Optimization of membrane dispersion ethanol precipitation process with a set of temperature control improved equipment

Ethanol precipitation is an important separation and purification process in the traditional Chinese medicines (TCMs) industry. In the present study, a membrane dispersion micromixer was applied to achieve good mixing for the ethanol precipitation process of Astragali radix concentrate. New experimental apparatus was set up to rapidly lower the temperature of ethanol solution before mixing with the concentrate. Ethanol precipitation process was optimized according to Quality by design concept. To identify critical material attributes (CMAs), ten batches of Astragali radix were used to prepare concentrates. Calycosin-7-O-β-D-glucoside content, the sucrose content, and the electrical conductivity were found to be CMAs after the correlation analysis and stepwise regression modelling. Definitive screening design was used to investigate the relationships among critical process parameters, CMAs, and process critical quality attributes (CQAs). Quadratic models were developed and design space was calculated according to the probability of attaining process CQA standards. A material quality control strategy was proposed. High quality and low quality Astragali radix concentrates can be discriminated by the inequalities. Low quality Astragali radix concentrates should not be released for ethanol precipitation process directly. Verification experiment results indicated accurate models and reliable design space. The temperature control method and control strategy are promising for ethanol precipitation process of other TCMs or foods.


Apparatus and ethanol precipitation process.
is the schematic diagram of the experimental setup. The membrane dispersion micromixer employed in this experiment was detailly described in previous work 13 . The sizes of mixing chamber (8 × 1 × 0.5 mm) were controlled by PTFE gaskets. The average pore size of stainless-steel membrane was 18 μm.
Ethanol solution at room temperature was cooled with a heat exchanger (K030-20M-NB4, Ningbo Gaori Technology Co., Ltd.). The heat exchanger was connected to the refrigeration circulation device (CA-1116A, Tokyo Rikakikai, Co. Ltd.). The heat transfer medium in the heat exchanger was 60% ethylene glycol-water (v/v). The low-temperature ethanol solution was served as the dispersed phase pumped into the micromixer by a gear pump (CT3001F, Baoding Reef Fluid Technology Co., Ltd.). A concentrate was pumped into the micromixer by an advection pump (2PB-20005II, Beijing Xingda Technology Development Co., Ltd.) as the continuous phase. The outlet mixture was collected in a jacketed beaker and magnetically stirred for 5 min. The temperature of the jacketed beaker was controlled by a thermostatic bath (THYD-1030W, Ningbo Tianheng instrument factory). The medium in the thermostatic bath connected to the jacketed beaker is 30% glycerol-water (v/v). The supernatant was collected after filtration. After an experiment, the apparatus was washed with 0.05% (wt.) Na 2 CO 3 solution and ethanol sequentially.
The low-temperature ethanol solution and a temperature-controlled jacketed beaker were used to control the temperature of the mixture together. The use of low temperature ethanol solution could reduce the temperature of the mixture in a short time. The temperature controlled jacketed beaker was used to control the temperature precisely. The temperature ranges are shown in Table S1.

Experimental design. The evaluation of ethanol precipitation process. The main active components of
Astragali radix were flavonoids and saponins. In Chinese Pharmacopeia, CG and Astragaloside IV are chosen as the representatives for flavonoids and saponins, respectively. In this study, three flavonoids of CG, PG, IFG, and two saponins of Astragaloside IV and Astragaloside II were measured as the representative active components of Astragali radix. In this work, the purity of five active components and dry matter removal were selected as the process CQAs.
CMAs identification. To identify the CMAs, the experiments were carried out with different concentrates under fixed process conditions. The mass ratio of ethanol solution to concentrate (ECR) was 1.5 g/g, the dry matter content of concentrates was 45%, the ethanol solution concentration was 92% (v/v), the refrigeration temperature was 15 °C, and the flow rate of concentrate was 60 mL/min. The electrical conductivity, flavonoid contents, saponin contents, and sugar contents of different concentrates were measured.
Definitive screening design. After identifying the CMAs, the definitive screening design was used to study the quantitative relationships among the potential CPPs, CMAs, and process CQAs. Many parameters can be studied in a very small number of experiments with the definitive screening experiment design 33 . Dry matter content www.nature.com/scientificreports/ (X 1 ), ECR (X 2 ), the ethanol solution concentration (X 3 ), and the temperature (X 4 ) were selected as potential CPPs because they were found to be CPPs in published works 30,34 . The experimental conditions are listed in Table 1. To study the effects of materials, different concentrates of Astragali radix were used in these experiments, as shown in Table 1.
Analytical methods. The contents of Astragaloside IV, Astragaloside II, CG, PG, and IFG were determined using a HPLC-ELSD method developed by Luo et al. 35 . A HPLC (1260, Agilent Technologies, USA) system with a UV detector and ELSD detector was used. Samples were diluted with 50% (v/v) methanol solution. Analyses were conducted on a Zorbax SB-C18 column (4.6 mm × 250 mm, 5 μm) with the column temperature controlled at 30 °C. The flow rate of solvent was maintained at 0.8 mL/min, while the injection volume of sample was set at 10 μL. The detection wavelength was fixed at 270 nm. The ELSD operation parameters were as follows: the evaporator temperature was fixed at 30 °C, nebulizer temperature was fixed at 80 °C, and gas flow rate was fixed at 1.6 L/min. The mobile phase was consisted of solvent A (0.2% (v/v) formic acid in water) and solvent B (acetonitrile  Figure S1. The HPLC method developed by Shao et al. was used to determine the contents of d-fructose and sucrose 36 . The HPLC system (1260, Agilent Technologies, USA) was equipped with an Alltech 2000ES ELSD detector. All the samples were diluted with 85% (v/v) acetonitrile-water mixture and carried out on a XBridge Amide column (5 μm, 4.6 × 250 mm; Waters, Milford, MA, USA). The column temperature was fixed at 34 °C. The solvent flow rate was fixed at 0.9 mL/min and the sample injection volume was set at 5 μL. The mobile phase solvent A was 0.3% (v/v) triethylamine in water and solvent B was 0.3% (v/v) triethylamine in acetonitrile. The solvent gradients were as follows: 0-37 min, 85-76% B; 37-38 min, 76-60% B; 38-48 min, 60-100% B. The re-equilibrium time was 10 min. The ELSD operation parameters were as follows: the nebulizer temperature was set at 65 °C, evaporator temperature was set at 60 °C, and gas flow rate was set at 1.8 L/min. A typical chromatogram of Astragali radix concentrate was shown in Figure S2.
Dry matter content was determined using a gravimetric method as described in previous work 28 . Each sample of the concentrates was diluted with water to a solution of 2% dry matter content. The conductivity of the diluted concentrates was measured using a portable conductivity meter (DDBJ-350, Hangzhou Qiwei Instrument Co., Ltd.  where Y is the process CQAs; a 0 is a constant; Z k represents a material attribute; and c k is the partial regression coefficient. Insignificant variables were removed by stepwise regression. The significance levels for adding terms and removing terms were both set to 0.1. The material attributes remaining in the model after stepwise regression were considered to be the CMAs. Quadratic models were developed based on the definitive screening designed experiment results. Equation (4) was used to model CPPs, CMAs, and process CQAs.
where n and m are the number of CPPs and CMAs, respectively; b and d are the partial regression coefficients; X i is a potential CPP; and Z C is a CMA. Stepwise regression was performed as before decribed. Data analysis was performed by Design Expert (version 11.0.0, Stat-Ease Inc., USA).

Results
Material attributes. The quality attributes of different Astragali radix concentrates are shown in Table 2.
The electrical conductivity was between 1442 and 2390 μS/cm, indicating different electrolyte contents in concentrates. The content of Astragaloside IV and Astragaloside II was lower than 2000 μg/g dry matter. The content of CG among the three flavonoid contents was the highest, which can exceed 1600 μg/g. The other flavonoids were less than 1000 μg/g dry matter. The sucrose content was higher than the d-fructose content, which could be more than 700 mg/g dry matter. At most occasions, sucrose was the main component of dry matter. The d-fructose content was lower than 30 mg/g dry matter.
The identification of CMAs. The results of the CMA identification experiments are shown in Table 3.
Though process conditions were fixed, the experimental results were quite different, indicating that the material attributes significantly affected the performance of Astragali radix ethanol precipitation process.
The correlation analysis of material attributes was carried out to find attributes with similar trends 37 . The Pearson coefficients are shown in Table 4. The Pearson coefficients among Astragaloside IV content (Z 1 ), Astragaloside II content (Z 2 ), PG content (Z 4 ), IFG content (Z 5 ) and electrical conductivity (Z 8 ) was higher than 0.90. It means that one of them can roughly represent other three material attributes because they contained similar information. Electrical conductivity (Z 8 ) was selected as the potential CMAs in the four material attributes because it is easy to measure. Other potential CMAs are CG (Z 3 ), d-fructose content (Z 6 ), and sucrose content (Z 7 ).
Stepwise regression method was used to determine CMAs 38 . In this method, the term left in linear equations after stepwise regression indicates a CMA 38 . The ANOVA results of multiple linear regression analysis of each CQA using Eq. (3) are shown in Table 5. The determination coefficient (R 2 ) of each model was higher than 0.70, indicating that the models can explain most of the variation of experimental data. However, these potential CMAs have no significant effect on the dry matter removal. It means that the determined material attributes were not main factors that influencing dry matter removal. According to the terms left in models, the CG content (Z 3 ), the sucrose content (Z 7 ), and the electrical conductivity (Z 8 ) were found to be CMAs. www.nature.com/scientificreports/ The effects of CMAs and CPPs. The partial regression coefficients and variance analysis results of the models are shown in Table 6. The P value of each model was less than 0.05, indicating that the model was significant. The adjusted determination coefficient ( R 2 adj ) of each model was higher than 0.84. The contour plots were obtained to analyze the effects of CPPs on CQAs, as shown in Figs. 3, 4, 5, 6. The dry matter removal increased as dry matter contents increased. The purity of CG decreased as as temperature increased. The purity of Astragaloside IV was mainly affected by CMAs.The dry matter removal was mainly affected by CPPs. The purity of other flavonoids and saponins was affected by both CPPs and CMAs.
Design space development. A Monte Carlo method was performed using a self-coded MATLAB program (R2016a, Version 9.0, The Math Works Inc., USA) to calculate the design space based on the specific goals of process CQAs. The calculation process was introduced in previous work 36 . The acceptable ranges of the CQAs and the probability requirements for compliance are shown in Table 7. 1000 simulations were carried out to get the probability of every possible condition.

Experimental no Concentrates
Purity of flavonoids and saponins in the supernatant (μg/g dry mater)

Dry matter removal (Y 6 )
Astragaloside IV (Y 1 )  www.nature.com/scientificreports/ The conditions of design space were listed in Table S2, and shown in Fig. 7a-d. The design space was an irregular region.

Astragaloside II (Y 2 ) CG (Y 3 ) PG (Y 4 ) IFG (Y 5 )
Control strategy of Astragali radix concentrates. In order to obtain a satisfactory supernatant, Inequalities (5) should be satisfied for CQA requirements listed in Table 7.  www.nature.com/scientificreports/ where superscripts refer to dry matter removal, purity of Astragaloside IV, purity of Astragaloside II, purity of CG, purity of PG, purity of IFG, and dry matter removal, respectively. The values of regression coefficients in Inequalities (5) can be found in Table 6. If the CMAs of a batch of Astragali radix concentrates meet Inequalities (5), the batch of Astragali radix concentrates is considered to be acceptable for ethanol precipitation. For a batch of acceptable Astragali radix concentrates, feasible process parameters can be chosen after calculation or selected from Table S2. A batch of Astragali radix concentrates is considered to be unacceptable when Inequalities (5) cannot be satisfied. In industry, the process parameters are usually fixed. If the process parameters are fixed as follows: the ECR is 1.5 g/g, the dry matter content of concentrates is 45%, the ethanol solution concentration is 92% (v/v), and the temperature is 15 °C, Inequalities (5) can be simplified to Inequalities (6).  www.nature.com/scientificreports/ If a batch of Astragali radix concentrates with CMAs meeting Inequalities (6), this batch of Astragali radix concentrates is considered to be high quality material for the current parameter fixing process. If not, it is considered a low-quality material and should not be released for ethanol precipitation directly.

Examples of material quality control and verification experiments. The CMAs of 3 batches of
Astragali radix concentrates were measured and are shown in Table 8. According to Inequalities (6), Astragali radix concentrates of N12 were low quality Astragali radix concentrates, and it should not be released. The design space calculation results of N12 was show in Figure S3. The results shown that when the materials were unqualified, no matter how to change the CPP, the standards of CQA can not be achieved with a high probability.
N11 and N13 were high quality Astragali radix concentrates. The verification experiment conditions and results are listed in Table 9 and Fig. 7e,f. All the predicted values were close to the experimental values, indicating that the models had good predictive performance.

Discussions
The effects of ethanol precipitation process on quality variation. The quality variations of most of Chinese medicines are from raw materials 39 . These variations may transmit from upstream intermediates to drug products. As a purification process, ethanol precipitation process is usually expected to reduce these variations.
The relative standard deviation (RSD) values of active component contents in the concentrates and ethanol precipitation supernatant of Experiment 1-10 were calculated, which are listed in Table 10. The RSD of active component contents were not decreased significantly. Active component contents in the precipitate of Experiment 2 was also analyzed. Astragaloside IV, Astragaloside II, CG, PG, and IFG were not found in the precipitate. It indicated that the variation of non-precipitated component contents could not be significantly reduced by ethanol precipitation process.
For a Chinese medicine prepared with a series of unit operations containing ethanol precipitation, it is not enough to ensure the quality consistency of drugs only by the control of ethanol precipitation process parameters. Strict quality control of raw materials is sometimes more important. Mixing different batches of raw materials was an effective way to improve the quality consistency of raw materials. This strategy was introduced in many published works 39,40 .    13 . Because of larger viscosity increase of an Astragali radix concentrate, we did not try to cool a concentrate before mixing with an ethanol solution in this work. Because of smaller viscosity change after lowering temperature, the ethanol solution was cooled before mixing with a concentrate.
Advantages and disadvantages. In industry, ethanol solutions and herbal concentrates were mixed in a stirred tank at most occasions. After that, the ethanol precipitation system was cooled by putting the stirred tank  www.nature.com/scientificreports/ in a cold storage, or pumping cooling water through tank jacket. In general, the larger the volume of a stirred tank, the slower the temperature decreases. In this work, the ethanol solution was cooled before mixing with an Astragali radix concentrate. Therefore, the system temperature would be lowered after mixing. Compared with conventional methods in industry, the cooling time can be probably shortened.
The effects of two mixing methods of micromixing and stirring on ethanol precipitation were compared in previous work using Codonopsis Radix concentrates as the processing objects 13 . The mixing effect was better when using a membrane dispersion micromixer, which led to less loss of the active component 13 .
There are some shortcomings of this work. Firstly, there was no online detection of liquid temperature, neither a feedback control of cooling. In the future, if some automatic control method can be applied, such as Programmable Logic Controller (PLC) programs, the control accuracy of temperature can be further improved. Secondly, the models were established based on the results of a small number of experiments carried out in this study. In industry, it is necessary to accumulate production big data and update the models regularly to make the prediction results more reliable. This idea is also in line with the concept of "continuous improvement" mentioned in Dr. Yu's paper 42 .

Conclusion
In this study, a membrane dispersion continuous ethanol addition device which can achieve rapid cooling was developed for Astragali radix ethanol precipitation. The ethanol precipitation process was then optimized according to QbD concept. The experiments were carried out with different concentrates under fixed process conditions to identified the CMAs. CG content, the sucrose content, and the electrical conductivity were found to be CMAs. Definitive screening design was used to investigate the relationships among CPPs, CMAs, and CQAs. After model development, it is found that dry matter removal was mainly affected by CPPs. The purity of Astragaloside IV was mainly affected by CMAs. The purity of Astragaloside II, PG, and IFG were affected by both CPPs and CMAs. The design space was then calculated according to the probability of attaining process CQA standards. A material quality control strategy was proposed. High quality and low quality Astragali radix concentrates can be discriminated by the inequalities. Low quality Astragali radix concentrates should not be released for ethanol process directly. Verification experiments were carried out for high quality Astragali radix concentrates. The experimental results agreed well with the prediction results. The control strategy proposed in this work is promising to be used in other processes to improve batch-to-batch consistency of TCMs or herbal medicines.

Data availability
All data generated or analyzed during this study are included in this published article and its supplementary information files. www.nature.com/scientificreports/