CN117355751A

CN117355751A - Multiparameter materials, methods, and systems for enhanced bioreactor fabrication

Info

Publication number: CN117355751A
Application number: CN202280037157.4A
Authority: CN
Inventors: C·奥马霍尼哈特内特; B·J·麦卡锡; C·W·罗德; R·海斯; F·马登; J·莫; F·C·马斯兰卡; K·克拉克; D·A·特劳特; P·拉纳; E·格拉齐亚; C·拉弗蒂; K·M·巴尔斯
Original assignee: Janssen Biotech Inc
Current assignee: Janssen Biotech Inc
Priority date: 2021-03-23
Filing date: 2022-03-22
Publication date: 2024-01-05

Abstract

Methods for determining glycan structures on glycosylated and/or glycosylated molecules on a molecule by using a combination of spectroscopic analysis and chemometric modeling are described. In addition, methods and systems for producing desired levels of glycan structures on molecules with desired levels of glycation and/or glycosylated molecules, including non-glycation molecules, are described.

Description

Multiparameter materials, methods, and systems for enhanced bioreactor fabrication

Cross Reference to Related Applications

The present application claims the benefit of U.S. Ser. No. 63/165,040 submitted at 3 months of 2021, U.S. Ser. No. 63/165,048 submitted at 3 months of 2021, U.S. Ser. No. 63/165,055 submitted at 3 months of 2021, and U.S. Ser. No. 63/165,057 submitted at 3 months of 2021, the disclosures of each of these patents being incorporated herein by reference in their entirety.

1. Technical field

The present disclosure relates in part to multiparameter materials, methods, and systems for enhanced bioreactor manufacturing. In particular, the present disclosure relates to methods and systems for controlling and monitoring glycosylation of therapeutic proteins [ e.g., recombinant proteins and/or monoclonal antibodies (mabs) ] during a production process.

2. Background art

Saccharification and glycosylation have been considered as key quality attributes (CQAs) to be considered in therapeutic protein production. For example, glycation can potentially affect the biological activity and molecular stability of therapeutic proteins. In addition, glycosylation of therapeutic proteins may affect protein aggregation, solubility, and stability in vitro and in vivo. Thus, the detection and characterization of glycation and/or glycosylation is an important aspect of therapeutic protein production.

3. Summary of the invention

In this context, the inventors of the present disclosure have discovered methods and systems for controlling and monitoring the saccharification and/or glycosylation of molecules (such as therapeutic proteins disclosed in section 5.1) during the manufacturing process.

In one aspect, provided herein is a method for determining glycan structures on a glycosylated molecule, the method comprising: obtaining, for each of a plurality of runs, a level of one or more glycan structures on the glycosylated molecule using a Process Analysis Technique (PAT) tool, wherein the obtaining is performed within one or more first bioreactors having a first volume equal to or less than a first threshold, the PAT tool obtaining spectral data; generating one or more regression models based on the obtained spectral data, the regression models correlating levels of the one or more glycan structures on the glycosylated molecule with the obtained spectral data; measuring the one or more glycan structures on the glycosylated molecule using the PAT tool, wherein the measuring is performed within one or more second bioreactors having a second volume equal to or greater than a second threshold to produce measured spectral data; and determining, by the at least one computing device, a level of the one or more glycan structures on the glycosylated molecules within the one or more second bioreactors using the generated one or more regression models and based on the measured spectral data.

In one embodiment, the method further comprises refining one or more regression models based on a combination of the obtained spectral data and the measured spectral data.

In one embodiment, the method further comprises maintaining one or more operating parameters of the one or more second bioreactors based on the determined levels to produce the desired glycosylated molecules.

In one embodiment, the method further comprises selectively modifying one or more operating parameters of the one or more second bioreactors based on the determined level to produce the desired glycosylated molecule. In one embodiment, the one or more operating parameters include pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof. In one embodiment, the nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration. In one embodiment, the concentration of glucose is automatically modified based on the measured spectral data.

In one embodiment, the method further comprises purifying the glycosylated molecule.

In one embodiment, the glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

In one embodiment, the obtaining is performed in two or more bioreactors having different volumes. In one embodiment, the first threshold is about 250 liters or less. In one embodiment, the first threshold is about 100 liters or less. In one embodiment, the first threshold is about 50 liters or less. In one embodiment, the first threshold is about 25 liters or less. In one embodiment, the first threshold is about 10 liters or less. In one embodiment, the first threshold is about 5 liters or less. In one embodiment, the first threshold is about 2 liters or less. In one embodiment, the first threshold is about 1 liter or less. In one embodiment, the second threshold is about 1,000 liters or greater. In one embodiment, the second threshold is about 2,000 liters or greater. In one embodiment, the second threshold is about 5,000 liters or greater. In one embodiment, the second threshold is about 10,000 liters or greater. In one embodiment, the second threshold is about 15,000 liters or greater. In one embodiment, the second threshold is about 10,000 liters to about 25,000 liters. In one embodiment, the second threshold is about 15,000 liters. In one embodiment, the second threshold is at least 5 times the first threshold. In one embodiment, the second threshold is at least 10 times the first threshold. In one embodiment, the second threshold is at least 100 times the first threshold. In one embodiment, the second threshold is at least 500 times the first threshold. In one embodiment, the second volume is from about 0.5 liters to about 250 liters. In one embodiment, the second volume is from about 1 liter to about 50 liters. In one embodiment, the second volume is from about 1 liter to about 25 liters. In one embodiment, the second volume is from about 1 liter to about 10 liters. In one embodiment, the second volume is from about 1 liter to about 5 liters. In one embodiment, the second volume is about 1,000 liters to about 25,000 liters. In one embodiment, the second volume is about 2,000 liters to about 25,000 liters. In one embodiment, the second volume is about 5,000 liters to about 25,000 liters. In one embodiment, the second volume is about 10,000 liters to about 25,000 liters. In one embodiment, the second volume is about 15,000 liters to about 25,000 liters.

In one embodiment, the PAT tool utilizes or otherwise includes raman spectroscopy.

In one embodiment, the one or more regression models include a Partial Least Squares (PLS) model.

In one embodiment, the glycosylation molecule comprises a monoclonal antibody (mAb). In one embodiment, the glycosylation molecule comprises a non-mAb.

In one embodiment, the determining step is performed in the field. In one embodiment, the determining step is performed off-site.

In one embodiment, the determining step is performed in-line, online, offline, or a combination thereof. In one embodiment, the determining step is performed on-line. In one embodiment, the determining step is performed online. In one embodiment, the determining step is performed offline. In one embodiment, the determining step is performed in-line.

In one embodiment, obtaining includes receiving data characterizing spectral data from a PAT tool.

In one embodiment, the generating is performed by one or more computing devices.

In one aspect, provided herein is a method of producing a glycosylated molecule having a desired glycan structure, the method comprising: measuring one or more glycan structures using a Process Analysis Technology (PAT) tool to produce spectral data, wherein the measurement is performed in a bioreactor having a volume equal to or greater than 1,000 liters; determining, by at least one computing device, a level of the one or more glycan structures within the bioreactor using one or more regression models and based on the measured spectral data, wherein the one or more regression models are generated using test runs from at least one bioreactor having a volume of less than or equal to 50 liters and at least one bioreactor having a volume of equal to or greater than 1,000 liters; and maintaining one or more operating parameters of the bioreactor when: the level of desired glycan structures is above a predetermined threshold or the level of undesired glycan structures is below a predetermined threshold; or selectively modifying one or more operating parameters of the bioreactor when: the level of desired glycan structures is below a predetermined threshold or the level of undesired glycan structures is above a predetermined threshold.

In one embodiment, the measurement is performed in-line, on-line, off-line, or a combination thereof. In one embodiment, the measurement is performed in-line. In one embodiment, the measurement is performed on-line. In one embodiment, the measurement is performed off-line. In one embodiment, the measurement is performed on a measurement line.

In one embodiment, the measurement is performed more than once per day. In one embodiment, the measurement is performed about every 5 minutes to 60 minutes. In one embodiment, the measurement is performed about every 10 minutes to 30 minutes. In one embodiment, the measurement is performed about every 10 minutes to 20 minutes. In one embodiment, the measurement is performed about every 12.5 minutes.

In one embodiment, the bioreactor volume is about 2,000 liters or more. In one embodiment, the bioreactor volume is about 5,000 liters or more. In one embodiment, the bioreactor volume is about 10,000 liters or more. In one embodiment, the bioreactor volume is about 15,000 liters or more. In one embodiment, the bioreactor volume is from about 10,000 liters to about 25,000 liters. In one embodiment, the bioreactor volume is about 15,000 liters.

In one embodiment, the desired glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

In one embodiment, the bioreactor is a batch reactor, a fed-batch reactor, a perfusion reactor, or a combination thereof.

In one embodiment, the one or more operating parameters include pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof. In one embodiment, the nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration. In one embodiment, the concentration of glucose is automatically modified based on the measured spectral data.

In one embodiment, the PAT tool comprises raman spectroscopy.

In one aspect, provided herein is a system for producing one or more glycosylated molecules, and the system comprising means for culturing a cell line producing the glycosylated molecules; means for measuring the level of one or more glycan structures, wherein the means generates spectral data; means for generating one or more regression models based on the spectral data; and means for measuring the level of one or more glycan structures in the cell line producing the glycosylated molecule.

In one embodiment, the cell line that produces the glycosylated molecule is a mammalian cell line. In one embodiment, the mammalian cell line is a non-human cell line.

In one embodiment, culturing comprises batch, fed-batch, perfusion, or a combination thereof.

In one embodiment, the culturing comprises a volume of about 2,000 liters or more. In one embodiment, the culturing comprises a volume of about 5,000 liters or more. In one embodiment, the bioreactor volume is about 10,000 liters or more. In one embodiment, the bioreactor volume is about 15,000 liters or more. In one embodiment, the culturing comprises a volume of about 10,000 liters to about 25,000 liters. In one embodiment, the culture comprises a volume of about 15,000 liters.

In one embodiment, the measurement is performed in-line, on-line, off-line, or a combination thereof. In one embodiment, the measurement is performed in-line. In one embodiment, the measurement is performed on a measurement line. In one embodiment, the measurement is performed on-line. In one embodiment, the measurement is performed off-line.

In one embodiment, the one or more glycan structures are selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

In one embodiment, the system further comprises means for selectively modifying one or more operating parameters to enhance the production of the desired one or more glycan structures. In one embodiment, the one or more operating parameters include pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof. In one embodiment, the nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration. In one embodiment, the concentration of glucose is automatically modified based on the spectral data.

In one embodiment, the one or more glycosylation molecules comprise monoclonal antibodies (mabs). In one embodiment, the one or more glycosylation molecules include non-mabs.

In one embodiment, the spectral data comprises raman spectra.

In one embodiment, the system further comprises means for isolating one or more glycosylated molecules.

In one aspect, provided herein is a system for producing one or more glycosylated molecules, and the system comprising: a bioreactor comprising a cell line producing glycosylated molecules; a Process Analysis Technology (PAT) tool that measures one or more glycan structures and generates spectral data; and a processor that correlates the level of one or more glycan structures with the spectral data using one or more regression models.

In one embodiment, the bioreactor is about 2,000 liters or more. In one embodiment, the bioreactor is about 5,000 liters or more. In one embodiment, the bioreactor volume is about 10,000 liters or more. In one embodiment, the bioreactor volume is about 15,000 liters or more. In one embodiment, the bioreactor is from about 10,000 liters to about 25,000 liters. In one embodiment, the bioreactor is about 15,000 liters.

Other aspects, features and advantages of the present invention will be apparent from the following disclosure, including the detailed description of the invention and its preferred embodiments, and the appended claims.

4. Description of the drawings

The foregoing summary, as well as the following detailed description of preferred embodiments of the present application, will be better understood when read in conjunction with the appended drawings. It should be understood, however, that the present application is not limited to the precise embodiments shown in the drawings.

FIGS. 1A and 1B show the Maillard reaction of saccharification of mAbs (FIG. 1A), and the common N-glycan structure associated with N-linked glycosylation of therapeutic proteins such as mAbs (FIG. 1B).

Fig. 2 shows a process flow of the development and testing of a raman-based PLS model.

Fig. 3A-3G show raman PLS model predictions (wavy blue line) and off-line measurements (stable red line) for monosaccharide (fig. 3A), non-saccharification (fig. 3B), G0F-GlcNac (fig. 3C), G0 (fig. 3D), G0F (fig. 3E), G1F (fig. 3F) and G2F (fig. 3G) in a reduced-scale (5L) batch (CTSS).

Fig. 4A-4G show raman PLS model predictions (wavy blue line) and off-line measurements (stable red line) for a manufacturing scale (2,000 l) batch (CTSS), mono-glycosylation (fig. 4A), non-saccharification (fig. 4B), G0F-GlcNac (fig. 4C), G0 (fig. 4D), G0F (fig. 4E), G1F (fig. 4F), and G2F (fig. 4G) without adding or removing additional data from the model.

Fig. 5A-5G show raman PLS model predictions (wavy blue line) and off-line measurements (stable red line) in a manufacturing scale (2000L) batch (CTSS) model, using a model supplemented with process data from a 2000L scale batch run (fig. 5A), non-saccharification (fig. 5B), G0F-GlcNac (fig. 5C), G0 (fig. 5D), G0F (fig. 5E), G1F (fig. 5F) and G2F (fig. 5G).

FIG. 6 illustrates an exemplary system for a process flow diagram 600 showing an exemplary system for providing real-time monitoring of saccharification and/or glycosylation profile of therapeutic proteins in a manufacturing-scale bioreactor.

Fig. 7 shows a process flow diagram 700 depicting the determination of glycan structures on glycosylated molecules.

Fig. 8 shows a process flow diagram 800 depicting the production of glycosylated molecules with a desired glycan structure.

Fig. 9 shows a process flow diagram 900 depicting the determination of saccharification on a molecule.

Fig. 10 shows a process flow diagram 1000 depicting the production of molecules having a desired saccharification level.

Fig. 11A-11F show trend superposition of variable projection importance (VIP) scores of PLS models (fig. 11A) mono-saccharification/non-saccharification, (fig. 11B) G0F-GlcNac, (fig. 11C) G0, (fig. 11D) G0F, (fig. 11E) G1F, and (fig. 11F) G2F (in flow 1 and flow 3).

Fig. 12A-12B show graphs of raman glucose values versus offline glucose values for CSS for glucose models for cell line 1 (fig. 12A) and cell line 2 (fig. 12B).

Fig. 13A-13C show cell line 1, (fig. 13A) lot a: bolus feed, (fig. 13B) batch B: automatic feedback control to 1g/l, (FIG. 13C) batch C: automatic feedback control to 1g/l raman PLS model prediction (fluctuating blue line) and off-line measurement (steady red line).

Fig. 14A-14C show cell line 2, (fig. 14A) lot D: bolus feed, (fig. 14B) batch E: automatic feedback control to 2g/l, (fig. 14C) batch F: automatic feedback control to 2g/l raman PLS model predictions (fluctuating blue line) and offline measurements (steady red line).

Figures 15A-15H show comparisons of the process trends of cell line 1 bolus feed and automatic feedback control batches (figure 15A), VCD (figure 15B), viability% (figure 15C), LDH (figure 15D), pH (figure 15E), O2 (figure 15F), glucose feed volume (figure 15G), titer and saccharification (figure 15H).

Fig. 16A-16H show a comparison of the process trends of cell line 2 bolus feeding and automatic feedback control batches (fig. 16A), VCD (fig. 16B), viability% (fig. 16C), LDH (fig. 16D), pH (fig. 16E), O2 (fig. 16F), glucose feeding volumes (fig. 16G), titers and saccharification (fig. 16H).

5. Detailed description of the preferred embodiments

Current methods for measuring glycation and glycosylation spectra include boric acid affinity chromatography, capillary isoelectric focusing colorimetric assay, and liquid chromatography mass spectrometry (LC/MS). These methods, while capable of providing accurate and precise results, are time consuming and resource consuming. Furthermore, sampling is usually at the cost of taking the product out of the bioreactor and there is a greater risk of contamination. Another disadvantage of the current methods is that once the batch is completed, product quality testing is involved.

Accordingly, there is an unmet need for methods and systems that can accurately monitor glycation and/or glycosylation throughout the production process of therapeutic proteins, which enable users to have the ability to control glycation and/or glycosylation in real-time during the production process.

The present disclosure relates in part to methods and systems for controlling and monitoring saccharification and/or glycosylation of molecules (such as therapeutic proteins disclosed in section 5.1) during a production process. For example, methods and systems are described herein that involve the cooperation of Process Analysis Technology (PAT) tools and chemometric modeling to develop predictive models (particularly focused on manufacturing-scale processes) that are capable of monitoring saccharification and glycosylation spectra, including single glycoforms (micro-heterogeneity). The present disclosure also relates in part to the following findings: by including manufacturing scale data in the stoichiometric modeling, the predictive power and robustness of the model can be improved. Such methods and systems enable identification of potential quality problems in therapeutic protein production processes before they affect batches, and may also help reduce process variability, reduce supply costs due to yield improvements, conduct real-time distribution of products, reduce lead time, and positively impact the technology transfer timeline of products due to reduced analytical method transfer and validation processes.

In some aspects, provided herein is a method for determining glycan structures on glycosylated molecules using PAT tools such as raman spectroscopy and generating a regression model that can be used by a computing device to determine the level of one or more glycan structures on a therapeutic protein within a bioreactor. In other aspects, provided herein is a method for producing glycosylated molecules having desired glycan structures using PAT tools that generate spectral data and determining the level of one or more glycan structures on a therapeutic protein within a bioreactor by using one or more regression models. One or more operating parameters may then be maintained or selectively modified based on the level of desired and/or undesired glycan structures. The present disclosure also provides a system for producing one or more glycosylated molecules. The system may include: a bioreactor comprising a cell line capable of producing glycosylated molecules; PAT tool measuring one or more glycan structures and generating spectral data; and a processor that correlates the level of the one or more glycan structures with the spectral data using the one or more regression models.

In some aspects, provided herein is a method for determining saccharification on a molecule using a Process Analysis Technology (PAT) tool, such as raman spectroscopy, and generating a regression model that can be used by a computing device to determine a level of saccharification on a molecule within a bioreactor. In other aspects, provided herein is a method of producing a molecule having a desired level of saccharification by measuring saccharification on the molecule using PAT tools that generate spectral data and determining the level of saccharification on the molecule using one or more regression models. One or more operating parameters may then be maintained or selectively modified horizontally based on the level of saccharification. The present disclosure also provides a system for producing non-glycosylated molecules. The system may include: a bioreactor comprising a cell line capable of producing non-glycosylated molecules; PAT tool measuring saccharification and generating spectral data; and a processor that correlates the saccharification level with the spectral data using one or more regression models.

Various publications, articles and patents are cited or described throughout the specification; each of these references is incorporated by reference herein in its entirety. The discussion of documents, acts, materials, devices, articles or the like which has been included in the present specification is intended to provide a context for the present invention. Such discussion is not an admission that any or all of these matters form part of the prior art base with respect to any of the inventions disclosed or claimed.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Otherwise, certain terms used herein have the meanings set forth in the specification. All patents, published patent applications, and publications cited herein are hereby incorporated by reference as if fully set forth herein.

It should be noted that, as used herein and in the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise.

Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", and variations such as "comprises" and "comprising", will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps. The term "comprising" as used herein may be replaced with the term "containing" or "including" or sometimes with the term "having" as used herein.

As used herein, "consisting of … …" excludes any element, step or ingredient not specified in the claim elements. As used herein, "consisting essentially of … …" does not exclude materials or steps that do not materially affect the basic and novel characteristics of the claims. Whenever used herein in the context of one aspect or embodiment of the present application, any of the foregoing terms "comprising," "including," "comprising," and "having" may be substituted with the terms "consisting of … …" or "consisting essentially of … …" to alter the scope of this disclosure.

As used herein, the connection term "and/or" between a plurality of recited elements is understood to encompass both single options and combined options. For example, where two elements are connected by an "and/or," a first option refers to the first element being applicable without the second element. The second option refers to the second element being applicable without the first element. A third option refers to the first element and the second element being adapted to be used together. Any of these options is understood to fall within the meaning and thus meet the requirements of the term "and/or" as used herein. Parallel applicability of more than one option is also understood to fall within the meaning and thus meet the requirements of the term "and/or".

As used herein, the term "molecule" generally refers to a protein or fragment thereof. Exemplary molecules are therapeutic proteins (e.g., recombinant proteins or mabs) or fragments thereof.

As used herein, the term "bioreactor" generally refers to a device that supports bioactive processes (such as the culture of cells). Exemplary bioreactors include stainless steel stirred tank bioreactors, airlift reactors, and disposable bioreactors.

As used herein, the term "spectral data" generally refers to an analytical output obtained using spectroscopy.

As used herein, the term "threshold" generally refers to an amount defining at least one limit (such as a lower limit or an upper limit) of a particular range or level. The threshold may be determined empirically or it may be a predetermined threshold set a priori.

As used herein, the term "selectively modifying" when used in reference to an operating parameter generally refers to purposefully adjusting one or more conditions to promote optimal production of a desired product and/or to reduce production of an undesired product.

As used herein, the term "automatically modify" generally means adjusting an operating parameter without requiring the user to manually adjust the operating parameter.

As used herein, the term "in the field" generally means that the measurement or determination is made in the same facility where the production occurs.

As used herein, the term "off-site" generally means that the measurement or determination is made in a facility other than where production occurs.

As used herein, the term "offline" generally means that measurement or determination is performed after the production process has been completed, and that collection of the sample is manual.

As used herein, the term "on-line" generally means that the measurement or determination is made within the production area and that the collection of the sample is manual.

As used herein, the term "in-line" generally means that the measurement or determination is made within the production area and that the collection of the sample is automated.

As used herein, the term "in-line" generally means that measurement or determination is performed in real-time within the production area by probes placed in the bioreactor, and collection of the sample is not required.

Practice of the embodiments provided herein will employ, unless otherwise indicated, conventional techniques of molecular biology, microbiology and immunology, which are within the skill of the art. Such techniques are well explained in the literature. Examples of text that is particularly suitable for consultation include the following: sambrookfal, molecular cloning: ALaboratoryManual, thirdEd., coldSpring Harbor Laboratory, new York (2001); ausubel et al Current Protocols in Molecular Biology, john Wiley and Sons, baltimore, MD (1999); glover, ed., DNA Cloning, volumes I and II (1985); freshney, ed., animal Cell Culture: immobilized Cells and Enzymes (IRL Press, 1986);et al,Plant Molecular Biology–A Laboratory Manual(Ed.by Melody S.Clark；Springer–Verlag,1997)；ImmunochemicalMethodsinCell andMolecularBiology(AcademicPress,London)；Scopes,Protein Purification:Principles and Practice(Springer Verlag,N.Y.,2d ed.1987)；National Research Council(US)Committee on Methods of ProducingMonoclonalAntibodies.MonoclonalAntibodyProduction.Washington(DC):National Academies Press(US)；1999；and Clausen H,et al.Glycosylation Engineering.2017.In:Varki A,Cummings RD,Esko JD,etal.,editors.Essentials of Glycobiology.3rd edition.Cold Spring Harbor(NY):Cold Spring Harbor Laboratory Press；2015-2017；andNational Research Council(US)Committee on Revealing Chemistry through Advanced Chemical Imaging.Visualizing Chemistry:The Progress and Promise of Advanced Chemical Imaging.Washington(DC):National Academies Press(US)；2006.3,Imaging Techniques:State ofthe Art and Future Potential.

to assist the reader of this application, the specification has been divided into individual paragraphs or chapters, or directed to various embodiments of this application. These divisions should not be considered as breaking apart the main content of a paragraph or section or embodiment from the main content of another paragraph or section or embodiment. Rather, those skilled in the art will appreciate that the present description has broad application and encompasses all combinations of individual chapters, paragraphs and sentences that may be envisioned. The discussion of any embodiment is intended to be exemplary only, and is not intended to suggest that the scope of the disclosure, including the claims, is limited to these examples. The present application contemplates the use of any suitable components in any combination, whether or not a particular combination is explicitly described.

5.1.Therapeutic proteins

The present disclosure relates in part to determining and/or measuring saccharification and/or glycosylation of molecules. In certain embodiments, the molecule is a therapeutic protein. Non-limiting examples of therapeutic proteins include, for example, antibody-based drugs (e.g., polyclonal or monoclonal antibodies (mabs)), fc fusion proteins, anticoagulants, blood factors, bone morphogenic proteins, engineered protein scaffolds, enzymes, growth factors, hormones, interferons, interleukins, recombinant proteins, and thrombolytics. Thus, in some embodiments, the molecule is a mAb. In some embodiments, the molecule is a recombinant protein. In some embodiments, the molecule is an Fc fusion protein. In some embodiments, the molecule is an anticoagulant. In some embodiments, the molecule is a blood factor. In some embodiments, the molecule is a bone morphogenic protein. In some embodiments, the molecule is an engineered protein scaffold. In some embodiments, the molecule is an enzyme. In some embodiments, the molecule is a growth factor. In some embodiments, the molecule is a hormone. In some embodiments, the molecule is an interferon. In some embodiments, the molecule is an interleukin. In some embodiments, the molecule is a thrombolytic agent.

5.2.Saccharification

The present disclosure relates in part to determining and/or measuring glycation of therapeutic proteins, such as those disclosed in section 5.1. Protein glycation is a non-enzymatic glycosylation on protein amine groups, typically occurring at the alpha amine terminus and epsilon amine groups on lysine side chains. Saccharification involves the process of covalent binding of reducing sugars (such as glucose, fructose and galactose) to proteins in a non-enzymatic reaction. Maillard in 1912 described for the first time the reaction between amino acids and reducing sugars. The sensitive amine groups are reversibly condensed with the aldehyde groups of the reducing sugar to form an unstable Schiff base intermediate that can undergo spontaneous multi-step Amadori rearrangement to form a more stable covalently bonded ketoamine. The reaction produces irreversible products, causing biophysical and structural changes in the protein.

Saccharification may potentially affect the biological activity and/or molecular stability of the therapeutic protein. For example, in the case of mabs representing exemplary therapeutic proteins of the present disclosure, saccharification may block biologically functional sites and/or cause degradation of the mAb. Degradation can further lead to aggregation of mabs. Thus, glycation of therapeutic proteins represents a potentially Critical Quality Attribute (CQA) of the production process.

5.3.Glycosylation

The present disclosure also relates in part to determining and/or measuring glycosylation of therapeutic proteins (such as those disclosed in section 5.1), including specific glycan structures.

Glycosylation is a complex post-translational modification involving attachment of glycans to specific sites on the therapeutic protein (e.g., N-linked and/or O-linked glycosylation). For example, N-linked glycosylation of mabs generally involves the attachment of glycans at the Asn-X-Ser/Thr sequence of the Fc portion of the mAb heavy chain, where X can be any amino acid other than proline (Ghaderi et al 2012). For example, a therapeutic protein may comprise an IgG1 molecule that contains a single N-linked glycan at Asn297 in each of the two heavy chains. N-glycosylation can also occur in the variable region of each heavy chain (e.g., cetuximab). O-linked protein glycosylation generally involves the linkage between the monosaccharide N-acetylgalactosamine and the amino acid serine or threonine.

These carbohydrate species (referred to as "glycans") can be derived from the arrangement of carbohydrate moieties covalently attached to specific regions of the therapeutic protein. For example, N-glycans include G0, G0F, G F-GlcNac, G1F, and G2F, which differ slightly in structure (see FIG. 1B). These changes in glycan structure can result in subtle changes in structure, which can have a significant impact on protein activity, conformation, safety and efficacy. Thus, glycosylation of therapeutic proteins plays a key role in their safety and efficacy (including immunogenicity), and proper glycosylation is one of the key quality attributes (CQAs) that must be demonstrated to ensure safety and efficacy of commercial therapeutic proteins such as mabs prior to approval by regulatory authorities.

5.4.Process Analysis Technology (PAT)

Current methods for measuring glycation and glycosylation spectra include boric acid affinity chromatography, capillary isoelectric focusing colorimetric assay, and liquid chromatography mass spectrometry (LC/MS). These methods, while capable of providing accurate and precise results, are time consuming and resource consuming. Furthermore, sampling using these methods generally comes at the cost of removing the product from the bioreactor and the risk of contamination. Another disadvantage of the current methods is that once the batch is completed, product quality testing is involved.

As provided herein, the present disclosure relates in part to measuring saccharification and/or glycosylation using Process Analysis Techniques (PAT). PAT is advantageous because it can be used to control and monitor the production process before the batch is completed. For example, PAT may allow for measuring samples in real time (such as by in-line measurement) or otherwise during the production process (such as by in-line or on-line measurement). For example, real-time monitoring can increase process control because it provides the opportunity to identify potential quality problems before they affect a lot, reduce process variability, reduce supply costs due to increased yield, conduct real-time distribution of products, reduce lead time, and positively impact the technology transfer timeline of products due to reduced analytical method transfer and validation processes. In contrast, complex and time-consuming methods that rely on measuring the frequency of off-line samples can only provide limited insight into these CQAs (such as saccharification and glycosylation), and are typically only evaluated after the bioreactor process is complete. Thus, by applying PAT during the production process, the quality of the product can be monitored more carefully and the final product produced will have a higher quality than relying on an off-line (end-of-line) test to filter out production of products that do not meet specifications.

5.4.1.PAT tool

As provided herein, detection and/or measurement of saccharification and/or glycosylation may be performed using PAT tools that involve spectroscopic techniques such as fluorescence spectroscopy, diffuse reflectance spectroscopy, infrared spectroscopy (e.g., near-infrared or mid-infrared), terahertz spectroscopy, transmission and absorption spectroscopy, raman spectroscopy, including Surface Enhanced Raman Spectroscopy (SERS), spatially Offset Raman Spectroscopy (SORS), transmission raman spectroscopy, and/or resonance raman spectroscopy.

By way of example, raman spectroscopy is a PAT tool capable of providing data that can cooperate with chemometric modeling (such as that described in section 5.4.2). It provides a clear, sharp spectrum and can be optionally recorded within the bioreactor (e.g., in situ) using an in situ probe during the production process. Raman spectroscopy is a vibrational spectroscopy technique that uses laser technology to provide chemical fingerprints of substances.

In addition, it has been used as a PAT tool to make non-destructive real-time measurements of many biotherapeutic process variables including metabolites, growth profiles, product levels, product quality attributes, nutrient supplements, and most recently culture pH. However, it should be understood that other spectroscopic techniques may also be used with the methods and systems provided herein, and that raman spectroscopy is merely an exemplary PAT tool.

The spectra obtained from any PAT tool described herein may be collected with a probe within the bioreactor (i.e., in situ), or by collecting a sample from the bioreactor and measuring the sample outside the bioreactor. The spectra may also be combined with multivariate analysis (MVA) to allow monitoring of other operating parameters besides saccharification and/or glycosylation, such as metabolites and/or cell concentrations. For example, data regarding one or more operating parameters (e.g., pH, temperature, pressure, dissolved oxygen, optical density, oxygen uptake rate, etc.) may be combined with support information (raw material analysis, timing and duration of feed, manual cell count, metabolite levels, etc.) to generate a large amount of high-dimensional data regarding the production process, which may be processed through chemometric modeling methods (such as chemometric modeling described in section 5.4.2). Spectra may also be collected before and after glucose feeding.

Thus, by combining spectroscopic analysis with chemometric modeling methods (such as chemometric modeling described in section 5.4.2), real-time monitoring of saccharification and/or glycosylation, and optionally additional information about the production process, can be achieved. Thus, spectroscopy (such as raman spectroscopy) provides the ability to monitor biological processes during the production process rather than after the run is complete. Thus, the spectrum may be used as a PAT tool to measure saccharification and/or glycosylation, optionally with the measurement being implemented as a feedback loop to control one or more operating parameters (such as nutrient feed) to produce a desired amount of saccharification and/or glycosylation, and optionally a specific glycan structure, on the therapeutic protein.

5.4.2.Chemometric modeling

The spectroscopic methods provided herein can generate a large amount of high-dimensional data. Generally, this data is processed by chemometric modeling methods using known techniques such as Partial Least Squares (PLS), classical Least Squares (CLS), or Principal Component Analysis (PCA). The model is then built with the captured absorption spectra and reference levels obtained offline, such as by liquid chromatography-mass spectrometry (LC-MS). The spectrum may be subjected to preprocessing methods such as first and second order or derivative, extended multiplicative scatter correction, average centering, and automatic scaling, etc. Pretreatment methods can be used to help mitigate disturbances such as turbidity or translucency of the fluid, instrument drift, and contamination accumulation on lenses in contact with the fluid. The pretreatment method also acts as a noise filter to enable the model to focus on the actual compositional changes in the fluid that can affect the final liquid vapor pressure. The chemometric model is then implemented as a calibration curve to the PAT tool analyzer to predict saccharification and/or glycosylation in real-time.

Thus, in some embodiments, the one or more regression models are selected from the group consisting of Partial Least Squares (PLS), classical Least Squares (CLS), and Principal Component Analysis (PCA). In some embodiments, the one or more regression models include a Partial Least Squares (PLS) model. In some embodiments, the one or more regression models include a Classical Least Squares (CLS) model. In some embodiments, the one or more regression models include Principal Component Analysis (PCA).

5.4.3.Measuring parameters

As provided herein, saccharification and/or glycosylation can be measured during the production process (e.g., online, and/or in-line) or after the production process is completed (e.g., offline) using PAT tools such as raman spectroscopy. Thus, in some embodiments, the measurement is performed in-line, online, offline, or a combination thereof. In some embodiments, the measurement is performed in-line. In some embodiments, the measurement is performed online. In some embodiments, the measurement is performed on a measurement line. In some embodiments, the measurement is performed off-line.

Saccharification and/or glycosylation can also optionally be measured on-site or off-site. In some embodiments, the measurement is performed in the field. In some embodiments, it is performed off-site.

The PAT tools disclosed herein that provide spectra can provide frequent measurements during the production process. For example, by using raman spectroscopy, when a single probe is used, it is possible to deliver predictions of a series of process variables that are typically limited to a single daily offline measurement at a frequency of once every 15 minutes, thereby generating up to 360 times the amount of information relative to a single daily offline measurement for a 16 day process.

Thus, in some embodiments, the measurement may be performed more than once per day. If more frequent measurements are desired, the measurements may be made about every 5 minutes to 60 minutes, about every 10 minutes to 30 minutes, about every 10 minutes to 20 minutes, or about every 12.5 minutes. Generally, more frequent measurements will be able to allow for more accurate predictions of glycation and/or glycosylation levels. This, in turn, will allow the user to better control the ability of the molecule to saccharification and/or glycosylation levels.

5.5.Yield of products

As provided herein, the present disclosure relates in part to measuring saccharification and/or glycosylation on a therapeutic protein (such as the therapeutic protein disclosed in section 5.1) produced by culturing a biologically active organism (such as by using the cell culture method described in section 5.5.2) in a bioreactor (such as the bioreactor described in section 5.5.1).

5.5.1.Bioreactor

As provided herein, the present disclosure relates in part to measuring glycation and/or glycosylation on a therapeutic protein grown in a bioreactor (such as the therapeutic protein disclosed in section 5.1). Various types of bioreactors may be used to produce therapeutic proteins. For example, the bioreactor may be a stainless steel stirred tank bioreactor (STR), an airlift reactor, a disposable bioreactor, or a combination thereof (e.g., a disposable bioreactor combined with an STR).

The bioreactor may have any suitable volume that allows for the culture and propagation of biological cells capable of producing a therapeutic protein, such as the therapeutic protein disclosed in section 5.1. For example, the volume of the bioreactor may be from about 0.5 liters (L) to about 25,000L. In some embodiments, the volume of the bioreactor may be less than or equal to about 250L. In some embodiments, the volume of the bioreactor may be from about 0.5 liters (L) to about 250L. In some embodiments, the volume of the bioreactor may be less than or equal to about 50L. In some embodiments, the volume of the bioreactor may be from about 1L to about 50L. In some embodiments, the volume of the bioreactor may be less than or equal to about 25L. In some embodiments, the volume of the bioreactor may be from about 1L to about 25L. In some embodiments, the volume of the bioreactor may be less than or equal to about 10L. In some embodiments, the volume of the bioreactor may be less than or equal to about 5L. In some embodiments, the volume of the bioreactor may be less than or equal to about 1L. In some embodiments, the volume of the bioreactor may be about 1L. In some embodiments, the volume of the bioreactor may be about 2L. In some embodiments, the volume of the bioreactor may be less than or equal to about 5L. In some embodiments, the volume of the bioreactor may be less than or equal to about 10L. In some embodiments, the volume of the bioreactor may be less than or equal to about 25L. In some embodiments, the volume of the bioreactor may be less than or equal to about 50L. In some embodiments, the volume of the bioreactor may be less than or equal to about 100L. In some embodiments, the volume of the bioreactor may be less than or equal to about 250L. In some embodiments, the volume of the bioreactor may be equal to or greater than 1,000l. In some embodiments, the volume of the bioreactor may be from about 1,000l to about 25,000L. In some embodiments, the volume of the bioreactor may be from about 10,000l to about 25,000L. In some embodiments, the volume of the bioreactor may be about 1,000l. In some embodiments, the volume of the bioreactor may be about 2,000l. In some embodiments, the volume of the bioreactor may be less than or equal to about 5,000l. In some embodiments, the volume of the bioreactor may be less than or equal to about 10,000l. In some embodiments, the volume of the bioreactor may be less than or equal to about 15,000l. In some embodiments, the volume of the bioreactor may be less than or equal to about 25,000L.

As provided herein, in certain aspects of the disclosure, the data obtained from a scaled-down bioreactor (e.g., less than 250L) may be first used to generate a stoichiometric model, and the robustness and predictive ability of the stoichiometric model may be increased by refining the model using data obtained from a manufacturing-scale bioreactor (e.g., greater than or equal to 2,000L, and preferably between 10,000L and 25,000L) for predicting production-scale saccharification and/or saccharification. Thus, in some embodiments, the methods and systems described herein involve two or more bioreactors having different volumes.

The addition of data obtained from a manufacturing scale bioreactor improves the predictive ability of the model. This may be due to variations in bioreactor culture that occur in manufacturing scale bioreactors, rather than during downscaling. For example, culture mixing time, CO ₂ The removal and oxygen transfer rates may vary depending on the scale of manufacture. By way of example, reduced oxygen transfer during manufacturing scale is likely to affect saccharification, particularly glycosylation profile, of mAb products in a manner that would not be seen at reduced scale. The reduced oxygen transfer results in the possible presence of a "dead zone" in the bioreactor where no oxygen is present for a short period of time (ash et al, 2019). This results in increased oxidative stress on the host cells in the bioreactor. Oxidative stress has been reported to have an effect on the glycosylation of mabs, as such stress reduces acetyl-CoA formation, which in turn results in a reduction of N-acetylglucosamine (GlcNac) (Lewis et al 2016), a key amide that forms part of the backbone structure of the glycosylation targets described herein in addition to G0F-GlcNac. Thus, by combining data obtained from a manufacturing scale process, these variations can be accounted for in the model and help to improve predictability of the model.

Thus, in some embodiments, the chemometric model relates to data generated from one or more first bioreactors and data generated from one or more second bioreactors. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 250L. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 100L. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 50L. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 25L. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 10L. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 5L. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 2L. In some embodiments, each of the one or more first bioreactors has a maximum threshold volume of less than or equal to about 1L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 250L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 100L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 50L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 25L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 10L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 5L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 2L. In some embodiments, each of the one or more first bioreactors has a volume of less than or equal to about 1L. In some embodiments, the volume of each of the one or more first bioreactors may be from about 0.5 liters (L) to about 250L. In some embodiments, the volume of each of the one or more first bioreactors may be from about 1L to about 50L. In some embodiments, the volume of each of the one or more first bioreactors may be from about 1L to about 25L. In some embodiments, the volume of each of the one or more first bioreactors may be from about 1L to about 10L. In some embodiments, the volume of each of the one or more first bioreactors may be from about 1L to about 5L.

In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of greater than or equal to about 1,000 l. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of greater than or equal to about 2,000 l. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of greater than or equal to about 5,000 l. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of greater than or equal to about 10,000 l. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of greater than or equal to about 15,000 l. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of greater than or equal to about 20,000 l. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of greater than or equal to about 25,000L. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of about 2,000l to about 25,000L. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of about 5,000l to about 25,000L. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume of about 10,000l to about 25,000L. In a preferred embodiment, each of the one or more bioreactors has a minimum threshold volume equal to the volume of the bioreactor used to produce the therapeutic protein. In some embodiments, each of the one or more second bioreactors has a volume of greater than or equal to about 1,000 l. In some embodiments, each of the one or more second bioreactors has a volume of greater than or equal to about 2,000 l. In some embodiments, each of the one or more second bioreactors has a volume of greater than or equal to about 5,000 l. In some embodiments, each of the one or more second bioreactors has a volume of greater than or equal to about 10,000 l. In some embodiments, each of the one or more second bioreactors has a volume of greater than or equal to about 15,000 l. In some embodiments, each of the one or more second bioreactors has a volume of greater than or equal to about 20,000 l. In some embodiments, each of the one or more second bioreactors has a volume greater than or equal to about 25,000L. In some embodiments, each of the one or more second bioreactors has a volume of about 1,000l to about 25,000L. In some embodiments, each of the one or more second bioreactors has a volume of about 2,000l to about 25,000L. In some embodiments, each of the one or more second bioreactors has a volume of about 5,000l to about 25,000L. In some embodiments, each of the one or more second bioreactors has a volume of about 10,000l to about 25,000L. In some embodiments, each of the one or more second bioreactors has a volume of about 15,000l to about 25,000L. In a preferred embodiment, each of the one or more bioreactors has a volume equal to the volume of the bioreactor used to produce the therapeutic protein.

In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least five times (5 times) the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least 10 times the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least 25 times the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least 50 times the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least 100 times the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least 250 times the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least 500 times the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a minimum threshold volume that is at least 1,000 times the maximum threshold volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least five times (5 times) the volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least 10 times the volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least 25 times the volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least 50 times the volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least 100 times the volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least 250 times the volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least 500 times the volume of each of the one or more first bioreactors. In some embodiments, each of the one or more second bioreactors has a volume that is at least 1,000 times the volume of each of the one or more first bioreactors.

5.5.2.Cell culture method

The production of therapeutic proteins of the present disclosure (such as those disclosed in section 5.1) can be performed with any suitable biologically active host cell type known in the art capable of producing therapeutic proteins. For example, mammalian cells and non-mammalian cells can be used as a platform for the production of therapeutic proteins. Non-limiting examples of mammalian host cell lines for the production of therapeutic proteins include Chinese Hamster Ovary (CHO), mouse myeloma derived NS0 and Sp2/0 cells, human embryonic kidney cells (HEK 293) and human embryonic retinal cell derived per.c6 cells. Non-limiting examples of non-mammalian hosts include, for example, pichia pastoris (Pichia pastoris), arabidopsis thaliana (Arabidopsis thaliana), nicotiana benthamiana (Nicotiana benthamiana), aspergillus niger (Aspergillus niger), and Escherichia coli (Escherichia coli). Thus, in some embodiments, the cell line that produces the therapeutic protein is a mammalian cell line. In some embodiments, the mammalian cell line is a non-human cell line.

Different host systems may express different glycosylases and transporters, thereby generating specificity and heterogeneity of the glycosylation profile of the therapeutic protein. Similarly, different host systems may have different effects on the glycation level of therapeutic proteins. Thus, host cell lines that can produce a desired therapeutic protein having, for example, a particular glycan structure and/or glycation level can be engineered or specifically selected.

The culturing and propagation of the bioactive cells can be performed using various methods known in the art, such as batch methods, fed-batch methods, continuous culture, or combinations thereof (e.g., mixing between fed-batch and perfusion).

In the batch process, all nutrients are supplied at the beginning of the culture and are no longer added during the subsequent biological process. No additional nutrients are added during the whole biological process, but control elements such as gases, acids and bases may optionally be added. The biological process then continues until the nutrients are consumed.

The fed-batch process adds nutrients as a bolus to the bioreactor at a set point in time or once the nutrients are depleted. Generally, the same medium used in the initial culture is also used for the feed, but in a more concentrated form. The feed solution composition may be designed to supply cells based on their metabolic state at different culture stages, such as by, for example, analyzing and identifying spent media nutrients that are consumed at a greater rate. In addition, the media used in the fed-batch process may even be modified to accommodate the needs of the cell culture or to facilitate the production of therapeutic proteins, such as to facilitate cell growth or to stimulate the production of therapeutic proteins. For example, in some embodiments, the culture medium may be modified to reduce or eliminate glycation of therapeutic proteins.

In continuous culture, nutrients are continuously added to the bioreactor and an equal amount of the converted nutrient solution is generally withdrawn from the system simultaneously. The three most common types of continuous culture are chemostat, turbidimeter and perfusion. The perfusion method circulates the medium in the growing culture, allowing for simultaneous removal of waste, nutrient supply and harvest of the product.

As provided herein, in some embodiments, the feeding is automatic. In certain aspects, the feeding is automatic and one or more operating parameters (e.g., the operating parameters described in section 5.5.3) are automatically controlled.

5.5.3.Operating parameters

Saccharification of therapeutic proteins can occur during the production process, wherein a reducing sugar (such as glucose) is used as an energy source for a biologically active organism (such as a cell culture producing the therapeutic protein). The level of saccharification may be affected by the amount of sugar added to the cell culture during mammalian cell culture. In addition, other factors such as pH level, nutrient level, time (e.g., frequency interval of medium addition), temperature, and ionic strength can affect the kinetics and extent of saccharification. Furthermore, the particular reactivity of the particular type of sugar (such as hexose) and accessible amino groups used in the cell culture medium can affect protein glycation.

Many process parameters can affect the glycosylation of therapeutic proteins, such as Dissolved Oxygen (DO) levels, the culture temperature of the bioreactor, host cell type, and nutrient or supplement availability.

Thus, in some embodiments, one or more operating parameters of the bioreactor are maintained or selectively modified based on the determined saccharification and/or glycosylation level to produce a therapeutic protein having an acceptable amount of saccharification and/or glycosylation. In some embodiments, the operating parameters include temperature, pH, dissolved Oxygen (DO) level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof.

In some embodiments, the operating parameter comprises temperature. For example, if the therapeutic protein has a desired amount of glycation and/or glycosylation, the temperature of the bioreactor may be maintained at or near the temperature in the bioreactor at which the level of glycation and/or glycosylation is measured. Alternatively, if the therapeutic protein has an undesirable amount of saccharification and/or glycosylation, the temperature of the bioreactor may be adjusted for a portion or all of the production process to achieve the desired amount of saccharification and/or glycosylation. In some embodiments, the temperature is a physiological temperature. In some embodiments, the temperature is maintained or adjusted to a temperature of about 25 ℃ to 42 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 35 ℃ to 39 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 35.5 ℃ to 37.5 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 35 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 35.5 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 36 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 36.5 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 37 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 37.5 ℃. In certain embodiments, the temperature in the bioreactor is a non-physiological temperature. In some embodiments, the temperature is maintained or adjusted to a temperature below 25 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 4 ℃ to about 25 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 4 ℃ to about 10 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 4 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 5 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 6 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 7 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 8 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 9 ℃. In some embodiments, the temperature is maintained or adjusted to a temperature of about 10 ℃.

In some embodiments, the operating parameter comprises a pH level maintained or modified depending on the measured level of glycation and/or glycosylation on the therapeutic protein. For example, if the therapeutic protein has a desired amount of saccharification and/or glycosylation, the pH in the bioreactor may be maintained at or near the pH level in the bioreactor at the time the saccharification and/or glycosylation level is measured. A high pH (. Gtoreq.7.0) may be beneficial for the initial cell growth phase. However, high pH, high lactate and high osmotic pressure cascades generally lead to delayed cell growth and accelerated cell death. Thus, it may also be necessary to adjust the pH level during the production process depending on the growth stage in which the bioactive cells are located. For example, the addition of carbon dioxide and/or sodium carbonate may be used to maintain or adjust the pH.

In some embodiments, the pH is a physiological pH. In some embodiments, the pH is between pH4.0 and pH 9.0. In some embodiments, the pH is between pH 5.0 and pH 8.0. In some embodiments, the pH is between pH6.0 and pH 7.0. In some embodiments, the pH is between pH4.0 and pH6.0. In some embodiments, the pH is between pH 5.0 and pH7. In some embodiments, the pH is between pH6.0 and pH 8.0. In some embodiments, the pH is maintained or adjusted to about pH 4.0. In some embodiments, the pH is maintained or adjusted to about pH 4.5. In some embodiments, the pH is maintained or adjusted to about pH 5.0. In some embodiments, the pH is maintained or adjusted to about pH 5.5. In some embodiments, the pH is maintained or adjusted to about pH6.0. In some embodiments, the pH is maintained or adjusted to about pH6.5. In some embodiments, the pH is maintained or adjusted to about pH6.6. In some embodiments, the pH is maintained or adjusted to about pH 6.7. In some embodiments, the pH is maintained or adjusted to about pH 6.8. In some embodiments, the pH is maintained or adjusted to about pH 6.9. In some embodiments, the pH is maintained or adjusted to about pH 7.0. In some embodiments, the pH is maintained or adjusted to about pH7.1. In some embodiments, the pH is maintained or adjusted to about pH 7.2. In some embodiments, the pH is maintained or adjusted to about pH 7.3. In some embodiments, the pH is maintained or adjusted to about pH 7.4. In some embodiments, the pH is maintained or adjusted to about pH 7.5. In some embodiments, the pH is maintained or adjusted to about pH 7.6. In some embodiments, the pH is maintained or adjusted to about pH 7.7. In some embodiments, the pH is maintained or adjusted to about pH 7.8. In some embodiments, the pH is maintained or adjusted to about pH 7.9. In some embodiments, the pH is maintained or adjusted to about pH 8.0.

In some embodiments, the operating parameter comprises media glucose concentration. For example, if the therapeutic protein has a desired level of saccharification and/or glycosylation, the medium glucose concentration added to the bioreactor (such as by fed-batch or perfusion culture or mixed fed-batch and perfusion glucose concentrations) may be maintained at or near the target concentration at which the saccharification and/or glycosylation level is measured.

In some embodiments, the operating parameter comprises a frequency interval of medium glucose addition. For example, if the therapeutic protein has a desired level of saccharification and/or glycosylation, the frequency interval of the bioreactor's media glucose addition (such as in fed-batch or perfusion culture or mixed fed-batch and perfusion glucose concentrations) can be maintained at or near the frequency at which the level of saccharification and/or glycosylation is measured. In some embodiments, the frequency is maintained or adjusted such that the medium addition is continuous. In some embodiments, the frequency is maintained or adjusted such that the medium additions are at separate intervals, for example six hours. In some embodiments, the frequency is maintained or adjusted such that the medium additions are at separate intervals, for example twelve hours. In some embodiments, the frequency is maintained or adjusted such that the medium addition is daily bolus, e.g., twenty-four hours. In some embodiments, the frequency is maintained or adjusted such that the medium addition is longer than every twenty-four hours.

In some embodiments, the operating parameter comprises a nutrient level. In some embodiments, the nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration. In some embodiments, the operating parameter comprises a concentration of glucose. In some embodiments, the operating parameter comprises the concentration of lactate. In some embodiments, the operating parameter comprises glutamine concentration. In some embodiments, the operating parameter comprises a concentration of ammonium ions.

In certain embodiments, the glucose concentration is maintained or adjusted to about 0.5g/L to about 40g/L, depending on the cell density. In certain embodiments, the glucose concentration is maintained or adjusted to about 0.5g/L to about 30g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 0.5g/L to about 20g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 0.5g/L to about 10g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 0.5g/L to about 5g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 5g/L to about 40g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 10g/L to about 40g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 20g/L to about 40g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 30g/L to about 40g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 35g/L to about 40g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 10g/L to about 30g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 10g/L to about 20g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 5g/L to about 10g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 10g/L to about 15g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 15g/L to about 20g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 20g/L to about 25g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 25g/L to about 30g/L. In certain embodiments, the glucose concentration is maintained or adjusted to about 30g/L to about 35g/L.

By way of example, a therapeutic protein having an acceptable level of glycation may maintain the concentration of a reducing sugar (such as glucose). Alternatively, as a further example, a therapeutic protein having an undesired amount of saccharification may have a reduced nutrient level, such as a concentration of reducing sugars, and/or the frequency of medium addition may be reduced. It is understood how many operating parameters, if any, should be adjusted will be within the skill of those in the art. Further, other alternative operating parameters may be adjusted depending on the saccharification level, and it should be understood that the above examples are intended to be exemplary only.

Similarly, therapeutic proteins having acceptable levels of specific glycan structures can maintain one or more operating parameters. Alternatively, a therapeutic protein having an undesired level of a particular glycan structure may have one or more operating parameters that are adjusted to control the particular glycan structure.

In certain aspects, one or more operating parameters are automatically modified based on the measured spectral data. For example, PAT tools (such as the PAT tool described in section 5.4.1) may be used to measure saccharification and/or glycosylation, and one or more operating parameters may be automatically maintained in the bioreactor if the level of desired glycan structures is above a predetermined threshold, or the level of undesired glycan structures is below a predetermined threshold. Alternatively, one or more operating parameters may be automatically modified in the bioreactor if the level of the desired glycan structure is below a predetermined threshold or the level of the undesired glycan structure is above a predetermined threshold.

The threshold for acceptable glycation levels on therapeutic proteins will generally be determined empirically. In some embodiments, the predetermined threshold for glycation will be less than 50% of the therapeutic protein in the bioreactor. In some embodiments, the predetermined threshold for saccharification will be less than 40%. In some embodiments, the predetermined threshold for saccharification will be less than 30%. In some embodiments, the predetermined threshold for saccharification will be less than 25%. In some embodiments, the predetermined threshold for saccharification will be less than 20%. In some embodiments, the predetermined threshold for saccharification will be less than 15%. In some embodiments, the predetermined threshold for saccharification will be less than 10%.

Thus, in some embodiments, at least 90% of the therapeutic proteins in the bioreactor are non-glycosylated. In some embodiments, at least 85% of the therapeutic protein in the bioreactor will be non-glycosylated. In some embodiments, at least 80% of the therapeutic protein in the bioreactor will be non-glycosylated. In some embodiments, at least 75% of the therapeutic protein in the bioreactor will be non-glycosylated. In some embodiments, at least 70% of the therapeutic protein in the bioreactor will be non-glycosylated. In some embodiments, at least 60% of the therapeutic protein in the bioreactor will be non-glycosylated. In some embodiments, at least 50% of the therapeutic protein in the bioreactor will be non-glycosylated.

The threshold for glycation level on the therapeutic protein will also typically be determined empirically. For example, a degree of variability in the batch-to-batch glycoform content is expected in the production environment, and in general, the threshold will be an acceptable limit that does not produce too much variability in the glycoform content. Thus, in some embodiments, the predetermined threshold for undesired glycan structures will be less than 50% of the therapeutic protein in the bioreactor. In some embodiments, the predetermined threshold for undesired glycan structures will be less than 40%. In some embodiments, the predetermined threshold for undesired glycan structures will be less than 30%. In some embodiments, the predetermined threshold for undesired glycan structures will be less than 25%. In some embodiments, the predetermined threshold for undesired glycan structures will be less than 20%. In some embodiments, the predetermined threshold for undesired glycan structures will be less than 15%. In some embodiments, the predetermined threshold for undesired glycan structures will be less than 10%.

5.5.4.Purification。

Generally, therapeutic proteins (such as those disclosed in section 5.1) are secreted from cells that are grown during the production process. After the production process is completed, the therapeutic protein may be isolated and purified from the cells using any technique known in the art suitable for protein purification, including purification according to FDA standards. Preferably, purification will remove process impurities such as host cell proteins, nucleic acids and/or lipids.

The purification process may involve one or more steps. For example, purification of the therapeutic protein may involve preliminary recovery, purification, and/or refining steps. Generally, the preliminary recovery step consists of: centrifugation and/or depth filtration to remove cells and cell debris from the culture broth, and to clarify the cell culture supernatant containing the therapeutic protein product. Additional techniques known in the art may be additionally included to improve the primary recovery process. For example, the use of flocculants such as simple acids, divalent cations, polycationic polymers, octanoic acids, and stimuli responsive polymers can enhance cell culture clarification and reduce the level of cells, cell debris, DNA, host Cell Proteins (HCPs), and/or viruses while retaining therapeutic proteins in the product stream.

Purification may also involve one or more chromatographic techniques (e.g., affinity, ion exchange, hydrophobic interactions), lectin-based purification, borate-based purification, and/or filtration techniques (e.g., ultrafiltration) that are used as a capture step in separating the product from smaller impurities and/or as a concentration step in reducing the total volume.

Optionally, a refining step may also be performed to remove viruses, aggregated proteins and any other impurities, for example, prior to storage of the therapeutic proteins. The refining step may include, for example, viral filtration, hydrophobic interaction chromatography, and/or filtration steps (e.g., ultrafiltration/diafiltration and/or aseptic filtration).

Thus, in some embodiments, the methods and systems provided herein relate to purifying therapeutic proteins.

5.6.Methods and systems for determining glycosylation and/or saccharification。

Referring to graph 600 of fig. 6, the present subject matter may utilize a Process Analysis Technology (PAT) tool 610, which may have probes 612 inside or extending into a bioreactor 620 to monitor or otherwise characterize aspects of reactions occurring within the bioreactor 620. The PAT tool 610 includes one or more data processors and memory into which the data processors may load instructions and execute the instructions. The PAT tool 610 may take various forms and, in some cases, utilize or otherwise include raman spectroscopy to characterize aspects of the reactions occurring within the bioreactor 620. In some cases, PAT tool 610 communicates over network 630 with one or more computing systems 640 (e.g., servers, personal computers, tablet computers, ioT devices, mobile phones, dedicated control units, etc.) that may execute various algorithms as described below. Computing system 640 may also be used to vary one or more operating parameters associated with bioreactor 620. In some cases, bioreactor 620 may also have network connectivity such that it may communicate with one or more remote computing systems 640, which in turn may cause one or more operating parameters of bioreactor 620 to change.

The PAT tool 610 as described above may be used to control and monitor a production process (such as in the bioreactor 620) in-line or online in real time. The PAT tool 610 may utilize spectroscopic techniques such as near infrared spectroscopy, fluorescence spectroscopy, and raman spectroscopy. Raman spectroscopy is itself a particularly suitable technique for use in bioreactor production processes because it provides a clear, sharp spectrum without some of the disadvantages of other techniques, such as water interference in the near infrared spectrum and other interference that may make the spectrum less sharp. The PAT tool 610 may include a laser that performs raman spectroscopy, a vibrational spectroscopy technique that provides chemical fingerprints of substances. The PAT tool 610 in the present context is technically advantageous because it allows non-destructive real-time measurements of many biotherapeutic process variables (including metabolites, growth profiles, product levels, product quality attributes, nutrient supplements, culture pH, etc.).

The computing system 640 may combine raman spectroscopy analysis from the PAT tool 610 with chemometric modeling to provide real-time monitoring of these variables. The spectral peaks obtained by raman spectroscopy are correlated with one or more process variables of interest (which may be predetermined and/or measured by offline analysis) using chemometric modeling software. Various signal processing techniques may be applied to identify and quantify the spectral peaks generated by the PAT tool 610. One type of signal processing technique is a Partial Least Squares (PLS) regression model, which can be used to model raman data given the linear nature of raman signals with respect to analyte concentration. In some embodiments, a linear PLS model may be employed. In certain embodiments, a non-linear PLS model may be employed.

The PAT tool 610 may be used to measure or otherwise characterize aspects related to large therapeutic proteins (such as mAb proteins) in the bioreactor 620. In particular, in some variations, PAT tool 610 may be used to provide real-time monitoring of the glycation and/or glycosylation profile of therapeutic proteins in a manufacturing-scale bioreactor. Manufacturing scale presents a number of technical difficulties compared to smaller scale laboratory bioreactors because of the need to address complex environmental conditions. The current subject matter addresses these technical difficulties and the complex nature of glycoprotein formation by utilizing PAT tool 610 and in some cases using raman-based PLS models.

Aspects of the PAT tool 610, bioreactor 620, and/or computing system 640 herein may be implemented in digital electronic circuitry, integrated circuit systems, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various implementations may include implementations in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

These computer programs (also known as programs, software applications or code) include machine instructions for a programmable processor, and may be implemented in a high-level procedural and/or object-oriented programming language, and/or in assembly/machine language. As used herein, the term "machine-readable medium" refers to any computer program product, apparatus and/or device (e.g., magnetic discs, optical disks, solid state drives, memory, programmable Logic Devices (PLDs)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.

Computing system 640 may include a back-end component (e.g., as a data server), a middleware component (e.g., an application server), or a front-end component (e.g., a client computer having a graphical user interface or a Web browser through which a user can interact with a particular implementation of the subject matter described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), and the Internet. Computing system 640 may include clients and servers. The client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

FIG. 7 is a process flow diagram 700 illustrating the determination of glycan structures on a glycosylated molecule, wherein at 710, for each of a plurality of runs, the level of one or more glycan structures on the glycosylated molecule is obtained using a Process Analysis Technology (PAT) tool that obtains or otherwise generates spectral data. These levels are obtained based on processes occurring within one or more bioreactors having volumes equal to or less than a first threshold. Thereafter, at 720, one or more regression models are generated based on the obtained spectral data, the regression models correlating the levels of the one or more glycan structures on the glycosylated molecule with the obtained spectral data. Subsequently, at 730, the one or more glycan structures on the glycosylated molecule are measured using PAT tools within one or more second bioreactors having volumes equal to or greater than a second threshold to produce measured spectral data. Then, at 740, the at least one computing device determines a level of one or more glycan structures on the glycosylated molecules within the one or more bioreactors using the generated one or more regression models and based on the measured spectral data.

One or more regression models may be refined based on a combination of the obtained spectral data and the measured spectral data.

One or more operating parameters of the one or more second bioreactors may be maintained and/or selectively modified (i.e., altered) based on the determined levels to produce the desired glycosylated molecules. The one or more operating parameters include pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof. The nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration. The concentration of glucose in the bioreactor may be automatically modified based on the measured spectral data.

In some variations, glycosylated molecules may be purified.

The glycan structure can take a variety of forms including G0F-GlcNac, G0F, G F, and G2F, or combinations thereof.

In some variations, two or more bioreactors with different volumes may be used for measurement.

The first threshold may have a different volume, including about 25 liters or less.

The second threshold may correspond to a different volume, including about 1,000 liters or greater, including 2,000 liters, 10,000 liters to about 25,000 liters, and about 15,000 liters.

The second threshold may be at least 5 times, and in some cases at least 10 times, and in some cases at least 100 times, and in other cases at least 500 times the first threshold.

PAT tools may utilize spectroscopic techniques, including raman spectroscopy.

The one or more regression models may include or otherwise use a Partial Least Squares (PLS) model.

The glycosylated molecule may comprise a monoclonal antibody (mAb). Alternatively, the glycosylated molecule comprises a non-mAb.

The determination may be made on-site and/or off-site. Further, the measurement and/or determination may be made in-line, online, offline, or a combination thereof.

Fig. 8 is a process flow diagram 800 illustrating the production of glycosylated molecules with a desired glycan structure. At 810, one or more glycan structures are measured using a Process Analysis Technology (PAT) tool to generate spectral data. This measurement is carried out in a bioreactor having a volume equal to or greater than 1,000 liters. Subsequently, at 820, the level of the one or more glycan structures within the bioreactor is determined by the at least one computing device using the one or more regression models and based on the measured spectral data. One or more regression models are generated using test runs from at least one bioreactor having a volume of less than or equal to 50 liters and at least one bioreactor having a volume of equal to or greater than 1,000 liters. At 830, one or more operating parameters of the bioreactor are maintained when the level of the desired glycan structure is above a predetermined threshold or the level of the undesired glycan structure is below a predetermined threshold. Further, at 840, one or more operating parameters of the bioreactor are modified when the level of the desired glycan structure is below a predetermined threshold or the level of the undesired glycan structure is above a predetermined threshold.

FIG. 9 is a process flow diagram 900 for determining saccharification on a molecule. At 910, for each of a plurality of runs, a level of saccharification on the molecule is obtained using a Process Analysis Technology (PAT) tool. The obtaining may be performed in one or more bioreactors having a volume equal to or less than a first threshold. The PAT tool may obtain or otherwise generate spectral data. At 920, one or more regression models are generated based on the obtained spectral data, the regression models correlating the level of saccharification on the molecule with the obtained spectral data. Saccharification on the molecule may then be measured using PAT tools at 930. The measurement may be performed within one or more second bioreactors having volumes equal to or greater than a second threshold to produce measured spectral data. Subsequently, at 940, a saccharification level on the molecules within the one or more bioreactors is determined by the at least one computing device using the generated one or more regression models and based on the measured spectral data.

One or more operating parameters of the one or more second bioreactors may be maintained and/or selectively modified (i.e., changed) based on the determined levels to produce a level of saccharification on the molecule. The one or more operating parameters include pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof. The nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration. The concentration of glucose in the bioreactor may be automatically modified based on the measured spectral data.

PAT tools may utilize spectroscopic techniques, including raman spectroscopy.

The molecule may comprise or be a mAb. Alternatively, the molecule comprises or is a non-mAb.

Fig. 10 is a process flow diagram 1000 illustrating the production of molecules having a desired level of saccharification. At 1010, saccharification on the molecule is measured using a Process Analysis Technology (PAT) tool to generate spectral data. The measurement can be performed in a bioreactor having a volume equal to or greater than 1,000 liters. Subsequently, at 1020, a level of saccharification on the molecules within the bioreactor is determined by the at least one computing device using the one or more regression models and based on the measured spectral data. The one or more regression models may be generated using test runs from at least one bioreactor having a volume of less than or equal to 50 liters and at least one bioreactor having a volume of equal to or greater than 1,000 liters. At 1030, one or more operating parameters of the bioreactor are maintained when the saccharification level on the molecule is below a predetermined threshold. Additionally, at 1040, one or more operating parameters of the bioreactor are selectively modified when the saccharification level on the molecule is above a predetermined threshold.

6.Description of the embodiments

A1. A method for determining glycan structures on a glycosylated molecule, the method comprising:

obtaining, for each of a plurality of runs, a level of one or more glycan structures on the glycosylated molecule using a Process Analysis Technique (PAT) tool, wherein the obtaining is performed within one or more first bioreactors having a first volume equal to or less than a first threshold, the PAT tool obtaining spectral data;

generating one or more regression models based on the obtained spectral data, the regression models correlating levels of the one or more glycan structures on the glycosylated molecule with the obtained spectral data;

measuring the one or more glycan structures on the glycosylated molecule using the PAT tool, wherein the measuring is performed within one or more second bioreactors having a second volume equal to or greater than a second threshold to produce measured spectral data; and

determining, by at least one computing device, a level of the one or more glycan structures on the glycosylated molecules within one or more second bioreactors using the generated one or more regression models and based on the measured spectral data.

A2. The method of embodiment A1, further comprising refining the one or more regression models based on a combination of the obtained spectral data and the measured spectral data.

A3. The method of embodiment A1 or embodiment A2, further comprising maintaining one or more operating parameters of the one or more second bioreactors based on the determined levels to produce a desired glycosylated molecule.

A4. The method of embodiment A1 or embodiment A2, further comprising selectively modifying one or more operating parameters of the one or more second bioreactors based on the determined levels to produce a desired glycosylated molecule.

A5. The method of embodiment A4, wherein the one or more operating parameters comprise pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof.

A6. The method of embodiment A5, wherein the nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration.

A7. The method of embodiment A6, wherein the concentration of glucose is automatically modified based on the measured spectral data.

A8. The method of any one of embodiments A1-A7, further comprising purifying the glycosylated molecule.

A9. The method of any one of embodiments A1-A8, wherein the glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

A10. The method of any one of embodiments A1-A9, wherein the obtaining is performed in two or more bioreactors having different volumes.

A11. The method of any one of embodiments A1-a 10, wherein the first threshold is about 250 liters or less.

A12. The method of any one of embodiments A1-a 10, wherein the first threshold is about 100 liters or less.

A13. The method of any one of embodiments A1-a 10, wherein the first threshold is about 50 liters or less.

A14. The method of any one of embodiments A1-a 10, wherein the first threshold is about 25 liters or less.

A15. The method of any one of embodiments A1-a 10, wherein the first threshold is about 10 liters or less.

A16. The method of any one of embodiments A1-a 10, wherein the first threshold is about 5 liters or less.

A17. The method of any one of embodiments A1-a 10, wherein the first threshold is about 2 liters or less.

A18. The method of any one of embodiments A1-a 10, wherein the first threshold is about 1 liter or less.

A19. The method of any one of embodiments A1-a 18, wherein the second threshold is about 1,000 liters or greater.

A20. The method of any one of embodiments A1-a 18, wherein the second threshold is about 2,000 liters or greater.

A21. The method of any one of embodiments A1-a 18, wherein the second threshold is about 5,000 liters or greater.

A22. The method of any one of embodiments A1-a 18, wherein the second threshold is about 10,000 liters to about 25,000 liters.

A23. The method of any one of embodiments A1-a 18, wherein the second threshold is about 15,000 liters.

A24. The method of any one of embodiments A1-a 18, wherein the second threshold is at least 5 times the first threshold.

A25. The method of any one of embodiments A1-a 18, wherein the second threshold is at least 10 times the first threshold.

A26. The method of any one of embodiments A1-a 18, wherein the second threshold is at least 100 times the first threshold.

A27. The method of any one of embodiments A1-a 18, wherein the second threshold is at least 500 times the first threshold.

A28. The method of any one of embodiments A1-a 27, wherein the first volume is from about 0.5 liters to about 250 liters.

A29. The method of any one of embodiments A1-a 27, wherein the first volume is from about 1 liter to about 50 liters.

A30. The method of any one of embodiments A1-a 27, wherein the first volume is from about 1 liter to about 25 liters.

A31. The method of any one of embodiments A1-a 27, wherein the first volume is from about 1 liter to about 10 liters.

A32. The method of any one of embodiments A1-a 27, wherein the first volume is from about 1 liter to about 5 liters.

A33. The method of any one of embodiments A1-a 32, wherein the second volume is about 1,000 liters to about 25,000 liters.

A34. The method of any one of embodiments A1-a 32, wherein the second volume is about 2,000 liters to about 25,000 liters.

A35. The method of any one of embodiments A1-a 32, wherein the second volume is about 5,000 liters to about 25,000 liters.

A36. The method of any one of embodiments A1-a 32, wherein the second volume is about 10,000 liters to about 25,000 liters.

A37. The method of any one of embodiments A1-a 32, wherein the second volume is about 15,000 liters to about 25,000 liters.

A38. The method of any one of embodiments A1-a 37, wherein the PAT tool utilizes or otherwise comprises raman spectroscopy.

A39. The method of any one of embodiments A1-a 38, wherein the one or more regression models comprise a Partial Least Squares (PLS) model.

A40. The method of any one of embodiments A1-a 39, wherein the glycosylation molecule comprises a monoclonal antibody (mAb).

A41. The method of any one of embodiments A1-a 40, wherein the glycosylation molecule comprises a non-mAb.

A42. The method of any one of embodiments A1-a 41, wherein the determining step is performed in situ.

A43. The method of any one of embodiments A1-a 41, wherein the determining step is performed off-site.

A44. The method of any one of embodiments A1-a 43, wherein the determining step is performed in-line, online, offline, or a combination thereof.

A45. The method of any one of embodiments A1-a 44, wherein the determining step is performed in-line.

A46. The method of any one of embodiments A1-a 44, wherein the determining step is performed online.

A47. The method of any one of embodiments A1-a 44, wherein the determining step is performed offline.

A48. The method of any one of embodiments A1-a 44, wherein the determining step is performed in-line.

A49. The method according to any one of the preceding embodiments, wherein the obtaining comprises: data characterizing the spectral data is received from the PAT tool.

A50. The method of any of the preceding embodiments, wherein the generating is performed by one or more computing devices.

B1. A method of producing a glycosylated molecule having a desired glycan structure, the method comprising

Measuring one or more glycan structures using a Process Analysis Technology (PAT) tool to produce spectral data, wherein the measurement is performed in a bioreactor having a volume equal to or greater than 1,000 liters;

determining, by at least one computing device, a level of the one or more glycan structures within the bioreactor using one or more regression models and based on the measured spectral data, wherein the one or more regression models are generated using test runs from at least one bioreactor having a volume of less than or equal to 50 liters and at least one bioreactor having a volume of equal to or greater than 1,000 liters; and

Maintaining one or more operating parameters of the bioreactor when:

the level of desired glycan structures is above a predetermined threshold, or

The level of undesired glycan structures is below a predetermined threshold; and is also provided with

Selectively modifying one or more operating parameters of the bioreactor when:

the level of desired glycan structures is below a predetermined threshold, or

The level of undesired glycan structures is above a predetermined threshold.

B2. The method of embodiment B1, wherein the measuring is performed in-line, online, offline, or a combination thereof.

B3. The method of embodiment B1 or embodiment B2, wherein the measuring is performed in-line.

B4. The method of embodiment B1 or embodiment B2, wherein the measurement is performed on-line.

B5. The method of embodiment B1 or embodiment B2, wherein the measuring is performed off-line.

B6. The method of embodiment B1 or embodiment B2, wherein the measuring is performed on a line.

B7. The method of any one of embodiments B1-B6, wherein the measuring is performed more than once per day.

B8. The method of any one of embodiments B1-B6, wherein measuring is performed about every 5 minutes to 60 minutes.

B9. The method of any one of embodiments B1-B6, wherein the measuring is performed about every 10 minutes to 30 minutes.

B10. The method of any one of embodiments B1-B6, wherein measuring is performed about every 10 minutes to 20 minutes.

B11. The method of any one of embodiments B1-B6, wherein the measuring is performed about every 12.5 minutes.

B12. The method of any one of embodiments B1-B11, wherein the bioreactor volume is about 2,000 liters or more.

B13. The method of any one of embodiments B1-B11, wherein the bioreactor volume is about 5,000 liters or greater.

B14. The method of any one of embodiments B1-B11, wherein the bioreactor volume is about 10,000 liters or greater.

B15. The method of any one of embodiments B1-B11, wherein the bioreactor volume is about 15,000 liters or more.

B16. The method of any one of embodiments B1-B11, wherein the bioreactor volume is from about 10,000 liters to about 25,000 liters.

B17. The method of any one of embodiments B1-B11, wherein the bioreactor volume is about 15,000 liters.

B18. The method of any one of embodiments B1-B17, wherein the determining step is performed in situ.

B19. The method of any one of embodiments B1-B17, wherein the determining step is performed off-site.

B20. The method of any one of embodiments B1 to B19, wherein the desired glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

B21. The method of any of embodiments B1-B20, wherein the bioreactor is a batch reactor, a fed-batch reactor, a perfusion reactor, or a combination thereof.

B22. The method of any one of embodiments B1-B21, wherein the one or more operating parameters comprise pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof.

B23. The method of embodiment B22, wherein the nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration.

B24. The method of embodiment B23, wherein the concentration of glucose is automatically modified based on the measured spectral data.

B25. The method of any one of embodiments B1-B24, wherein the PAT tool utilizes or otherwise comprises raman spectroscopy.

B26. The method of any one of embodiments B1-B25, wherein the one or more regression models comprise a Partial Least Squares (PLS) model.

C1. A system for producing one or more glycosylated molecules, the system comprising:

means for culturing a cell line producing the glycosylated molecule;

means for measuring the level of one or more glycan structures, wherein the means generates spectral data;

means for generating one or more regression models based on the spectral data; and

a device for measuring the level of one or more glycan structures in a cell line producing a glycosylated molecule.

C2. The system of embodiment C1, wherein the cell line that produces the glycosylated molecule is a mammalian cell line.

C3. The system of embodiment C2, wherein the mammalian cell line is a non-human cell line.

C4. The system of any one of embodiments C1-C3, wherein the culturing comprises batch, fed-batch, perfusion, or a combination thereof.

C5. The system of any one of embodiments C1-C4, wherein culturing comprises a volume of about 2,000 liters or more.

C6. The system of any one of embodiments C1-C4, wherein culturing comprises a volume of about 5,000 liters or more.

C7. The system of any one of embodiments C1-C4, wherein culturing comprises a volume of about 10,000 liters or more.

C8. The system of any one of embodiments C1-C4, wherein culturing comprises a volume of about 15,000 liters or more.

C9. The system of any one of embodiments C1-C4, wherein culturing comprises a volume of about 10,000 liters to about 25,000 liters.

C10. The system of any one of embodiments C1-C4, wherein culturing comprises a volume of about 15,000 liters.

C11. The system of any one of embodiments C1-C4, wherein the measurement is performed in-line, online, offline, or a combination thereof.

C12. The system of any one of embodiments C1-C11, wherein measuring is performed in-line.

C13. The system of any one of embodiments C1-C11, wherein measuring is performed on a line.

C14. The system of any one of embodiments C1-C11, wherein the measurement is performed online.

C15. The system of any one of embodiments C1-C11, wherein the measurement is performed offline.

C16. The system of any one of embodiments C1-C15, wherein the measuring is performed more than once per day.

C17. The system of any one of embodiments C1-C15, wherein the measurement is performed about every 5 minutes to 60 minutes.

C18. The system of any one of embodiments C1-C15, wherein the measurement is performed about every 10 minutes to 30 minutes.

C19. The system of any one of embodiments C1-C15, wherein the measurement is performed about every 10 minutes to 20 minutes.

C20. The system of any one of embodiments C1-C15, wherein the measurement is performed about every 12.5 minutes.

C21. The system of any one of embodiments C1-C20, wherein the one or more glycan structures are selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

C22. The system of any one of embodiments C1-C21, further comprising means for selectively modifying one or more operating parameters to enhance production of the desired one or more glycan structures.

C23. The system of embodiment C22, wherein the one or more operating parameters comprise pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof.

C24. The system of embodiment C23, wherein the nutrient level is selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration.

C25. The system of embodiment C24, wherein the concentration of glucose is automatically modified based on the spectral data.

C26. The system of any one of embodiments C1-C25, wherein the one or more glycosylation molecules comprise a monoclonal antibody (mAb).

C27. The system of any one of embodiments C1-C25, wherein the one or more glycosylated molecules comprise a non-mAb.

C28. The system of any one of embodiments C1-C27, the spectral data comprising raman spectra.

C29. The system of any one of embodiments C1-C28, wherein the one or more regression models comprise a Partial Least Squares (PLS) model.

C30. The system of any one of embodiments C1-C29, further comprising a device for isolating the one or more glycosylated molecules.

D1. A system for producing one or more glycosylated molecules:

a bioreactor comprising a cell line producing glycosylated molecules;

A Process Analysis Technology (PAT) tool that measures one or more glycan structures and generates spectral data; and

a processor that correlates the level of one or more glycan structures with the spectral data using one or more regression models.

D2. The system of embodiment D1, wherein the bioreactor is about 2,000 liters or more.

D3. The system of embodiment D1, wherein the bioreactor is about 5,000 liters or greater.

D4. The system of embodiment D1, wherein the bioreactor is about 10,000 liters or more.

D5. The system of embodiment D1, wherein the bioreactor is about 15,000 liters or more.

D6. The system of embodiment D1, wherein the bioreactor is about 10,000 liters to about 25,000 liters.

D7. The system of embodiment D1, wherein the bioreactor is about 15,000 liters.

D8. The system of any one of embodiments D1-D7, wherein the glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

D9. The system of any one of embodiments D1-D8, wherein the cell line that produces the glycosylated molecule is a mammalian cell line.

D10. The system of embodiment D9, wherein the mammalian cell line is a non-human cell line.

D11. The system of any of embodiments D1-D10, wherein the PAT tool utilizes or otherwise comprises raman spectroscopy.

D12. The system of any one of embodiments D1-D11, wherein the one or more regression models comprise a Partial Least Squares (PLS) model.

7.Examples

The following examples of the present patent application will further illustrate the nature of the present patent application. It will be appreciated by those skilled in the art that changes could be made to the embodiments described above without departing from the broad inventive concept thereof. It is understood, therefore, that this invention is not limited to the particular embodiments disclosed, but it is intended to cover modifications within the spirit and scope of the present invention as defined by the present specification.

Example 1: experimental design for monitoring saccharification and glycosylation in real time

The aim of this study was to develop a PLS model of raman spectroscopy for monitoring saccharification and glycosylation (both CQAs) in real time in representative CHO cell cultures at production scale.

Model development of reduced-scale data is initially evaluated. Robustness of the developed model is then taken into account by supplementing the scaled-down data with the manufacturing-scale data. Product quality was considered throughout the manufacture of the biotherapeutic mAb products, which was achieved by monitoring CQA in real-time throughout the upstream production process using raman spectroscopy.

Eighteen batches of mAb-producing CHO cell lines were used in the experimental design. The stirred bioreactor scale is disposable bag (SUB) 2000L (Thermo Fisher Scientific, waltham, MA), glass 5L (Applikon inc., schiedam, netherlands) and glass 1L (Eppendorf, hamburg, germany) and the fed-batch process lasts for 16-18 days. A total of nine 1L batches, seven 5L batches and two 2000L batches were included. Each 2000L batch was inoculated using commercial seed expansion (seed train) on day 0, ten downscale bioreactors were inoculated using day 1 cell cultures from the 2000L batch, and six downscale bioreactors were inoculated using laboratory seed expansion on day 0. Basal medium and daily feed medium were used for each batch performed, and process control was performed with all batches.

Two feeding strategies were employed during the execution of the bioreactor run, each consisting of two composite feeds, feeding each bioreactor starting on day 3. Twelve batches (2×2000L, 7×5L, 3×1L) were fed two composite feeds per day based on a defined percentage of the vessel volume. The first composite feed was fed to another 6 batches (6 x 1L) based on a defined percentage of the vessel, and the second composite feed was delivered multiple times during the day to maintain a predetermined target level of glucose in the bioreactor as part of the feed strategy study. All bioreactors were inoculated within the same inoculation density limit.

Each batch had the same Dissolved Oxygen (DO), pH and temperature control targets. DO was controlled at 40% by aeration and oxygen injection. The pH target of 6.95 was maintained using the addition of carbon dioxide and 2.0M sodium carbonate. The temperature was controlled at a set point of 36.5 ℃ (35.5 ℃ -37.5 ℃) throughout the cell culture process. The scale-related process parameters were stirred across scales using per-volume power calculations. Offline samples were collected from each bioreactor daily and a set of metabolites including glucose, lactate, titer, viable CELL density and% viability were measured using a Vi-CELL MetaFLEX (Beckman Coulter, brea, calif.) Vi-CELL XR CELL viability analyzer (Beckman Coulter) and a Cedex Bio (Roche Holding AG, switzerland) offline analyzer. After each batch was completed, each daily culture sample was retained and frozen at-70 ℃ for further testing.

Saccharification and glycosylation analysis

The saccharification and glycosylation of the mAb was characterized off-line by LC/MS analysis. Briefly, protein samples for saccharification analysis were first pre-treated with EndoS enzyme to remove N-linked carbohydrates in order to eliminate glycoform-based sample heterogeneity. The protein samples were then separated by High Performance Liquid Chromatography (HPLC) and analyzed by online electrospray ionization quadrupole time-of-flight mass spectrometry. Mass/charge data collected across chromatographic peaks were then summed and deconvolved using Empower software (Waters Corp, milford, MA). The detected glycoisoforms are assigned based on deconvolution mass spectrometry analysis and their relative abundance is calculated using the peak intensities of the center deconvolution mass spectrum. Protein samples for glycosylation were first pre-treated with 1M dithiothreitol to separate the individual mAb heavy and light chains. Protein samples were then analyzed by LC/MS as before and the glycosylated isoforms were assigned and quantified as before.

Raman spectrum acquisition during bioreactor batches

Raman spectra were collected for all batches using two instruments, each using a multichannel raman RXN2 system (Kaiser Optical Systems inc., ann Arbor, MI) containing a 785-nm laser source and a Charge Coupled Device (CCD) at-40 ℃. The detector is connected to an MR probe (Kaiser Optical Systems inc.) consisting of a fiber optic excitation cable and a fiber optic collection cable. Data were collected by an MR probe attached to a bIO-Optic-220 stainless steel probe (Kaiser Optical Systems inc.) inserted into a sterile bioreactor. The ramanan software 4.1 (Mettler Toledo Autochem, columbia, MD) was used to control the ramanaxn 2 for spectral acquisition for all scaled-down batches (1L, 5L). Raman Runtime HMI (Kaiser Optical Systems inc.) was used for spectral acquisition in all 2000L batches. All raman spectral data were collected using a system setup of 10 second exposure for 75 scans, which produced the spectrum of the probe after 15 minutes (including an overhead time of 2.5 minutes). The Raman spectrum acquisition range is 100-3,425cm-1 of wave number. The reduced-scale containers are protected from light by aluminum foil. (since the disposable bioreactor is provided in the manufacturing kit, this is not required as it is enclosed in stainless steel). The intensity calibration of the instrument was performed with Hololab Calibration Accessory (HCA) (Kaiser Optical Systems inc.) and the internal calibration was set to be performed every 24 hours throughout the bioreactor process, before each use of the system.

Example 2: development of PLS model based on Raman Spectroscopy for reduced-scale saccharification and glycosylation (scheme 1)

This example demonstrates that a raman spectrum based PLS model can be developed for saccharification and glycosylation spectra of exemplary therapeutic protein producing cell lines during production in a bioreactor.

In this example, raman and off-line data from a scaled down (1L and 5L) cell culture process were used to develop a set of 7 stoichiometric PLS models for the glycation and glycosylation profile of mabs. Two models were considered for saccharification (mono-saccharification, non-saccharification) and 5 models were considered for glycosylation profile (G0F-GlcNac, G0F, G1F, G F). Chemometric modeling was performed with Simca 15.1 (immerics inc., san Jose, CA).

Offline measurements of saccharification and glycosylation were aligned with raman spectra based on the time they were collected, starting on day 05 for each batch. The decision to model from this point in time was based on empirical knowledge of the process and its production of mAb products by HPLC at detectable levels. All data, raman spectra and off-line measurements prior to day 05 were excluded from model building and testing.

Each model consisted of 15 batches (9×1L and 6×5L) of Calibration Sample Sets (CSSs) (for model development) and 1 batch (5L) of Calibration Test Sample Sets (CTSS) (used as blind data sets for testing with PLS models generated for each CQA). The batch was randomly selected as CTSS from the available 5L batch data. The X-variable for each model in the flow is raman spectrum (centered) and the Y-variable is the off-line value of: % mono-saccharification, non-saccharification, G0F-GlcNac, G0F, G F and G2F (univariate reduction). The wavenumbers of the Raman spectra of all models were chosen to be 415-1800cm-1 and 2800-3100cm-1. The spectral filters applied to all PLS models were Savitsky-Golay first derivative second (31 cm-1 point) and standard normal variables (SNV; data not shown).

Each PLS model was built and errors were assessed by using a leave-in cross-validation method (leaving each batch in the model development once). Model errors were averaged based on model-to-omitted lot predictions to identify cross-validated root mean square (RMSEcv). RMSEcv indicates the predictive capability of the model based on the data used to build the model. A lower mean error (RMSEcv) indicates that the model is improved. This enables better informed decisions as to which component amounts are to be used for the model generated for testing for the blind data set. The predictive Root Mean Square Error (RMSEP) of the predictions was identified with respect to the predictive ability of CTSS using the predictive function test model in Simca 15.1, which indicates the predictive ability of the model with respect to the unseen dataset. Regression (R2) values, coefficients of variation, were recorded for each PLS model. The R2 value is used to determine the amount of change in the Y variable that the model predictor (X variable) can interpret. The closer the R2 value is to 1, the greater the model interpretation Y variable. Model performance is assessed based on the respective RMSEcv, RMSEP, and R2 values for each model. The variable projection importance (VIP) of each model is also considered, which is a parameter summarizing the importance of the X variable in the model in predicting the Y variable. An X variable having a value greater than 1 is considered to be most relevant for interpreting a Y variable.

Flow 1 evaluates a calibration dataset built using only scaled-down data as PLS model. Each of the 7 models developed (2 for saccharification (mono-saccharification, non-saccharification) and 5 glycosylation spectra (G0F-GlcNac, G0F, G1F, G F)) were tested by predicting the blind dataset of the scaled-down batch data. Each model in scheme 1 consisted of a 15 lot downscaling process in CSS, where one complete lot (5L) was used for testing as a blind CTSS.

Leave-in cross-validation is used to determine the optimal number of components per model. This is determined as the lowest number of components with the lowest matching RMSEcv value. The optimal component amounts for each model were evaluated individually. The average R2 results from cross-validation show that there is a large degree of variability for each model, which can be explained by the reported R2 values >0.8 for each of the models considered (table 1). The acceptance criteria for the accuracy of each model are determined based on the range of measurements observed in the model CSS. The accuracy acceptance criterion was set to 10% of the range of values used in the calibration data set according to the chemometric model development criterion of JSI. This is determined to be a sufficient acceptance criterion based on the potential error sources for each model, including acceptable errors in the method employed in the offline analysis. The error in each model was evaluated by comparing the root mean square error (RMSEcv, RMSEP) with the calculated acceptance criteria (table 1). RMSEcv informs of the optimum component amounts. The value also indicates the ability of the model to predict a variable of interest based on the calibration data set. The number of components selected for each model gives RMSEcv that falls within acceptable criteria for prediction error and is therefore considered suitable for model development.

Table I: scheme 1: model development statistics for raman PLS models for saccharification and glycosylation

Complementary results were observed when the predictive accuracy of each model was tested with the blind dataset (5L batch) (table 1). R of mono-glycosylated, non-glycosylated, G0F, G1F, G2F, G0 ² >0.85, and G0F-GlcNac<0.7. When testing against a blind data set, all R' s ² The values all showed a decrease, however, exceptExcept for G0, mono-saccharification and non-saccharification, RMSEP values for all models showed lower errors in prediction than those observed for RMSEcv. R is R ² >0.9 does not necessarily indicate that the model is better and enhances binding of R during model evaluation ² The values take into account the needs of a variety of factors including RMSEcv and RMSEP.

Although testing against scaled-down blind data sets showed R ² The model errors decreased and observed for the G0, mono-saccharification and non-saccharification models increased slightly, but it should be noted that these models still fall within accepted standards and thus can be considered suitable for predicting reduced scale glycosylation. The observed trends for the predictive tests (fig. 3A-3G) confirm the suitability of the model used, with the mono-and non-saccharification trends closely following the off-line data trends and complementing each other over the duration of the process as expected. G0F-GlcNac, G0F, G F and G2F similarly follow an offline trend and are within acceptable error criteria, although slight deviations are observed in each.

The raman spectral regions identified in each model with VIP scores >1 indicate an acceptable degree of model specificity (fig. 11A-11F). The saccharification model (mono-saccharification, non-saccharification) indicates a similar raman spectral region correlation to that previously identified for the glucose PLS model, whereas a degree of secondary correlation may be due to the structure of the glycated protein, avoiding the undesired correlation of these models with glucose by ensuring that the data used in the model CSS is obtained at intervals both before and after glucose feeding. When executing a bioreactor batch for data collection, different glucose feeding strategies are also employed in order to break the undesired glucose correlation. Glycosylation pattern models (G0F-GlcNac, G0F, G F, and G2F) showed VIP scores >1 in regions previously associated with glycan components (e.g., mannose, fucose, n-acetylglucosamine). This suggests that the model identifies peaks in the raman spectrum associated with each glycan.

Initial evaluation of the internal relationship graph for each model shows that linearity reaches a satisfactory level as model evaluation proceeds, however, in future work, nonlinear PLS methods may improve model accuracy taking into account secondary correlations with product titer. The model performance in each case was acceptable and closely matched with the daily offline samples with expected profiles supported the model development decisions made in the flow.

The ability to monitor CQA in real time during the bioreactor process, such as saccharification and glycosylation, is consistent with the key objective of QbD, with an assessment of quality built into the process (pda..2012). Using CQA monitoring during use, a causal relationship can be established between Critical Process Parameters (CPP) and product quality. When process characterization has not been fully completed, establishing such a method early in product development can reduce the time required for development and scale-up (kozlwoski., 2006) as well as reduce manufacturing inefficiency, which allows for tighter control of product quality and improved yields (Yu et al, 2014).

The results presented here support decisions made in model development and demonstrate that raman spectrum based PLS models are able to predict both saccharification and glycosylation of the CHOmAb-producing bioreactor process, focusing especially on glycan spectra.

Example 3: prediction of manufacturing-scale saccharification and glycosylation using models developed with reduced-scale data (scheme 2)

In this example, the ability to accurately predict at the manufacturing scale was explored using each model developed by CSS described above in example 2. No adjustments or additional data are used to develop the model in this embodiment.

Each model (mono-glycosylated, non-glycosylated, G0F-GlcNac, G0F, G F and G2F) was tested and evaluated against a new CTSS consisting of data from a single manufacturing scale batch (2000L batch a) of the cell culture process by using the predictive function in Simca 15.1. The batch was randomly selected as CTSS from the available 2000L batch data. CTSS was then used to explore the ability of raman models developed using only scaled-down data (1L, 5L) to predict manufacturing-scale CQAs. Leave-on cross-validation was not repeated as no changes were made to the model and thus no changes in RMSEcv were observed. Each of the RMSEP and R2 values for the 7 models was compared to the output of the model generated in example 2 to determine if a separate scale-down was sufficient for PLS model development.

In flow 2, model robustness was tested against manufacturing scale data (2000L). Each model of flow 1 described above in example II was tested by predicting a new blind CTSS consisting of process data from a 2000L batch. Model statistics for RMSEcv as observed in flow 1, because no additional data is added or deleted from the model. In the case of all models, when predicting a 2000L blind dataset, an average R is observed ² Value of>0.80 (Table 2) showing good variability within each model.

Table 2: scheme 2: model development statistics for raman PLS models for saccharification and glycosylation。

Saccharification models all performed well, with 0.3553% RMSEP observed with each model (mono-saccharification and non-saccharification). Surprisingly, the RMSEP values observed here are lower than those observed in scheme 1 (described in example 2). This would suggest that reduced-scale data alone may be sufficient to develop a model for monitoring the glycation profile of a therapeutic protein (e.g., mAb) process. The interactions that produce protein glycation depend on the level of reducing sugars in the bioreactor, temperature and time (Quan et al, 2008). While specific saccharification site occupancy can be difficult to control (Wei et al, 2017), the overall level of saccharification can be maintained at certain levels during production scale-up. Thus, process variables that normally contribute to saccharification are maintained at comparable levels throughout the bioreactor scale. For this reason, using reduced-scale data may be sufficient to develop a robust saccharification model.

Statistical information related to the glycosylation pattern showed more variability when tested against manufacturing scale data And (5) responding. The RMSEP value for the G2F model was observed to be 0.4074%, which is outside of accepted standards, and therefore the model was considered unsuitable for monitoring on a manufacturing scale. Interestingly, R of the model ² Value of>0.85, further enhancing the importance of consulting a plurality of statistical information when evaluating the PLS model.

Showing the trend of raman model predictions versus offline values during 2000L batch progress (fig. 4A-4G) demonstrates model statistics where G2F deviates from the offline measurement trend of the entire batch (fig. 4G). The G0F-GlcNac, G0F, G F models met the acceptance criteria and visually appeared to follow the trend of offline measurements. G0 and G0F-GlcNac perform best, having values that are comparable to or lower than the RMSEP values obtained in scheme 1. G0F and G1F showed a significant increase in RMSEP values and from the trend (fig. 4E and fig. 4F, respectively) it was observed that the predictive power of these models began to decrease after day 07 in the process. Prior knowledge suggests that the process goes from the exponential growth phase to the stationary phase on day 07/08. As CHO cell cultures progress and enter the stationary phase of growth, they reached the peak of antibody production. The cell size and metabolic rate change with increasing cell output at twice the rate of exponential growth phase (Templete et al, 2013; xiao et al, 2017).

The model presented in scheme 2 will show that the process variations that occur at this point in time have different effects on the scaled-down versus manufacturing-scale saccharification spectra. Even after introducing variability into the model using spectra of multiple scales (1L and 5L), different raman probes/optics, different glycemic feed strategies (6 x 1L) and glycemic peaks (1 x 5L), the predictive ability of the model based on reduced-scale data alone was found to be not robust at manufacturing scale.

Taking the model as a combined spectrum, the glycosylation model built using the reduced-scale data performed worse when the manufacturing data was predicted, with increased prediction errors observed in G0, G1F, and G2F. In particular, the prediction error of G2F increases significantly (79%) so that it is outside acceptable limits for the prediction error (RMSEP). While the scaled-down bioreactor process is designed to represent a manufacturing process, it is apparent here that the manufacturing-scale glycosylation kinetics are somewhat different, whereby the scaled-down data is not sufficient for robust PLS model development alone.

Example 4: incorporating manufacturing scale data into a raman spectrum-based PLS model to predict manufacturing scale saccharification And glycosylation (scheme 3)

This example describes the construction of each PLS model using an updated CSS, which consists of CSS data from the small scale experiment described above in example 2, supplemented with data from a single manufacturing scale (2000L lot B) lot. The calibration dataset of the model developed in scheme 1 (described above in example 2) was supplemented with single lot manufacturing scale data (2000L) and predictive testing was performed according to scheme 2 (described above in example 3) using the same 2000L blind CTSS.

Then, 16 batches (9×1l, 6×5l, and 1×2000L) of CSS were used to develop 7 updated models as in example 2. Once completed, each of the 7 models is evaluated by leave-in cross-validation to derive updated RMSEcv and component numbers for each model. The predictive Root Mean Square Error (RMSEP) of the predictions and the R2 value of the updated model were identified using the predictive function test model in simca15.1 with respect to the predictive ability of the new CTSS. As in the previous procedure, RMSEcv, RMSEP, R and VIP values output from each model were evaluated to identify the model's ability to predict at the manufacturing scale. They were also used to compare predictive power with the models developed and used in schemes 1 and 2 to determine the necessity to include manufacturing scale data in the CSS of the raman PLS model for saccharification and glycosylation.

The model herein (flow 3) attempts to explore the effect of scale data on the predictive power and robustness of the model constructed in flow 1 (described above in example 2). The model from scheme 1 was supplemented with process data from a 2000L scale batch run. Adding the 2000L lot data means that 16 lots are used to develop each model. Cross-validation is again used here to determine the optimal number of components for model development and predictive testing. At the position ofAverage R was observed in cross-validation of all models ² Value of>0.8, wherein each model also had RMSEcv within acceptable standards (table 3).

Table 3: scheme 3: model development statistics for raman PLS models for saccharification and glycosylation

For this flow, the prediction test was completed using the manufacturing-scale blind test dataset from flow 2. The average R of each model was observed ² >0.80. Notably, all models showed a decrease in RMSEP values, indicating a decrease in prediction error for all models. Moreover, similar RMSEcv for each model or a slight decrease for each model was observed, which would indicate that adding manufacturing scale data to these models did not affect the model's ability to predict at small scale. When CSS was supplemented with manufacturing scale data, the G2F model was greatly improved as demonstrated by the reduction of RMSEP from 0.4074% to 0.0919% (relative improvement in prediction error of 77.5%). In addition, the trends seen in fig. 5A-5G indicate that the deviations observed in scheme 2 (see fig. 4A-4G) have been explained and that the model can be accurately predicted throughout the manufacturing scale.

Comparing the regions of each model with VIP score >1 to the model created in flow 1, each model designates similar or identical regions as important contributors to the model, in some cases the score of the identified regions changes due to manufacturing scale data addition, further supporting decisions made during model development (fig. 11A-11F). It should be noted that during this work, models were developed for both mono-saccharification and non-saccharification, each model having the same result in all three flows. Thus, a single model can be created and the complementary result inferred. For the purposes of this work, two models were created that attempted to support the decisions made in model development, and that showed acceptable values obtained in each case when tested against unknown data sets.

In summary, the results presented herein emphasize the importance of model robustness and the factors that must be considered when designing a model calibration dataset. Furthermore, including manufacturing scale data (although representing only 4.4% of the total data observations available in CSS in this study) allows for significant improvements in the predictive ability of the model and helps ensure robust model development.

Example 5: assessment of the impact of raman-based glucose feedback control on cellular bioreactor process development

The aim of this study was to explore the impact of continuous raman-based feedback control strategies using GMP-ready PAT management tools on cellular bioreactor processes (e.g., CHO cellular bioreactor processes). As an exemplary model, two CHO cell bioreactor processes were selected based on the respective development phases and development strategies. The impact of raman-based feedback control strategies on each CHO cell bioreactor process is considered in terms of cell growth, metabolism and productivity, as well as a number of key process parameters and quality attributes when compared to bolus fed-batch bioreactor processes. The results indicate that raman spectroscopy is an effective PAT tool in process development and optimization.

Example 5.1: materials and methods

The bioreactor is operated. Two mAb-producing CHO cell lines (cell line 1, cell line 2) were used in the experimental design. Three bioreactor batches of cell line 1 (batch a, batch B, batch C) and three bioreactor batches of cell line 2 (batch D, batch E, batch F) were performed in light-shielded 1L bioreactors (Eppendorf, hamburg, germany), with the process of each cell line lasting 14-15 days. On day 0, each batch was inoculated from the expanded seed propagated using the medium and target within the same inoculation density limit. Basal medium and feed medium were used for each batch and process control. Cell line 1 was used as a standard cell line for normal production. Cell line 2 was used as a high-output cell line with higher growth rate and resource requirements.

Two feed strategies were employed during each of the bioreactor runs. The feeding strategy of cell line 1 consisted of composite feeding and glucose feeding. From day 3, both the feed composite feed and the glucose feed are injected into batch a once a day based on a defined percentage of the container volume. Starting on day 3, batch B and batch C were fed a composite feed once a day based on a defined percentage of the vessel, and glucose feeds were automated and delivered as needed to maintain a predetermined target level of 1g/l glucose in the bioreactor based on raman glucose values, which was determined a posteriori. The feeding strategy of cell line 2 consisted of two complex feeds and glucose feeds. Batch D was injected with the feed composite feed and glucose feed once daily starting on day 3. The composite feed is delivered using a defined percentage of the container volume, and the glucose feed is based on achieving a defined glucose target concentration for that day. Starting on day 3, as a defined percentage of the vessel, batch E and batch F were fed two composite feeds once a day, and based on the raman glucose values, glucose feeds were automated and delivered as needed to maintain a predetermined target level of 2g/l glucose in the bioreactor, which was also determined a posteriori. Each batch had the same Dissolved Oxygen (DO), pH and temperature control targets. DO was controlled at 40% by aeration and oxygen injection. The pH target of 6.95 was maintained using the addition of carbon dioxide and 2.0M sodium carbonate. The temperature was controlled at a set point of 36.5 ℃ (35.5 ℃ -37.5 ℃) throughout the cell culture process.

Evaluation of bioreactor performance: offline sample analysis and online parameter trend. Offline samples were collected from each bioreactor daily and glucose, lactate, titer, viable CELL density and% viability were measured using a Vi-cellneta flex (Beckman Coulter, break, CA) Vi-CELL XR CELL viability analyzer (Beckman Coulter) and a Cedex Bio (Roche Holding AG, switzerland) offline analyzer. After each batch was completed, each daily culture sample was retained and frozen at-70 ℃ for further testing.

Saccharification is considered when evaluating the effect of glucose feedback control. Offline samples from day 4 to day 14/15 of each of the six bioreactor batches were thawed for product quality analysis of mabs produced throughout the batch. Saccharification of mAb was characterized off-line by LC/MS analysis. Briefly, all samples were initially purified using protein a purification. Protein samples for saccharification analysis were then pre-treated with EndoS enzyme to remove N-linked carbohydrates in order to eliminate glycoform-based sample heterogeneity. Protein samples were then separated by ultra-High Performance Liquid Chromatography (HPLC) using a Waters core, milford, MA using a gradient of acetonitrile and trifluoroacetic acid on a reverse phase column and analyzed by on-line electrospray ionization quadrupole time-of-flight mass spectrometry using a Xevo G2-XS mass spectrometer (Waters core, milford, MA). Mass/charge data collected across chromatographic peaks were then summed and deconvolved using Masslynx (Agilent Technologies, santa Clara, CA) or unici software (Waters Corp, milford, MA). The detected glycoisoforms are assigned based on deconvolution mass spectrometry analysis and their relative abundance is calculated using the peak intensities of the center deconvolution mass spectrum.

The on-line trends of pH and O2 delivery were trend analyzed and compared throughout the batch execution of both cell lines using Dasware control software (Hamburg, germany) as key indicators of bioreactor environment and process conditions in the bioreactor. The evaluation of the on-line and off-line data provides an indication of any process differences observed when using automatic feedback control of glucose.

And (5) Raman spectrum acquisition. For each batch, raman spectra were collected using a multichannel raman RXN2 system (Kaiser Optical Systems inc., ann Arbor, MI) containing a 785-nm laser source and a Charge Coupled Device (CCD) at-40 ℃. The detector is connected to an MR probe (Kaiser Optical Systems inc.) consisting of a fiber optic excitation cable and a fiber optic collection cable. Data were collected by an MR probe attached to a bIO-Optic-220 stainless steel probe (Kaiser Optical Systems inc.) inserted into a sterile bioreactor. Raman Runtime HMI (Kaiser Optical Systems inc.) was used for spectral acquisition in all batches. All raman spectral data were collected using a system setup of 10 second exposure for 75 scans, which produced the spectrum of the probe after 15 minutes (including an overhead time of 2.5 minutes). The Raman spectrum acquisition range is 100-3,425cm-1 of wave number. The reduced-scale containers are protected from light by aluminum foil. The intensity calibration of the instrument was performed with Hololab Calibration Accessory (Kaiser Optical Systems inc.) and the internal calibration was set to be performed every 24 hours throughout the bioreactor process, before each use of the system.

Raman-based PLS model development of glucose. A cell line specific raman-based glucose PLS model was developed in this study for cell line 1 and cell line 2 to facilitate the real-time generation of glucose concentration in the bioreactor and to facilitate the performance of real-time feedback control of glucose. All stoichiometrical modeling was performed with SIMCA 15.0 (immerics inc., san Jose, CA). Offline measurements of glucose are aligned with raman spectra based on the time they were acquired. Each model consists of a scaled down Calibration Sample Set (CSS) and manufacturing scale data (if available). A cell line 1 model CSS consisting of 12 batches (6×5L, 3×2000L and 3×1L) (for model development) and 1 batch (1L) of Calibration Test Sample Set (CTSS) was used as the blind data set for the PLS model. The batch was randomly selected as CTSS from the 1L scale available batch data. Cell line 2 model CSS consisting of 7 batches (3×5L and 4×1L) (for model development) and 1 batch (1L) of Calibration Test Sample Set (CTSS) was used as the blind data set for the PLS model. The batch was randomly selected as CTSS from the 1L scale available batch data. CTSS was chosen as the 1L batch because this is the scale at which both models will be deployed for real-time feedback control. All model development data was collected from laboratory scale studies and manufacturing scale batches previously performed. Robustness within the CSS is ensured by including process and technology variability for each of the generated PLS models. Different sampling strategies are employed in many lots included in the CSS in order to break any spurious correlations with lot progress.

The X-variable for each model is raman spectrum (centered) and the Y-variable is the off-line value of glucose (single variable reduction). The wavenumbers of the Raman spectra for each model were chosen to be 415-1800cm-1 and 2800-3100cm-1. The spectral filters applied to each PLS model were Savitsky-Golay first derivative second order (15 cm-1 points) and standard normal variables (SNV; data not shown). Each PLS model was built and errors were assessed by using a leave-in cross-validation method (leaving each batch in the model development once). Model errors were averaged based on model-to-omitted lot predictions to identify cross-validated root mean square errors (RMSEcv). RMSEcv indicates the predictive capability of the model based on the data used to build the model. A lower mean error (RMSEcv) indicates that the model is improved. This enables better informed decisions as to which component amounts are to be used for the model generated for testing for the blind data set. Each model is also evaluated using an estimated root mean square error (RMSEE), a model performance index related to the residuals of the data points in the CSS (i.e., the fit of the model). The predictive Root Mean Square Error (RMSEP) of the predictions was identified with respect to the predictive ability of CTSS using the predictive function test model in Simca 15.1, which indicates the predictive ability of the model with respect to the unseen dataset. Regression (R2) values, coefficients of variation, were recorded for each PLS model. The R2 value is used to determine the amount of change in the Y variable that the model predictor (X variable) can interpret. The closer the R2 value is to 1, the greater the model interpretation Y variable. Model performance was evaluated based on the corresponding RMSEE, RMSEcv, RMSEP and R2 values for each model.

Implementation of raman-based feedback control of glucose. Automation of glucose feeding was facilitated by PAT management tool synTQ (Optimal Industrial Automation Limited, UK). Briefly, the schedule (recipe) created within synTQ connects the raman instrument to the glucose PLS model and bioreactor system to perform feedback control. The initiation programmed in synTQ notifies the raman instrument to begin collecting spectra via an Open Platform Communication (OPC) connection. The generated spectral data is then sent to synTQ, which is then fed into the glucose PLS model contained in the orchestrated SIMCA Q engine block. The spectral data is converted to glucose readings for each point at which the spectral data is generated, which is then used to calculate the glucose feed to be delivered within the schedule. The resulting calculation of glucose feed was then transferred from synTQ to Dasware bioreactor control software (Eppendorf, germany) to start and stop the glucose feed pump on the bioreactor system after the feed had been delivered by OPC connection. The glucose target concentration is maintained by starting glucose feed to reach the target concentration whenever the raman and glucose model data indicate that the glucose concentration in the bioreactor has fallen below the target value. For each raman spectrum generated, the process is repeated for the duration of the batch.

Example 5.2: results

Raman-based PLS model development of glucose. Raman spectra collected from previously completed batches and offline glucose data were used to develop a raman-based PLS model prior to execution of a glucose feedback control batch. Cell line specific glucose models were developed for cell line 1 and cell line 2. Cell line 1 model contained 170 data points in its CSS and cell line 2 model contained 169 data points in its CSS. A leave-in cross-validation method is used to determine the optimal number of components per model.

FIGS. 12A and 12B summarize the final model for each cell line, with a RMSEE of 0.1845g/l for the cell line 1 glucose model and 0.3532g/l for the cell line 2 glucose model. The RMSEE values of both models combined with R2 values >0.95, indicating that there is a strong variability within each model and is used to support the decisions made in the data included in each model CSS. The acceptable prediction accuracy criteria for each model were determined to be ∈0.5g/l. This is determined relative to the current glucose feeding targets employed in each cell line bioreactor process.

Glucose measurements outside of this range may have a negative impact on the process for process control purposes, as it may result in an over-or under-feeding of the bioreactor. Thus, the accuracy of each model was further verified based on the comparison of RMSEcv and RMSEP for each glucose model to the acceptance criteria. Cell line 1 and cell line 2 models were both determined to have an optimal component number of 6, corresponding to RMSEcv values of 0.2695g/l and 0.3895g/l, respectively. The number of components selected for each model gives RMSEcv that falls within acceptable criteria for prediction error, indicating the ability of the model to predict variables of interest based on the calibration data set, and is therefore considered suitable for model development. When each model is tested for the corresponding CTSS, complementary RMSEP values are observed. Cell line 1 glucose model had a RMSEP of 0.2926g/l and cell line 2 model had a RMSEP of 0.2573 g/l. Both values are within established acceptance criteria (.ltoreq.0.5 g/l) for model development. Good agreement between model statistics suggests that design considerations in the development of both models are reasonable and ensure that each model is robust and will accurately predict and control glucose levels in the bioreactor process for each cell line.

Model CSS for cell lines 1 and 2 differ in terms of data availability at the time of model creation due to differences in the process development stage of each cell line. Manufacturing scale batches may be used for inclusion in the cell line 1 model, while only laboratory scale data may be used for inclusion in the cell line 2 model. Nevertheless, for the purposes of this work, two models were developed with acceptable accuracy. The cell line 1 model may be more easily scaled up in its current state, while the cell line 2 model may require additional data and model adjustments. The availability of additional data collected at the manufacturing scale may improve these models for in-line deployment on the scale.

Implementation of raman-based feedback control of glucose. For this study, an automatic feedback control strategy to maintain glucose at a predetermined setpoint concentration was considered. Two cell lines were selected to test the strategy, and each strategy was repeated once and compared to the strategy currently employed by each cell line (once daily bolus feed in both cell lines). Each study was performed using a single raman instrument, so the spectrum acquisition was limited to once every 45 minutes. This is determined as an acceptable interval for automatic feedback control of glucose based on previous data generated for the process set point and glucose consumption rate for each cell line considered.

The glucose trend of cell line 1 lot is summarized in fig. 13A-13C. Fig. 13A shows raman glucose trend for bolus feed lot of cell line 1 lot a. Raman measurements of glucose performed well throughout the batch, with RMSEP observed at 0.2917g/l. Raman measurements provide a deeper understanding of glucose trends and consumption for the entire batch and are used to indicate whether manual delivery of glucose is sufficiently targeted for each feeding event. In lot a, it was observed that the bolus feeding strategy of glucose produced a very large daily peak of glucose when feeding was started from day 03, which was not observed if only offline daily measurements were used. This suggests that when raman is used as a PAT tool, there is additional process information available that would otherwise be unavailable to the process development team, even if it does not control feed delivery. The glucose concentration throughout batch A fluctuated between 1-3 g/l.

In contrast, the automatic feedback control batches of cell lines 1, batch B and batch C (fig. 13A and 13C) showed tighter feed curves, with glucose concentration maintained at 1g/l in each batch. In each automatic feedback control batch, feeding was started from day 03, which is comparable to the timing of the initial bolus feeding of batch a. No significant concentration peaks were observed and consistent glucose concentrations were maintained for 12 days. The RMSEP for lots B and C were 0.2887g/l and 0.2940g/l, respectively, and the low RMSEP in each case was further used here to demonstrate that the glucose model for cell line 1 was strong and performed well. The low RMSEP in each case indicated that the model was able to provide an accurate indication of the glucose concentration of the process and well inform the feed delivery within +/-0.5g/L of the established acceptance criteria.

Cell line 2 glucose trend showed similar results to cell line 1 and can be seen in fig. 14A-14C. Batch D trend (fig. 14A) shows a typical trace of the bolus feeding process for cell line 2 with a very large expected daily glucose peak from a single feeding delivery. Furthermore, raman monitoring reveals a more informative process than single daily offline measurements, which more reflects glucose trends. Because of the greater demand for glucose by cell line 2, the concentration range was distributed over a wider range than cell line 1, with a process range of 0.5-4.5g/l. The RMSEP for the glucose model in lot D was 0.2369g/l, which is within the accepted standards of performance and indicates that the data provided by the model accurately reflects the actual process values. Batches E and F (FIGS. 14B and 14C, respectively) were controlled to set point 2g/l for 11 days.

The key difference observed in these batches was that the feed control was started on day 03, and the manual bolus of the feed batch was started on day 04 as the glucose level was further reduced. Batch E and batch F both had good control of glucose and RMSEP was 0.2516g/l and 0.2119g/l, respectively, indicating that the model performance was very strong in each case. Note that the batch F glucose set point drifts upward from the target 2g/l from day 09 of the process, which is believed to be due to calibration problems with the pump used to deliver the glucose feed. Nonetheless, lot F had a strict glucose control range and exceeded the target acceptance criteria +/-0.5g/l only on the last day of the process.

The raman-based glucose feedback control of both cell line 1 and cell line 2 provides a more stable glucose supply to the bioreactor environment and prevents abrupt changes in glucose concentration as seen in bolus feed batches. mAb bioreactor process development and optimization has focused on producing high product titers with well-defined and controlled product quality characteristics. The feedback control mechanism presented herein is repeatable and consistent and provides a well-defined feed profile for both cell line 1 and cell line 2.

PAT tools (including raman spectroscopy) are themselves a relatively low cost process development tool beyond initial investment. By performing high throughput and experimental design of bioreactor batches, large amounts of process data are collected very rapidly at an early stage of process development. Inclusion of raman probes for data collection in early development batches, such as these allow dense data collection and model creation phases to occur very early, which does not interfere with the strict time constraints of process development (e.g., mAb process development), and would allow advanced feedback control strategies, such as those proposed for cell line 1 and cell line 2, to be implemented. As has been shown in this study, this technique can provide a large amount of additional process information as a monitoring tool, with both lot a and lot D identifying peaks and peaks in glucose trend that would otherwise be ignored when trend analysis is performed on a single daily offline measurement. The combination of such data-rich methods with process control strategies may support changing and improving the methods employed in process development and enhancement.

Evaluation of bioreactor performance: offline sample analysis and online parameter trend. The objective of this study was to deploy raman-based automated feedback control for two exemplary cell line processes and evaluate the impact of this feedback control on each process. Both processes are considered in terms of the effects of automatic feedback control on process variables associated with bioreactor health, productivity, and environmental conditions. Any observed effects are discussed with respect to the development phase of each process and how PAT-centric approaches at this phase support process development and enhancement.

Fig. 15A-15H show the process trend of bolus feeding and automatic feedback control batches for cell line 1. It can be observed that the growth and health of each bioreactor batch was comparable, as indicated by VCD, viability and LDH trends in fig. 15A, 15B and 15C. The VCD trend for each batch peaks between 8.00-8.75X10-6 cells/ml on day 07/08 and gradually decreases for the remainder of the process. For each batch, the last day viability was observed to be between 70-75% and the last day LDH was observed to be between 1400-1450 IU/ml. These trends and small degree of variability observed between each batch indicate that raman-based feedback control has no adverse effect on the cell line 1 process.

Variability in bioreactor environment was observed depending on which feed strategy was employed for cell line 1. The pH trends for batches A, B and C (fig. 15D) are similar until day 07, at which point bolus feed batch a continues to increase to pH 7.3, thereby requiring constant addition of CO2 to maintain the pH of the batch within the set point from day 07 to day 12. Raman controlled batches B and C showed a different trend than batch a and remained within the set point range throughout the process, especially during days 07 to 12, with pH maintained at-pH 7.2. The on-line O2 trends for batches A, B and C are different from each other, as can be seen in fig. 15E. Repeated batches B and C showed a gradual increase in demand for O2 delivered from day 0 to day 15. The O2 requirement for batch A varies with the increase and decrease in O2 requirement throughout the bioreactor process. Trend differences started to appear at day 07 for both O2 and pH, consistent with the time the process reached peak VCD. The differences in these trends observed in bolus feed batch a versus feedback control batches B and C would potentially indicate that there is a difference in cell metabolism of the bolus and feedback control batches after the peak VCD. The increasing O2 demand and pH trend not reaching the upper control limit observed throughout feedback control batches B and C may indicate that the environment maintained in the feedback control batches is more beneficial in terms of cell metabolism, possibly due to more consistent delivery of glucose feed.

The total amount of glucose fed in the automatic feedback control batch of cell line 1 was less than the bolus fed batch. Fig. 15F shows that the last day glucose feed volumes for lots B and C were 51.5ml and 54.5ml compared to the last day glucose feed volume 59ml for lot a, indicating a 12.7% and 7.6% reduction in the glucose feed delivered in lots B and C, respectively. Since feedback control batches B and C automatically maintain a target concentration of 1g/L at each interval in which a raman measurement is obtained, the level of control over the glucose concentration in the bioreactor is higher compared to bolus fed batch a. Here, automatic glucose delivery is based on real-time measurements, which capture glucose demand in the bioreactor more accurately than a single daily off-line sample. Bolus delivery of glucose is based on a feed target that has been performed from a previous batch and established by measuring glucose concentration offline. This approach does not take into account the current performance of the bioreactor and may result in overestimated glucose demand, as identified in lot a herein.

The automatic feedback control of cell line 1 resulted in comparable product titers compared to the bolus feed strategy, with fig. 15G showing that the last day titer of each batch was between 2.2-2.5G/L, as expected. When compared to bolus feed batch a, a reduction in total product saccharification was observed in both batch B and batch C. The last day of lot a was observed to have a saccharification of 6.68% while lot B was observed to have a 3.78% and lot C was observed to have a 2.81%, indicating a reduction in saccharification of 43.4% and 57.9%, respectively, as seen in fig. 15H. The last day titers were comparable in all 3 cell line 1 batches, further serving to support that no negative impact on the process was observed when raman-based feedback control was used. Of note is the significant improvement in product saccharification, which can be attributed to the lower total glucose volume in both batches B and C as a result of raman-based feedback control. The total concentration of glucose in the bioreactor directly affects the level of saccharification, and this control of CQA identifies a key process improvement associated with the PAT method that would not otherwise be achieved during process development.

Similar process variation assessments were considered for bolus feeding and automatic feedback control batches for cell line 2. Fig. 16A-16H summarize comparisons for each batch. VCD, viability and LDH trends indicate that there is a difference in the growth curves for batches D, E and F. FIG. 16A shows that bolus feed lot D and auto feedback control lot E reached a peak VCD of 10-10.5X10-6 cells/ml on day 08 of the process, while auto feedback control lot F reached a peak of 11.97X10-6 cells/ml on day 10. The VCD in automatic feedback control batches E and F decreased below the drop in bolus feed batch D, which values for batches E and F were 4.81 x 10 x 6 cells/ml and 8.07 x 10 x 6 cells/ml, respectively, compared to 3.65 x 10 x 6 cells/ml for batch D for the last day. The viability and LDH trend further underscores the differences in bioreactor health status, as can be seen in fig. 16B and 16C. Of the two automatic feedback controlled batches, bioreactor viability remained higher than the bolus fed batch, with batches E (56%) and F (72.5%) exhibiting 11.1% and 27.6% higher viability than batch D (44.9%) on the last day of the process. From day 08 to day 14, LDH was observed to be higher in the bolus fed batch than in the automatic feedback control batch, with batches E (6053 IU/ml) and F (4834.5 IU/ml) observed to be 21% and 37% lower than batch D (7694 IU/ml) on day 14. Less severe degradation of VCD and viability for batches E and F and lower LDH would indicate that a glucose set point of 2g/L for raman-based feedback control strategy has a positive impact on cell growth and health in the bioreactor. A more stable glucose concentration in the bioreactor environment appears to favor cell line 2 processes than the daily peaks and valleys associated with bolus feeding seen in lot D. The ability to maintain process viability and produce product output for longer periods of time serves as another reason for including PAT in the process development stage, as the same long-term growth levels may not be obtained using conventional development strategies.

For each process, bioreactor pH, O2 flow and glucose feed volume were tracked online until day 13, the connection problem prevented recording the remainder of the process data in each case. The pH trend was observed to be similar for each batch, and fig. 16D shows daily peaks up to pH 7.1, consistent with daily addition of the composite feed. Fig. 16E shows that the O2 demand is also consistent across all 3 batches, with increasing demand for O2 per batch, up to day 10. Starting on day 10, the lot demand for O2 was different because lots E and F O2 flow were maintained at >0.5sl/h, while lot D O flow gradually decreased below 0.5 sl/h. FIG. 15F shows that the last day of batches D, E and F has a similar glucose feed volume of 26ml. The consistency of these parameters suggests that there is no negative impact on the cell line 2 process associated with raman-based feedback control. The decrease in O2 demand in bolus lot D may be due to the decrease in the amount of living cells in the lot since day 10. Interestingly, the total glucose delivered in bolus feed batch D and raman-based feedback control batches E and F was comparable, which would indicate that the previously observed effect on bioreactor health was not due to a smaller amount of feed delivered, but rather to the glucose feed delivery system, i.e. single daily bolus feed versus multiple smaller feeds per day.

When automatic feedback control of glucose is used, the productivity of the cell line 2 process increases. FIG. 16G shows that the last day titers of batch E (5.75G/L) and batch F (5.77G/L) were 25% higher than bolus fed batch D (4.6G/L). The quality of product titer was also improved in the automatic feedback control batch (fig. 16H), where the last day of the bolus fed batch D was observed to have 3.06% product glycosylation, while batches E and F were observed to be lower, 2.13% and 2.37%, respectively. Although all three batches showed similar total volumes of added glucose, saccharification results would indicate that continuous control of setpoint concentration is a more advantageous strategy in terms of product quality. Improvements in growth, product yield and quality indicate that more robust processes (such as those of cell line 2) using PAT development strategies can further drive development and produce results that may not be possible with conventional bioreactor processes.

The once daily bolus delivery of glucose as a platform feeding strategy is widely considered in mAb bioreactor process development, as it has been demonstrated that nutrients can be adequately delivered to the bioreactor to promote growth and product output. The main problem with this method of nutrient delivery is the impact on product quality. In this study, large daily fluctuations in glucose concentration could be observed in bolus feed batches (fig. 13A and 14A). The effect of glucose on product quality is mainly manifested in its promoting effect on product saccharification. Saccharification is a key quality attribute of mAb bioreactor process, the direct impact of saccharification on mAb is different, and its impact is likely to depend on the mAb site being saccharified and the total amount of saccharification present.

Automation of the process eliminates the additional resource requirements for more frequent bioreactor feed plans and increases QbD considerations for process development, which may allow further improvements to be made using techniques that supplement current development methodologies. Automatic feedback control of setpoint glucose concentration using raman-based PLS model as shown in this study has a direct positive impact on each cell line process without any additional resource requirements. The automatic feedback control of cell line 1 resulted in reduced overall product saccharification in both batch B and batch C as compared to bolus feed batch a, while maintaining comparable growth curves and product yields. This improvement can be attributed to the consistent delivery and overall reduced volume of glucose feed required for automatic feedback control batches B and C. Cell line 1 represents a platform process in post-development. While important process development has been completed for this cell line, the introduction of PAT tools (such as raman-based automatic feedback control) may support beneficial process development without any expected negative impact on the currently designed process.

Cell line 2 represents a more intensive process in an earlier development stage than cell line 1. Process enhancement in mAb-producing bioreactors depends primarily on the biological limitations of the process. Medium formulation and enrichment and feed strategy optimization are important to maintain higher productivity of VCD, cellular metabolism and enhancement processes. Automation of glucose feeding for cell line 2 allowed for testing of more optimized feeding strategies. Although the total volume of glucose delivered in bolus and automatic feedback control batches was similar, many process improvements were observed. In both automated feedback control batches of cell line 2, the improved growth and viability characteristics, as well as greater product output and improved product quality underscores the effectiveness of PAT tools in the development process. The ability to directly affect product quality and output in process development is consistent with the objectives of QbD initiative.

Implementing PAT techniques (such as raman spectroscopy) at early or late development stages can have a significant impact on process development, amplification, and robustness of the manufacturing process. Here, when the PAT method was employed in the development stage, the two CHO cell line bioreactor processes showed an indication that the efficiency of their current process was improved.

The use of methods such as automatic feedback control using raman in the development stage can generate valuable data with low risks involved. The ability to normalize PAT in process development can result in a more efficient and cost effective manufacturing process. As a direct result of the more efficient development, increasing product output and/or improving product quality at the manufacturing scale may affect the patient's dosage requirements, which may result in more doses of product being generated per batch, saving significant cost and time. Consideration of techniques such as this in the development and implementation of commercial manufacturing will yield an end-to-end QbD approach that will maximize process efficiency and meet the need for new mAb therapies.

The results presented herein demonstrate that including raman spectra as PAT tools in process development can increase product output and reduce overall product saccharification. For example, cell line 1 process was improved by reducing total glucose delivery to the bioreactor in raman-based feedback control batches, thereby reducing total product saccharification by 43.7% and 57.9% while maintaining comparable growth curves and product output. In addition, the cell line 2 process is improved, the cell health is prolonged, and as the process proceeds, the decrease in cell viability is greatly reduced. This improvement produces a linkage effect that increases the product output of the process by 25%. Thus, including raman spectra as PAT tools in process development can have a significant impact on process development, amplification, and robustness of the manufacturing process.

***

In the description and claims above, phrases such as "at least one of … …" or "one or more of … …" may appear after a connected list of elements or features. The term "and/or" may also occur in a list of two or more elements or features. Such phrases are intended to mean any one of the listed elements or features alone or in combination with any one of the other listed elements or features unless otherwise implicitly or explicitly contradicted by context in which it is used. For example, the phrase "at least one of a and B"; "one or more of A and B"; and "a and/or B" are each intended to mean "a alone, B alone, or a and B together". Similar explanations apply to a list comprising three or more items. For example, the phrase "at least one of A, B and C"; "one or more of A, B and C"; and "A, B and/or C" are each intended to mean "a alone, B alone, C, A and B together, a and C together, B and C together, or a and B and C together. The use of the term "based on" in the foregoing and claims is intended to mean "based at least in part on" such that unrecited features or elements are also permitted.

The subject matter described herein may be embodied in systems, devices, methods, and/or articles of manufacture, depending on the desired configuration. The implementations set forth in the foregoing description are not intended to represent all implementations consistent with the subject matter described herein. Rather, they are merely examples of some of the many aspects that may be relevant to the described subject matter. Although a few variations have been described in detail above, other modifications or additions are possible. In particular, additional features and/or variations may be provided in addition to those set forth herein. For example, implementations described above may involve various combinations and subcombinations of the disclosed features and/or combinations and subcombinations of several further features disclosed above. Furthermore, the logic flows depicted in the figures and/or described herein do not necessarily require the particular order shown, or sequential order, to achieve desirable results. Other implementations may be within the scope of the following claims.

Claims

1. A method for determining glycan structures on a glycosylated molecule, the method comprising:

2. The method of claim 1, further comprising refining the one or more regression models based on a combination of the obtained spectral data and the measured spectral data.

3. The method of claim 1 or 2, further comprising maintaining one or more operating parameters of the one or more second bioreactors based on the determined levels to produce a desired glycosylated molecule.

4. The method of claim 1 or 2, further comprising selectively modifying one or more operating parameters of the one or more second bioreactors based on the determined level to produce a desired glycosylated molecule,

wherein the one or more operating parameters optionally include pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof,

wherein the nutrient level is optionally selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration, and wherein the glucose concentration is optionally automatically modified based on the measured spectral data.

5. The method of any one of claims 1 to 4, further comprising purifying the glycosylated molecule.

6. The method of any one of claims 1 to 5, wherein the glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

7. The method of any one of claims 1 to 6, wherein the obtaining is performed within two or more bioreactors having different volumes.

8. The method of any one of claims 1 to 7, wherein

(a) The first threshold is:

(i) About 250 liters or less;

(ii) About 100 liters or less;

(iii) About 50 liters or less;

(iv) About 25 liters or less;

(v) About 10 liters or less;

(vi) About 5 liters or less;

(vii) About 2 liters or less;

(viii) About 1 liter or less; and/or

(b) The second threshold is:

(i) About 1,000 liters or more;

(ii) About 2,000 liters or more;

(iii) About 5,000 liters or more;

(iv) About 10,000 liters or more;

(v) About 15,000 liters or more;

(vi) About 10,000 liters to about 25,000 liters;

(vii) At least 5 times the first threshold;

(viii) At least 10 times the first threshold;

(ix) At least 100 times the first threshold; or alternatively

(x) At least 500 times the first threshold.

9. The method of any one of claims 1 to 8, wherein

(a) The first volume is:

(i) About 0.5 liters to about 250 liters;

(ii) About 1 liter to about 50 liters;

(iii) About 1 liter to about 25 liters;

(iv) About 1 liter to about 10 liters; or alternatively

(v) About 1 liter to about 5 liters; and/or

(b) The second volume is:

(i) About 1,000 liters to about 25,000 liters;

(ii) About 2,000 liters to about 25,000 liters;

(iii) About 5,000 liters to about 25,000 liters;

(iv) About 10,000 liters to about 25,000 liters; or alternatively

(v) About 15,000 liters to about 25,000 liters.

10. The method according to any one of claims 1 to 9, wherein the PAT tool utilizes or otherwise comprises raman spectroscopy.

11. The method of any one of claims 1 to 10, wherein the one or more regression models comprise a Partial Least Squares (PLS) model.

12. The method of any one of claims 1 to 11, wherein the glycosylated molecule comprises a monoclonal antibody (mAb) or a non-mAb.

13. The method of any one of claims 1 to 12, wherein the determining step is performed as follows

(a) On site or off site; and/or

(b) In-line, on-line, off-line, or a combination thereof.

14. The method of any of the preceding claims, wherein the obtaining comprises: data characterizing the spectral data is received from the PAT tool.

15. The method of any of the preceding claims, wherein the generating is performed by one or more computing devices.

16. A method of producing a glycosylated molecule having a desired glycan structure, the method comprising:

maintaining one or more operating parameters of the bioreactor when:

the level of desired glycan structures is above a predetermined threshold, or

The level of undesired glycan structures is below a predetermined threshold; and

selectively modifying one or more operating parameters of the bioreactor when:

the level of desired glycan structures is below a predetermined threshold, or

The level of undesired glycan structures is above a predetermined threshold.

17. The method of claim 16, wherein the measuring is performed in-line, online, offline, or a combination thereof.

18. The method of claim 16 or 17, wherein the measuring is performed as follows:

(a) More than once a day;

(b) About every 5 minutes to 60 minutes;

(c) About every 10 minutes to 30 minutes;

(d) About every 10 minutes to 20 minutes; or alternatively

(e) About every 12.5 minutes.

19. The method of any one of claims 16 to 18, wherein the measuring is performed in a bioreactor having the following volumes:

(a) About 2,000 liters or more;

(b) About 5,000 liters or more;

(c) About 10,000 liters or more;

(d) About 15,000 liters or more;

(e) About 10,000 liters to about 25,000 liters; or alternatively

(f) About 15,000 liters.

20. The method of any one of claims 16 to 19, wherein the determining step is performed on-site or off-site.

21. The method of any one of claims 16 to 20, wherein the desired glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

22. The method of any one of claims 16 to 21, wherein the bioreactor is a batch reactor, a fed-batch reactor, a perfusion reactor, or a combination thereof.

23. The method of any one of claims 16 to 22, wherein the one or more operating parameters comprise pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof,

Wherein the nutrient level is optionally selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration; and is also provided with

Wherein the concentration of glucose is optionally automatically modified based on the measured spectral data.

24. The method according to any one of claims 16 to 23, wherein the PAT tool utilizes or otherwise comprises raman spectroscopy.

25. The method of any one of claims 16 to 24, wherein the one or more regression models comprise a Partial Least Squares (PLS) model.

26. A system for producing one or more glycosylated molecules, the system comprising:

means for culturing a cell line producing the glycosylated molecule;

27. The system of claim 26, wherein the cell line that produces the glycosylated molecule is a mammalian cell line, optionally wherein the mammalian cell line is a non-human cell line.

28. The system of claim 26 or 27, wherein the culturing comprises batch, fed-batch, perfusion, or a combination thereof.

29. The system of any one of claims 26 to 28, wherein culturing comprises the following volumes:

(a) About 2,000 liters or more;

(b) About 5,000 liters or more;

(c) About 10,000 liters or more;

(d) About 15,000 liters or more;

(e) About 10,000 liters to about 25,000 liters; or alternatively

(f) About 15,000 liters.

30. The system of any one of claims 26 to 29, wherein the measurement is performed in-line, online, on-line, off-line, or a combination thereof.

31. The system of any one of claims 26 to 30, wherein the measurement is performed as follows:

(a) More than once a day;

(b) About every 5 minutes to 60 minutes;

(c) About every 10 minutes to 30 minutes;

(d) About every 10 minutes to 20 minutes; or alternatively

(e) About every 12.5 minutes.

32. The system of any one of claims 26 to 31, wherein the one or more glycan structures are selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

33. The system of any one of claims 26 to 32, further comprising means for selectively modifying one or more operating parameters to enhance production of a desired glycan structure or structures,

Wherein the one or more operating parameters include pH level, nutrient level, medium concentration, frequency interval of medium addition, or a combination thereof,

wherein the nutrient level is optionally selected from the group consisting of: glucose concentration, lactate concentration, glutamine concentration and ammonium ion concentration, and

34. The system of any one of claims 26 to 33, wherein the one or more glycosylated molecules comprise monoclonal antibodies (mabs) or non-mabs.

35. The system of any one of claims 26 to 34, wherein the spectral data comprises raman spectra.

36. The system of any one of claims 26 to 35, wherein the one or more regression models comprise a Partial Least Squares (PLS) model.

37. The system of any one of claims 26 to 36, further comprising means for isolating the one or more glycosylated molecules.

38. A system for producing one or more glycosylated molecules:

a bioreactor comprising a cell line producing glycosylated molecules;

39. The system of claim 38, wherein the bioreactor is:

(a) About 2,000 liters or more;

(b) About 5,000 liters or more;

(c) About 10,000 liters or more;

(d) About 15,000 liters or more;

(e) About 10,000 liters to about 25,000 liters; or alternatively

(f) About 15,000 liters.

40. The system of claim 38 or 39, wherein the glycan structure is selected from the group consisting of: G0F-GlcNac, G0F, G F, or G2F, or combinations thereof.

41. The system of any one of claims 38 to 40, wherein the cell line that produces the glycosylated molecule is a mammalian cell line, optionally wherein the mammalian cell line is a non-human cell line.

42. The system of any of claims 38-41, wherein the PAT tool utilizes or otherwise includes raman spectroscopy.

43. The system of any one of claims 38 to 42, wherein the one or more regression models include a Partial Least Squares (PLS) model.