Potential sources of interference on Abeta immunoassays in biological samples

Therapeutic products that depend on the use of an in vitro diagnostic biomarker test to confirm their effectiveness are increasingly being developed. Use of biomarkers is particularly meaningful in the context of selecting the patient population where the therapeutic treatment is believed to be efficacious (patient enrichment). Currently available 'research-use-only' assays for Alzheimer's disease diagnosis all suffer from non-analyte and analyte-specific interferences. The impact of these interferences on the outcome of the assays is not well understood. The confounding factors are hampering correct value determination in biological samples and are intrinsic to the assay concept, the assay design, the presence in the sample of heterophilic antibodies and auto-antibodies, or might be the result of the therapeutic approach. This review focuses on the importance of assay interferences and considers how these might be minimized with the final aim of making the assays more acceptable as in vitro diagnostic biomarker tests for theranostic use.


Introduction
Th e development of therapeutic compounds that depend on the use of an in vitro diagnostic biomarker test (IVD) to confi rm their eff ectiveness will become more common in the future. Companion diagnostics will ultimately shorten the development time for Alzheimer's disease (AD) therapeutic trials and increase their success rates. When the therapeutic product becomes available, assay information will be used to select (stratifi cation) or exclude (risk assessment) patient populations for a particular clinical study, to optimize dosing regimens, or to identify subjects who will most likely respond to treatment and will not suff er from side eff ects (responders, safety). If the outcome of a diagnostic assay determines how a patient will be treated, it is obvious that health care professionals must be able to rely on the quality of the result. Inadequate performance charac teristics of an IVD or companion diagnostic biomarker test could expose a patient to preventable treatment risks.
Several research assays for AD biomarkers in cere brospinal fl uid (CSF) evolved over the past decade from proof-of-concept to tools with promising or accepted clinical value. In this disease fi eld, no US Food and Drug Administration-approved assay is available yet on the market, due in part to some drawbacks in their analytical performance characteristics. Th e US Food and Drug Administration provides more detailed relevant policies for the safety and eff ectiveness of IVD com panion diagnostic devices as used with therapeutics [1].
Th e AD community has considered for several decades that the β-amyloid protein (Aβ) might be at the origin of AD, although amyloidopathy is not absolutely specifi c for AD [2][3][4]. A full understanding of its clinical relevance is hampered by (i) the intrinsic nature of Aβ, including its aggregation and adsorption properties, (ii) the complexity and heterogeneity of Aβ isoforms, including modifi cations or diff erent conformational forms, (iii) the presence of confounding factors, (iv) low concentrations of Aβ in biological fl uids, (v) high variability in outcomes of each assay between study centers, and (vi) the absence of a reference method or reference materials (relative quantitative assays) [5,6].

Problem statement
Immunoassays that use antibodies are easy to perform, specifi c for an epitope or conformation of an analyte, and highly vulnerable towards confounding factors or interferences [5] (in this context, an interference is an eff ect of a substance present in the sample that alters the correct value of the result). Detailed understanding of the nature, the prevalence, the complexity, the technologyor protocol-dependency, as well as the interactions between diff erent confounding factors is key to defi ne solutions and improve the robustness of the test methods. Cost-effi cient and user-friendly integration in the Abstract Therapeutic products that depend on the use of an in vitro diagnostic biomarker test to confi rm their eff ectiveness are increasingly being developed. Use of biomarkers is particularly meaningful in the context of selecting the patient population where the therapeutic treatment is believed to be effi cacious (patient enrichment). Currently available 'researchuse-only' assays for Alzheimer's disease diagnosis all suff er from non-analyte and analyte-specifi c interferences. The impact of these interferences on the outcome of the assays is not well understood. The confounding factors are hampering correct value determination in biological samples and are intrinsic to the assay concept, the assay design, the presence in the sample of heterophilic antibodies and autoantibodies, or might be the result of the therapeutic approach. This review focuses on the importance of assay interferences and considers how these might be minimized with the fi nal aim of making the assays more acceptable as in vitro diagnostic biomarker tests for theranostic use.
product design of assay modifi cations to reduce interferences, without having an impact on the clinical accuracy, is a major challenge.
Assay interferences are often underestimated, but highly relevant; they have an eff ect on sample homogeneity and stability, assay precision, or clinical interpretation. Every false result will generate extra cost for the lab and will introduce preventable concerns (through the incorrect message given) for patients, families, and caregivers.
Immunoassays measure the presence (qualitative assay), concentrations (quantitative assay), or changes in concentrations of one or several analytes in a complex mixture of proteins. Th e affi nity of the antibody for the analyte is related to its thermodynamic property (association and dissociation capacity). Antibodies and antigens (or antigen conformations) are in a state of dynamic equilibrium that is concentration dependent. Only a fraction of the total amount of analyte might be detectable by the immunoassays. Notwithstanding the well-known pre-analytical variables [5], the measurement of Aβ by classical immunoassays is complicated by artifi cial or induced confounding factors, which are illustrated in Figure 1 and discussed here. Th is review will not focus on antibody-independent techniques, as this could be the subject of future discussions, but discusses in more detail the confounding factors and some possibilities for overcoming them.

Confounding factors
Non-analyte-specifi c interferences are not necessarily directly linked to one specifi c analyte, but might be relevant also for other proteins in the sample. Several non-analyte-specifi c parameters in the product design have a direct eff ect on the equilibrium constant of the antigen-antibody reaction (for example, temperature, pH, ionic strength), while others have not (for example, antigen and antibody concentration, duration of incubation) [7]. Underdeveloped parameters can explain discrepancies between study results. Problems occur at the level of the assay, the raw materials, or the sample. For example, the pH of the assay diluent modifi es the analytical sensitivity of Aβ1-42 assays [8] and aff ects the oligomerization state of Aβ [9]. It is well-known that the pH of CSF increases rapidly after storage at room temperature for a few hours [10], although Bjerke and colleagues [10] did not observe any diff erence in Aβ1-42 concentration when CSF samples were tested at diff erent pH levels. Aβ1-42 detection in CSF is further infl uenced by temperature, especially within the AD group [11], aff ect ing its diagnostic use. Non-specifi c binding can occur also when other components (for example, secondary conjugates, detector mAbs, substrate) directly bind to the solid phase or when aggregated proteins in a β-pleated sheet conformation bind to each other. Th e latter was documented for biotinylated mAbs depending on the pH of the storage buff er [12], an eff ect that can be prevented by the use of arginine [13].
Each component in the product has to be screened for its infl uence on the intended use of the assay. If needed, test instructions have to be re-evaluated, even when the product is already available on the market.

Heterophilic antibodies
Heterophilic antibodies (HAs) are found in a number of healthy and diseased patient samples (5 to 40%) [14]. Th ese endogenous (mostly polyclonal human) antibodies have a broad spectrum of activity against non-immune immunoglobulins from several species, as well as reactivity towards poorly defi ned (self ) antigens. In contrast, human anti-animal antibodies are directed against well-defi ned antigens and show high affi nity. Th eir presence can be the consequence of the administration of an exogenous antibody, such as from treatment with therapeutic antibodies. Th e relevance, occurrence, and importance of HAs as a confounding factor for the measurement of CSF Aβ have yet to be fully described.
HAs have weak affi nity and are multispecies specifi c. Th e interfering antibodies are of the IgG, IgM or IgA type. Generally, HAs are directed to the Fc part (antiisotypical interference, such as rheumatoid factor), while anti-idiotypical interfering antibodies bind to the highly variable Fab portion of the molecule (an anti-idiotypical antibody is directed towards the antigen-binding site of another antibody; the antigen binding site of this antibody can be similar in structure to the original antigen).
Th e concentration of HAs is higher in blood samples than in CSF [15]. If plasma samples from the same subject are collected and tested over time, the interference will be visible in each sample from the aff ected subject, independent of the technology platform. Also, for low abundant proteins such as Aβ oligomers, positive signals are eliminated (plasma) or reduced (CSF) when assays are repeated in the presence of HA-blocking factors [16]. Small amounts of HAs will immediately aff ect the quantifi cation of low-abundance proteins, espe cially when the dilution factor for the sample in the assay is limited, unless HA-reducing buff ers were included in the test concept during the development phase of the product.
HAs are not directly visible in single analyte immunoassays. Th ey can be identifi ed by replacement of one antibody (capture mAb, detector mAb) with a nonanalyte-specifi c antibody [17], using antibodies from another species, by selection of assays from other vendors, or (maybe in part) by the absence of parallelism in serial dilutions of the sample. In the latter case, no experimental evidence is available to date that HAs are the major confounders.
HA interference can be minimized by assay modifi cations or sample pre-treatment. At the assay level, interference is minimized by (i) replacement of mAbs by Fab fragments, humanized antibodies, affi bodies or aptamers [18], (ii) the use of ready-to-use blocking buff ers, detergentia, heterophilic blocking tubes (for example, Scantibodies) or affi nity discriminator buff ers (for example, LowCross buff er from CANDOR Biosciences, Wangen im Allgäu, Germany), or (iii) the addition of well-controlled non-immune mouse IgG or animal serum (this will not neutralize the anti-idiotypic or anti-anti-idiotypic forms of HAs [19]).
Sample-pretreatment procedures have already been implemented in commercial assays for Aβ1-42. Increased levels of free Aβ1-42 were noted after dilution of EDTAplasma samples in detergent-containing buff er [20] or after dilution of CSF [10,21,22]. If compatible with the properties of the analyte, HAs can be removed by heating [8,10], although this is not a customer-friendly approach and might be detrimental to non-Aβ analytes to be analyzed in the sample. Bjerke and colleagues [10] reported a higher increase of CSF Aβ1-42 after heating of AD samples, resulting in reduced clinical accuracy. Extraction of HAs prior to analysis by depletion of IgG (protein A, protein G) [16], poly ethylene glycol precipitation, or pre-absorption with human gamma globulin-coated beads/Protein L-coated beads [23] has been proposed for other analytes. Th ese extraction procedures will add more cost and workload to the assay used in routine clinical testing labs. Th e improvement methods described above might be an excellent approach for one analyte, detrimental to another, and only applicable to one specifi c sample type. Th e exact impact of any assay changes on clinical outcome has to be qualifi ed.

Aβ autoantibodies
Analyte-specifi c interferences will directly compete with the quantifi cation of Aβ in immunoassays. Th e prevalence of analyte-specifi c interferences in modern (blocked) two-site immunoassays is very low (<0.05%). Th e relevance, occurrence, and importance of auto-antibodies as a confounding factor for CSF Aβ measurements are uncertain at present due to a lack of wellcharacterized and validated detection systems and are a subject for future studies.
Th e clinical relevance of auto-antibodies against Aβ is still unclear, related in part to the urgent need for harmonization of the detection methodologies. Th e introduction of Aβ immunotherapies will further accelerate the development of harmonized immune bio markers (Table 1).
Auto-antibodies in AD can infl uence the IgG-based clearance system for amyloidogenic proteins [24], modulate plaque removal [25], be indicative for dysfunctional  immune signaling [26], refl ect blood-brain barrier breakdown [27], or reduce neuritic dystrophy and astrogliosis in mouse models [28]. When low concentrations of an analyte under investigation are present in a sample, such as in the case of Aβ oligomers as a potential neurotoxin, auto-antibodies may pose additional particular challenges linked to the analytical sensitivity of the technology platforms used for their detection [16].
Naturally occurring antibodies against Aβ are found in CSF and plasma of healthy and diseased subjects. Patients with AD have lower, higher, or identical levels of serum anti-Aβ antibodies to healthy age-matched individuals [24]. Th e serum anti-Aβ antibody titer was higher in patients with mild cognitive impairment who progressed to AD than in stable cases [29]. Th e level of auto-antibodies to Aβ did not correlate with the 'likelihood of developing AD [30], although titers were negatively correlated with the cognitive status [31]. Henkel and colleagues [24] showed that the concentration of the large Aβ-binding particles (LAPs) was highly variable among individuals and that the levels were not disease-specifi c. Th e abovementioned studies use diff erent technologies and are not always qualifi ed extensively for the intended use.
Th e immunoreactivity was 30 to 230 times lower in CSF than in plasma [32]. Anti-Aβ IgGs were restricted to the IgG1 and IgG3 subclass, which are the most potent IgG subclasses in activating the complement classical pathway through C1q binding [33]. Several types of antibodies recognize the higher order Aβ assemblies, includ ing oligomers and post-translationally modifi ed peptides, rather than the monomeric forms [32]. Th e signal in unvaccinated nonTg mice was specifi c for the fulllength Aβ peptide, while for anti-Aβ vaccines, the predominant epitope is in the amino-terminal domain [34].
In the study of Henkel and colleagues [24], confocal microscopy for single molecular imaging was used to identify and classify distinct types of LAPs. Double labeling with anti-Aβ and anti-IgG antibodies characterized LAP-3 and LAP-4 as immune complexes of Aβ and IgG. Th e LAPs are most probably directed against either a non-pathogenic form of Aβ or aggregated Aβ1-42.
Th e antibody-antigen complex can be dissociated by incubation at low pH [34,35]. Th e increase in Aβ autoantibody levels is higher in AD than in controls, again possibly aff ecting clinical accuracy. Diff erences between studies may relate to a diversity of methods used for anti-Aβ assessment, the validation status of the assays, and detailed investigation of the relevance of the generated output. One approach to solve this issue could be the development of a multi-analyte assay format, either (i) by inclusion of an extra region with a non-analyte specifi c mAb used as capture antibody (direct bridging between mAbs) [17], or (ii) by the integration of an internal control with Aβ coupled to a specifi c region to verify the presence of anti-Aβ antibodies. Th is approach will help to understand the contribution of Aβ auto-antibodies to the output signal of the assay.

Analyte-specifi c interferences induced by therapeutics
Many pharmaceutical and biotechnology companies perform studies with Aβ-lowering drugs: vaccination with peptides (active) or immunization with mAbs (passive immunization) ( Table 1). Vaccination improves the brain pathology and protects against cognitive decline, while passive immunization with Aβ antibodies reduces plaque burden [2]. Th erapeutic compounds, however, can directly interfere with assay design and assay modifi cations are required dependent on the therapeutic approach. Short amino-terminal-specifi c Aβ-peptides used for immunization will compete for binding with the mAbs if amino-terminal specifi c mAbs are used, while the therapeutic mAbs can mask the binding site of the analyte, resulting in an underestimation of the levels in a sample.
Analyte-specifi c interferences induced by a specifi c therapeutic (with a defi ned molecular nature) can be identifi ed easily and reproducibly by spiking of the drug in relevant concentrations in the sample of interest, followed by its quantifi cation using diff erent technologies and/or antibody pair combinations. In active immunization trials with a more heterogeneous antibody spectrum, it might be more relevant to include antibodyindependent techniques (for example, mass spectrometry). Although the latter might not be relevant for verifi cation of effi cacy of drug treatment since these techniques quantify the total amount of Aβ in a sample. Th e monomeric, non-bound Aβ could be the most relevant Aβ isoform to verify for a disease-modifying eff ect. As such, it will be obligatory to develop assays in biological fl uids that make a distinction between free analyte, and the analyte bound to confounding factors (for example, endogenous antibodies, therapeutic compounds), or conformational forms. Comparison with the total amount of Aβ, independent of the isoform, could be an extra advantage for clinical interpretation of the data.

Conclusion
Measurements of low concentrations of Aβ1-42 in biological samples are impeded by the presence of nonspecifi c interactions. Th ese documented confounders need to be addressed during the development phase of the product to improve robustness, respecting the balance between the required assay sensitivity and precision. Th e impact of assay modifi cations on the accuracy of clinical decisions still has to be determined.