Exocrine lyophilization process based on open spectrum RD data processing

Abstract

The present application relates to the technical field of biological product preservation, process analysis and automatic control, in particular to a freeze-drying process of excretion based on open spectrum RD data processing; comprising a special freeze-drying protective solution, gradient cooling freezing, one-time and two-time drying processes, and an online spectrum acquisition and analysis module; the system combines the characteristics of exosomes, collects near-infrared or Raman spectrum in the drying stage, and calculates the activity retention rate, residual moisture prediction value and abnormal index through open spectrum RD data processing; the core is to use multi-parameter fusion for stage conversion and end point joint determination, and dynamically feedback adjust temperature and pressure parameters in the process design space; the present application solves the problem that the traditional fixed program cannot cope with the batch difference of exosomes, overcomes the lag defect of offline sampling evaluation, and realizes non-contact continuous monitoring and dynamic optimization of the freeze-drying process.

Description

Technical Field

[0001] This invention relates to the fields of biological product preservation, lyophilized formulations, process analysis technology and automated control technology, specifically to an exocrine lyophilization process based on open-spectrum RD data processing. Background Technology

[0002] With the widespread application of exosomes in the biomedical field, the demand for long-term stable preservation of exosome preparations has increased significantly. The complexity of this biomembrane structure and its contents brings many challenges, especially in the thermodynamic management of the freeze-drying process and continuous monitoring of bioactivity. Currently, freeze-drying processes are typically performed by setting fixed plate temperatures and chamber pressures. Technicians then perform destructive sampling after freeze-drying and assess the particle size distribution and membrane integrity of exosomes using cell function assays and physicochemical testing tools, recording the final quality information for each batch of product. However, traditional fixed-process freeze-drying and offline evaluation methods rely on post-process sampling and delayed biological testing, requiring technicians to perform cumbersome efficacy determinations only after the process is complete. While these methods can provide detailed information on the state of freeze-dried exosomes, the evaluation is usually severely delayed, making it difficult to detect vesicle collapse and activity impairment caused by batch differences in materials or process thermodynamic fluctuations in real time. Therefore, how to promptly detect abnormalities in the exosome freeze-drying process and prevent loss of activity has become an urgent problem to be solved in this field. Summary of the Invention

[0003] The purpose of this invention is to provide a highly active exocrine lyophilization process. Existing exosome lyophilization technologies have limitations in addressing batch-to-batch variations of exosomes, continuous monitoring of complex state changes during the lyophilization process, endpoint determination based on exosome structure and functional state, and offline sampling for activity evaluation. Therefore, this invention provides a process that combines online spectral detection, offline biological calibration, and constrained feedback control to achieve non-contact continuous monitoring and dynamic optimization of the lyophilization process. The specific technical solution is as follows: The exocrine lyophilization process based on open-spectrum RD data processing includes: preparing a lyophilization protectant containing non-reducing sugars, structural support agents, and buffers, and adjusting its pH; mixing the lyophilization protectant with the exosome preparation and filling it; subjecting the filled preparation to gradient cooling freezing, a first vacuum drying, and a second heating drying in sequence; acquiring near-infrared spectroscopy, Raman spectroscopy, or a combination thereof at at least one drying stage, and obtaining the predicted value of activity retention rate, the predicted value of residual moisture, the prediction uncertainty, and the spectral anomaly index through open-spectrum RD data processing; The aforementioned open-spectrum RD data processing refers to processing the online acquired spectral data and fusing the processing results with the measured process parameters of product temperature, plate temperature, and cavity pressure obtained synchronously according to the sampling time.

[0004] Preferably, the preparation of the lyophilization protective solution includes dissolving non-reducing sugars, structural support agents, and buffers in water for injection, adjusting the pH to 6.8-7.4; adding a surface stabilizer and allowing it to stand to remove bubbles; and mixing it with the exosome preparation after sterile filtration; wherein the non-reducing sugars are selected from one or more of trehalose, sucrose, and raffinose, and their total concentration is 2%-12%; The structural support is selected from one or more of mannitol and glycine, with a concentration of 0.2%-5%; the buffer is selected from histidine, citrate or phosphate, with a concentration of 5-25 mmol / L; the surface stabilizer is selected from polysorbate 20, polysorbate 80 or poloxamer, with a concentration of 0.001%-0.05%.

[0005] Preferably, the particle concentration of the exosome preparation is [missing information]. The total protein concentration is 0.05-5 mg / mL, the particle size range is 30-300 nm, and the osmotic pressure is 250-400 mOsm / kg.

[0006] Preferably, the gradient cooling freezing includes cooling to -55°C to -35°C at a rate of 0.1-2°C / min and holding at that temperature for 0.5-3 hours, followed by annealing at -30°C to -15°C for 0.5-2 hours; the chamber pressure of the primary vacuum drying is 5-50 Pa, and the upper limit of the product temperature is controlled to be the temperature value obtained by subtracting 2°C to 5°C from the preset collapse temperature threshold; the secondary heating drying includes heating to 10-35°C at a rate of 0.05-0.5°C / min and holding at that temperature for 2-10 hours; The secondary heating and drying process ends when the predicted residual moisture value reaches the preset range, the predicted activity retention rate fluctuates less than the preset threshold within the continuous sampling period, and the prediction uncertainty is within the preset range.

[0007] Preferably, the open spectrum RD data processing includes performing dark current correction, empty bottle background subtraction, freeze-drying protective liquid blank correction, abnormal spectrum removal, spectral preprocessing, and feature extraction on the original spectrum; and fusing moisture characteristics, carbohydrate solid state characteristics, protein or lipid-related characteristics with product temperature, plate temperature, and cavity pressure.

[0008] Preferably, the spectral preprocessing includes one or more of the following: standard normal variable transformation, multiplicative scattering correction, detrending, Savitsky-Gorye smoothing, first derivative, second derivative, fluorescence background subtraction, baseline correction, or peak position correction.

[0009] Preferably, the predicted activity retention rate is obtained by a partial least squares regression, support vector regression, random forest, or neural network model based on the spectral data of at least two batches of exosomes and the offline detection results of unit particle functional efficacy, membrane integrity, particle number recovery rate, particle size distribution, key markers, or combinations thereof.

[0010] Preferably, a collapse temperature threshold is preset. When the product temperature reaches the collapse temperature threshold, the predicted value of activity retention rate continuously decreases, the spectral anomaly index exceeds a preset limit, or the cavity pressure fluctuation exceeds a preset fluctuation threshold, the plate temperature is reduced, the heating rate is reduced, the cavity pressure is adjusted, or the heat preservation time of the current drying stage is extended. When the spectrum deviates from the preset applicable range of the model training set, the prediction deviation exceeds the preset verification range, the reference sample calibration fails, or the spectral equipment malfunctions, the automatic control based on the predicted value of activity retention rate is suspended, and the process is switched to the backup freeze-drying program.

[0011] Preferably, the freeze-drying endpoint is also determined jointly based on the difference between the product temperature and the plate temperature, the difference between the cavity pressure and the reading of the capacitor diaphragm vacuum gauge, the pressure rise test results, and the moisture-related spectral characteristics.

[0012] The beneficial effects of this invention are as follows: By combining non-reducing sugar protectants, structural support agents, and buffer systems, the effects of freeze-drying concentration, dehydration, and interfacial effects on the structure and function of exosome membranes are reduced; controlled freezing, primary drying, and secondary drying are used to match the product temperature with the formulation's critical temperature, reducing the risk of collapse, shrinkage, and structural instability; and continuous or intermittent monitoring of the freeze-drying process is achieved by collecting near-infrared spectroscopy, Raman spectroscopy, or a combination thereof at least one drying stage. Detailed Implementation

[0013] The technical solutions in the embodiments of the present invention will be clearly and completely described below; the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0014] Example 1

[0015] This embodiment uses an exosome preparation derived from mesenchymal stem cells, and specifically includes the following steps: S1. Preparation of exosome preparations The exosome preparation, purified by size exclusion chromatography and concentrated by ultrafiltration, was subjected to buffer replacement to adjust the particle concentration of the exosome preparation to [value missing]. The total protein concentration was 0.42 mg / mL, the main particle size range was 65-185 nm, the pH was 7.0, and the osmotic pressure was 310 mOsm / kg. Before lyophilization, the number of particles, particle size distribution, membrane integrity, CD9, CD63 and CD81 markers, and receptor cell uptake capacity were detected, and the results of the pre-lyophilization detection were used as the baseline value for calculating the activity retention rate. S2, Preparation of lyophilization protective solution Weigh out trehalose, mannitol and histidine buffer, add them to water for injection and dissolve them to make the concentration of trehalose 6.0% w / v, mannitol 1.0% w / v and histidine buffer 10 mmol / L. Adjust the pH of the protective solution to 7.0, add polysorbate 20 to a concentration of 0.005% w / v, and allow it to stand to remove bubbles; after sterile filtration, test the pH, appearance and osmotic pressure of the protective solution. Lyophilized protective solution without exosomes, empty bottles, and reference samples with known activity levels were used for spectral background correction and model suitability checks. S3, Low-shear mixing and filling The lyophilization protectant was mixed with the exosome preparation at a volume ratio of 1:1. The mixing process was carried out using a low-shear method to avoid shear rates exceeding the set threshold and prolonged stirring. After mixing, the mixture was filled into 2 mL lyophilization vials, with each vial containing 0.75 mL. Production samples, spectral monitoring samples, and offline destructive sampling samples all used the same formulation, volume, bottle type, and stopper condition, and were placed in validated representative positions. S4, Freezing Phase After filling, the lyophilized vials are placed in a lyophilizer and cooled to -45°C at a rate of 0.5°C / min. The temperature is maintained for 1 hour to allow the formulation to complete a controlled phase change and crystallize and solidify completely. During the freezing stage, product temperature, plate temperature, cavity pressure, and near-infrared spectrum are collected. The conditions for determining whether freezing is complete are that the difference between product temperature and plate temperature is within a set range, the product temperature remains stable, and the rate of change of moisture-related spectral characteristics is lower than a set threshold. S5, First Drying Stage Adjust the pressure in the freeze dryer chamber to 20 Pa, raise the plate temperature to -25°C, and approach the set plate temperature at a heating rate of 0.15°C / min. The collapse temperature of this formula was determined to be -32℃ by freeze-drying microscopy and thermal analysis. During the single drying process, the product temperature was controlled within the range of no higher than -35℃, so that the product temperature was more than 3℃ lower than the collapse temperature. Moisture-related characteristics were collected using near-infrared spectroscopy, and protein and lipid-related characteristics were collected using Raman spectroscopy. The spectral characteristics were then matched with the product temperature, plate temperature, cavity pressure, and pressure rise test results. When the product temperature and the plate temperature tend to be consistent, the difference between the cavity pressure and the reading of the capacitor diaphragm vacuum gauge reaches the set range, the pressure rise test shows that the sublimation rate is lower than the set limit, and the moisture-related spectral characteristics are continuous and stable, it is determined that one drying cycle is completed. S6, Secondary Drying Stage After the first drying, the temperature of the plate was raised to 25℃ at a heating rate of 0.2℃ / min and held for 4 hours; during the second drying, near-infrared spectra were continuously collected, and the predicted values ​​of residual moisture, activity retention rate, prediction uncertainty, and spectral anomaly index were calculated. When the predicted residual moisture content reaches the set range, the predicted activity retention rate remains stable within the continuous sampling period, and the prediction uncertainty and spectral anomaly index are within the allowable range, the secondary drying process ends. S7, Data Processing and Feedback Control The original spectrum was sequentially subjected to dark current correction, empty bottle background subtraction, lyophilization protective liquid blank correction, abnormal spectrum identification, standard normal variable transformation, first derivative processing and feature extraction. The data related to moisture, solid state of carbohydrates, protein and lipids were fused with product temperature, plate temperature and cavity pressure, and then input into a partial least squares regression model established by spectral data from multiple exosome batches and offline detection results. Offline calibration metrics include functional efficacy per unit particle, membrane integrity, particle number recovery, particle size distribution, and retention rates of CD9, CD63, and CD81 markers; model outputs include predicted activity retention, predicted residual moisture, prediction uncertainty, and spectral anomaly index. When the predicted activity retention rate decreases continuously within the applicable range of the model, and the prediction uncertainty and spectral anomaly index meet the control conditions, reduce the plate temperature or reduce the plate heating rate; when the spectrum deviates from the applicable range of the model training set, the prediction deviation exceeds the validation range, or the reference sample calibration fails, pause the automatic control based on the predicted activity retention rate and switch to the validated conservative freeze-drying procedure. After freeze-drying, the product is plugged and sealed under vacuum conditions; To clarify the terminology, the open-spectrum RD data processing referred to in this embodiment refers to processing the online acquired near-infrared spectrum, Raman spectrum, or a combination thereof, and fusing the processing results with the process measurement parameters obtained synchronously according to the sampling time; wherein, open-spectrum refers to the online acquisition of the near-infrared spectrum, Raman spectrum, or a combination thereof, and RD refers to the original spectral data and the process measurement parameters synchronized with it; The raw spectral data includes sampling time, wavelength or Raman shift, spectral intensity and reference sample number. The measured process parameters include product temperature, plate temperature and cavity pressure. Each sampling moment forms a data record. After processing, the predicted residual moisture value, predicted activity retention rate value, prediction uncertainty and spectral anomaly index are output. The output values ​​and sampling time are written into the control record. In specific processing, first subtract the detector background from the dark current spectrum, then subtract the empty bottle spectrum and the blank spectrum without exosome lyophilization protection solution; spectral intensity saturation, signal-to-noise ratio below 20, or Q deviation from the principal component model of the same stage are considered. If all statistics exceed the 99% control limit, the spectrum is removed; the retained spectra are smoothed using an 11-point, second-order polynomial Savitzky-Golay method, and then subjected to standard normal variable transformation and first-order derivative processing. Near-infrared spectroscopy was used to extract moisture-related variables at 970-990 nm, 1400-1500 nm, and 1880-1980 nm, while Raman spectroscopy was used to extract... The peak area, peak height, and peak position variables of proteins and lipids within the range; the spectral variables, along with product temperature, plate temperature, and cavity pressure, are spliced ​​together after batch-level standardization to form a fusion input matrix; The model was built using 8 independent batches, with 12 predetermined time points collected for each batch: one time point at the start of the freezing stage, one time point when the product temperature reaches -30℃, one time point when the product temperature reaches -45℃, and one time point at the end of the -45℃ holding period of 1 hour (i.e., the end of freezing); one time point at the 2nd, 4th, and 6th hours after the start of the first drying stage, and one time point at the end of the first drying stage; and one time point at the 1st, 2nd, 3rd, and 4th hours of the second drying holding period. Spectral data acquisition and offline destructive sampling were performed at each of the above time points, and the corresponding product temperature, plate temperature, cavity pressure, and process stage were recorded, resulting in a total of 96 paired samples. Five-fold cross-validation grouped by batch was used to determine the number of latent variables for partial least squares regression, with the minimum root mean square error of cross-validation and the number of latent variables not exceeding 10 as selection criteria. When support vector regression was used, the radial basis function was selected as the kernel function, and the penalty parameter and kernel parameter were determined through grid search combined with five-fold cross-validation. When using a random forest model, the number of decision trees is set to 100 to 500, and the maximum number of features for node splits is the square root of the total number of input variables. When using a neural network model, the network structure is a multilayer perceptron with at least two hidden layers, the number of nodes in the hidden layers decreases layer by layer, the activation function is a linear rectified function, and an adaptive moment estimator optimizer is used for gradient descent. Activity retention was calculated as the percentage of each index after lyophilization relative to the baseline value before lyophilization, and normalized according to the weights of 0.40, 0.20, 0.15, 0.15 and 0.10 for unit particle functional efficacy, membrane integrity, particle number recovery, retention of CD9, CD63 and CD81 markers and particle size distribution retention, respectively. Residual moisture content was measured as a mass percentage using the Karl Fischer method, serving as a modeling reference. In external validation, the coefficient of determination for the activity retention model was used. The coefficient of determination for the residual moisture model is 0.93, with a root mean square error of 3.2 percentage points; The value is 0.96, and the root mean square error is 0.18 percentage points. Prediction uncertainty is expressed as the 95% prediction interval half-width and calculated according to the model bias standard deviation and sample leverage value; when the prediction uncertainty is no more than 5 percentage points, the predicted value of the activity retention rate is allowed to participate in automatic control; the spectral anomaly index is calculated according to the robust Mahalanobis distance of the fused input variable relative to the center of the modeled sample, and when the spectral anomaly index is no more than 3.0, the spectrum is determined to be within the applicable range of the model; The residual moisture content during secondary drying is controlled between 0.8% and 2.0% w / w. The maximum difference in the predicted activity retention rate over three consecutive sampling periods does not exceed 2 percentage points, and the prediction uncertainty is no greater than 5 percentage points and the spectral anomaly index is no greater than 3.0. The secondary drying process is terminated when the sampling period is 5 minutes. The specific criteria for determining the completion of freezing are: the difference between the product temperature and the plate temperature is no greater than 3℃ and maintained continuously for 10 minutes, and the relative change rate of the moisture-related spectral characteristics in three consecutive samples does not exceed 2%; the specific criteria for determining the completion of drying in one step are: the difference between the product temperature and the plate temperature is no greater than 2℃ and maintained continuously for 10 minutes, the difference between the chamber pressure and the reading of the capacitive diaphragm vacuum gauge is no greater than 5Pa, the pressure rise test is: the pressure rise value after stopping evacuation for 30 seconds is no greater than 3Pa, and the relative change rate of the moisture-related spectral characteristics in three consecutive samples does not exceed 2%. When the product temperature is higher than -35℃, the predicted activity retention rate decreases by more than 2 percentage points for three consecutive sampling cycles, the spectral anomaly index is greater than 3.0, or the pressure rise is greater than 3 Pa / 30 seconds, the plate temperature is reduced by 2℃ or the heating rate is reduced by 0.05℃ / min; after adjustment, maintain the adjustment for 15 minutes and re-evaluate; when the deviation between the chamber pressure and the set value exceeds 5 Pa for two consecutive sampling cycles, the chamber pressure is adjusted back to 20 Pa and the heating is paused; If the spectral anomaly index is greater than 3.0 for three consecutive times, the prediction uncertainty is greater than 5 percentage points, the reference sample calibration deviation exceeds 5 percentage points, or the spectroscopic equipment malfunctions, then the spectral prediction value should be discontinued, and a conservative freeze-drying program should be switched to a plate temperature not higher than -25℃, a chamber pressure of 20Pa, a heating rate of 0.10℃ / min, and a secondary drying and heat preservation time extended by 1 hour.

[0016] Example 2

[0017] This embodiment uses the same exosome preparations obtained from the same source, with the same preparation process and quality standards as Example 1, and each batch is prepared independently. The composition of the lyophilization protective solution, the particle concentration of the exosome preparation, the fill volume, and the freezing and drying parameters are adjusted based on Example 1, as follows; The lyophilization protection solution contained 4.0% w / v trehalose, 0.5% w / v mannitol, 15 mmol / L histidine buffer, 0.002% w / v polysorbate 20, and pH 6.8. The particle concentration of the exosome preparation was adjusted to The particle count was 0.5 mL; during the freezing stage, the temperature was lowered to -35°C at 0.2°C / min and held for 0.5 hours; during the first drying stage, the chamber pressure was 30 Pa and the plate temperature was -30°C; for the formulation of this embodiment, after the protective solution and exosome preparation were mixed at a 1:1 ratio, the glass transition characteristics were measured by lyophilization microscopy, and the temperature was observed at 30 Pa with a heating rate of 0.1°C / min. The temperature at which the first collapse of the lyophilized layer structure occurred was taken as the collapse temperature, and the collapse temperature of this formulation was measured to be -34°C; and the glass transition characteristics were verified by differential scanning calorimetry; during the first drying process, the product temperature was controlled within the range not exceeding -38°C, so that the product temperature was 4°C lower than the collapse temperature of this formulation; The secondary drying stage involves increasing the temperature to 20°C at a rate of 0.1°C / min and holding for 6 hours. The methods for spectral acquisition, background correction, model prediction, endpoint determination, and feedback control are the same as in Example 1.

[0018] Example 3

[0019] This embodiment uses the same exosome preparations obtained from the same source, with the same preparation process and quality standards as Example 1, and each batch is prepared independently. The composition of the lyophilization protective solution, the particle concentration of the exosome preparation, the fill volume, and the freezing and drying parameters are adjusted based on Example 1, as follows; The lyophilization protectant contained 8.0% w / v trehalose, 2.0% w / v mannitol, 10 mmol / L histidine buffer, 0.010% w / v polysorbate 20, and pH 7.2. The particle concentration of the exosome preparation was adjusted to particles / mL, with a volume of 1.0mL; during the freezing stage, the temperature was lowered to -50℃ at a rate of 1.0℃ / min, held for 2 hours, and then annealed at -20℃ for 1 hour; The chamber pressure during the first drying stage was 10 Pa, and the plate temperature was -20°C. For the formulation obtained by mixing the protective solution and exosome preparation in a 1:1 ratio in this embodiment, lyophilization microscopy was used. The temperature was increased by 0.1°C / min at 10 Pa, and the collapse temperature was determined to be -30°C, with the temperature at which the first collapse of the lyophilized layer structure occurred. Differential scanning calorimetry was used to verify its glass transition characteristics. During the first drying process, the product temperature was controlled within a range not exceeding -32°C, ensuring that the product temperature was 2°C below the collapse temperature of the formulation. In the secondary drying stage, the temperature is increased to 30℃ at a rate of 0.3℃ / min and held for 3 hours. Near-infrared spectroscopy and Raman spectroscopy were used for joint acquisition; near-infrared spectroscopy was used to extract residual moisture and solid state characteristics of sugars, and Raman spectroscopy was used to extract protein and lipid-related characteristics; when the difference between the prediction results of the two spectra was within a set range, the fusion model was used to output the predicted value of activity retention rate; other data processing and feedback control methods were the same as in Example 1.

[0020] Example 4

[0021] This embodiment uses the same exosome preparation from the same source and batch as Example 1, and the operation steps are the same as in Example 1, only the protection system and drying parameters are adjusted. The freeze-drying protectant solution uses a combination of trehalose and sucrose, with trehalose concentration of 3.0% w / v, sucrose concentration of 3.0% w / v, mannitol concentration of 0.8% w / v, histidine buffer concentration of 20 mmol / L, polysorbate 20 concentration of 0.003% w / v, and pH of 7.4. During the freezing stage, the temperature is lowered to -40℃ at a rate of 0.8℃ / min and held for 1.5 hours. During the primary drying stage, the chamber pressure is 40Pa, the plate temperature is -28℃, and the product temperature is controlled 3℃ below the verified collapse temperature. In the secondary drying stage, the temperature is increased to 15°C at a rate of 0.05°C / min and held for 10 hours. The residual moisture content is predicted by near-infrared spectroscopy, and the activity retention rate is predicted by the fusion results of near-infrared and Raman spectroscopy. The prediction uncertainty and spectral anomaly index are used as automatic control permission conditions. Other operations are the same as in Example 1.

[0022] Comparative Example 1: The difference between this comparative example and Example 1 is that online spectral acquisition, activity retention rate prediction, and feedback control based on prediction results are omitted. The first and second drying are completed using fixed plate temperature, fixed chamber pressure, and fixed drying time. Other operating steps and process parameters are exactly the same as in Example 1. The first drying stage uses a chamber pressure of 20 Pa and a plate temperature of -25 °C, and the first drying is completed within the preset time. The second drying stage uses a plate temperature of 25 °C and is kept at that temperature for 4 hours. The process parameters are not adjusted based on the predicted values ​​of residual moisture, activity retention rate, or prediction uncertainty.

[0023] Comparative Example 2: The difference between this comparative example and Example 1 is that the structural support agent mannitol is omitted, the concentration of trehalose in the freeze-drying protective solution is still 6.0% w / v, and the concentrations of other components, mixing methods, freezing process, primary drying process, secondary drying process, spectral acquisition and feedback control methods are exactly the same as in Example 1.

[0024] Comparative Example 3: The difference between this comparative example and Example 1 is that the non-reducing sugar trehalose is omitted, and the volume of the protective solution is made up with an equal volume of histidine buffer. Mannitol, histidine buffer and polysorbate 20 are retained. Other operating steps and process parameters are exactly the same as in Example 1.

[0025] Comparative Example 4: The difference between this comparative example and Example 1 is that the product temperature in the first drying stage is controlled to be 0.5°C lower than the verified collapse temperature, rather than 3°C lower than the collapse temperature. The other operating steps and process parameters are exactly the same as in Example 1.

[0026] Comparative Example 5: The difference between this comparative example and Example 1 is that online spectral acquisition and activity retention rate prediction are retained, but prediction uncertainty and model applicability judgment are omitted. No control is made on whether the spectrum deviates from the model training set range, prediction bias, and reference sample calibration results. Other operating steps and process parameters are exactly the same as in Example 1.

[0027] Comparative Example 6: The difference between this comparative example and Example 1 is that the chamber pressure in the first drying stage is adjusted from 20 Pa to 80 Pa, which exceeds the pressure range of a single vacuum drying stage. Other operating steps and process parameters are exactly the same as in Example 1.

[0028] Performance Tests and Data Sheets: Particle number recovery was calculated using particle number detection before and after lyophilization; particle size distribution change and polydispersity index were calculated using particle size analysis; membrane integrity retention was calculated using membrane integrity testing; and unit particle functional efficacy retention was calculated using receptor cell uptake and application-specific functional efficacy testing. Residual moisture was determined using the Karl Fischer method. Reconstitution time was calculated based on the time required for the formation of a homogeneous liquid after adding the specified reconstitution medium to the lyophilized bottle. Batch-to-batch quality fluctuations were expressed as the relative standard deviation of functional efficacy retention. Table 1. Performance test results of lyophilized serotonin formulations in each example and comparative example.

[0029] Each of the above embodiments and comparative examples underwent three independent batch tests, with 10 vials filled in each batch. The exosome preparations from the same source and batch mentioned in Examples 2-4 refer to those with the same exosome source, preparation process, and quality standards, rather than using the same measured production batch as in Example 1. Five vials from each batch were randomly selected for testing of particle number, particle size distribution, membrane integrity, marker retention rate, and functional efficacy, with three technical replicates performed for each vial and the average value taken. Five vials from each batch were randomly selected for testing of residual moisture, with one test per vial. Ten vials from each batch were randomly selected for testing of reconstitution time, with one test per vial. The results in the table are the overall average of three independent batch results. The relative standard deviation of functional efficacy between batches is calculated by dividing the standard deviation of the three batch means by the batch means and then multiplying by 100%. Except for the relative standard deviation of functional efficacy listed in the table, the relative standard deviation of other indicators in each group between batches does not exceed 8.0%. Particle count was determined using a nanoparticle tracking analyzer, and the sample was diluted to approximately [value missing] with sterile buffer. - For each sample, five 60-second video segments were collected. The particle count recovery rate was calculated by multiplying the ratio of the particle concentration after freeze-drying to the particle concentration before freeze-drying by 100%, and corrected for dilution factor. The average particle size and particle size distribution were determined using a nanoparticle tracking analyzer, and the polydispersity index was determined using a dynamic light scattering instrument. The change in average particle size was calculated by dividing the absolute value of the difference between the average particle size after freeze-drying and the average particle size before freeze-drying by the average particle size before freeze-drying and then multiplying by 100%. Membrane integrity was determined by a combination of nanoflow cytometry and a green nucleic acid fluorescent dye membrane impermeability assay. The sample was added to a final concentration of 1 μmol / L with green nucleic acid fluorescent dye and incubated at room temperature in the dark for 10 minutes before detection. Samples from the same batch treated with 0.1% polyethylene glycol octylphenyl ether were used as controls for complete membrane destruction. The membrane integrity retention rate was calculated by dividing the proportion of green nucleic acid fluorescent dye negative particles after lyophilization by the proportion of green nucleic acid fluorescent dye negative particles before lyophilization and then multiplying by 100%. CD9, CD63, and CD81 were captured by magnetic beads and then detected by flow cytometry with fluorescently labeled antibodies. The marker retention rate was calculated by multiplying the ratio of the relative fluorescence signal after lyophilization to the relative fluorescence signal before lyophilization by 100%. The uptake capacity of recipient cells was detected by DiD fluorescent labeling of exosomes and flow cytometry. Exosomes were incubated at 37°C in the dark for 20 minutes with a final DiD concentration of 1 μmol / L. After removing the free dye, they were co-incubated with recipient cells for 4 hours. Extracellular fluorescence was quenched by adding 0.2% trypan blue, and then the intracellular DiD positivity rate and average fluorescence intensity were detected. Functional effectiveness per unit particle The normalized composite value of the DiD positivity rate of the receptor cells corresponding to each exosome and the average fluorescence intensity is expressed as follows: the retention rate of functional efficacy per unit particle is calculated by multiplying the composite value after lyophilization to the composite value before lyophilization by 100%. Residual moisture was determined using a coulometric Karl Fischer moisture analyzer. 20-50 mg of lyophilized powder was taken from each vial, sealed, and transferred to the detection cell. The drift was measured until it did not exceed 10 μg / min for 60 consecutive seconds. The result was expressed as a percentage of the mass of the lyophilized powder. For reconstitution, 0.75 mL of water for injection was added to each vial. The vial was gently inverted at 25±2℃, and the results were observed every 10 seconds. The endpoint was reached when the lyophilized cake completely disappeared and the liquid was homogeneous. The time from the addition of the reconstitution medium to the reaching of the endpoint was recorded. All test items were based on the same batch of exosome preparations before lyophilization. Two-sided one-way ANOVA was used to compare the examples and comparative examples. Welch correction was used when the variances were unequal. Tukey correction was used for multiple comparisons between groups. The significance level was set at 0.05. The online model prediction results were independently validated against the offline detection results. A comparison was made between 36 paired samples from 3 batches that were not involved in the modeling. The root mean square error (RMSE) of the predicted activity retention rate compared to the offline comprehensive activity retention rate was 3.2 percentage points, and the mean absolute error was 2.5 percentage points. The RMS error of the predicted residual moisture compared to the Karl Fischer method results was 0.18 percentage points, and the mean absolute error was 0.14 percentage points. The deviations between the predicted values ​​and the offline results were all within the pre-set 95% prediction range. To ensure the reproducibility of the calculation of overall activity retention rate, the normalization baseline for all indicators of the same batch of exosome formulations before lyophilization was set to 100%; Represents the retention rate of functional effectiveness per unit particle, in Indicates the membrane integrity retention rate, in The particle number recovery rate is expressed as... This represents the arithmetic mean of the retention rates of the three markers CD9, CD63, and CD81, expressed as... This represents the particle size distribution retention rate, and the overall activity retention rate. according to calculate; Calculated by summing the retention rates of CD9, CD63, and CD81 and dividing by 3; Using a fixed particle size range of 30-300 nm, the particles were divided into bins, and the proportion of particles in each particle size range to the total number of particles was calculated before and after freeze-drying. and and in accordance with calculate; The supplementary data, calculated using the same average of three independent batches as the performance table above, is as follows: Table 2. Results of retention rates of oocyst markers and overall activity retention rates in each embodiment and comparative example.

[0030] The above steps involve first calculating the baseline value for each test sample before lyophilization in the same batch, then averaging the results of samples within the same batch, and finally averaging the batch averages of three independent batches. The offline reference value for the comprehensive activity retention rate model is obtained using the same sample-by-sample weighted calculation method. As can be seen from the comparison of the test results of Example 1 and Comparative Example 1 in Tables 1 and 2, after omitting online spectral monitoring, activity retention rate prediction and feedback control, the unit particle functional efficacy retention rate decreased from 94.2% to 72.4%, and the particle number recovery rate, membrane integrity retention rate and resolution performance also decreased; fixed plate temperature, chamber pressure and drying time cannot reflect the changes in product temperature, residual moisture and sublimation rate at different drying stages; When the product temperature is close to the collapse temperature or the drying process is insufficient, the fixed program cannot reduce the heat load or extend the drying time in time, which can easily cause local structural collapse, high residual moisture, and damage to the exosome membrane structure. The decrease in membrane integrity and the increase in aggregation further reduce the uptake capacity of the receptor cells, resulting in a decrease in the retention rate of functional efficacy per unit particle. As can be seen from the comparison of the test results of Example 1 and Comparative Example 2 in Tables 1 and 2, after omitting mannitol structural support, the retention rate of functional efficacy per unit particle decreased from 94.2% to 78.1%, and the particle size change and reconstitution time increased; mannitol can form a rigid physical framework support in the freeze-dried cake, improving the drying channels and the mechanical strength of the freeze-dried cake; Omitting mannitol reduces the supporting capacity of the amorphous solids mainly formed by sugars, makes the pore structure more prone to shrinkage during drying, increases the resistance to water transfer, and raises the local residual moisture content. During reconstitution, the solid-liquid mass transfer interface area between the shrinkage region and the reconstitution medium is significantly reduced, leading to an increase in reconstitution time. At the same time, the increased local stress caused by freeze concentration and dehydration results in a decrease in membrane integrity and functional efficiency. As can be seen from the comparison of the test results of Example 1 and Comparative Example 3 in Tables 1 and 2, after omitting trehalose and making up the volume with histidine buffer, the retention rate of functional efficacy per unit particle decreased to 61.7%, the particle number recovery rate decreased to 76.8%, and the residual moisture increased to 3.05%. Trehalose can protect the exosome membrane structure and the hydration layer on the membrane surface during freeze concentration and dehydration, and form a solid protective phase with an amorphous glassy network structure after drying. Omitting trehalose weakens the water replacement effect around membrane lipids and proteins during exosome dehydration, subjecting the membrane structure to extreme dehydration stress exceeding the tolerance threshold; the continuity of the solid structure of the protective fluid decreases, local solute concentration and molecular migration increase, leading to exosome aggregation, decreased membrane integrity and reduced functional efficiency; the shrinkage of the solid structure also reduces water escape efficiency, resulting in increased residual water. As can be seen from the comparison of the test results of Example 1 and Comparative Example 4 in Tables 1 and 2, after controlling the product temperature in the first drying stage to be within 0.5°C below the collapse temperature, the unit particle functional efficiency retention rate decreased from 94.2% to 69.5%, and the particle size change and polydispersity index significantly exceeded the set process control limits. When the product temperature is too close to the collapse temperature during the primary drying stage, the amorphous structure of the freeze-dried cake is prone to viscous flow and local collapse during sublimation. The drying channels are compressed, and the escape of water vapor is hindered. Local water retention and solute concentration increase the contact opportunities between exosomes, promote vesicle aggregation or fusion, and generate mechanical and permeational stress on the lipid bilayer membrane. As a result, the particle size distribution becomes wider, the polydispersity index increases, and the membrane integrity and the retention rate of functional efficiency per unit particle decrease. As can be seen from the comparison of the test results of Example 1 and Comparative Example 5 in Tables 1 and 2, after omitting the prediction uncertainty and model applicability judgment, the unit particle functional power retention rate decreased from 94.2% to 76.8%, and the batch-to-batch functional power relative standard deviation increased from 3.8% to 12.5%. The spectral signal may be affected by condensation in the observation window, empty bottle background, changes in fiber optic position, instrument drift, or changes in the solid state of the preservative. Without model suitability assessment, spectra that deviate from the training set range may still be input into the activity prediction model, resulting in predicted values ​​that are inconsistent with the actual product state. When erroneous predicted values ​​are used to adjust plate temperature, cavity pressure, or heating rate, they may cause an increase in heat load or premature transition of the drying stage, resulting in different degrees of structural damage to different batches. Prediction uncertainty and model suitability assessment can identify spectral data that are not suitable for automatic control and reduce process misadjustments caused by abnormal signals. As can be seen from the comparison of the test results of Example 1 and Comparative Example 6 in Tables 1 and 2, after increasing the chamber pressure in the primary drying stage from 20 Pa to 80 Pa, the unit particle functional efficiency retention rate decreased from 94.2% to 74.3%, the residual moisture increased to 2.59%, and the resolution time increased to 82 s. The increase in chamber pressure will reduce the sublimation driving force and change the water vapor transfer conditions in the product layer, thus reducing the moisture removal rate in the primary drying stage. Moisture retention prolongs the time exosomes are in a frozen, concentrated environment with high solute concentrations, increasing the osmotic stress on the membrane structure and the risk of local phase separation. If there is still unremoved moisture exceeding the residual control threshold at the end of the first drying process, the temperature rise during the second drying process will increase the thermal exposure of exosome membrane proteins and lipid components, leading to decreased membrane integrity, increased particle size variation, and reduced functional efficacy. The test results from Examples 1 to 4 show that when the non-reducing sugars, structural support agents, buffer systems, surface stabilizers, freezing processes, and drying processes are all within the set ranges, the unit particle functional efficiency retention rate reaches over 87.9%, the particle number recovery rate reaches over 92.5%, and the membrane integrity retention rate reaches over 90.7%. Example 2 uses a sugar concentration and freezing temperature with a lower limit set in the process design space, resulting in increased residual moisture and reconstitution time; Example 3 uses higher sugar and mannitol concentrations, which can maintain a particle recovery rate that meets the target parameter thresholds, but the secondary drying temperature is at the upper limit set, and the functional efficacy retention rate is lower than that of Example 1. Example 4 uses a combination of trehalose and sucrose and extends the secondary drying and heat preservation process. The residual moisture content is within an acceptable range, but the reconstitution time and batch-to-batch fluctuations are higher than in Example 1. The above results indicate that the composition of the protective agent, freezing conditions, product temperature, chamber pressure, and secondary drying conditions need to be determined within a validated process design space. Changing a single parameter may affect the exosome membrane integrity, residual moisture content, and reconstitution performance. The above results were statistically analyzed using three independent batches. The functional efficacy retention rate and residual moisture results listed in the table were compared with the offline detection results of receptor cell uptake and the Karl Fischer method detection results, respectively. The mean absolute error between the online activity retention rate prediction and the offline comprehensive activity retention rate in Example 1 was 2.5 percentage points, and the root mean square error was 3.2 percentage points; the mean absolute error between the residual moisture prediction and the Karl Fischer method result was 0.14 percentage points, and the root mean square error was 0.18 percentage points. This indicates that the online spectral model can reflect the changing trends of activity retention rate and residual moisture in independent batches that were not involved in the modeling.

[0031] The above are merely specific embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any conventional modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the protection scope of the technical solution of the present invention.

Claims

1. An exocrine lyophilization process based on open-spectrum RD data processing, characterized in that, include: Prepare a lyophilization protection solution containing non-reducing sugars, structural support agents, and buffers, and adjust its pH. The freeze-drying protective solution is mixed with the exosome preparation and then filled into the container; the filled preparation is subjected to gradient cooling freezing, one vacuum drying and two heating drying in sequence; near-infrared spectroscopy, Raman spectroscopy or a combination thereof are collected in at least one drying stage, and the predicted value of activity retention rate, the predicted value of residual moisture, the prediction uncertainty and the spectral anomaly index are obtained after spectral RD data processing. The aforementioned open-spectrum RD data processing refers to processing the online acquired spectral data and fusing the processing results with the measured process parameters of product temperature, plate temperature, and cavity pressure obtained synchronously according to the sampling time.

2. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 1, characterized in that, The preparation of the lyophilization protective solution includes dissolving non-reducing sugars, structural support agents, and buffers in water for injection, adjusting the pH to 6.8-7.4; adding a surface stabilizer and allowing it to stand to remove bubbles; and mixing it with the exosome preparation after sterile filtration. The non-reducing sugars are selected from one or more of trehalose, sucrose, and raffinose, with a total concentration of 2%-12%. The structural support is selected from one or more of mannitol and glycine, with a concentration of 0.2%-5%; the buffer is selected from histidine, citrate or phosphate, with a concentration of 5-25 mmol / L; the surface stabilizer is selected from polysorbate 20, polysorbate 80 or poloxamer, with a concentration of 0.001%-0.05%.

3. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 1, characterized in that, The particle concentration of the exosome preparation is The total protein concentration is 0.05-5 mg / mL, the particle size range is 30-300 nm, and the osmotic pressure is 250-400 mOsm / kg.

4. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 1, characterized in that, The gradient cooling freezing includes cooling to -55℃ to -35℃ at a rate of 0.1-2℃ / min and holding at that temperature for 0.5-3 hours, followed by annealing at -30℃ to -15℃ for 0.5-2 hours. The chamber pressure for the first vacuum drying is 5-50 Pa, and the upper limit of the product temperature is the temperature value obtained by subtracting 2℃ to 5℃ from the preset collapse temperature threshold. The second heating drying includes heating to 10-35℃ at a rate of 0.05-0.5℃ / min and holding at that temperature for 2-10 hours. The second heating drying ends when the predicted residual moisture value reaches the preset range, the fluctuation rate of the predicted activity retention rate is less than the preset threshold within the continuous sampling period, and the prediction uncertainty is within the preset range.

5. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 1, characterized in that, The open spectrum RD data processing includes performing dark current correction, empty bottle background subtraction, freeze-drying protective liquid blank correction, abnormal spectrum removal, spectral preprocessing and feature extraction on the original spectrum; and fusing moisture characteristics, carbohydrate solid state characteristics, protein or lipid related characteristics with product temperature, plate temperature and cavity pressure.

6. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 5, characterized in that, The spectral preprocessing includes one or more of the following: standard normal variable transformation, multiplicative scattering correction, detrending, Savitsky-Gorye smoothing, first derivative, second derivative, fluorescence background subtraction, baseline correction, or peak position correction.

7. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 1, characterized in that, The predicted activity retention rate is obtained by a partial least squares regression, support vector regression, random forest, or neural network model based on the spectral data of at least two batches of exosomes and the offline detection results of unit particle functional efficacy, membrane integrity, particle number recovery rate, particle size distribution, key biomarkers, or combinations thereof.

8. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 1, characterized in that, A collapse temperature threshold is preset. When the product temperature reaches the collapse temperature threshold, the predicted value of the activity retention rate decreases continuously, the spectral anomaly index exceeds the preset limit, or the cavity pressure fluctuation exceeds the preset fluctuation threshold, the plate temperature is reduced, the heating rate is reduced, the cavity pressure is adjusted, or the heat preservation time of the current drying stage is extended. When the spectrum deviates from the preset applicable range of the model training set, the prediction deviation exceeds the preset verification range, the reference sample calibration fails, or the spectroscopic equipment malfunctions, the automatic control based on the predicted value of activity retention rate is suspended and switched to the backup freeze-drying program.

9. The exocrine lyophilization process based on open-spectrum RD data processing according to claim 4, characterized in that, The freeze-drying endpoint is also determined by combining the difference between the product temperature and the plate temperature, the difference between the cavity pressure and the reading of the capacitor diaphragm vacuum gauge, the pressure rise test results, and the moisture-related spectral characteristics.

Images