EA046325B1 - APPLICATION OF RAMAN SPECTROSCOPY FOR POST-PURIFICATION - Google Patents

Description

Technical field of the invention

The invention is generally directed to systems and methods for monitoring and controlling one or more critical quality attributes (CQAs) or parameters in downstream protein purification processes.

State of the art

Over the past decade, there has been a significant increase in the number of monoclonal antibodies (mAbs) approved for therapeutic use. This is due in part to improvements in large-scale manufacturing processes that make it easier to manufacture large quantities of mAbs. In addition, initiatives such as the Food and Drug Administration's (FDA) Process Analytical Technology (PAT) concept have led to innovative solutions for process design, process analysis, and process control to better understand processes and control product quality. Efficient extraction and purification of mAbs from cell culture media is an important part of the manufacturing process. The purification process must ensure that mAbs are obtained that are safe for use in humans. This includes monitoring of critical quality attributes (CQAs), which include protein properties and impurities such as host cell proteins, DNA, viruses, endotoxins, aggregates, concentrations, excipients and other agents that may affect safety, efficacy for the patient , or activity. Protein concentration is also often a CQA in purified material, and the corresponding protein concentration in process intermediates can be a critical parameter when assessing the performance of a typical process. These CQAs need to be tracked throughout production as well as throughout the program life cycle.

To ensure that the final mAb formulations do not contain impurities above certain levels, mAb products are tested at various downstream processing steps. Quality control in the manufacture of bioproducts such as monoclonal antibodies is typically accomplished by analyzing purification intermediates and compounded drug samples using offline methods for each batch. Samples are removed from process equipment, such as a UF/DF unit, and subjected to offline tests to evaluate product CQAs such as protein concentration (g/L), buffer excipients, and size variations. Real-time process monitoring and analysis during manufacturing is not available, resulting in longer processing times and a higher risk of batch defects due to non-compliance with CQA. Accordingly, there is a need for fast, on-line methods for real-time quality control of monoclonal antibodies.

It is therefore an object of this invention to provide systems and methods for real-time monitoring of critical quality indicators during a downstream purification process.

The essence of the invention

This invention provides in situ Raman spectroscopy methods and systems for characterizing or quantifying a protein purification intermediate during a manufacturing or manufacturing process. In one embodiment of the present invention, in situ Raman spectroscopy is used to characterize or quantify critical quality attributes of a protein preparation during downstream processing (ie, after collection of the protein purification intermediate from the cell culture fluid). For example, the described in situ Raman spectroscopy methods and systems can be used to characterize and quantify critical quality attributes of protein purification intermediates as protein purification intermediates are purified before formulation into a finished drug product for sale or administration. Critical quality indicators include, but are not limited to, protein concentration, excipients, high molecular weight (HMW) agents, antibody titer, and drug-to-antibody ratio.

In one embodiment, the present invention provides a method for producing a concentrated protein purification intermediate by determining the concentrations of the protein purification intermediate in real time using in situ Raman spectroscopy while concentrating/diafiltrating the protein purification intermediate and adjusting the parameters of the concentration step in real time to obtain predetermined target concentrations and levels of excipients required to formulate the drug substance. The protein purification intermediate may have a concentration of 5 mg/ml to 300 mg/ml, preferably 50 mg/ml to 300 mg/ml for subsequent formulation steps. In one embodiment, the protein purification intermediate is concentrated to the desired target concentration using ultrafiltration during primary or final concentration. Diafiltration is used during processing after primary concentration as a means of buffer replacement to achieve

- 1 046325 desired components of the final composition. The protein purification intermediate can be obtained from a bioreactor, fed-batch culture, or continuous culture. In another embodiment of the present invention, determination of the concentration of a protein purification intermediate can occur continuously or intermittently in real time. Quantitative determination of protein concentration can be carried out at intervals of approximately 5 seconds to 10 minutes, hourly or daily. The protein purification intermediate may be an antibody or an antigen binding fragment thereof, a fusion protein, or a recombinant protein. Spectral data may be collected in one or more wavenumber ranges selected from the group consisting of 977-1027 cm ^-1 , 1408-1485 cm ^-1 , 1621-1711 cm ^-1 , 2823-3046 cm ^-1 and combinations thereof .

In another embodiment, the present invention provides a method for producing a protein purification intermediate by independently performing Raman spectroscopy analysis on a plurality of protein purification intermediates to obtain a universal model capable of quantifying any of a plurality of protein purification intermediates. The concentration of the protein purification intermediate can be determined by in situ Raman spectroscopy with a universal model during the concentration of the protein purification intermediate from the beginning to the end of the concentration of the protein purification intermediate. In another embodiment, the present invention provides a method for producing a protein purification intermediate to obtain a protein-specific model capable of quantifying protein concentrations, which can be used to enable commercial protein production.

This model can be constructed using fractional least squares regression analysis for the raw spectral data and using the orthogonal method for the offline protein concentration data. Preprocessing techniques, such as standard deviation normal variable (SNV) and/or 1st derivative data smoothing with 21 cm ^-1 smoothing, can be performed on Raman spectroscopy data to reduce model variability and prediction error. Further refinement of the model can be performed to highlight spectral regions that correlate with CQA predictive data, such as protein concentration. In one embodiment of the present invention, said model has a margin of error of <5%, preferably a margin of error of <3%.

In yet another embodiment, the present invention provides a method for monitoring and controlling excipient levels in the collected culture liquid and/or protein purification intermediate during subsequent purification by determining excipient concentrations in real time using in situ Raman spectroscopy during purification of the culture liquid. cells or protein purification intermediate, and adjusting parameters of the purification step in real time to obtain or maintain predetermined amounts of excipients in the collected culture fluid and/or protein purification intermediate. Said excipient may be acetate, citrate, histidine, succinate, phosphate, hydroxymethylaminomethane (Tris), proline, arginine, sucrose or combinations thereof. Said excipient may be a surface excipient such as polysorbate 80, polysorbate 20 and poloxamer 188.

Brief description of graphic materials

Fig. 1 is a flow diagram showing a typical protein purification process.

Fig. 2 is a representative spectrograph showing the development of an initial model for in-line Raman spectroscopy. The X axis represents the Raman shift. The Y axis represents intensity. The legend on the right represents the protein concentration. Spectral regions used during initial model development include, from left to right, 977-1027 cm ^-1 (ring structure), 1408-1485 cm ^-1 (arginine), 1621-1711 cm ^-1 (secondary structure) and 2823-3046 cm ^{-1 1} (C-H stretch).

Fig. 3 is a bar graph showing the protein concentration (g/L) for mAb 1 during a standard ultrafiltration/diafiltration process. Empty bars represent concentrations determined using an offline UV-Vis based method with a SoloVPE system (C-technologies) with an error bar of ±5%, which is the target for on-line Raman prediction. Columns with wide shading represent the model, referred to herein as the universal model without mAb1, included in the Raman prediction, and bars with narrow shading represent the universal model with mAb 1. Cross-shaded bars correspond to predictions from the initial universal model with the range of 0-120 g/L . The shaded bars correspond to predictions from the initial universal model >120 g/L.

Fig. 4 is a histogram showing the absolute error of the Raman model for various mAbs from the development of the initial generic model. Filled bars represent the initial universal model 0-120 g/L (primary concentration and diafiltration), and empty bars represent the initial universal model >120 g/L. (Final end

- 2 046325 testing). The horizontal line represents the Raman model target with <5% error.

Fig. 5 is a histogram showing the absolute error of the Raman model for various mAbs. Columns with wide shading represent 0-120 g/L (initial concentration and diafiltration), and bars with narrow shading represent >120 g/L (final concentration). The horizontal line represents the Raman model target with <5% error. Two versions of the universal Raman model are demonstrated. The shaded bars represent the primary universal model, and the cross-shaded bars represent the updated universal model.

Fig. 6 is a schematic illustration of an ultrafiltration/diafiltration system including locations for an in-line Raman probe.

Fig. 7 is a histogram showing protein concentration for mAb10 using the generic model or mAb10-specific model for final concentrated pool (FCP) measurements. Protein concentrations are shown for real-time in-line Raman prediction (wide-shaded columns), updated models (narrow-shaded columns), and SoloVPE (empty column). The X-axis represents the experimental group and the Y-axis represents the protein concentration.

Fig. 8A,B are histograms showing the error of the Raman model for the laboratory DoE simulation with respect to protein concentration. In fig. Figure 8A demonstrates the error of the Raman model with respect to real-time prediction at various stages of UV/DF processing (primary concentration, diafiltration and final concentrated pool). In fig. Figure 8B demonstrates the Raman model error for the final DoE model at various stages of UV/DF processing (primary concentration, diafiltration, and final concentrated pool).

Fig. 9A is a bar graph showing the percentage error of the model prediction relative to the scaled-up model reproduction on the experimental mAb11 processing equipment. Data from the laboratory model were compared with data from a laboratory model incorporating pilot-scale data. The X-axis represents the experimental groups and the Y-axis represents the percentage model prediction error (%). Fig. 9B is a bar graph demonstrating the model's predictive capabilities regarding the processing of various monoclonal antibodies on experimental processing equipment. The X-axis represents the experimental groups and the Y-axis represents the percentage error of the model prediction (%).

Fig. 10A is a diagram of a typical automated batch UV/DF process with Raman feedback. Fig. 10B is a schematic diagram of a typical automated direct-flow TFF with Raman feedback.

Fig. 11 is a graph showing the concentration of mAb14 during UF/DF treatment. The X axis represents the productivity (L/m ² ), and the Y axis represents the mAb 14 concentration (g/L).

Fig. 12A is a bar graph showing the percentage of high molecular weight (HMW) agents in mAb2 at various processing steps (primary concentration, diafiltration, final concentration) determined by SE-UHPLC or Raman simulation. The X-axis represents the experimental group and the Y-axis represents the percentage (%) of mAb2 HMW.

Fig. 12B is a bar graph showing the percentage of high molecular weight (HMW) agents predicted using different scan times (10, 20, 30 s) for mAb 15. The X-axis represents the experimental group and the Y-axis represents the percentage HMW (%).

Fig. 13 is a scatter plot showing the actual titer (g/L) versus the titer predicted by Raman for monoclonal antibody samples from mAb 14. The X-axis represents the titer predicted by Raman and the Y-axis represents the actual titer (g/L) .

Fig. 14A is a scatter plot showing Raman prediction of histidine concentration for various monoclonal antibodies. The x-axis represents the histidine concentration predicted by Raman modeling, and the y-axis represents the actual histidine concentration determined by amino acid analysis. Fig. 14B is a dot plot showing the actual histidine concentration versus the histidine concentration predicted by Raman modeling for a monoclonal antibody sample. Fig. 14C is a scatter plot showing the actual arginine concentration versus the Raman modeling predicted arginine concentration for a monoclonal antibody sample.

Fig. 15A is a scatter plot showing the actual drug-to-antibody ratio (DAR) compared to the DAR predicted by Raman modeling for mAb3 mAb samples. The X-axis represents the DAR predicted

- 3 046325 bath using Raman modeling, and the Y axis represents the actual DAR determined using UV spectroscopy. Fig. 15B is a scatter plot showing the actual drug-antibody ratio (DAR) versus the DAR predicted by Raman modeling for monoclonal antibody samples from mAb 1. The X-axis represents the DAR predicted by Raman modeling and the Y-axis represents the actual DAR.

Detailed description of the invention

I. Definitions.

It should be understood that this description is not limited to the compositions and methods described herein, nor to the experimental conditions described, since these may vary. It should also be understood that the terminology used herein is used only to describe certain embodiments and is not intended to be limiting, as the scope of this description is limited only by the appended claims.

Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this description pertains. Although any compositions, methods and materials similar or equivalent to those described herein can be used in the practice or testing of this invention. All publications mentioned are incorporated herein by reference in their entirety.

The use of terms and references indicating the singular in the context of the description of a presently claimed invention (especially in the context of the claims) shall be construed to include both the singular and the plural, unless otherwise stated herein or clearly inconsistent with the context.

The listing of ranges of values herein is simply intended to serve as a concise means of individually referring to each individual value falling within that range unless otherwise noted herein, and each individual value is included in the description as if separately stated herein.

The use of the term about is intended to describe values above or below a specified value in the range of approx. ± 10%; in other embodiments of the present invention, the values may be in the range of values above or below the specified value in the range of approx. ± 5%; in other embodiments of the present invention, the values may be in the range of values above or below the specified value in the range of approx. ± 2%; in other embodiments of the present invention, the values may be in the range of values above or below the specified value in the range of approx. ± 1%. The above ranges are intended to provide context and no further limitations are intended. All methods described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly inconsistent with the context. The use of any and all examples or illustrative language (eg, such as) presented herein is intended to better describe the invention only and does not constitute a limitation on the scope of the invention unless otherwise stated. Nothing in the specification should be construed as indicating any element not claimed as essential to the practice of this invention.

As used herein, the term protein refers to a molecule containing two or more amino acid residues linked to each other by a peptide bond. The protein includes polypeptides and peptides, and may also include modifications such as glycosylation, lipid addition, sulfation, gamma-carboxylation of glutamic acid residues, alkylation, hydroxylation and ADP-ribosylation. Proteins may be of scientific or commercial interest, including protein-based drugs, wherein proteins include, but are not limited to, enzymes, ligands, receptors, antibodies, and chimeric or hybrid proteins. Proteins are produced by various types of recombinant cells using well-known cell culture techniques and are typically introduced into the cell by transfection of genetically engineered nucleotide vectors (such as a chimeric protein coding sequence or a codon-optimized sequence, an intronless sequence, etc. ), where the vectors can be in the form of an episome or integrated into the cell genome.

The term antibody refers to an immunoglobulin molecule consisting of four polypeptide chains, two heavy (H) chains and two light (L) chains, linked together by disulfide bonds. Each heavy chain contains a heavy chain variable region (HCVR or VH) and a heavy chain constant region. The heavy chain constant region contains three domains: CH1, CH2 and CH3. Each light chain has a light chain variable region and a light chain constant region. The light chain constant region consists of a single domain (CL). The VH and VL regions can be further subdivided into hypervariable regions called complementarity determining regions (CDRs), interspersed with more conserved regions called framework regions (FRs). Each VH and VL consists of three CDRs and four FRs, located from the amino terminus to the carboxy terminus.

- 4 046325 si-ends in the following order: FR1, CDR1, FR2, CDR2, FR3, CDR3, FR4. The term antibody includes reference to both glycosylated and non-glycosylated immunoglobulins of any isotype or subclass. The term antibody includes antibody molecules produced, expressed, generated or isolated by recombinant methods, such as antibodies isolated from a host cell transfected to express the antibody. The term antibody also includes a bispecific antibody that includes a heterotetrameric immunoglobulin that can bind to more than one dissimilar epitope. Bispecific antibodies are generally described in US Patent Application Publication No. 2010/0331527, which is incorporated herein by reference.

The term secondary structure refers to the local folded structures that form within the polypeptide due to interactions between the atoms of the main chain. The most common types of secondary structures are α-helix and β-sheet. Both structures are held in shape by hydrogen bonds that form between the carbonyl O of one amino acid and the amino H of the other.

In this context, the term excipient refers to a pharmacologically inactive substance that is used as a stabilizing agent for long-term storage of the formulated drug substance. Typically, additional excipients are added to the final concentrated pool to produce a formulated drug substance. At the same time, UV/DF treatment times control excipient levels to ensure that these levels do not affect the desired formulation performance. Excipients add bulk to a pharmaceutical formulation, promote absorption or solubility of the drug, and provide stability and prevent denaturation. Common pharmaceutical excipients include, but are not limited to, amino acids, excipients, binders, disintegrants, coatings, sorbents, buffering agents, chelating agents, lubricants, lubricants, preservatives, antioxidants, flavorings, sweeteners, colorants, solvents and cosolvents. , as well as viscosity imparting agents. In one embodiment of this invention, the excipient is polyethylene glycol, including, but not limited to, PEG-3550.

Polyethylene glycol or PEG is a polyether ethylene oxide polymer commonly used in food, medicine and cosmetics. It is a nonionic macromolecule that can be used as a molecule that reduces the solubility of a biomolecule. PEG is commercially available in various molecular weights from 300 g/mol to 10,000,000 g/mol. Examples of PEG types include, but are not limited to, PEG 20000, PEG 8000 and PEG 3350. PEG is available in a variety of geometries including linear, branched (3-10 chains attached to a central core), star (10-100 chains attached to a central core ) and comb (many chains attached to the polymer backbone).

Raman spectroscopy is a spectroscopic technique used to measure the wavelength and intensity of inelastically scattered light from molecules. This method is based on the principle that monochromatic incident radiation on materials will be reflected, absorbed or scattered in a specific manner, which depends on the specific molecule or protein that receives this radiation. Most energy is scattered at the same wavelength, which is called elastic or Rayleigh scattering. A small amount (<0.001%) is scattered at different wavelengths, which is called inelastic or Raman scattering. Raman scattering is associated with transitions of rotational, vibrational and electronic levels. Raman spectroscopy reveals the chemical and structural composition of samples.

As used herein, the term ultrafiltration refers to a membrane process that is widely used for protein concentration for downstream processing of protein therapeutics during the purification of recombinant proteins. Ultrafiltration is a size separation in which particles larger than the pores of the membrane are retained while smaller agents pass freely. During processing, the protein solution is pumped tangentially across the surface of a semi-permeable parallel flat-sheet membrane. The membrane is permeable to buffers and buffer salts, but is generally impermeable to monoclonal antibodies. The driving force for permeation is the application of transmembrane pressure (TMP ₎ induced by the flow restriction at the outlet of the membrane flow channel (TMP = (P _feed + P _retentate )/2 - P _permeate ).

In this context, the term primary concentration refers to the initial step in which transmembrane pressure moves water and salts across the permeable membrane, which reduces the volume of liquid and thus increases the protein concentration. The degree of concentration during primary concentration can be optimized to balance throughput, protein stability, processing time, and buffer consumption.

As used herein, the term diafiltration refers to a technique that uses a semi-permeable membrane to exchange a product of interest from one liquid medium to another. Buffer exchange and salt removal are usually performed using a diafiltration mode in which

- 5 046325 small impurities and buffer components are effectively washed out of the product by continuous addition of a buffer, which is designed to condition the protein to a stable pH and concentrate the excipient, resulting in a highly concentrated product. This can be done in a continuous or intermittent manner depending on the processing technology.

Diafiltration is often combined with ultrafiltration to achieve the desired volume reduction while simultaneously removing impurities and salts. UV/DF is the final typical processing step in the downstream purification process that conditions the mAb to achieve pH, excipient content, and protein concentration to facilitate long-term storage, as well as the addition of stabilizing excipients to produce a formulated drug substance (FDS).

In this context, the term final concentration refers to the final step in which transmembrane pressure forces water and salts across the permeable membrane, which reduces the volume of liquid and thus increases the protein concentration to the desired target for storage and/or formulation. The pool obtained from the final concentration step is the final concentrated pool (FCP). This final concentration step can be performed in continuous or intermittent processing mode.

The terms bioproduct and protein purification intermediate can be used interchangeably and refer to any antibody, antibody fragment, modified antibody, protein, glycoprotein or fusion protein, as well as the final drug substances purified after a bioreactor.

The terms control and monitoring refer to adjusting the amount or concentration level of a critical quality indicator in the collected culture fluid to a predetermined set point.

In this context, the term pretreatment refers to the first step during which antibodies or therapeutic proteins are produced in bioreactors, typically by bacterial or mammalian cell lines. Pretreatment includes media preparation, cell culture, and cell separation and collection. When the cells reach the desired density, they are collected and moved to the next part of the bioprocess. The term downstream refers to the isolation and purification that occurs after the antibody or therapeutic protein is collected from the bioreactor. Typically this means extracting the product from an aqueous solution in several different ways. The harvested product is processed to ensure it meets purity and critical quality requirements during downstream processing.

As used herein, the term protein purification intermediate refers to the protein that has been collected from the bioreactor and refers to any intermediate product obtained during downstream processing.

The term concentrated protein purification intermediate refers to a protein purification intermediate with a concentration greater than 5 mg/mL. More preferably, the concentration is from 50 to 300 mg/ml.

The terms monitoring and monitoring refer to regularly checking the quantity or concentration level of a critical quality indicator in a cell culture or in the collected culture fluid.

The term harvested culture fluid refers to the fluid that is removed from a bioreactor containing cells that have been engineered to secrete proteins of interest. The collected culture fluid optimally contains the secreted protein of interest, such as a monoclonal antibody.

As used herein, the term critical quality attribute (CQA) refers to a physical, chemical, biological or microbiological property or characteristic that must be within an appropriate range, range or distribution in order to ensure the desired quality of a biological therapeutic drug. These characteristics may affect safety, efficacy and/or potency. Critical quality indicators include, but are not limited to, protein concentration, high molecular weight agents, buffering excipients, and pH.

As used herein, the term compounded drug substance refers to the active ingredient that is intended to provide pharmacological activity, but does not include intermediates used in the synthesis of such ingredient.

In this context, the term universal model refers to the mathematical correlation of the spectral properties of different recombinant proteins used to predict a critical quality indicator.

In this context, the term mAb specific model refers to the mathematical correlation of the spectral properties of one specific protein used to predict a critical quality indicator.

In this context, the term titer refers to the number of antibody or protein molecules in solution.

- 6 046325

II. Systems and methods for characterizing products of subsequent protein purification.

The present invention provides systems and methods for monitoring and controlling protein concentration during protein production. Concentrated protein solutions are difficult to measure accurately due to the high correlated viscosity of the solutions (>10 cP). Accurate quantification requires special off-line equipment and typically dilution of the solution. With high concentration UV/DF, final levels of excipients are a function of protein concentration due to the Gibbs-Donnan effect. Real-time monitoring and analysis during the manufacturing process is not available, resulting in longer processing times and a higher risk of batch failure due to non-compliance with CQA. The systems and methods described herein can be used to monitor protein concentration and other critical quality indicators on-line .

In one embodiment of the present invention, the Raman spectroscopy system is an in-line or in situ Raman spectroscopy system used during the production of a final concentrated pool, which may be highly concentrated (>150 g/L). Typically, a Raman spectroscopy system is used after the production of a protein purification intermediate, for example, during processing of a protein purification intermediate after collection from a bioreactor or fed culture system and subsequent purification. In fig. Figure 1 demonstrates a typical protein purification process. Typically, the protein purification intermediate is collected from cell culture (100) and processed through various purification steps such as affinity capture (110), virus inactivation (120), final chromatographic purification (130 and 140), virus retention filtration (150) and ultrafiltration/diafiltration (160) to produce a final concentrated pool, which is then formed into a drug substance. In one embodiment of the present invention, the protein concentration in the collected culture fluid is monitored using in situ Raman spectroscopy.

A. Raman spectroscopy.

In one embodiment of the present invention, the protein concentration in the collected culture fluid is monitored and controlled using Raman spectroscopy. Raman spectroscopy is a form of vibrational spectroscopy that provides molecular vibrational information that can be used to identify samples and quantify critical quality attributes. In situ Raman analysis is a method of analyzing a sample at its original location without the need to remove part of the sample for analysis in a Raman spectrometer. In situ Raman analysis is advantageous because Raman spectroscopy analyzers are non-invasive and non-destructive, reducing the risk of contamination and loss of protein quality. On-line Raman analysis can be implemented to provide continuous processing while monitoring protein concentrations in the collected culture fluid, protein purification intermediates, and/or the final concentrated pool.

In situ Raman analysis can provide real-time estimates of protein concentrations in protein purification intermediates. For example, raw spectral data obtained from in situ Raman spectroscopy can be used to obtain and monitor the ongoing protein concentration of protein purification intermediates. In this aspect, to ensure that the original spectral data is continually updated, said Raman spectroscopy spectral data must be collected approximately every 5 seconds to 10 hours. In another embodiment of the present invention, spectral data must be collected approximately every 15 minutes to 1 hour. In yet another In an embodiment of the present invention, spectral data must be collected approximately every 20-30 minutes.

Monitoring the protein concentration of the protein purification intermediate can be analyzed using any commercially available Raman spectroscopy analyzer that allows in situ Raman analysis. The in situ Raman analyzer must be capable of obtaining raw spectral data from the protein purification intermediate. For example, a Raman analyzer must be equipped with a probe that can be installed in the fluid line. Suitable Raman analyzers include, but are not limited to, RamanRXN2 and RamanRXN4 analyzers (Kaiser Optical Systems, Inc., Ann Arbor, MI).

Raw spectral data obtained from in situ Raman spectroscopy can be compared with offline protein concentration measurements to correlate peaks in the spectral data with protein concentration. Offline protein concentration measurements can be used to determine which spectral regions exhibit a protein signal. Offline measurement data can be collected using any suitable analytical method. In the case of protein concentration, for example, offline measurements can be obtained using SoloVPE (Ctechnologies). In addition, any type of multiparameter software package, such as SIMCA 13 (MKS Data Analytic Solutions, Umeå, Sweden), can be used to correlate peaks in raw spectral data with offline protein concentration measurements. However, in some embodiments of the present invention, it may be necessary to pre-process the original spectral data using spectral filters to remove any changing original

- 7 046325 states. For example, the raw spectral data may be preprocessed using any type of data smoothing or normalization technique. Normalization may be required to correct for any changes in the probe, optics, laser power, and Raman analyzer exposure time. In one embodiment of the present invention, the raw spectral data can be processed using a data smoothing method, such as 1st derivative with 21 cm ^-1 smoothing, and a normalization method, such as standard deviation of a normally distributed random variable (SNV). These preprocessing techniques can be combined for specific spectral regions to improve model prediction.

Chemometric modeling can also be performed on the obtained spectral data. In this aspect, one or more multivariate methods can be applied to the spectral data, including, but not limited to, fractional least squares (PLS), principal component analysis (PCA), orthogonal fractional least squares (OPLS), multivariate regression, canonical correlation, factorial analysis, cluster analysis, graphical methods, etc. In one embodiment of the present invention, the acquired spectral data is used to create a PLS regression model. A PLS regression model can be created by projecting predicted variables and observed variables into a new space. In this aspect, a PLS regression model can be created using measurement values obtained from Raman analysis and offline measurement values. The PLS regression model provides predicted process values, such as the predicted value of protein concentration. In one embodiment of the present invention, the model provides predicted protein concentration values with an error of <5% compared to offline protein concentration values. In a preferred embodiment of the present invention, the model provides predicted protein concentration values with an error of <3% compared to offline protein concentration values.

After chemometric modeling, a signal processing technique can be applied to the predicted protein concentration values. In one embodiment of the present invention, this signal processing technique will reduce model variability and prediction error. In this aspect, one or more pretreatment methods can be applied to the predicted protein concentration values. Any pre-treatment methods known to those skilled in the art may also be used. For example, the noise reduction technique may include data smoothing and/or signal rejection. Smoothing is achieved through a series of smoothing algorithms and filters, while noise reduction uses signal characteristics to identify data that should not be included in the spectral data being analyzed. In one embodiment of the present invention, the predicted protein concentration values are reduced by a noise suppression filter. The noise suppression filter provides the final predicted protein concentration values. In this aspect, the noise reduction method combines the original measurements with a model-based estimate of what the measurement should produce according to the model. In one embodiment of the present invention, the noise reduction method combines the current predicted protein concentration with its uncertainties. Uncertainties can be determined by the repeatability of predicted protein concentration values and current protein concentration values. As soon as the next predicted protein concentration value is observed, the estimate of the predicted protein concentration value is updated using a weighted average, which gives more weight to estimates with higher confidence. Using an iterative approach, the final protein concentration values can be updated based on the previous measurement and the current measurement. In this aspect, the algorithm must be recursive and be able to operate in real time to use the current predicted protein concentration value, the previous value and experimentally determined constants. The noise reduction technique improves the reliability of Raman measurements and PLS predictions by reducing the noise that the automatic recall controller will act upon.

B. Methods of application.

The described methods can be used to monitor and control the protein concentration in the collected culture fluid and/or protein purification intermediate fluids during subsequent protein purification processes. Common downstream purification processes include, but are not limited to, centrifugation, direct depth filtration, protein A affinity purification, virus inactivation steps, ion exchange chromatography, hydrophobic interaction chromatography, size exclusion chromatography, ultrafiltration/diafiltration, virus retention filtration, and combinations thereof. These typical operations are carried out in a specific sequential combination to isolate the protein of interest and provide control of impurities and/or critical quality parameters before obtaining the formulated drug substance. In one embodiment of the present invention, the described methods include a Raman probe installed in line

- 8 046325 liquid circuit. In another embodiment of the present invention, the described methods can be used to make a concentrated protein purification intermediate or a final concentrated pool.

1. Antibody titer and protein concentration.

Both antibody titer and protein concentration are important factors in the purification of bioproducts. The antibody titer measured after the initial collection of cell culture fluid is important to determine column loading and ensure a reliable purification process to remove impurities. Monitoring protein concentrations throughout all purification steps is important both to ensure proper concentration of the final product and to ensure proper performance of typical purification operations. Inadequate protein concentrations can result in ineffective drugs or the production of ineffective drug substances.

In one embodiment of the present invention, the collected culture liquid is subjected to Raman spectral analysis immediately after collection, but before any further purification begins. Raman spectral data can be used after collection to quantify the antibody titer in the collected culture fluid. Protein concentration can be measured using the described methods during multiple steps during the protein purification process, for example, during affinity capture, during final chromatographic purification, during virus retention filtration, or during ultrafiltration/diafiltration. In-line Raman probes can detect Raman scattering in the collected culture fluid and/or protein purification intermediate in the liquid loop without removing the sample from the system, thereby providing analytical performance that is typically determined off-line.

In one embodiment of the present invention, if the protein concentration falls outside the predetermined concentration during ultrafiltration/diafiltration, the system is notified and the protein purification intermediate is modified accordingly. For example, if the protein concentration of the protein purification intermediate is below the predetermined protein concentration, the protein purification intermediate can be further concentrated by ultrafiltration/diafiltration.

In one embodiment of the present invention, the concentration step is carried out using Protein A affinity chromatography.

2. Drug-antibody ratio.

In another embodiment of the present invention, the described methods can be used to monitor and control the drug-antibody ratio (DAR). DAR is a quality indicator that is monitored during the development of antibody-drug conjugates (ADCs), antibody-radionuclide conjugates (ARCs), and general protein conjugates (potent steroids, noncytotoxic loads, etc.) to ensure consistent product quality and facilitate subsequent product labeling regarding load. DAR represents the average amount of drugs or other therapeutic molecules conjugated to antibodies and is an important quality indicator in the manufacture of therapeutic conjugates. The DAR value influences the efficacy of a conjugate drug because a low drug load reduces the activity of that drug, while a high load can negatively impact pharmacokinetics and safety.

In one embodiment of the present invention, antibody-radionuclide conjugates are subjected to Raman spectral analysis immediately after conjugation, but before any further purification. Raman spectral data can be used after conjugation to determine DAR.

In one embodiment of the present invention, if the DAR goes beyond a predetermined concentration during processing, the system is notified and the ADC intermediate is modified accordingly. For example, if the DAR in an ADC intermediate is below a predetermined DAR, components of the conjugation reaction may be modified, such as reagent concentrations may be optimized, linker type may be changed, and temperature or other manufacturing process variables may be optimized.

3. Buffer auxiliaries.

The described methods can be used to monitor and control the levels of buffer excipients in the collected culture fluid and/or protein purification intermediate during subsequent purification. Buffering excipients that are commonly used in the manufacture of monoclonal antibodies include, but are not limited to, acetate, citrate, histidine, succinate, phosphate and hydroxymethylaminomethane (Tris), proline and arginine. Surfactant excipients include, but are not limited to, polysorbate 80 (Tween 80), polysorbate 20 (Tween 20), and poloxamer 188. Polyol/disaccharide/polysaccharide excipients include, but are not limited to, mannitol, sorbitol, sucrose, and dextran 40. Antioxidant excipients include, but are not limited to, ascorbic acid, methionine

- 9 046325 and ethylenediaminetetraacetic acid (EDTA). Two commonly used amino acid excipients are histidine and arginine. In a preferred embodiment of the present invention, the excipient that is to be monitored and controlled is histidine and arginine.

The excipient concentrations of the final concentrated pool differ from the composition of the diafiltration buffer due to a combination of exclusion volume and Donnan effects. The Donnan effect is a phenomenon that occurs due to the retention of pure, positively charged protein by the membrane during UV/DF, coupled with the requirement for charge neutrality in both the retentate and permeate. To balance the positively charged protein, the negatively charged components of the buffer are enriched in the retentate relative to the diafiltration buffer, and the positively charged components of the buffer are removed. This effect can cause the FCP pH and buffer excipient concentrations to be significantly different from the diafiltration buffer composition (Stoner, et al., J Pharm Sci, 93: 2332-2342 (2004)).

Exclusion volume describes the characteristics of highly concentrated samples in which the protein occupies a significant portion of the solution volume. The buffer is excluded from the volume occupied by the protein, causing the solute concentrations in the buffer to decrease as the protein concentration, expressed as moles (or mass) of solutes per volume of solution, increases. The buffer is excluded from the volume occupied by the protein, causing the solute concentrations in the buffer to decrease as the protein concentration, expressed as moles (or mass) of solutes per volume of solution, increases.

Based on both of these principles and the fact that buffer excipient levels are critical quality indicators, in-line Raman probes minimize off-line analytical characterization and provide further process insight to ensure that excipient levels are sufficient for formulation.

4. High molecular weight impurities.

In the manufacture of monoclonal antibodies, low levels of product-related impurities are often detected even after extensive purification steps. High molecular weight (HMW) agents (eg, antibody dimer species) are product-associated impurities that contribute to the size heterogeneity of mAb products. The formation of HMW agents in a therapeutic mAb drug through protein aggregation can potentially compromise both the efficacy and safety of the drug. HMW agents are considered CQAs that are subject to regular monitoring during drug development and as part of testing for the release of purified protein drug products during manufacture.

In one embodiment of the present invention, the described methods can be used to identify protein drug products that contain HMW agents. HMW agents can be detected using Raman spectroscopy at various stages during the purification process, including, but not limited to, during affinity capture, during virus inactivation, during final chromatographic purification, during virus retention filtration, during ultrafiltration/diafiltration or combinations thereof.

In one embodiment of the present invention, HMW agents are detected in the collected culture fluid using the described methods, and said fluid is further processed to remove the HMW agents. Methods for removing HMW agents from culture fluid involve additional purification steps including, but not limited to, cation exchange chromatography and anion exchange chromatography.

C. UV/DF systems.

In fig. Figure 6 demonstrates the ultrafiltration/diafiltration processing system, including various localizations for in-line Raman probes (circled numbers 1-4). The protein purification intermediate is pumped into the retentate vessel 205 using a diafiltration pump 200, which may be a peristaltic, rotary, transfer or diaphragm pump. Liquid from retentate reservoir 205 flows to feed pump 210, which may be a rotary, peristaltic or diaphragm pump, past feed pressure valve 215 and into tangential flow filtration module (TFFM) 220. In TFFM 220, the protein purification intermediate is ultrafiltrated through a membrane. The bioproduct of interest is retained in a liquid (retentate) while water and low molecular weight solutes, including buffer auxiliaries, pass through the membrane in the permeate (filtrate), which exits the system through a permeate pressure control valve 225 into waste reservoir 230. The retentate exits the TFFM 220 and passes through the retentate pressure control valve 235, the transmembrane pressure control valve 240, and the retentate return channel 245, through which it flows back to the retentate vessel 205. The described process can be repeated as needed. , to concentrate the bioproduct, remove impurities and ensure CQAs are within acceptable limits. During diafiltration, the same flow path as described above is applied, where

- 10 046325 rum permeable solutes are replaced as new buffer is washed into the product stream. When new buffer is added at the same rate as permeate is removed from the system, the sum of the retentate reservoir volume and the brake retention volume determines the system volume. Volume per turnover (TOV) is defined as the amount of diafiltration buffer added during the ultrafiltration/ultrafiltration process equal to the system volume. Typically, replacing 8 times the system volume (8 TOV) provides >99.9% buffer replacement (Schwarts, L., Scientific and Technical Report, PN 33289)

Additionally, during the UV/DF process, it is necessary to agitate the protein solution in the retentate vessel 205. Due to the density differences between the diafiltration buffer, the return retentate, and the bulk retentate during diafiltration, it is necessary that there is sufficient agitation in the vessel to ensure adequate exchange buffer, but moderate enough to avoid shearing as this has been observed to lead to protein aggregation and generation of invisible particles (SVPs) in certain products. Additionally, it is important to ensure adequate mixing of the return retentate during the concentration steps to prevent polarization of the protein concentration in the retentate reservoir, resulting in higher protein concentrations being delivered to the UV/DF membranes.

In one embodiment of the present invention, the Raman probe is placed in position 1, after the diafiltration pump 200 in the retentate reservoir 205. In an alternative embodiment, the Raman probe can be placed in position 2 after the retentate reservoir 205 in line before the feed pump 210. In another embodiment, of this invention, the Raman probe is placed in line at position 3 between the pressure pump 215 and the tangential flow filtration module 220. In yet another embodiment of the present invention, the Raman probe is placed on the return retentate line 245. The location of the Raman probe is critical to ensure accurate linear measurements in complex system that has technical and technological limitations.

D. Cell culture.

The collected culture fluid can be recovered from a bioreactor containing cells engineered to produce monoclonal antibodies. The term cell includes any cell that is suitable for expressing a recombinant nucleic acid sequence. Cells include prokaryotic and eukaryotic cells, such as bacterial cells, mammalian cells, human cells, non-human animal cells, avian cells, insect cells, yeast cells or fused cells such as, for example, hybridomas or quadromas. In some embodiments of the present invention, said cell is a human, monkey, primate, hamster, rat or mouse cell. In other embodiments of the present invention, said cell is selected from the following cells: Chinese hamster ovary (CHO) cell (e.g., CHO K1, DXB-11 CHO, Veggie-CHO), COS (e.g., COS-7), retinal cell, Vero, CV1, kidney (eg HEK293, 293 EBNA, MSR 293, MDCK, HaK, BHK21), HeLa, HepG2, WI38, MRC 5, Colo25, HB 8065, HL60, lymphocyte such as Jurkat (T-lymphocyte) or Daudi ( B lymphocyte), A431 (epidermal), U937, 3T3, L cell, C127 cell, SP2/0, NS-0, MMT cell, stem cell, tumor cell and a cell line derived from the above cells. In some embodiments of the present invention, said cell contains one or more viral genes, for example, a retinal cell that expresses a viral gene (eg, a PER.C6® cell). In some embodiments of the present invention, said cell is a CHO cell. In other embodiments of the present invention, said cell is a CHO K1 cell.

In the protein manufacturing process, fed-batch cell culture or fed-batch culture refers to a fed-batch culture where cells and culture medium are first introduced into a culture vessel, and additional culture nutrients are introduced into the culture during culture in slow, discrete increments, with periodic collection cells and/or products before cessation of culture or without such collection. Fed-batch culture includes a semi-continuous fed-batch culture in which the entire culture (which may include cells and media) is periodically removed and replaced with fresh media. Fed-batch culture differs from batch culture in that in batch culture, all cell culture components (including animal cells and all culture nutrients) are supplied to the culture vessel at the beginning of the culture process. Fed-batch culture may differ from perfusion culture because the supernatant is not removed from the culture vessel during the standard fed-batch process, whereas in perfusion culture cells are maintained in culture, for example by filtration, and culture medium is continuously or periodically introduced and removed from the vessel for cultivation. However, during fed-batch cell culture, samples are expected to be removed for testing purposes. The feeding process continues until it is determined that maximum working volume and/or maximum protein production has been reached, and the protein is subsequently harvested.

The phrase continuous cell culture refers to a technique used to continuously grow cells, usually at a specific growth phase. For example, if a constant supply is required

- 11 046325 cell culture or the production of a particular protein of interest is required, the cell culture may need to be maintained at a specific growth phase. Thus, to maintain cells in this particular phase, conditions must be constantly monitored and adjusted accordingly.

The terms cell culture medium and culture medium refer to a nutrient solution used for growing mammalian cells that typically provides essential nutrients to enhance cell growth, such as an essential carbohydrate energy source, essential (e.g., phenylalanine, valine, threonine, tryptophan, methionine, leucine, isoleucine, lysine and histidine) and non-essential (e.g. alanine, asparagine, aspartic acid, cysteine, glutamic acid, glutamine, glycine, proline, serine and tyrosine) amino acids, trace elements, energy sources, lipids, vitamins, etc. d. Cell culture media may contain extracts, such as whey or peptones (hydrolysates), that provide the starting ingredients to support cell growth. The medium may contain extracts of yeast or soy origin instead of extracts of animal origin. Chemically defined medium refers to a cell culture medium in which all chemical components are known (ie, have a known chemical structure). The chemically defined medium is completely free of animal derived ingredients such as whey or animal derived peptones. In one embodiment of the present invention, said medium is a medium of a specific chemical composition.

The solution may also contain components that increase growth and/or survival beyond the minimum, including hormones and growth factors. The solution can be formulated with a pH and salt concentration that is optimal for the survival and proliferation of the particular cultured cell.

E. Proteins of interest.

Using the methods described, any protein of interest suitable for expression in prokaryotic or eukaryotic cells can be monitored. For example, a protein of interest includes, but is not limited to, an antibody or an antigen-binding fragment thereof, a chimeric antibody or an antigen-binding fragment thereof, a ScFv or a fragment thereof, an Fc fusion protein or a fragment thereof, a growth factor or a fragment thereof, a cytokine or a fragment thereof, or an extracellular domain cell surface receptor or fragment thereof. Proteins of interest may be simple polypeptides consisting of a single subunit or complex polysubunit proteins containing two or more subunits. The protein of interest may be a biopharmaceutical product, a dietary supplement or preservative, or any protein product subject to purification and quality standards.

In some embodiments of the present invention, the antibody is selected from the group consisting of an anti-programmed cell death protein 1 antibody (e.g., an anti-RP1 antibody as described in US Pat. No. 9,987,500), an anti-programmed cell death protein 1 ligand antibody (e.g., an anti-PD-LT as described in US Pat. No. 9,938,345), anti-EP4 antibodies, anti-angiopoietin-2 antibody (e.g., anti-ANG2 antibody as described in US Pat. No. 9,402,898), anti-angiopoietin-like protein 3 antibodies ( e.g., anti-AngPt13 antibodies as described in US Pat. No. 9,018,356), anti-platelet growth factor receptor antibodies (e.g., anti-PDGFR antibody as described in US Pat. No. 9,265,827), anti-Erb3 antibodies, anti-prolactin receptor antibodies (e.g. anti-PRLR antibody as described in US Pat. No. 9302015), anti-complement 5 antibodies (e.g. anti-C5 antibody as described in US Pat. No. 9795121), anti-TNF antibodies, anti-epidermal growth factor receptor antibodies (e.g. anti-EGFR, as described in US Pat. No. 9,132,192, or anti-EGFRvIn antibodies, as described in US Pat. No. 9,475,875), anti-subtilisin-kexin type 9 proprotein convertase antibodies (e.g., anti-PCSK9 antibody, as described in US Pat. No. 8,062,640 or US Pat. No. 9,540,449), anti-growth differentiation factor 8 antibodies (e.g., anti-GDF8 antibody, also known as anti-myostatin antibody, as described in US Pat. Nos. 8,871,209 or 9,260,515), anti-glucagon receptor antibodies (e.g. anti-GCGR antibody as described in US Pat. Nos. 9,587,029 or 9,657,099), anti-VEGF antibodies, anti-IL antibodies, anti-interleukin 4 receptor antibodies (e.g., anti-IL4I antibody as described in US Patent Application Publication No. 2014/0271681A1 or in US patents 8735095 or 8945559), anti-interleukin 6 receptor antibodies (for example, anTH-IL6R antibody. as described in US patent Nos. 7582298, 8043617 or 9173880), anti-III antibodies, anti-[B2 antibodies. anti-III3 antibodies, anti-III4 antibodies, anti-III5 antibodies, anti-III6 antibodies, anti-IL? antibodies, anti-interleukin 33 antibodies (for example, anti-III33 antibody, as described in US patent No. 9453072 or 9637535), anti-respiratory syncytial virus antibodies (e.g., anti-RSV antibody as described in U.S. Patent Application No. 9,447,173), anti-cluster of differentiation 3 antibodies (e.g., anti-CD3 antibody as described in U.S. Pat. Nos. 9,447,173 and 9,447,173 and application US Patent No. 62/222605) antibodies against cluster of differentiation 20 (for example, anti-CD20 antibody, as described in US Patent Nos. 9657102 and US 20150266966 A1, as well as in US Patent No.

- 12 046325

7879984), aHTu-CD19 antibody, aHTu-CD28 antibody, anti-cluster of differentiation-48 antibody (eg, anti-CD48 antibody as described in US Pat. No. 9,228,014), anti-Fel d1 antibody (eg, as described in US Pat. No. 9079948), anti-Middle East respiratory syndrome virus antibodies (e.g., anti-MERS antibody as described in US Pat. No. 9,718,872), anti-Ebola virus antibodies (e.g., as described in US Pat. No. 9,771,414), anti-Zika virus antibodies, anti-gene antibodies lymphocyte activation 3 antibody (eg, anti-LAG3 antibody or anti-CD223 antibody), anti-nerve growth factor antibody (eg, anti-NGF antibody, as described in US Patent Application No. US 2016/0017029 and US Patent Nos. 8309088 and 9353176) and antibodies against activin A.

In some embodiments of the present invention, the bispecific antibody is selected from the group consisting of anti-CD3 x anti-CD20 bispecific antibody (as described in US Pat. No. 9,657,102 and US Patent Application No. US 20150266966 A1), anti-CD3 x anti-CD2 bispecific antibody mucin 16 (eg, anti-CD3 x ημή-Μ^^ bispecific antibody) and anti-CD3 x bispecific antibody against specific membrane antigen (eg, α-anti-CD3 x and anti-PSMA bispecific antibody). In some embodiments of the present invention, the protein of interest is selected from the group consisting of abciximab, adalimumab, adalimumab-atto, adotrastuzumab, alemtuzumab, alirocumab, atezolizumab, avelumab, basiliximab, belimumab, benralizumab, bevacizumab, bezlotoxumab, blinatumomab, brentuximab vedotin, brodalumab, canakinumab, capromab pendetide, certolizumab pegola, cemiplimab, cetuximab, denosumab, dinutuximab, dupilumab, durvalumab, eculizumab, elotuzumab, emicizumab-kxwh, emtansinealirocumab, evinacumab, evolocumab, fasinumab, golimumab, guselkumab, Ibrit umomab tiuxetan, idarucizumab, infliximab, infliximab-abda , infliximab-dyyb, ipilimumab, ixekizumab, mepolizumab, necitumumab, nesvacumab, nivolumab, obiltoxaximab, obinutuzumab, ocrelizumab, ofatumumab, olaratumab, omalizumab, panitumumab, pembrolizumab, pertuzumab, ramucirumab, ranibizumab, cancer sibakumab, reslizumab, rinukumab, rituximab, sarilumab, secukinumab , siltuximab, tocilizumab, trastuzumab, alarmrumab, ustekinumab and vedolizumab.

Examples

Example 1. Universal in-line protein concentration model for UV/DF applications.

Materials and methods.

Data acquisition for the model included spectral analyzer data from Raman Rxn2 and Rxn 4 (Kaiser Optical Systems, Inc., Ann Arbor, MI) using the MR-Probe-785 and RamanRxn Probehead-758 (Kaiser Optical Systems, Inc., Ann Arbor, Michigan). Additionally, several different optical elements were used during the development process depending on availability. The operating parameters of the Raman analyzers were set to a scan time of 10 s for 6 acquisitions repeated 5 times. SIMCA 13 (MKS Data Analytic Solutions, Umeå, Sweden) was used to correlate peaks in the spectral data with offline protein concentration measurements. On-line measurements were performed at various stages of a typical UV/DF process, including primary concentration, diafiltration and final concentration. Autonomic protein concentrations were determined using SoloVPE (C Technologies, Inc.). SoloVPE measurements were performed in triplicate.

In fig. Figure 2 shows the spectral regions that were used to create chemometric models. The regions included the 1-977-1027 cm ^-1 region (ring structure), the 2-14081485 cm ^-1 region (arginine), the 3-1621-1711 cm ^-1 region (secondary structure) and the 4-2823-3046 cm ^-1 region (C-H stretch). The following spectral filtering was performed on the original spectral data: 1st derivative with data smoothing of 21 cm ^-1 to remove different original states.

Results.

To determine the technical applicability of a universal in-line protein concentration model for ultrafiltration/diafiltration (UF/DF), mAb1 was analyzed using Raman spectroscopy. Protein concentration was measured before diafiltration (primary concentration), during diafiltration (diafiltration), and after diafiltration (final concentration). The calculated concentrations from the model were compared with the protein concentration determined using SoloVPE (Fig. 3). The model error for 0-120 g/L (initial concentration and diafiltration) was 3.1%, and the model error for >120 g/L (final concentration) was 1.8% when training set data was included (mAb1 spectra included to the PLS model).

Processing errors can be detected using Raman spectroscopy. In fig. 3 demonstrated that air entrainment into the system during final recirculation through the UV/DF system was detected using Raman spectral data. Boxed bars demonstrate that the SoloVPE predicted mAbl concentration was >200 g/L, whereas the Raman prediction was ~65 g/L.

In fig. Figure 4 demonstrates the absolute error of the Raman model for ten representative mAbs. These data confirmed the successful development of models for the demonstrated mAbs, which included different mAb isotypes (IgG1 and IgG4) as well as bispecific molecules. 14 of 17

- 13,046,325 model forecasts had an error <5%. However, specific models (0–120 g/L and >120 g/L) were created for each probe used during development because the variability probe increased the errors predicted by the PLS models.

To optimize the model, different probes and lasers were tested in line with multiple mAbs. Model refinement was performed in cases where only one spectral region was in focus of the updated universal model: 2823-3046 cm ^-1 (C-H stretch) with spectral filtering of the standard deviation of a random variable with a normal distribution (SNV) to correct for changes in laser power and probe variability as an adjustment to the initial state. A comparison of the two developed model components is presented in Table. 1. Fractional least squares (PLS) regression models were created with corresponding offline SoloVPE measurements performed in triplicate. Details of the fractional least squares regression model are given in Table. 2. The updated data set predicted by the optimized laser/probe universal model showed that 15 of 17 model-based predictions had <5% error compared to the previous 14 of 17 (Figure 5).

Table 1

Comparison of universal model components

Component Description Universal model (item 1) Universal model (U. 2) Laser 1 3 Optics 2 6 Pre-filter 1st derivative with data smoothing 21 cm' ¹ Standard deviation of a normally distributed random variable (SNV) Spectral regions 977-1027 cm' ¹ 1408-1485 cm' ¹ 1621-1711 cm' ¹ 2823-3046 cm' ¹ 2823-3046 cm' ¹

table 2

Universal fractional least squares regression model

for protein concentration (v. 2). detailed information Final models 0-120 g/l >120 g/l Sample size 1412 879 R ² X 0.993 0.987 a ² 0.984 0.958 RMSECV 3.34 7.77

R2X - percentage of variation explained by the model, target: R ² >0.9;

Q ² - percentage of variation predicted by the model during cross-validation, target: Q ² >

0.8;

RMSECV: root mean square error of cross-validation.

Example 2: Improving the performance of protein concentration models.

Materials and methods.

The optimized universal models (v.2) (see Table 1) were tested using a scale-up 1/2 experiment with a disposable tangential flow filtration system (Pall Corporation) with mAb10. The mAb10 loading material was a drug substance and not a typical loading material. The FDS material was diluted to a representative UV/DF loading source, including protein concentration and buffer auxiliaries. However, due to the presence of additional excipients in the loading source that were not tested during development of the generic model, mAb10-specific models were generated. Example 1 describes Raman data collection methods and scan duration information. The mAb-specific model for concentrations > 120 g/L used the same spectral regions and preprocessing methods as the universal model (v. 2). However, in the mAb-specific model for the 0120 g/L concentration, four spectral regions were used, as demonstrated in FIG. 2, with preliminary processing of the standard deviation of a random variable with a normal distribution. Differences in the 0-120 g/L model can be attributed to additional excipients in the loading source.

- 14 046325

Results.

The <5% model error target was achieved for 2/4 of the updated models when the training set was included in the model prediction, as shown in Table 1. 3. Based on the noticeable differences such as loading source, laser, and scale (lab scale vs. scaled up) during the preliminary experiment at IOPS, a second run was conducted. Before the second experiment, changes were made to the mAb 10-specific model with a concentration of 0-120 g/L. Using SNV preprocessing, all four regions were included, but three regions: 977-1027 cm ^-1 , 1408-1485 cm ^-1 and 1621-1711 cm ^-1 additionally applied the 1st derivative with a data smoothing of 21 cm ^-1 . A summary of the generalized experimental results is presented in Table. 3. During the second experiment, the <5% model error target was achieved for 3/4 of the updated models when the training set was included in the model prediction, as shown in Table 1. 3. Additional observation made during data analysis confirmed that the loading source was a major contributor to the increase in error. When the loading pattern was removed, the error of the generic model decreased from 8.6 to 5.7%. In fig. Figure 7 shows the results of in-line prediction (real-time, wide-hatched bars) and updated model results (narrow-hatched bars) for the final concentrated pool, comparing protein concentration with offline measurement using SoloVPE (empty bars). The in-line universal model and updated model had an error of 2.7%, whereas the error of the in-line and updated mAb10 model was 12.0 and 4.2%, respectively, for the final concentrated pool. The increased magnitude of error observed in the mAb 10 model may be due to limited data in the ~250 g/L range, whereas the generic model had a larger data set. An additional factor contributing to the magnitude of error is the inability of the models to extrapolate beyond the characterized range (i.e. >250 g/L in mAb 10 model).

Table 3

Average Model Error for Predicting Protein Concentration as Upscaled

IOPS, experiment No. 1 IOPS, experiment No. 2 In-line Updated In-line Updated ¹ Universal 0-120 g/l 13.8% 8.9% 9.9% 8.6% Universal > 120 g/l 8.8% 2.8% 5.1% 5.0% mAb 11 0-120 g/l 6.0% 5.5% 4.1% 3.5% mAb 11 > 120 g/l 12.0% 1.5% 8.8% 3.6%

Removing the loading pattern reduces the magnitude of the error; Universal (0-120 g/l): from 8.6% to 5.7% and mAbll (0-120 g/l): from 3.5% to 2.7%

Example 3: Upscaling a protein concentration model for processing on pilot scale equipment.

Materials and methods.

In the development of commercially available processes, UV/DF is characterized as a standard process operation in accordance with the principles of Quality established by the design to understand critical process parameters as well as critical quality indicators. To improve understanding of the process and provide an optimized approach to model development, a Raman approach was used to develop mAb11. See Example 1 for the Raman data acquisition method except for the scan duration, which was adjusted from 10 to 5 s. The developed mAb11 model used SNV preprocessing for all four spectral regions except three: 977-1027 cm ^-1 , 1408-1485 cm ^-1 and 1621-1711 cm ^-1 , for which the 1st derivative with data smoothing was additionally applied 21 cm ^-1 .

Results.

The laboratory model was created using four DoE experiments and four spectral regions. In fig. Figure 8A demonstrates the error of the Raman model for real-time prediction from four DoE experiments. In fig. Figure 8B demonstrates the error of the Raman model for 15 additional experiments using the laboratory scale model. mAb-specific protein concentration models were generated for 0-120 g/L and >120 g/L with <5% error.

Using the laboratory model (n=15), experimental runs (n=3) had a prediction error of 0.6-10.6% for mAb11 (Figure 9A). Experimental-scale prediction error decreased to 0.6–2.2% when experimental-scale data were included in the laboratory model (n=18). The increased error magnitude of the laboratory model is likely a result of the effect of temperature on the Raman spectrum shifts due to heat dissipation associated with the scaling hardware. Temperature will be a factor taken into account in future Raman model development and model verification.

- 15 046325

In seven pilot experiments with three different monoclonal antibodies (mAb J, mAb K, and mAb L), the final model prediction error was 0.6–2.2%; which was well within the target 5% (Fig. 9B).

Example 4: Application of Raman models to real-time concentration determination for decision making.

Materials and methods.

Since a similar protocol was used for Raman automation to collect and model Raman spectra, see Example 3 for more information. An automated control strategy was developed using the Raman spectra data to achieve final target protein concentrations. Using the generated predictive models, data was filtered and used for input into the UV/DF instruments until the typical operation was stopped when the target protein concentration was reached. SoloVPE measurements were performed in triplicate.

Results.

In fig. 10A and 10B show an example of screen setup for automatic monitoring of protein concentration under UV/DF and direct flow TFF. In fig. 11 demonstrates that predictive modeling can be used to monitor mAb14 concentrations in real time at various processing steps and run a typical concentration process until stopping when the desired target concentration is reached. The final concentrated pool was predicted by the Raman model to be 260 g/L, compared to the offline SoloVPE measurement of 262 g/L, resulting in an error of 0.8%, corresponding to a target error value of <5%. The Raman approach is a useful tool that can be used to make automated processing decisions that verify the desired target protein concentration is achieved.

Example 5: Proof of Concept for Modeling High Molecular Weight (HMW) Agent Levels in UV/DF.

Materials and methods.

Data collection for HMW agent level modeling included spectral data from the Raman Rxn2 (Kaiser Optical Systems, Inc., Ann Arbor, MI), Raman Rxn Probehead-758 (Kaiser Optical Systems, Inc., Ann Arbor, MI) analyzers ). In addition, several different optical elements were used during the development process depending on their availability. The operating parameters of the Raman analyzers were set to a scan time of 72 s for 1 acquisition, repeated 25 times. On-line measurements were performed at various stages of a typical UV/DF process, including primary concentration, diafiltration and final concentration. The spectral range was 110-3415 cm ^-1 . The raw spectral data were preprocessed using SNV and further filtered using 1st derivative with a data smoothing of 21 cm ^-1 . Off-line measurements of HMW agent levels were performed using size exclusion ultra-performance liquid chromatography.

Results.

The level of HMW agents is another indicator that is considered a preliminary critical quality indicator in protein purification. Current technologies cannot monitor HMW agent levels in real time during processing. In fig. 12A demonstrates that the described Raman modeling technique can be used to monitor HMW agents during protein purification. Prediction of HMW agent levels using Raman modeling was compared with measurements obtained using SE-UHPLC during the purification process (primary concentration, diafiltration, and final concentration). Raman modeling effectively predicted the percentage of high molecular weight compounds in real time during protein processing. This model had an average error of 3.4%.

Example 6: Proof of Concept for Modeling High Molecular Weight (HMW) Agents in Final Chromatographic Purification.

Materials and methods.

Data acquisition for the HMW agent model included spectral data from Raman Rxn2 analyzers (Kaiser Optical Systems, Inc., Ann Arbor, MI) using the MR-Probe-785. The operating parameters of the Raman analyzer were set to scan times of 10, 30, or 60 s for 1 acquisition with 5 repeated measurements. Offline measurements were performed using anion exchange chromatography (AEX) pools with total levels of HMW agents of 6.2–76.2%. The spectral ranges used for modeling and preprocessing methods are described in Table. 4. Autonomous measurements of the level of HMW agents were carried out using size exclusion ultra-performance liquid chromatography.

Results.

The described Raman modeling method can be used to monitor the level of HMW agents during the final chromatographic purification of proteins. HMW Level Prediction

- 16,046,325 agents by Raman modeling were compared with measurements obtained by SE-UHPLC from the generated AEX pools. As demonstrated in Table. 4, the RMSEP value of the evaluated methods ranges from 3.2 to 7.6%. In fig. 12B, a compressed data set (6.2-19.7% HMW) was used to evaluate the model generated with the spectral region 350-3100 cm ^-1 and SNV preprocessing methods. By reducing the HMW range, the RMSEP value decreased to 1.2, 1.4 and 2.1% at 10, 30 and 60 s, respectively. Based on these results, the content of HMW agents can be determined using the Raman model in AEX pools.

Table 4

HMW predictive model error (RMSEP) for HMW agent level from 6.2 to 76.2%

10 sec. 30 sec. 60 sec. 1550-1725 cm ¹ 7.62% 5.31% 3.21% I ^am produced 350-3100 cm ¹ 3.48% 3.37% 4.63% SNV 990-1020, 1550-1725 cm ¹ 7.12% 7.48% 4.98% I ^am produced

Example 7: Proof of Concept for Title Modeling.

Materials and methods.

The training sample model was 35 protein A flowing through samples spiked with FCP (265 g/L) to achieve titers ranging from 0.36 to 9.8 g/L. The model was evaluated on diluted filtrate samples with titers ranging from 1.3 to 8.8 g/L. Data acquisition for the titer model included spectral data from Raman Rxn2 (Kaiser Optical Systems, Inc., Ann Arbor, MI) using MR-Probe-785. The immersion probe was used offline to generate spectral data with operating parameters set to a scan time of 20 s for 1 acquisition repeated 5 times. The spectral ranges were as follows: 977-1027, 1408-1485, 1621-1711 and 2823-3046 cm ^-1 . The raw spectral data were preprocessed using SNV and further filtered using 1st derivative with a data smoothing of 21 cm ^-1 . The characteristics of the model are given in table. 5.

Table 5

Characteristics of the antibody titer model

Spectral areas (cm ^-1 ) 3046-2823, 1711-1621, 1485-1408, 1027-977 Pre-treatment methods ^{I am} derivative and SNV Accumulations and length 5 x 20 seconds Average model error 25.7%

Results.

Antibody titer is a protein purification process parameter that is needed to inform subsequent typical purification operations, including affinity column loading, product consistency, and process intermediate volume limitations. Incorrect column loading can affect subsequent preliminary critical quality indicators, hence the need for monitoring techniques such as Raman spectroscopy. In fig. Figure 13 shows the actual antibody titer in comparison with the monoclonal antibody titer predicted using the Raman approach. In this experiment, the model error was 26%, which is higher than the desired target of <5%. Increasing the scan duration, as well as developing a model using diluted and undiluted deep filtrate, will reduce the magnitude of model error.

Example 8: Raman models for buffer excipient measurements that are within the current orthogonal quantitation error of about 10%.

Materials and methods.

Data were collected from previous rounds of concentration model development for various antibodies. See Example 1 for the Raman data collection method. In table 6 bring

- 17 046325 new model components for detecting histidine and arginine in samples. Spectral regions were based on known histidine/arginine peaks (Zhu, et al., Spectrochim Acta A Mol Biomol Spectrosc, 78(3): 1187-1195 (2011)). After initial development of the histidine and arginine model, further characterization of the model was carried out using mAb14. Using a non-contact optical sensor, the operating parameters of the Raman analyzer were set to a scan time of 20 s for 5 acquisitions. The spectral range for histidine was 1200-1480 cm ^-1 , and for arginine 860-1470 cm ^-1 . For both buffer excipients, the raw spectral data were preprocessed using SNV and further filtered using ^1st derivative with a data smoothing of 21 cm ^-1 . Off-line measurements of histidine and arginine species were performed using the ultra-performance liquid chromatography (UHPLC) amino acid quantification method.

Results.

To determine whether the described method and Raman modeling system could be applied to measure buffer excipients in a processed antibody sample, data collected from previous rounds of preconcentration model development were analyzed for histidine and arginine. The predicted values obtained by Raman modeling were compared with the values calculated by the amino acid method from the UHPLC results, as can be seen in FIG. 14A. Predicted values and average model error for preliminary Raman modeling of histidine/arginine levels are presented in Table. 6. In FIG. 14B shows a scatter plot of the predictive histidine model versus the actual histidine model for mAb14 with an average model error of 8.2%, which is within the <10% target for buffer excipients. The <10% target is based on current variability in orthogonal UHPLC quantitation. In fig. A 14C scatter plot for the predictive arginine model versus the actual arginine model is demonstrated for mAb14 with an average model error of 2.9%, which is within the <10% target.

These data demonstrate that Raman modeling can be used to predict buffer excipient levels in UV/DF and FCP processes. Successful quantification of these excipients ensures that UV/DF will provide a final concentrated pool that will enable subsequent formulation.

Table 6

Model components and data collected for histidine/arginine when running the universal concentration model (Example 1)

G istidin Arginine Spectral area (cm ^-1 ) 1200-1480 970-1100, 1300-1500 Pre-treatment technique ^{I am} derivative and SNV ^{I am} derivative and SNV ^R2Y 0.940 0.964 Q ² 0.938 0.963 RMSEP 1.20 mM 6.19 mM Average model error 10.4% 7.39%

histidine range: 0-25 mM; arginine range: 0-81 mM;

SNV - standard deviation of a random variable with a normal distribution - mean centered and normalized;

R ² - percentage of variation in the training set explained by the model, R ² >0.9;

Q ² is the percentage of variation in the training set predicted by the model during cross-validation;

Q ² >0.8;

RMSEP: Root Mean Square Error Prediction.

Example 9. Raman models for measuring the drug-antibody ratio.

Materials and methods.

DAR is a quality indicator that is monitored during the development of antibody-drug conjugates (ADCs), antibody-radionuclide conjugates (ARCs), and general protein conjugates (potent steroids, noncytotoxic loads, etc.) to ensure consistent product quality and facilitate subsequent product labeling regarding load. The Raman approach was evaluated as a technology for monitoring DAR levels, which could then be used as a reaction control strategy. Two different mAbs that were in development (mAb1 and mAb3) were evaluated for their application to detect DAR using a Raman approach. Using a non-contact optical sensor, the operating parameters of the Raman analyzer were set to a scan time of 10 s for 10 acquisitions. The spectral range of 350-3100 cm ^-1 was used with the original spectral data pre-processed using SNV and further filtered using the 2nd derivative with smooth

- 18 046325 based on data 21 cm ^-1 . In table Figure 7 shows the components of the model for determining DAR in samples. Off-line DAR measurements were performed using a UV spectroscopy-based method.

Table 7

Components of a model for measuring the drug-antibody ratio

Name models REGN2810 UV-DAR REGN910 UV-DAR Y range 0.81-4.39 0.79 -3.57 Best model 2nd ^derivative R2Y=0.788 Q2=0.548 RMSECV=0.64 ^2nd derivative R2Y=0.994 Q2=0.776 RMSEC V=0.58

Results.

In fig. 15A-15B demonstrate that Raman modeling can be used to measure the drug-antibody ratio for iPET-drug conjugates for mAb1 and mAb3. For both models, the cross-validation root mean square error was 0.6 DAR. Current orthogonal UV quantification has a variability associated with 0.3 DAR (one standard deviation). The primary Raman prediction is within two standard deviations and suggests that with further refinement of the model, the DAR can be successfully predicted using the Raman approach.

Although this invention has been described in the foregoing description in connection with certain embodiments thereof and many details have been set forth for the purpose of illustration, it will be apparent to those skilled in the art that the invention is capable of additional embodiments and that some of the details described herein may vary significantly without deviating from the basic principles of the invention.

All references cited herein are incorporated by reference in their entirety. The present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof, and accordingly reference should be made to the appended claims rather than to the foregoing description as an indication of the scope of the invention.

Claims (31)

1. A method for producing a concentrated protein purification intermediate, comprising: providing a universal model obtained using a plurality of protein purification intermediates; determining concentrations of the protein purification intermediate in real time using in situ Raman spectroscopy using said universal model when concentrating the protein purification intermediate; adjusting the parameters of the concentration step in real time to obtain a concentrated protein purification intermediate and/or a final concentrated pool, where the concentration of the protein purification intermediate is at least 150 mg/ml, and wherein said universal model includes a mathematical correlation of the spectral properties of various recombinant proteins, used to predict a critical quality indicator. 2. The method of claim 1, wherein the protein purification intermediate is concentrated by ultrafiltration, buffer exchange, or both. 3. The method of claim 1 or 2, wherein the protein purification intermediate is collected from a bioreactor, fed-batch culture or continuous culture. 4. The method according to any one of claims 1 to 3, wherein the protein purification intermediate is an antibody or an antigen-binding fragment thereof, a fusion protein or a recombinant protein. 5. A method for producing a protein purification intermediate, comprising: independently performing Raman spectroscopy analysis on a plurality of protein purification intermediates to obtain a universal model capable of quantifying any of the plurality of protein purification intermediates; determination of protein purification intermediate concentrations using in situ Raman spectroscopy with a universal model during concentration of the protein purification intermediate; obtaining a concentrated protein purification intermediate when said concentrated protein purification intermediate reaches a predetermined concentration or target of the final concentrated pool, - 19 046325 wherein the concentration of the protein purification intermediate is at least 150 mg/ml, and wherein said universal model includes a mathematical correlation of the spectral properties of various recombinant proteins used to predict a critical quality indicator. 6. The method of claim 5, wherein said model is generated using fractional least squares regression analysis of the raw spectral data and offline protein concentration data. 7. The method of claim 6, further comprising performing a method of normalizing or smoothing the Raman spectroscopy data. 8. The method according to claim 7, wherein said normalization method includes the standard deviation of a random variable with a normal distribution. 9. The method according to claim 7, where the data smoothing includes the 1st derivative with a smoothing of 21 cm ^-1 . 10. The method according to any one of claims 5 to 9, wherein said model provides predicted protein concentration values for a plurality of protein purification intermediates with an error of <5% compared to offline protein concentration values. 11. The method according to any one of claims 5 to 9, wherein said model provides predicted protein concentration values for a plurality of protein purification intermediates with an error of <3% compared to offline protein concentration values. 12. The method according to claim 5, wherein said concentrated protein purification intermediate has a concentration of 150 to 300 mg/ml. 13. The method of claim 5, wherein said concentrated protein purification intermediate is concentrated by ultrafiltration, diafiltration, or both. 14. The method of claim 5, wherein said concentrated protein purification intermediate is collected from a bioreactor, fed-batch culture, or continuous culture. 15. The method according to any one of claims 1 to 14, where the determination of the concentration of the protein purification intermediate product occurs continuously or intermittently in real time. 16. The method according to any one of claims 1 to 15, where the quantitative determination of protein concentration is carried out at intervals of 30 seconds to 10 minutes, hourly or daily. 17. The method according to any one of claims 5 to 16, wherein said protein purification intermediate is an antibody or an antigen binding fragment thereof, a fusion protein or a recombinant protein. 18. A method for monitoring and controlling critical quality attributes in a protein purification intermediate during protein purification downstream processing, comprising: providing a universal model derived from multiple protein purification intermediates; quantifying one or more critical quality attributes in a protein purification intermediate using in situ Raman spectroscopy; adjusting said one or more critical quality attributes in the protein purification intermediate to meet predetermined levels of critical quality attributes; wherein the concentration of the protein purification intermediate is at least 150 mg/ml, and wherein said universal model includes a mathematical correlation of the spectral properties of various recombinant proteins used to predict a critical quality indicator. 19. The method according to claim 18, wherein said critical quality indicators are selected from the group consisting of antibody titer, protein concentration, high molecular weight agents, antibody drug ratio and buffer excipients. 20. The method of claim 18 or 19, wherein the subsequent protein purification process is ultrafiltration/diafiltration. 21. A method for monitoring and controlling the levels of excipients in the collected culture fluid and/or protein purification intermediate during subsequent purification, comprising: providing a universal model derived from multiple protein purification intermediates; real-time determination of excipient concentrations using in situ Raman spectroscopy for purification of cell culture liquid or protein purification intermediate; adjusting the parameters of the purification step in real time to obtain or maintain predetermined amounts of excipients in the collected culture liquid and/or protein purification intermediate, and wherein said universal model includes a mathematical correlation of the spectral properties of various recombinant proteins used to predict a critical quality indicator. - 20 046325 22. The method of claim 21, wherein said excipient includes a buffer excipient. 23. The method according to claim 21 or 22, wherein said excipient is selected from the group consisting of acetate, citrate, histidine, succinate, phosphate, hydroxymethylaminomethane (Tris), proline, arginine, sucrose or combinations thereof. 24. The method according to claim 21, wherein said excipient comprises a surfactant. 25. The method according to claim 21 or 24, wherein said surfactant excipient is selected from the group consisting of polysorbate 80, polysorbate 20 and poloxamer 188. 26. The method according to claim 5, where the specified excipient includes polyethylene glycol and/or sucrose. 27. The method according to claim 1, further including: real-time determination of excipient concentrations using in situ Raman spectroscopy during purification of collected culture fluid and/or protein purification intermediate; and adjusting the parameters of the purification step in real time to obtain and maintain predetermined amounts of excipients in the collected culture fluid and/or protein purification intermediate. 28. The method according to claim 1, further including: quantifying one or more critical quality attributes in a protein purification intermediate using in situ Raman spectroscopy; and adjusting said one or more critical quality attributes in the protein purification intermediate to meet predetermined levels of critical quality attributes, wherein said critical quality attributes are selected from the group consisting of antibody titer, protein concentration, high molecular weight agents, drug-antibody ratio and buffer auxiliaries. 29. The method of claim 1, wherein spectral data is collected in one or more wavenumber ranges selected from the group consisting of 977-1027 cm ^-1 , 1408-1485 cm ^-1 , 1621-1711 cm ^-1 , 28233046 cm ^-1 , and their combinations. 30. The method of claim 1, wherein the protein purification intermediate is a fusion protein. 31. The method of claim 1, wherein the protein purification intermediate comprises an extracellular domain of a cell surface receptor or a fragment thereof.