MSc Chemistry
Analytical Sciences
Literature Thesis
The potential path to real-time decision
making for biopharmaceutical
manufacturing: in-line spectroscopic
multidetector systems
Process analytical technology (PAT)
by
Tobias
Homburg
10759026
January 2020
12 EC
October 2019 to December 2019
Supervisor/Examiner:
Examiner:
Adrian Apetri, Scientific Director
Rob Haselberg, Professor
Johnson & Johnson – Janssen Vaccine and
Prevention
ad lib
logo
1
Table of contents:
Abstract……….. 2
Abbreviations ..……….. 3
Introduction.……… 5
Chapter 1. PAT detectors……… 8
1.1 Review on PAT techniques………. 9
1.2 Additive techniques ……… 18
1.3 Multidetector systems ………. 19
Chapter 2. Impurity detection ……… 22
2.1 Process-related impurities ..………. 22 2.2 Product-related impurities ……… 24 2.3 Product-related modifications ……… 26 Discussion ………. 29 Conclusion ……… 32 References ……… 34 Appendix ……… 45 Supplementary material ……….51
2
Abstract
Currently, most biopharmaceuticals are analysed off-line on large scale. To create these products, strict protocols are followed. This has some advantages, namely reliability and simplicity. Indeed, batch-to-batch variability is higher for off-line analyses even when protocols are followed, thereby reducing the consistency of products. Furthermore, off-line processing is time-consuming, labour-intensive and requires strict following of the protocols, which opens up the possibility for human errors. Moreover, in some cases off-line analyses may be hazardous. To prevent this, pharmaceutical companies are increasingly looking towards process analytical technology (PAT). PAT helps in understanding the correlations between manufacturing process and product. Here, a review will be presented on the different biopharmaceutical-analysing PAT techniques, their (dis)advantages, opportunities and limits. In addition, common impurities in biopharmaceutical samples will be given, as well as their interference and detection. The goal of this article is to educate the reader on current progress in in-line PAT implementation and provide them with a list of available techniques. In addition, the possibilities, limits, benefits and drawbacks of those techniques will be described. New technological developments in the PAT field are also presented. Many detectors are impaired with low resolution, which makes them insufficient by themselves for complex analyses. Therefore, the case is made for increasingly expanding detector numbers to form sophisticated multidetector systems. This is supported by revealing interesting preliminary multidetector systems and their applications. It is further backed by showing analyses that can be done upon combining their measured data.
3
Abbreviations
PAT Process Analytical Technology FDA Food and Drug Administration QbD Quality by Design
CQA Critical Quality Attribute RTRT Real-time release testing CPP Critical Process Parameter USP Upstream processing DSP Downstream processing mAb monoclonal Antibody
IEX Ion exchange chromatography
HIC Hydrophobic interaction chromatography SEC Size-exclusion chromatography
ATR Attenuated total reflection FT Fourier transform
UV/UV-Vis Ultraviolet-Visible spectroscopy PLS Partial least-squares
VP Variable pathlength spectroscopy
FlowVPE Brand name of variable pathlength spectroscopic technology dRI Differential refractive index
MS Mass spectrometry IR Infrared
NIR Near-infrared spectroscopy MIR Mid infrared spectroscopy FIR Far infrared spectroscopy HCP Host cell protein
CD Circular dichroism FL Fluorescence
λmax Wavelength at maximum intensity Rg Radius of gyration
4 RM Mass radius
RH Hydrodynamic radius RR Rotational radius DLS Dynamic light scattering SLS Static light scattering MALS Multiangle light scattering
Mw Weight-average molecular weight MALLS Multiangle laser light scattering IV Viscometry
[η] Intrinsic viscosity
Mv Viscosity-molecular weight Viscostar III Brand name of viscometer
TEM Transmission electron miscroscopy NTA Nanoparticle tracking analysis PTA Particle tracking analysis AFM Atomic force microscopy
PVM Particle vision and measurement
FBRM Focussed beam reflectance measurement VLP Virus-like particle
ρ Shape factor
PTM Posttranslational modification ELISA Enzyme-linked immunosorbent assay SPR Surface plasmon resonance
LC Liquid chromatography DNA deoxyribonucleic acid RNA Ribonucleic acid
EMA European medicines Agency
qPCR Quantitative polymerase chain reaction PEG Polyethylene glycol
5
Introduction
Process analytical technology
In 2004 the U.S. Food and Drug Administration (FDA) posted the process analytical technology (PAT) guidance [33)], which was set out to gain a better understanding of the pharmaceutical manufacturing processes, as described by many others [34)35)58). This initiative has the goal of assuring consistency in pharmaceutical product quality and encourages innovation of new PAT. By implementation of Quality by Design (QbD), pre-analysis is required to understand risks in the process and prevent failure. Systems are often referred to as off-line, at-line, on-line or in-line. The former does not have PAT as goal. As a consequence, off-line systems are not able to control the process in real-time and are limited in their understanding of it.
At-line analysis: the measurement is done at the production area. This makes the analysis closer to real-time than an off-line measurement, but requires manual sampling, which means that the measurement is not in situ. There are higher risks of contamination and there is still chance of human error.
On-line analysis: the process is sampled by an autosampler and subsequently redirected towards the detector(s). Sometimes before that, a sample prep is done. Since this is automated, there are less risks of human errors and process sterility can be maintained. Hence, on-line detection systems are often considered powerful measurement setups. The measurements are, however, not in real-time and the sampling procedure and/or sample prep may affect the analytes.
In-line analysis: the process is seen ‘as is.’ In some cases, a redirecting tube towards the detector is used. In a way, this could be considered sampling without touching the sample. In other cases, direct measurements or fiber optic probes are used. The aspects that are involved in implementing such probes is described by T. Redman et al [38). The fact that no sample prep is done, means that in-line systems are deprived of many detectors. As such, spectroscopic detectors are the confined choice in such systems. Which also means that many high resolution techniques, such as nuclear magnetic resonance and X-ray crystallography, are inaccessible. Further, detector calibration for both on- and in-line systems remains tricky. S. Kadam et al describe steps to do this for on-line systems [68). Nevertheless, line detection is the golden standard of process automation, because it allows in-process measurements without having to change the in-process conditions. A detector ideally does real-time in situ measurements with high resolving power, while also being sensitive, non-destructive and robust [39). Currently, this hypothetical detector does not exist, but a system can come a long way by combining of multiple detectors with some of those properties.
Regulatory aspects make PAT implementation difficult [35), but for pharmaceutical companies there is a lot to be potentially gained [37)]:
• Reduced production time.
• Reduced lab work, less possibilities for contamination and less exposure to dangerous situations (e.g. high pressure, radioactive or toxic material).
• Improved process efficiency, which leads to reduced waste and energy consumption. • Cost reduction, due to all the above.
6 • Real-time release testing (RTRT).
In biopharmaceutical manufacturing, small initial variations in the process can lead to large variations in the final product. Therefore, product quality has to be evaluated along the process. In order to do this, critical quality attributes (CQAs) must be defined and monitored, so as to know that the process conditions are right. This allows for development of optimal process conditions. So called critical process parameters (CPPs) are determined and must be maintained to consistently obtain CQAs. This minimizes batch-to-batch variability. First and foremost, analytical techniques are necessary to monitor CQAs. Subsequently, data generated by these techniques is often processed by mathematical and statistical methods to extract relevant information. In this article, there will be no elaborations of these methods. For this, reference is made to others, since many have described these methods in detail and used them successfully [32)37)44)48)105). Using this processed information, process condition adjusting technologies are added to the system. In essence, this is RTRT [58),139). RTRT is the endgame of PAT, since it would allow real-time in situ evaluation of CQAs. In-line systems are the only feasible systems for RTRT, since they measure the process as it is. This is crucial for a system that allows fast decision making. Measuring in real-time demands fast analysis, which often translates into lower accuracy and/or less information about the product. This time vs information trade-off further limits detection capabilities.
Currently, PAT is mostly applied to manufacturing of small molecules [35)]. Although uncommon, there are projects that have managed to create intelligent decision support systems [71)106)]. These systems are mostly in the initial phases of lab scale automation [36)]. Actual real-time release is still limited [35)], especially for biopharmaceuticals. This is mainly due to the dependence on organisms and their inherent high variability [58)]. To this day, there have been no regulatory approval of complete and fully self-sufficient functioning real-time release systems for the manufacturing of biopharmaceuticals. Only relatively simple intermediate steps in the process have been regulatorily approved [31)].
The biopharmaceutical manufacturing process
The production of biopharmaceuticals proceeds in three stages [22)37)]. The first stage is cell culturing, also known as upstream processing (USP). To automate this process, multiple analysers and regulators are necessary to monitor and maintain the process. Monitoring biopharmaceutical concentrations is difficult in this stage, due to the high abundance of other (proteinaceous) compounds. Hence, analysers are often chosen for their ability to monitor secondary process indicators, such as metabolite, nutrient and biomass content. On-line analysis of this stage is relatively well established [37),31)], as feasibility of many detectors has been proven and extensive organism-specific metabolome databases have been created.
Next, cells are harvested and crude purification steps are undertaken to remove multiple components from the cell culture fluid. These steps may include cell lysis, centrifugation, flocculation, liquid-liquid extraction and filtration [166)] and they are intended to remove whole cells, cell substrates and particles. In this stage, analysers suitable for large particles are mainly useful to see whether the cells and debris has been removed properly.
Finally, the target protein is isolated from the sample and purified. This stage is also known as downstream processing (DSP) and includes separation by chromatographic techniques, reconcentration by diafiltration and (remaining) virus removal by ultrafiltration [109)]. CQA evaluation is most interesting in this stage, since most interfering compounds have already been removed and only co-eluting compounds have to be differentiated. Therefore, direct information can often be gained on the target compound and process conditions can be correlated to CQAs.
7
As an example, the first step in DSP purification of monoclonal antibodies (mAbs) is universally protein A affinity chromatography. This is followed by several separation steps, like ion exchange chromatography (IEX), hydrophobic interaction chromatography (HIC), size-exclusion chromatography (SEC) or affinity chromatography (AC) [18)19),20),24),25)]. Problematically, the eluate of these separation techniques is rather different. Other than different elution order, the composition necessary for separation also widely differs. To illustrate, HIC and IEX use high levels of salt, while AC may use e.g. high pH and SEC specific solvent systems. Therefore, a detector may be perfect after one step, but horrible after another. Here, no more attention will be given to these separation techniques, since they have already been described extensively and this article is directed specifically towards in-line detection methods. For more on them, reference is made to others [18)21)27)28)29)].
In all three stages, detection methods are in demand. These methods would provide critical information about the effect of process conditions on product quality. In that light, in-line PAT detectors are described along with their capabilities, benefits and limitations. In addition, common impurities and their current off-line as well as PAT detection methods are described. Furthermore, results of the latter are shown and their systems described. On account of the results, the case is made for multidetector systems. As will be shown, even low-resolution detectors allow relatively complex analyses by coupling them, thereby allowing monitoring of multiple CQAs.
8
Chapter 1. PAT detectors
In the first chapter, special attention is given to techniques that are used as PAT, their PAT applicability, analytical significance, interesting applications that have already been analysed by the techniques and more. An overview of the PAT techniques was made (see table 1). In the table, some techniques show the ability for multiple levels of analyses. However, these are rarely able at the same time. In the next section, there will be information about technologies that are not necessarily about the product creation itself, but are definitely helpful tools when creating a PAT system. In the last section, interesting in-line and on-line multidetector systems and their applications will be shown.
Table 1: PAT detectors and their properties for protein analysis.
Properties & techniques
PAT applic-cability
Sensitivity Main in-formational significance
Technical limitations Remarks
Ultraviolet-Visible spectroscopy
In-line High Quantitative (or tertiary structure)
Target compound must contain chromophores
Low selectivity.
Fluorescence spectroscopy
In-line Very high Quantitative (or tertiary structure)
Target compound must contain fluorophores. Quenching components and light scattering interference in spectra. Quantitative: high selectivity. Differential refractive index
In-line Moderate Quantitative Gradients, high pressures and/or flow rate cannot be used. Sensitive to all changes.
Universal detector
Mass
spectrometry
On-line Very high Primary, quaternary and quantitative
Involatile buffers and high salt cannot be used. Regular clean-ups. Continuous monitoring does not allow high flow rates. Detector conditions are denaturing.
Very high selectivity
infrared spectroscopy
In-line Moderate Secondary structure
High spectrum interference from solvents and high temperature.
Raman spectroscopy
In-line Moderate Secondary (or tertiary) structure
Relevant groups must be polarizable. Fluorescence interference.
High selectivity and no solvent interference
Circular dichroism
In-line Moderate/weak Secondary ( or tertiary) structure
Can only handle low concentrations > small cells > low flow rates. Further, limited to achiral buffers and solvents. Cells need frequent clean-ups.
Mostly done at lower wavelengths, for increased sensitivity.
9
Dynamic light scattering
In-line Moderate Quaternary structure Fluorescence interference. High viscosity samples invalidate Brownian motion assumption. Low resolution technique. Biased towards large molecules. Multiangle light scattering
In-line Moderate Quaternary structure
Size measurement works for particle sizes above 10 nm.
Low resolution technique.
Viscometry In-line Moderate Quaternary structure
Frequent clean-ups. Low resolution technique.
1.1 Review on PAT techniques
Information from measurements can be classified into multiple levels. The first classification is quantitative and qualitative analysis. Quantitative analysis is divided into relative and absolute measurements. Relative measurements are done by using an internal standard, reference, et cetera. Hence, the measurement reflects the analyte in comparison to some other compound. On the contrary, absolute measurements are direct determinations of content. In biopharmaceutical research, qualitative analysis can be divided into four levels: primary, secondary, tertiary and quaternary. Primary structure involves the building blocks of the protein (i.e. amino acid sequence). Secondary structure contains structures that arise from interactions between atoms in the backbone. Examples include α-helices and β-sheets, which are created by (mostly) hydrogen bonds. Tertiary structure is affected by (predominantly hydrophobic) interactions between the amino acid residues within the protein. In essence, these interactions influence the three-dimensional conformation of the protein as a whole. Finally, quaternary structure involves multiple polypeptide chains (subunits) that interact through the same interactions as tertiary structure formation, but then outside the protein. This is the three-dimensional conformation of the subunits combined. In these levels, detectors vary with their measurement capabilities. Some detectors solely measure at one level, while others at multiple levels, as shown in table 1. Later, those detectors will be discussed, but first some spectroscopic considerations will be mentioned, as most of the detectors are spectroscopy-based.
Spectroscopic considerations
For in-line PAT, spectroscopic techniques are the main option. This is because spectroscopic techniques allow non-destructive in situ real-time measurements. Spectroscopic techniques allow multiple detection modes, like transmission, transflectance, reflection and attenuated total reflection. The first two modes, are sensitive to oversaturation of the system and light scattering influences, thus should be used with caution in turbid media. Transflectance is better than transmission for low abundance detection and low sensitivity detectors, as longer pathlengths are used in this mode. The third mode, reflection, provides the user with index of refraction and is mainly used for analysis of solids and hence has value in tablet and powder characterisation [139),140)]. Finally, attenuated total reflection (ATR) is done by total reflection of light in a medium next to the sample. In this mode the penetration depth into the sample is rather small. Therefore, sensitivity is generally low. The sensitivity can be enhanced by increasing the interaction length, since this increases the amount of interactions between the light and sample. As will be evident from later references, this mode is especially useful for in-line PAT in combination with Fourier Transform (FT) operations. Such mathematical operations enable distinction
10
between otherwise overlapping (absorption) bands, which allows for rapid signal-to-noise discernment.
Quantitative analysis
Ultraviolet-Visible spectroscopy detectors (UV-Vis) are often used for the absolute determination of
molar concentration, due to high sensitivity, easy-to-use and fast measurements. For comprehensive description of this technique (and others), reference is made to others [39)]. Using models, like partial least-squares (PLS) calibration, UV-Vis has been used for many analyses. Among which chemical oxygen demand, total organic carbon and nitrate concentration in wastewater [39),40)]. These analyses may be useful for USP. In pharmaceutical analyses, UV-Vis is used for quantification of proteins and peptides. Quantification is mostly done at 280 nm, because aromatic amino acids absorb at that wavelength. Needless to say, this works solely for proteins and peptides that contain those amino acids. Now, due to the time constraints in in-line PAT systems, there is a trade-off between time and amount of gathered data. If possible, it is recommended to record a spectrum, because this would reveal potential process impurities. In some cases, however, detection at 280 nm only may be sufficient. Therefore, preference is situation dependent and should be thoroughly investigated in advance.
Data extraction from UV-Vis spectra can be maximised using variable pathlength (VP) UV-Vis spectroscopy [97)]. In combination with PLS models, distinctions can be made between co-eluting species. FlowVPE is such a VP spectroscopic technology. It is transmission based and can be implemented in-line on lab scale. The pathlength of detection is adjusted by moving a mobile optical fiber perpendicular to the flow through the tubing. This allows measurements of absorption as a function of pathlength (see Figure 6 in appendix). Accordingly, FlowVPE avoids saturation-induced nonlinear behaviour. This promises no necessity for dilutions or background corrections, leaving the sample as is. FlowVPE can been used to differentiate between co-eluting species. As such, N. Brestich et al used the technology to distinguish mAbs from its aggregates [97)]. Herewith, they were able to make pooling decisions based on the mAb purity. Distinction was made through the difference in absorption spectra of mAb and its aggregates. Likewise, host cell and target protein can potentially be distinguished. A measurement cycle was roughly 30 seconds, which will be hard to further decrease, the authors mentioned, due to the accuracy of the optical fibre positioning. Another interesting application of FlowVPE is monitoring of chromatographic protein A capture step [98). This allows real-time monitoring of continuous chromatography equipment. In conclusion, FlowVPE is a promising new development for PAT that will likely be increasingly used.
As alternative, fluorescence spectroscopy (FL) is often used for absolute quantification. Fluorescence spectroscopy (FL) involves excitation of a fluorophore by a photon, which emits a lower energy photon back. Due to the specificity of these excitation and emission energies, FL is a high selectivity and low noise technique. For a more extensive explanation of this technique reference is made to [146)]. Multiple forms exist, of which intrinsic FL and fluorescent dyes are the most interesting for PAT. Dyes, such as 8-anilino-1-naphthalenesulfonic acid (ANS), allow rapid and reliable quantification of protein content [172)]. As alternative, intrinsic FL can be used [170)]. This form is implemented more easily for in-line systems, as nothing has to be added to the sample. However, fluorescent dyes are more robust for quantification, since dyes are insensitive to their environment, whereas intrinsic FL is not. Furthermore, dyes are not dependent on aromatic amino acids. Thus, for quantification fluorescent dyes are mostly chosen. Quenching, i.e. fluorescence suppression, interferes with quantifications done by FL. Therefore, possible quenchers should be identified and (preferably) removed in advance.
11
In USP, FL has been used to measure biomass and cell mass concentration in cell culture [55),56),57). Known for its universal applicability, differential refractive index (dRI) is used for relative quantification. dRI is based on the linear dependence of concentration and refractive index. This dependence is material specific. For this technique, the refractive index is compared between sample and a reference solution, i.e. background signal of the solvent and buffer. This allows mass concentration measurements. dRI responds to all changes in refractive index, which includes fluctuations in temperature, pressure, mobile phase and buffer composition, et cetera. As a consequence, gradient analyses are frequently avoided in this technique. Provided that these fluctuations are properly accounted for, dRI allows easy-to-use and rapid in situ quantification. Currently, dRI detectors are considered low sensitivity. They are also unable to cope with simultaneous high pressures and flow rates. In PAT applications, high pressures and flow rates are often used to increase throughput. New detector types are therefore necessary. Despite all of this, coupling to other detectors makes dRI a favourable technique. Later this will become evident.
Evidently, multiple detectors can be used to determine concentration. Sensitive detectors can be chosen for quantification. Moreover, they are suitable for biopharmaceutical monitoring in many situations. On a side note, quantification of the aforementioned techniques is, to some degree, all interfered by (proteinaceous) impurities, as will be the case for the following techniques. This shows how crucial it is to be able to differentiate between target protein and impurities. More on that will be discussed in chapter 2.
Primary structure analysis
Mass spectrometry (MS) can be used to obtain extensive product information. MS is known for its
high resolution and sensitivity. Tandem MS reveals highly detailed information on primary structure, i.e. amino acid sequence. By using soft ion sources, quaternary structural information (e.g. molar mass) can be obtained too. Therefore, this technique is often used in off- and on-line proteomics [149)]. Additionally, MS can be used to do other relevant biopharmaceutical analyses, such as impurity detection, protein-protein interactions and glycoform analysis. It can also be used for determination of mass concentration, but this requires addition of an internal standard. Problematically, the technique is known to change sample composition (e.g. precipitation or denaturation), samples must be removed from the process stream to do analysis, instrument maintenance is high and buffering is restricted to volatile species. Moreover, some substances, such as detergents and salts, interfere with the measurements; therefore sample clean-ups are often needed. In addition, on-line analyses may take more than several minutes. Overall, MS cannot be implemented in-line and hence measurements are neither in real-time nor in situ. Therefore, data from MS cannot be used for RTRT at this stage of technology evolution. After the fact, however, it may be used in a supplementary way to other techniques. Nevertheless, future developments may soon enable MS as an in-line detector, as improvements are constantly made to the detector. Such developments would have large ramifications for in-line PAT.
Clearly, primary structure analysis techniques are limited. For in-line PAT, there are no technologies available for amino acid sequence determination, i.e. no primary structure analysis techniques. Thus, development of new methods is profoundly needed in this area.
Secondary structure analysis
The first and foremost detector that can be used to analyse secondary structure, is infrared
12
detectors. The entirety of the IR spectral region consists of the near-infrared (NIR), mid infrared (MIR) and far infrared (FIR). While rather arbitrary, NIR is defined to be between 12,500-4000 cm-1 or 0.8-2.5 µm, MIR between 4000-400 cm-1 or 2.5-25 µm and FIR (or terahertz spectroscopy) between 400-25 cm-1 or 25-400 µm [49)]. Specific information about the analyte can be gained from all of these regions. The region that is chosen is situation dependent. However, NIR has the most potential for PAT, as it can obtain the highest signal-to-noise.
Implementing in-process IR spectroscopy used to be problematic, due to the lack of machine mobility and warming of the detector. The latter significantly increases all noise in IR measurements. On the other end, cooling of IR detectors substantially decreases noise [42)], which increases sensitivity of the detector. Nowadays, fiber optic probes allow the machine to be further away from the process, thereby eliminating those problematic aspects. Though, cooling is rather costly and difficult to implement in-line.
IR has many advantages, measurements can be rapidly done in situ and no sample preparation is required. Furthermore, quantitative and qualitative analyses are possible [141)]. IR spectra show absorption bands that belong to certain vibrations. These vibrations are bond specific. The changes in the content of these specific bonds report changes in vibrational capacity of those bonds, which indicates changes in secondary structure. Proteins have many characteristic structures (e.g. α-helices and ß-sheets). These structures have specific vibrational patterns. Changes in these patterns suggest changes in characteristic structures. Hence, IR can be used to monitor in-solution structural changes of protein [41),45),46),47)]. For more on IR spectroscopy in protein analysis, reference is made to others [58),141)].
In many low sensitivity techniques (e.g. IR and Raman), lasers are used. This increases signal, though caution should be kept, as degradation may be induced by lengthy exposure to high intensity irradiation [108)]. One of the main problems with using IR for the characterisation of biopharmaceuticals, is presence of high levels of water. Water absorbs strongly and dominates IR spectra [44)]. In fact, water absorbs so strongly that IR is often used for detection of low abundance moisture content [67)142)]. Water interference in IR spectra is mostly mitigated by use of multivariate tools, but also by use of reference cells that help determining the absorption of the solvent, buffer and column. At present, most reference cells do not have the ability to match changing process conditions. However, there is a promising new development that has more accurate reference cells, called AQS3pro. This technology was developed by Redshiftbio and uses microfluidic modulation spectroscopy (MMS). In MMS, the reference cell fluid is rapidly adjusted to match the background of the sample. This increases sensitivity, dynamic range and accuracy of the measurements. Currently, the measurements can only be done off-line, but future developments to in-line detection would minimize many of the present problems that IR has. As alternative, more active measures against water interference can be undertaken. For example, V. Acha et al showed an ATR-MIR detection system that could lessen interference of water by using a hydrophobically coated sensor [43)]. This is an interesting invention. However, this may not work for proteins, since they may stick to such coatings.
F. Capito et al showed the ability of at-line MIR to quantification between mAbs and aggregates in cell culture fluid [99)]. They were able to differentiate between the two and their method allowed determination of content as low as 1% aggregate. ATR-FTIR and multivariate tools can be used to differentiate host cell protein (HCP) from mAbs [60)]. Multiple articles describe using FTIR to monitor folding, unfolding, refolding and misfolding [100),135),136),137)]. ATR-FTIR can also be applied to monitor protein A column loading [101)]. This method allows semi real-time decision-making.
13
Currently, FTIR works well off-line, but is not feasible for in-line detection yet. This is due to low sensitivity and the amount of time necessary for obtaining an accurate spectrum.
Nevertheless, the amount of applications shows the feasibility of IR as a tool for fast detection of multiple CQAs, which makes it a useful tool for process control purposes.
In USP, IR has purpose too. (ATR-)NIR and MIR have been used mainly in fermentation processes. Here all kinds of target molecules involved in metabolism are monitored, among others fatty acids, glycerol, biomass, methane, ethanol, phosphate, nitric oxide, ammonia content and many more (to exemplify some extractable information from MIR region, see Figure 7 in appendix) [39),43),96). Additionally, NIR has been used to monitor optical density, dissolved carbon dioxide, oxidation-reduction potential and turbidity.
Raman spectroscopy takes on a clear second place. An in-depth description and implementation of
Raman as a PAT can be found elsewhere [50),66)]. Briefly, Raman is based on inelastic collision of photons with specific polarizable groups. These collisions are rare (about 1 in 107), therefore Raman is considered a low sensitivity technique. Additionally, the activity of these collisions is sensitive to its environment, which is also true for the polarizable groups of the peptide backbone. This enables detection of changes in structure [141)]. Raman spectra show clear distinctions between secondary structural patterns, such as α-helices and ß-sheets (see Figure 8 in appendix) [175)].
There are many types of Raman-based spectroscopy techniques. These types, their applicability and examples of their have been extensively described elsewhere [87)]. Raman is done anywhere between 200 and 1064 nm. Doing Raman at UV-Vis has the advantage of high scattering efficiency. Yet in most biopharmaceutical applications a wavelength in NIR region (between 785 and 830 nm) is used to minimize interference of fluorescence. Fluorescent sample components are the main problem of Raman and they make in-line Raman hard, if not unattainable [87),89)]. Raman is already being used extensively for real-time measurements of solid pharmaceuticals [86)]. Applying of Raman for in-solution biopharmaceutical measurements seems to be a growing interest. Reason for this trend is mostly that Raman has many advantages. Raman works on all sample types, has very high selectivity, water (and other solvents) do not dominate Raman spectra and non-destructive measurements can be rapidly done in situ [82)]. Comprehensive description of the many aspects that come with using in-line Raman is shown by R. Hart et al [88)], in which the authors were able to setup a system capable of monitoring a heterogeneous etherification reaction.
Raman can also be used in upstream processing, as it can monitor carotenoid production [53)] and metabolite concentrations in cell culture [54)].
Lastly, circular dichroism (CD) can be used for secondary structure analysis. CD spectra offer unique information about the overall orientation of the target protein that cannot always be gained from IR and Raman spectra. CD is based on the difference in left- and righthanded light absorption of the analyte. This difference comes from analyte chirality. Except for glycine, all (natural) amino acids are chiral, which means their average orientation in proteins can be detected. As these differences are small, CD is considered to be a low sensitivity technique. To account for this, measurement times are usually increased, which is contrary to in-line PAT objectives. Consequently, extractable information from in-line detected CD spectra is often minimal. For more on CD, reference is made to [102)]. CD can be done in multiple wavelength regions, such as IR, near-UV and far-UV. IR CD, also known as vibrational CD, can provide information on the secondary structure of proteins [102)]. Based on the photoelectric effect, IR detection is lower in sensitivity than far-UV. Cumulatively, IR CD is even lower.
14
Therefore, its usefulness as PAT is questionable. Near-UV CD has applications in tertiary structure monitoring, which is discussed later.
Far-UV CD measures mostly peptide bonds. Hence, secondary structure contents can be determined. It should be noted that quantitative analyses of this kind only work if protein concentration (and therefore its differential extinction coefficient) is known. Far-UV CD has been used for monitoring aggregation in mAbs [111)] and protein unfolding as a function of denaturing agent concentration [103)] (for the latter, see Figure 9 in appendix). Of the three regions, Far-UV CD obtains the best signal (highest sensitivity). It is therefore the best candidate for most PAT applications. Absorption patterns have already been correlated to specific structures, such as random coil, α-helices and ß-sheets, -turns and -barrels. This allows for rapid in situ structural analysis. In addition, CD is not influenced by water and most other solvents. As chiral buffers and solvent should be avoided, CD is partially limited in choice. Furthermore, there are structural changes in which the total orientation of the structure remains the same. In such cases, CD spectra do not show changes [111)]. Moreover, CD cells need frequent clean-ups, which makes it hard to work with in a (sterilised) in-line setup [103)]. Lastly, proteinaceous impurities in biopharmaceuticals interfere strongly in CD spectra [113)]. CD spectra can be obtained rapidly, but there is a trade-off with resolution of the spectrum. As always, the decision depends on the level of detail necessary for the analysis.
Decidedly, for in-line secondary structure analysis multiple methods are available. However, high sensitivity detectors are lacking. This is especially unfortunate, since secondary structure PAT analyses would allow defining and monitoring of multiple CQAs. Thus, in-line secondary structure detector choice is conclusively limited.
Tertiary structure analysis
Additionally, CD can be used for tertiary structure analysis. In the near-UV region (250 nm or higher) exclusively cysteine and aromatic amino acids absorb. Absorption of these amino acids is environment dependent. This is also known as the solvatochromatic effect. As hydrophobic (e.g. aromatic) residues are mainly located in the interior of the protein in its native structure, there they are comfortably shielded from (most of) the solvent molecules and the intramolecular interactions of the protein mostly determine their chromophore activity. As such, the electronic ground state and excited state are different from when the residue is interacting with solvent molecules. Upon denaturation, these residues expose more readily to solvent molecules. The change in energy level is detected by near-UV CD as a shift in wavelength at maximum absorption (λmax). Consequently, tertiary structural changes can be monitored using this method, such as unfolding and ligand binding [102),169)].
As alternative to quantification, FL can be used to monitor tertiary structural changes (see Figure 10 in appendix) [96),158),171)]. This is generally done by intrinsic fluorescence of tryptophan. For FL too, changes in signal occur due to the solvatochromatic effect. Other aromatic amino acids (tyrosine and phenylalanine) can be used too, but are frequently not, because of lower fluorescence efficiency. FL for tertiary structure analysis is a low sensitivity method, since the effect is relatively small. The structural changes are often simply detected by following the λmax. Problematically, light scattering commonly contributes to emission spectra [158)], which renders λmax unreliable. Thus, it should be used with caution or other models should be used [159)]. Regardless, FL is commonly used to detect unfolding transitions [78),80)] and ligand interactions [79)]. As PAT, FL has been used to correlate sample purity to fluorescence signal after a misfold separation step [143)]. Using this method, the authors were able to make pooling decisions in semi real-time (within minutes). Reports on calibration, preprocessing and implementation of FL can be found elsewhere [96),145)]. FL does have a few
15
drawbacks in monitoring of biopharmaceuticals. Any impurity with aromatic groups may fluoresce as well (e.g. HCPs), thereby interfering in the spectra. Furthermore, some compounds may cause quenching, i.e. fluorescence suppression. These quenchers should be identified in advance.
Raman can be used for tertiary structure measurements as well. Again, changes in polarizability influenced by the solvatochromatic effect. For instance, the microenvironment of tryptophan or tyrosine can be analysed [82)]. A. Mungikar et al showed that these methods allow monitoring of thermal aggregation (see Figure 11 in appendix). In the spectrum, clear changes are observed. These are directly indicative of changes in polarization of the functional groups in tryptophan and tyrosine, which can be correlated to alternating structure as consequence of aggregation.
Last and (arguably) least, UV-Vis spectroscopy can be used. UV-Vis is a low selectivity technique; because of this hardly any qualitative information can be obtained. Therefore, UV-Vis is not a strong candidate for conformational studies. Despite this, M. Rüdt et al used UV-Vis to monitor changes in tertiary structure of virus-like particles (VLPs) [69)]. This was done by measuring UV absorption at 285 nm (a) and 294 nm (b). By looking at the relative abundance of a and b (a/b-ratio), changes in the tertiary structure were monitored. This a/b-ratio method monitors changes in absorption of tyrosine, due to the solvatochromatic effect. The effect was emphasized through second derivative spectroscopy. As first defined by R. Ragone et al, this method allows detection of structural changes, such as unfolding [70)]. Although an interesting concept, even the smallest amount of impurities would completely invalidate the results, due to the low sensitivity of the method. Hence, this method is not robust enough for most PAT applications.
There are multiple in-line detectors for tertiary structure analysis. For detection, all of them are depend on the solvatochromatic effect. This effect is only small. Additionally, analysis is limited to proteins that contain aromatic amino acids. Except for near-UV CD, which also has the additional ability to monitor cysteine bonds. To conclude, tertiary structure analysis in PAT is limited in choice. Only very low sensitivity methods are available and measurements are mostly limited to aromatic amino acid containing proteins.
Quaternary structure analysis
Technically, quaternary structure is not directly measured by these detectors. However, changes in quaternary structure are detectable by monitoring certain biophysical properties, such as size. Size can be expressed in multiple terms. Now, radius-based terms will be explained (Figure 1 shows a representation). These radius-based expressions assume the particle to have spherical form. Firstly, radius of gyration (Rg) is the radius of a sphere, if it were to be of the same (uniformly distributed) mass as the protein. Secondly, the mass radius (RM) is the equivalent radius of a sphere with the same mass and specific volume as the protein. Thirdly, the hydrodynamic radius (RH) is the radius of a sphere that takes in the same hydrodynamic volume as the protein. Finally, the rotational radius (RR) is the radius of the sphere that is created upon rotation over the geometric center of the protein. The techniques below are based on varying physical properties and therefore measure different radii.
16
Figure 1: Protein with its different descriptive radii. Picture used with permission of J&J.
Dynamic light scattering (DLS), also known as photon correlation spectroscopy or quasi-elastic light
scattering, is a technique that detects changes in light scattering over time. These changes are correlated to the movement pattern of particles in solution. Extensive reviews have been written on the specifics of this technique [123),125). DLS has the ability to measure in situ particle size in real-time. In general, small particles in solution move fast and large particles move slowly. The autocorrelation function describes this phenomenon mathematically. In this function, the analyte is assumed to move according to Brownian motion. Using that assumption, the diffusion coefficient can be obtained from the autocorrelation function (see equation 1 of supplementary material). DLS is considered a low-resolution technique, since it gives an average value. However, sample heterogeneity can be detected in the autocorrelation function, that is, if particle sizes diverge enough and measurement time is sufficient. In turn, RH can be calculated from the diffusion coefficient through the Stokes-Einstein equation (see equation 2).
𝑅𝐻 = 𝑘𝐵𝑇
6𝜋𝜂𝐷𝑡 , (2)
where kB is the boltzmann constant (~1.38 ∙ 10−23
𝑚2𝑘𝑔
𝑠2𝐾 ), T is temperature, η is sample viscosity and
Dt is the translational diffusion coefficient.
In order to obtain the best results, absorptive and refractive indices should be accurately determined before the analysis. To increase sensitivity and resolution, measurement time is frequently increased, which, again, shows the time/information trade-off that is often encountered in in-line analyses. Additionally, wavelength can be decreased, because scattered intensity is proportional to 1/λ4 (this can be reduced from equation 3 and 4 from supplementary material). This increases sensitivity, but simultaneously increases fluorescence interference. These trade-offs are especially important to consider. DLS allows easy-to-interpret and fast in situ measurements. Furthermore, DLS analysis is not interfered by buffer and solvent. However, high ionic strength and viscosity does influence diffusion coefficient estimations.
DLS has been used to characterise aggregates, follow folding events and intermolecular interactions in solution [83),123),124),125),127). In DLS, the detected intensity is proportional to RH6 (radius to the power sixth), which means that large particles strongly overshadow smaller particles [75)]. This makes DLS unsuitable for small macromolecule detection in many complex samples. If applied for that purpose especially, even the tiniest amounts of process impurities, such as cell debris and dust particles, should be spotted. Otherwise, the measurement is worthless [76),83)].
17
Multi-angle light scattering (MALS) is the most common static light scattering technique (SLS). MALS
measures total light scattering of the particles in solution. This is correlated to the absolute molar mass and particle size. To specify, the amount of light scattering is correlated to the weight-average molecular weight (MW) using a concentration (e.g. dRI) detector through the Debye equation (see
equation 3, 4 and 5 from the supplementary material) [167)]. The MALS detector can be used to create
a 𝑘𝑐𝑅
𝜃 plot. Using this plot, Mw can be found by extrapolating to 𝑐 → 0 and 𝜃 → 0 (see equation 6)
[167)].
𝑘𝑐
𝑅𝜃=
1
𝑀𝑤 , (6)
Where k is an optical constant dependent on many system parameters (see equation 4 from supplementary material), c is concentration and Rϴ is the excess Rayleigh ratio dependent on many system parameters (see equation 5 from supplementary material).
For measurements at high pressure, Mw can only be reliably determined of particles above 1 kDa and low flow rates (<5 mL/min) can be used. Mw is an average of all co-eluting compounds combined and is used to follow multiple in-solution transformations. One of which is the oligomeric state of a vaccine protein [85)]. Next, the particle size (Rg) and second virial coefficient (A2) can be calculated using the Mw (see equation 7 and 8 from the supplementary material). However, this determination is only accurate for particles above roughly 10 nm in size. A differential viscometer can be added for lower size determinations. For more on MALS, reference is made to [126)].
Indeed, DLS and MALS both measure light scattering, but the difference is that DLS measures the change in light scattering over a given time period, while MALS detects the total light scattering at a given time interval and averages it. Consequently, MALS measures in real-time. Similar to DLS, MALS is a low-resolution technique, as often only one average values can be obtained. Using a UV-MALS detection system, in-solution mAb and aggregate Mw and concentration can be determined [147)]. B. Patel et al were able to make a pool decision-making system that could identify mAb purity and automatically stopped collecting, when aggregate content became too high. J. Mou et al used MALLS-dRI (MALLS: Laser MALS) to measure Mw across a SEC-dRI chromatogram (see Figure 12 in appendix) [122)]. The authors were able to conclude that their sample was polydisperse. In the SEC-dRI chromatogram only one peak is observed, hence this conclusion would have been overlooked, were it not for the added MALLS detector. As is evident, the combined data of coupled detectors yields interesting results.
While not a new technique, use of viscometry (IV), or viscosimetry, in biopharmaceutical science is still relatively uncommon. A viscometer has the ability to measure relative viscosity. By coupling a detector capable of measuring concentration (e.g. dRI), the intrinsic viscosity ([η]) can be obtained by extrapolation to 𝑐 → 0 in a
𝜂 𝜂0−1
𝑐 vs concentration plot (see equation 9) [173)].
[𝜂] =
𝜂 𝜂0−1
𝑐 , (9)
where η is the solution viscosity, η0 is the viscosity of the pure solvent and c is the concentration. [η] is a measure of the average viscosity contribution of the particles in solution, which is correlated to hydrodynamic volume and viscosity-molecular weight (Mv) through the hydrodynamic volume-intrinsic viscosity relationship and Mark-Houwink equation, respectively (see equation 10 and 11 from
18
supplementary material). These equations were made for synthetic polymers and for Mv its applicability to proteins is unclear. To obtain the hydrodynamic volume, MW must be known, which means a MALS detector has to be coupled. For further elaboration on the technique reference is made to others [118), 121)].
Evidently, MALS coupled to viscometry and DLS can determine a similar biophysical property. DLS ought to be used for large macromolecules, because it is more sensitive towards them, while viscometry should be used for small macromolecules and strong fluorescent samples. The latter is due to interference of fluorescence in the measurements. Disadvantages of viscometry are that instrument cleanup is more extensive than for DLS and the technique is overall less sensitive. There are no comparative studies, so the point at which one technique exceeds the other is unclear. In contrast to DLS and MALS, viscometry is a relative method, so universal calibration is needed [168)].
In 2016, Wyatt introduced a new viscometer, called ViscoStar III onto the market. As of yet, literature on this technique is rare. The biggest problems of old viscometers was that pump pulses created high noise in the measurements. As a consequence, the dynamic range of viscometers was often limited. Conversely, ViscoStar III circumvents these problems by using pump filters and noise cancelers. Consequently, it is currently the lowest-noise viscometer available. This lab scale technique is especially interesting upon coupling to other detectors. Using a (SEC-)MALS-dRI-IV system, the [η], Mw, (Mv,) Rg, RH, and/or the Mark-Houwink-Sakurada parameters can be determined, the latter are related to the material and solvent properties [119),121),122)]. Molecular weight, Rg and RH can be used for posttranslational modification analysis. In chapter 2, more on that will be discussed.
To conclude, there are multiple techniques for in-line quaternary structure analysis. However, all have low resolving power. Furthermore, unlike the previous detectors, these detectors do not measure quaternary structure directly, but rather measure features that can be correlated to quaternary structural changes. Despite this, they all offer specific product properties that are valuable in defining CQAs.
Microscopy-based techniques
Lastly, microscopy-based techniques, such as fluorescence microscopy, transmission electron microscopy (TEM), (nano)particle tracking analysis (NTA/PTA), atomic force microscopy (AFM) and particle vision and measurement (PVM), are all useful techniques for quantification and/or characterisation of particles [71),72),73),74)]. These techniques are often used for their high resolution compared to MALS and DLS [74),75)]. Furthermore, the measurements are not prejudiced towards larger particles as for DLS, which makes their measurements more reliable. However, all microscopy-based techniques need longer analysis times (minutes or more) to obtain those reliable measurements. Additionally, many require sample prep, skilled operators and measurement by transmission. This often makes them unsuitable for in-line detection in a flow through setup.
1.2 Additive PAT methods
This section is about techniques that may not necessarily characterize the biopharmaceuticals themselves, but rather help with process validation and/or control. Most commonly measured process variables in bioreactors are ionic strength, buffering, pH, flowrates, dissolved oxygen, temperature and stirring speed [51)]. The former three can all be monitored and maintained by an in-line conditioning system, for instance. Such technologies have been around for longer and are well established, so no further elaborations on them will be given.
19
In both upstream and downstream processing, one commonly employed technology is fibre optic sampling probes. These probes have made it possible for many spectroscopic techniques (e.g. Raman and ATR-FTIR) to do reliable in situ real-time measurements. The ability for non-contact measurements makes them ideal for in-line PAT. Additionally, the probes are heat and electrically resistant, which allows operation of detectors that are otherwise unable to cope with such conditions [107)]. Therefore, they are an essential part in setting up a well-functioning system. Unfortunately, some sensitivity is lost when using probes. Additionally, deposit builds up on and around probes. This can be mitigated by frequent cleaning and proper positioning of the probes. T. Redman et al describe this in detail [38)]. In order to maintain product quality, the process must be kept properly sterilised and functioning. This implies that the process equipment has to be monitorable without jeopardising the process sterility, which urges the need for such capable technologies. Using FL, M. Pathak et al showed that it is possible to monitor protein A chromatography resin fouling [145)]. Alternatively, ATR-FTIR has been used [101)]. These techniques are useful tools for optimizing resin replacement and clean up without unnecessarily compromising the process sterility.
A considerable mention in process automation in DSP is focused beam reflectance measurement (FBRM). This technique is most commonly employed to detect large particles. Its detection limit is particles of roughly 1 µm [131)]. Biopharmaceuticals are much smaller than this. Therefore, FBRM cannot be used to analyse them. FBRM can be used to detect bubbles, suspension flocculation and emulsions [128)129),130),132)], but also remaining cells after cell removal steps. Other than detecting disturbances, FBRM may even help in characterising them. This allows proper process adjustments. It is important to detect their presence before detectors, such as DLS, measure them, since they are very sensitive to these kinds of process faults. Consequently, these false positives may lead to the wrong conclusions about the process. In upstream processing, FBRM also has an interesting application. FBRM has be used to monitor changes in biomass concentration in cell culture [133)]. Using this technique, non-invasive in situ detection of cell growth can be monitored. In conclusion, FBRM as a process control detector has its uses.
1.3 Multidetector systems
By themselves, most in-line techniques can only give so much information about the process, but combining them allows increasingly complex analyses to be done. Many such multidetector systems have been used for interesting applications. In this section and the next chapter examples of in-line implementable multidetector systems will be shown along with their applications.
UV-DLS-SLS detection system
M. Rüdt et al were able to monitor virus-like particle (VLP) assembly using a UV-DLS-SLS detection system during a diafiltration step [69)]. They used UV spectroscopy at 280 nm to determine VLP concentration. DLS was used for determining z-average (i.e. hydrodynamic size). SLS was used to measure scattered light intensities. Scattered light intensities are correlated to particle size (radius of gyration). Intuitively, the scattered light increases with Mw, Rg and/or concentration (this can also be obtained from equation 7). Accordingly, VLP assembly could be monitored. Essentially, they could have method used a parameter, called the shape factor (ρ) (see equation 12).
ρ = R𝑔
20
By monitoring this parameter, at any time in a spectrum the average particle shape can be determined. Arguably, this would have been more representative. The significance of ρ is represented in Figure 2. As VLP proteins are disassembled, their structure looks roughly like the representation on the right. When the VLP is assembled, its structure looks more like the representation on the left. Therefore, measuring changes in shape factor can be used to monitor assembled VLP content or (larger) unfolding events. This detection system allowed 6 measurements per minute. M. Rüdt et al were able to determine three phases in VLP assembling. In the first phase only disassembled VLP proteins were dominating. In the second phase, VLP assembling occurred until the maximum VLP concentration. In the final phase, VLP concentration started decreasing due to degradation. This was reinforced by off-line SEC measurements, which showed aggregates formed that were unable to assemble. VLP formation was checked by off-line DLS and TEM measurements.
Figure 2: Schematic representation of differences in Rg and RH, that lead to differences in shape factor. Used with
permission of J&J.
Similarly, F. Capito et al used the same detection system to measure drug-loaded liposomes [84)]. They further made the case to support the detection system by adding another detector. Detectors such as analytical ultracentrifugation, differential centrifugal sedimentation, TEM or NTA add resolving power to the detection system. This would allow for complex analyses in real samples. In spite of their truth, these technologies do not allow real-time in situ measurements.
UV-FL-IR-MALS-dRI detection system
D. Sauer et al used prediction models based on data analysis tools and in combination with a UV-IR-FL-dRI-MALS as well as pH and conductivity probes, in order to create a sophisticated detection system [144)]. In the article, the authors compared basic model, a medium model and an extensive model. The basic model consisted of a standard chromatographic workstation. This workstation contained pH and conductivity probe in combination with a UV detector. To obtain maximal information, UV detection was done at 214, 260 and 280 nm. This allows concentration measurements of the peptide backbone (214 nm), DNA/RNA concentration (260 nm) and aromatic amino acids (280 nm). Despite absorption overlap between these chromophores, quantification was possible using data analysis tools.
The medium model had dRI and MALS detector added to the detection system. In the chromatography step, they used a gradient. As already discussed, dRI detector is sensitive to gradients. The authors
21
avoided this by using preprocessing operations. MALS detection was done at three different angles. According to them, these models were both simple to use and implement.
Oppositely, the extensive model was not simple in use. This model further expanded the detection system to include FL and IR detectors. Fluorescence excitation was done at seven wavelengths and emission spectra between 236-795 nm were detected for each excitation wavelength. The longest measurement cycle was that of the fluorescence emission spectra, which took 16 seconds. Thus, this method allows for rapid detection. This measurement cycle is within the time window of relevant process flow rates. An ATR probe was used along with FT operations for the IR measurements. The separation was done with pure water as the solvent composition. The IR detector is also sensitive to the gradient and the solvent composition as a whole. The authors avoided this by subtracting the buffer background using preprocessing operations.
Interestingly, their detection system allowed monitoring of multiple CQAs simultaneously. The authors were able to simultaneously quantify double-stranded DNA, host cell protein and target protein after an ion exchange chromatography step in the presence of other impurities. Quantification was done in actual E. coli cell culture fluid eluate. The eluate was fractionated and the authors knew where to find relevant fractions using their model. They concluded that the extensive model was best at predicting the CQAs. Especially crucial was the IR detector, because it allowed significantly better structural differentiation between target and impurity than the UV detector. The authors mentioned that this system cannot (yet) replace end-product testing, but off-line testing can be reduced, since it allows rapid determination of multiple CQAs.
22
Chapter 2. Impurity detection
In real biopharmaceutical samples, chemists often deal with all kinds of impurities. Whether those impurities are proteinaceous or not, they must (almost universally) be removed completely or reduced to acceptable levels for a product to end up on the market. During DSP steps, most of those impurities are indeed removed, though some amount co-elutes with the target protein [110)]. In order to guarantee CQAs, these compounds must be identified, quantified and removed during the manufacturing process or, otherwise, afterwards. Thus, a well-functioning PAT system contains detection methods capable of measuring target protein without interference of impurities, while also being able to quantify impurity content. Impurities can be process- or product-related. Process-related impurities are created by the manufacturing process and not (entirely) separated from the biopharmaceutical during purification. Examples include process residuals, column leachables and bacterial/viral contaminations. Product-related impurities are the interactions of the product itself. Examples include degradation and aggregation products, unfolding and (unwanted) intermolecular interactions. Here, some common impurities will be given and at least one possible PAT method to analyse them.
2.1 Process-related impurities
Host cell proteins
During the cultivation of cells, the host cell creates proteins for survival. In biopharmaceuticals, these host cell proteins (HCPs) are considered a process-related impurity, due to their many unwanted interactions. Some HCPs degrade the medically active substance (e.g. proteases) [2),3)], while others cause immunogenicity [1)4)]. Therefore, it is of utmost importance that HCP content be quantified. In Chinese hamster ovary (CHO)-cell culture fluid, common (total) HCP levels in e.g. mAbs range from 337,000 to 7,450,000 µg/L [52)]. Currently for biopharmaceuticals, an acceptable concentration of (total) HCP is less than 100 µg/L [5)], which means that methods with high sensitivity must be applied to assure reliable enough detection.
One technique is specifically used in the detection of HCPs, called enzyme-linked immunosorbent assay (ELISA). This technique is known for its high selectivity and sensitivity, which easily enables detection of low abundance predetermined compounds. A disadvantage of using this technique is that the HCP of interest must be known, in order to create the anti-HCP antibodies necessary for ELISA prior to analysis. Additionally, other proteins than the ones specified cannot be detected. Furthermore, variants of ELISA has only been implemented in-line [7)] or at larger scale [8)], not both. Alternatives to ELISA do exist, like surface plasmon resonance (SPR) of antigen-antibody complexes [6)], but for this system prior knowledge of the target HCPs is still necessary. Furthermore, process conditions often induce degradation of the HCP capture structures. As alternative to ELISA, LC-MSN is often used [26),52)]. However, applying this is limited in its ability for real-time detection, as mentioned before. As shown in section 1.3, there are extensive PAT systems able to detect HCP content. Though, simpler detection systems have also been somewhat successful. F. Capito et al have used at-line ATR-FTIR and data analysis tools for analysing mAb-producing CHO-cell culture fluid [60)]. These findings agreed with an ELISA-assay. Unfortunately, the analysis was only able to quantify HCP titers ranging from 20,000 to 200,000 µg/L, which is not enough to guarantee CQA.
23
Host cell (deoxy)ribonucleic acid
Even though still debated, insertion of deoxy- (DNA) and ribonucleic acid (RNA) likely has multiple effects on mammalian cells. Injection (along with biopharmaceuticals) of residual host cell DNA/RNA may cause immunogenicity, mutagenesis and cancer [9),10),11)]. Hence, the FDA and European Medicines Agency (EMA) put strict guidelines on this impurity that mention a maximum of 10 ng per dose is acceptable [12),13)]. As such for an injection of 1 mL, the maximum acceptable dose has a concentration 10 µg/L. This urges the need for very sensitive detection techniques.
Unfortunately for low abundance DNA/RNA, these techniques entail amplification, which is undesired in the product itself, but can be done by process sampling and doing at- or on-line quantifications [15),16)]. Amplification techniques, such as quantitative polymerase chain reaction (qPCR), are fast, cheap and easy to use. Though, the technique does need use of specific primers. As a consequence, only the predetermined DNA sequence will be amplified, which means only predetermined contaminations will be detectable.
Alternative techniques to qPCR use probes that take advantage of chemiluminescence, colorimetric, fluorescent or radioactive labelling [17)]. These probes allow high sensitivity detection [150)], but have to be removed from the process after use or the process has to be representatively sampled. Therefore, just as with qPCR their applicability for real-time decision making remain dubious.
Instead of solely detecting at 280 nm, extending detection towards multiple absorption wavelengths or e.g. a 240-300 nm spectrum would provide significant additional information, because DNA/RNA absorb at 260 nm. As do aromatic acids [151)] and many other phenyl-containing impurities. At the same time, it must be taken into account that recording of a spectrum would increase analysis time and the sensitivity of this detection method is already rather low, as it works for concentrations above ~1 mg/L [150)]. Hence, this technique is not most robust, but it can detect presence of higher DNA/RNA content.
Bacterial and viral contaminations
Other potential hazards in biopharmaceutical products are bacteria and viruses. To avoid infectivity, such contaminations must not be present in biopharmaceuticals. Ultrafiltration is the typical technique used to remove viral contaminations. If the process has been sterilised properly, then only extremely tiny amounts of these contaminations are present to begin with. As complete elimination is impossible, viral and bacterial contamination absence has to be proven. Thus, manufacturers have to be able to guarantee low abundance in the final product. For this analysis, qPCR is again often used. In combination with lysing agents, specific viral and microbial contaminations can be detected [139)]. FTIR has also been used [116),117)]. Unlike qPCR, FTIR has potential for in-line implementation. However, FTIR sadly does not have the sensitivity to provide regulatory-acceptable guarantees. Moreover, presence of water and other interfering compounds further reduces sensitivity. In conclusion, research into other (new) methods is necessary for robust in-line detection of host cell proteins, DNA, viral and bacterial contaminations.
