|Home | About | Journals | Submit | Contact Us | Français|
The success of clinical and commercial therapeutic proteins is rapidly increasing, but their potential immunogenicity is an ongoing concern. Most of the studies that have been conducted over the past few years to examine the importance of various product-related attributes (in particular several types of aggregates and particles) and treatment regimen (such as dose, dosing schedule and route of administration) in the development of unwanted immune responses have utilized one of a variety of mouse models. In this review we discuss the utility and drawbacks of different mouse models that have been used for this purpose. Moreover, we summarize the lessons these models have taught us and some of the challenges they present. Finally, we provide recommendations for future research utilizing mouse models to improve our understanding of critical factors that may contribute to protein immunogenicity.
Therapeutic proteins such as monoclonal antibodies (mAbs), cytokines, blood factors, growth factors and hormones have become an important class of drugs.1 Despite their clinical and commercial successes, however, an ongoing concern with the use of therapeutic proteins is their potential immunogenicity, i.e., some patients receiving these drugs produce anti-drug antibodies (ADAs) which may be associated with various potential clinical consequences.2 Most protein therapeutics are immunogenic in at least some patients3; with some protein products (e.g., interferon beta4) greater than 30–50% of patients experience immune responses that can result in reduced efficacy. Protein immunogenicity depends on multiple factors that are related not only to the product,5–12 but also to the patient (e.g., disease state) and the treatment regimen.5–7 Unfortunately, there are still major gaps in our understanding with respect to the relative importance of each of these factors and how they mutually influence each other. In an attempt to overcome these shortcomings, several different preclinical in silico models13–15, in vitro models13–19 and in vivo models14,15,20–22 have been developed by a number of groups to study protein immunogenicity, and especially the relative contributions of product-related factors. Although at present none of these approaches can completely simulate the course of development of unwanted immune responses in the clinic, the currently available preclinical models are valuable tools for investigating various product-related attributes or different molecule variants. For the purposes of the discussion in this commentary product-related attributes include different types of protein aggregates and fragments, and chemical degradation products. The effect of the presence of foreign materials (e.g., glass, stainless steel, silicone oil) on protein immunogenicity has been studied as well. In vivo models can also be used to some extent for evaluating the relative impact of administration-related factors. Animal models have been particularly popular, because, unlike in silico and in vitro models, they provide an intact immune system. Thus, though they are different than humans, murine models present the only option to study these, as clinical studies on the impact of product-related attributes cannot be performed for ethical reasons. The formation of ADAs, the endpoint of the humoral response, can be directly measured in the in vivo models. In contrast, in vitro assays usually rely on indirect measurements, such as cytokine release or immune cell stimulation which represent a single step in the complex immune response. Although historically a variety of animal models has been used to study immunogenicity of proteins, most of the studies that have been conducted over the past few years to examine the potential importance of various product-related attributes (as defined above) have been conducted in murine models.21–29 Others have used these in vivo models (see Table 1 for an overview) to study the effects of route of delivery,26,30 as well as the potential molecular mechanisms behind immune responses that could help mitigating such reactions in the future.31 The mouse strains used in these studies were very diverse. For instance, mice capable of forming large antibody repertoires and a diverse response of human antibodies following immunization were used to evaluate ADA responses to attributes of a human monoclonal IgG1.22,23 Similarly, a human immunoglobulin G2 (IgG2)-tolerant and immune-competent heterozygous mouse model (Xeno-het) derived from a C57BL/6J wild-type strain expressed both mouse and human immunoglobulin G (IgG) genes, resulting in B-cells expressing human and mouse IgG, and secretion of human and mouse immunoglobulins into serum.21 The cross breeding with the C57BL/6J strain ensured robust B cell signaling and diverse repertoires.
Some groups have also utilized the nude and immune competent BALB/c mouse strains as they are more suitable to distinguish the T-cell dependent and T-cell independent B-cell responses that can be distinguished through the secreted IgG isoforms.32 Some clear trends have begun to emerge on the utility and drawbacks of different murine models as well as on the product quality attributes that elicit the greatest immune response in these model systems. Below we summarize the lessons we have learned from these studies, as well as some of the challenges that they present.
Since the 1960s, there have been numerous studies in which the immunostimulatory effects of aggregates and particles formed from non-murine proteins have been characterized in wild-type mice. Some studies suggest that in the absence of aggregates and particles, nominally foreign proteins were tolerated immunologically. For instance, Dresser reported that formulations of bovine γ globulin were not immunogenic in mice if they were treated by centrifugation to remove insoluble particles prior to injection.33 More recently, Fradkin et al. showed that the immunogenicity of recombinant human growth hormone (rhGH) in mice decreased significantly when aggregate and particle levels were reduced by high-pressure treatment.34 However, in some studies in wild-type mice, the immunogenicity induced by the native recombinant human therapeutic protein is so high that it is difficult to observe any additional effects from factors (e.g., aggregation or chemical degradation) that are believed to increase the risk of immunogenicity.21,35–41 Nevertheless, wild-type mice might be useful for studying effects of therapeutic replacement proteins that are used to treat patients who are unable to produce certain enzymes, hormones and clotting factors; in these patients the therapeutic human proteins might be perceived by the immune system as foreign.
A more recent approach utilizes murine variants of human proteins, as has been done for growth hormone,26,32 interferon-β42 and monoclonal IgG.27,28,43 These models have been useful for investigations on the relative impacts of product-related factors on immunogenicity, as well as immune mechanisms involved in the breaking of immune tolerance (for a description of the concept of immune tolerance, see Goodnow et al.44). A practical advantage of this approach is that the mice in principle are suitable for testing the immunogenicity of any murine protein. In addition, commercially available mouse strains can be used and their breeding is straightforward.
Transgenic mice have been used in several studies of immune responses to therapeutic proteins. Why should one take the trouble to generate transgenic mice that express human proteins of therapeutic interest? The reason for this is that in many cases, patients are receiving fully human or humanized proteins to which they have a high level of immune tolerance, and with the transgenic mouse the same human protein can be used for immunogenicity studies. The mouse model used should mimic a state of tolerance to self-protein that has to be broken to generate ADAs, because the mechanism of immunogenicity may differ from that of a classical response against a foreign antigen, such as in vaccinations.
Transgenic models have been established for investigating the immunogenicity of recombinant human therapeutic proteins, such as recombinant human interferon alfa (rhIFNα),45 interferon beta (rhIFNβ),35,38 rhGH,34 insulin25,46 and monoclonal IgG.21–23,47 The major advantage of this approach is that these animals can be used to study the effects of different factors that may lead to immunogenicity of recombinant human protein drugs for which they are immune tolerant within an intact immune system, just like many patients at the start of a therapy. Disadvantages of this approach (in addition to the differences between humans and animals such as physiology and dosing regimen that are common to all animal model systems) are that the mice can be used for only one type of protein, the expressed human protein may exert unpredictable biological effects in mice, and expression levels may not match those in humans. Also, construction of the transgenic mouse strains is time consuming and the breeding may be cumbersome, depending on the model. Moreover, one should keep in mind that these mice still have a murine immune system and there is no representation of relevant human HLA and antigen presenters, which would be needed to evaluate sequence-based as well as long-term adaptive phase associated immunogenicity risk.
Several factors are crucial for the successful application of a mouse model. The choice of the mouse strain is of great importance, as the immune response is strongly dependent on the strain.38,43,46 For instance, BALB/c and C57BL/6 mice differ in their immunological and functional architecture and their sensitivity to antigenic stimuli. In BALB/c mice, T-cell functionality and the cytokine release capability is much higher than in C57BL/6 mice, making BALB/c mice better models for studying adaptive phase responses. Similarly, C57BL/6 mice are characterized by higher cytostatic activity of splenic NK cells, making them better models to investigate innate phase responses.48 Additionally, C57BL/6 and BALB/c mice differ in their responsiveness to antigen induced pro-inflammatory and immune suppressive cytokines.49
Another important factor is the timing for the withdrawal of blood samples from mice. Because detection of anti-drug antibodies in the presence of circulating drug may be difficult, the optimal sampling schedule may be influenced by the serum half-life of the therapeutic protein in the mice, the applied dose and the dosing schedule. The sampling schedule may also depend on the type of immune response that is anticipated. For example, detection of long-term memory responses may require extended periods of time between doses.
Determining a suitable administration/sampling schedule can be very time-consuming, depending on the protein and the analytical assays used to detect immunogenicity in the presence of circulating drug.50 For instance, a dose of 5 μg/injection and 15 injections over a period of three weeks for hIFNα and hIFNβ transgenic mice, following the protocol originally published by Braun et al.,45 was successful in generating ADAs. However, for monoclonal IgG and insulin this administration protocol did not result in measurable ADA levels even in the (non-immune-tolerant) wild-type mice, and new protocols had to be developed in order to obtain satisfactory results.23,25 It is worth noting that dosing and administration schemes used for these studies in general are not meant to translate directly to the dosing scheme in humans.
Even in transgenic strains, there are several challenges to their use to mimic potential immune responses in humans. Although the mice may be tolerant for the human protein, the predominant background of the immune response is murine, which may limit their ability to predict immunogenicity in humans. For instance, the HLA restriction to a single allele results in a homozygous genotype, which is very rare in humans, and these mice also often lack the ability to mount a robust polyclonal and affinity-matured immune response.51 Another important consideration when interpreting results from these model systems is that these are inbred populations with little overall genetic variability. This decreases the variability of the background against which the results are obtained and could make them easier to interpret, but it also implies that the differences in genetic makeup, age, concurrent treatments and other variables that are seen with human populations in the clinic will be missed.
There are several product-specific factors that may impact the immunogenicity of a therapeutic protein. These include potential immunogenic T-cell epitope content in the protein’s sequence, presence of host cell proteins and/or DNA, type and amount of degradants and impurities (such as aggregates and particulates) present, glycosylation patterns, immunomodulatory activity of the therapeutic protein, protein conformation, and type and extent of chemical modifications, to name a few. For example, the rhIFNβ products Avonex, Rebif and Betaferon have shown varying levels of immunogenicity in the clinic, with Betaferon in general being the most immunogenic.8,52,53 However, in addition to the product itself, the dose, dosing frequency and route of administration are different for these products, and different assays were used to assess immunogenicity. In a head-to-head comparison (same dose, dosing schedule and administration route) in an immune-tolerant mouse model, Betaferon was clearly the most immunogenic product of the three, breaking immune tolerance in all animals to which it was administered.38 This effect might be related to Betaferon’s relatively high aggregate content,54,55 but other differences (e.g., sequence, glycosylation pattern, chemical degradation, excipients) may also play a role. Altogether, these studies demonstrate that differences in product quality may affect immunogenicity, highlighting the importance of understanding the role of storage, handling, administration, process, and formulation conditions during commercial manufacturing and use of therapeutic proteins to minimize impact on product quality.
The presence of aggregates has been highlighted as an important product-related risk factor for immunogenicity.6,31,56–59 Hence, many of the studies with immune-tolerant mouse models so far have focused on probing the role of aggregates in protein immunogenicity. The collective results from immune-tolerant mouse studies have shown that aggregates and particles can lead to the breaking of tolerance and enhanced immunogenicity, but also that not all types of aggregates or particles are equally immunogenic and that some aggregates and particles appear to not be immunogenic at all. Instead, the immunogenicity of protein aggregates and particles seems highly dependent on their characteristics (including chemical modifications) which in turn depend on the way they are created.23,34,60,61
Aggregates created by metal-catalyzed oxidation have consistently shown high immunogenicity compared to other aggregates and unstressed protein, as demonstrated for rhIFNα-2b,36,62 rhIFNβ-1a,40 monoclonal IgG23 and insulin.25 Significant oxidation has been shown to induce, besides aggregation, some conformational changes.23,25,40,60,62 Moreover, all of these aggregates showed some oxidation of histidine61,63 and the aromatic amino acid residues.63,64 Hydrogen peroxide-mediated oxidation,21,40 stirring61 and UV-light exposure22,24,28 have also been used to generate protein aggregates which have enhanced immunogenicity in the murine model system tested compared to that of the protein before stress treatment. It is important to note that all of the chemical modifications described above were induced by treatments that increased the amount of such modifications to levels that and are not representative of what typically would be seen in the clinic. Several studies have shown that some chemically modified monomeric proteins have not been immunogenic,21,22,62 suggesting that certain chemical changes may not be sufficient to break immune tolerance.
Compared to unstressed protein, rhIFNα-2b aggregates36 and monoclonal IgG1 aggregates23 created by heat and/or pH stress showed enhanced immunogenicity, albeit to a lesser extent than aggregates of the corresponding protein created by (metal-catalyzed) oxidation. These aggregates were shown to be composed of protein molecules that appear to be structurally perturbed but not fully denatured (retaining some folded structure).23,36 Stirring-induced IgG2 aggregates induced a low and transient response in Xeno-het mice21 and also retained a significant degree of their folded structure.60 Aggregates of recombinant murine growth hormone (rmGH) formed by freeze-thawing or agitation contained protein molecules with native-like secondary structure and were immunogenic in wild-type mice.32 Interestingly, the antibody isotypes produced in response to these aggregates were different and somewhat lower in titers than those produced when rmGH was adsorbed to glass or aluminum hydroxide microparticles, where the protein structure was even more native-like.32 In line with these findings, native-like aggregates of recombinant factor VIII were reported to be more immunogenic than the native protein in hemophilia A mice.65 In contrast, aggregates consisting of largely denatured protein created by heating28,62,65 or guanidine treatment40, 62 were not able to break immune tolerance to the native protein, suggesting that some residual ‘native-like’ structure is needed for breaking immune tolerance to the native protein. These results suggest that aggregates composed of protein with a somewhat perturbed structure may be more immunogenic (or perhaps produce immune responses that are more cross-reactive with native protein molecules) than aggregates consisting of either native-like or denatured protein. Alternatively, these results might indicate that the modification of B cell epitopes, being predominantly conformational, is not sufficient to break tolerance. Another study showed that IgG1 aggregates composed of largely native-like protein molecules created by freeze-thaw or shaking stress were also not immunogenic in mice.23
Aggregates can differ in size by several orders of magnitude.66 Aggregates as small as dimers produced by severe UV-light stress of an IgG1 induced only a very minor ADA response in transgenic mice,22 with titers being significantly lower than those induced in response to oligomers generated by the same stress method. Similarly-sized aggregates produced by process-related conditions and low pH, as well as covalently modified monomers, were not immunogenic.22 A few other publications suggest that larger aggregates tend to be more immunogenic than smaller ones with otherwise similar properties. For instance, in a study that used murine IgG2c, a fraction containing micron-sized aggregates (containing conformationally perturbed protein) induced by UV-light exposure was more immunogenic than a fraction containing soluble oligomers.28 In the same study, subvisible aggregates created by mechanical stress also showed enhanced immunogenicity, albeit to a lower extent.
The results of studies on the impact of subvisible protein aggregates on immunogenicity are somewhat contradictory. On the one hand, rhIFNβ-1a immunogenicity tended to be enhanced in samples containing elevated levels of micron-sized aggregates, rather than being dependent on total aggregate content.39 On the other hand, Filipe et al. observed that the immunogenicity of differently stressed monoclonal IgG samples did not correlate with the concentrations of micron-sized or submicron-sized aggregates in these samples,23 implying that factors other than subvisible particle concentration played a more important role.
In conclusion, aggregate size may have an impact on protein immunogenicity. However, the results from different studies are somewhat conflicting, most likely because aggregate attributes other than size (such as chemical and physical modifications), which may depend on the treatment protocol, and the molecule being tested, are at least as important. Furthermore, current available technology used to fractionate, purify, stabilize and deliver aggregates and particles in narrow, well-defined size ranges is limited.67 Even when aggregates are fractionated, they may be unstable, and any changes in aggregate distributions that may occur upon administration are unknown. The reversibility/dissociability of aggregates under in vivo conditions could be important in this regard.
Non-proteinaceous particles may be a risk factor especially if proteins are prone to adsorb to them. A number of experiments have been carried out using suspensions of various non-proteinaceous particles (see below) to evaluate their effects on potential immune reactions. One common caveat of such model studies is that the types of the particles used (nature, amount, size-distribution and homogeneity) are often not representative of the non-proteinaceous particles present in therapeutic products. It remains to be seen to what extent these limitations impact the relevance of such studies.
RhIFNβ-1a, when adsorbed to 14-μm stainless steel particles, showed enhanced immunogenicity as compared to native, unadsorbed rhIFNβ-1a.68 Although rhIFNβ-1a also adsorbed to 0.2-μm carboxylated polystyrene particles and to 1.1-μm glass particles, these formulations did not result in increased immunogenicity compared to monomeric protein controls. However, when rmGH was adsorbed to the same 1.1-μm glass particles, it showed a response that was as high as when the protein was adsorbed to aluminum hydroxide particles, a classical vaccine adjuvant.32 Also, when rmGH immunogenicity was tested with the same stainless steel microparticles used in the rhIFNβ-1a study of van Beers et al., there was no enhancement of immunogenicity.32 The addition of aluminum hydroxide microparticles to rhIFNβ-1a did not increase the protein’s immunogenicity compared to that of the monomeric protein alone.69 High numbers of IgG2-coated microspheres that were 5 μm in size were able to induce a low and transient ADA response in the Xeno-het mouse, in contrast to protein-coated nanospheres that did not induce a response.21 The immunogenicity of a murine anti-TNFα IgG2c/κ antibody in mice was enhanced when the protein was adsorbed to the surfaces of micron-sized glass or aluminum hydroxide microparticles, or to the surface of micron-sized silicone oil droplets.27 The antibody showed minimal conformational change upon adsorption to glass or aluminum hydroxide,43 but was still immunogenic. Most recently, an emulsion of silicone oil droplets added to an ovalbumin formulation was shown to enhance immune responses to the adsorbed protein (in the absence of surfactant).29 The different results obtained in these studies may be due to different mouse strains, molecule types, particle numbers, particle chemical compositions, extents of adsorption, reversibilities of adsorption and conformational changes, amongst other factors.
A few studies in immune-tolerant mouse models have shown a positive correlation between aggregate dose and immunogenicity. Hermeling et al. observed an increase in anti-rhIFNα-2b IgG response with increasing aggregate dose, while the total rhIFNα-2b dose was kept constant.36 More recently, in a study with rhIFNβ-1b (Betaferon, a severely aggregated protein formulation54) a clear trend towards higher IgG responses with an increasing total protein dose was seen.30 In the latter study, an increase in number of doses or dosing frequency also led to higher anti-rhIFNβ IgG levels, but one dose was found to be sufficient to break immune tolerance (for a discussion on immune tolerance and how it relates to natural abundance of proteins and other factors, please see44,70–72). However, the type of aggregates and their chemical modification were not assessed in these studies. Stirring-induced aggregates of an IgG2 solution (containing > 106 micron-sized particles/ml, with most in the size range of 2–10 μm) were found to break tolerance in transgenic mice.21 However, a 100-fold dilution of these aggregates in the same total protein concentration (containing > 104 micron-sized particles/ml) did not break tolerance.73 A similar result was obtained in a study using wild-type mice, where rmGH solutions containing microparticles (1.6 ng/dose, corresponding to ~106 micron-sized particles/ml, again with most in the size range of 2–10 μm) were immunogenic, but no immune response was detected when the particle concentration was reduced 100-fold to 0.02 ng/dose (~104 particles/ml).32
It has long been known that low or high doses of antigens may induce “low zone” or “high zone” tolerance, respectively.74 These dose zones are dependent on the antigen in question. In a study using murine IgG, low doses of aggregated protein adsorbed to the surface of non-proteinaceous microparticles or silicone oil droplets provoked stronger immune responses in mice than higher doses.27 Similarly, repeated administration of high doses of IgG-coated microparticles resulted in significantly lower ADA responses as compared to low doses.21 These results highlight that it is challenging to choose an appropriate dosing schedule in murine models for evaluating immunogenicity of therapeutic proteins. One further complication is the difficulty in quantifying and comparing different dosing regimens, particularly when heterogeneous, aggregated material is used.
The route and site of administration will affect the biodistribution of a therapeutic protein, its residence time, and the likelihood of encounter with antigen-presenting cells, and therefore its immunogenicity. Especially for injections into subcutaneous (SC) tissue, filtration effects may cause larger aggregates and particles (and some excipients) to concentrate in a small region at the site of injection.75 In contrast, in intravenous (IV) administration, dilution effects may cause quick dispersion of impurities or degradants, as well as possible dissociation of aggregates and particles. It is not clear how these in vivo effects would scale from mice to humans, making it difficult to compare particle doses and effects of administration route. This question is further complicated by the fact that mice and humans exhibit major anatomical and physiological differences, particularly with respect to the structure of their skin and subcutaneous space. It has been stated that IV administration has a lower immunogenicity risk than SC administration,5,10 but in fact there are few preclinical and clinical data supporting this dogma.
The frequently-cited preclinical study on the effect of administration route on the immunogenicity of rhIFNa-2α by Braun et al.45 was performed in (non-immune-tolerant) wild-type mice rather than in immune-tolerant mice. Moreover, recently opposite results were observed in immune-tolerant mouse models, i.e., IV injected Betaferon was more immunogenic than SC, intramuscular or intraperitoneal injection30 and IV administration of rmGH particles elicited higher IgG responses than the same formulation injected SC.26
The aggregates associated with drug product administered IV may have a faster disposition than those delivered via the SC route. One study using whole mouse fluorescent imaging suggested that IgG1 aggregates could be retained at the SC injection site for over 1 month, thereby possibly increasing the chances for an immune response.75 In another study, optical imaging in live mice was used to show that fluorescently tagged IgG2 can form high amounts of aggregate in vivo in the SC space, where it was retained for an extended period of time and could be taken up by immune cells.76 However, this retention at the injection site did not cause significant ADA formation or major loss in exposure.77
An early hypothesis was that immune tolerance for recombinant human therapeutic proteins would be broken via a T-cell independent mechanism through the direct activation of B-cells recognizing repetitive epitopes78,79 presented on protein aggregates.34,36,58,62 Additionally, T-independent isotype switching has been reported in mice through concomitant Toll-like receptor and B-cell receptor activation.80
Although we cannot rule out this mechanism as a contributor to the immune response elicited by protein aggregates, it has become clear that T-cell dependent mechanisms do play a role. For example, depletion of T-helper cells led to abolition of the immune response of immune-tolerant mice against both aggregated rhIFNβ-1b41 and aggregated IgG1.23 The occurrence of isotype switching in other immune-tolerant mouse models also provides evidence of the involvement of T-helper cells in the breaking of immune tolerance.24,26,28,32,43 Moreover, the timing of T-cell recruitment and the number of germinal centers formed after administration of aggregated rhIFNβ-1b were almost identical in immune-tolerant and wild-type mice.69 Nevertheless, there are several indications that the breaking of immune tolerance occurs via mechanisms that differ from a classical immune response against foreign antigens. Marginal zone B-cells, a B-cell subset involved in T-cell independent responses, were shown to play a role in antibody production against aggregated rhIFNβ-1b in transgenic immune-tolerant mice, suggesting involvement of direct, T-cell independent B-cell activation.41 Van Beers et al. observed an apparent lack of immunological memory in transgenic immune-tolerant mice, but not in wild-type mice, against aggregated rhIFNβ-1b39 and very similar results were found for rhIFNα-2a.81 Interestingly, conjugation of rhIFNβ-1a to the cholera toxin subunit B, a strong T-cell dependent adjuvant, increased the anti-rhIFNβ antibody titers but did not trigger immunological memory.69 A similar observation was made for mAb-based biologics where highly stressed forms of mAbs were able to impact T-cell driven responses.21 However, in this case the response was short-lived as evident from the low magnitude and transient ADA response in the human IgG2 tolerized mice.
Mouse models have served as important tools to investigate the effect of product-related attributes on immunogenicity. By employing these models, some factors that increase the potential risk of immunogenicity have been identified. The presence of high numbers of aggregates, and in particular aggregates composed of proteins with extensive chemical modifications such as those induced by oxidation or photolysis, in association with limited loss of native conformation, seems to increase the relative risk of immunogenicity. However, the amount and type of aggregates in these studies were not necessarily representative of what is typically present in drug products. In some studies larger protein particles, up to about 10 micrometers in size, appear to be more immunogenic than small oligomers, but results are anecdotal and it has become clear that attributes other than size may be more important; more research is needed to rigorously establish the effect of aggregate size on immunogenicity risk. Furthermore, non-proteinaceous subvisible particles with adsorbed protein can sometimes lead to enhanced immunogenicity, although not consistently for all types of particles and all proteins. It is worth noting that since particle size is recognized as a contributor to immune responses to vaccines82–84, it is reasonable to assume that size may also be an important attribute to consider in the context of immunogenicity of therapeutic protein aggregates and particles.
Although the level of risk to human patients cannot be defined from results with animal studies, the collective findings from these studies suggest that it is sensible to minimize levels of aggregates and particulate impurities in formulations of therapeutic proteins, as low as reasonably possible. In this context, one should keep in mind that improper storage, compounding or administration practices might lead to the formation of substantially increased numbers of degradants. Therefore, instructing end-users about proper handling of therapeutic protein products, as well as creating general awareness of the risk of improper handling, remains of utmost importance.
The authors thank Drs. Gregory Flynn, Michael Hall and Antonio Iglesias for critically reviewing the manuscript.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.