Commercial antibodies are extensively used in research, in diagnostic and therapeutic applications. While diagnostic and therapeutic antibodies are subject to stringent regulations by health authorities throughout the world, there are no standards or third-party quality controls for research antibodies. Research antibodies do not always perform as advertised. Lot-to-lot variation exists in terms of specificity and performance [2, 3]. The lot-to-lot difference for polyclonal antibodies is inherent if different lots indicate different bleeds/animals. The lot-to-lot difference for monoclonal antibodies from hybridoma is also likely, since a significant number of hybridomas have additional productive, nonspecific variable chains [4], including those originated from myeloma fusion partners, as in the case of anti-PIVKA II hybridoma 3C10 mouse IgG1 clone [5]. Variations among different vials from the same lot have been documented as well [6]. In addition, well-characterized antibodies might yield discordant results due to spatial variation in the case of arrays [7], unexpected cross-reactivities such as against AF700 fluorophore [8], genetic variation of samples [9], post-mortem interval and preservation [10], laboratory settings [11], or sample processing/handling [12], or non-specific associations, for example, the non-specific binding between IgG and short unmodified DNA repeats [13], endogenous biotinylated proteins with biotin-avidin detection systems and shedding of protein A from Sepharose beads [14], or in the case of ELSA, the common practice of using E coli-derived recombinant proteins as standards [15]. In addition, fake data in the publications can not be easily ruled out [16]. Researchers have recognized this critical quality issue, and over the years, have evaluated some research antibodies and published their findings. This article enumerates these findings and summarizes the causes of antibody nonperformance.

[enlarge]

Figure 1. NLRP3 antibodies are evaluated with regular, knock-out, or over-expressed samples. From [1].

Most commercially available antibodies are affinity purified, but it is unknown if they meet specificity requirements. There are no uniform or enforceable standards for antibody validation in the market [17]. Investigators have been repeatedly frustrated by inconsistent ChIP results and questionable immunohistology data [18]. Authors had to withdraw their papers because the antibodies they used against novel markers were found to stain tissues in knock-out mice that lack the markers [19-21]. In 2015, it was found that over 80 published irisin studies had been based on commercial ELISA kits detecting unknown proteins [22]. This antibody quality problem also happens to commonly studied genes, for example, only one out of thirteen estrogen receptor beta antibodies, a rarely used antibody (ThermoFisher/Invitrogen PPZ0506 417100), was specific, despite numerous citations for the other antibodies [23]. A publication about the role of neuropsin/KLK8 in Alzheimers' disease was criticised ("a cardinal sin") for lack of antibody validation [24]. Antibody validation with knockout samples indicates that a commonly used AGO2 monoclonal antibody 11A9 bound to a different protein SMARCC1 [25], and a phospho(Ser418)-CYLD antibody reacted with an unknown protein with the same molecular weight migration band, whose phosphorylation was stimulated similarly by the same chemicals as CYLD [26]. Michael G Weller lists ten basic rules when using/purchasing antibodies in terms of antibody quality [27].

Table 1 lists these evaluation investigations to highlight the scope of antibody quality problem with a focus on monoclonal antibodies, which should always be prefered over polyclonal antibodies, especially for multiplexing of tumor biomarkers [28]. The table omits the evaluation studies using only polyclonal antibodies, such as those against cannabinoid receptor 2 (none of them were specific in IHC) [29] or others [30]. Mechetner L et al reported that nearly one out of every 11 complementarity-determining regions (CDRs) was occupied by a hitchhiking antigen in commercial histone H1 antibodies [31]. Even more strikingly, Egelhofer TA et al tested the antibodies for histone modification research and discovered more than 25% failed specificity tests by dot blot or WB, and more than 20% failed in chromatin immunoprecipitation experiments [32]. Their results are available at http://compbio.med.harvard.edu/antibodies/ and new histone antibody specificity test can be posted there. Rothbart SB et al examined the binding characteristics of over 100 commonly used histone modification antibodies from nine suppliers with a peptide microarray technique and revealed several significant problems [33]. Li et al performed a thorough analysis of several commercial antibodies against the interferon regulatory factor 5 (IRF5), a known regulator of genes important for innate and adaptive immunity, cell growth regulation and apoptosis [34]. By using appropriate positive and negative controls, that include human cell lines expressing high, low or insignificant levels of IRF5, human siRNA knockdowns and murine irf5 gene knockouts, as well as a previously validated but no longer available antibody – Cell Signaling #3257 (cs3257), the authors verified the specificity of 16 different antibodies from 7 companies (Table 1). In addition, they established the suitability of using these antibodies in applications like WB, immunoprecipitation, flow cytometry, immunohistochemistry, and immunofluorescence. Their results reveal that each antibody behaves differently. Some showed single specificity to either human or mouse samples, while others worked on both. Most antibodies were good for some techniques, but not for others. Another example of this application-dependent specificity is the evaluation of anti-GABARAP antibodies by Simons IM et al [35]. Using knockout samples, they determined that three commercial antibodies selectively detected GABARAP in immunoblotting upon SDS-PAGE, however, all failed in immunocytochemical staining [35]. Morishita R et al examined the cross-reactivities of two monoclonal antibodies anti-HA (MBL clone TANA2) and anti-PD-1 (CST D4W2J) with CF-PA2Vtech, a cell-free human protein array technology, and found them able to recognize unrelated human proteins which shared epitope sequences [36].

During the updating of Table 1 and this article, references and discussions about earlier studies are removed. They are [37-42]. Rigorous researchers also validated individual antibodies, typically those central to their conclusions, such as the LepR antibody [43]. These studies are not discussed here.

The vast majority of antibody suppliers strive to provide high-quality antibodies, although, due to a large number of enterprises involved in the production and distribution of research antibodies, unscrupulous providers cannot be ruled out. More often, different suppliers/distributors adopt different quality control policies, ranging from extensive to non-existent. Antibody users are encumbered to communicate with specific antibody providers to discover and discuss their quality control practices. Adding to the confusion is the practice of some antibody suppliers/distributors to rebrand antibody catalog numbers and to re-assign antibody clone names. For example, Laflamme C et al was able to cull more than 100 antibodies for C9ORF72 to 14 distinct commercial C9ORF72 antibodies [47]. The antibody rebranding, in combination with the low quality of some antibodies, generates one of the most perverse aspects of antibody business. That is, when an antibody fails, the antibody user tends to purchase a second antibody from a different supplier, while not knowing the second antibody is a re-branded version of the failed one. Some of the failed antibodies can in the short term even generate significant revenue for the antibody supplier/manufacturer. Thus, it is imperative for any antibody user to ascertain that the second antibody is not a re-branded version of the first one. Different QC data sheets may not indicate different products, as this point has been thoroughly discussed in an article [76]. In worst cases, some of the reagents might be too diluted by legit or bogus distributors or simply counterfeit, especially in countries like China [77].

The few nefarious players aside, the presence of a large number of non-performing research antibodies is indicative of 1) difficulty in generating specific antibodies against certain antigens, 2) less optimal antibody purification protocols, 3) contamination and storage problems, and 4) other issues.

Groups of the proteins share significant homologies, for example, the aforementioned G-protein coupled receptors, within the same antigen species and also between antigen target species and antibody host species. Similarly, developing antibodies against glycans is problematic [78], and the presumably same antibodies from different suppliers may yield different specificities [79]. The development of antibodies able to distinguish highly homologous proteins is challenging. Only after extensive quality testing can the specificity of antibodies be ascertained.

Antibodies are commonly purified through affinity chromatography, based on the reversible interactions between an antibody and its cognate ligand, or between staphylococcal protein A or G and the Fc region of immunoglobulins [80]. When the affinity purification is based on protein A or G, which is more common due to its simplicity and low cost, the CDRs of the antibodies may have already bound with so-called 'hitchhiker antigens' [80, 81]. The hitchhiker antigens are protein fragments often produced during cell death or even during cell lysis. The presence of hitchhiker antigens in the 'purified' antibodies renders inconsistent antibody specificity and performance, and has been postulated to be one of the major factors affecting antibody quality [31, 82]. Attentions on the presence of hitchhiker antigens have led to the development of downstream processing techniques for monoclonal antibodies [82-84]. More stringent purification methods were proposed, for example, Mechetner L et al compared the effects of three purification schemes for hitchhiker removal: a traditional approach using a one-step, low pH elution buffer (pH 2.5), and a gentler approach using a pH gradient elution scheme (pH 7 down to pH 2.5) were inadequate for hitchhiker removal; while a more stringent purification using a quaternary amine guard column and a high salt wash during antibody immobilization on the protein G obtained a higher score by eliminating more attached histones and DNA [31]. Luhrs KA et al used a similar measure on protein A chromatography and gained satisfactory results as well [85]. These simple measures include the addition of a quarternary amine guard column to the protein A, adjusting the ionic strength of the cell culture supernatant to 400 mM sodium chloride, and establishing a mobile phase gradient from 400 mM to 2 M during protein A chromatography [85].

Storage and handling may also generate variations for antibody experiment results. Labome has an extensive discussion on this topic. For example, antibody storage in dry ice with non-airtight vessels and vigorous shaking of antibody vials or incubation apparatus may degrade/denature/aggregate antibodies. In addition, high-intensity lighting introduced vial-to-vial variation, and BD Biosciences issued a rare recall for research antibodies conjugated with fluorophores in 2017 due to this lighting issue [86].

Various strategies have been proposed to address the antibody quality issue. Parseghian proposed a control to be added to ChIP assays [18]. Couchman JR suggested using WB with precaution for immunohistochemical antibody [2]. Bordeaux J et al proposed an approach developed by their laboratory (the Rimm Lab Algorithm) for antibody validation [17]. Dr. Alexander E. Kalyuzhny stressed the need for the antibody users to communicate with the antibody providers [3]. Drs. Andrew Bradbury and Andreas Plückthun proposed that antibodies must be defined by their sequences and generated as recombinant proteins [87]. Others suggested criteria for research antibodies [88]. Dr. Weller provided a comprehensive list of problems/issues associated with the antibody quality, and proposed several interesting approaches to address them [89]. Despite the significant antibody quality issues, a substantial number of researchers do not validate their antibodies purchased commercially or in-house, according to a 2016 survey [90]. Sortase-based specific, covalent conjugation of engineered antibodies can remove the uncertainty of chemical conjugation methods [91, 92]. NIH Common Fund initiated the NIH Protein Capture Reagents Program (PCRP), and generated 1,406 mAbs, targeting 737 unique human transcription factors and their associated proteins. These antibodies work for immunoprecipitation, immunoblotting, ChIP-seq, or immunohistochemistry (IHC) [93].

One of the important considerations is for the antibody suppliers to report any quality data using standardized protocols. Marcon E et al evaluated a mass spec-based procedure for assessing the immunoprecipitation of recombinant antibodies [94]. The protocol remains to be improved, since only 9 out of 20 antibodies yielded consistent antibody quality data from all 5 laboratories [94].

Antibody experiment results usually go through image analysis and data processing, which is not a significant contributory factor to the irreproducibility. In fact, different digital image analysis systems yielded excellent reproducibility [95].

Lot-to-lot variability is an important issue for studies applying various immunology and molecular biology techniques using antibodies or other reagents in research and clinical settings. Different results may be caused by the variations between the reagents, decomposition of the control samples or even technical malfunctions. These differences are often noticed and may have a negative influence on both scientific results and the health care system, if not detected prior to the practical applications.

Previously, lot-to-lot variations in median fluorescence intensity (MFI) were reported for anti-HLA antibodies applied in the Luminex assay [96]. Incorporation of control serum in each assay lot was suggested to be an appropriate solution for the correction of MFI parameters. In addition, significant variability was found between lots of blood glucose measurement systems [97]. The study suggested that both diabetic patients and clinicians should be informed about the possibility of these lot-to-lot differences, which may have certain clinical importance. Also, the levels of lot-to-lot differences for several reagents, including antibodies to hepatitis B surface antigen (HBsAg), α-fetoprotein (AFP) and ferritin, were previously evaluated by chemiluminescence immunoassay [98]. Observed variability ranged from 0.1 to 17.5% for AFP, 1.0 to 18.6% for ferritin and 0.6 to 16.2% for HBsAg.

Lot-to-lot variations were evaluated for immunoturbidimetric HbA1c assay (A1C3), which is applied for the measurement of hemoglobin A1c and diagnosis of diabetes mellitus [99]. The detected variability between the A1C3 reagents was approximately 0.5%. Also, different lots of fluorochrome-conjugated monoclonal antibodies show variabilities. These detected differences depend on the type of the fluorochrome [100]. In particular, the lowest variability was observed for FITC, while PE from different lots demonstrated lower stability. Moreover, several other studies reported differences between reagent lots for anticardiolipin antibodies for ELISA [101], antibodies to hepatitis C virus for enzyme immunoassay and chemiluminescence assay [102] and the reagents for detection of insulin-like growth factor 1 (IGF-1) [103]. Besides, significant bias between lots were found for HLA antibodies applied for HLA detection assay [104].

Thus, lot-to-lot variability may significantly affect the results of both basic research and clinical screening tests and requires further investigation and optimization. It is recommended for each research group or clinical unit to identify the appropriate range of lot-to-lot variabilities for each product [98]. In particular, a statistical analysis based on the Student t-test combined with cumulative regression slopes across reagent lots was previously suggested for the evaluation of these differences and may help to identify both proportional and constant drifts between products from different lots [105].

The authors acknowledge Dr. Andrew Bradbury from Las Alamos National Laboratory for his suggestion of two publications [65, 66].