| Hyaluronic acid (HA) is a linear glycosaminoglycan consisting of repeating disaccharide units. HA is one of the major components of the extracellular matrix. The high molecular weight of HA together with its special viscoelastic features and biological functions have made HA as an attractive material to prepare biocompatible devices with applications in drug delivery and tissue engineering. Presently it has reached prominence in cosmetic practice where it is now the injectable dermal filler of choice for most surgeons. Molecular weight and content are two important fundamental parameters characterizing HA. HA of different molecular weight displays different physical and chemical properties which decide its application fields. Content is an important index in HA fermentation process. Determination of HA content in fermentation broth is not only one of the necessary steps when screening HA high-yielding bacterial strains but also a routine measurement during fermentation which is significant for process optimization and control. Various methods can determine these two parameters now. But all those traditional methods are time-consuming and reagent consuming. It is not useful for fast release in laboratory and on-line inspection during fermentation process.Under this context, this paper chooses HA solid powder and fermentation broth as research object. HA molecular model and HA content model in fermentation broth were established to achieve rapid analysis. The details are listed below:Firstly, due to the lack of theoretical foundation of molecular weight determination by NIRS technology and the low HA content in fermentation, the feasibility of establishment of HA molecular weight model and low HA content model were done. The overall results sufficiently demonstrate the feasibility of the method.Secondly, to acquire more homogeneous molecular weight gradient, we developed samples of different molecular weight by heating degradation. The diffuse reflectance spectra of samples collected were associated with the molecular weight data which is determined using intrinsic viscosity by PLS algorithm to establish the molecular weight model. During modeling process, the selections of optimal pretreatment method and optimal spectral region by correlation coefficient method were made. With the purpose of comprehensive consideration of the correlation coefficient and root mean square error, we introduce the concept of objective function into model estimation. And finally, MSC was used as the best pretreatment method. After leave-one-out cross-validation, the correlation coefficient of calibration set (Rc) between the measured and predicted values was 0.9847 and RMSECV was 85.06 with the selected ranges of wavelength (1452-1542nm,1658-1848nm,1952-2048nm and 2218-2300nm). Predict the samples of validation set by the established model and obtain that the RMSEP was 66.35 and Rp was 0.9946 after the calculation. Repeatability tests were performed by repeated measurement of spectra. The result indicated that the repeatability standard deviations belonged to the same population and the biases were all within the range of permissible error in factories. The effect of different primary analysis method and modeling algorithm were investigated. The result proved that both the statistical significance of molecular weight and the accuracy of primary analysis method had effects on model. The viscometric-average molecular weight and weight-average molecular weight were both suitable for establishing the linear models.Thirdly, the transmission spectra of fermentation broth under the same fermentation conditions were associated with the content data which is determined using carbazole and sulfuric acid spectrophotometric method by PLS algorithm to establish the HA content model. During modeling process, the selections of optimal pretreatment method and optimal spectral region by the correlation coefficient method were done. With the purpose of comprehensive consideration of the correlation coefficient and root mean square error, we introduce the concept of objective function into model estimation. And finally, first derivative with Norris 5 smoothing points was used as the best pretreatment method. After leave-one-out cross-validation, the correlation coefficient of calibration set (Rc) between the measured and predicted values was 0.9951 and RMSECV was 0.223 with the selected ranges of wavenumber (7312.7-10000cm-1). Predict the samples of validation set by the established model and obtained that the RMSEP was 0.376 and Rp was 0.9809 after the calculation. The consistency of submodel was investigated and the result proved that the optimal results can still be achieved by different division of sample sets.Finally, the process design for applying NIRS technology to fermentation process was done. But due to the time restriction, it is a long way to go for process control of HA by NIRS technology. |
PDF Full Text Request