nitrogen, and phosphate concentration on polymer to biomass can vary depending on culture condi-
cell growth and PHA accumulation is in line with other tions, initial biomass or substrate, or choice of harvesting
studies [53]. time. The economic benefit from improvement in PHA pro-
The quadratic model equations (Eqs. 8 and 9) represent the duction must be greater than the inflated budget for pre-
relationship between independent and dependent variables. treatment and downstream processing of increased amount
The importance of individual predictors and their optimal set- of biomass [16]. In this context, composite desirability can
tings can be estimated in order to maximize polymer produc- be adopted to co-optimize such contrasting objectives, giv-
tion or minimize biomass growth. Though the coefficient of ing appropriate weight to each of them [19].
the model equations are very specific to the problem under Considering the biomass and PHA production as the
investigation, the relative importance of the predictors and objectives, the individual and composite desirability func-
their nature of influence (synergistic/antagonistic) can be tions are used to find the optimal combination of process
extended to studies for cost-effective PHA production [54]. variables. The individual desirability of biomass decreases
The response surface and contour plots for PHA accu- (data not shown) with an increase in either incubation time,
mulation, drawn as a function of two independent variables urea, or phosphate concentration. However, the desirability
(holding the other at respective center-point, phosphate 2.05 of PHAs initially increases with time or substrate concentra-
g ­l-1, time 47 h, and urea 1.05 g l­-1), are shown in Fig. 2. tions, reaches a maxima, and then declines. Although opti-
At the lower end of incubation time and urea concentra- mizing individual responses provide a better result for each
tion, the PHA accumulation seems to be considerably low, response, the optimal parameter settings are different from
which increases until reaching maxima and then declines that of the composite desirability [55]. As polymer produc-
with further increase of the variables (Fig. 2a). An optimum tion is often growth-associated, high composite desirability
combination of independent variables for PHA accumulation in media optimization can be expected when the target is
can be traced to the clear peak in the response surface within set to maximize either or both of these responses [56]. High
the design boundary. According to Eq. (8), a maximum PHA desirability can be attained in optimization if PHA produc-
production (2.40 g ­l-1) can be expected after 61.87 h of incu- tion remains the only objective [57].
bation in the experimental run starting with 2 g ­l-1 of urea The assessment of composite desirability for PHA and
and 4 g ­l-1 phosphate. biomass production, while changing their relative impor-
tance in polynomial or ANN models, is shown in Table 5.
3.3 Artificial neural network–based model The first scenario focusing on the maximization of poly-
and optimization mer accumulation suggests the production of 2.40 g ­l-1 of
PHA along with 1.10 g l­ -1 biomass in the polynomial model.
The GA-based optimization projected six neurones in the However, ANN suggests a little higher PHA accumulation
hidden layer (Fig. S1) while minimizing composite normal- (2.69 g l­-1) with similar biomass production (1.01 g l­-1) on
ized error for both PHA and biomass in the dataset. The similar culture conditions as the polynomial model. This

13
 Biomass Conversion and Biorefinery

suggested that a higher yield of PHA with low biomass can process conditions [62] has been observed to be very effec-
be achieved, thereby decreasing the cost of downstream tive to increase the microbial PHA concentration (Table 6),
processing. Significantly lower polymer production (1.02- though proportionate improvement in polymer content in
2.11 g ­l-1) with a higher proportion of biomass (2.26-3.41 microbial biomass is not always evident. In some cases,
g ­l-1) has been reported in other studies, utilizing either the PHA content has been observed to decrease due to an
defined or complex media under optimized conditions [47, out-of-proportion increase in residual biomass when com-
48]. Increasing the importance for biomass minimization to pared to the polymer [46, 67]. Therefore, multi-objective
0.25, 0.50, or 0.75 projected a drop in biomass accumulation optimization with composite desirability is essential to
as well as in PHA production, for both the model variants. attain a compromised solution between the two conflict-
However, a relative decrease in PHA to that of the biomass ing aspects [53]. ANN models can be adapted to monitor
was found to be lower in the ANN model than that of the pilot-scale polymer production in real-time by integrating
polynomial regression. The polymer-to-biomass ratio con- them with on-site sensors while using the optimal process
tinued to increase for the ANN model. The culture condi- parameters as a reference. Biomass is usually pre-treated to
tions projected for reduced priority of biomass (weightage 0) facilitate polymer recovery through alkali, oxidants, heat,
are somewhat similar for both model variants. However, the and mechanical treatment [70] for digestion of non-PHA
predicted incubation time, nitrogen, and phosphate concen- cell mass, followed by extraction into an appropriate solvent
tration at higher desirability weight for biomass is higher in and reprecipitation. In the pilot scale, PHA-rich biomass
ANN. The combination of RSM and ANN model to predict is harvested through flocculation or simple gravity settling
PHA and biomass has reported superiority of ANN model in from which either wet biomass or lyophilized cells are used
comparison to polynomial regression [24, 58, 59]. for polymer extraction [71]. However, pre-treatment of bio-
mass used on a pilot or commercial scale may not be neces-
3.5 Characterization of the biopolymer sary, and PHA is extracted either with organic solvents (e.g.,
chloroform) or a mixture of organic and aqueous phase [71,
The FTIR spectroscopy was performed to investigate the dif- 72]. The conditions of extraction protocol can also affect the
ferent functional groups of PHA produced by C. necator. The polymer recovery where the optimal conditions for the lab
FTIR spectrum of PHA recorded between 4000 and 400 c­ m−1 scale might not be suitable for the pilot scale.
(Fig. 4) shows a sharp absorption band at 1730.15 ­cm−1 which Though the use of ANN and multi-objective optimiza-
corresponds to carbonyl (C=O) stretching of the ester [60]. tion are rarely reported in microbial production process of
The bands of minor relevance at 3446.79 c­ m−1 are related to PHAs, the results from both the approaches corroborate each
a terminal OH group [20]. The band at 1109.07 ­cm-1 is char- other in optimizing the process conditions as evident from
acteristic of C-O-C group stretching. Accompanying bands of the total phenolic content and total flavonoid content recov-
the C-O-C groups emerge in the spectral region from 1100 to ery in phytochemical extraction process [73]. The RSM and
1500 ­cm-1. The absorption band at 619.15 c­ m-1 was assigned ANN modeling approaches have been successfully adopted
to the C-C stretching vibration in the amorphous phase [61]. to minimize the hardness while maximizing the toughness of
1
H NMR analysis (Fig. 5) revealed four resonance an egg shell reinforced composite with multi-objective opti-
signals indicating the presence of different types of pro- mization using GA [74]. The ANN model developed from
tons. Triplet and doublet peaks at a chemical shift of the CCD matrix integrated with a multi-objective GA has
0.842–0.869 ppm and 1.275–1.296 ppm were identified as been used to predict optimal media and process condition to
terminal methyl (-CH3) group of 3HB unit. Multiplet peak enhance phycobiliproteins and cell biomass in cyanobacteria
at 2.447–2.659 ppm represented methylene (-CH2) group growth [75].
of C2 carbon atom while chemical shift at 5.236–5.301 In the present work, RSM and GA-ANN models provide
ppm attributed to methane (-CH) groups of 3HB and 3HV important insights into the effect of incubation time, urea,
monomer units. The results obtained were in agreement and phosphate concentration while co-optimizing biomass
with earlier reports [1, 20]. Along with crotonic acid assay, and PHA accumulation, which can be extended to select
both 1H NMR and FTIR spectra of the extracted polymer appropriate renewable resources. However, prior economic
confirm the PHB form from the production process. analysis of PHA recovery processes at the proposed multi-
objective solution needs to be validated. The improvement
3.6 Practical implication of the work and future in the upstream process while applying different polymer
research prospects recovery routes such as high pressure and heat has been
deemed beneficial in decreasing the production cost [76].
The improvements in biopolymer yield while minimizing The concept of partial withdrawal of culture medium and
biomass accumulation is an important pre-requisite for replenishing with a fresh one (referred to as cyclic fed-batch
sustained PHA production. The optimization of media and fermentation), based on the optimized growth condition,

13
Biomass Conversion and Biorefinery

can increase the PHA productivity [77]. Further, the adop- Declarations
tion of a bioprocess model-based optimal feeding strategy
can be invaluable to boost polymer accumulation in hetero- Ethical approval Not applicable
trophic biomass. The application of optimal feeding strategy Competing interests The authors declare no competing interest.
on an industrial-scale becomes essential as it helps opti-
mize the bioprocess conditions which would differ from
that of lab-scale experiments [78]. The optimal process
conditions evaluated for biomass growth and product for- References
mation in this laboratory study cannot be extrapolated to
1. Rao A, Haque S, El-Enshasy HA et al (2019) RSM–GA based
the industrial-scale unless the scale-up parameters relevant optimization of bacterial PHA production and in silico modula-
to the mass-transfer process are appropriately considered. tion of citrate synthase for enhancing PHA production. Biomol-
Additional validation studies may be required while con- ecules 9:872. https://​doi.​org/​10.​3390/​biom9​120872
sidering the bioreactor operations for polymer production 2. Bhattacharya S, Dubey S, Singh P et al (2016) Biodegradable
polymeric substances produced by a marine bacterium from a
on an industrial-scale. surplus stream of the biodiesel industry. Bioengineering 3:34.
https://​doi.​org/​10.​3390/​bioen​ginee​r ing3​040034
3. Singh AK, Srivastava JK, Chandel AK et al (2019) Biomedical
applications of microbially engineered polyhydroxyalkanoates:
4 Conclusions an insight into recent advances, bottlenecks, and solutions. Appl
Microbiol Biotechnol 103:2007–2032. https://​doi.​org/​10.​1007/​
The selection of appropriate carbon and nitrogen sources and s00253-​018-​09604-y
identifying their optimal concentrations are very important 4. United Nations (2022) The sustainable development Goals
report, pp 1–65
for an economical microbial biopolymer production process.
5. Gregson N, Crang M, Fuller S, Holmes H (2015) Interrogating
In this study, a combination of sucrose and urea offered bet- the circular economy: the moral economy of resource recovery
ter PHA production by Cupriavidus necator when compared in the EU. Econ Soc 44:218–243. https://​doi.​org/​10.​1080/​03085​
with other tested carbon and nitrogen sources. CCD-based 147.​2015.​10133​53
6. Gomes Gradíssimo D, Pereira Xavier L, Valadares Santos A
analysis suggested a significant influence of incubation time,
(2020) Cyanobacterial polyhydroxyalkanoates: a sustainable
urea, and phosphate concentration on the PHA production. alternative in circular economy. Molecules 25:4331. https://​
Though both the polynomial and GA-ANN models showed doi.​org/​10.​3390/​molec​ules2​51843​31
reasonably good agreement with experimental data, the pre- 7. Albuquerque MGE, Martino V, Pollet E et al (2011) Mixed cul-
ture polyhydroxyalkanoate (PHA) production from volatile fatty
dictability of the latter for PHA production was found to be
acid (VFA)-rich streams: effect of substrate composition and
superior. Multi-objective optimization for increasing PHA feeding regime on PHA productivity, composition and proper-
production and decreasing residual biomass predicted an ties. J Biotechnol 151:66–76. https://​doi.​org/​10.​1016/j.​jbiot​ec.​
increase in polymer to biomass ratio in the GA-ANN model. 2010.​10.​070
8. Guerra-Blanco P, Cortes O, Poznyak T et al (2018) Polyhydroxy-
The exploration of different initial biomass and substrate
alkanoates (PHA) production by photoheterotrophic microbial
concentrations, a detailed cost-benefit analysis, and valida- consortia: effect of culture conditions over microbial population
tion of the PHA production are highly recommended for the and biopolymer yield and composition. Eur Polym J 98:94–104.
process to be implemented in an industrial-scale. https://​doi.​org/​10.​1016/j.​eurpo​lymj.​2017.​11.​007
9. Bellini S, Tommasi T, Fino D (2022) Poly(3-hydroxybutyrate)
Supplementary Information The online version contains supplemen- biosynthesis by Cupriavidus necator: a review on waste substrates
tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 13399-0​ 23-0​ 4043-w. utilization for a circular economy approach. Bioresour Technol
Reports 17:100985. https://​doi.​org/​10.​1016/j.​biteb.​2022.​100985
Acknowledgements PL and BM acknowledge the support from the 10. Dalsasso RR, Pavan FA, Bordignon SE et al (2019) Polyhydroxy-
Karunya Institute of Technology and Sciences. SKB thanks VIT Vel- butyrate (PHB) production by Cupriavidus necator from sugar-
lore (India) for the staff-time support to collaborate with researchers cane vinasse and molasses as mixed substrate. Process Biochem
from the Karunya Institute of Technology and Sciences. Standard ethi- 85:12–18. https://​doi.​org/​10.​1016/j.​procb​io.​2019.​07.​007
cal and professional conduct has been followed. There are no conflicts 11. Sabapathy PC, Devaraj S, Meixner K et al (2020) Recent develop-
of interest (financial or otherwise) associated with this publication. ments in polyhydroxyalkanoates (PHAs) production – a review.
Bioresour Technol 306:123132. https://d​ oi.o​ rg/1​ 0.1​ 016/j.b​ iorte​ ch.​
Data availability Experimental data pertinent to this study can be avail- 2020.​123132
able from the corresponding author upon reasonable request. 12. Surendran A, Lakshmanan M, Chee JY et al (2020) Can poly-
hydroxyalkanoates be produced efficiently from waste plant and
Author contributions PL contributed to conceptualization-supporting, animal oils? Front Bioeng Biotechnol 8:169. https://​doi.​org/​10.​
formal analysis-equal, visualization-lead, writing—original draft-lead, 3389/​FBIOE.​2020.​00169/​BIBTEX
writing—review and editing-equal. BM contributed to conceptualiza- 13. Alsafadi D, Ibrahim MI, Alamry KA et al (2020) Utilizing the
tion-equal, project administration-lead, supervision-lead, writing— crop waste of date palm fruit to biosynthesize polyhydroxyal-
original draft-supporting, writing—review and editing-equal. SKB kanoate bioplastics with favorable properties. Sci Total Environ
contributed to editing-equal. 737:139716. https://​doi.​org/​10.​1016/j.​scito​tenv.​2020.​139716

13
 Biomass Conversion and Biorefinery

14. Mohd Zain NF, Paramasivam M, Tan JS et al (2021) Response 28. Haldar D, Shabbirahmed AM, Mahanty B (2023) Multivariate
surface methodology optimization of polyhydroxyalkanoate regression and artificial neural network modelling of sugar yields
production by Burkholderia cepacia BPT1213 using waste glyc- from acid pretreatment and enzymatic hydrolysis of lignocellu-
erol from palm oil-based biodiesel production. Biotechnol Prog losic biomass. Bioresour Technol 370:128519. https://​doi.​org/​10.​
37:e3077. https://​doi.​org/​10.​1002/​btpr.​3077 1016/j.​biort​ech.​2022.​128519
15. Sen KY, Baidurah S (2021) Renewable biomass feedstocks for 29. Liu L, Moayedi H, Rashid ASA et al (2020) Optimizing an ANN
production of sustainable biodegradable polymer. Curr Opin model with genetic algorithm (GA) predicting load-settlement
Green Sustain Chem 27:100412. https://​doi.​org/​10.​1016/j.​cogsc.​ behaviours of eco-friendly raft-pile foundation (ERP) system. Eng
2020.​100412 Comput 36:421–433. https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 00366-0​ 19-0​ 0767-4
16. Yang Y-H, Brigham CJ, Budde CF et al (2010) Optimization 30. Fourati M, Smaoui S, Ennouri K et al (2019) Multiresponse
of growth media components for polyhydroxyalkanoate (PHA) optimization of pomegranate peel extraction by statistical versus
production from organic acids by Ralstonia eutropha. Appl artificial intelligence: predictive approach for foodborne bacterial
Microbiol Biotechnol 87:2037–2045. https://​doi.​org/​10.​1007/​ pathogen inactivation. Evidence-Based Complement Altern Med
s00253-​010-​2699-8 2019:1–18. https://​doi.​org/​10.​1155/​2019/​15426​15
17. Ganesh Saratale R, Cho S-K, Dattatraya Saratale G et al (2021) 31. Jensen WA (2008) DOE simplified: practical tools for effective
A comprehensive overview and recent advances on polyhy- experimentation, second edition. J Qual Technol 40 124–125.
droxyalkanoates (PHA) production using various organic waste https://​doi.​org/​10.​1080/​00224​065.​2008.​11917​717
streams. Bioresour Technol 325:124685. https://​d oi.​o rg/​1 0.​ 32. Erzurum Cicek ZI, Kamisli Ozturk Z (2021) Optimizing the
1016/j.​biort​ech.​2021.​124685 artificial neural network parameters using a biased random key
18. Sawall M, von Harbou E, Moog A et al (2018) Multi-objec- genetic algorithm for time series forecasting. Appl Soft Comput
tive optimization for an automated and simultaneous phase 102:107091. https://​doi.​org/​10.​1016/j.​asoc.​2021.​107091
and baseline correction of NMR spectral data. J Magn Reson 33. Gonçalves JF, Resende MGC (2011) Biased random-key genetic
289:132–141. https://​doi.​org/​10.​1016/j.​jmr.​2018.​02.​012 algorithms for combinatorial optimization. J Heuristics 17:487–
19. Indumathi S, Diwakar RV, Krishnaiah G (2014) Optimization 525. https://​doi.​org/​10.​1007/​s10732-​010-​9143-1
of machining parameters using desirability function analysis 34. Darvishi H, Farhudi Z, Behroozi-Khazaei N (2020) Multi-objec-
and ANOVA for thermo-mechanical form drilling. Int J Ind Eng tive optimization of savory leaves drying in continuous infrared-
Technol 4:19–26 hot air dryer by response surface methodology and desirability
20. Trakunjae C, Boondaeng A, Apiwatanapiwat W et al (2021) function. Comput Electron Agric 168:105112. https://​doi.​org/​10.​
Enhanced polyhydroxybutyrate (PHB) production by newly 1016/j.​compag.​2019.​105112
isolated rare actinomycetes Rhodococcus sp. strain BSRT1-1 35. Yetilmezsoy K (2012) Integration of kinetic modeling and desira-
using response surface methodology. Sci Rep 11:1–14. https://​ bility function approach for multi-objective optimization of UASB
doi.​org/​10.​1038/​s41598-​021-​81386-2 reactor treating poultry manure wastewater. Bioresour Technol
21. Vinayaka AS, Mahanty B, Rene ER, Behera SK (2021) Bio- 118:89–101. https://​doi.​org/​10.​1016/j.​biort​ech.​2012.​05.​088
diesel production by transesterification of a mixture of pon- 36. Jain A, Jain R, Jain S (2020) Quantitative analysis of reducing
gamia and neem oils. Biofuels 12:187–195. https://​doi.​org/​10.​ sugars by 3, 5-dinitrosalicylic acid (DNSA method). Humana,
1080/​17597​269.​2018.​14648​74 New York, NY, pp 181–183
22. Arumugam A, Senthamizhan SG, Ponnusami V, Sudalai S 37. Langenfeld NJ, Payne LE, Bugbee B (2021) Colorimetric determi-
(2018) Production and optimization of polyhydroxyalkanoates nation of urea using diacetyl monoxime with strong acids. PLoS
from non-edible Calophyllum inophyllum oil using Cupriavidus One 16:e0259760. https://​doi.​org/​10.​1371/​journ​al.​pone.​02597​60
necator. Int J Biol Macromol 112:598–607. https://​doi.​org/​10.​ 38. Habibah N, Dhyanaputri IGAS, Karta IW et al (2018) A simple
1016/j.​ijbio​mac.​2018.​02.​012 spectrophotometric method for the quantitative analysis of phos-
23. Desai KM, Survase SA, Saudagar PS et al (2008) Compari- phate in the water samples. JST (Jurnal Sains dan Teknol) 7:198.
son of artificial neural network (ANN) and response surface https://​doi.​org/​10.​23887/​jst-​undik​sha.​v7i2.​13940
methodology (RSM) in fermentation media optimization: case 39. Pradhan S, Dikshit PK, Moholkar VS (2018) Production, ultra-
study of fermentative production of scleroglucan. Biochem Eng sonic extraction, and characterization of poly (3-hydroxybutyrate)
J 41:266–273. https://​doi.​org/​10.​1016/j.​bej.​2008.​05.​009 (PHB) using Bacillus megaterium and Cupriavidus necator. Polym
24. Imandi SB, Karanam SK, Nagumantri R et al (2022) Neural Adv Technol 29:2392–2400. https://​doi.​org/​10.​1002/​pat.​4351
networks and genetic algorithm as robust optimization tools for 40. Cristea A, Baricz A, Leopold N et al (2018) Polyhydroxybutyrate
modeling the microbial production of poly-β-hydroxybutyrate production by an extremely halotolerant Halomonas elongata
(PHB) from Brewers’ spent grain. Biotechnol Appl Biochem. strain isolated from the hypersaline meromictic Fără Fund Lake
https://​doi.​org/​10.​1002/​bab.​2412 (Transylvanian Basin, Romania). J Appl Microbiol 125:1343–
25. Nalini S, Inbakandan D, Stalin Dhas T, Sathiyamurthi S (2021) 1357. https://​doi.​org/​10.​1111/​jam.​14029
Optimization of biosurfactant production by marine Strepto- 41. Mohanrasu K, Rao RGR, Dinesh GH et al (2020) Optimization
myces youssoufiensis SNSAA03: a comparative study of RSM of media components and culture conditions for polyhydroxyal-
and ANN approach. Results Chem 3:100223. https://​doi.​org/​10.​ kanoates production by Bacillus megaterium. Fuel 271:117522.
1016/j.​rechem.​2021.​100223 https://​doi.​org/​10.​1016/j.​fuel.​2020.​117522
26. Rayavarapu B, Tallapragada P, Ms U (2021) Optimization 42. Khanna S, Srivastava AK (2005) Statistical media optimization
and comparison of ℽ-aminobutyric acid (GABA) produc- studies for growth and PHB production by Ralstonia eutropha.
tion by LAB in soymilk using RSM and ANN models. Beni- Process Biochem 40:2173–2182. https://​doi.​org/​10.​1016/j.​procb​
Suef Univ J Basic Appl Sci 10:14. https://​d oi.​o rg/​1 0.​1 186/​ io.​2004.​08.​011
s43088-​021-​00100-3 43. Nascimento VM, Silva LF, Gomez JGC, Fonseca GG (2016)
27. Olalere OA, Abdurahman NH, R bin M Y, Alara OR (2019) Multi- Growth of Burkholderia sacchari LFM 101 cultivated in glucose,
response optimization and neural network modeling for parameter sucrose and glycerol at different temperatures. Sci Agric 73:429–
precision in heat reflux extraction of spice oleoresins from two 433. https://​doi.​org/​10.​1590/​0103-​9016-​2015-​0196
pepper cultivars (Piper nigrum). J King Saud Univ - Sci 31:789– 44. Wang J, Yue Z-B, Sheng G-P, Yu H-Q (2010) Kinetic analysis on
797. https://​doi.​org/​10.​1016/j.​jksus.​2017.​09.​010 the production of polyhydroxyalkanoates from volatile fatty acids

13
Biomass Conversion and Biorefinery

by Cupriavidus necator with a consideration of substrate inhibi- 59. Chester Jules A. Tantoco, Princess J. Requiso CGA, Jewel A.
tion, cell growth, maintenance, and product formation. Biochem Capunitan, Fidel Rey P. Nayve Jr. and J-RSV (2023) Response
Eng J 49:422–428. https://​doi.​org/​10.​1016/j.​bej.​2010.​02.​005 surface methodology and artificial neural network optimization
45. Marudkla J, Lee W-C, Wannawilai S et al (2018) Model of acetic and modeling of the saccharification and fermentation condi-
acid-affected growth and poly(3-hydroxybutyrate) production by tions of the polyhydroxybutyrate from corn stover. Philipp J Sci
Cupriavidus necator DSM 545. J Biotechnol 268:12–20. https://​ 152:357–374
doi.​org/​10.​1016/j.​jbiot​ec.​2018.​01.​004 60. Geethu M, Vrundha R, Raja S et al (2019) Improvement of the pro-
46. Campos MI, Figueiredo TVB, Sousa LS, Druzian JI (2014) The duction and characterisation of polyhydroxyalkanoate by Bacillus
influence of crude glycerin and nitrogen concentrations on the endophyticus using inexpensive carbon feedstock. J Polym Envi-
production of PHA by Cupriavidus necator using a response ron 27:917–928. https://​doi.​org/​10.​1007/​s10924-​019-​01397-z
surface methodology and its characterizations. Ind Crops Prod 61. R SK (2012) FTIR analysis of polyhydroxyalkanoates by novel
52:338–346. https://​doi.​org/​10.​1016/j.​indcr​op.​2013.​11.​008 Bacillus sp. AS 3-2 from soil of Kadi region, North Gujarat, India.
47. Lee SM, Lee H-J, Kim SH et al (2021) Screening of the strictly J Biochem Tech 3:380–383
xylose-utilizing Bacillus sp. SM01 for polyhydroxybutyrate 62. Zafar M, Kumar SS, Kumar SS, Dhiman AK (2012) Optimi-
and its co-culture with Cupriavidus necator NCIMB 11599 for zation of polyhydroxybutyrate (PHB) production by Azohyd-
enhanced production of PHB. Int J Biol Macromol 181:410–417. romonas lata MTCC 2311 by using genetic algorithm based
https://​doi.​org/​10.​1016/j.​ijbio​mac.​2021.​03.​149 on artificial neural network and response surface methodology.
48. Vu DH, Mahboubi A, Root A et al (2022) Thorough investiga- Biocatal Agric Biotechnol 1:70–79. https://​doi.​org/​10.​1016/j.​
tion of the effects of cultivation factors on polyhydroalkanoates bcab.​2011.​08.​012
(PHAs) production by Cupriavidus necator from food waste- 63. Biglari N, Ganjali Dashti M, Abdeshahian P et al (2018) Enhance-
derived volatile fatty acids. Fermentation 8:605. https://​doi.​org/​ ment of bioplastic polyhydroxybutyrate P(3HB) production from
10.​3390/​ferme​ntati​on811​0605 glucose by newly engineered strain Cupriavidus necator NSDG-
49. Israni N, Shivakumar S (2020) Polyhydroxyalkanoate (PHA) bio- GG using response surface methodology. 3 Biotech 8:330. https://​
synthesis from directly valorized ragi husk and sesame oil cake doi.​org/​10.​1007/​s13205-​018-​1351-7
by Bacillus megaterium strain Ti3: Statistical optimization and 64. Evangeline S, Sridharan TB (2019) Biosynthesis and statistical
characterization. Int J Biol Macromol 148:20–30. https://​doi.​org/​ optimization of polyhydroxyalkanoate (PHA) produced by Bacil-
10.​1016/j.​ijbio​mac.​2020.​01.​082 lus cereus VIT-SSR1 and fabrication of biopolymer films for sus-
50. Blunt W, Levin D, Cicek N (2018) Bioreactor operating strategies tained drug release. Int J Biol Macromol 135:945–958. https://d​ oi.​
for improved polyhydroxyalkanoate (PHA) productivity. Polymers org/​10.​1016/j.​ijbio​mac.​2019.​05.​163
(Basel) 10:1197. https://​doi.​org/​10.​3390/​polym​10111​197 65. Morya R, Sharma A, Kumar M et al (2021) Polyhydroxyalkanoate
51. Bhatia SK, Kim J-HHJ, Kim M-SS et al (2018) Production of synthesis and characterization: a proteogenomic and process optimi-
(3-hydroxybutyrate-co-3-hydroxyhexanoate) copolymer from zation study for biovalorization of industrial lignin. Bioresour Tech-
coffee waste oil using engineered Ralstonia eutropha. Bio- nol 320:124439. https://​doi.​org/​10.​1016/j.​biort​ech.​2020.​124439
process Biosyst Eng 41:229–235. https:// ​ d oi. ​ o rg/ ​ 1 0. ​ 1 007/​ 66. Sangkharak K (2011) Utilization of biodiesel waste as a feedstock
s00449-​017-​1861-4 for the production of polyhydroxybutyrate by Cupriavidus necator.
52. Sathiyanarayanan G, Kiran GS, Selvin J, Saibaba G (2013) Opti- African J Biotechnol 10. https://​doi.​org/​10.​5897/​AJB11.​2184
mization of polyhydroxybutyrate production by marine Bacillus 67. Dubey S, Mishra S (2021) Efficient production of polyhydroxyal-
megaterium MSBN04 under solid state culture. Int J Biol Macro- kanoate through halophilic bacteria utilizing algal biodiesel waste
mol 60:253–261. https://​doi.​org/​10.​1016/j.​ijbio​mac.​2013.​05.​031 residue. Front Bioeng Biotechnol 9. https://d​ oi.o​ rg/1​ 0.3​ 389/f​ bioe.​
53. Papaneophytou CP, Kyriakidis DA (2012) Optimization of poly- 2021.​624859
hydroxyalkanoates production from Thermus thermophilus HB8 68. Bhatia SK, Gurav R, Choi T-R et al (2019) Bioconversion of plant
using response surface methodology. J Polym Environ 20:760– biomass hydrolysate into bioplastic (polyhydroxyalkanoates)
773. https://​doi.​org/​10.​1007/​s10924-​012-​0451-4 using Ralstonia eutropha 5119. Bioresour Technol 271:306–315.
54. Das R, Pal A, Paul A (2022) Enhanced production of poly(3- https://​doi.​org/​10.​1016/j.​biort​ech.​2018.​09.​122
hydroxybutyrate-co-3-hydroxyvalerate) copolymer by endophytic 69. Tanamool V, Imai T, Danvirutai P, Kaewkannetra P (2013) An
Bacillus cereus RCL 02 utilizing sugarcane molasses as sole alternative approach to the fermentation of sweet sorghum juice
source of carbon: a statistical optimization approach. BioTechno- into biopolymer of poly-β-hydroxyalkanoates (PHAs) by newly
logia 103:283–300. https://​doi.​org/​10.​5114/​bta.​2022.​118671 isolated, Bacillus aryabhattai PKV01. Biotechnol Bioprocess Eng
55. Mobin M, Mousavi SM, Komaki M, Tavana M (2018) A hybrid 18:65–74. https://​doi.​org/​10.​1007/​s12257-​012-​0315-8
desirability function approach for tuning parameters in evolution- 70. De Donno NL, Moreno Sayavedra S, Rene ER (2021) Polyhy-
ary optimization algorithms. Measurement 114:417–427. https://​ droxyalkanoate (PHA) production via resource recovery from
doi.​org/​10.​1016/j.​measu​rement.​2017.​10.​009 industrial waste streams: a review of techniques and perspectives.
56. Kumar M, Gupta A, Thakur IS (2016) Carbon dioxide seques- Bioresour Technol 331:124985. https://d​ oi.o​ rg/1​ 0.1​ 016/j.b​ iorte​ ch.​
tration by chemolithotrophic oleaginous bacteria for production 2021.​124985
and optimization of polyhydroxyalkanoate. Bioresour Technol 71. Lorini L, Martinelli A, Capuani G et al (2021) Characterization
213:249–256. https://​doi.​org/​10.​1016/j.​biort​ech.​2016.​02.​038 of polyhydroxyalkanoates produced at pilot scale from different
57. Israni N, Venkatachalam P, Gajaraj B et al (2020) Whey valoriza- organic wastes. Front Bioeng Biotechnol 9. https://​doi.​org/​10.​
tion for sustainable polyhydroxyalkanoate production by Bacillus 3389/​fbioe.​2021.​628719
megaterium: production, characterization and in vitro biocompat- 72. Valentino F, Moretto G, Lorini L et al (2019) Pilot-scale polyhy-
ibility evaluation. J Environ Manage 255:109884. https://​doi.​org/​ droxyalkanoate production from combined treatment of organic
10.​1016/j.​jenvm​an.​2019.​109884 fraction of municipal solid waste and sewage sludge. Ind Eng Chem
58. Bose SA, Rajulapati SB, Velmurugan S et al (2023) Process Res 58:12149–12158. https://​doi.​org/​10.​1021/​acs.​iecr.​9b018​31
intensification of biopolymer polyhydroxybutyrate production by 73. Oke EO, Adeyi O, Okolo BI et al (2020) Phenolic compound
pseudomonas putida SS9: a statistical approach. Chemosphere extraction from Nigerian Azadirachta Indica leaves: response
313:137350. https://​doi.​org/​10.​1016/j.​chemo​sphere.​2022.​137350 surface and neuro-fuzzy modelling performance evaluation

13
 Biomass Conversion and Biorefinery

with Cuckoo Search multi-objective optimization. Results Eng enhanced polyhydroxyalkanoate production by Bacillus thuring-
8:100160. https://​doi.​org/​10.​1016/j.​rineng.​2020.​100160 iensis using a glucose-rich hydrolyzate. Bioresour Bioprocess
74. Nwobi-Okoye CC, Uzochukwu CU (2020) RSM and ANN mod- 8:11. https://​doi.​org/​10.​1186/​s40643-​021-​00361-x
eling for production of Al 6351/ egg shell reinforced composite: multi 78. Magdouli S, Saffar T, Guedri T et al (2018) Practical aspects and
objective optimization using genetic algorithm. Mater Today Com- case studies of industrial scale fermentation. In: Microbial Sens-
mun 22:100674. https://​doi.​org/​10.​1016/j.​mtcomm.​2019.​100674 ing in Fermentation. John Wiley & Sons, Ltd, Chichester, UK, pp
75. Saini DK, Rai A, Devi A et al (2021) A multi-objective hybrid 267–298
machine learning approach-based optimization for enhanced bio-
mass and bioactive phycobiliproteins production in Nostoc sp. Publisher’s Note Springer Nature remains neutral with regard to
CCC-403. Bioresour Technol 329:124908. https://​doi.​org/​10.​ jurisdictional claims in published maps and institutional affiliations.
1016/j.​biort​ech.​2021.​124908
76. Pavan FA, Junqueira TL, Watanabe MDB et al (2019) Economic Springer Nature or its licensor (e.g. a society or other partner) holds
analysis of polyhydroxybutyrate production by Cupriavidus neca- exclusive rights to this article under a publishing agreement with the
tor using different routes for product recovery. Biochem Eng J author(s) or other rightsholder(s); author self-archiving of the accepted
146:97–104. https://​doi.​org/​10.​1016/j.​bej.​2019.​03.​009 manuscript version of this article is solely governed by the terms of
77. Singh S, Sithole B, Lekha P et al (2021) Optimization of cul- such publishing agreement and applicable law.
tivation medium and cyclic fed-batch fermentation strategy for

13