Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

In-line spectroscopy combined with multivariate analysis methods for endpoint determination in column chromatographic adsorption processes for herbal medicine

In-line spectroscopy combined with multivariate analysis methods for endpoint determination in... Objective: In a chromatographic cycle, the adsorption process is a critical unit operation that has a significant impact on downstream processes and, ultimately, the quality of the final products. The development of a rapid method to determine the endpoints of adsorption processes in a large-scale manufacturing is of substantial importance for herbal medicine (HM) manufacturers. Methods: In this study, the adsorption of saponins on a macroporous resin column chromatograph, a critical unit operation in Panax notoginseng (Burkill) F.H.Chen injection manufacturing, was considered as an example. The evaluation results of in-line ultraviolet and visible spectra combined with various multivariate analysis methods, including the moving block standard deviation (MBSD), difference between the moving block average and the target spectrum (DMBA-TS), soft-independent modeling of class analogy (SIMCA), and partial least-squares discriminant analysis (PLS-DA), were compared. Results: MBSD was unsuitable for adsorption processes. The relative standard errors of prediction between the predicted and experimental endpoints were 13.2%, 4.67%, and 5.71% using DMBA-TS, SIMCA, and PLS-DA, respectively. Conclusions: Among the considered analysis methods, SIMCA and PLS-DA were more effective for endpoint determination. The results of this study provide a more comprehensive overview of the effectiveness of various multivariate analysis methods to facilitate the selection of the most suitable method. This study was also conducive to address the issues of the in-line detection of adsorption endpoints to guide practical HM manufacturing. Keywords: Adsorption, Endpoint, Herbal medicine, Multivariate analysis, Ultraviolet and visible Graphical abstract: http://links.lww.com/AHM/A23 Introduction plays an important role in improving the safety and effi- cacy of the final products. It has been used to separate Column chromatography has a wide range of applica- several types of bioactive substances from medicinal tions in many fields, including herbal medicine (HM) herbs, including saponins, polyphenols, flavonoids, and manufacturing, biological medicine manufacturing, and [1] alkaloids . Each chromatographic cycle typically con- wastewater treatment. During the manufacturing of HM, sists of adsorption, washing, and elution. The adsorp- especially botanical injections, column chromatography tion process is a critical unit operation and may have a significant impact on the downstream processes and quality of the final products. Insufficient adsorption time College of Pharmaceutical Sciences, Zhejiang University, Hangzhou, China; Zhejiang Provincial Key Laboratory of Cancer Prevention and may result in a low efficiency of adsorbent utilization, Treatment Technology of Integrated Traditional Chinese and Western whereas excessive adsorption time may result in high Medicine, Zhejiang Academy of Traditional Chinese Medicine, Tongde variability in the final product quality and low efficiency Hospital of Zhejiang Province, Hangzhou, China; Innovation Center [2] of solute adsorption . The determination of the optimal in Zhejiang University, State Key Laboratory of Component-Based adsorption endpoint is of considerable importance in a Chinese Medicine, Hangzhou, China large-scale manufacturing. * Corresponding author. Haibin Qu, College of Pharmaceutical Breakthrough time, which is defined as the time Sciences, Zhejiang University, Hangzhou 310058, China, E-mail: required for a certain number of active components quhb@zju.edu.cn. to be detected in the effluent, is generally considered a Copyright © 2022 Tianjin University of Traditional Chinese Medicine. [3] suitable endpoint . However, several factors may influ- This is an open-access article distributed under the terms of the ence the breakthrough time, such as variations in load- Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download ing solutions, fluctuations in operating parameters, and and share the work provided it is properly cited. The work cannot be variations in the adsorption capacity of the adsorbents. changed in any way or used commercially without permission from the Consequently, traditional HM manufacturing, which is journal. conducted with a fixed adsorption time as the endpoint, Acupuncture and Herbal Medicine (2022) 2:4 cannot adapt to changes in the above-mentioned factors. Received 15 January 2022 / Accepted 11 July 2022 The breakthrough time can be determined by an offline http://dx.doi.org/10.1097/HM9.0000000000000035 assessment of the active components in the effluents via 253 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Materials and Methods chromatographic techniques. However, the long duration required for analysis may ultimately delay critical man- Samples, chemicals, and reagents ufacturing decisions. Therefore, the development of a The medicinal herb Panax notoginseng (Burkill) F.H.Chen rapid method to determine the endpoints of adsorption was purchased from the Wenshan Panax Notoginseng processes is of critical importance for HM manufacturers. International Trade Center (Yunnan, China). The stan- Spectroscopy is an effective choice for process anal- dard compounds notoginsenoside R , ginsenoside ysis because it requires no sample preparation and can Rg , ginsenoside Re, ginsenoside Rb , and ginseno- provide real-time information about the processes. Thus 1 1 side Rd were purchased from Ronghe Pharmaceutical far, several studies have used spectroscopic techniques Technology Co., Ltd. (Shanghai, China) with purities for endpoint determination in pharmaceutical processes, of over 99%. D-101 macroporous resin was purchased including blending, fluidized drying, fluidized bed gran- from Haiguang Chemical Co., Ltd. (Tianjin, China). [4-9] ulation, coating, extraction, and chemical synthesis . HPLC-grade acetonitrile was purchased from Merck Spectroscopic techniques have also been reported to (Darmstadt, Germany). Ultrapurified water was pre- monitor the column chromatographic elution processes [10] pared using a Milli-Q academic water purification sys- of Coptis chinensis Franch and Salvia miltiorrhiza tem (Millipore, Billerica, MA, USA). [11] Bunge . Ultraviolet and visible (UV-VIS) spectroscopy is a form of electron spectroscopy that is based on the exci- [12] tation of electrons . It possesses the advantages of high Process description sensitivity and inexpensive instrumentation. However, the performance of in-line UV-VIS spectroscopy in the In this study, seven experiments were conducted. Experiments 1 to 6 were designed to establish the models, determination of adsorption endpoints during HM man- ufacturing is not yet explored. while experiment 7 evaluated the predictive capability of the models. The loading solution was prepared using Numerous qualitative methods have been applied for process monitoring and can be classified into several several steps. Briefly, dry Panax notoginseng (Burkill) F.H.Chen powders with different harvest seasons (spring categories. The first consists of calibration-free meth- ods, including moving block standard deviation (MBSD) and winter) and medicinal components (rhizome and taproot) were refluxed twice using twice the amount and monitoring specific spectral peak intensity of active [13-14] substances ; these approaches are relatively straight- of 90% ethanol-water solution (v/w); the extracts were then filtered. Thereafter, the filtrates were merged and forward and require no additional information for mod- eling. The second category relies on methods that estimate concentrated. Subsequently, the concentrated solutions were diluted with water to different mass concentrations the difference between the spectra obtained during the process and the target spectrum, including the differ- (0.2, 0.4, and 1.0 g/mL) and some precipitates formed. The loading solutions were obtained via vacuum suc- ence between the moving block average and the target spectrum (DMBA-TS) and spectra linear superposition tion filtration. The harvest season, medicinal compo- [13,15] nents, and weight concentration corresponding to each method . The third category consists of chemometric methods based on principal component analysis (PCA), loading solution are presented in Supplementary Table S1 (http://links.lww.com/AHM/A23). The details of the including the soft-independent modeling of class anal- ogy (SIMCA) and principal component score distance loading solution preparation can be found in our previ- [16-18] [21] ous report . analysis (PC-SDA) . The fourth category consists of chemometric methods based on a classical partial least The adsorption process in each experiment was con- ducted based on several procedures. D-101 macroporous squares regression (PLSR) where the response variable is a categorical variable, including partial least-squares resin column chromatography with a 1.5-cm inner diam- [19-20] eter and 11-cm bed height was employed. The column discriminant analysis (PLS-DA) . However, no criti- cal study has been conducted to benchmark the poten- was first equilibrated with five bed volumes (BVs) of water and subsequently loaded with the loading solu- tial of various multivariate analysis methods for HM adsorption processes thus far. Therefore, we conducted tion at various flow rates (0.47, 1.0, and 2.8 mL/min). The flow rate was controlled using a BT300-2J peristal- a comparative study to evaluate the effectiveness of mul- tivariate analysis methods and facilitate the selection of tic pump (Longer, Hebei, China). Following adsorption, the column was regenerated with five BVs of 95% etha- the most appropriate method for the in-line detection of adsorption endpoints during HM manufacturing. nol-water (v/v), water, 1 mol/L HCl solution, water, and 1 mol/L NaOH solution. A schematic of the experimen- This study compared existing multivariate analysis methods to determine adsorption endpoints using in-line tal setup is depicted in Supplementary Figure S1 (http:// links.lww.com/AHM/A23). The flow rates for adsorption UV-VIS spectra. Four representative qualitative methods, including MBSD, DMBA-TS, SIMCA, and PLS-DA, were in each experiment are listed in Supplementary Table S1 (http://links.lww.com/AHM/A23). investigated. The adsorption of saponins on a D-101 macroporous resin column chromatograph, a critical Throughout the adsorption process, the effluents were monitored using a T6 spectrophotometer (Purkinje, unit operation in Panax notoginseng (Burkill) F.H.Chen injection manufacturing, was considered as an example. Beijing, China) integrated with the UVwin5 software. A flow cell with a 5-mm pathway was used for the in-line This study aims to provide a comprehensive and holistic overview of the advantages and disadvantages of vari- collection of spectra. UV-VIS spectra were collected in the region of 200 to 720 nm with a data interval of ous multivariate analysis methods to effectively facilitate their application in guiding practical HM adsorption 1 nm. Ultrapure water was used as the background for the calculation of absorbance before the daily operation. processes. 254 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com The scan speed was rapid and each spectrum could be from the process spectrum to the SBEP model domain acquired within approximately 1.7 min. The effluents higher than the critical distance (which was defined as were collected every 4 min using a BS-100N automatic the 95% confidence interval of distance) was defined compositional collector (Huxi, Shanghai, China) for the as the predicted endpoint. The detailed computa- offline ultraperformance liquid chromatography (UPLC) tional formula for SIMCA can be found in the relevant [26] analysis of notoginsenoside R , ginsenoside Rg , ginseno- reference . 1 1 side Re, ginsenoside Rb , and ginsenoside Rd. Details of PLS-DA was established based on the classical PLSR. the UPLC analytical method have been presented in a pre- The objective of PLS-DA is to find a model that separates [22] [27] vious article . All time intervals were calculated by sub- classes of observations based on X variables . In this tracting the dead time, which is defined as the first point study, the X matrix contained the in-line UV-VIS spectra in time at which the potassium chloride concentration of the effluents in the calibration set, while the Y matrix in affluent/concentration in the loading solution (C/C ) contained dummy variables that described the class was higher than 10% as detected by a conductivity membership of each spectrum; “1” was used to encode [22] meter . Finally, the first point in time at which notogin- the SBEP and “0” was used to encode the SAEP. The first senoside R C/C was observed to be higher than 10%, time point with a predicted Y less than 0.5 was defined as 1 0 as detected using UPLC, was determined to be the end- the predicted endpoint. The detailed computing formula [23] [28] point of each experiment . for PLS-DA can be found in the relevant reference . The predictive capacities of various models were eval- uated by comparing the relative standard error of pre- Multivariate modeling diction (RSEP) between the predicted and experimental [29] Based on the UPLC results, the in-line UV-VIS spectra endpoints, as shown in Equation (1) : were divided into two categories. One category consisted 2 t − t e e of spectra collected before the endpoint (SBEP), while RSEP = e (1) the other category consisted of spectra collected after the endpoint (SAEP). PCA provides correlations between fea- t where t is the experimental endpoint, and is the pre- tures by constructing orthogonal principal components dicted endpoint. All data analyses were performed using [24] (PCs) . PCs are linear combinations of feature vectors MATLAB (version 7.5.0, MathWorks, USA) scripts writ- [24] oriented in the direction of maximum variance . In this ten in-house. study, PCA was first performed to visualize the spectra of the two categories and to display the trajectory of the Results processes. The in-line UV-VIS spectra based on all wave- Overview of the spectral profile lengths in the calibration set (Experiments 1–6) were used to establish the PCA model; the spectra in the val- The UV-VIS spectra of the effluents were collected idation set (Experiment 7) were then projected onto the using an inline flow cell. The UV-VIS spectra of model. The detailed computing formula of PCA can be Experiment 1 are shown in Supplementary Figure S1 [25] found in the relevant literature . (http://links.lww.com/AHM/A23). As the adsorption Various multivariate data analysis methods, includ- process progressed, the absorbance in the UV-VIS ing MBSD, DMBA-TS, SIMCA, and PLS-DA, were then spectra gradually increased. The absorbance of sev- investigated for endpoint determination. MBSD calcu- eral wavelengths was higher than 2, indicating that lates the SD for consecutive blocks of spectra over time. less than 1% of the incident light was transmitted. To A time point with an SD lower than a certain thresh- minimize the influence of spectral noise, the absor- old is usually defined as the predicted endpoint. In this bance at these wavelengths was set to 2. The endpoints study, three consecutive spectra based on all wavelengths of each experiment obtained using offline UPLC are were used, while one spectrum was moved at a time. The listed in Table 1. detailed computing formula for MBSD can be found in PCA was performed based on the in-line UV-VIS spec- [13] the relevant reference . tra in Experiments 1 to 6. The score plots (PC -PC ) of 1 2 DMBA-TS calculates the difference between the spectra the in-line UV-VIS spectra and FSAEP in each experi- recorded during the process and the target spectrum. In ment are shown in Figure 1. PC explained 71.6% of the this study, the target spectrum was defined as the average variation, while PC explained 18.3% of the variation. of the first spectra collected after the endpoint (FSAEP) in All runs followed comparable trajectories. It was demon- the calibration set. A block of three consecutive spectra strated that the UV-VIS spectra could provide informa- recorded during the adsorption process was utilized. The tion on process evolution. time point with the minimum SD was defined as the pre- dicted endpoint for each experiment, wherein one spectrum Endpoint determination using MBSD was moved at a time. The detailed computing formula for [13] DMBA-TS can be found in the relevant reference . MBSD was then applied to evaluate the endpoints of SIMCA is a well-known pattern-recognition method the adsorption processes. The SD values throughout based on PCA. It describes each class separately in PC the adsorption process for each experiment are shown space and builds a distinct confidence region around in Figure  2. A gradual decrease in the SD values was [26] each class . In this study, the SIMCA model was estab- observed. Endpoints of the adsorption processes were lished based on two spectral classes (SBEP and SAEP) in difficult to determine because the SD threshold varied the calibration set. The first time point with the distance with changes in the loading solutions and flow rates. 255 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Table 1 Experimental endpoints determined using ultra-performance liquid chromatography; predicted endpoints and predicted errors using various multivariate analysis methods (min) Predicted endpoint errors using different Predicted endpoints using different methods methods No. Experimental endpoints DMBA-TS SIMCA PLS-DA DMBA-TS SIMCA PLS-DA 1 122.0 130.9 119.0 119.0 8.9 −3.0 −3.0 2 86.0 105.9 89.4 93.8 19.9 3.4 7.8 3 70.0 74.3 65.5 67.2 4.3 −4.5 −2.8 4 2.0 0.3 0.3 0.3 −1.7 −1.7 −1.7 5 17.0 21.2 14.4 21.2 4.2 −2.6 4.2 6 4.3 9.0 4.1 5.7 4.7 −0.2 1.4 7 58.0 59.2 53.9 57.4 1.2 −4.1 −0.6 DMBA-TS: difference between the moving block average and target spectrum; PLS-DA: partial least-squares discriminant analysis; SIMCA: soft-independent modeling of class analogy. Figure 1. Scores plots of in-line ultraviolet and visible spectra by principal component analysis. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. PC : Principal component 1; PC : Principal component 2. First spectra 1 2 being collected after end-point: ○; Spectra being collected before end-point: ▲; Spectra being collected after endpoint: ◇. All runs followed similar process trajectories indicated that the in-line ultraviolet and visible spectra could provide information on the process evolution. Therefore, MBSD was not sufficiently sensitive to detect R values of all extracted components were 0.996 and the adsorption endpoints. 0.998 for the SBEP and SAEP models, respectively. The cumulative Q values for the extracted components were 0.995 and 0.998 for the SBEP and SAEP models, respec- Endpoint determination using DMBA-TS tively. The distances from the spectra to the SBEP and Next, DMBA-TS was conducted to evaluate the end- SAEP model domains during the adsorption processes points. The SD values throughout the adsorption process are shown in Figure 4. The SBEP and SAEP were distin- for each experiment are shown in Figure  3. A gradual guishable. The predicted endpoints and errors of each decrease in SD values was observed prior to the end- experiment using SIMCA are listed in Table 1. The RSEP points, whereas they gradually increased beyond the between the predicted and experimental endpoints using endpoints. The predicted endpoints and errors of each SMICA was 4.67%. experiment using DMBA-TS are listed in Table  1. The RSEP between the predicted and experimental endpoints Endpoint determination using PLS-DA using DMBA-TS was 13.2%. The PLS-DA model was established based on the in-line UV-VIS spectra in the calibration set. There were six Endpoint determination using SIMCA 2 components in the PLS-DA model. The cumulative R SIMCA was performed based on two classes of spectra. value of all the extracted components was 0.998, while the cumulative Q for the extracted components was The numbers of components for the SBEP and SAEP models were four and three, respectively. The cumulative 0.931. The predicted Y values for each experiment 256 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Figure 2. Endpoints determination by moving block SD. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. SD: standard deviation. Spectra being collected before endpoint: ▲; Spectra being collected after end-point: ◇. The threshold changed with the changes of loading solutions and flow rates indicated that the moving block SD was not sensitive to detect the adsorption endpoints. Figure 3. Endpoints determination by difference between the moving block average and the target spectrum. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. SD: standard deviation. Predicted end-point: ○; Spectra being collected before end-point: ▲; Spectra being collected after endpoint: ◇. The predicted end-points were defined to be time points with min- imum SD values. Discussion throughout the adsorption process using PLS-DA are shown in Figure  5. SBEP and SAEP could be distin- In general, adsorption reached saturation when the com- guished as follows: SBEP possessed a predicted Y above ponent concentration in the effluent was 10% of that in 0.5, whereas SAEP possessed a predicted Y below 0.5. [30] the loading solution . Notoginsenoside R , ginsenoside The predicted endpoints and errors of each experiment Rg , ginsenoside Re, ginsenoside Rb , and ginsenoside 1 1 using PLS-DA are listed in Table  1. The RSEP between Rd are the major bioactive components of the Panax the predicted and experimental endpoints using PLS-DA notoginseng (Burkill) F.H.Chen injection. As revealed in was 5.71%. 257 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Figure 4. Endpoints determination by soft independent modeling of class analogy. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. SAEP: Spectra being collected after endpoint; SBEP: Spectra being collected before endpoint. Predicted endpoint: ○; Spectra being collected before endpoint: ▲; Spectra being collected after end-point: ◇. The predicted endpoints were defined to be the first time point with distance from process spectrum to spectra being collected before endpoint model domain outside the 95% confidence intervals of distance. Figure 5. Endpoints determination by partial least-squares discriminant analysis. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. Predicted endpoint: ○; Spectra being collected before endpoint: ▲; Spectra being collected after endpoint: ◇. The predicted endpoints were defined to be the first time point with predicted Y below 0.5. our previous study, notoginsenoside R is easily lost in were altered, both the endpoints of the adsorption [23] the effluent of these five saponins . Therefore, it was processes and composition of the effluent changed. To used as the indicator component for endpoint determina- build multivariate analysis models with wide applica- tion. The first time-point with notoginsenoside R C/C bility, the spectra included in the calibration set were 1 0 higher than 10% to be detected through UPLC was obtained from Experiments 1 to 6, which were con- determined to be the experimental endpoint. ducted with various flow rates, harvest seasons, medic- The quality of the loading solution was dependent inal components, and dilution ratios. In Experiment 4, on the harvest season, medicinal components, and dilu- saponins leaked rapidly from the macroporous resin. tion ratio. When the flow rates or loading solutions Compared with other adsorption processes, the weight 258 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com concentration of the loading solution diluent in detection of adsorption endpoints to guide practical Experiment 4 was extremely high. These results indi- HM manufacturing. cate that a high concentration of saponins and limited number of active sites may result in the incomplete Conflict of interest statement adsorption of saponins on the macroporous resin. The authors declare no conflict of interest. MBSD was found to be unsuitable for adsorption processes. This may be explained by the fact that MBSD is particularly well-suited for processes with a gradual Funding increase or decrease in the components. Prior to the end- This work was supported by the National Natural point of adsorption, no saponins were detected in the Science Foundation of China (82104383) and National effluent. However, components such as pigments, whose S&T Major Project of China (2012ZX09101201-003). adsorption capacity on the resin is weaker than that of saponins, may be detected in the effluent. Additionally, Author contributions the threshold was extremely high in Experiments 5 Cheng Jiang was participated in the writing (original and 6. Compared with the other adsorption processes, draft) of the paper, the performance of the research and the flow rates in Experiments 5 and 6 were high. The data analysis. Haibin Qu was participated in conceptu- results indicate that a high flow rate may result in a rapid alization, research design, and writing (review & editing) change in the spectra as a consequence high threshold of of the paper. MBSD. Although MBSD has been widely recommended for endpoint determination in the blending processes of chemical medicine, it is incompatible with the adsorp- Ethical approval of studies and informed consent tion process of HM. This article does not contain any research on human or The RSEPs between the predicted and experi- animal subjects performed by any of the authors. mental endpoints were 13.2%, 4.67%, and 5.71% using DMBA-TS, SIMCA, and PLS-DA, respectively. Compared with MBSD, DMBA-TS can be used to Acknowledgments assess the closeness of the process spectrum to the None. target spectrum; however, the description of the tar- get spectrum is required prior to modeling. The RSEPs References obtained using SIMCA and PLS-DA were smaller [1] Bai Y, Ma J, Zhu W, et al. Highly selective separation and purifica- than those obtained via DMBA-TS, indicating that tion of chicoric acid from Echinacea purpurea by quality control SIMCA and PLS-DA were more suitable for endpoint methods in macroporous adsorption resin column chromatogra- determination in adsorption processes. A possible rea- phy. J Sep Sci 2019;42(5):1027–1036. son might be that the MBSD was a calibration-free [2] Cren EC, Cardozo L, Silva EA, et al. Breakthrough curves for oleic acid removal from ethanolic solutions using a strong anion method, whereas the DMBA-TS model was established exchange resin. Sep Purif Technol 2009;69(1):1–6. based on the target spectrum. As shown in Figure  1, [3] Fayaz M, Zarifi MH, Abdolrazzaghi M, et al. A novel technique the in-line UV-VIS spectra during the adsorption pro- for determining the adsorption capacity and breakthrough time of adsorbents using a noncontact high-resolution microwave res- cesses can also provide information on process evo- onator sensor. Environ Sci Technol 2017;51(1):427–435. lution. Therefore, SIMCA and PLS-DA, which were [4] Wu YJ, Jin Y, Li YR, et al. NIR spectroscopy as a process analyti- modeled based on all spectra in the calibration set, cal technology (PAT) tool for on-line and real-time monitoring of contained more information. Moreover, models using an extraction process. Vib Spectrosc 2012;58:109–118. [5] Geskovski N, Stefkov G, Gigopulu O, et al. Mid-infrared spectros- both MBSD and DMBA-TS were established based on copy as process analytical technology tool for estimation of THC all wavelengths in the UV-VIS spectral region. SIMCA and CBD content in Cannabis flowers and extracts. Spectrochim and PLS-DA are representative methods for informa- Acta A Mol Biomol Spectrosc 2021;251:119422. [6] Wu SJ, Qiu P, Li P, et al. A near-infrared spectroscopy-based end- tion extraction, which are helpful for classifying large point determination method for the blending process of Dahuang amounts of data. It was demonstrated that models soda tablets. J Zhejiang Univ Sci B 2020;21(11):897–910. containing more information and processing with [7] Tewari J, Strong R, Boulas P. At-line determination of phar- appropriate information extraction enabled the favor- maceuticals small molecule’s blending end point using che- mometric modeling combined with Fourier transform near able detection of endpoints. SIMCA and PLS-DA can infrared spectroscopy. Spectrochim Acta A Mol Biomol Spectrosc therefore be used for the effective in-line detection of 2017;173:886–891. adsorption endpoints. [8] Khorasani M, Amigo JM, Bertelsen P, et al. Detecting blend- ing end-point using mean squares successive difference test and near-infrared spectroscopy. J Pharm Sci 2015;104(8):2541–2549. [9] Igne B, de Juan A, Jaumot J, et al. Modeling strategies for phar- Conclusions maceutical blend monitoring and end-point determination by This study compared four representative multivari- near-infrared spectroscopy. Int J Pharm 2014;473(1–2):219–231. [10] Yang NL, Cheng YY, Qu HB. An approach to purifying process ate analysis methods, including MBSD, DMBA-TS, analysis of Chinese herbal extracts using NIRS. Acta Chim Sinica SIMCA, and PLS-DA, to determine the endpoints of 2003;61(5):742–747. HM adsorption processes. SIMCA and PLS-DA were [11] Yan B, Qu H. Multivariate data analysis of UV spectra in mon- most effective for endpoint determination. The results itoring elution and determining endpoint of chromatography using polyamide column. J Sep Sci 2013;36(7):1231–1237. of this study provide a more comprehensive overview [12] Jiang C, Liu Y, Qu HB. Data fusion strategy based on near infra- of the various multivariate analysis methods to facili- red spectra and ultraviolet spectra for simultaneous determina- tate the selection of the most suitable one. This study tion of ginsenosides and saccharides in Chinese herbal injection. was conducive to addressing the issues of the in-line Anal Methods-Uk 2013;5:4467–4475. 259 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com [13] Blanco M, Cueva-Mestanza R, Cruz J. Critical evaluation of [22] Jiang C, Gong XC, Qu HB. A strategy for adjusting macroporous methods for end-point determination in pharmaceutical blending resin column chromatographic process parameters based on raw processes. Anal Methods-Uk 2012;4(9):2694–2703. material variation. Sep Purif Technol 2013;116:287–293. [14] Wu Y, Jin Y, Ding H, et al. In-line monitoring of extraction [23] Jiang C, Gong X, Qu H. Multivariate modeling and prediction of process of scutellarein from Erigeron breviscapus (vant.) breakthrough curves for herbal medicine adsorption on column chro- Hand-Mazz based on qualitative and quantitative uses of near-in- matography: a case study. Sep Sci Technol 2015;50(7):1030–1037. frared spectroscopy. Spectrochim Acta A Mol Biomol Spectrosc [24] Tranas R, Lovvik OM, Tomic O, et al. Lattice thermal conductiv- 2011;79(5):934–939. ity of half-Heuslers with density functional theory and machine [15] Wu HQ, Khan MA. Quality-By-Design (QbD): an integrated learning: enhancing predictivity by active sampling with principal approach for evaluation of powder blending process kinetics and component analysis. Comp Mater Sci 2022;202:110938. determination of powder blending end-point. J Pharm Sci-Us [25] Wold S, Esbensen K, Geladi P. Principal component analysis. 2009;98(8):2784–2798. Chemometr Intell Lab 1987;2(1–3):37–52. [16] Costa EB, Silva RC, Espejo-Roman JM, et al. Chemometric meth- [26] Wold S. Pattern-recognition by means of disjoint principal com- ods in antimalarial drug design from 1,2,4,5-tetraoxanes ana- ponents. Pattern Recogn 1976;8(3):127–139. logues. SAR QSAR Environ Res 2020;31(9):677–695. [27] Ledauphin J, Le Milbeau C, Barillier D, et al. Differences in the [17] Rukundo IR, Danao MC. Identifying turmeric powder by volatile compositions of French labeled brandies (Armagnac, source and metanil yellow adulteration levels using near-in- Calvados, Cognac, and Mirabelle) using GC-MS and PLS-DA. J frared spectra and PCA-SIMCA modeling. J Food Prot Agr Food Chem 2010;58(13):7782–7793. 2020;83(6):968–974. [28] Frank IE, Friedman JH. A statistical view of some chemometrics [18] Puchert T, Holzhauer CV, Menezes JC, et al. A new PAT/QbD regression tools. Technometrics 1993;35(2):109–135. approach for the determination of blend homogeneity: combina- [29] Luis ML, Garcia JM, Jimenez F, et al. Simultaneous determina- tion of on-line NIRS analysis with PC scores distance analysis tion of chlorthalidone and spironolactone with univariate and (PC-SDA). Eur J Pharm Biopharm 2011;78(1):173–182. multivariate calibration: wavelength range selection. J AOAC Int [19] Jimenez-Carvelo AM, Martin-Torres S, Ortega-Gavilan 1999;82(5):1054–1063. F, et al. PLS-DA vs sparse PLS-DA in food traceability. A [30] Liu W, Zhang S, Zu YG, et al. Preliminary enrichment and case study: authentication of avocado samples. Talanta separation of genistein and apigenin from extracts of pigeon 2021;224:121904. pea roots by macroporous resins. Bioresource Technol [20] Vieira LS, Assis C, de Queiroz M, et al. Building robust models for 2010;101(12):4667–4675. identification of adulteration in olive oil using FT-NIR, PLS-DA and variable selection. Food Chem 2021;345:128866. [21] Jiang C, Qu H. A comparative study of using in-line near-in- How to cite this article: Jiang C, Qu HB. In-line spectroscopy combined frared spectra, ultraviolet spectra and fused spectra to monitor with multivariate analysis methods for endpoint determination in column Panax notoginseng adsorption process. J Phar Biomed Anal chromatographic adsorption processes for herbal medicine. Acupunct Herb 2015;102:78–84. Med 2022;2(4):253–260. doi: 10.1097/HM9.0000000000000035 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Acupuncture & Herbal Medicine Wolters Kluwer Health

In-line spectroscopy combined with multivariate analysis methods for endpoint determination in column chromatographic adsorption processes for herbal medicine

Acupuncture & Herbal Medicine , Volume 2 (4) – Dec 22, 2022

Loading next page...
 
/lp/wolters-kluwer-health/in-line-spectroscopy-combined-with-multivariate-analysis-methods-for-f0oJv0T0CG
Publisher
Wolters Kluwer Health
Copyright
Copyright © 2022 Tianjin University of Traditional Chinese Medicine.
ISSN
2765-8619
DOI
10.1097/hm9.0000000000000035
Publisher site
See Article on Publisher Site

Abstract

Objective: In a chromatographic cycle, the adsorption process is a critical unit operation that has a significant impact on downstream processes and, ultimately, the quality of the final products. The development of a rapid method to determine the endpoints of adsorption processes in a large-scale manufacturing is of substantial importance for herbal medicine (HM) manufacturers. Methods: In this study, the adsorption of saponins on a macroporous resin column chromatograph, a critical unit operation in Panax notoginseng (Burkill) F.H.Chen injection manufacturing, was considered as an example. The evaluation results of in-line ultraviolet and visible spectra combined with various multivariate analysis methods, including the moving block standard deviation (MBSD), difference between the moving block average and the target spectrum (DMBA-TS), soft-independent modeling of class analogy (SIMCA), and partial least-squares discriminant analysis (PLS-DA), were compared. Results: MBSD was unsuitable for adsorption processes. The relative standard errors of prediction between the predicted and experimental endpoints were 13.2%, 4.67%, and 5.71% using DMBA-TS, SIMCA, and PLS-DA, respectively. Conclusions: Among the considered analysis methods, SIMCA and PLS-DA were more effective for endpoint determination. The results of this study provide a more comprehensive overview of the effectiveness of various multivariate analysis methods to facilitate the selection of the most suitable method. This study was also conducive to address the issues of the in-line detection of adsorption endpoints to guide practical HM manufacturing. Keywords: Adsorption, Endpoint, Herbal medicine, Multivariate analysis, Ultraviolet and visible Graphical abstract: http://links.lww.com/AHM/A23 Introduction plays an important role in improving the safety and effi- cacy of the final products. It has been used to separate Column chromatography has a wide range of applica- several types of bioactive substances from medicinal tions in many fields, including herbal medicine (HM) herbs, including saponins, polyphenols, flavonoids, and manufacturing, biological medicine manufacturing, and [1] alkaloids . Each chromatographic cycle typically con- wastewater treatment. During the manufacturing of HM, sists of adsorption, washing, and elution. The adsorp- especially botanical injections, column chromatography tion process is a critical unit operation and may have a significant impact on the downstream processes and quality of the final products. Insufficient adsorption time College of Pharmaceutical Sciences, Zhejiang University, Hangzhou, China; Zhejiang Provincial Key Laboratory of Cancer Prevention and may result in a low efficiency of adsorbent utilization, Treatment Technology of Integrated Traditional Chinese and Western whereas excessive adsorption time may result in high Medicine, Zhejiang Academy of Traditional Chinese Medicine, Tongde variability in the final product quality and low efficiency Hospital of Zhejiang Province, Hangzhou, China; Innovation Center [2] of solute adsorption . The determination of the optimal in Zhejiang University, State Key Laboratory of Component-Based adsorption endpoint is of considerable importance in a Chinese Medicine, Hangzhou, China large-scale manufacturing. * Corresponding author. Haibin Qu, College of Pharmaceutical Breakthrough time, which is defined as the time Sciences, Zhejiang University, Hangzhou 310058, China, E-mail: required for a certain number of active components quhb@zju.edu.cn. to be detected in the effluent, is generally considered a Copyright © 2022 Tianjin University of Traditional Chinese Medicine. [3] suitable endpoint . However, several factors may influ- This is an open-access article distributed under the terms of the ence the breakthrough time, such as variations in load- Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download ing solutions, fluctuations in operating parameters, and and share the work provided it is properly cited. The work cannot be variations in the adsorption capacity of the adsorbents. changed in any way or used commercially without permission from the Consequently, traditional HM manufacturing, which is journal. conducted with a fixed adsorption time as the endpoint, Acupuncture and Herbal Medicine (2022) 2:4 cannot adapt to changes in the above-mentioned factors. Received 15 January 2022 / Accepted 11 July 2022 The breakthrough time can be determined by an offline http://dx.doi.org/10.1097/HM9.0000000000000035 assessment of the active components in the effluents via 253 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Materials and Methods chromatographic techniques. However, the long duration required for analysis may ultimately delay critical man- Samples, chemicals, and reagents ufacturing decisions. Therefore, the development of a The medicinal herb Panax notoginseng (Burkill) F.H.Chen rapid method to determine the endpoints of adsorption was purchased from the Wenshan Panax Notoginseng processes is of critical importance for HM manufacturers. International Trade Center (Yunnan, China). The stan- Spectroscopy is an effective choice for process anal- dard compounds notoginsenoside R , ginsenoside ysis because it requires no sample preparation and can Rg , ginsenoside Re, ginsenoside Rb , and ginseno- provide real-time information about the processes. Thus 1 1 side Rd were purchased from Ronghe Pharmaceutical far, several studies have used spectroscopic techniques Technology Co., Ltd. (Shanghai, China) with purities for endpoint determination in pharmaceutical processes, of over 99%. D-101 macroporous resin was purchased including blending, fluidized drying, fluidized bed gran- from Haiguang Chemical Co., Ltd. (Tianjin, China). [4-9] ulation, coating, extraction, and chemical synthesis . HPLC-grade acetonitrile was purchased from Merck Spectroscopic techniques have also been reported to (Darmstadt, Germany). Ultrapurified water was pre- monitor the column chromatographic elution processes [10] pared using a Milli-Q academic water purification sys- of Coptis chinensis Franch and Salvia miltiorrhiza tem (Millipore, Billerica, MA, USA). [11] Bunge . Ultraviolet and visible (UV-VIS) spectroscopy is a form of electron spectroscopy that is based on the exci- [12] tation of electrons . It possesses the advantages of high Process description sensitivity and inexpensive instrumentation. However, the performance of in-line UV-VIS spectroscopy in the In this study, seven experiments were conducted. Experiments 1 to 6 were designed to establish the models, determination of adsorption endpoints during HM man- ufacturing is not yet explored. while experiment 7 evaluated the predictive capability of the models. The loading solution was prepared using Numerous qualitative methods have been applied for process monitoring and can be classified into several several steps. Briefly, dry Panax notoginseng (Burkill) F.H.Chen powders with different harvest seasons (spring categories. The first consists of calibration-free meth- ods, including moving block standard deviation (MBSD) and winter) and medicinal components (rhizome and taproot) were refluxed twice using twice the amount and monitoring specific spectral peak intensity of active [13-14] substances ; these approaches are relatively straight- of 90% ethanol-water solution (v/w); the extracts were then filtered. Thereafter, the filtrates were merged and forward and require no additional information for mod- eling. The second category relies on methods that estimate concentrated. Subsequently, the concentrated solutions were diluted with water to different mass concentrations the difference between the spectra obtained during the process and the target spectrum, including the differ- (0.2, 0.4, and 1.0 g/mL) and some precipitates formed. The loading solutions were obtained via vacuum suc- ence between the moving block average and the target spectrum (DMBA-TS) and spectra linear superposition tion filtration. The harvest season, medicinal compo- [13,15] nents, and weight concentration corresponding to each method . The third category consists of chemometric methods based on principal component analysis (PCA), loading solution are presented in Supplementary Table S1 (http://links.lww.com/AHM/A23). The details of the including the soft-independent modeling of class anal- ogy (SIMCA) and principal component score distance loading solution preparation can be found in our previ- [16-18] [21] ous report . analysis (PC-SDA) . The fourth category consists of chemometric methods based on a classical partial least The adsorption process in each experiment was con- ducted based on several procedures. D-101 macroporous squares regression (PLSR) where the response variable is a categorical variable, including partial least-squares resin column chromatography with a 1.5-cm inner diam- [19-20] eter and 11-cm bed height was employed. The column discriminant analysis (PLS-DA) . However, no criti- cal study has been conducted to benchmark the poten- was first equilibrated with five bed volumes (BVs) of water and subsequently loaded with the loading solu- tial of various multivariate analysis methods for HM adsorption processes thus far. Therefore, we conducted tion at various flow rates (0.47, 1.0, and 2.8 mL/min). The flow rate was controlled using a BT300-2J peristal- a comparative study to evaluate the effectiveness of mul- tivariate analysis methods and facilitate the selection of tic pump (Longer, Hebei, China). Following adsorption, the column was regenerated with five BVs of 95% etha- the most appropriate method for the in-line detection of adsorption endpoints during HM manufacturing. nol-water (v/v), water, 1 mol/L HCl solution, water, and 1 mol/L NaOH solution. A schematic of the experimen- This study compared existing multivariate analysis methods to determine adsorption endpoints using in-line tal setup is depicted in Supplementary Figure S1 (http:// links.lww.com/AHM/A23). The flow rates for adsorption UV-VIS spectra. Four representative qualitative methods, including MBSD, DMBA-TS, SIMCA, and PLS-DA, were in each experiment are listed in Supplementary Table S1 (http://links.lww.com/AHM/A23). investigated. The adsorption of saponins on a D-101 macroporous resin column chromatograph, a critical Throughout the adsorption process, the effluents were monitored using a T6 spectrophotometer (Purkinje, unit operation in Panax notoginseng (Burkill) F.H.Chen injection manufacturing, was considered as an example. Beijing, China) integrated with the UVwin5 software. A flow cell with a 5-mm pathway was used for the in-line This study aims to provide a comprehensive and holistic overview of the advantages and disadvantages of vari- collection of spectra. UV-VIS spectra were collected in the region of 200 to 720 nm with a data interval of ous multivariate analysis methods to effectively facilitate their application in guiding practical HM adsorption 1 nm. Ultrapure water was used as the background for the calculation of absorbance before the daily operation. processes. 254 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com The scan speed was rapid and each spectrum could be from the process spectrum to the SBEP model domain acquired within approximately 1.7 min. The effluents higher than the critical distance (which was defined as were collected every 4 min using a BS-100N automatic the 95% confidence interval of distance) was defined compositional collector (Huxi, Shanghai, China) for the as the predicted endpoint. The detailed computa- offline ultraperformance liquid chromatography (UPLC) tional formula for SIMCA can be found in the relevant [26] analysis of notoginsenoside R , ginsenoside Rg , ginseno- reference . 1 1 side Re, ginsenoside Rb , and ginsenoside Rd. Details of PLS-DA was established based on the classical PLSR. the UPLC analytical method have been presented in a pre- The objective of PLS-DA is to find a model that separates [22] [27] vious article . All time intervals were calculated by sub- classes of observations based on X variables . In this tracting the dead time, which is defined as the first point study, the X matrix contained the in-line UV-VIS spectra in time at which the potassium chloride concentration of the effluents in the calibration set, while the Y matrix in affluent/concentration in the loading solution (C/C ) contained dummy variables that described the class was higher than 10% as detected by a conductivity membership of each spectrum; “1” was used to encode [22] meter . Finally, the first point in time at which notogin- the SBEP and “0” was used to encode the SAEP. The first senoside R C/C was observed to be higher than 10%, time point with a predicted Y less than 0.5 was defined as 1 0 as detected using UPLC, was determined to be the end- the predicted endpoint. The detailed computing formula [23] [28] point of each experiment . for PLS-DA can be found in the relevant reference . The predictive capacities of various models were eval- uated by comparing the relative standard error of pre- Multivariate modeling diction (RSEP) between the predicted and experimental [29] Based on the UPLC results, the in-line UV-VIS spectra endpoints, as shown in Equation (1) : were divided into two categories. One category consisted 2 t − t e e of spectra collected before the endpoint (SBEP), while RSEP = e (1) the other category consisted of spectra collected after the endpoint (SAEP). PCA provides correlations between fea- t where t is the experimental endpoint, and is the pre- tures by constructing orthogonal principal components dicted endpoint. All data analyses were performed using [24] (PCs) . PCs are linear combinations of feature vectors MATLAB (version 7.5.0, MathWorks, USA) scripts writ- [24] oriented in the direction of maximum variance . In this ten in-house. study, PCA was first performed to visualize the spectra of the two categories and to display the trajectory of the Results processes. The in-line UV-VIS spectra based on all wave- Overview of the spectral profile lengths in the calibration set (Experiments 1–6) were used to establish the PCA model; the spectra in the val- The UV-VIS spectra of the effluents were collected idation set (Experiment 7) were then projected onto the using an inline flow cell. The UV-VIS spectra of model. The detailed computing formula of PCA can be Experiment 1 are shown in Supplementary Figure S1 [25] found in the relevant literature . (http://links.lww.com/AHM/A23). As the adsorption Various multivariate data analysis methods, includ- process progressed, the absorbance in the UV-VIS ing MBSD, DMBA-TS, SIMCA, and PLS-DA, were then spectra gradually increased. The absorbance of sev- investigated for endpoint determination. MBSD calcu- eral wavelengths was higher than 2, indicating that lates the SD for consecutive blocks of spectra over time. less than 1% of the incident light was transmitted. To A time point with an SD lower than a certain thresh- minimize the influence of spectral noise, the absor- old is usually defined as the predicted endpoint. In this bance at these wavelengths was set to 2. The endpoints study, three consecutive spectra based on all wavelengths of each experiment obtained using offline UPLC are were used, while one spectrum was moved at a time. The listed in Table 1. detailed computing formula for MBSD can be found in PCA was performed based on the in-line UV-VIS spec- [13] the relevant reference . tra in Experiments 1 to 6. The score plots (PC -PC ) of 1 2 DMBA-TS calculates the difference between the spectra the in-line UV-VIS spectra and FSAEP in each experi- recorded during the process and the target spectrum. In ment are shown in Figure 1. PC explained 71.6% of the this study, the target spectrum was defined as the average variation, while PC explained 18.3% of the variation. of the first spectra collected after the endpoint (FSAEP) in All runs followed comparable trajectories. It was demon- the calibration set. A block of three consecutive spectra strated that the UV-VIS spectra could provide informa- recorded during the adsorption process was utilized. The tion on process evolution. time point with the minimum SD was defined as the pre- dicted endpoint for each experiment, wherein one spectrum Endpoint determination using MBSD was moved at a time. The detailed computing formula for [13] DMBA-TS can be found in the relevant reference . MBSD was then applied to evaluate the endpoints of SIMCA is a well-known pattern-recognition method the adsorption processes. The SD values throughout based on PCA. It describes each class separately in PC the adsorption process for each experiment are shown space and builds a distinct confidence region around in Figure  2. A gradual decrease in the SD values was [26] each class . In this study, the SIMCA model was estab- observed. Endpoints of the adsorption processes were lished based on two spectral classes (SBEP and SAEP) in difficult to determine because the SD threshold varied the calibration set. The first time point with the distance with changes in the loading solutions and flow rates. 255 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Table 1 Experimental endpoints determined using ultra-performance liquid chromatography; predicted endpoints and predicted errors using various multivariate analysis methods (min) Predicted endpoint errors using different Predicted endpoints using different methods methods No. Experimental endpoints DMBA-TS SIMCA PLS-DA DMBA-TS SIMCA PLS-DA 1 122.0 130.9 119.0 119.0 8.9 −3.0 −3.0 2 86.0 105.9 89.4 93.8 19.9 3.4 7.8 3 70.0 74.3 65.5 67.2 4.3 −4.5 −2.8 4 2.0 0.3 0.3 0.3 −1.7 −1.7 −1.7 5 17.0 21.2 14.4 21.2 4.2 −2.6 4.2 6 4.3 9.0 4.1 5.7 4.7 −0.2 1.4 7 58.0 59.2 53.9 57.4 1.2 −4.1 −0.6 DMBA-TS: difference between the moving block average and target spectrum; PLS-DA: partial least-squares discriminant analysis; SIMCA: soft-independent modeling of class analogy. Figure 1. Scores plots of in-line ultraviolet and visible spectra by principal component analysis. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. PC : Principal component 1; PC : Principal component 2. First spectra 1 2 being collected after end-point: ○; Spectra being collected before end-point: ▲; Spectra being collected after endpoint: ◇. All runs followed similar process trajectories indicated that the in-line ultraviolet and visible spectra could provide information on the process evolution. Therefore, MBSD was not sufficiently sensitive to detect R values of all extracted components were 0.996 and the adsorption endpoints. 0.998 for the SBEP and SAEP models, respectively. The cumulative Q values for the extracted components were 0.995 and 0.998 for the SBEP and SAEP models, respec- Endpoint determination using DMBA-TS tively. The distances from the spectra to the SBEP and Next, DMBA-TS was conducted to evaluate the end- SAEP model domains during the adsorption processes points. The SD values throughout the adsorption process are shown in Figure 4. The SBEP and SAEP were distin- for each experiment are shown in Figure  3. A gradual guishable. The predicted endpoints and errors of each decrease in SD values was observed prior to the end- experiment using SIMCA are listed in Table 1. The RSEP points, whereas they gradually increased beyond the between the predicted and experimental endpoints using endpoints. The predicted endpoints and errors of each SMICA was 4.67%. experiment using DMBA-TS are listed in Table  1. The RSEP between the predicted and experimental endpoints Endpoint determination using PLS-DA using DMBA-TS was 13.2%. The PLS-DA model was established based on the in-line UV-VIS spectra in the calibration set. There were six Endpoint determination using SIMCA 2 components in the PLS-DA model. The cumulative R SIMCA was performed based on two classes of spectra. value of all the extracted components was 0.998, while the cumulative Q for the extracted components was The numbers of components for the SBEP and SAEP models were four and three, respectively. The cumulative 0.931. The predicted Y values for each experiment 256 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Figure 2. Endpoints determination by moving block SD. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. SD: standard deviation. Spectra being collected before endpoint: ▲; Spectra being collected after end-point: ◇. The threshold changed with the changes of loading solutions and flow rates indicated that the moving block SD was not sensitive to detect the adsorption endpoints. Figure 3. Endpoints determination by difference between the moving block average and the target spectrum. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. SD: standard deviation. Predicted end-point: ○; Spectra being collected before end-point: ▲; Spectra being collected after endpoint: ◇. The predicted end-points were defined to be time points with min- imum SD values. Discussion throughout the adsorption process using PLS-DA are shown in Figure  5. SBEP and SAEP could be distin- In general, adsorption reached saturation when the com- guished as follows: SBEP possessed a predicted Y above ponent concentration in the effluent was 10% of that in 0.5, whereas SAEP possessed a predicted Y below 0.5. [30] the loading solution . Notoginsenoside R , ginsenoside The predicted endpoints and errors of each experiment Rg , ginsenoside Re, ginsenoside Rb , and ginsenoside 1 1 using PLS-DA are listed in Table  1. The RSEP between Rd are the major bioactive components of the Panax the predicted and experimental endpoints using PLS-DA notoginseng (Burkill) F.H.Chen injection. As revealed in was 5.71%. 257 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com Figure 4. Endpoints determination by soft independent modeling of class analogy. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. SAEP: Spectra being collected after endpoint; SBEP: Spectra being collected before endpoint. Predicted endpoint: ○; Spectra being collected before endpoint: ▲; Spectra being collected after end-point: ◇. The predicted endpoints were defined to be the first time point with distance from process spectrum to spectra being collected before endpoint model domain outside the 95% confidence intervals of distance. Figure 5. Endpoints determination by partial least-squares discriminant analysis. (A) Experiment 1; (B) Experiment 2; (C) Experiment 3; (D) Experiment 4; (E) Experiment 5; (F) Experiment 6; (G) Experiment 7. Predicted endpoint: ○; Spectra being collected before endpoint: ▲; Spectra being collected after endpoint: ◇. The predicted endpoints were defined to be the first time point with predicted Y below 0.5. our previous study, notoginsenoside R is easily lost in were altered, both the endpoints of the adsorption [23] the effluent of these five saponins . Therefore, it was processes and composition of the effluent changed. To used as the indicator component for endpoint determina- build multivariate analysis models with wide applica- tion. The first time-point with notoginsenoside R C/C bility, the spectra included in the calibration set were 1 0 higher than 10% to be detected through UPLC was obtained from Experiments 1 to 6, which were con- determined to be the experimental endpoint. ducted with various flow rates, harvest seasons, medic- The quality of the loading solution was dependent inal components, and dilution ratios. In Experiment 4, on the harvest season, medicinal components, and dilu- saponins leaked rapidly from the macroporous resin. tion ratio. When the flow rates or loading solutions Compared with other adsorption processes, the weight 258 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com concentration of the loading solution diluent in detection of adsorption endpoints to guide practical Experiment 4 was extremely high. These results indi- HM manufacturing. cate that a high concentration of saponins and limited number of active sites may result in the incomplete Conflict of interest statement adsorption of saponins on the macroporous resin. The authors declare no conflict of interest. MBSD was found to be unsuitable for adsorption processes. This may be explained by the fact that MBSD is particularly well-suited for processes with a gradual Funding increase or decrease in the components. Prior to the end- This work was supported by the National Natural point of adsorption, no saponins were detected in the Science Foundation of China (82104383) and National effluent. However, components such as pigments, whose S&T Major Project of China (2012ZX09101201-003). adsorption capacity on the resin is weaker than that of saponins, may be detected in the effluent. Additionally, Author contributions the threshold was extremely high in Experiments 5 Cheng Jiang was participated in the writing (original and 6. Compared with the other adsorption processes, draft) of the paper, the performance of the research and the flow rates in Experiments 5 and 6 were high. The data analysis. Haibin Qu was participated in conceptu- results indicate that a high flow rate may result in a rapid alization, research design, and writing (review & editing) change in the spectra as a consequence high threshold of of the paper. MBSD. Although MBSD has been widely recommended for endpoint determination in the blending processes of chemical medicine, it is incompatible with the adsorp- Ethical approval of studies and informed consent tion process of HM. This article does not contain any research on human or The RSEPs between the predicted and experi- animal subjects performed by any of the authors. mental endpoints were 13.2%, 4.67%, and 5.71% using DMBA-TS, SIMCA, and PLS-DA, respectively. Compared with MBSD, DMBA-TS can be used to Acknowledgments assess the closeness of the process spectrum to the None. target spectrum; however, the description of the tar- get spectrum is required prior to modeling. The RSEPs References obtained using SIMCA and PLS-DA were smaller [1] Bai Y, Ma J, Zhu W, et al. Highly selective separation and purifica- than those obtained via DMBA-TS, indicating that tion of chicoric acid from Echinacea purpurea by quality control SIMCA and PLS-DA were more suitable for endpoint methods in macroporous adsorption resin column chromatogra- determination in adsorption processes. A possible rea- phy. J Sep Sci 2019;42(5):1027–1036. son might be that the MBSD was a calibration-free [2] Cren EC, Cardozo L, Silva EA, et al. Breakthrough curves for oleic acid removal from ethanolic solutions using a strong anion method, whereas the DMBA-TS model was established exchange resin. Sep Purif Technol 2009;69(1):1–6. based on the target spectrum. As shown in Figure  1, [3] Fayaz M, Zarifi MH, Abdolrazzaghi M, et al. A novel technique the in-line UV-VIS spectra during the adsorption pro- for determining the adsorption capacity and breakthrough time of adsorbents using a noncontact high-resolution microwave res- cesses can also provide information on process evo- onator sensor. Environ Sci Technol 2017;51(1):427–435. lution. Therefore, SIMCA and PLS-DA, which were [4] Wu YJ, Jin Y, Li YR, et al. NIR spectroscopy as a process analyti- modeled based on all spectra in the calibration set, cal technology (PAT) tool for on-line and real-time monitoring of contained more information. Moreover, models using an extraction process. Vib Spectrosc 2012;58:109–118. [5] Geskovski N, Stefkov G, Gigopulu O, et al. Mid-infrared spectros- both MBSD and DMBA-TS were established based on copy as process analytical technology tool for estimation of THC all wavelengths in the UV-VIS spectral region. SIMCA and CBD content in Cannabis flowers and extracts. Spectrochim and PLS-DA are representative methods for informa- Acta A Mol Biomol Spectrosc 2021;251:119422. [6] Wu SJ, Qiu P, Li P, et al. A near-infrared spectroscopy-based end- tion extraction, which are helpful for classifying large point determination method for the blending process of Dahuang amounts of data. It was demonstrated that models soda tablets. J Zhejiang Univ Sci B 2020;21(11):897–910. containing more information and processing with [7] Tewari J, Strong R, Boulas P. At-line determination of phar- appropriate information extraction enabled the favor- maceuticals small molecule’s blending end point using che- mometric modeling combined with Fourier transform near able detection of endpoints. SIMCA and PLS-DA can infrared spectroscopy. Spectrochim Acta A Mol Biomol Spectrosc therefore be used for the effective in-line detection of 2017;173:886–891. adsorption endpoints. [8] Khorasani M, Amigo JM, Bertelsen P, et al. Detecting blend- ing end-point using mean squares successive difference test and near-infrared spectroscopy. J Pharm Sci 2015;104(8):2541–2549. [9] Igne B, de Juan A, Jaumot J, et al. Modeling strategies for phar- Conclusions maceutical blend monitoring and end-point determination by This study compared four representative multivari- near-infrared spectroscopy. Int J Pharm 2014;473(1–2):219–231. [10] Yang NL, Cheng YY, Qu HB. An approach to purifying process ate analysis methods, including MBSD, DMBA-TS, analysis of Chinese herbal extracts using NIRS. Acta Chim Sinica SIMCA, and PLS-DA, to determine the endpoints of 2003;61(5):742–747. HM adsorption processes. SIMCA and PLS-DA were [11] Yan B, Qu H. Multivariate data analysis of UV spectra in mon- most effective for endpoint determination. The results itoring elution and determining endpoint of chromatography using polyamide column. J Sep Sci 2013;36(7):1231–1237. of this study provide a more comprehensive overview [12] Jiang C, Liu Y, Qu HB. Data fusion strategy based on near infra- of the various multivariate analysis methods to facili- red spectra and ultraviolet spectra for simultaneous determina- tate the selection of the most suitable one. This study tion of ginsenosides and saccharides in Chinese herbal injection. was conducive to addressing the issues of the in-line Anal Methods-Uk 2013;5:4467–4475. 259 Jiang and Qu • Volume 2 • Number 4 • 2022 www.ahmedjournal.com [13] Blanco M, Cueva-Mestanza R, Cruz J. Critical evaluation of [22] Jiang C, Gong XC, Qu HB. A strategy for adjusting macroporous methods for end-point determination in pharmaceutical blending resin column chromatographic process parameters based on raw processes. Anal Methods-Uk 2012;4(9):2694–2703. material variation. Sep Purif Technol 2013;116:287–293. [14] Wu Y, Jin Y, Ding H, et al. In-line monitoring of extraction [23] Jiang C, Gong X, Qu H. Multivariate modeling and prediction of process of scutellarein from Erigeron breviscapus (vant.) breakthrough curves for herbal medicine adsorption on column chro- Hand-Mazz based on qualitative and quantitative uses of near-in- matography: a case study. Sep Sci Technol 2015;50(7):1030–1037. frared spectroscopy. Spectrochim Acta A Mol Biomol Spectrosc [24] Tranas R, Lovvik OM, Tomic O, et al. Lattice thermal conductiv- 2011;79(5):934–939. ity of half-Heuslers with density functional theory and machine [15] Wu HQ, Khan MA. Quality-By-Design (QbD): an integrated learning: enhancing predictivity by active sampling with principal approach for evaluation of powder blending process kinetics and component analysis. Comp Mater Sci 2022;202:110938. determination of powder blending end-point. J Pharm Sci-Us [25] Wold S, Esbensen K, Geladi P. Principal component analysis. 2009;98(8):2784–2798. Chemometr Intell Lab 1987;2(1–3):37–52. [16] Costa EB, Silva RC, Espejo-Roman JM, et al. Chemometric meth- [26] Wold S. Pattern-recognition by means of disjoint principal com- ods in antimalarial drug design from 1,2,4,5-tetraoxanes ana- ponents. Pattern Recogn 1976;8(3):127–139. logues. SAR QSAR Environ Res 2020;31(9):677–695. [27] Ledauphin J, Le Milbeau C, Barillier D, et al. Differences in the [17] Rukundo IR, Danao MC. Identifying turmeric powder by volatile compositions of French labeled brandies (Armagnac, source and metanil yellow adulteration levels using near-in- Calvados, Cognac, and Mirabelle) using GC-MS and PLS-DA. J frared spectra and PCA-SIMCA modeling. J Food Prot Agr Food Chem 2010;58(13):7782–7793. 2020;83(6):968–974. [28] Frank IE, Friedman JH. A statistical view of some chemometrics [18] Puchert T, Holzhauer CV, Menezes JC, et al. A new PAT/QbD regression tools. Technometrics 1993;35(2):109–135. approach for the determination of blend homogeneity: combina- [29] Luis ML, Garcia JM, Jimenez F, et al. Simultaneous determina- tion of on-line NIRS analysis with PC scores distance analysis tion of chlorthalidone and spironolactone with univariate and (PC-SDA). Eur J Pharm Biopharm 2011;78(1):173–182. multivariate calibration: wavelength range selection. J AOAC Int [19] Jimenez-Carvelo AM, Martin-Torres S, Ortega-Gavilan 1999;82(5):1054–1063. F, et al. PLS-DA vs sparse PLS-DA in food traceability. A [30] Liu W, Zhang S, Zu YG, et al. Preliminary enrichment and case study: authentication of avocado samples. Talanta separation of genistein and apigenin from extracts of pigeon 2021;224:121904. pea roots by macroporous resins. Bioresource Technol [20] Vieira LS, Assis C, de Queiroz M, et al. Building robust models for 2010;101(12):4667–4675. identification of adulteration in olive oil using FT-NIR, PLS-DA and variable selection. Food Chem 2021;345:128866. [21] Jiang C, Qu H. A comparative study of using in-line near-in- How to cite this article: Jiang C, Qu HB. In-line spectroscopy combined frared spectra, ultraviolet spectra and fused spectra to monitor with multivariate analysis methods for endpoint determination in column Panax notoginseng adsorption process. J Phar Biomed Anal chromatographic adsorption processes for herbal medicine. Acupunct Herb 2015;102:78–84. Med 2022;2(4):253–260. doi: 10.1097/HM9.0000000000000035

Journal

Acupuncture & Herbal MedicineWolters Kluwer Health

Published: Dec 22, 2022

References