Rapid quantification of pullulan in fermentation broth using UV-visible spectroscopy and partial least squares regression

Quantification of exopolysaccharide (EPS) production in fermentation broth requires solvent precipitation of the polymer, followed by acid or enzymatic hydrolysis, and colorimetric or chromatographic analysis. This lengthy multistep sample preparation and analysis is a major bottleneck in bioprocess monitoring. The development of a nondestructive analytical method requiring minimal sample preparation is warranted. In this study, partial least squares (PLS) regression models were developed to quantify pullulan in cell-free supernatant (PCS) and precipitated pullulan redissolved in distilled water (PDW) from spectral data (204-400 nm). Genetic algorithm, particle swarm optimization, competitive adaptive reweighted sampling, and adaptive bottom-up space exploration strategies were employed to select optimal spectral regions. The full-spectrum model on the PCS (5 latent variables, RMSE_CV: 0.020 g l⁻¹, R_CV²: 0.997) outperformed the PDW (3 latent variables, R_CV²: 0.990). Adaptive bottom-up space exploration achieved the lowest RMSE_CV (0.009 g l⁻¹ for the PCS, 0.027 g l⁻¹ for the PDW), retaining just 16 and 21 spectral variables, respectively. The residual predictive deviation (RPD) for all PLS model variants remains satisfactory (>6.559). The method's limit of detection (0.021 g l⁻¹) was suitable for quantifying pullulan in fermentation broth. The proposed method can be extended to other structurally similar biopolymers where PLS-based soft sensor integration would enable real-time monitoring and bioprocess control.