WO2025051772A1

WO2025051772A1 - Spectroscopy methods

Info

Publication number: WO2025051772A1
Application number: PCT/EP2024/074664
Authority: WO
Inventors: Michael Moses; Jan SCHWICHTENBERG; Katharina Alena BECKER; Janina BELTZ
Original assignee: Csl Behring Gmbh
Priority date: 2023-09-04
Filing date: 2024-09-04
Publication date: 2025-03-13

Abstract

The invention relates to methods for monitoring of parameters in solutions or suspensions and the application of same in methods for purifying solutions comprising human plasma proteins and other components. In one aspect, the invention provides a method for determining the concentration of an analyte in a sample obtained from processing of blood-derived plasma, the method comprising applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating test spectra and predictive models, comparing the test spectra with reference spectra obtained from reference samples having known concentrations of the analyte, to determine the concentration of the analyte in the sample.

Description

SPECTROSCOPY METHODS

Cross-reference to related application

This application claims priority from Australian provisional application no. 2023902846, the entire contents of which are incorporated herein by reference.

Field of the invention

The invention relates to methods for monitoring of parameters in solutions or suspensions and the application of same in methods for purifying solutions comprising human plasma proteins and other components.

Background of the invention

Human blood plasma contains over 100 proteins that cover a wide range of essential functions (e.g., defense against pathogens, blood coagulation or mass transport). Therapeutic proteins purified from human blood plasma play an important role in the treatment of life-threatening diseases. In orderto obtain plasma proteins for therapeutic purposes, these have to be isolated from a variety of plasma donations. The collection of human blood plasma can be performed, for example, by plasmapheresis, where whole blood is extracted from the donor and separated by physical separation procedures. During this process, cellular components are returned to the donor while the plasma is collected in a reservoir. Proteins are then purified from a pool of thawed plasma or cryo-poor plasma after removing cryo-precipitate by utilizing different purification procedures. For example, a subsequent cold ethanol fractionation, which is characterized by increasing ethanol concentrations and decreasing pH values, is considered as a central fractionation method for therapeutic proteins since the 1940s. Various combinations of negative temperatures, pH and ethanol concentration cause the stepwise precipitation of proteins such as immunoglobulins and albumin.

Process automation and Process Analytical Technology (PAT) is becoming increasingly important in the pharmaceutical industry. PAT enables real-time monitoring, providing high- value approaches to process understanding, control and optimization for efficient continuous manufacturing processes. The establishment of PAT in in the industry of purifying proteins from human plasma is currently at an early stage but also offers substantial advantages in this area. This could be process control based on or correlated to the direct measurement of a product or intermediate and/or critical quality attributes (CQAs) and/or process parameters in real time.

There is a need for new and/or improved methods for determining the concentration of various analytes in complex solutions during plasma processing to improve downstream efficiency, reduction in waste, and/or improve final product yield and/or quality. Due to the heterogeneity of plasma-derived product solutions and suspensions, the quantification of key chemical components, such as quantification of different proteins, excipients and/or other components, is complex and to date, has only been achieved by use of off-line analytical methods that require sampling effort and analysis lead times of commonly several days. Therefore, there is an increasing need for new analytical processes to determine concentration of analytes in solutions or suspensions from the processing of blood-derived plasma.

Reference to any prior art in the specification is not an acknowledgment or suggestion that this prior art forms part of the common general knowledge in any jurisdiction or that this prior art could reasonably be expected to be understood, regarded as relevant, and/or combined with other pieces of prior art by a skilled person in the art.

Summary of the invention

In one aspect, the present invention provides a method for determining the concentration of an analyte in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating test spectra, comparing the test spectra with reference spectra obtained from reference samples having known concentrations of the analyte, to determine the concentration of the analyte in the sample.

Typically, inelastic scattering produced spectra contain hundreds of variables and therefore some form of mathematical and/or statistical analyses, e.g. multivariate data analysis method, is preferably used to analyze raw data from the measurements. Such multivariate data analysis methods, e.g. learning methods, are well known in the art and include Partial least squares regression (PLS); PLS Discriminant Analysis (PLS-DA); Ordinary Least Squares (OLS) regression; MLR (multiple linear regression); OPLS (Orthogonal-PLS); SVM (support vector machines); GLD (general discriminant analysis); GLMC (generalized linear model); GLZ (generalized linear and non-linear model); LDA (Linear Discriminant Analysis); classification trees; cluster analysis; neural networks; and Pearson correlation.

In another aspect, the present invention provides a method for determining the presence of, or concentration of, an analyte in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of the analyte to determine the presence of, or concentration of, the analyte in the sample.

In one aspect, the present invention provides a method for determining the concentration of an analyte in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma, measuring inelastic scattering from the test sample, thereby generating test spectra, comparing the test spectra to a reference data set in the form of a model generated using multivariate analysis of processed reference spectra of reference samples having known concentrations of the analyte, to determine the concentration of the analyte in the sample.

In another aspect, the present invention provides a method for generating a training or reference spectrum or spectra, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma, wherein the test sample has a known concentration of an analyte or is known to have an analyte present; measuring inelastic scattering or Raman shift from the test sample, thereby generating a training or reference spectrum or spectra. Optionally the steps in the method may be repeated with test samples having different concentrations of an analyte, preferably the different concentrations are across a desired concentration range. Preferably the training or reference spectrum or spectra are used to generate a model, wherein the model is generated by using multivariate analysis of processed training or reference spectrum or spectra of reference samples having known concentrations of the analyte.

In any aspect, the analyte is total protein or alcohol (e.g. ethanol). In any aspect, the analyte is a specific protein present in human plasma, such as IgG or albumin. In these embodiments, the methods can then be used to determine the presence of, or concentration of, total protein, a specific plasma protein (such as IgG, albumin), or ethanol in a test sample obtained from processing of blood-derived plasma.

In one aspect, the present invention provides a method for determining the presence of, or concentration of, ethanol in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of ethanol to determine the presence of, or concentration of, the analyte in the sample.

In one embodiment, the present invention provides a method for determining the presence of, or concentration of, ethanol in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma prior to, during, and/or after ethanol precipitation; measuring inelastic scattering or Raman shift from the test sample, thereby generating test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of ethanol to determine the presence of, or concentration of, the analyte in the sample.

In one embodiment, the present invention provides a method for determining the presence of, or concentration of, ethanol in a sample obtained from an ethanol precipitation step during the processing of blood-derived plasma, the method comprising: applying a light source to test samples obtained from processing of blood-derived plasma at different times during ethanol precipitation; measuring inelastic scattering or Raman shift from the test samples, thereby generating a series of test spectra, comparing the test spectra with a reference spectrum that contains a Raman signature of ethanol to determine the presence of, or concentration of, the analyte in the sample over the course of the precipitation process.

In another aspect, the present invention provides a method for determining the presence of, or concentration of, IgG in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of IgG to determine the presence of, or concentration of, the analyte in the sample.

In another aspect, the present invention provides a method for precipitating IgG during the processing of blood-derived plasma, the method comprising: adding ethanol to an IgG containing solution or suspension; applying a light source to a test sample from the solution or suspension after the addition of ethanol; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of IgG to determine the presence of, or concentration of, IgG in the sample, optionally, adding more ethanol, or continuing to add ethanol, preferably until a comparison of a test spectrum with a reference spectrum that contains a Raman signature of IgG indicates the desired level of IgG.

In another aspect, the present invention provides a method for determining the presence of, or concentration of, albumin in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of albumin to determine the presence of, or concentration of, albumin in the sample.

In another aspect, the present invention provides a method for precipitating albumin during the processing of blood-derived plasma, the method comprising: adding ethanol to an albumin containing solution or suspension; applying a light source to a test sample from the solution or suspension after the addition of ethanol; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of albumin to determine the presence of, or concentration of, albumin in the sample, optionally, adding more ethanol, or continuing to add ethanol, preferably until a comparison of a test spectrum with a reference spectrum that contains a Raman signature of albumin indicates the desired level of albumin.

In another aspect, the present invention provides a method for determining the presence of, or concentration of, total protein in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of total protein to determine the presence of, or concentration of, total protein in the sample.

In another aspect, the present invention provides a method for precipitating total protein during the processing of blood-derived plasma, the method comprising: adding ethanol to a protein solution or suspension; applying a light source to a test sample from the solution or suspension after the addition of ethanol; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of total protein to determine the presence of, or concentration of, total protein in the sample, optionally, adding more ethanol, or continuing to add ethanol, preferably until a comparison of a test spectrum with a reference spectrum that contains a Raman signature of total protein indicates the desired level of total protein. In any aspect, the Raman signature comprises a selected number of peaks and associated peak ranges that allow for the evaluation (e.g., identification) of an analyte in the sample.

In any aspect, the Raman signature comprises a selected number of peaks and associated peak ranges that allow for the evaluation or determination of the level or concentration of an analyte in the sample.

In some embodiments, a Raman spectrum is obtained using Surface Enhanced Raman Spectroscopy (SERS), resonance Raman spectroscopy, tip-enhanced Raman spectroscopy, polarized Raman spectroscopy, stimulated Raman spectroscopy, transmission Raman spectroscopy, spatially offset Raman spectroscopy, difference Raman spectroscopy, Fourier Transform (FT) Raman, or hyper Raman spectroscopy. In some embodiments, a Raman spectrum is obtained using a Raman analyzer configured with a laser (e.g. laser diode) or other suitable light source that operates at, but is not limited to, wavelengths in a range of 325 nm to 1064 nm.

In any embodiment, the light source has a wavelength of at least about 500nm, at least about

525nm, at least about 550nm, at least about 575nm, at least about 600nm, at least about

625nm, at least about 650nm, at least about 675nm, at least about 700nm, at least about

725nm, at least about 730nm, at least about 735nm, at least about 740nm, at least about

745nm, at least about 750nm, at least about 755nm, at least about 760nm, at least about

765nm, at least about 770nm, at least about 775nm, at least about 780nm, at least about

785nm, at least about 790nm, at least about 795nm, at least about 800nm, at least about

805nm, at least about 810nm, at least about 815nm, at least about 820nm, at least about

825nm, at least about 830nm, at least about 835nm, at least about 840nm, at least about

845nm, at least about 850nm, at least about 875nm, at least about 900nm, at least about

925nm, at least about 950nm, or at least about 1000nm.

In any embodiment, the light source has a wavelength of about 500nm, about 525nm, about 550nm, about 575nm, about 600nm, about 625nm, about 650nm, about 675nm, about 700nm, about 725nm, about 730nm, about 735nm, about 740nm, about 745nm, about 750nm, about 755nm, about 760nm, about 765nm, about 770nm, about 775nm, about 780nm, about 785nm, about 790nm, about 795nm, about 800nm, about 805nm, about 810nm, about 815nm, about 820nm, about 825nm, about 830nm, about 835nm, about 840nm, about 845nm, about 850nm, about 875nm, about 900nm, about 925nm, about 950nm, or about 1000nm.

In any embodiment, the light source has a wavelength of 500nm, 525nm, 550nm, 575nm,

600nm, 625nm, 650nm, 675nm, 700nm, 725nm, 730nm, 735nm, 740nm, 745nm, 750nm,

755nm, 760nm, 765nm, 770nm, 775nm, 780nm, 785nm, 790nm, 795nm, 800nm, 805nm,

810nm, 815nm, 820nm, 825nm, 830nm, 835nm, 840nm, 845nm, 850nm, 875nm, 900nm,

925nm, 950nm, or 1000nm.

In a preferred embodiment, the light source has a wavelength of one or more of about 532nm, about 785nm, and about 993nm. Typically, the light source has a wavelength of about 785nm.

In a preferred embodiment, the light source has a wavelength of one or more of 532nm, 785nm, and 993nm. Typically, the light source has a wavelength of 785nm.

Preferably, the wavelength is in the visible spectrum.

In some embodiments, a Raman spectrum comprises spectral signal in the visible, near infrared, infrared, near ultraviolet, or ultraviolet (UV) range.

In any embodiment, the spectra may comprise measurements of inelastic scattering or Raman shift in a Raman Shift range from about 5000cm-¹ to about 0cm ¹ or about 3500cm ¹ to about 0cm ¹. In any embodiment, the spectra may comprise measurements of inelastic scattering or Raman shift in a Raman Shift range from 5000cm ¹ to 0cm ¹ or 3500cm ¹ to 0cm ¹.

In any embodiment, the spectra comprise measurements of inelastic scattering or Raman shift from about 3200cm ¹ to about 400cm ¹, preferably from about 1900cm ¹ to about 400cm ¹. In yet another embodiment, the spectra comprise measurements of inelastic scattering or Raman shift from about 1800cm ¹ to about 600cm ¹. In another embodiment, the spectra comprise measurements of inelastic scattering or Raman shift from about 1725cm ¹ to about 475cm ¹. In any embodiment, the spectra comprise measurements of inelastic scattering or Raman shift from 3200cm ¹ to 400cm ¹, preferably from 1900cm ¹ to 400cm ¹. In yet another embodiment, the spectra comprise measurements of inelastic scattering or Raman shift from 1800cm ¹ to 600cm ¹. In another embodiment, the spectra comprise measurements of inelastic scattering or Raman shift from 1725cnr¹ to 475cm ¹.

In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from about 3500cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 350cm ¹. In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman Shift range from about 3100cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 350cm ¹. In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from 3500cm ¹ to 2650cm ¹ and/or from 1800cm ¹ to 350cm ¹. In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman Shift range from 3100cm ¹ to 2650cm ¹ and/or from 1800cm ¹ to 350cm ¹.

In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from about 3320cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 200cm ¹. In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from 3320cm ¹ to 2650cm ¹ and/or from 1800cm ¹ to 200cm ¹.

In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from about 5000cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 0cm ¹. In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from 5000cm ¹ to 2650cm ¹ and/or from 1800cm ¹ to 0cm ¹.

In embodiments where methods of the invention relate to determination of concentration of alcohols, such as ethanol, the spectra comprise measurements of inelastic scattering or Raman shift from about 1455 cm ¹ to about 830 cm ¹ and/or about 1100 cm ¹ to about 1000 cm ¹. In another embodiment, the spectra may comprise measurements of inelastic scattering or Raman shift from 1455 cm ¹ to 830 cm ¹ and/or 1100 cm ¹ to 1000 cm ¹. In another embodiment, the spectra may comprise measurements of inelastic scattering or Raman shift from about 1100 cm ¹ to about 1000 cm ¹, at about 900 cm ¹ and at about 1455 cm ¹. In some embodiments where methods of the invention relate to determination of concentration of alcohols, such as ethanol, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from about 3500 cnr¹ to about 2650 cm ¹, and/or from about 1800 cm ¹ to about 350 cnr¹. In another embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from about 3320 cm ¹ to about 2650 cm ¹ , and/or from about 1800 cm ¹ to about 200 cm ¹. In another embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift in a Raman shift range from 3320 cnr¹ to 2650 cnr¹, and/or from 1800 cm ¹ to 200 cnr¹.

In a preferred embodiment, after data pretreatments, a data filter may be used to exclude wavelength ranges or Raman shift regions that are not of interest. In a preferred embodiment, after data pretreatments, a data filter is used to select wavelength ranges or Raman shift regions that are of interest, preferably wherein the Raman shift ranges comprise from about 3500 cm ¹ to about 2650 cm ¹, from about 3100 cm ¹ to about 2650 cm ¹, from about 1800 erm ¹ to about 1540 cnr¹, from about 1500 cm ¹ to about 1410 cnr¹, from about 1110 cm ¹ to about 1020 cm ¹, and/or from about 910 cm ¹ to about 830 cm ¹. Preferably wherein the Raman shift ranges comprise from 3500 cm ¹ to 2650 cm ¹, from 3100 cm ¹ to 2650 cm ¹, from 1800 cm ¹ to 1540 cnr¹, from 1500 cm ¹ to 1410 cnr¹, from 1110 cm ¹ to 1020 cnr¹, and/or from 910 cm ¹ to 830 cnr¹.

In embodiments where methods of the invention relate to determination of concentration of protein, such as immunoglobulin (Ig) preferably IgG, the spectra comprise measurements of inelastic scattering or Raman shift between about 475 - about 770 cm ¹, about 920 - about 970 cnr¹, about 990 - about 1010 cm ¹, about 1150 - about 1370 cnr¹, and/or about 1550 - about 1725 cnr¹. In this embodiment, typically the spectra comprise measurements of inelastic scattering or Raman shift between 475 - 770 cnr¹, 920 -970 cm ¹, 990 - 1010 cnr¹, 1150 - 1370 cm ¹ , and/or 1550 - 1725 cm ¹.

In embodiments where methods of the invention relate to determination of concentration of protein, such as immunoglobulin (Ig), preferably IgG, and/or albumin, the spectra may also comprise measurements of inelastic scattering or Raman shift from about 1800 cm ¹ to about 350 cnr¹. In a preferred embodiment, a data filter is used to select inelastic scattering or Raman shift of specific wavelength ranges or specific Raman shift regions, preferably wherein the ranges comprise from about 1660 cm ¹ to 1500 cnr¹, from about 1410 cm ¹ to about 1110 cm- ¹, from about 1020 cnr¹ to about 910 cm ¹, and/or from about 830 cm ¹ to about 480 cnr¹. Preferably wherein the ranges comprise from 1660 cm ¹ to 1500 cm ¹, from 1410 cm ¹ to 1110 cm ¹, from 1020 cm ¹ to 910 cm ¹, and/or from 830 cm ¹ to 480 cnr¹. In another preferred embodiment, Raman shift regions are selected wherein the ranges comprise from about 1800 cnr¹ to about 1510 cm ¹, from about 1400 cm ¹ to about 1310 cm ¹, from about 1230 cm ¹ to about 1110 cm ¹ , from about 1020 cm ¹ to about 915 cm ¹ , and/or from about 830 cm ¹ to about 480 cm ¹. In another preferred embodiment, Raman shift regions are selected wherein the ranges comprise from 1800 cm ¹ to 1510 cm ¹, from 1400 cm ¹ to 1310 cm ¹, from 1230 cm ¹ to 1110 cm ¹, from 1020 cm ¹ to 915 cm ¹, and/or from 830 cm ¹ to 480 cnr¹. In another preferred embodiment, the Raman Shift region is only selected wherein the range comprise from about 1020 cnr¹ to about 980 cm ¹. In another preferred embodiment, the Raman shift region is selected wherein the range comprises from 1020 cnr¹ to 980 cm ¹.

In some embodiments, range(s) of the data filter for selecting Raman shift region(s) used for model building may be determined by a Variable Importance in Projection (VIP) plot. The VIP plot outlines the area of the spectra which are important for the prediction of the concentration of an analyte. Accordingly, region(s) with a low VIP may not be considered by the data filter. In some embodiments, although specific Raman Shift regions are displayed in the VIP plot as being important for the prediction of the concentration of an analyte, those regions may be manually excluded due to being influenced by the presence of substances other than the analyte of interest.

In any embodiment, the spectra may also comprise measurements of inelastic scattering or Raman shift at any wavelength or Raman shift region or from and to the boundaries of any wavelength or Raman shift range described in the Examples.

In any aspect, the model of the processed reference spectra is a model generated using, for example, Peak Integration (PI), Hard Modelling, Partial Least Squares (PLS) regression or other multivariate statistics of processed spectra of samples having known concentrations of the analyte are generated using a method described herein.

In any aspect, the model of the processed reference spectra is a model generated using, for example, Peak Integration of processed spectra of samples having known concentrations of the analyte. In some embodiments where methods of the invention relate to determination of concentration of alcohols, such as ethanol, the peak(s) of inelastic scattering or Raman shift for integration may be at about 880 cm ¹ , at about 1046 cm ¹ , at about 1086 cm ¹ , or any combination thereof. In some embodiments where the method of the invention relates to determination of concentration of alcohols, such as ethanol, the peak(s) of inelastic scattering or Raman shift for integration may be at about 437 cm ¹ , at about 879 cm ¹, at about 1046 erm ¹ , at about 1086 erm¹, at about 1279 erm¹, at about 1455 erm¹, at about 1483 erm¹, at about 2934 erm¹ or any combination thereof. In some embodiments where the method of the invention relates to determination of concentration of alcohols, such as ethanol, the peak(s) of inelastic scattering or Raman shift for integration may be at 437 erm¹ , at 879 erm¹ , at 1046 erm¹ , at 1086 erm¹, at 1279 erm¹, at 1455 erm¹, at 1483 erm¹, at 2934 erm¹ or any combination thereof. In embodiments where methods of the invention relate to determination of concentration of total protein, the peak of inelastic scattering or Raman shift for integration may be at about 1003 erm¹, or 1003 erm¹.

In another aspect, the present invention provides a method for generating a model to determine the concentration of an analyte in a sample obtained from plasma processing, the method comprising: providing training samples obtained from processing of blood-derived plasma, wherein the samples have known concentrations of the analyte, applying a light source to the training samples, measuring inelastic scattering from the training samples, thereby generating training spectra, selecting inelastic scattering or Raman shift regions of interest in the training spectra; optionally applying at least one spectral pre-treatment; generating a model by applying Peak Integration (PI), Hard Modelling, Partial Least Squares (PLS) regression or other multivariate analysis to the spectra to provide a correlation with known concentration of the analyte, thereby obtaining a model for determining the concentration of an analyte in a sample obtained from processing of blood-derived plasma. Optionally the multivariate analysis is selected from Partial least squares (PLS) regression; peak integration (PI); Hard Modelling; PLS Discriminant Analysis (PLS-DA); Ordinary Least Squares (OLS) regression; MLR (multiple linear regression); OPLS (Orthogonal-PLS); SVM (support vector machines); GLD (general discriminant analysis); GLMC (generalized linear model); GLZ (generalized linear and nonlinear model); LDA (Linear Discriminant Analysis); classification trees; cluster analysis; neural networks; and Pearson correlation. An inelastic scattering or Raman shift region(s) of interest may be any wavelength, wavelength range or Raman shift range described herein (including those in the Examples), such as those described herein to be used to determine the presence of, or concentration of, protein(s) or alcohols (e.g. ethanol).

In any aspect, the training samples are obtained from routine manufacture of blood-derived plasma products and/or experimental laboratory studies, for example, that have been scaled down from processes at commercial scale, as further described herein and such as immunoglobulins, and other proteins derived from blood plasma including albumin and clotting factors.

In any aspect, the spectral pre-treatment of Raman spectra generated is the selection of Raman shift regions of interest, baseline correction, baseline nodes (rubber band subtraction), 1^st order derivative, 2^nd order derivative, vector normalization, smoothing, standardization, Standard Normal Variate (SNV), or a combination of both 1^st order derivative and vector normalization, or a combination of any of 1^st order derivative, 2^nd order derivative, vector normalization, smoothing, standardization and Standard Normal Variate (SNV). Alternatively, the spectral pre-treatment is min-max normalisation. Optionally wherein the standardization is performed by area normalization.

In any aspect, the type and number of pre-treatment applied are as outlined in the Examples.

In any aspect, the spectral pre-treatment may comprise smoothing with a filter, preferably smoothing with a filter length of 5, 7, 9, 11 , 13, 15, 17, 19, 21 , 23, 25, 27, 29, 31 , 33, 35, 37, 39, 41 , 43, 45, 47, 49, 51 , 53, 55, 57, or 59. The higher the number, the more the smoothing effect has.

In any aspect, the spectral pre-treatment may comprise a standardisation. In some embodiments, the standardisation may be performed by the area normalization approach, for example, by using pre-selected Raman Shift regions of interest, for example in case the sample matrix is changing overtime (e.g. during ethanol precipitation). In any other aspect, the Raman intensities may be normalised based on the peak area of the probe peak from 380 cm- ¹ to 420 cnr¹ with a linear fit baseline (e.g. to compare the Raman results when different Raman equipment is used). Due to standardization, spectra measured with different exposure settings can be compared with each other.

In any aspect, the method further comprises a step of identifying inelastic scattering or Raman shift of the buffer or background solution in which the analyte of interest in present and either disregarding that inelastic scattering or Raman shift, or identifying inelastic scattering or Raman shift of the analyte of interest that does not overlap with or affected by the inelastic scattering or Raman shift of the buffer or background solution.

In any aspect, where the analyte is protein, the concentration of protein in reference or training samples may be determined using any means known in the art, for example the Dumas assay, or the immunoturbidimetric assay to quantify specific proteins such as albumin and/or IgG, or any means described herein including the Examples.

In any aspect, where the analyte is an alcohol such as ethanol, the concentration of alcohol (e.g. ethanol) in the reference or training samples may be determined using any means known in the art, or using theoretical values, or any means described herein (e.g. gas chromatography or enzymatic ethanol determination) including the Examples.

In any aspect, the methods of the invention allow determination of protein concentration of a range of about 0 g/kg to about 10 g/kg, 0 g/kg to 10 g/kg, of about 10 g/kg to about 150 g/kg, 10 g/kg to 150 g/kg, about 15 g/kg to about 45 g/kg, 15 g/kg to 45 g/kg, about 20 g/kg to about 35 g/kg, 20 g/kg to 35 g/kg, about 100 g/kg to about 150 g/kg, 100 g/kg to 150 g/kg, about 150 g/kg to about 300 g/kg, or 150 g/kg to 300 g/kg. In one embodiment, where the protein in the test sample is predominantly, or contains a significant amount of, IgG the protein concentration range may be about 0 g/kg to about 15 g/kg, 0 g/kg to 15 g/kg, about 15 g/kg to about 40 g/kg, 15 g/kg to 40 g/kg, about 16 g/kg to about 42g/kg, 16 g/kg to 42 g/kg, about 20 g/kg to about 35 g/kg, or 20 g/kg to 35 g/kg. In one embodiment, where the protein in the test sample is predominantly, or contains a significant amount of, albumin, the protein concentration range may be about 0 g/kg to about 100 g/kg, 0 g/kg to 100 g/kg, about 100 g/kg to about 150 g/kg, 100 g/kg to 150 g/kg, about 150 g/kg to about 300 g/kg, or 150 g/kg to 300 g/kg. In any aspect, the methods of the invention allow determination of alcohol (e.g. ethanol) concentration of a range of about 1 % v/v to about 65% v/v, or 1 % v/v to 65% v/v, or about 8% to about 44% v/v, or 8% to 44% v/v.

In any aspect, the methods of the invention allow determination of total protein, a specific protein present in human plasma (such as IgG or albumin), or ethanol concentration of a range typically used during the fractionation of blood plasma, including to produce any, or all of, Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV (including Cohn Fraction IVi, IV₄), and Cohn Fraction V and other similar variant fractions or precipitates. Further, in any aspect, the methods of the invention allow determination of total protein, a specific protein present in human plasma (such as IgG or albumin), acetate or ethanol concentration of a range typically used during the fractionation of blood plasma, including to produce any, or all of, Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Precipitate B, Kistler/Nitschmann Fraction IV, and Kistler Nitschmann Precipitate C and other similar variant fractions or precipitates.

As used herein, Cohn Fraction (l+)ll+lll includes Cohn Fraction l+ll+lll or Cohn Fraction ll+lll. It is also equivalent to Kistler/Nitschmann Precipitate A and other similar variant fractions or precipitates.

As used herein, Cohn Fraction IV includes Cohn Fraction IVi and IV₄

In any aspect, the methods of the invention allow determination of total protein, a specific protein present in human plasma (such as albumin), or ethanol concentration of a range typically used during the fractionation of blood plasma to produce Cohn Fraction V. Further, in any aspect, the methods of the invention allow determination of ethanol concentration of a range typically used during the fractionation of blood plasma to produce either, or both of, Kistler/Nitschmann Precipitate C.

In any aspect, the methods of the invention allow determination of total protein, a specific protein present in human plasma (such as IgG or albumin), or ethanol concentration of a range typically used during the dilution or resuspension of plasma fractions including any, or all of, Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV (including Cohn Fraction IVi, IV₄), and Cohn Fraction V and other similar variant fractions or precipitates. Further, in any aspect, the methods of the invention allow determination of total protein, a specific protein present in human plasma (such as IgG or albumin), or ethanol concentration of a range typically used during the dilution or resuspension of plasma fractions including any, or all of, Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Precipitate B, Kistler/Nitschmann Fraction IV, and Kistler Nitschmann Precipitate C and other similar variant fractions or precipitates.

In an embodiment, the total protein, a specific protein present in human plasma (such as IgG), or ethanol concentration is measured during resuspension of any, or all of, Cohn Fraction I, Cohn Fraction ll+lll, Cohn Fraction l+ll+lll, or Kistler/Nitschmann Precipitate A or other similar variant fractions or precipitates.

In an embodiment, the total protein, a specific protein present in human plasma (such as albumin or IgG) or ethanol concentration is measured during resuspension of Cohn Fraction IV paste (including Cohn Fraction IVi , IV₄ or other similar variant fraction or precipitate).

In an embodiment, the total protein, a specific protein present in human plasma (such as IgG) or ethanol concentration is measured after resuspension of any, or all of, Cohn Fraction I, Cohn Fraction ll+lll, Cohn Fraction l+ll+lll, or Kistler/Nitschmann Precipitate A paste and priorto any filtration (e.g. clarifying filtration) of the resuspended paste or any significant reduction in turbidity of the resuspended paste.

In an embodiment, the total protein, a specific protein present in human plasma (such as albumin or IgG) or ethanol concentration is measured after resuspension of Cohn Fraction IV paste (including Cohn Fraction IVi , IV₄ or other similar variant fraction or precipitate) and prior to any filtration (e.g. clarifying filtration) of the resuspended paste or any significant reduction in turbidity of the resuspended paste.

In an embodiment, Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV paste (including Cohn Fraction IVi, IV₄ or other similar fraction or precipitate), Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Fraction IV or Kistler/Nitschmann Precipitate B, or other similar fraction or precipitate, paste is resuspended by the addition of one or more diluting agents, such as distilled water. Typically, the paste is resuspended by the addition of one or more diluting agents at a ratio of dilution agent between 1-7 x the weight of the Precipitate paste. In an embodiment, the paste is resuspended at a temperature below 26°C, including 25°C, 24°C, 23°_c 22°C, 21 °C, 20°C, 19°C, 18°C, 17°C, 16°C, 15°C, 14°C, 13°C, 12°C, 11 °C, 10°C, 9°C, 8°C, 7°C, 6°C, 5°C, 4°C, 3°C, 2°C, 1 °C, 0°C, -1 °C, -2°C, -3°C, -4°C, -5°C, -6°C, -7°C, -8°C, - 9°C, -10°C, -11 °C, or -12°C. In an embodiment the resuspension temperature is <21 °C.

In an embodiment, the ethanol concentration in the resuspended Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV paste (including Cohn Fraction IVi, IV₄ or other similar fraction or precipitate), Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Fraction IV or Kistler/Nitschmann Precipitate B, or other similar fractions or precipitates, paste is between the range of about 2% (w/w) to about 30% (w/w), about 2% (w/w) to about 20% (w/w), about 5% (w/w) to about 30% (w/w), about 5% (w/w) to about 20% (w/w), about 5% (w/w) to about 15% (w/w), or about 5% (w/w) to about 10% (w/w).

In an embodiment, the total protein concentration in the resuspended Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV paste (including Cohn Fraction IVi, IV₄ or other similar fraction or precipitate), Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Fraction IV or Kistler/Nitschmann Precipitate B, or other similar fractions or precipitates, paste is between the range of about 5% (w/w) to about 15% (w/w), 5% (w/w) to 15% (w/w), about 5% (w/w) to about 30% (w/w) or 5% (w/w) to 30% (w/w), typically about 10% (w/w) to about 15% (w/w) or 10% (w/w) to 15% (w/w).

In an embodiment, optionally, filter aid is added to the resuspended Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV paste (including Cohn Fraction IVi, IV₄ or other similar fractions or precipitates), Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Fraction IV or Kistler/Nitschmann Precipitate B, or other similar fractions or precipitates, paste prior to any filtration (e.g. clarifying filtration) step or prior to any significant reduction in turbidity of the resuspended paste.

In any aspect, the methods of the invention allow determination of total protein, a specific protein present in human plasma (such as albumin) or ethanol concentration of a range typically used during the dilution or resuspension of Cohn Fraction V. Further, in any aspect, the methods of the invention allow determination of ethanol concentration of a range typically used during the dilution or resuspension of Kistler/Nitschmann Precipitate C. In an embodiment, the total protein, a specific protein present in human plasma (such as albumin), or ethanol concentration is measured during resuspension of Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste.

In an embodiment, the total protein, a specific protein present in human plasma (such as albumin), or ethanol concentration is measured after resuspension of Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste and prior to any filtration (e.g. clarifying filtration) of the resuspended paste or any significant reduction in turbidity of the resuspended paste.

In an embodiment, Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste is resuspended by the addition of one or more diluting agents, such as distilled water. Typically, the Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste is resuspended by the addition of one or more diluting agents at a ratio of dilution agent between 1-3 x, preferably 1- 2 x, the weight of the Precipitate paste. In an embodiment, the Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste is resuspended at a temperature below 26°C, preferably at or below 25°C, at or below 24°C, at or below 23°C, at or below 22°C, at or below 21 °C, at or below 20°C, at or below 19°C, at or below 18°C, at or below 17°C, at or below 16°C, at or below 15°C, at or below 14°C, at or below 13°C, at or below 12°C, at or below 11 °C, at or below 10°C, at or below 9°C, at or below 8°C, at or below 7°C, at or below 6°C, at or below 5°C, at or below 4°C, at or below 3°C, at or below 2°C, at or below 1 °C or 0°C. In an embodiment the resuspension temperature is <21 °C.

In an embodiment, the ethanol concentration in the resuspended Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste is between the range of about 5% (w/w) to about 15% (w/w), 5% (w/w) to 15% (w/w), about 5% (w/w) to about 30% (w/w) or 5% (w/w) to 30% (w/w), typically about 5% (w/w) to about 10% (w/w) or 5% (w/w) to 10% (w/w).

In an embodiment, the total protein concentration in the resuspended Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste is between the range of about 5% (w/w) to about 15% (w/w), 5% (w/w) to 15% (w/w), about 5% (w/w) to about 30% (w/w) or 5% (w/w) to 30% (w/w), typically about 10% (w/w) to about 15% (w/w) or 10% (w/w) to 15% (w/w). In an embodiment, filter aid is added to the resuspended Cohn Fraction paste or Kistler/Nitschmann Precipitate C paste prior to any filtration (e.g. clarifying filtration) step or prior to any significant reduction in turbidity of the resuspended paste.

In any aspect, the training samples include concentrations of analyte across the concentration range for test sample determination. For example, if the possible concentration of an analyte in a test sample is within a range of X g/kg to Y g/kg, then the training samples include concentrations of analyte at, and between, X g/kg to Y g/kg.

In any aspect, the training samples and the test sample are exposed to a light source at a temperature in the range of about -8°C to about 37°C or -8°C to 37°C. Typically the temperature is in the range of about 10°C to about 37°C, preferably in the range of about 15°C to about 30°C. The temperature may be about 15°C, about 16°C, about 17°C, about 18°C, about 19°C, about 20°C, about 21 °C, about 22°C, about 23°C, about 24°C, about 25°C, about 26°C, about 27°C, about 28°C, about 29°C, or about 30°C. In any embodiment, the temperature is 18°C, 19°C, 20°C, 2°C, 22°C, 23°C, or 24°C.

In any aspect, the light source is applied to the training samples and/or the test sample using a probe adapted to emit light having wavelengths in the relevant range. Optionally the probe is configured for inclusion in an industrial protein mixing, filtration or purification apparatus, including for use for in-line measurement of inelastic scattering from the training samples or test samples.

In any aspect, the light source is applied to the training samples and/or the test sample during mixing of the samples. The light source may be applied to the sample(s) at an angle that is parallel to the direction of fluid stream during mixing of the sample(s). Alternatively, the light source may be applied to the sample(s) at an angle that is non-parallel to the direction of fluid stream during mixing of the sample(s), for example the light source may be applied to the sample(s) at, or about, 45° to the direction of the fluid stream during mixing of the sample(s).

In any aspect, the quality of the model generated may be judged using the following statistical parameters:

Number of latent variables (PLS factors) in the model, • Bias,

• RMSEC

• RMSECV,

• RMSEP for independent test samples,

• Rank,

• R²,

• RPD value,

• Uncertainty, and/or

• MAPE.

In any aspect, the sample comprising the analyte is obtained from processing of blood-derived plasma including any plasma sample derived from blood, preferably human blood. In certain embodiments, the sample is obtained or derived from the processing of blood-derived plasma that comprises fresh plasma, cryo-poor plasma, cryo-precipitate, or cryo-rich plasma. In other words, the source of plasma may be blood, preferably human blood, preferably fresh plasma, cryo-poor plasma, cryo-precipitate, or cryo-rich plasma. The plasma may be obtained from a number of donations and/or subjects, and pooled. The plasma may be hyperimmune plasma.

In any aspect, the sample comprising the analyte is a resuspension of a precipitate or paste obtained from blood-derived plasma and as further described herein.

In any aspect, the sample comprising the analyte is a filtrate obtained from blood-derived plasma and as further described herein. For example, the filtrate may have been obtained from separation of a precipitate, including any precipitate described herein.

In any aspect, the sample contains octanoic acid and/or other precipitants. Therefore, the octanoic acid containing sample also contains blood derived plasma or is obtained or derived from the processing of blood-derived plasma.

In any aspect of the present invention, the sample comprising the analyte is a blood-plasma fraction (intermediate). In particular embodiments the fraction is a Cohn Fraction. In a particularly preferred embodiment the plasma fraction is selected from the group consisting of Cohn Fraction I (Fr I), Cohn Fraction (l+)ll+lll (such as Cohn Fraction ll+lll (Fr ll+lll), and Cohn Fraction l+ll+lll (Fr l+ll+lll)), Cohn Fraction II (Fr II), Cohn Fraction III (Fr III), Cohn Fraction IV (Fr IV), Cohn Fraction V (Fr V), Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Precipitate B, Kistler/Nitschmann Precipitate C. In another embodiment, the plasma fraction is selected from the group consisting of Cohn Fraction I (Fr I), Cohn Fraction (l+)ll+lll (such as Cohn Fraction ll+lll (Fr ll+lll), and Cohn Fraction l+ll+lll (Fr l+ll+lll)), or Kistler/Nitschmann Precipitate A (KN A, PPT A or Fr A). The plasma fraction may be a combination of different fractions. For example, the plasma fraction may be a combination of KN A and one or more of Fr I, Fr ll+lll and Fr l+ll+lll. In particular embodiments, the fraction may be an albumin enriched fraction, for example the fraction has been treated to deplete components such IgG.

In any aspect, the sample comprising the analyte may comprise filter aid (for example, diatomaceous earth and perlite; or cellulose or silica gel).

In any aspect, the sample comprising the analyte is a turbid solution or suspension. In any embodiment, the turbid solution or suspension may have Nephelometric Turbidity Units (NTU) and/or Formazine Turbidity Units (FTU) of equal to or greater than 0 NTU, equal to or greater than 10 NTU, equal to or greater than 15 NTU, equal to or greater than 20 NTU, equal to or greater than 25 NTU, equal to or greater than 30 NTU, equal to or greater than 35 NTU, equal to or greater than 40 NTU, equal to or greater than 45 NTU, equal to or greater than 50 NTU, equal to or greater than 55 NTU, equal to or greater than 60 NTU, equal to or greater than 65 NTU, equal to or greater than 70 NTU, equal to or greater than 75 NTU, equal to or greater than 80 NTU, equal to or greater than 85 NTU, equal to or greater than 90 NTU, equal to or greater than 95 NTU, equal to or greater than 100 NTU, equal to or greater than 150 NTU, equal to or greater than 200 NTU, equal to or greater than 250 NTU, equal to or greater than 300 NTU, equal to or greater than 350 NTU, equal to or greater than 400 NTU, equal to or greater than 450 NTU, equal to or greater than 500 NTU, equal to or greater than 550 NTU, equal to or greater than 600 NTU, equal to or greater than 650 NTU, equal to or greater than 700 NTU, equal to or greater than 750 NTU, equal to or greater than 800 NTU, equal to or greater than 850 NTU, equal to or greater than 900 NTU, equal to or greater than 950 NTU, equal to or greater than 1 ,000 NTU, equal to or greater than 1 ,500 NTU, equal to or greater than 2,000 NTU, equal to or greater than 2,500 NTU, equal to or greater than 3,000 NTU, equal to or greater than 3,500 NTU, equal to or greater than 4,000 NTU, equal to or greater than 4,500 NTU, equal to or greater than 5,000 NTU, equal to or greater than 5,500 NTU, equal to or greater than 6,000 NTU, equal to or greater than 6,500 NTU, equal to or greater than 7,000 NTU, equal to or greater than 7,500 NTU, equal to or greater than 8,000 NTU, equal to or greater than 8,500 NTU, equal to or greater than 9,000 NTU, equal to or greater than 9,500 NTU, or equal to or greater than 10,000 NTU. In any embodiment, the turbid solution or suspension may have NTU of 10 NTU to 100 NTU, 10 NTU to 90 NTU, 10 NTU to 80 NTU, 10 NTU to 70 NTU, 10 NTU to 60 NTU, 10 NTU to 50 NTU, 10 NTU to 40 NTU, 10 NTU to 30 NTU, 10 NTU to 20 NTU, 20 NTU to 100 NTU, 30 NTU to 100 NTU, 40 NTU to 100 NTU, 50 NTU to 100 NTU, 60 NTU to 100 NTU, 70 NTU to 100 NTU, 80 NTU to 100 NTU, or 90 NTU to 100 NTU. In any embodiment, the turbid solution may have a maximum NTU of 10,000 NTU, 9,500 NTU, 9,000 NTU, 8,500 NTU, 8,000 NTU, 7,500 NTU, 7,000 NTU, 6,500 NTU, 6,000 NTU, 5,500 NTU, 5,000 NTU, 4,500 NTU, 4,000 NTU, 3,500 NTU, 3,000 NTU, 2,500 NTU, 2,000 NTU, 1 ,500 NTU, 1000 NTU, 950 NTU, 900 NTU, 850 NTU, 800 NTU, 750 NTU, 700 NTU, 650 NTU, 600 NTU, 550 NTU, 500 NTU, 450 NTU, 400 NTU, 350 NTU, 300 NTU, 250 NTU, 200 NTU, 150 NTU, 100 NTU or 50 NTU.

In any aspect, the sample comprising the analyte is not a turbid solution or suspension. In any embodiment, the sample or the solution or suspension from which the sample is taken may have Nephelometric Turbidity Units (NTU) and/or Formazine Turbidity Units (FTU) of less than 10 NTU, equal to or less than about 9 NTU, equal to or less than about 8 NTU, equal to or less than about 7 NTU, equal to or less than about 6 NTU, equal to or less than about 5 NTU, equal to or less than about 4 NTU, equal to or less than about 3 NTU, equal to or less than about 2 NTU, or equal to or less than about 1 NTU. In any embodiment, the solution or suspension may have NTU of less than 10 NTU to about 0.1 NTU, about 9 NTU to about 0.1 NTU, about

8 NTU to about 0.1 NTU, about 7 NTU to about 0.1 NTU, about 6 NTU to about 0.1 NTU, about

5 NTU to about 0.1 NTU, about 4 NTU to about 0.1 NTU, about 3 NTU to about 0.1 NTU, about

2 NTU to about 0.1 NTU, about 1 NTU to about 0.1 NTU, about 9 NTU to about 0.1 NTU, about

9 NTU to about 0.2 NTU, about 9 NTU to about 0.3 NTU, about 9 NTU to about 0.4 NTU, about

9 NTU to about 0.5 NTU, about 9 NTU to about 0.6 NTU, about 9 NTU to about 0.7 NTU, about 9 NTU to about 0.8 NTU, about 9 NTU to about 0.9 NTU, about 9 NTU to about 1 NTU, about 9 NTU to about 2 NTU, about 9 NTU to about 3 NTU, about 9 NTU to about 4 NTU, about 9 NTU to about 5 NTU, about 9 NTU to about 6 NTU, about 9 NTU to about 7 NTU, about 9 NTU to about 8 NTU, about 1 NTU to about 5 NTU, about 1 NTU to about 4 NTU, about 1 NTU to about 3 NTU, or about 1 NTU to about 2 NTU. In any embodiment, the solution or suspension may have Nephelometric Turbidity Units (NTU) of less than 10 NTU, equal to or less than 9 NTU, equal to or less than 8 NTU, equal to or less than 7 NTU, equal to or less than 6 NTU, equal to or less than 5 NTU, equal to or less than 4 NTU, equal to or less than 3 NTU, equal to or less than 2 NTU, or equal to or less than 1 NTU. In any embodiment, the solution or suspension may have NTU of less than 10 NTU to 0.1 NTU, 9 NTU to 0.1 NTU, 8 NTU to 0.1 NTU, 7 NTU to 0.1 NTU, 6 NTU to 0.1 NTU, 5 NTU to 0.1 NTU, 4 NTU to 0.1 NTU, 3 NTU to 0.1 NTU, 2 NTU to 0.1 NTU, 1 NTU to 0.1 NTU, 9 NTU to 0.1 NTU, 9 NTU to 0.2 NTU, 9 NTU to 0.3 NTU, 9 NTU to 0.4 NTU, 9 NTU to 0.5 NTU, 9 NTU to 0.6 NTU, 9 NTU to 0.7 NTU, 9 NTU to 0.8 NTU, 9 NTU to 0.9 NTU, 9 NTU to 1 NTU, 9 NTU to 2 NTU, 9 NTU to 3 NTU, 9 NTU to 4 NTU, 9 NTU to 5 NTU, 9 NTU to 6 NTU, 9 NTU to 7 NTU, 9 NTU to 8 NTU, 1 NTU to 5 NTU, 1 NTU to 4 NTU, 1 NTU to 3 NTU, or 1 NTU to 2 NTU.

Any references herein to a particular NTU value or range is intended to provide basis for a FTU value or range.

One example of a solution that is not turbid is a solution in plasma processing prior to addition of an alcohol (e.g. ethanol) for the purpose of protein precipitation.

The present invention allows use during in-line monitoring including monitoring an analyte, such as an alcohol (e.g. ethanol) or, total or specific protein (e.g. IgG or albumin), in a solution that has low turbidity but during the processing becomes a higher, or highly, turbid solution. Further, the in-line monitoring also allows similar monitoring of solutions that are highly turbid but during processing become low, or have lower, turbidity.

In any aspect or embodiment, any or all steps of the method are performed in-line, at-line, offline and/or on-line.

Those skilled in the field will understand and appreciate that plasma fractionation processes have some adaptability and have been optimized and varied over the years, for example, to suit different manufacturers and different product profile goals. An example of such a modification is the presence or absence of Cohn fractionation step IV- 1 , which can be used to extract alpha-1 -antitrypsin. Thus, it should be understood that the methods and products described herein can be practiced with modifications and variations of human plasma fractionation processes, and that such modifications and variations are included within the scope of this disclosure. In a preferred embodiment, the training samples comprise a representative set of samples that cover variables, such as different paste type, sample temperature, instrument variability, operator handling, raw materials, and plasma source.

As used herein, except where the context requires otherwise, the term "comprise" and variations of the term, such as "comprising", "comprises" and "comprised", are not intended to exclude further additives, components, integers or steps.

Further aspects of the present invention and further embodiments of the aspects described in the preceding paragraphs will become apparent from the following description, given by way of example and with reference to the accompanying drawings.

Brief description of the drawings

Figure 1 : Raman spectra before and after precipitation in comparison

The Raman intensity can be seen as a function of the Raman shift [cm ¹]. The spectrum before the precipitation is shown in dark grey and the spectrum after precipitation is shown in light grey. It can be seen that a good distinction between the two spectra is possible and that some peaks change.

Figure 2: Overview of experimental design and execution

Three of the four experiments use frozen cryopoor plasma (CPP) as starting material. The samples generated from this are denoted as training samples and used for model building and training. Another experiment using fresh CPP as starting material was used to verify the model. The samples generated are designated as test samples.

Figure 3: IgG concentration as a function of ethanol concentration

A reproducible descending IgG concentration with increasing alcohol concentration can be observed (not linear) throughout the precipitation process.

Figure 4: Predicted vs true plot for ethanol

The upper graph shows that the samples (both training samples and test samples) are lying close to the predicted identity line. The lower graph plots the differences of the mean value of the test set (dashed line) against the mean value of the calibration (training samples, grey line at 0) as it progresses with the individual samples.

Figure 5: Predicted vs true plot for IgG

The upper graph shows that the samples (both training samples and test/validation samples) are closely distributed around the predicted identity line. An implied clot of values in the upper region and a slight gap in the middle of the graph are due to ethanol concentration-dependent IgG precipitation behavior. The lower graph shows the differences between the samples based on the mean ethanol concentration of the training samples (grey line at 0) and test/ validation samples (dashed line).

Figure 6: As an example, change in Raman spectrum observed at different time points during the ethanol mediated precipitation of cryopoor plasma in a Raman Shift range of 0 to 3500 cm'¹

Raman spectra have been obtained at different time points during the ethanol precipitation process of cryopoor plasma and overlayed to highlight the change in Raman intensity overtime. The precipitation process can be monitored based on the change in Raman intensity in a Raman Shift range of 0 to 3500 cnr¹.

Figure 7: As an example, change in Raman spectrum observed at different time points during the ethanol mediated precipitation of cryopoor plasma in a Raman Shift range of 300 to 1800 cm'¹ that was used for further evaluation

Raman spectra have been obtained at different time points during the ethanol precipitation process of cryopoor plasma and overlayed to highlight the change in Raman intensity overtime. The precipitation process can be monitored based on the change in Raman intensity in a Raman Shift range of 300 to 1800 cnr¹. As an example, the spectra generated in this Raman Shift range were used for further evaluation of this specific ethanol precipitation step, as described as follows (see section describing the examples).

Figure 8: As an example, Predicted vs. True plots and Difference vs. True plots of IgG for an optimized IgG model

Figure 9: Predicted vs. True plots and Difference vs. True plots of ethanol (EtOH) for an optimized EtOH model Figure 10: As an example, EtOH assigned peaks/ranges in Raman spectrum

Figure 11 : As an example, protein assigned peaks/ranges in Raman spectrum (0 - 1800 cm^-1)

Figure 12: As an example, EtOH integration peak at a Raman shift of ~879 cm'¹, where Raman intensity correlated linearly with increasing ethanol concentration

Figure 13: As an example, correlation between total protein and the representative protein peak at 1003.22 cm'¹ from (A) pastes containing filter aid and (B) pastes after filter aid removal

Figure 14: As an example, Predicted vs. True plot for Peak Integration model for total protein determination of dissolved pastes containing filter aid

Figure 15: As an example, Predicted vs. True plot for Peak Integration model for total protein determination of dissolved pastes without filter aid

Figure 16: As an example, linear correlation between EtOH concentration and peak area of Raman spectra measured from dissolved pastes (with filter aid) and their respective dissolving buffers

Figure 17: As an example, linear correlation between EtOH concentration and peak area of Raman spectra measured from dissolved pastes (after removal of filter aid) and their respective dissolving buffers

Figure 18: As an example, Predicted vs. True plot - EtOH protein prediction model calibration, dissolved pastes with filter aid

Figure 19: As an example, Predicted vs. True plot - EtOH prediction model calibration, dissolved pastes after filter aid removal at 1045.9 cm'¹ Figure 20: As an example, Predicted vs. True plot - EtOH model calibration, dissolved pastes after filter aid removal using peak at 1085.81 cm'¹

Figure 21 : As an example, correlation between total protein and the representative protein peak at approximately 1003 cm'¹ from (A) pastes containing filter aid and (B) pastes after filter aid removal

Figure 22: As an example, linear correlation between EtOH concentration and peak area of Raman spectra measured from dissolved V pastes (A) (with filter aid), (B) without filter aid and, water for injection (WFI) - ethanol mixtures (C) with filter aid and (D) without filter aid

Figure 23: As an example, Predicted vs. True plot - EtOH prediction model calibration, dissolved pastes after filter aid removal

Figure 24: As an example, Predicted vs. True plot - total protein (TP) prediction model calibration, dissolved pastes after filter aid removal

Figure 25: As an example, Predicted vs. True plot - EtOH prediction model calibration, dissolved pastes with filter aid

Figure 26: As an example, Predicted vs. True plot - total protein (TP) prediction model calibration, dissolved pastes with filter aid

Figure 27: As an example, spectra of both IV precipitation runs after application of preprocessing developed for (l+)ll+lll precipitation spectra

Figure 28: As an example, RMSE vs. Function plot for the calibration of the EtOH Peak Integration model of IV precipitation spectra

Figure 29: As an example, Predicted vs. True plot - EtOH prediction model calibration, IV precipitation Figure 30: As an example, correlation between total protein and the representative protein peak intensity at ~1003 cm'¹ from all sampling points during the IV Precipitation process

Figure 31: As an example, RMSE vs. Rank plot for the calibration of the total protein (TP) PLS model, IV precipitation spectra

Figure 32: As an example, Predicted vs. True plot - total protein (TP) prediction model calibration, IV precipitation

Figure 33: As an example, Variable Importance in Projection (VIP) plot for calibration of PLS model for a total protein model, IV precipitation spectra

Figure 34: As an example, validation of the Peak Integration model used for ethanol prediction

Figure 35: As an example, RMSE vs. Rank plot for calibration of a PLS model for total protein (TP), NC precipitation spectra

Figure 36: As an example, Predicted vs. True plot for calibration of PLS model for total protein, NC precipitation spectra

Detailed description of the embodiments

Reference will now be made in detail to certain embodiments of the invention. While the invention will be described in conjunction with the embodiments, it will be understood that the intention is not to limit the invention to those embodiments. On the contrary, the invention is intended to coverall alternatives, modifications, and equivalents, which may be included within the scope of the present invention as defined by the claims.

One skilled in the art will recognize many methods and materials similar or equivalent to those described herein, which could be used in the practice of the present invention. The present invention is in no way limited to the methods and materials described. It will be understood that the invention disclosed and defined in this specification extends to all alternative combinations of two or more of the individual features mentioned or evident from the text or drawings. All of these different combinations constitute various alternative aspects of the invention.

All of the patents and publications referred to herein are incorporated by reference in their entirety.

For purposes of interpreting this specification, terms used in the singular will also include the plural and vice versa.

Due to the heterogeneity of plasma-derived product solutions and suspensions, the quantification of key chemical components, such as protein (e.g. IgG and albumin) or alcohol (e.g. ethanol), is complex and to date, has only be achieved by use of off-line analytical methods. This can substantially impact on process efficiency.

The present invention seeks to address some of the deficiencies of prior approaches to processing of plasma-derived products by providing in-line systems for determining the concentration of various analytes in complex solutions during blood plasma processing. The methods of the present invention have the advantage of improving downstream efficiency, reduction in waste and/or improve final product yield.

The approach also enables quantification of analytes in various starting materials used during preparation of blood-plasma derived products without prior sample preparation, as is required with current in-line procedures. A further benefit of the methods of the present invention is the ability to monitor progression of product processing (such as resuspension or precipitation) and other reactions in real-time leading to reduction of cycle time. The invention defined herein has been applied to determine analyte concentration during purification of blood-plasma derived proteins.

A particular advantage of the present invention is the ability to quantify analytes in the presence of filter aid that is used during plasma processing. Filter-aid is used to remove precipitated protein(s) (e.g. after cold ethanol precipitation of human plasma) and/or impurities but increase the complexity of a solution. However, the increased complexity of a solution, for example due to filter aid, does not prevent the present methods from quantifying analytes. Definitions

The term “a sample obtained from processing of blood-derived plasma” is intended to refer to any material, especially protein-containing material, derived from the fractionation or processing of blood plasma. The sample may be a suspension or concentrate, eluate, or filtrate of a “protein-comprising precipitate”, wherein the “protein-comprising precipitate” is derived from blood plasma.

The term “protein-comprising precipitate” is intended to refer to any precipitated material containing a protein and derived from blood plasma. This term may refer to plasma, serum, precipitates produced from plasma or serum. Typically, in the context of the present invention, it refers to precipitates from plasma, such as Cohn or Oncley ethanol precipitates, or Kistler- Nitschmann precipitates. For example, the precipitates may be any one of Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV (including Cohn Fraction IVi , IV₄), and Cohn Fraction V and other similar variant fractions or precipitates. Further, the precipitates may be any one of Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Precipitate B, Kistler/Nitschmann Fraction IV, and Kistler Nitschmann Precipitate C and other similar variant fractions or precipitates.

It will be appreciated that the samples (including the test sample) analysed in accordance with the methods described herein, do not need to be “isolated” samples. In other words, the term “sample” is intended to simply indicate a small part or quantity of a larger whole or bulk or the larger whole or bulk itself. The latter one is relevant in terms of an inline measurement approach. The methods of the present invention are therefore intended to include at-line, inline and off-line methods whereby the light source is applied to a small part of a larger bulk solution and where the light source can be applied to the small part of the bulk solution in situ, or to an aliquot of the solution that has been removed (isolated) from the larger bulk.

As used herein, the term “in-line” refers to a method of analysis whereby a probe, or sampling interface or sensor (e.g. for providing a light source) can be placed directly in a process vessel or tubing or in line with a stream of flowing material to conduct the analysis. The process may involve placing a probe in a flow system, precipitation vessel or processing unit. Such process may allow analysis without having to remove the probe or any material or samples from the bulk (ie the sample remains “in situ" for the analysis). In some embodiments, multiple Raman spectra are obtained from different locations within a solution, for example an in-line processing solution. The data from such multiple spectra may be averaged if appropriate.

As used herein “on-line” refers to a method of analysis without having to remove the material or samples from the bulk. However, it may involve separating from the main process line and performing measurements on just a portion of the bulk. This may be accomplished by adding a sampling loop which directs a sample of the bulk material towards the probe or sensor, and whereby the diverted sample may be re-introduced to the process stream, flow or bulk of material, or disposed of, depending on the application.

As used herein, the term “at-line” refers to a method which includes manual sampling followed by discontinuous sample preparation, measurement and evaluation. When measuring at-line, analysis is typically completed at or near the process stream, flow or bulk of material.

As used herein, the term "off-line” refers to a method that involves the most physical difference between the process stream, flow or bulk of material and the analysis of the sample. Similarly to at-line measurement, off-line measurement involves removing an analytical sample from the larger bulk of material. Off-line analysis typically involves taking the sample or sometimes multiple samples to be analysed in a formal lab setting.

In any method of the invention, the applying a light source step may be in-line, at-line, off-line and/or on-line using cuvette measurement.

As used herein, the coefficient of determination “R²” indicates the percentage of variance explained by the prediction model. The higher the coefficient, the better the correlation between the reference data and spectral data.

As used herein, “bias” is the Systematic averaged deviation between the reference values and the predicted values. P-i Xc i ~ Y_C i) ^{c =} IRS predicted value

Bias = — - - ^: - — yc = Reference method value n n = Number of samples As used herein, for “cross validation” or “cross-validation” (also referred to as internal validation), individual leave-out samples (defined by the user) are removed from the calibration or training set. Using the remaining samples, a chemometric model is established and used to predict the previously extracted sample. A comparison of the predicted with the actual values determined by the reference method shows how well the model predicts the samples.

As used herein, k-fold cross validation is an iterative validation procedure performed based on the parameter k, meaning k iterations are done during the cross-validation process. The training dataset is split into k subgroups. Then, for each iteration k-1 subgroups are used for the training of the model whereas the remaining subgroup functions as validation dataset. The root mean square error of cross-validation is obtained from the average root mean square error of all iteration of the cross validation. The root mean square error of prediction is obtained from the original test dataset and therefore represents the model performance on a dataset outside the training dataset. The root mean square error of calibration is obtained based on the complete training dataset. For PLS models a rank must be selected which is the number of multivariate factors used to explain the variance of the dataset. Increasing the rank results in an increased complexity of the model which could lead to an overfitting of the model to the training data making it unable to perform precise predictions on new data. Therefore, the best rank is as small as possible but also leads to a small root mean square error and a large R².

‘Partial Least Squares’ (Regression) is a statistical technique that reduces the predictors to a smaller set of uncorrelated components and performs least squares regression on these components, instead of on the original data.

As used herein, ‘Peak Integration’ calculates peak areas by integrating the measured signal within specified ranges.

As used herein, ‘RMSE’ is ‘root mean square error’ and a common metrics in regression. It relates to the average difference between the values predicted by a model and the actual values, and provides an estimation of how well the model can predict the target value (accuracy). In some embodiments, the ‘actual value’ describes the value that is measured with an analytical test system other than Raman. The different RMSE-based performance measures used are described as follows. As used herein ‘RMSECV’ is ‘root mean square error of cross validation’ and is a quantitative measure for the predictive ability of the model during cross validation. The RMSECV is comparable to the RMSEP for the external validation using an independent test set of samples.

Yc = NIRS predicted value of calibration set sample yc = Reference method value

n = Number of samples

As used herein ‘RMSEP’ is ‘root mean square error of prediction’ and is a quantitative measure for the predictive ability of the model during external validation using an independent test set of samples. The RMSEP is comparable to the RMSECV for cross validation.

YT = NIRS predicted value of independent test set sample

RMSEP = yr = Reference method value

n = Number of samples

As used herein, ‘RPD’ is Ratio of standard deviation (SD) and standard error of prediction (SEP).

SD = Standard deviation of reference values SEP = Standard error of prediction

Standard deviation may be determined by: value

As used herein, ‘SEP’ is ‘standard error of prediction’ and is the RMSEP corrected by the bias.

Yc = NIRS predicted value of independent test set sample yc = Reference method value

n - Number of samples As used herein, ‘MAPE’ is ‘mean absolute percentage error’.

A = actual value

Pt = predicted value by the model

n = number of fitted points

As used herein, ‘uncertainty’ relates to how the estimated value predicted by the model might differ from the true value to understand the degree of confidence in the prediction.

The skilled person will understand that parameters including bias, RMSEC, RMSECV, RMSEP, R², RPD value, uncertainty and/or MAPE may be referenced to determine or adjust parameters of a model, including the optimal function for a calibration curve type (e.g. on the basis of a RMSE vs. Function plot) and the rank of a PLS model (e.g. on the basis of a RMSE vs. rank plot).

As used herein, ‘calibration curve’ is the regression curve. The skilled person will understand that in general the type of a calibration curve may be simple, linear, quadratic, exponential, logarithmic, or any other type of curve.

As used herein, reference to a particular range of X - Y is intended to include X, Y and all values from X to Y, unless explicitly stated otherwise. For example, between X-Y includes X, Y and all values from X to Y, unless stated otherwise.

Samples comprising analytes

The methods of the present invention relate to determining the amount or concentration of various analytes present in blood-plasma, fractions, or in derivatives or resuspensions of precipitated material derived therefrom or during precipitation. Typically, the analyte being determined comprises total protein, IgG, and/or albumin, but may also comprise alternative components present in the samples, including alcohol (e.g., ethanol). The analyte may be an additive, i.e. an exogenous component added during the process, and is not naturally found in blood-plasma. The plasma may be fresh plasma, “normal” plasma, “hyperimmune” plasma, cryo-poor plasma (also referred to as cryosupernatant), or cryo-rich plasma. Optionally, the plasma has been treated to remove or deplete components such IgG, as C1 -inhibitor, PCC (Prothrombin Complex Concentrate), AT-III and/or other plasma proteins. The plasma may be obtained from a number of donations and/or individuals, and pooled.

The term “cryosupernatant” (also called cryo-poor plasma, cryoprecipitate-depleted plasma and similar) refers to plasma (derived from either whole blood donations or plasmapheresis) from which the cryoprecipitate has been removed. Cryoprecipitation is the first step in most plasma protein fractionation methods in use today, for the large-scale production of plasma protein therapeutics. The method generally involves pooling frozen plasma that is thawed under controlled conditions (e.g. at or below 6 °C) and the precipitate is then collected by either filtration or centrifugation. The supernatant fraction, known to those skilled in the art as a "cryosupernatant", is generally retained for use. The resulting cryo-poor plasma has reduced levels of Factor VIII (FVIII), von Willebrand factor (VWF), Factor XIII (FXIII), fibronectin and fibrinogen. Cryosupernatant provides a common feedstock used to manufacture a range of therapeutic proteins, including, but not limited to, alpha 1 -antitrypsin (AAT), apolipoprotein A-l (APO), antithrombin III (ATI 11) , prothrombin complex comprising the coagulation factors (II, VII, IX and X), FXIII, albumin (ALB), haptoglobin, hemopexin, transferrin, ceruloplasmin, and immunoglobulins such as immunoglobulin G (IgG).

The term “cryo-rich plasma” refers to plasma (derived from either whole blood donations or plasmapheresis) that has been frozen and then thawed, but from which the cryoprecipitate has not been removed.

Where plasma has been frozen for transport from a collection location, the frozen plasma is thawed and then collected in a pooling tank before centrifugation. The cryoprecipitate is removed by continuous centrifugation. The cryo-depleted plasma may be pumped into a stainless-steel fractionation tank and sampled for in-process controls.

The plasma, whether pooled from more than one or several hundred individuals, or whether obtained from a single individual, may be hyperimmune plasma. For example, the plasma may be obtained from the blood of individual(s) who have/has mounted an immune response to an infection, and have recovered (and are therefore otherwise healthy individuals). The sample comprising the analyte of interest may be a precipitate or fraction derived from processing of blood plasma. Many different methods can be used to selectively precipitate proteins from solution, for instance by the addition of salts, alcohols and/or polyethylene glycol with the combination of pH adjustment and/or a cooling step. It is therefore anticipated that the present invention will be applicable to most protein precipitates, such as immunoglobulin G- containing protein precipitates, regardless of how they are initially prepared. It should be noted that the present invention can also be implemented in separating other types of protein including albumin, immunoglobulins (Ig), such as IgA, IgD, IgE or IgM, either each type of immunoglobulin alone or a mixture thereof.

The sample may be any plasma derived IgG or albumin-containing material (e.g. in form of a paste, or precipitate) or derived from a starting material such as a solution from which the IgG or albumin can be precipitated by for example one or more of the methods explained above. In other embodiments, the sample may be any alcohol (e.g. ethanol) containing material or derived from a starting material such as a solution to which alcohol (e.g. ethanol) has been added to promote precipitation.

In orderto obtain the immunoglobulins or albumin from plasma, the plasma is usually subjected to alcohol fractionation, which may be combined with other purification techniques like chromatography, adsorption or precipitation. However, other processes can also be used. For instance, the protein-comprising precipitate can be the ll+lll precipitate according to the Cohn’s methods such as the Method 6, Cohn et. al. J. Am; Chem. Soc., 68 (3), 459-475 (1946), the Method 9, Oncley et al. J. Am; Chem. Soc., 71 , 541-550 (1946), or the l+ll+lll precipitate, the Method 10, Cohn et.al. J. Am; Chem. Soc., 72, 465-474 (1950); as well as the Method of Deutsch et.al. J. Biol. Chem. 164, 109-118 (1946) or the Precipitate-A of Nitschmann and Kistler Vox Sang. 7, 414-424 (1962); Helv. Chim. Acta 37, 866-873 (1954). Alternative precipitates comprising the protein of interest include but are not limited to other immunoglobulin G or albumin-containing Oncley fractions, Cohn fractions, ammonium sulphate precipitates from plasma described by Schulze et al. in U.S. patent 3,301 ,842. Further alternative precipitates comprising the protein of interest include but are not limited to octanoic acid precipitates, as described, for example, in EP893450. "Normal plasma", "hyperimmune plasma" (such as hyperimmune anti-D, tetanus or hepatitis B plasma) or any plasma equivalent thereto can be used as a starting material in the cold ethanol fractionation processes described herein.

The supernatant of the 8 % ethanol-precipitate (method of Cohn et al.; Schultze et al. (see above), p. 251), precipitate ll+lll (method of Oncley et al.; Schultze et al. (see above) p. 253) or precipitate B or IV (method of Kistler and Nitschmann; Schultze et al. (see Schultze above), p. 253) are examples of a source of IgG compatible with industrial scale plasma fractionation. The starting material for a purification process to gain IgG or albumin in high yield can alternatively be any other suitable material from different sources like fermentation and cell culture or other protein suspensions.

Particular protein-comprising precipitates or suspensions thereof can comprise plasma proteins, peptide hormones, growth factors, cytokines and polyclonal immunoglobulins proteins, plasma proteins selected from human and animal blood clotting factors including fibrinogen, prothrombin, thrombin, prothrombin complex, FX, FXa, FIX, FIXa, FVII, FVIIa, FXI, FXIa, FXI I , FXIIa, FXI 11 and FXI I la, von Willebrand factor, transport proteins including albumin, transferrin, ceruloplasmin, haptoglobin, hemoglobulin and hemopexin, protease inhibitors including p-antithrombin, a-antithrombin, a-2-macroglobulin, C1 -inhibitor, tissue factor pathway inhibitor (TFPI), heparin cofactor II, protein C inhibitor (PAI-3), Protein C and Protein S, a-1 esterase inhibitor proteins, a-1 antitrypsin, antiangionetic proteins including latent- antithrombin, highly glycosylated proteins including a-1 -acid glycoprotein, antichymotrypsin, inter-a-trypsin inhibitor, a-2-HS glycoprotein and C-reactive protein and other proteins including histidine-rich glycoprotein, mannan binding lectin, C4-binding protein, fibronectin, GC-globulin, plasminogen, blood factors such as erythropoietin, interferon, tumor factors, tPA, yCSF.

In certain embodiments, the methods of the present invention can be applied to determining the presence of, or concentration of, an analyte, eg total protein, during resuspension of a precipitate derived from blood-derived plasma or during precipitation, such as ethanol based precipitation. In particular, the methods can be used for assessing protein concentration in real-time during resuspension or precipitation and to assist in determining total protein concentration to facilitate determination of the amount or number of subsequent reagents to be added to the resuspension or precipitation. The advantage of the methods of the invention is that the manufacturer does not need to manually sample the protein-containing sample to then manually calculate the amount of subsequent reagent to add. Moreover, the progression of protein dissolution during resuspension of the protein-containing precipitate or paste, such as described herein, can be monitored in real-time, enabling more efficient determination of when the resuspension is complete, optimum time for adding the subsequent reagents or performing the next step in product processing and thereby reducing unnecessary cycling time. Similarly, during precipitation protein-containing precipitate, such as described herein, can be monitored in real-time, enabling more efficient determination of when the precipitation is complete, optimum time for adding the subsequent reagents or performing the next step in product processing and thereby reducing unnecessary cycling time.

In any aspect, the analyte is any plasma protein, peptide hormone, growth factor, cytokine, polyclonal immunoglobulin. Exemplary plasma proteins are selected from human and animal blood clotting factors including fibrinogen, prothrombin, thrombin, prothrombin complex, FX, FXa, FIX, FIXa, FVII, FVIIa, FXI, FXIa, FXII, FXIIa, FXIII and FXIIIa, von Willebrand factor, transport proteins including albumin, transferrin, ceruloplasmin, haptoglobin, hemoglobulin and hemopexin, protease inhibitors including p-antithrombin, a-antithrombin, a-2-macroglobulin, C1 -inhibitor, tissue factor pathway inhibitor (TFPI), heparin cofactor II, protein C inhibitor (PAI- 3), Protein C and Protein S, a-1 esterase inhibitor proteins, a-1 antitrypsin, antiangionetic proteins including latent-antithrombin, highly glycosylated proteins including a-1 -acid glycoprotein, antichymotrypsin, inter-a-trypsin inhibitor, a-2-HS glycoprotein and C-reactive protein and other proteins including histidine-rich glycoprotein, mannan binding lectin, C4- binding protein, fibronectin, GC-globulin, plasminogen, blood factors such as erythropoietin, interferon, tumor factors, tPA, yCSF.

In particular embodiments the analyte is a plasma protein selected from immunoglobulins such as immunoglobuilin G, albumin, fibrin, thrombin, prothrombin complex, fibrinogen, plasminogen, alpha 1 -antitrypsin, C1 -inhibitor, apolipoprotein A1 , alpha acid glycoprotein, haptoglobin, hemopexin, transferrin and coagulation factors such as Factor VII, Factor VIII and Factor IX.

Much of the core methodology used to extract plasma proteins is largely based on the cryoprecipitation and ethanol fractionation. Albumin and IgG were the first proteins to have been fractionated from human plasma using multiple-step, sequential cold ethanol processes. Cohn and his colleagues were the pioneer in plasma fractionation by using low temperature and by the addition of ethanol from 8% to 40% v/v for separation of albumin. In Cohn’s method, five fractions could be obtained which are from fraction I to fraction V, each of which is prepared by adjusting parameters such as the concentration of ethanol, concentration of protein, temperature, and pH.

There are various stages during the processing of plasma into specific protein rich fractions that involve the use of ethanol and the present invention can be used to determine the amount of ethanol present in a complex solution which can then inform any adjustments that are required. Further, the present invention can be used to determine when a certain concentration of ethanol has been reached during a step of ethanol addition.

As described herein, the sample comprising the analyte may be a turbid solution or suspension, particularly a highly turbid solution or suspension. In any embodiment, (a) the turbid solution or suspension may have NTU of equal to or greater than any value described herein, (b) the turbid solution or suspension may have NTU of any value described herein, or (c) the turbid solution or suspension may have a maximum NTU of any value described herein. Typically, turbidity is measured using various methods of photometry of turbid media, such as nephelometry, optometry, turbidimetry. Turbidity measurements are made using an instrument such as a turbidity meter or nephelometer. Typically, this is a photoelectric detector that measures the light scattered by a liquid. In particular, it is the scattering of light by suspensions that makes it possible to estimate the concentration of substances suspended in a liquid. Usually, this device consists of a white light or infrared light source. In nephelometry, scattered light is measured at 90° and 25° with respect to the incident light. In turbidimetry, scattered light is measured using a sensor located on the axis of the incident light. Such turbidity analysis methods are well known in the art and a wide range of instruments are available for turbidity analysis, including hand-held and in-line sensors, for example the Hach TL2360 hand-held turbidimeter that measures turbidity in nephelometric turbidity units (NTU) at a 90° angle. In any method of the invention described herein, the method further provides a step of determining the turbidity of a sample obtained from processing of blood-derived plasma. Preferably, the turbidity of the sample is any value or range described herein. Preferably, the step of determining the turbidity of a sample comprises measuring turbidity in a 10 mL volume of a test sample obtained from processing of blood derived plasma in 11 mm glass tubes using a Hach TL2360 turbidimeter calibrated with NTU primary Formazin solution standards, at a 90° angle.

Methods for obtaining wavelength spectra

The skilled person will be familiar with standard eguipment that can be used for applying light sources. In the context of the present invention, and in the preferred embodiments relating to determining protein concentration or ethanol concentration during processing of blood-derived plasma samples, the eguipment may include use of a Raman probe adapted for use in a (large) vessel which comprises the samples of interest.

In one embodiment the Raman spectroscopy instrument is arranged to analyze the test sample during mixing in a (large) tank and provide inelastic scattering or Raman shift data in real-time. In certain embodiments, several probes may be connected to a single spectrometer. In an embodiment, a first probe may thus be arranged at the first position while a second probe is arranged at the second position and, if applicable, a third probe is arranged at the third position. All such probes may be connected to the same spectrometer. The skilled person will appreciate that the use of multiple Raman probes may assist with providing a more accurate range of data relating to test samples or training samples comprising the analyte of interest.

The probe of the Raman spectroscopy instrument may be in the form of an immersion probe or constitute a part of a flow cell. The whole process flow or a side stream of the flow can be lead through such a flow cell.

In preferred embodiments, the Raman probe is configured to enable measurement of inelastic scattering or Raman shift during mixing of a sample. The optical slit of the Raman probe may be oriented parallel to the direction of the fluid stream during mixing. Typically, the optical slit of the Raman probe is oriented so that it is not directly facing the flow of the fluid stream during mixing. For example, the optical slit may be perpendicular or at an angle relative to the fluid stream during mixing. In other words, the Raman probe may be oriented downwards alongside the wall of the vessel.

Advanced data analysis models can be developed (e.g., partial least squares regression) to ultimately quantify the analytes at several time points within the process flow. To predict ethanol concentrations or specific protein concentrations (e.g., Immunoglobulin G (IgG) typically intrinsic florescence effects must be avoided in the protein-rich plasma solution. Intrinsic fluorescence is a rarely occurring characteristic of several proteins mainly caused by tryptophan residues and insignificantly caused by tyrosine side chains. This can negatively influence the sensitive Raman measurement.

Mathematical and/or statistical analyses, e.g. multivariate models

Any appropriate statistical model may be used in methods disclosed herein. In some embodiments, the model is a regression model that relates predicted variables (e.g., protein or ethanol concentration) and observable variables (e.g., Raman spectral data). In some embodiments, the regression model is a partial least squares model. In some embodiments, the model is a bilinear factor model that projects predicted variables (e.g., protein or ethanol concentration) and observable variables (Raman spectral data) into a new space. In some embodiments, the model is a regression model that uses principal components analysis (PCA) for estimating unknown regression coefficients in the model. However, other multivariate analytical techniques may be used including, for example, support vector machines, multivariate linear regression, and others.

Typically, inelastic scattering produced spectra contain hundreds of variables and therefore some form of multivariate data analysis method is preferably used to analyze raw data from the measurements. Such multivariate data analysis methods are well known in the art and includes Partial least squares regression (PLS); PLS Discriminant Analysis (PLS-DA); Ordinary Least Squares (OLS) regression; MLR (multiple linear regression); OPLS (Orthogonal-PLS); SVM (support vector machines); GLD (general discriminant analysis); GLMC (generalized linear model); GLZ (generalized linear and non-linear model); LDA (Linear Discriminant Analysis); classification trees; cluster analysis; neural networks; and Pearson correlation.

Fluorescent background can also be managed by employing preprocessing and baseline normalization techniques such as smoothing and/or rubber band subtraction, background correction algorithms or derivative spectroscopy, to Raman spectral data, including first and second differentiation, Savitzky-Golay smoothing differentiation, SNV, multiplicative signal correction (MSC), extended multiplicative signal correction (EMSC) polynomial fitting, Fourier Transform, wavelet analysis, orthogonal signal correction (OSC), and extended inverted signal correction (EISC) among others.

Raman spectroscopy

The principle of the Raman effect is based on inelastic light scattering. The scattering behavior depends on the vibrational properties of molecules. The basic vibration or energy level of individual molecules is affected by the energy transition from the Raman laser (photons). If, for example, a high-energy laser beam hits a molecule, a distinction is made between three different types of light scattering depending on the vibrational properties and electrical polarizability of the molecules and the energy transition: The Anti-Stokes Raman scattering, the Stokes scattering (inelastic scatter) and the Rayleigh scattering (elastic scatter). For the Anti-Stokes Raman scattering the energy transition takes place from the molecule to the photon, so that the energy level of the molecule is then lower, and the energy level and frequency of the photon is higher. For the Stokes Raman scattering the energy transition takes place from the photon to the molecule, so that the energy level of the molecule is then higher, and the energy level and frequency of the photon is lower. For the Rayleigh scattering the photon bounces off the molecule without any change of energy.

The change in energy level and frequency, also called Raman shift, is measurable by the Raman probe and, like a fingerprint, results in a unique spectrum depending on the molecular properties of the solution under investigation.

In some embodiments, the Raman spectroscopy may be performed in the visible, near infrared, infrared, near ultraviolet, or ultraviolet (UV) range. In some embodiments, a signal enhancement technique known as Surface Enhanced Raman Spectroscopy (SERS), which relies on a phenomenon known as surface plasmonic resonance, may be used. In some embodiments, resonance Raman spectroscopy, tip-enhanced Raman spectroscopy, polarized Raman spectroscopy, stimulated Raman spectroscopy, transmission Raman spectroscopy, spatially offset Raman spectroscopy, difference Raman spectroscopy, Fourier Transform (FT) Raman, or hyper Raman spectroscopy may be used. In some embodiments, a Raman analyzer can be used that is configured with a laser (e.g. laser diode) or other suitable light source that operates at appropriate wavelengths (e.g., those described herein). In some embodiments, the spectra are trimmed by removing peaks that are distorted. In some embodiments, peaks that are distorted are peaks that are laterally shifted or inverted. However, it should be appreciated that distorted peaks may include any peak that fails to meet certain criteria (e.g., intensity, signal-to-noise (S/N) ratio, shape, closeness to other peaks). Distorted peaks can be identified by visual inspection or by using a computer program that identifies (and removes) peaks that do not meet certain criteria. For example, peaks may be excluded because they are laterally shifted or inverted by at least 5%, at least 10%, at least 20%, at least 30%, at least 40%, or at least 50% compared with a reference peak (e.g., a non-distorted peak). Similarly, peaks may be excluded because they have a S/N ratio that is at least 5%, at least 10%, at least 20%, at least 30%, at least 40%, or at least 50% less than the S/N of a reference peak (e.g., a non-distorted peak).

In some embodiments, only a portion of a Raman spectrum is evaluated. For example, data relating to only a portion of the Raman spectrum is evaluated and the remaining data is filtered or otherwise removed prior to analysis. In some embodiments, the distorted peaks that are removed are lateral peak shifts. In some embodiments, a lateral peak shift looks like a 2- dimensional peak that has been stretched out. This peak distortion is likely the result of a component in the culture medium that is interacting with one of the bonds on the molecule of interest/analyte, the presence of a bond with similar character, solvent distortion, or any combination of these phenomena. In some embodiments, the laterally shifted peak or inverted peak is shifted by more than 5 cm^-1 in a concentration dependent fashion. In some embodiments, the lateral peak is removed if it is shifted by more than 1 cm^-1, more than 2 cm^-1, more than 5 cm^-1, more than 10 cm^-1, or more than 20 cm^-1 or more. In some embodiments, the lateral peak is removed if it is shifted by more than 1 cm^-1, more than 2 cm^-1, more than 5 cm^-1, more than 10 cm^-1, or more than 20 cm^-1 or more, in a concentration dependent fashion.

In some embodiments, the distorted peaks that are removed are inversion peaks (also called “inverted peaks” herein). An inversion peak is a peak where it appears that the lower concentration data is higher in magnitude than the high concentration data, when this relationship did not exist in the basis peaks. This type of distortion is usually due to a molecular species within the media that has similar vibrational properties and therefore similar peaks. In some embodiments, the inverted peak is removed if there is a lack of baseline. Methods for generating models/reference data sets

The skilled person will be familiar with general approaches for preparing a reference data set or developing a model of representative spectra against which the spectra from test samples can be compared for the purposes of determining analyte presence or concentration.

The reference data set may be from one or more samples comprising a known concentration of the analyte, wherein the concentration of the analyte has been determined by a method that is appropriate given the composition of the reference and test samples. For example, in the context of turbid solutions or suspensions comprising proteins, the most appropriate method for confirming protein concentration may be the Dumas method which is based on determining total nitrogen content, rather than other methods for determining protein concentration, such as the Biuret assay, BCA assay, Bradford assay or absorbance at 280 nm.

Representative spectra can then be obtained for the reference or training samples for which protein concentration has been determined, such that the representative spectra can be used to form the basis of a model against which test spectra can be assessed.

The skilled person will appreciate that in most cases the greater the number of representative spectra or training spectra provided, the greater the accuracy of the model.

There may be a need to apply spectral pre-treatments to data (whether the test spectra or the reference or training spectra used to derive a suitable model). These pre-treatments can be applied to emphasise spectral changes. Examples of suitable spectral pre-treatments include vector normalisation, first order derivative, min-max normalisation, straight line subtraction, multiplicative scatter correction, 2^nd order derivative, baseline corrections and combinations thereof. Preferably, the pre-treatment applied to the test spectra or the reference or training spectra used to derive a suitable model is vector normalisation or 1^st order derivative. In one embodiment, the pre-treatment applied to the test spectra or the reference or training spectra used to derive a suitable model is vector normalisation in combination with 1^st order derivative. Optionally, the pre-treatment further comprises Standard Normal Variate (SNV). Optionally, the pre-treatment further comprises a smoothing. Optionally, the pre-treatment further comprises a standardization, preferably wherein the standardization is performed by area normalization. The model may be generated using a multivariate calibration algorithm, such as Multiple Linear Regression (MLR), Principal Component Regression (PCR), or Partial Least Squares (PLS)-Regression. Preferably the model is generated using Partial Least Squares (PLS)-Regression, such as that described herein. The PLS algorithm is described in (Haaland, Thomas, Anal. Chem 60 (1998) 1193; Martens, Naes, Multivariate Calibration, J. Wiley & Sons, New York (189): Chapter 3.5; Brown, Apply. Spectosc. 49, No. 12 (1995) 14A; and Bouveresse, Hartmann, Massard, Last, Prebble, Anal. Chem. 68, No. 6 (1996) 982).

Methods for assessing the quality of a given model (including to then determine whether further training data are required for further developing the model) are described herein.

In certain examples, criteria that may be considered when assessing the model quality of the different chemometric models or multivariate models include:

• Rank: corresponds to the number of factors of the chemometric model. A lower rank usually leads to increased model stability.

• Root mean square error of cross validation (RMSECV): The RMSECV should be minimized.

• Residual prediction deviation (RPD): model performance indicator. The RPD should be maximized.

• R²: coefficient of determination, describes the relation between spectral data and the concentration data. The R² should be maximized to close to 100.

The following criteria may also be considered when assessing the predictive ability of the chemometric models or multivariate models on an independent data set:

• Bias: Average difference between reference values and predicted values. Should be close to 0.

• Root mean square error of prediction (RMSEP): accuracy indicator for prediction of independent test samples. The RMSEP should be minimized.

• R²: coefficient of determination, describes the relation between spectral data and the concentration data. The R² should be maximized to close to 100. MAPE: mean absolute percentage error, measures the prediction accuracy of a forecasting model. The MAPE should be minimized.

The generation of the model may involve training samples that may comprise a representative set of samples that cover variables, such as different paste type, sample temperature, instrument variability, operator handling, raw materials, and plasma source. Using such varied reference samples to capture such variables in the generation of the training model will further ensure the robustness of the model when it comes to assessing a variety of samples comprising analytes of unknown concentration.

In any method described herein, the method may further comprise a step of determining the presence of, or concentration of, an analyte in a sample after comparing the test spectra with reference spectra, comparing the test spectrum with a reference spectrum, or comparing the test spectra to a reference data set.

Raman signature

In some embodiments, the Raman signature comprises a selected number of peaks and associated peak ranges that allow for the evaluation (e.g., identification) of an analyte in a sample.

In some embodiments, the Raman signature comprises a selected number of peaks and associated peak ranges that allow for the evaluation of the level of an analyte in a sample.

In some embodiments, a Raman signature of an analyte comprises multiple combinations of identifying peaks. It should be appreciated that a minimal number of peaks may define a Raman signature. However, additional peaks may help refine the Raman signature. Thus, for instance, a Raman signature consisting of 4 peaks may provide a 95% certainty that a sample that shows those peaks contains the analyte associated with the Raman signature. However, a Raman signature consisting of 10 peaks may provide a 99% certainty that a sample that shows those peaks contains the analyte associated with the Raman signature. Similarly, a Raman signature consisting of 4 peaks may provide a 90% certainty that a sample that shows those peaks contains the analyte at the level of the analyte associated with the Raman signature. However, a Raman signature consisting of 10 peaks may provide a 98% certainty that a sample that shows those peaks contains the analyte at the level of the analyte associated with the Raman signature.

In another aspect, the present invention provides a method for precipitating IgG during the processing of blood-derived plasma, the method comprising: adding ethanol to an IgG containing solution or suspension; applying a light source to a test sample from the solution or suspension after the addition of ethanol; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of IgG to determine the presence of, or concentration of, IgG in the sample, optionally, adding more ethanol, or continuing to add ethanol, preferably until a comparison of a test spectrum with a reference spectrum that contains a Raman signature of IgG indicates the desired level of IgG. In another aspect, the present invention provides a method for determining the presence of, or concentration of, albumin in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating a test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of albumin to determine the presence of, or concentration of, albumin in the sample.

In the methods of the invention where a test spectrum is compared to a reference spectrum that contains a Raman signature, there is no need to generate a model.

In any aspect, the reference spectrum may contain a Raman signature for any plasma protein, peptide hormone, growth factor, cytokine, polyclonal immunoglobulin. Exemplary plasma proteins are selected from human and animal blood clotting factors including fibrinogen, prothrombin, thrombin, prothrombin complex, FX, FXa, FIX, FIXa, FVI I , FVIIa, FXI, FXIa, FXI I , FXIIa, FXI 11 and FXI I la, von Willebrand factor, transport proteins including albumin, transferrin, ceruloplasmin, haptoglobin, hemoglobulin and hemopexin, protease inhibitors including |3- antithrombin, a-antithrombin, a-2-macroglobulin, C1 -inhibitor, tissue factor pathway inhibitor (TFPI), heparin cofactor II, protein C inhibitor (PAI-3), Protein C and Protein S, a-1 esterase inhibitor proteins, a-1 antitrypsin, antiangionetic proteins including latent-antithrombin, highly glycosylated proteins including a-1 -acid glycoprotein, antichymotrypsin, inter-a-trypsin inhibitor, a-2-HS glycoprotein and C-reactive protein and other proteins including histidine-rich glycoprotein, mannan binding lectin, C4-binding protein, fibronectin, GC-globulin, plasminogen, blood factors such as erythropoietin, interferon, tumor factors, tPA, yCSF.

In particular embodiments the reference spectrum contains a Raman signature for a plasma protein selected from immunoglobulins such as immunoglobulin G, albumin, fibrin, thrombin, prothrombin complex, fibrinogen, plasminogen, alpha 1 -antitrypsin, C1 -inhibitor, apolipoprotein A1 , alpha acid glycoprotein, haptoglobin, hemopexin, transferrin and coagulation factors such as Factor VII, Factor VIII and Factor IX.

It will be understood that the invention disclosed and defined in this specification extends to all alternative combinations of two or more of the individual features mentioned or evident from the text or drawings. All of these different combinations constitute various alternative aspects of the invention.

Examples

Example 1 - Materials and Methods

Calibration of the Raman Rxn2 probe

The Raman Rxn2 probe was selected since it is designed for the analytical in-situ consideration with model transfer capabilities in chemical, pharmaceutical, biopharmaceutical and food and beverage applications. It promises reliable real-time measurements with embedded control software to convert acquired Raman spectra into process knowledge through integrated multivariate predictors. The calibration of Raman Rxn2 probe was performed prior to experiments by using a calibration accessory with corresponding optical adapter (Raman calibration accessory, Raman Rxn2 analyzer and Raman Rxn-10 probe from Endress+Hauser Group Services AG). After a 12.5 minutes warm-up phase in “Intensity” mode, the “Calibration” mode was started and documented. Subsequent verification was performed in a sample chamber filled with 70% isopropanol using “Verify” mode and documented. Finally, 40 - 70 % saturation of the detector was checked. During PoC study, calibration and verification was done accordingly, with the exception that a different Raman probe in combination with suitable sample chamber was used (Raman Rxn2 analyzer with Raman enclosed sample compartment from Endress+Hauser Group Services AG).

Proof-of-Concept study

The PoC study was performed by using 5 ml of deep frozen pre- and post-precipitation aliquots. Samples were thawed at 37°C prior to testing. The Raman measurements were performed for 5 minutes in triplicates (wavelength 2 = 785 nm; exposure time = 10 s; count = 30) with calibrated equipment.

Plasma fractionation experiments with inline Raman measurements

In total, a set of four fractionation experiments was performed at development scale. For the model training three experiments were performed. Frozen starting material was thawed for 4h at +37°C and subsequently cooled to +1 (±1)°C prior to the start of the fractionation procedure. The samples taken during the precipitation were used as training samples to “train” the model. The model validation experiment was performed by usage of fresh starting material. The samples taken in this process were accordingly designated as validation samples. Protein precipitation was done according to standard procedures for five hours in a darkened incubator whilst recording of Raman spectra. The taken samples were deep frozen at -80°C until further use. Data used for model building were generated in “continuous” measure mode, using parameters as described above. Sampling interval was set to 30 min for training samples, and to 60 min for validation samples, respectively. During the training experiments two Raman probes were installed in the experimental approach.

Sample analytics

Prior to analysis, frozen samples were thawed in a water bath at +2°C and sediments separated by centrifugation (+2°C, 20000 x g, 60 min). Sediments were discarded and supernatants homogenized and kept at +2°C until analysis. IgG and ethanol concentrations were determined on the Cedex Bio HT device from Roche Diagnostics GmbH using test kits from Roche (material number of the IgG test kit: 06608540001 , material number of the ethanol test kit: 08055661001). The "IGGHD", "ETOHB" and "ETOHD" tests were calibrated at the appropriate concentrations according to the manufacturer's instructions.

Creation of a prediction model using Raman spectroscopy and PLS regression

The partial least squares regression (PLS) was selected for prediction model building. In the quantification of systematically variable analyte concentrations, multivariate PLS analysis was applied because this modelling relates the variation of the measured spectrum to the systematic variation of the concentrations of different target analytes. (Drishya Rajan Parachalil, 2020) For model building, data such as concentration determination of IgG and ethanol at respective timepoints and related Raman spectra were required (see Figures 6 and 7). IgG and ethanol concentrations measured were sorted according to the corresponding spectra and loaded into the software-bound data table (software: PEAXACT, Version 4.7 from S-PACT GmbH). Furthermore, the "Usage" which means the purpose of the sample for model building was selected. Samples generated in the runs performed with frozen starting material were classified as "Training" samples, whereas samples generated in the run performed with fresh starting material were classified as "Test" samples. Moreover, run numbers have been assigned, categorized by numbers 1 - 4 (experiment with fresh cryo-poor plasma = 1 , experiment with frozen cryo-poor plasma = 2/3/4). Afterwards, the spectra were inspected in a 2D and 3D plot and checked for visual irregularities. Thereafter, the model building was started. In the partial window "Data pretreatment", rubber band subtraction was selected for baseline correction of the y-axis. Moreover, a global range of 350 - 3100 cm ¹ was selected for the x- axis, excluding 1800 - 2650 erm¹. Additionally, a baseline node (inflection point) at 1370 cm ¹ was added. Finally, a smoothing with a filter length of five was inserted and pretreatment completed. Then a "Data filter" was selected including a range of about ~ 800 cm ¹ - 3000 erm ¹ for ethanol. Furthermore, an IgG "Data filter" was added in the range of ~ 800 cm ¹ - 500 erm¹ and 2600 erm¹ - 3000 erm¹. This was followed by the calibration of the model. The "Partial Least Squares Regression" (PLS) was selected as calibration method. A check mark at "Predictive features" was selected and corresponding IgG or ethanol "Data filter" were selected. As regression variable "Centered" with a maximum rank of ten was defined. For partitioning under the item "Cross-validation", "K-fold" with a quantity of three was defined. In the "Calibration report" the most suitable ranks for IgG and ethanol were selected out of two regression models "Root mean square error" (RMSE) respectively R squared (R²) aiming the smallest possible RMSE and largest possible R² values. Finally, other reports were evaluated to find possible multivariate outliers. In the end, the quality of the calibration was checked using the "Predicted vs true report" function. Both calibrations, IgG and ethanol, showed strong overlapping identity and recovery lines and a good fit for sample points.

Based on this, the model was validated in a "Predicted vs true report". Here, test samples were compared with the training samples.

Example 2 - Results of Proof-of-Concept study

In the PoC study, the potential use of the Raman Rxn2 probe in plasma-based samples was tested to check the general feasibility. This included the verification of the diversity of the spectra before and after IgG precipitation using ethanol, as well as the reproducibility and quality of the spectra. Three spectra of each sample (before and after precipitation) were recorded according to the settings selected (see Figure 1). As a first result, different spectra could be recorded with high signal to noise ratio and distinct peaks before and after precipitation. Based on these Raman spectra, a general feasibility, replicability of the measurements was determined. The background noise is low and therefore close to the baseline. In the spectra of the sample before precipitation peak signals were found at a Raman shift of about 1630 cnr¹ (maximum Raman intensity about 77.500), 1450 cnr¹ (maximum Raman intensity about 37.500), 1375 cnr¹ - 1150 cm ¹ (more a broad peak area, maximum Raman intensity between 18.000 - 5.000), 1150 cm ¹ - 980 cm ¹ (more a broad peak area with a pointed tip at 1000 cm ¹, maximum Raman intensity between 31.000 - 5.000) and 980 cm ¹ - 600 cnr¹ (more a broad peak area, maximum Raman intensity between 15.000 - 2.000). In the spectra of the sample after precipitation peak signals were found at a Raman shift of about 1630 cnr¹ (maximum Raman intensity about 5.000), 1450 cnr¹ (maximum Raman intensity about 34.000), 1275 cnr¹ (maximum Raman intensity about 8.000), 1080 cnr¹ + 1050 cm ¹ (double peak, maximum Raman intensity about 19.000 + 22.500) and 875 cm ¹ (maximum Raman intensity about 75.000).

In the PoC, the potential use of the Raman Rxn2 probe in plasma-based samples was tested to check the general feasibility as an initial comparison. The cuvette method applied offers the advantage of being able to measure samples flexibly at any time outside a running process. Based on literature research and interviews with technical experts, the wavelength of 2 = 785 nm was assumed to be suitable. It affects the intensity, resolution, background fluorescence, acquisition time and the cost incurred. According to the manufacturer, detector saturation should be in the range of 10 - 80 %. Consequently, the exposure time was set to 10 seconds and the count at 30 (10 s multiplied with 30 count = 3000 s = 5 min) which resulted in a detector saturation of 12 %. Furthermore, the signal-to-noise ratio, i.e., the ratio between the signal height and the height of the noise, is very high, since the background noise is close to the baseline and the peaks reach high intensities. If the principle of the precipitation process is applied to the spectrum course it was expected that the signal caused by IgG decreases, while the signal caused by ethanol increases significantly. Thereby, it is found that the resulting double peak between 1100 cnr¹ and 1000 erm¹, as well as the signals at about 900 cm ¹ and 1455 cnr¹ can be found in the known ethanol spectrum. In contrast, the identification of IgG is more complex since the chemical structure is multidimensional. In addition, the subclasses of IgG are characterized by different glycosylation patterns, which means that the spectra of IgG subclasses are not identical. IgG can mainly be localized together with other proteins in the area between ~ 700 - 1900 erm¹. Based on the spectrum, the areas in which a decrease in the signal can be seen, can be assigned to a protein precipitation, and following to an IgG depletion. These areas are located between 475 - 770 erm¹, 920 - 970 erm¹, 990 - 1010 erm ¹, 1150 - 1370 erm¹, 1550 - 1725 erm¹. The areas are mainly characterized by carbon bonds (single and double, aliphatic, and aromatic), hydrocarbons, amide bonds of the beta sheets, tryptophane, sulfur carbon bonds and halogen carbon bonds. Using this technique, both aromatic amino acid side chains (e.g., phenylalanine at 1001 erm¹ or tryptophan at 707 erm¹, 1225 erm¹, 1366 erm¹, 1455 erm¹, 1487 erm¹) and the secondary structure characteristic for IgG, the beta-sheet structure (1237cm ¹), could be characterized. Furthermore, the phenomenon of intrinsic fluorescence was excluded based on the high signal-to-noise ratio in the recorded spectra.

Example 3 - Results of plasma fractionation experiments with inline Raman measurements

In total, four experiments were performed. The aim of the investigation was to evaluate comparability of the spectra depending on the conditions of the starting material (fresh vs frozen). In three of the four experiments frozen starting material (CPP) was used. In addition, two Raman probes were applied in these experiments to consider the influence of the probe's locality. The samples generated from this and further measured ethanol and IgG concentrations are denoted as training samples and used for model building and training. Another experiment using fresh starting material was used to verify the model. Therefore, the generated samples are designated as test samples (see Figure 2). The spectra were similar regardless of the location of the probe and the condition of the starting material.

Example 4 - Results of the determination of IgG and ethanol concentration

In general, the four fractionation experiments focus on the building of a model to predict IgG and ethanol concentrations at real time. As part of the implementation of a suitable prediction model, the IgG concentrations as well as the ethanol concentrations during the precipitation were determined. The course of precipitation (IgG concentration as a function of ethanol concentration, determined by simultaneous determination of IgG and ethanol (EtOH) concentration) is shown in Figure 3. Decreasing IgG concentrations with corresponding increasing ethanol values were observed. The course is reproducible (from the condition of the starting material) but not linear (experiment 1 = fresh starting material, experiments 2,3,4 = frozen starting material). Furthermore, a higher decrease of IgG from an ethanol concentration of > 100 mg/mL can be observed. Even with a division of the graph at 100 mg/mL of ethanol, the resulting mean values of the slopes show that the precipitation effect is four times higher (mean value for slope of ethanol concentration < 100 mg/mL: -0.01 , mean value for slope of ethanol concentration > 100 mg/mL: - 0.04) at ethanol concentrations > 100 mg/mL.

Example 5 - Results of the creation of a prediction model using Raman spectroscopy and PLS regression

Spectra were recorded during each precipitation experiment. Within the three experiments used for the model training and building, two Raman Rxn2 probes were installed to ensure a repeat determination. In a visual inspection in a three-dimensional field, the same Raman shifts, and signal strengths off were determined. Thus, the Raman measurement method is confirmed regarding its quality and reproducibility. The prediction model is built using an empirical PLS regression, a classical multivariate method that provides reliable predictions. Within the spectra, the characteristic signals of the known ethanol spectrum can be found unambiguously, but this does not apply to IgG. IgG can mainly be localized together with other proteins in the area between ~ 700 - 1900 cm ¹. Based on the spectra, the areas in which a decrease in the signal can be seen, can be assigned to a protein precipitation, and following to an IgG depletion. These areas are located between Raman shifts of 475 - 770 cm ¹, 920 - 970 cm ¹, 990 - 1010 cm ¹, 1150 - 1370 cm ¹ and 1550 - 1725 cm ¹. In PLS, the regression is always performed separately for each feature, which in this case is IgG and ethanol. By means of the root mean square error (RMSE) and R squared (R²), calculated in the calibration/training process of the model, the smallest possible ranks for IgG and ethanol have been selected (see Table 1). This resulted in rank 3 for ethanol and rank 4 for IgG. For the evaluation of the quality of the model calibration/training, the PEAXACT software calculates several error indices during the validation process. This includes the root mean squared error of calibration (RMSEC), the root mean squared error of prediction (RMSEP) and the root mean squared error of cross- validation (RMSECV) (see Table 1).

Table 1 : Significant statistical values for selecting the rank

Based on R ², RMSEC, RMSECV and RMSEP rank 4 was selected for IgG and rank 3 for ethanol (EtOH).

The resulting "Predicted vs true" plots allowed the modelling to be visually examined.

Looking at the "Predicted vs true" validation plot for ethanol (see Figure 4) the identity line, which is calibrated to the recovery line, is surrounded by training samples from the calibration (= training samples, light grey) and test samples (= validation samples, dark grey) These are lying very close to the identity line which indicates that the predicted values represent values that are very close to the actual measured values. The dot cloud around (0| 0) , which correlates with an ethanol concentration near 0 mg/mL, is attributed to the samples before precipitation (starting material CPP and after pH adjustment/before precipitation). Based on the graphical progression during the precipitation itself, it is obvious that the test samples/ validation samples are located amidst the training samples and thus within the prediction prognosis. In addition, a gentle waveform of the course of the samples is visible in the "Difference vs true" plot (lower plot). Nevertheless, no strong expression of this can be found in the "Predicted vs true" (upper Plot).

The validation samples in the IgG plot are located amidst the training samples (see Figure 5). Due to the precipitation kinetics of IgG, there is an accumulation of points at the upper end (> 4 mg/mL) and a volatilization at the lower end (< 3.5 mg/mL) of the concentration spectrum. In the lower plot some values scatter up to ± 0.5 mg/mL around the mean value.

Discussion

After the use of this PAT technology was deemed possible in the PoC, precipitation experiments were carried out to record Raman spectra. To record spectra with the highest possible quality from the outset, dark conditions are recommended. By preventing further light irradiation from outside, the noise of the spectra is further reduced. As mentioned above, following the principle of double determination, two Raman probes were used in three of the four experiments. This was to exclude that the localization of the probe has no influence on the precipitation process and reverse the precipitation process is reproducible at two different locations within the reaction vessel.

The prediction model was built using an empirical PLS regression. This type of model is a classical multivariate method that provides reliable predictions. Since no defined peaks can be unambiguously assigned to IgG, the change in the entire spectrum in the corresponding wavenumber ranges were considered for analysis. IgG can be quantitatively detected in the range of 0.08 mg/mL and 7.5mg/mL.

The samples originated from the experiment with fresh cryo-poor plasma as starting material were selected as validation samples, so that the model, which was trained with frozen starting material, could directly be applied and verified for the use of the fresh cryo-poor plasma samples.

During the cross-validation process, the data were first divided into a certain number of segments of equal size. Then, depending on the selected "k" (number of iterations, here 3), k iterations of training and validation were performed, so that in each iteration a different segment was used for validation, while the remaining k - 1 segments were used for training. Through this the root mean squared errors of cross-validation and prediction were derived. The rank selected therefrom is the number of multivariate factors used to explain the variance of the data. The best rank should be as small as possible, resulting in the smallest possible RMSE and a large R². Depending on the selection of the rank, the calibration quality and the resulting prediction quality differ.

In the "Predicted vs true" validation plot (see Figure 4) for ethanol and IgG the identity line, which was calibrated to the recovery line, is closely surrounded by training samples from the calibration (= training samples, light grey) and test samples (= validation samples, dark grey), which indicates low error prediction. The dot cloud at (0|0) in the ethanol plot, which is attributable to samples taken before precipitation, can be disregarded in the consideration because their irrelevance for the prediction during precipitation. This negatively influences the prediction accuracy for the values, as shown in the "Difference vs true" plot below (variation from - 15 mg/mL to + 20 mg/mL of the mean value). Finally, the predicted values for the samples before precipitation won’t have any importance for the precipitation process and were therefore hidden. Much more important are the predictions during the precipitation. Here, the "Predicted vs true" plot also shows more reliable predictions. Based on the graphical progression, it is obvious that the validation samples are located amidst the training samples and thus within the prediction prognosis. In addition, a gentle waveform of the course of the samples is visible in the "Difference vs true" plot. Nevertheless, no strong expression of this can be found in the "Predicted vs true" plot.

When looking at the IgG plot (see Figure 4) the training samples as well as the validation samples are close to the identity line. The validation samples are located amidst the training samples and thus within the prediction prognosis. However, there is an accumulation of points at the upper end (> 4 mg/mL) and a volatilization at the lower end (< 3.5 mg/mL) of the concentration spectrum. However, it can be assumed that a close location of the points would also be found in the middle area. In terms of the process, the samples with higher IgG concentration are located at the beginning of the process, where ethanol concentrations of < 100 mg/mL are predominant. After reaching this concentration, the IgG precipitation behavior changes (as shown in Figure 3) to a more strongly and rapidly precipitation. Due to this, there is a slight thinning of the values in the concentration range between 4.5 mg/mL - 1.5 mg/mL. Because of very similar concentrations of IgG after the precipitation an accumulation of dots at the lower left end of the graph (< 1.5 mg/mL) can be observed. Moreover, this indicates a reproducible precipitation process. In the "Difference vs true" plot (see Figure 4) it can also be seen that the values < 4.5 mg/mL show a larger dispersion around the mean (± 0.5 mg/mL) value. This signifies, that the prediction at the beginning of the precipitation could be more imprecise than in the end. Nevertheless, the initial concentration of IgG depends on the starting material and usually has a wider fluctuation than after the precipitation.

Finally, to evaluate the quality of the prediction, the average error to be expected within future predictions (RMSEP), estimated from the validation set is focused on in the validation. RMSEP is calculated by summing all squared prediction errors during a cross-validation and characterizes the goodness of a model. In principle, a low RMSEP value indicates a good prediction model. But RMSE values are also dataset dependent, which means it depends on the range of concentration of each parameter. The ethanol concentrations in the whole precipitation process ranges from 0 - 200 mg/mL and have a RMSEP of 9.93 mg/mL (see Table I). The IgG concentrations ranges from 1.5 - 6.1 mg/mL throughout the process and have a RMSEP of 0.26 mg/mL (see Table I). For an estimated precision for predicted concentrations, it is recommended to use 2 x RMSEP, because this is comparable to a 95 % confidence interval. This would further mean that an analytical accuracy of ± 19.86 mg/mL for ethanol and of ± 0.52 mg/mL for IgG would be achieved with each future inline Raman measurement. These accuracies are also considered in the lower "Difference vs true" plots.

In summary, the precipitation behavior of IgG during the precipitation step was first investigated in detail at development scale. Continuing with Raman spectroscopy, a prediction model for IgG and ethanol concentration was built, trained, and validated so that it can now be used for real-time concentration determinations for IgG and ethanol during ethanol precipitation of human plasma.

The IgG precipitation behavior was evaluated as a function of the ethanol concentration and initially assessed using Raman spectroscopy as a PAT tool. From a concentration of about 100 mg/mL ethanol, a fourfold increase in the slope of IgG depletion is observed. From this, optimization approaches can be derived in which the flow rate is increased to varying degrees until the mentioned concentration is reached to shorten the overall process time. Furthermore, the use of Raman spectroscopy in plasma-based precipitation approaches was found to be feasible and useful. A statistical PLS prediction model was developed and validated, which can be used in the future in combination with inline Raman spectroscopy to perform real-time concentration determinations of IgG and ethanol. Thus, long waiting times for results and delays can be avoided. In addition, further models for subsequent process steps or for other essential process parameters can be created to provide even more detailed and comprehensive insights.

Table 2 - Different turbidity units measured during cryopoor plasma (CPP) precipitation

* 6685 NTU were detected by utilizing the Hach turbidity measurement system as offline measurement

Example 6 - Optimization on Raman-based inline monitoring of IgG and ethanol concentrations during cold ethanol plasma fractionation of IgG at lab scale

The initially created proof-of-concept (PoC) PLS prediction model for IgG and ethanol (EtOH) (Examples 1-5) was evaluated against different model pretreatments and calibration options to further decrease root mean squared errors of the calibration (RMSEC), the prediction (RMSEP) and the cross-validation (RMSECV).

Methods

The PEAXACT software, Version 5.8, was used for spectra evaluation and the subsequent model building process in the model optimization. The training and test samples were used in the same way as in the PoC study, but under different modelling conditions (pretreatment and calibration). The optimized models characterized by their key parameters are shown in the table below. Table 3 - Key parameters of the PoC, IgG and ethanol (EtOH) PLS models

Calculated model errors

Based on the chosen model parameters, PEAXACT software calculated the coefficient of determination (R²) and root mean square errors for calibration (RMSEC), cross-validation (RMSECV) and prediction (RMSEP), which serve as an initial quality attribute for model performance. Results of the calculation are shown in Tables 4 and 5.

Table 4 - Overview of calculated model errors for IgG at concentrations ranging from 1.53 - 6.1 g/L

Table 5 - Overview of calculated model errors for ethanol (EtOH) at concentrations ranging from 0.51 (<0.51 ) - 200.8 g/L

RMSE, the average difference between the values predicted by a model and the actual values, provides an estimation of how well the model can predict the target value (accuracy). Consequently, lower values, approaching hypothetical state (expected value was predicted), are desirable. Simultaneously, R², a quality measure of the linear regression generated in the calibration, should be strived as high as possible. Optimized models for IgG (no. 2) and EtOH (no. 3) were characterized by lower RMSEs and higher R², indicating promising predictions.

The PoC model for both parameters, IgG and EtOH, were optimized through model adaptions (e.g., 1^st order derivation, area normalization, more precise data filter in pretreatment and higher ranks in calibration) and consequently serve as low-error models. Predictive ability of the models

To evaluate the predictive ability of those models, the predicted concentration values were compared to the actual measured values (true values) with regard to their absolute and percentage deviations (n = 42). Hereby, the corresponding uncertainty (95%) is calculated by the software for every sample. For simplified representation, the mean absolute percentage error (MAPE) is represented for each model (Tables 6 and 7). In addition, the minimum deviations are tabularly contrasted with the maximum deviations to outline the range of deviation and beyond that the amount (n) of deviations > 10% are considered for evaluation.

Table 6 - Overview of main evaluation parameters for IgG models

Out of 42 observations

Table 6 shows that the uncertainty (95%) was reduced by half through model adaptions, striving lowest possible value. The MAPE, which represents the mean deviation across all measured values in a dataset, indicates the accuracy of a prediction model in relation to the to the actual values, performing best without outliers. Even if they are calculated regardless of the magnitude of the errors, it is considered suitable for an overall comparison of the quality of the IgG models. The MAPE for the IgG model was improved from 6.41 to 3.62%. The IgG model further provides a minimum deviation of 0.0, and a maximum deviation of 19%. The optimized prediction ability was identified in the predicted vs. true plots where the recovery line and trainings samples (fresh starting material) approximate closer to the identity line (Figure 8). The test samples (frozen starting material) still fit in well and demonstrate the transferability of the prediction. Table 7 - Overview of main evaluation parameters for ethanol (EtOH) models

** Excluding values of lower quantification limit (0.51 mg/mL)

The uncertainty (95%) for ethanol prediction (Table 7) was reduced by a factor of three by remodelling. The MAPE values improved from 7% (PoC model) to 2%. The minimum deviation was improved from 0.6% to 0.0% and is found in the presence of ethanol (during precipitation), whereas the maximum deviations are unambiguously attributable to the values below the detection limit (before precipitation). This effect is also recognizable in the predicted vs. true plots, where scattering of data points is mainly related to low concentration range. Here again, by excluding these values (< 0.51 mg/mL), maximum deviation improved to 11% and 32%, respectively. The recovery line and trainings samples (fresh starting material) are located very close to the identity line (Figure 9). The test samples (frozen starting material) continue to match and again demonstrate the transferability of the prediction.

Discussion

The PLS prediction model initially created served as a proof-of-concept and thus as a base for further optimizations. Therefore, it was evaluated against different model pretreatment and calibration options to further decrease RMSEC, RMSEP and RMSECV and following improve predictive ability. For the re-modelling, the same data (spectra and concentration determinations of IgG and EtOH and classification of training/ test samples) were applied.

For model optimization, other model types than PLS were assessed (e.g. peak integration, hard modelling), but PLS was considered as most suitable prediction method. Finally, the optimized models are both characterized by a R² of 1.00, lower RMSEs, uncertainties and MAPEs, promising more accurate prediction of IgG and ethanol concentrations during cold ethanol plasma fractionation. Example 7 - Raman-based inline monitoring of total protein and ethanol (EtOH) concentrations during cold ethanol plasma fractionation of an albumin enriched plasma fraction

Six experiments including inline Raman spectroscopy were performed to build prediction models for total protein (TP) and ethanol (EtOH) concentrations during the Kistler-Nitschmann (KN) or Cohn (C) fractionation of an albumin enriched plasma fraction to generate a protein precipitate comprising residual proteins at laboratory scale using a pre-processed cryo-poor plasma derivative. Each experiment involved two parallel runs with the same starting material but at different EtOH dosage speeds.

In this study report, based on the data generated, different models were evaluated to create suitable prediction models for ethanol (EtOH) and total protein (TP) concentrations, which monitored the protein and EtOH concentration in real-time during the protein precipitation step using inline Raman spectroscopy.

Partial-Least-Squares (PLS) modelling was used for the prediction of total protein concentrations. A Peak Integration (PI) approach was used for the prediction of total EtOH concentrations. As a base, different PLS and PI prediction models were created to evaluate different spectra pre-treatment and model calibration methods. The focus was on decreasing root mean squared errors of the calibration (RMSEC), the prediction (RMSEP) and the cross- validation (RMSECV), to enable an accurate concentration prediction for the key parameters (TP and EtOH).

Methods

A Bio Optics Raman probe (Endress+Hauser GmbH + Co. KG) was installed inline and samples were taken directly after spectrum recording. The selected probe parameters for spectra measurement during protein precipitation were: wavelength = 785 nm, channel no. = 1 , exposure time = 10 s, and count = 30 for runs 1 , 3, 5, 7, 9 and 11 ; and wavelength = 785 nm, channel no. = 2, exposure time = 10 s, and count = 15 for runs 2, 4, 6, 8, 10 and 12. The spectrum recording and the associated sampling intervals were set to an approximately quarter-hourly cycle. Furthermore, data acquisition (spectra recording and sampling) was not limited only to the precipitation and post-stirring steps, but was extended to previous process steps.

Model building using PEAXACT software

Depending on the recorded spectra and analyzed samples, for each model, separate model training sets and a model test/validation set were defined to create the prediction models (Table 8). Table 8 - Definition of training and test/validation samples

The key parameters for EtOH and TP spectra pre-treatment and model calibration are shown in Table 9.

Table 9 - Key parameters of the TP PLS model and the EtOH PI model

Visual spectra evaluation

EtOH

In the course of protein precipitation due to EtOH addition, the spectra showed the occurrence of peaks with consistently identical Raman shifts, which differed in their intensity and were used for modelling. In general, peaks whose intensities decrease over time could be assigned to protein signals as precipitation caused the concentration of dissolved protein to decrease. Vice versa, peaks whose intensities increased overtime could be related to ethanol signals as concentration increased. Consequently, the following spectra ranges could be related to ethanol: 420 - 450 cm ¹, 830 - 915 cm ¹, 1020 - 1100 cm ¹, 1250 - 1300 cm ¹, 1410 - 1500 cnr¹ and 2650 - 3500 cm ¹ (Figure 10).

Total protein

In contrast, other Raman spectra ranges can be assigned to proteins, more precisely protein typical bonds: 480 - 830 cm ¹, 915 - 1020 cm ¹, 1113 - 1228 cm ¹, 1310 - 1398 cm ¹ and 1506 - 1800 cnr¹ (Figure 11).

The Raman spectra of thawed, and thawed and centrifuged supernatant were also analysed. Thawing and centrifugation had an influence on the Raman intensity, but was not crucial to the Raman shift. Furthermore, thawing and centrifugation did not significantly impact on concentration determinations when using consistent at-line measurement conditions.

Calculated model errors

Based on chosen model parameters (Table 9), PEAXACT software calculated the coefficient of determination (R²) and root mean square errors for calibration (RMSEC), cross validation (RMSECV) and prediction (RMSEP), which served as an initial quality attribute for model performance (Table 10). The percentages were calculated by dividing the absolute value by the maximal concentration value and multiplying it by a factor 100. Table 10: Overview of calculated model errors for EtOH model and TP model

TP concentration range: 20.4 - 35.7 g/L.

The percentual value enables a simpler assessment and was calculated based on a division on the maximum concentration value each (EtOH: 198.093 g/L, TP: 35.7 g/L). As a result, EtOH PI model was characterized by an absolute RMSEC of 4.96 (2.54%), an absolute RMSECV of 5.05 (2.58%), an absolute RMSEP of 2.489 (1.48%) and R2 reaching 0.99. In addition, TP PLS model was characterized by an absolute RMSEC of 0.77 (2.16%), an absolute RMSECV of 0.79 (2.21%), an absolute RMSEP of 0.64 (1.78%) and R2 reaching 0.96. Both models indicated promising predictions since calibration and prediction errors were at low levels. An overfitting of the model for the training dataset was not present as no significant increase from RMSEC to RMSEP was found.

Peak integration model for EtOH

For EtOH, a peak integration approach was tested due to the non-overlapping peak (peak: ~ 879 cm ¹) and intensity increase caused by the EtOH addition. Figure 12 shows the linear correlation between signal intensity and EtOH concentration one of the most prominent peaks, which was tested for building a Peak Integration model since it was a characteristic and nonoverlapping high intensity peak which was not interacting with other Raman active substances/elements in this matrix and did not exist in the absence of EtOH (before precipitation).

Partial least squares model for TP

Proteins could be identified as Raman signals in areas that decreased over the course of precipitation and correlated with a decreasing protein concentration. These areas could be attributed to Raman shifts of ~1655 cm ¹, which showed the maximum Raman intensity and corresponded to amide I vibrations, ~1350 - 1300 cm ¹, which related to deformation of CH bonds, and ~1003 cm ¹, which was associated with phenylalanine for example. However, Raman signal intensity did not correlate linearly with the decreasing protein concentration. A PLS modelling was selected for TP concentration prediction.

Predictive ability of the models

To evaluate the predictive ability of those models, the predicted concentration values were compared to the actual measured values (true values) with respect to their absolute and percentage deviations (n = 178). Hereby, the corresponding uncertainty (95%) was calculated by the software for every sample. For simplified representation, the mean absolute percentage error (MAPE) was represented for each model (Table 11). In addition, the minimum deviations were tabularly contrasted with the maximum deviations to outline the range of deviation and beyond that the amount (n) of deviations > 10% was considered for evaluation.

Table 11 - Overview of main evaluation parameters for EtOH and TP models

Out of 178 observations

** Excluding values of lower quantitation limit (<0.51), out of 166 observations

Evaluation of the EtOH model showed highest deviations in the absence of EtOH (before precipitation), This effect was significantly reduced by excluding EtOH free conditions. The EtOH PI prediction model delivered promising predictions for protein precipitation using the pre-processed cryo-poor plasma fraction. Assessing the TP model, low MAPE, low maximum deviation and a small number (n) of deviations were achieved, demonstrating a high predictive ability.

Predictive ability of the models for different EtOH dosage speeds

The predictive ability of the built models was evaluated for the six parallel runs where the ethanol dosage speed was different. The model was validated against the six runs conducted by utilizing a faster precipitation time. The software calculated the RMSEP for this validation dataset based on the predicted concentration values which were compared with the true concentration values. These in turn were compared to the actual measured values (true values) regarding their absolute and percentage deviations (n = 164). Again, the corresponding uncertainty (95%) was calculated by the software for every sample. For simplified representation, the mean absolute percentage error (MAPE) was represented for each model (Table 12). In addition, the minimum deviations were tabularly contrasted with the maximum deviations to outline the range of deviation and beyond that, the amount (n) of deviations > 10% was considered for evaluation.

Table 12 - Overview of main evaluation parameters for EtOH and TP models applied to the six parallel runs

EtOH concentration range: <0.50677 - 202.483 g/L;

TP concentration range: 20.1 - 36.6 g/L;

** Out of 164 observations;

*** Excluding values of lower quantitation limit (<0.51), out of 153 observations

Comparing the main evaluation parameters between the model building set (Tables 10 and 11) and the validation set (Table 12), RMSEPs increased slightly when applying the built model to different EtOH dosage speed conditions. While MAPE values, max deviations as well as the number (n) of deviations increased, the uncertainty (95%) and min deviations for EtOH and TP remained the same.

Discussion

For the prediction of TP during protein precipitation step using pre-processed cryo-poor plasma, a PLS model was built, which achieved low calibration, cross validation and prediction errors, and high predictive abilities. For the prediction of EtOH during protein precipitation using pre-processed cryo-poor plasma, a PI model using the EtOH peak at ~880 cnr¹ was built, which demonstrated low calibration, cross validation and prediction errors, and high predictive abilities.

Therefore, the models are well suited for the prediction of analyte concentration during plasma fractionation. Furthermore, the prediction models could be applied to a protein precipitation step with a different EtOH dosage rate, with only slightly increased prediction errors.

Example 8 - Raman spectroscopy as a tool for real-time monitoring of the quality of paste resulting from the first cold ethanol precipitation step

This study was conducted to assess whether Raman spectroscopy can be used as an inline measurement tool to evaluate total protein and ethanol concentration in Kistler-Nitschmann (KN) and Cohn IgG comprising samples as well as samples from Kistler-Nitschmann (KN) and Cohn comprising residual proteins of an albumin enriched fraction, and the effect of filter aid present in the dissolved pastes on Raman measurements. Raman spectroscopy was applied as an inline technology to generate real-time data. This Example explored the use of Raman during a dissolution step.

Materials

Cryopoor plasma (CPP) and CPP after a purification step was used as starting material for the standard fractionation process such as KN or Cohn.

These KN and C samples were divided into three groups.

Methods

KN and C samples were resuspended in buffer in appropriate quantities (ratio: 1 part of paste + 5 parts of buffer (w/w)). The KN buffer was prepared with 17.06 g glacial acetic acid 100% and 21.18 g sodium acetate per 2 kg of the buffer with pH adjusted to 4.37, whereas the C buffer was prepared with 10.76g glacial acetic acid 100% and 35.22 g sodium acetate per 2 kg of the buffer with pH adjusted to 4.81 . Resuspended samples were immediately measured using the inline Raman device. Thereafter, all the samples were centrifuged to collect filter aid-free supernatants, which were also subjected to Raman, pH, conductivity, OD and Cedex Bio HT measurements, wherein the Cedex measurement provided the concentration of EtOH, IgG, total protein, albumin and OD protein. As negative controls, both paste resuspension buffers were measured with and without filter aid, respectively.

The Raman measurements were performed (wavelength 2 = 785 nm; exposure time = 50 s; count = 3) with calibrated equipment. The measured Raman spectra were analyzed based on an individually developed preprocessing strategy using PEAXACT software (Version 5.9). After the global range was defined and regions were excluded during preprocessing, Raman shifts from 370 - 427 cm ¹, 490 - 1567 cm ¹ and 2680 - 3100 cm ¹ were used for further evaluation. An additional preprocessing step for baseline correction using “Rubber band subtraction” was applied to eliminate the background. The remaining background was reduced by applying a baseline node at 1425 cm ¹ to pull down the peaks to the baseline. To exclude possible variance between the different probes and Rxn2 analyzers, a standardization is applied based on the peak area of the probe peak from 380 cm ¹ to 420 cm ¹ with a linear fit baseline.

Determination of total protein content in the pastes

A peak integration model using the peak area of the 1003.23 cm ¹ peak (Figure 13) was established to predict the protein concentration of the dissolved paste containing filter aid and after filter aid removal, respectively. Each model was calibrated based on spectral data of pastes containing filter aid and not containing filter aid from of two of the three groups. The dissolving buffer with and without filter aid was also included in the respective training dataset to represent 0 mg/mL of total protein. The third group was used as the test dataset.

Using the predicted vs. true plot of the calibration of the model used for filter aid containing pastes, a precise prediction could be shown for the total protein content with a low RMSEP of 1.94 (measurement range: 0 - 30 mg/mL) and a difference of -2 to +2 mg/mL (Figure 14).

The model built for the prediction of total protein concentration after filter aid removal necessitated an additional baseline node to separate the evaluated peak at a Raman shift of 1003.23 cm ¹ (Figure 13). The model was calibrated with the identical dataset used for the model above. The predicted vs. true plot shows that the prediction is even more precise (Figure 15). This model also uses a linear calibration function resulting in an RMSEP of 0.93.

Although the RMSEP of the model for filter-aid-containing pastes is higher, the Uncertainty value is lower compared to the model without filter aid (3.3 - 3.7 g/L vs. 3.9 - 4.4 g/L at 95%). However, as RMSEP is an overall metric and “Uncertainty” is calculated for each spectrum, such a possibility can occur that “Uncertainty” is higher although the RMSEP is lower.

Determination of Ethanol concentration in the pastes

Figure 16 demonstrates that dissolved pastes containing filter aid also showed peaks at the same positions that were evaluated as being relevant for EtOH. One of the most representative peaks was at 880 cnr¹ but in this region, the dissolving buffer also showed a specific peak which additionally influenced the ethanol peak. Therefore, the peak integration model was built based on a smaller but isolated peak at 1045.9 cm ¹. For the spectra of dissolving paste after filter aid removal shown in Figure 17, the linear correlation was even better.

Two models were calibrated using the filter aid dataset and the filter aid-free dataset respectively. Again, two of the three groups were used as training samples and the third group for the testing of the model against unknown data. The model for dissolved pastes containing filter aid was calibrated with a linear function resulting in a RMSEP of 2.07 g/L and a difference of -2 to 2 g/L EtOH (Figure 18).

The model for dissolved pastes after filter aid removal was calibrated against a quadratic function as this reduced the RMSE errors significantly compared to a simple or linear function. Choosing quadratic calibration function resulted in a RMSEP of only 0.71 g/L and a difference in a range from -1 to 1 g/L (Figure 19).

For spectra of pastes after filter aid removal, the peak at 1085.81 cm ¹ showed an even higher linear correlation between the peak area and EtOH concentration (Figure 17). Therefore, another model was built accordingly using this peak instead of the one at 1045.9 cm ¹. The pre-processing was slightly adjusted using alternative baseline nodes to elaborate the peak (using baseline nodes at 1035.43 cm ¹, 1058.73 cm ¹ and 1066.38 cm ¹ for the 1085.81 cm ¹ model, as compared to 1045.43 cm ¹, 1066.00 cm ¹ and 1425.00 cm ¹ for the 1045.9 cm ¹ model). The calibration with the same dataset used for the other models reached a RMSEP of 0.58 if using a quadratic calibration function (Figure 20).

Table 13 - Summary of all EtOH prediction models created in this example

Discussion

A peak integration model was established based on the protein peak 1003.23 cnr¹ which showed a linear correlation based on total protein values obtained from at-line analytical data. In total, three peak integration models were established for the determination of ethanol concentration in the pastes. The most representative peak for ethanol was at approximately 880 cm ¹, however a peak arising from the dissolving buffer overlapped with it. Additionally, it was shown in Example 6 that the peak area at 880 cm ¹ does not correlate linearly with the related ethanol concentration. Therefore, the peak integration model was built based on a less intense but isolated peak at 1045.9 cm ¹ or 1085.81 cm ¹. For the spectra of dissolved pastes after filter aid removal, a linear correlation between ethanol concentration and peak area could be observed.

Example 9 - Raman spectroscopy as a tool for real-time monitoring of the quality of Paste V (albumin precipitate) resulting from cold ethanol precipitation

This study was conducted to assess whether Raman spectroscopy can be used as an inline measurement tool to evaluate total protein and ethanol concentration in albumin samples of Fraction or Precipitate C, and the effect of filter aid present in the dissolved pastes on Raman measurements. Raman spectroscopy was applied as an inline technology to generate realtime data.

Materials

Following the steps of generating KN samples from CPP, the albumin manufacturing process is performed to produce an enriched albumin product from the following starting intermediates generated using cold ethanol fractionation: Precipitate C (PPT C from KN process) or Fraction precipitate (paste from Cohn process).

Methods

The samples were resuspended in water in appropriate quantities (ratio: 1 part of paste + 2 parts of water (w/w)). These resuspended samples were immediately measured using the inline Raman device. Thereafter, all the samples were centrifuged at 4000 g for 25 min and 22 °C, their filter-aid free supernatants were collected, and also subjected to Raman and Cedex measurements. As negative controls, ethanol of various concentrations with filter aid and after filter aid removal were also measured. The filter aid-free supernatants were remeasured after one freeze/thaw cycle by freezing them at <-60°C overnight and thawing conditions of +37°C for 20 minutes. These freeze/thawed samples were used as “test” dataset for samples from which filter aid was removed.

The Raman measurements were performed (wavelength 2 = 785 nm; exposure time = 20 s; count = 3) with calibrated equipment. The measured Raman spectra were analyzed based on an individually developed preprocessing strategy using PEAXACT software (Version 5.9). The range used after defining the global range and excluding ranges was 340 - 500 cm ¹ and 800 - 3100 cm ¹. An additional preprocessing step for baseline correction using “Rubber band subtraction” was applied to eliminate the background. The remaining background was reduced by applying a baseline node at 857, 915, 1036, 1066 and 1113 cm ¹ to pull down the peaks to the baseline. To exclude possible variance between the different probes and Rxn2 analyzers, a standardization is applied based on the peak area of the probe peak from 380 cm ¹ to 420 cm ¹ with a linear fit baseline.

Identification of peak related to total protein

The difference in protein concentration could be observed for the spectra of pastes before and after filter aid removal when focusing on the area of the representative protein peak at approximately 1003 cm ¹ (Figure 21).

Two peak integration models both using the area of the 1003 cm ¹ peak were established to predict the protein concentration of the dissolved paste containing filter aid and after filter aid removal, respectively. Each model was calibrated based on spectral data of pastes containing filter aid and not containing filter aid. The dissolving buffer (water) with and without filter aid was also included in the respective training dataset to represent 0 mg/mL of total protein. The remeasured dataset of samples after filter aid removal was used as the test dataset for the model of the matrix without filter aid.

Identification of peaks related to total ethanol

Figure 22 demonstrates that dissolved pastes containing filter aid also show peaks at the same positions that were evaluated as being relevant for EtOH. The most representative peaks are at around 880 cm ¹, 1047 cm ¹ and 1086 cm ¹. Therefore, the peak integration models were built based on all three peaks. For the spectra of dissolved pastes after filter aid removal (Figure 22B), the linear correlation is better compared to the spectra of dissolved pastes containing filter aid (Figure 22A).

To investigate the effect of filter aid on the behaviour of ethanol present in the paste (Figure 22C & 22D), linear correlation plots of WFI-ethanol mixtures of various concentrations were also created.

Building of peak integration models

Two separate models were built based on sample data with and without filter aid as filter aid appears to suppress the Raman signal and therefore the peak area of the same concentration is not comparable between spectra of samples with and without filter aid (Figure 22). The same preprocessing (see Methods in this example) was used for the models with filter aid, as well as for without filter aid.

Using the predicted vs. true plot of the calibration of the model used for filter aid removed pastes, a precise prediction could be shown for the ethanol content using a linear calibration function with a low RMSEP of 2.49 (measurement range: 0 - 72.3 g/L) and a difference of -2 to +2 g/L (Figure 23).

For the feature “total protein”, the model for filter aid-removed pastes was calibrated with a linear function resulting in a RMSEP of 3.48 g/L and a difference of -4.8 to 4.8 g/L total protein (Figure 24).

As no test data was available, the RMSEP could not be obtained for total protein and ethanol in filter aid-containing samples (Figure 25 and Figure 26). Therefore, only RMSEC and RMSECV could be estimated. The EtOH prediction model for pastes with filter aid was calibrated with a linear function resulting in a difference of -8.4 to 8.4 g/L total EtOH; the total protein prediction model for pastes with filter aid was calibrated with a linear function resulting in a difference of -5.5 to 5.5 g/L. Table 14 - Summary of calibration for all prediction models created in this example

Discussion

A peak integration model for total protein was established based on the protein peak at approximately 1003 cm ¹ which showed a linear correlation based on total protein values obtained from at-line analytical data.

An integration model built on the ethanol peaks at approximately 880 cm ¹ , 1047 cm ¹ and 1086 cm ¹ was established for the determination of ethanol concentration in the pastes, however for spectra of dissolved pastes containing filter aid, a reduced linear correlation between ethanol concentration and peak area could be observed. In order to decipher whether filter aid was involved in the high variation seen with regard to the ethanol peaks, water-ethanol mixtures with increasing ethanol concentrations containing filter aid were used for further analysis. However, it was revealed that a linear correlation between ethanol concentration and peak area could be observed in mixtures containing filter aid as well as after its removal. For the spectra of dissolved pastes after filter aid removal, a stronger linear correlation between ethanol concentration and peak area could be observed.

Example 10 - Raman spectroscopy as a tool for real-time monitoring of Fraction IV precipitation

This study focuses on the Fraction IV precipitation step and the feasibility of using in-line Raman spectroscopy to monitor protein and ethanol (EtOH) concentrations during protein precipitation with increasingly high volumes of ethanol to generate real-time data with the aim to understand protein precipitation behaviour.

Materials

Two identical sets of experiments were run - Run 1 and Run 2 with an albumin enriched filtrate (l+)ll+lll material. During the course of the fractionation process, real time Raman spectra were continuously recorded at intervals of 20 minutes.

In this example, the albumin enriched (l+)ll+lll filtrate was treated under pH and alcohol conditions to form a fraction IV suspension to which filter aid was subsequently added. imultaneously, samples were collected at each interval and frozen overnight at < -60°C. For at-line analysis, particle-free supernatants of centrifuged frozen samples were collected and subjected to at-line measurements for an estimation of total protein and ethanol concentration. The analytical at-line results for total protein and ethanol content for the respective fraction were used as true value.

Methods

Data from Run 1 was used as the training dataset and Run 2 as the test dataset for model establishment. The Raman measurements were performed (wavelength 2 = 785 nm; exposure time = 10 s; count = 30) with calibrated equipment. The range setting used for further evaluation was 200 - 1800 cm ¹ and 2650 - 3320 cm ¹. An additional preprocessing step for baseline correction using “Linear fit subtraction” was applied to eliminate the background. A smoothing step was applied with a filter length of 5. Standardization is applied based on area normalization.

Modelling of ethanol concentration

The application of pre-processing parameters on the spectra of both IV precipitation runs showed an approximately linear correlation between the peak area and the ethanol concentration present (Figure 27). The inventors observed previously for the spectra from ‘albumin enriched filtrate’, intermediates of the present study - ‘after 1. pH adjustment’ and ‘after 2. pH adjustment’ could not be detected as outliers although their matrices differed in comparison to the product solution during precipitation, especially with regard to their turbidity. Hence, the turbidity does not affect the peak area of the ethanol peak, if the spectra are pre- processed using these settings.

A new model was established and calibrated against the two runs using Run 1 as training data and Run 2 as test data for the evaluation of the model performance on spectra outside the training set. Based on the ‘RMSE vs. Function’-plot (Figure 28), a linear function was chosen as calibration curve type.

The ‘Predicted vs. True’-plot showed a high comparability between recovery line and identity line (Figure 29). A difference of up to 10 g/L ethanol between Predicted vs. True concentration was seen at approximately 270 g/L ethanol. Nevertheless, this difference was still within an acceptable range. The intermediates of the ‘albumin enriched filtrate’, ‘after 1. pH adjustment’ and ‘after 2. pH adjustment’ also lay well within the calibration curve and could therefore also be considered, although their matrices differed in comparison to the precipitation samples. The samples of the post-stirring time also did not show any irregularities.

Modelling of total protein concentration

The following parameters were selected for pre-processing of the Total Protein PLS model: a global range of 0cm ¹ - 5000cm ¹, an excluded range of 1800cm ¹ - 2650cm ¹, a baseline correction of rubber band subtraction, a baseline node of 1370cm ¹, a smoothing filter length of 5, and a standardization of SNV normalization.

Figure 30 shows that a weaker linear correlation between peak area and total protein value is visible. The protein peak signal is relatively weak. Therefore, a PLS model approach is applicable.

A PLS model was implemented and calibrated on the IV precipitation-spectra with the respective at-line analytics values for total protein. Run 1 was used as training data and Run 2 acts as test data. The model was cross validated using the k-fold approach with a grouping of 3, meaning the training dataset was split into 3 subgroups and 3 iterations. For each iteration, two subgroups were used for training and the remaining group was used for model validation. Rank 3 was chosen based on the RMSE vs. Rank-plot rank, as this rank achieves the lowest RMSEP and a high R² (Figure 31). Higher ranks improved the RMSEC but increased the RMSEP.

The ‘Predicted vs. True’-plot (Figure 32) showed a strong comparability between the recovery line and identity line. The difference between the true and the predicted total protein concentration was in a range from -2 to +1 g/L. No overall difference between the train samples and test samples was visible. Also, the prediction for ‘Filtrate l+ll+lll’ intermediates - ‘after 1. pH adjustment’ and ‘after 2. pH adjustment’ showed no large differences.

The Variable Importance in Projection (VIP) plot outlines the area of the spectra which may be important for the prediction of the total protein concentration. A data filter of Raman shift regions comprising the ranges of 480.158 cnr¹ - 830 cm ¹, 915 cm ¹ - 1020 cm ¹, 1112.48 cm ¹ - 1227.55 cm ¹, 1310.22 cm ¹ - 1398.47 cm ¹, and 1505.72 cm ¹ - 1800 cm ¹ was used to select these specific Raman shifts for the modelling of total protein. For this example, although some of these Raman shift regions showed low importance in the VIP plot (Figure 33), these regions were found to be specific for protein measurement. As areas of high importance shown in the VIP plot are related to ethanol in this example, these areas were excluded to not negatively influence total protein prediction. This strategy increases specificity and robustness of the protein model and reduces negative effects caused by the ethanol peaks predominantly present.

Table 15 - Summary of all prediction models created in this example

Conclusion

The spectral intensity of peaks characteristic for ethanol and total protein recorded during the two runs of IV precipitation correlated to their respective at-line analytical data.

Spectra of IV precipitation after preprocessing showed a strong linear correlation between ethanol concentration and the peak area of the specific ethanol peak at a Raman shift of 880 cm ¹. Therefore, a peak integration model was calibrated against the IV precipitation spectra using Run 1 as train samples and Run 2 as test samples. Linear function was chosen as calibration curve type which resulted in a model with a RMSEP of only 6.8 g/L ethanol which equals 2.0 % based on the upper range of the covered concentration.

PLS as a model approach for predicting total protein concentration was possible. The PLS model was calibrated on the IV precipitation spectra. Rank 3 was chosen as it led to the lowest RMSEP. Increasing the complexity of the model by higher ranks, leads to an increase of the RMSEP due to overfitting. Overall, the calibration resulted in an acceptable RMSEP of 1.11 g/L total protein which equals 5.5 %.

Overall, this study showed the possibility of using Raman spectroscopy in combination with predictive modelling to enable real-time monitoring of total protein and ethanol during IV precipitation.

Example 11 - Raman spectroscopy as a tool for real-time monitoring of the KN precipitation (albumin precipitation step)

This study focuses on PPT C precipitation at lab scale and the feasibility of introducing in-line Raman spectroscopy therein to monitor protein and ethanol (EtOH) concentrations during protein precipitation (mainly albumin precipitation) to generate real-time data.

As described herein, PPT C precipitation involves the following. The filtrate IV including the post wash of the filter press is pH adjusted to a pH of 4.80 while maintaining a constant alcohol concentration from the previous precipitation step. The temperature of the NC suspension is at -7.0°C. Materials

KN fractionation was performed with fresh CPP used as starting material. During the course of the PPT C precipitation step of the fractionation process, real-time Raman spectra were continuously recorded at intervals of 20 minutes.

Simultaneously, samples were collected at each interval and frozen overnight at <-60°C. For at-line analysis, particle-free supernatants of centrifuged frozen samples were collected and subjected to at-line measurements for an estimation of total protein and ethanol concentration.

Methods

The Raman measurements were performed (wavelength 2 = 785 nm; exposure time = 10 s; count = 30) with calibrated equipment. Additional measurement settings were consistent with the settings as described in Example 10.

Modelling of EtOH concentration

Ethanol is not added during the PPT C precipitation but is still an important parameter to monitor as the presence of a final concentration of 40% ethanol is important for albumin precipitation. Therefore, the 1.1 M acetic acid buffer used in this example also contains 40% of ethanol, to prevent a change in ethanol concentration caused by the addition of the buffer.

The peak integration model developed for the IV precipitation step in Example 10 was applied on the PPT C precipitation data. The Predicted vs. True Plot (Figure 34) shows a high precision of the prediction for most of the samples. The Predicted vs. Time plot showed that the higher differences occurred during the start of the precipitation and that predictions 100 min thereafter showed a high precision eluding to differences observed due to the changes in the matrix caused by buffer addition.

Overall, the model validation reports a RMSEP of 11.9 g/L ethanol which equals 3.5 % when taking the concentration range into account. Modelling of total protein concentration

The PLS model developed for the IV precipitation step in Example 10 was recalibrated using the PPT C precipitation spectra. As this study contains only one set of PPT C precipitation samples, an evaluation of the performance of a recalibrated PLS model cannot be conducted due to the missing test dataset which is necessary for the evaluation of the model performance on data outside the training dataset.

To demonstrate the general application of the PLS model used forthe prediction of total protein during the previous precipitation, a duplicate of the model is calibrated on the Raman spectra of the PPT C precipitation. A low rank of 2 resulted in a low RMSEC and RMSECV (Figure 35).

The Predicted vs. True plot shows a high comparability between the recovery line and the identity line which is the basis for a precise prediction (Figure 36).

Conclusion

Using the peak integration model calibrated on IV precipitation in Example 10, spectra for the prediction of ethanol concentration during PPT C precipitation was evaluated as possible. Although the model was not recalibrated on PPT C precipitation spectra, only a moderate increase for RMSEP to 11 .9 g/L (3.5 %) was perceived. Especially after 100 min from the start of precipitation, a high precision for the prediction was visible. This shows the high robustness of peak integration models also for a changing matrix.

A duplicate of the PLS model used during IV precipitation with identical pre-processing and data filters were recalibrated on the PPT C precipitation data which resulted in a PLS model with a low complexity (Rank 2) and a strong overlay of identity and recovery line.

Claims

1. A method for determining the concentration of an analyte in a sample obtained from processing of blood-derived plasma, the method comprising applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating test spectra, comparing the test spectra with reference spectra obtained from reference samples having known concentrations of the analyte, to determine the concentration of the analyte in the sample.

2. The method of claim 1 , wherein the test spectra are subjected to peak integration, hard modelling or multivariate data analysis.

3. A method for determining the concentration of an analyte in a sample obtained from processing of blood-derived plasma, the method comprising applying a light source to a test sample obtained from processing of blood-derived plasma, measuring inelastic scattering from the test sample, thereby generating test spectra, comparing the test spectra to a reference data set in the form of a model generated using peak integration, hard modelling or multivariate analysis of processed reference spectra of reference samples having known concentrations of the analyte, to determine the concentration of the analyte in the sample.

4. A method for generating a model to determine the concentration of an analyte in a sample obtained from plasma processing, the method comprising providing training samples obtained from processing of blood-derived plasma, wherein the samples have known concentrations of the analyte, applying a light source to the training samples, measuring inelastic scattering from the training samples, thereby generating training spectra, selecting inelastic scattering or Raman shift regions of interest in the training spectra; optionally applying at least one spectral pre-treatment; generating a model by applying peak integration, hard modelling, multivariate analysis to the spectra to provide a correlation with known concentration of the analyte, thereby obtaining a model for determining the concentration of an analyte in a sample obtained from processing of blood-derived plasma.

5. The method of claim 4, wherein the training samples are obtained from routine manufacture and/or experimental laboratory studies of blood-derived plasma products.

6. The method of any one of claims 1 to 5, wherein the analyte is protein, and the concentration of protein in reference or training samples is determined using the Dumas assay or the immunoturbidimetric assay.

7. The method of any one of claims 4 to 6, wherein the training samples include concentrations of analyte across the concentration range for test sample determination.

8. The method of any one of claims 2 to 7, wherein the multivariate analysis is selected from Partial least squares regression (PLS); peak integration (PI); Hard Modelling; PLS Discriminant Analysis (PLS-DA); Ordinary Least Squares (OLS) regression; MLR (multiple linear regression); OPLS (Orthogonal-PLS); SVM (support vector machines); GLD (general discriminant analysis); GLMC (generalized linear model); GLZ (generalized linear and non-linear model); LDA (Linear Discriminant Analysis); classification trees; cluster analysis; neural networks; and Pearson correlation.

9. The method of any one of claims 3 to 8, wherein the model is a model generated using peak integration (PI), hard modelling or partial least squares (PLS) regression of processed wavelength spectra of samples having known concentrations of the analyte.

10. The method of any one of claims 3 to 9, wherein the model generated is judged using the following statistical parameters:

• Number of latent variables (PLS factors) in the model,

• Bias,

• RMSEC,

• RMSECV,

• RMSEP for independent test samples,

• Rank,

• R²,

• RPD value,

• Uncertainty, and/or

• MAPE.

11. The method of any one of claims 1 to 10, wherein the method comprises applying at least one spectral pre-treatment to the wavelength spectra.

12. The method of claim 11 wherein the spectral pre-treatment is 1^st order derivative, 2^nd order derivative, vector normalization, standardization, Standard Normal variate (SNV) smoothing, or a combination of any of 1^st order derivative, 2^nd order derivative, vector normalization, standardization and SNV, smoothing, optionally wherein the standardization is performed by area normalization.

13. The method of claim 12 wherein the spectral pre-treatment includes selection of specific Raman shift regions and/or data filters.

14. A method for generating a training or reference spectrum or spectra, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma, wherein the test sample has a known concentration of an analyte or is known to have an analyte present; measuring inelastic scattering or Raman shift from the test sample, thereby generating a training or reference spectrum or spectra.

15. The method of claim 14, wherein the steps in the method are repeated with test samples having different concentrations of an analyte.

16. The method of claim 15, wherein the training or reference spectrum or spectra are used to generate a model, wherein the model is generated by using peak integration, hard modelling or multivariate analysis of processed training or reference spectrum or spectra of reference samples having known concentrations of the analyte.

17. The method of any one of claims 1 to 16, wherein the Raman spectrum is obtained using Surface Enhanced Raman Spectroscopy (SERS), resonance Raman spectroscopy, tip- enhanced Raman spectroscopy, polarized Raman spectroscopy, stimulated Raman spectroscopy, transmission Raman spectroscopy, spatially offset Raman spectroscopy, difference Raman spectroscopy, Fourier Transform (FT) Raman, or hyper Raman spectroscopy.

18. A method for determining the presence of, or concentration of, an analyte in a sample obtained from processing of blood-derived plasma, the method comprising: applying a light source to a test sample obtained from processing of blood-derived plasma; measuring inelastic scattering or Raman shift from the test sample, thereby generating test spectrum, comparing the test spectrum with a reference spectrum that contains a Raman signature of the analyte to determine the presence of, or concentration of, the analyte in the sample.

19. The method of any one of claims 1 to 18, wherein the Raman spectrum is obtained using a Raman analyzer configured with a laser or other suitable light source that operates at defined wavelengths, for example in a range of 325 nm to 1064 nm.

20. The method of any one of claims 1 to 19, wherein the light source has a wavelength of at least about 500nm, at least about 525nm, at least about 550nm, at least about 575nm, at least about 600nm, at least about 625nm, at least about 650nm, at least about 675nm, at least about 700nm, at least about 725nm, at least about 730nm, at least about 735nm, at least about 740nm, at least about 745nm, at least about 750nm, at least about 755nm, at least about 760nm, at least about 765nm, at least about 770nm, at least about 775nm, at least about 780nm, at least about 785nm, at least about 790nm, at least about 795nm, at least about 800nm, at least about 805nm, at least about 810nm, at least about 815nm, at least about 820nm, at least about 825nm, at least about 830nm, at least about 835nm, at least about 840nm, at least about 845nm, at least about 850nm, at least about 875nm, at least about 900nm, at least about 925nm, at least about 950nm, or at least about 1000nm.

21 . The method of any one of claims 1 to 19, wherein the light source has a wavelength of about 500nm, about 525nm, about 550nm, about 575nm, about 600nm, about 625nm, about 650nm, about 675nm, about 700nm, about 725nm, about 730nm, about 735nm, about 740nm, about 745nm, about 750nm, about 755nm, about 760nm, about 765nm, about 770nm, about 775nm, about 780nm, about 785nm, about 790nm, about 795nm, about 800nm, about 805nm, about 810nm, about 815nm, about 820nm, about 825nm, about 830nm, about 835nm, about 840nm, about 845nm, about 850nm, about 875nm, about 900nm, about 925nm, about 950nm, or about 1000nm.

22. The method of any one of claims 1 to 19, wherein light source has a wavelength of

500nm, 525nm, 550nm, 575nm, 600nm, 625nm, 650nm, 675nm, 700nm, 725nm,

730nm, 735nm, 740nm, 745nm, 750nm, 755nm, 760nm, 765nm, 770nm, 775nm,

780nm, 785nm, 790nm, 795nm, 800nm, 805nm, 810nm, 815nm, 820nm, 825nm,

830nm, 835nm, 840nm, 845nm, 850nm, 875nm, 900nm, 925nm, 950nm, or 1000nm.

23. The method of any one of claims 1 to 19, wherein the light source has a wavelength of one or more of about 532nm, about 785nm, and about 993nm, preferably 532nm, 785nm, and 993nm.

24. The method of any one of claims 1 to 19, wherein the light source has a wavelength of about 785nm, preferably 785nm.

25. The method of any one of claims 1 to 19, wherein the light source is in the visible spectrum.

26. The method of any one of claims 1 to 19, wherein the Raman spectrum comprises a spectral signal in the visible, near infrared, infrared, near ultraviolet, or ultraviolet (UV) range.

27. The method of any one of claims 1 to 26, wherein the spectra comprise measurements of inelastic scattering or Raman shift from about 5000cnr¹ to about 0cm ¹, preferably from about 5000cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 0cm ¹.

28. The method of any one of claims 1 to 27, wherein the spectra comprise measurements of inelastic scattering or Raman shift from about 3500cm ¹ to about 0cm ¹, preferably from about 3200cm ¹ to about 400cm ¹ or from about 1900cm ¹ to about 400cm ¹; the spectra comprise measurements of inelastic scattering or Raman shift from about 3500cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 350cm ¹, preferably from about 3100cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 350cm ¹; or the spectra comprise measurements of inelastic scattering or Raman shift from about 3320cm ¹ to about 2650cm ¹ and/or from about 1800cm ¹ to about 200cm ¹.

29. The method of any one of claims 1 to 28, wherein the spectra comprise measurements of inelastic scattering or Raman shift from about 1800cm ¹ to about 600cm ¹.

30. The method of any one of claims 1 to 28, wherein the spectra comprise measurements of inelastic scattering or Raman shift from about 1725cm ¹ to about 475cm ¹.

31 . The method of any one of claims 1 to 30, wherein the analyte is total protein, a specific protein, or an alcohol.

32. The method of claim 31 , wherein the analyte is total protein.

33. The method of claim 31 , wherein the alcohol is ethanol.

34. The method of claim 31 , wherein where the analyte is an alcohol, such as ethanol, the spectra comprise measurements of inelastic scattering or Raman shift from about 1455 cnr¹ to about 830 cnr¹ and/or from about 1100 cnr¹ to about 1000 cnr¹ ; from about 3500 cm ¹ to about 2650 cnr¹, and/or from about 1800 cm ¹ to about 350 cm ¹; and/or from about 3320 cm ¹ to about 2650 cnr¹, and/or from about 1800 cm ¹ to about 200 cnr¹.

35. The method of claim 31 , wherein where the analyte is an alcohol, such as ethanol, the peak(s) of inelastic scattering or Raman shift for analysis is at about 437 cm ¹, at about 879cm ¹, at about 880 cnr¹, at about 900 cm ¹, at about 1046 cm ¹, at about 1086 cnr¹, at about 1279 cm ¹, at about 1455 cm ¹, at about 1483 cm ¹, at about 2934 cm ¹, or any combination thereof.

36. The method of claim 31 , wherein where the analyte is total protein the spectra comprise measurements of inelastic scattering or Raman shift between about 475 - about 770 erm 1 , about 920 - about 970 cm ¹ , about 990 - about 1010 cm ¹ , about 1150 - about 1370 cm ¹ , and/or about 1550 - about 1725 cm ¹.

37. The method of claim 31 , wherein the spectra comprise measurements of inelastic scattering or Raman shift between 475 - 770 erm¹, 920 -970 erm¹, 990 - 1010 erm¹, 1150 - 1370 erm¹, and/or 1550 - 1725 cm ¹.

38. The method of any one of claims 1 to 30, wherein the analyte is IgG.

39. The method of any one of claims 1 to 30, wherein the analyte is albumin.

40. The method of claim 31 , wherein where the analyte is a protein, the spectra comprise measurements of inelastic scattering or Raman shift from about 1660 cm ¹ to 1500 cm ¹ , from about 1410 cm ¹ to about 1110 cm ¹, from about 1020 cm ¹ to about 910 cm ¹, and/or from about 830 cm ¹ to about 480 cm ¹.

41. The method of claim 31 , wherein where the analyte is a protein, the spectra comprise measurements of inelastic scattering or Raman shift from about 1800 cm ¹ to about 1510 cm ¹, from about 1400 cm ¹ to about 1310 cm ¹, from about 1230 cm ¹ to about 1110 erm 1, from about 1020 cm ¹ to about 915 cm ¹, and/or from about 830 cm ¹ to about 480 erm 1, preferably from about 1020 erm¹ to about 980 erm¹.

42. The method of any one of claims 1 to 41 , wherein the sample comprising the analyte is a sample obtained from processing of blood-derived plasma obtained from human blood.

43. The method of any one of claims 1 to 42, wherein the sample is obtained or derived from the processing of blood-derived plasma that comprises fresh plasma, cryo-poor plasma, or cryo-rich plasma.

44. The method of claim 43, wherein the plasma is obtained from a number of donations and/or subjects, and pooled.

45. The method of any one of claims 1 to 44, wherein the sample is obtained or derived from hyperimmune plasma.

46. The method of any one of claims 1 to 45, wherein the sample comprising the analyte is a resuspension of a precipitate or paste obtained from blood-derived plasma.

47. The method of any one of claims 1 to 45, wherein the sample comprising the analyte is a filtrate obtained from blood-derived plasma.

48. The method of any one of claims 1 to 47, wherein the sample comprising the analyte is a resuspension of a precipitate or paste, or is a fraction selected from: Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV (including Cohn Fraction IVi, IV₄), and Cohn Fraction V, Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Precipitate B, Kistler/Nitschmann Fraction IV, and Kistler Nitschmann Precipitate C and other similar variant fractions or precipitates.

49. The method of claim 48, wherein the sample comprising the analyte is a resuspension of a precipitate or paste, or is a fraction selected from Cohn Fraction I, Cohn Fraction (l+)ll+lll, Cohn Fraction IV paste (including Cohn Fraction IVi, IV₄ or other similar fractions or precipitates), Kistler/Nitschmann Precipitate A, Kistler/Nitschmann Fraction IV or Kistler/Nitschmann Precipitate B, or other similar fractions or precipitates.

50. The method of claim 49, wherein the sample comprising the analyte is a fraction prepared under conditions to generate Cohn Fraction V paste or Kistler/Nitschmann Precipitate C paste.

51. The method of any one of claims 1 to 50, where the analyte is an alcohol such as ethanol, the concentration of alcohol (e.g. ethanol) in the reference or training samples may be determined using theoretical values, gas chromatography or enzymatic ethanol determination.

52. The method of any one of claims 1 to 51 , wherein a. the methods allow determination of protein concentration of a range of about 0 g/kg to about 10 g/kg, 0 g/kg to 10 g/kg, about 10 g/kg to about 150 g/kg, 10 g/kg to 150 g/kg, about 15 g/kg to about 45 g/kg, 15 g/kg to 45 g/kg, about 20 g/kg to about 35 g/kg, 20 g/kg to 35 g/kg, about 100 g/kg to about 150 g/kg, 100 g/kg to 150 g/kg, about 150 g/kg to about 300 g/kg or 150 g/kg to 300 g/kg; b. the analyte in the sample is predominantly, or contains a significant amount of, IgG, and the methods allow determination of protein concentration of a range of about 0 g/kg to about 15 g/kg, 0 g/kg to 15 g/kg, about 15 g/kg to about 40 g/kg, 15 g/kg to 40 g/kg, about 16g/kg to about 42g/kg, 16g/kg to 42g/kg, about 20 g/kg to about 35 g/kg, or 20 g/kg to 35 g/kg; or c. the analyte in the sample is predominantly, or contains a significant amount of, albumin, and the methods allow determination of protein concentration of a range of about 0 g/kg to about 100 g/kg, 0 g/kg to 100 g/kg, about 100 g/kg to about 150 g/kg, 100 g/kg to 150 g/kg, about 150 g/kg to about 300 g/kg or 150 g/kg to 300 g/kg.

53. The method of any one of claims 1 to 51 , wherein the methods allow determination of alcohol (e.g. ethanol) concentration of a range of about 1% v/v to about 65% v/v, or 1% v/v to 65% v/v, or about 8% to about 40% v/v, or 8% to 40% v/v.

54. The method of any one of claims 1 to 44, wherein the sample comprising the analyte is a turbid solution or suspension.

55. The method of claim 54, wherein the turbid solution or suspension has Nephelometric Turbidity Units (NTU) and/or Formazine Turbidity Units (FTU) of equal to or greater than 10 NTU, equal to or greater than 15 NTU, equal to or greater than 20 NTU, equal to or greater than 25 NTU, equal to or greater than 30 NTU, equal to or greater than 35 NTU, equal to or greater than 40 NTU, equal to or greater than 45 NTU, equal to or greater than 50 NTU, equal to or greater than 55 NTU, equal to or greater than 60 NTU, equal to or greater than 65 NTU, equal to or greater than 70 NTU, equal to or greater than 75 NTU, equal to or greater than 80 NTU, equal to or greater than 85 NTU, equal to or greater than 90 NTU, equal to or greater than 95 NTU, equal to or greater than 100 NTU, equal to or greater than 150 NTU, equal to or greater than 200 NTU, equal to or greater than 250 NTU, equal to or greater than 300 NTU, equal to or greater than 350 NTU, equal to or greater than 400 NTU, equal to or greater than 450 NTU, equal to or greater than 500 NTU, equal to or greater than 550 NTU, equal to or greater than 600 NTU, equal to or greater than 650 NTU, equal to or greater than 700 NTU, equal to or greater than 750 NTU, equal to or greater than 800 NTU, equal to or greater than 850 NTU, equal to or greater than 900 NTU, equal to or greater than 950 NTU, equal to or greater than 1 ,000 NTU, equal to or greater than 1 ,500 NTU, equal to or greater than 2,000 NTU, equal to or greater than 2,500 NTU, equal to or greater than 3,000 NTU, equal to or greater than 3,500 NTU, equal to or greater than 4,000 NTU, equal to or greater than 4,500 NTU, equal to or greater than 5,000 NTU, equal to or greater than 5,500 NTU, equal to or greater than 6,000 NTU, equal to or greater than 6,500 NTU, equal to or greater than 7,000 NTU, equal to or greater than 7,500 NTU, equal to or greater than 8,000 NTU, equal to or greater than 8,500 NTU, equal to or greater than 9,000 NTU, equal to or greater than 9,500 NTU, or equal to or greater than 10,000 NTU.

56. The method of claim 55, wherein the turbid solution or suspension has NTU or FTU of 10 NTU to 100 NTU, 10 NTU to 90 NTU, 10 NTU to 80 NTU, 10 NTU to 70 NTU, 10 NTU to 60 NTU, 10 NTU to 50 NTU, 10 NTU to 40 NTU, 10 NTU to 30 NTU, 10 NTU to 20 NTU, 20 NTU to 100 NTU, 30 NTU to 100 NTU, 40 NTU to 100 NTU, 50 NTU to 100 NTU, 60 NTU to 100 NTU, 70 NTU to 100 NTU, 80 NTU to 100 NTU, or 90 NTU to 100 NTU.

57. The method of claim 56, wherein the turbid solution has a maximum NTU or FTU of 10,000 NTU, 9,500 NTU, 9,000 NTU, 8,500 NTU, 8,000 NTU, 7,500 NTU, 7,000 NTU, 6,500 NTU, 6,000 NTU, 5,500 NTU, 5,000 NTU, 4,500 NTU, 4,000 NTU, 3,500 NTU, 3,000 NTU, 2,500 NTU, 2,000 NTU, 1 ,500 NTU, 1000 NTU, 950 NTU, 900 NTU, 850 NTU, 800 NTU, 750 NTU, 700 NTU, 650 NTU, 600 NTU, 550 NTU, 500 NTU, 450 NTU, 400 NTU, 350 NTU, 300 NTU, 250 NTU, 200 NTU, 150 NTU, 100 NTU or 50 NTU.

58. The method of any one of claims 1 to 44, wherein the sample comprising the analyte is not a turbid solution or suspension.

59. The method of claim 58, wherein the sample or the solution or suspension from which the sample is taken may have Nephelometric Turbidity Units (NTU) and/or Formazine Turbidity Units (FTU) of less than 10 NTU, equal to or less than about 9 NTU, equal to or less than about 8 NTU, equal to or less than about 7 NTU, equal to or less than about 6 NTU, equal to or less than about 5 NTU, equal to or less than about 4 NTU, equal to or less than about 3 NTU, equal to or less than about 2 NTU, or equal to or less than about 1 NTU. In any embodiment, the solution or suspension may have NTU of less than 10 NTU to about 0.1 NTU, about 9 NTU to about 0.1 NTU, about 8 NTU to about 0.1 NTU, about 7 NTU to about 0.1 NTU, about 6 NTU to about 0.1 NTU, about 5 NTU to about 0.1 NTU, about 4 NTU to about 0.1 NTU, about 3 NTU to about 0.1 NTU, about 2 NTU to about 0.1 NTU, about 1 NTU to about 0.1 NTU, about 9 NTU to about 0.1 NTU, about 9 NTU to about 0.2 NTU, about 9 NTU to about 0.3 NTU, about 9 NTU to about 0.4 NTU, about 9 NTU to about 0.5 NTU, about 9 NTU to about 0.6 NTU, about 9 NTU to about 0.7 NTU, about 9 NTU to about 0.8 NTU, about 9 NTU to about 0.9 NTU, about 9 NTU to about 1 NTU, about 9 NTU to about 2 NTU, about 9 NTU to about 3 NTU, about

9 NTU to about 4 NTU, about 9 NTU to about 5 NTU, about 9 NTU to about 6 NTU, about

9 NTU to about 7 NTU, about 9 NTU to about 8 NTU, about 1 NTU to about 5 NTU, about

1 NTU to about 4 NTU, about 1 NTU to about 3 NTU, or about 1 NTU to about 2 NTU.

In any embodiment, the solution or suspension may have Nephelometric Turbidity Units (NTU) of less than 10 NTU, equal to or less than 9 NTU, equal to or less than 8 NTU, equal to or less than 7 NTU, equal to or less than 6 NTU, equal to or less than 5 NTU, equal to or less than 4 NTU, equal to or less than 3 NTU, equal to or less than 2 NTU, or equal to or less than 1 NTU. In any embodiment, the solution or suspension may have NTU of less than 10 NTU to 0.1 NTU, 9 NTU to 0.1 NTU, 8 NTU to 0.1 NTU, 7 NTU to 0.1 NTU, 6 NTU to 0.1 NTU, 5 NTU to 0.1 NTU, 4 NTU to 0.1 NTU, 3 NTU to 0.1 NTU, 2 NTU to 0.1 NTU, 1 NTU to 0.1 NTU, 9 NTU to 0.1 NTU, 9 NTU to 0.2 NTU, 9 NTU to 0.3 NTU, 9 NTU to 0.4 NTU, 9 NTU to 0.5 NTU, 9 NTU to 0.6 NTU, 9 NTU to 0.7 NTU, 9 NTU to 0.8 NTU, 9 NTU to 0.9 NTU, 9 NTU to 1 NTU, 9 NTU to 2 NTU, 9 NTU to 3 NTU, 9 NTU to 4 NTU, 9 NTU to 5 NTU, 9 NTU to 6 NTU, 9 NTU to 7 NTU, 9 NTU to 8 NTU, 1 NTU to 5 NTU, 1 NTU to 4 NTU, 1 NTU to 3 NTU, or 1 NTU to 2 NTU.

60. The method of claim 58, wherein the solution that is not turbid is a solution in plasma processing prior to addition of an alcohol (e.g. ethanol) for the purpose of protein precipitation.

61. The method of any one of claims 1 to 60, wherein any or all steps of the method are performed in-line, at-line, off-line or on-line.

62. The method of any one of claims 1 to 61 , further comprising a step of applying preprocessing and baseline normalization techniques, background correction algorithms or derivative spectroscopy to the spectrum or spectra to manage the background fluorescence.