2.2. Optimizing BioLector fermentation screening through central
composite design of response surface methodology
The DoE packagefacilitated the generation of 30 experiments using the
RSM to optimize carefully selected model terms, encompassing
MgSO4, glycerol, glucose, and certain LB broth
concentration (referred as TB). The experimental runs were conducted
utilizing the BioLector microfermenter in a 1 mL screening culture that
maintained a constant concentration of 10 mM KCI, 15 mM
MgCI2, 15 mM KH2PO4, 10
mM thiamine, and 50 g/L 2X LB Miller media (Supplementary Table. 2 Fig.
5), complemented by 0.25 mM IPTG tailored to each specific design point.
Once conducting analyses using various models, the quadratic model
emerged as a suitable candidate for predicting and validating optimal
conditions. Notably, the model exhibited a significant p-value of
0.0037, thereby providing a reliable prediction framework. Additionally,
the model’s lack of fit displayed insignificance at 0.4479 relative to
the pure error, suggesting the error’s immateriality in influencing the
accuracy of the proposed model (Table 2). The model’s R2 value of 0.6866
indicated that it reasonably fits the experimental data. Furthermore,
the adequate precision value of 7.861 indicates the sufficiency of the
signal for the biological system under scrutiny. The perturbation
analysis, factorial contour plots and 3D response surface of the central
composite design was operated to gain a deeper understanding of the
impact of independent variables’ major effects and interactions on
insulin yield (Fig. 6, Fig. 7, and Fig. 8). Perturbation of CCD analysis
specifically clarified the primary marks of variables A, B, C, and D, on
insulin yield, as depicted in Fig. 6. The two-dimensional contour and
three-dimensional response surface plot effectively portrayed the
intricate relationship between the independent variables and the ensuing
insulin yield (Fig. 7, Fig.8). These independent variables were
deliberately manipulated across a predefined range to systematically
investigate their combined influence on insulin production. The contour
plots (Fig.7), in particular, offer invaluable insights into the
inherent characteristics and magnitude of the interactions among various
contributing factors. Notably, when glycerol levels were elevated, there
was a substantial reduction in the resultant insulin yield, plummeting
from 12.6259 mg/ml to 7.25534 mg/ml. Likewise, the simultaneous increase
of glycerol and media content to elevated levels (exceeding 30 g/l)
resulted in a decrease in insulin yield from 11.124 mg/ml to 6.84772
mg/ml. Conversely, the presence or absence of MgSO4exhibited no discernible impact on the insulin yield of significance.
The investigation of the interplay between the dependent and independent
variables was further elucidated by implementing 3D response surface
plots (Fig. 8). These plots effectively affirmed the intricate and
interactive influence of specific glucose, glycerol,
MgSO4 concentrations, and media content on insulin
yield. Notably, when considering the actual factors of LB (30 g/l) and
MgSO4 (5mM), it was regarded that a lower concentration
of glycerol and the highest concentration of glucose displayed a
positive correlation with the production of insulin at higher levels
compared to the other variables. Consequently, based on these findings,
four distinct scenarios were statistically derived and subsequently
subjected to experimental validation. The experimentation was achieved
using the BioLector micro-fermenter, with each scenario being replicated
three times to ensure the reliability and accuracy of the results (Table
3). Contempt the absence of substantial discrepancies among the
scenarios, it is noteworthy that scenario III, characterized by a
glucose concentration of 8.78 mM, glycerol concentration of 10 g/L, LB
of 30 g/l, and MgSO4 concentration of 15 mM, exhibits
the slightest deviation and most heightened insulin product in
comparison to the additional scenarios (Table 3).