Antibody performance in western blot applications is context-dependent

© 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 1

1 Introduction

One of the key challenges in biological research is todevelop genome-wide resources of validated affinityreagents to explore the expression patterns and functionsof the different gene products. The affinity reagents needto be both specific and sensitive to allow for detection ofproteins present in concentrations varying more than 106

in cells and tissues and more than 1010 in human plasmaor serum [1]. There are different types of affinity reagents,

Research Article

Antibody performance in western blot applications is context-dependent

Cajsa Älgenäs1, Charlotta Agaton2, Linn Fagerberg3, Anna Asplund4, Lisa Björling3, Erik Björling1, Caroline Kampf4, Emma Lundberg3, Peter Nilsson3, Anja Persson1, Kenneth Wester4, Fredrik Pontén4, Henrik Wernérus2, Mathias Uhlén3, Jenny Ottosson Takanen1 and Sophia Hober5,*

1 Division of Proteomics, School of Biotechnology, Albanova University Center, KTH – Royal Institute of Technology, Stockholm,Sweden

2 Atlas Antibodies AB, Albanova University Center, Stockholm, Sweden3 Science for Life Laboratory, KTH – Royal Institute of Technology, Solna, Sweden4 The Rudbeck Laboratory, Department of Genetics and Pathology, Uppsala University, Uppsala, Sweden5 Division of Protein Technology, School of Biotechnology, Albanova University Center, KTH – Royal Institute of Technology,Stockholm, Sweden

An important concern for the use of antibodies in various applications, such as western blot (WB)or immunohistochemistry (IHC), is specificity. This calls for systematic validations using well-designed conditions. Here, we have analyzed 13 000 antibodies using western blot with lysatesfrom human cell lines, tissues, and plasma. Standardized stratification showed that 45% of theantibodies yielded supportive staining, and the rest either no staining (12%) or protein bands ofwrong size (43%). A comparative study of WB and IHC showed that the performance of antibod-ies is application-specific, although a correlation between no WB staining and weak IHC stainingcould be seen. To investigate the influence of protein abundance on the apparent specificity of theantibody, new WB analyses were performed for 1369 genes that gave unsupportive WBs in the ini-tial screening using cell lysates with overexpressed full-length proteins. Then, more than 82% ofthe antibodies yielded a specific band corresponding to the full-length protein. Hence, the vastmajority of the antibodies (90%) used in this study specifically recognize the target protein whenpresent at sufficiently high levels. This demonstrates the context- and application-dependence ofantibody validation and emphasizes that caution is needed when annotating binding reagents asspecific or cross-reactive. WB is one of the most commonly used methods for validation of anti-bodies. Our data implicate that solely using one platform for antibody validation might give mis-leading information and therefore at least one additional method should be used to verify theachieved data.

Keywords: Immunohistochemistry · Monoclonal antibodies · Polyclonal antibodies · Validation · Western blot

Correspondence: Prof. Sophia Hober, Division of Protein Technology,School of Biotechnology, Albanova University Center, KTH – Royal Institute of Technology, SE-106 91 Stockholm, SwedenE-mail: [email protected]

Abbreviations: CAB, commercial antibody; CMA, cell-microarrays; ER-1,estrogen receptor 1; HPA, Human Protein Atlas; HSA, human serum albu-min; IHC, immunohistochemistry; IS, IHC staining intensities; N/A, notapplicable; TMA, tissue-microarrays; WB, western blot

Biotechnol. J. 2014, 9 DOI 10.1002/biot.201300341

www.biotechnology-journal.com

BiotechnologyJournal

Received 03 AUG 2013Revised 12 DEC 2013Accepted 03 JAN 2014Accepted article online 08 JAN 2014

2 © 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

including polyclonal and monoclonal antibodies and vari-ous recombinant binding reagents often obtained by invitro selection methods, such as scFvs, Fabs [2, 3], affi-body molecules [4], aptamers [5], intrabodies [6], andankyrins [7]. At present, the most common bindingreagents for analysis of human proteins are the polyclon-al antibodies, as exemplified by the fact that approxi-mately 70% of the more than 730 000 antibody reagentstoward human targets in the community-based databaseresource Antibodypedia (www.antibodypedia.org) arepolyclonal antibodies. This antibody portal lists antibod-ies available to the research community through variouscommercial and academic providers and antibodies arescored depending on the presence or absence of antibodyvalidations in specific assays [8]. However, although stan-dardized exchange formats for affinity reagents are underdiscussion, there are still no agreed standardizations forthe criteria to be used for validating antibodies in differ-ent applications.

Two fundamental criteria must be fulfilled for a func-tional antibody, (i) ability to bind the target molecule and(ii) low cross-reactivity. Specificity of an antibody isdependent on both the physical characteristics of thebinding site, as well as the environment in the experi-mental setup in which the antibody is to be used. Sincethe complexity of a full proteome from an organism isenormous, there will always be possibilities to interactwith molecules similar to the intended target. Theobtained signal will be dependent on both the strength ofthe binding and importantly also on the availability/con-centration of competing molecules in the analyzed sam-ple. Hence, the origin, type, and treatment of the biosam-ple will influence the experimental outcome to a greatextent since different epitopes might be exposed and dif-ferent concentrations of available proteins will be presentin the sample preparations.

A commonly used method for validation of antibodiesis WB analysis [9] since the method yields a size meas-urement of the target protein and often reveals if the anti-body is cross-reactive or binds to other, non-relevant, pro-teins. This makes WB analysis the most common valida-tion method for antibodies, although there are many lim-itations associated with the technique. First, the use ofsodium dodecyl sulfate polyacrylamide gel electrophore-sis (SDS–PAGE) in the separation step often leads to irre-versible denaturation of all proteins in the lysate andthereby the reactive sites, the epitopes, on the fractionat-ed molecules might be disrupted. This feature influencesboth the detection of the target protein and possible inter-actions with other proteins in the lysate. Secondly, themigration of the target protein on the SDS–PAGE systemis not only affected by protein size, but also by several oth-er parameters such as degree of denaturation, the chargeof the proteins, and the presence or absence of hydropho-bic regions. Hence, simple comparison with a molecularweight ladder might give erroneous conclusions. Thirdly,

proteins might have post-translational modifications thatinfluence the apparent molecular weight in theSDS–PAGE, such as glycosylation or cleavage of the pep-tide backbone, and thus there is not a strict relationshipbetween the molecular weight predicted from thegenome sequence and the true gene product with post-translational modifications.

Despite all these limitations, most antibodies availableto the research community have been validated using aWB assay. However, so far no systematic analysis of anti-body validation based on WB with a standardized assayhas been described. Here, we describe a large-scale effortto validate more than 13 000 polyclonal and monoclonalantibodies using a standardized set of validation criteriaas part of the Human Protein Atlas (HPA) project [10]. Thecross-reactivity pattern in WB was also compared withdata achieved from IHC experiments. Furthermore, alarge group of the antibodies that showed unspecificstaining in the standardized WB were used to detect thetarget protein in protein lysates with recombinant full-length target proteins. The cross-reactivity pattern wasexplored in these experiments and the implication of thefindings for the validation process of antibodies producedwithin large-scale proteomics initiatives are discussed.

2 Materials and methods

2.1 Antibodies

All antibodies, produced within the HPA project, weregenerated against human protein fragments, affinity puri-fied, and quality controlled as previously described (HPA)[11]. The externally produced monoclonal and polyclonalantibodies were provided by 48 commercial suppliers and13 academic scientists (CAB).

2.2 Cell culturing

Glioma cell line U251MGsp [12] and urinary bladder can-cer cell line RT4 (German Collection of Microorganismsand Cell Cultures, DSMZ [http://www.dsmz.de/]) werecultured in a humidified atmosphere at 5% CO2 and 37°C.Confluent cells were split (1:2) 16–24 h before harvest andgrown and harvested in proliferation phase. By inocula-tion on indicator cells and Hoechst staining (SVA,Mycoplasma laboratory, Uppsala, Sweden), all cell lineswere tested for mycoplasma infection [13].

2.3 Preparation of cell and tissue lysates

Protein lysates were prepared from cell lines and humantissues from liver and tonsil. Prior to protein extraction,frozen tissue material was first crushed mechanically in amortar containing liquid nitrogen and subsequently pul-verized using a Micro-Dismembranator (Sartorius, Goet-

www.biotechnology-journal.com www.biotecvisions.com

BiotechnologyJournal Biotechnol. J. 2014, 9

© 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 3

tingen, Germany) for 1–2 × 60 s at maximum speed. Pul-verized tissue was stored at −70°C until protein extrac-tion.

Protein extraction from frozen pellets of cultured cellsor pulverized tissue was done using ProteoExtract Com-plete Mammalian Proteome Extraction Kit® (Calbiochem,Darmstadt, Germany) according to manufacturer’sinstructions. The kit includes a hypotonic buffer and adetergent to encourage lysis of cells and to solubilize pro-teins. Enzyme (benzonase) is included in the proteinextraction step for degradation of nucleic acids. Finally,protein concentrations were determined using Non-Inter-fering Protein Assay Kit® (Calbiochem) according to man-ufacturer’s instructions. Pooled human plasma from maleand female donors were depleted of IgG and humanserum albumin (HSA) using IgG and HSA specific Affi-body ligands according to manufacturers recommenda-tions (Affibody AB, Bromma, Sweden).

2.4 Western blot

The entire WB protocol including dilution of both primary-and secondary antibodies and the final detection stepwas performed in a routine manner and no specific exper-imental optimization was made for individual antibodies.Briefly, 15 µg of total protein lysates, from selected celllines and tissues (RT-4, U-251MG, HSA- and IgG-deplet-ed plasma, liver, and tonsil) were loaded on precast10–20% criterion SDS–PAGE gradient gels (Bio-Rad Lab-oratories) and ran under reducing conditions, followed bytransfer to PVDF membranes (Bio-Rad Laboratories)according to the manufacturer’s recommendations. TheCriterion SDS–PAGE gradient gels make it possible toanalyze protein in the sizes ranging from 11 to 230 kDa.For the analysis of the human estrogen receptor, ER-1,100 ng of recombinant protein (Sigma–Aldrich St. Louis,MO, US) was applied on the SDS–PAGE gradient gel. Twomicrograms of over-expression HEK293T cell lysate (OriGene Technologies, Rockville, MD) was used as a pos-itive control for 723 of the antibodies. Membranes wereactivated in methanol before blocking (5% dry milk, 0.5%Tween 20, 1× TBS; 0.1 M tris–HCl, 0.5 M NaCl) for 1 h atroom temperature during constant shaking. The mem-branes were then incubated with primary antibody, HPA(polyclonal antibody from the HPA-project) diluted 1:250or CAB (commercial antibody) diluted 1:500, for 1 h, fol-lowed by washing (0.1  M tris–HCl, 0.5  M NaCl, 0.1%Tween) and incubation with the secondary peroxidase-conjugated antibody (swine anti-rabbit 1:3000, or goatanti-mouse 1:7000 depending on the primary antibodyused, Dakocytomation). A CCD-camera (Bio-Rad Labora-tories) was used for detection of signal from the substrate(SuperSignal West Dura Extended Duration Substrate[Pierce]).

www.biotecvisions.comwww.biotechnology-journal.com

BiotechnologyJournal Biotechnol. J. 2014, 9

2.5 Scoring of western blot analysis

All images were manually annotated following a commonannotation scheme based on scores ranging from 1 to 8.WBs were first scored lane-by-lane and subsequently given an overall score taking all five lanes into considera-tion. Overall scores, were categorized into three maingroups as either supportive, uncertain, or not supportiveas judged from the presence/absence of protein bands ofthe expected size according to ENSEMBL predictionsand, when available, SwissProt/UniProt data [14–16].Annotation categories for western blot analysis on theHPA were as follows (Fig. 1). In the first category, sup-portive, there are two different validation scores that dis-tinguish results with one single band at the expectedmolecular weight (± 20%) (WB-score 1) and those wherealso other few additional weaker bands were found (WB-score 2). In the group with uncertain validations, differentaspects have been taken into consideration. Here, anti-bodies that gave a single band larger than predicted solely from the genomic sequence, but with other bioin-formatic data supporting the result, were included. Thesebioinformatic data predicted the proteins to have a signalsequence or transmembrane region. Hence, they areexpected to enter the endoplasmatic reticulum with ahigh likelihood of being glycosylated, resulting in a high-er molecular weight (WB-score 3) [17]. The second sub-group with uncertain validation includes all antibodiesthat do not give any detectable bands (WB-score 4). Witha WB that does not show any band, it is impossible toknow if the reason for this is lack of target protein in theprotein lysate or antibodies that do not recognize theexpected protein. The third subgroup with uncertain val-idation constitutes the antibodies that give single bandswith a molecular weight not corresponding to what hasbeen predicted by Ensembl/UniProt and also lack bioin-formatic data to support such a result (WB-score 5). Thereason to sort these in the group of uncertain antibodiesis that the bioinformatical prediction of transcripts still israther unsure and hence, the detected protein size could,due to protein processing or the presence of an unknownisoform, be correct. In the last validation group, not sup-portive, all antibodies that give multiple bands in the WBanalysis are placed. The common feature of antibodiesgrouped in the last category, not supportive, is that whenusing them in WB, the most prominent band/bands are ofother molecular weights than expected (WB-score 6 an 7).The last group (WB-score 8) constitutes antibodies forwhich this standardized analysis is not applicable (N/A),the target is too small/large to be analyzed with the pres-ent setup.

2.6 Immunohistochemistry and image analysis

Tissue- (TMA) and cell-microarrays (CMA) were immuno -stained using the same protocol. In brief, over night, slides

4 © 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

were baked in 50oC and deparaffinized in xylene followedby hydration in graded alcohols and blocking for endoge-nous peroxidase using 0.3% hydrogen peroxide in 95%ethanol. In a decloaking chamber® (Biocare Medical, Wal-nut Creek, CA) the antibody retrieval was done by immer-sion and boiling in citrate buffer for 4 min at 125oC and pH6(Lab Vision, Freemont, CA, USA). Slides were then left tocool down to 90°C. Using an Autostainer 480 instrument®

(Lab Vision), IHC was performed. Primary antibodies wereincubated on the slide followed by incubation with a dex-tran polymer visualization system (UltraVision LP HRPpolymer® Lab Vision). Each incubation was carried out for30 min in room temperature and followed by washing inwash buffer® (Lab Vision). Slides were subsequently, devel-oped using diaminobenzidine (Lab Vision) as chromogenand thereafter counterstained in Mayers hematoxylin (His-tolab, Gothenburg, Sweden) and coversliped using Pertex®

(Histolab) as mounting medium [18, 19].Annotation of digital TMA images were performed by

experienced pathologists and TMAx software application

(Beecher Instruments, Sun Prairie, WI, USA) was used foranalysis of the digital CMA images [20].

2.7 Antibody validation

Available bioinformatics data such as the presence orabsence of trans-membrane regions and predicted size ofthe protein product based on the genome sequence wereconsidered when evaluating the antibodies. Apparently,also existing experimental data from scientific literaturewere included when scoring the antibodies. Based on asummary of this comprehensive set of data, each anti-body was given a reliability score and all antibodies witha supportive or uncertain validation score in the IHCapplication are published on the publicly available HPAportal (www.proteinatlas.org). The high-resolution IHCimages as well as the supportive WBs can be found at theHPA portal that, presently (August 2013) contains morethan 21 984 antibodies corresponding to more than 80% ofall human genes [21].

www.biotechnology-journal.com www.biotecvisions.com

BiotechnologyJournal Biotechnol. J. 2014, 9

Figure 1. Representative western blots of the scores used in the antibodyapproval process. The black box shows the expected molecular weight ofthe target protein as calculated from the amino acid sequence. WB-score1: supportive (single band of predicted size (±20%), WB-score 2: support-ive (band of predicted size (±20%) with additional weaker bands present),WB-score 3: uncertain (single band larger than predicted size but partlysupported by other available data), WB-score 4: uncertain (no proteinsdetected), WB-score 5: uncertain (single band differing more than ±20%from predicted size), WB-score 6: not supportive (weak band of predictedsize but additional higher intensity bands also present), and WB-score 7:not supportive (no band of correct size).

© 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 5

2.8 Immunofluorescent confocal microscopy

The immunofluorescent staining of cells and confocalimage acquisition was performed as previously described[22]. Briefly, cells were fixed with paraformaldehyde andpermeabilized using Triton X-100 before staining withprimary antibodies and fluorescently labeled secondaryantibodies. Besides the HPA antibody of interest, the cellswere also stained with organelle specific probes outliningthe cytoskeleton, endoplasmatic reticulum, and nuclei.Images were acquired using a Zeiss LSM 510 Meta, con-focal laser scanning microscope with a 63×/1.4 NA oilimmersion objective.

2.9 Validation of the immunohistochemistry pattern

All antibodies were also evaluated by IHC applied onTMAs representing 17 normal tissues, 11 tumor types,and 2 cell lines, in duplicate or triplicate, resulting in totalof 70 individual spots. For each antibody, the observedstaining pattern was compared, when available, to theexpected protein distribution as described in existingexperimental data from scientific literature and/or bioin-formatic data based on various prediction algorithms.Staining patterns from antibodies expected to recognizethe same protein were compared and scored, similar,partly similar, or not similar. The general immunostainingquality was also considered with special reference to theintra-cellular staining pattern. Based on a comprehensiveevaluation of the obtained data, each antibody was givena final validation score: very low, low, medium, or high.The validation score indicates how well the quality assur-ance data supports the specificity of the antibody towardthe expected human target protein. The antibodies wereclassified into four main categories: (i) high (two inde-pendent antibodies targeting one protein yielding similarstaining patterns. Staining pattern consistent with exper-imental and/or bioinformatic data), (ii) medium (stainingpattern consistent with experimental and/or bioinformat-ic data), (iii) low (staining pattern partly consistent withexperimental and/or bioinformatic data), and (iv) very low(no experimental and/or bioinformatic data available, orstaining pattern not consistent with experimental and/orbioinformatic data) (www.proteinatlas.org).

3 Results

A large-scale effort to systematically compare the resultsfrom WB analysis and IHC using 13 000 antibodies is pre-sented. Most of the analyzed antibodies were affinity-purified rabbit polyclonals generated through the HPAeffort [11, 23–25] to map the human proteome. In addition,3100 antibodies were obtained from commercial pro -viders and these were either polyclonal or monoclonalantibodies generated by hybridoma cell lines [26]. A com-

parative study has been undertaken in which the anti-bodies have been analyzed on a standardized WB plat-form and the results have been compared with the per-formance in IHC [25].

To make WB validation possible, a high throughputassay was developed including protein extracts from twohuman cell lines (RT-4 and U-251MG), human plasma(depleted of HSA and IgG), and protein lysates fromhuman liver and tonsil. The origin of the protein extractshave been chosen to cover a large part of the human pro-teome, but clearly full coverage is not feasible and someproteins are not present in any of the selected cells or tis-sues. The seven validation scores can be divided intothree main categories: supportive, uncertain, and notsupportive (Fig. 1). For a more detailed explanation, seeSection 2. All WB analyses were scored based on the pre-dicted molecular weight obtained from the genomesequence [15].

3.1 Annotation statistics of the analyzed antibodies

To further investigate the behavior of antibodies in differ-ent applications, the performance in the standardized IHCand WB platforms was analyzed using 12 929 antibodies.Nine thousand eight hundred sixteen of these were HPA-generated antibodies while 3113 were acquired fromexternal suppliers. The results of all WB analysis (Fig. 2A)show that 45% of all antibodies analyzed gave supportiveWB results (WB-score 1–2), while 3% were uncertain withWB-score 3, 12% with WB-score 4, and 9% were uncertainwith WB-score 5, altogether 24% of the antibodies wereshown to be uncertain in this setup. Thus, 31% were concluded to be non-supportive (WB-score 6–7) due tobinding to proteins of incorrect sizes in the WB-analysis(Fig. 2A).

Interestingly, information generated by the externalantibody providers, suggests that 84% of the antibodiesshould yield correct staining in western blot. In our stan-dardized WB analysis, only 48% of these antibodies gavea supportive result. Eighteen percent of the antibodieswas annotated as uncertain, and among these 12% gaveno band in the analysis. Of the externally provided anti-bodies analyzed, 33% showed a non-supportive result inour WB assay. This discrepancy is probably explained bythe fact that a standardized WB analysis have been usedhere and that the correct protein targets were not overex-pressed or might not even been present for many of theantibodies.

The annotation data for all 12 929 antibodies showsthat almost half of the antibodies are able to bind to thetarget protein under these circumstances (Fig. 2A). How-ever, about one-third shows a non-supportive pattern inthe WB analysis.

www.biotecvisions.comwww.biotechnology-journal.com

BiotechnologyJournal Biotechnol. J. 2014, 9

6 © 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

3.2 Antibody validation using overexpression lysates

To further understand the behavior of the antibodies, asubset of the HPA-generated antibodies, scored uncer-tain or not supportive when earlier analyzed in WB, wasselected for further validation. For this analysis, proteinlysates from a human kidney cell line (HEK293T), overex-pressing the respective target protein was used.HEK293T lysate, lacking the overexpressed target pro-tein, was used as a negative control. The subset, com-prising 1544 antibodies, was scored with the same cate-gorization as earlier described (Fig. 1). The analysisrevealed that 82% of the antibodies that in the first analy-sis were unable to selectively recognize a protein ofexpected molecular weight now were able to identify theoverexpressed target protein among all other proteins inthe lysate (Fig. 2B). The subset of antibodies that wereshown to be not supportive in the WB assay decreasedfrom 49 to 5%, when using a lysate with overexpressedtarget protein. The number of antibodies scored as WB-score 4, not giving any detectable bands, decreased from26 to 6%. Antibodies that were assigned to WB-score 3decreased from 6 to 1% and WB-score 5 from 19 to 2%.Hence, the restricted number of lysates used in the rou-tine setup is shown to limit the possibility to correctly val-idate the antibodies and consequently a lack of target pro-

tein may explain the high amount of antibodies that givesan uncertain or not supportive result.

3.3 Antibodies targeting the human estrogenreceptor 1

In order to further elucidate the selectivity and specificityof antibodies in different platforms, antibodies directed tovarious parts of a common protein were produced andcompared in two different experimental platforms, name-ly WB and IHC. Here, this is exemplified by three anti-bodies, one of external origin (CAB) and two generatedwithin the HPA project, toward the human estrogenreceptor. These antibodies were used in the standardassays IHC and WB (Fig. 3). The antibodies weredesigned to recognize different parts of the full-lengthform of the estrogen receptor 1 (ER-1), a ligand-activatedDNA-binding transcription factor. The resulting stainingin IHC on human tissues showed presence of the proteintarget in endometrium, cervix, and other endocrinefemale tissues (Fig. 3A). Distinct nuclear staining pat-terns were seen for all antibodies in the endometrial tis-sue (Fig. 3B). Thus, all the three independent antibodiestoward the same protein target showed congruent IHCstaining patterns in agreement with the expected targetdistribution [27]. However, the analysis on the standard-

www.biotechnology-journal.com www.biotecvisions.com

BiotechnologyJournal Biotechnol. J. 2014, 9

Figure 2. Distribution of WB-scores of the antibodies. Black shows percentage of antibodies with supportive WB (WB-score 1–2), gray represents antibod-ies with an uncertain WB (dark gray shows antibodies with WB-score 3; gray, antibodies with WB-score 4 and light gray, antibodies with WB-score 5) andwhite represents antibodies with non-supportive WB results (WB-score 6–7). (A) Results from cell/tissue-specific annotation of WB analysis of 12 929 anti-bodies of which 9816 were produced within the Human Protein Atlas project and 3113 were provided by external suppliers. (B) Data from the WB analysisof 1544 antibodies that in the first analysis were scored as uncertain or not supportive were reanalyzed with cell lysate from cells overexpressing the spe -cific target protein included in the analysis.

© 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 7

ized WB show different patterns for all three antibodies,with multiple bands ranging from 20 kDa to more than200 kDa (Fig. 3C). To investigate if the lack of a specificband could be due to the absence of protein target in theassay, the WB analysis was repeated with the addition ofrecombinant ER-1 receptor. As can be seen in Fig. 3D, theantibodies now clearly recognized the target protein at

66 kDa, confirming the specificity of the antibodies forER-1 and the IHC-results. The results show that back-ground reactivity toward other human proteins becomesvisible if no target protein is present. Thus, a non-sup-portive WB (WB-score 6 or 7) is obtained for the antibod-ies raised against the ER-1 receptor using the standard-ized protocol (Fig. 3C) despite the fact that the antibody

www.biotecvisions.comwww.biotechnology-journal.com

BiotechnologyJournal Biotechnol. J. 2014, 9

Figure 3. Validation of three antibodies targeting the human estrogen receptor 1. (A) The protein expression profiles in 48 different normal tissues areshown. Intensity and abundance of immunoreactivity are given as a color code (red = strong, orange = moderate, yellow = weak, and white = no staining).Each colored square represents one tissue type, samples from three different patients. The combined data from staining by three different antibodies (ninesamples) are used for the annotated expression, blue squares (dark blue = high, blue = medium, light blue = low, and white = no expression). (B) Highmagnification of a TMA section showing immunohistochemical staining in endometrium, comparison between all three antibodies. (C) WB-analysis withthe standard protein lysates used in the HPA-project, molecular weight marker in kDa (lane 1), urinary bladder cell line RT-4 (lane 2), glioblastoma cell lineU-251MG (lane 3), human plasma depleted of HSA and IgG (lane 4), liver (lane 5), and tonsil (lane 6). Neither CAB000037 nor HPA000450 are able to rec-ognize a protein with the predicted molecular weight, but when using HPA000449 a band with the expected molecular weight is detected. (D) WB analysiswith the same antibodies but in this setup using recombinantly produced ER-1 as positive control material. All antibodies recognize a protein band withthe correct molecular weight (lane 2). The expected molecular weight of the full length ER-1 is 66 kDa.

8 © 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

strong nucleic staining is shown in the majority of thecells of both cell lines and the automated software hasscored the intensity as strong (IS-3). The WB analysisusing the same antibody reveals a strong band around95 kDa in both cell lines (Fig. 5A). According to bioinfor-matics data, only one transcript is known and the molec-ular weight of the corresponding protein is expected to be76 kDa. However, this protein is also denoted AKAP 95due to its apparent migration in SDS–PAGE around95 kDa supporting the molecular weight seen in the WBanalysis [30]. Another example is the antibody raisedtoward the CD44 antigen precursor (Fig. 5B), a proteinthat is known to be involved in cell–cell and cell–matrixinteractions [31]. The WB-score is differing between thetwo cell lines, uncertain (WB-score 4) for RT-4 and sup-portive (WB-score 2) for U-251MG resulting in a support-ive WB-score (WB-score 2) for this antibody (Fig. 5B). Con-cordant results could be observed in the IHC-images,where RT-4 showed no IHC positivity (IS-0) whereas in U-251MG a clear staining could be seen in the membraneof the cell line and the image analysis software graded thestaining as strong (IS-3). The last example shown (Fig. 5Cand 5D) is two antibodies targeting the same protein. Theantibodies used are raised against the transcription fac-tor C/EBP epsilon. This protein is predicted to be neces-sary for the maturation of committed granulocyte progen-itor cells and is therefore, expected to be found in bonemarrow [32, 33]. Indeed, these antibodies showed selec-tive and reliable IHC results with staining in bone marrow,promyelocytic leukemia and cell lines from chronic myel-ogenous leukemia. However, the WB analysis renderedmultiple bands of different sizes (left panel) despite thefact that the IHC-analysis only shows weak staining in thesame tissue/cell samples (IS-1). When using lysates froma human kidney cell line overexpressing the protein tar-get for the WB analysis (right panel), the results clearlyshow that the antibody is able to recognize the correct tar-get protein (Fig. 5C and 5D). Despite the indefinite resultswhen using the antibodies for WB-analysis of proteinlysate from cells or tissue, the combined results from thetwo antibodies targeting the same protein could be usedfor informed decisions whether the reactivity shown inthe IHC analysis is a correct pattern or if it emerges fromcross-reactivity.

4 Discussion

Here, we present the results from validation of 13 000 anti-bodies by WB and IHC and the results illustrate chal-lenges associated with high throughput approaches forgeneration and validation of such binding molecules.Since antibodies are important tools in biological andmedical research, the fidelity of antibody-based analysisand thereby the validation of these molecules is of utmostimportance.

www.biotechnology-journal.com www.biotecvisions.com

BiotechnologyJournal Biotechnol. J. 2014, 9

is apparently specific for the target protein (Fig. 3B and3D).

3.4 Relation between western blot scores andimmunohistochemistry staining intensities

The results from 3364 WB- and IHC analyses on samplesfrom human cell lines RT-4 and U-251MG were analyzedin order to explore the relationship between WB-score andthe signal intensity/protein quantities as determinedfrom IHC stained cell lines. Staining intensities werescored by an automated image analysis software [20] andcompared with the WB results on protein lysates from thesame cell lines (Fig. 3C). The IHC staining intensitieswere scored from IS-0 to IS-3, where IS-0 implies no stain-ing and IS-3 very strong staining. Data from the 3364 pairwise analyses clearly showed that the number of uncer-tain WB-scores was over-represented when the IHC reac-tivities were low (IS-1) or very low (IS-0) in the correspon-ding cell line. For the group showing very low IHC-reac-tivity (IS-0), around 60% of the corresponding WB analy-ses resulted in no band in the WB analysis (Fig. 4).Furthermore, the fraction of antibodies with a supportiveWB result increased with increasing IHC reactivity. In thegroup with strong reactivity in the IHC analysis (IS-3),50% of the tested antibodies gave a supportive WB.

To exemplify the correlation between the intensity ofIHC-staining and the result in the WB analysis, dataachieved from different antibodies are shown in Fig. 5.Analyses using an antibody recognizing the A-kinaseanchoring protein 8, previously described to be located inthe nuclear matrix [28, 29] is shown in Fig. 5A. A clear and

Figure 4. Summary of a pair-wise comparison of WB-score and computer-based image-analysis of IHC-staining intensity on human cell lines RT-4and U-251MG. The IHC intensity is shown on the X-axis where cell scoreIS-0 represents no staining and IS-3 symbolizes strong staining. WB-scores for RT4 are shown by the leftmost bars and data from analysis ofU-251MG are illustrated by the rightmost bars.

© 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 9

In the standardized WB analysis, the predicted molec-ular weight has been used. Due to the diverse experi-mental behavior of different proteins in SDS–PAGE, set bythe stability of the three dimensional structure of the pro-tein of interest, the inherent charge and also the possibil-ity to bind SDS [34, 35] a broader definition of the molec-ular weight has been used (±20% of the molecularweight). This standardized WB reveals that almost half

(45%) of the antibodies gave supportive data with a singlemajor band of the correct size (Fig. 1 and 2A). However, byexpressing the protein target recombinantly in a humankidney cell line and analyze the lysates with WB, as manyas 82% of the antibodies with unsupportive data in thestandardized assays yielded supportive staining in thisassay with recombinant protein present (Fig. 2B). Hence,90% of the used antibodies were able to selectively rec-

www.biotecvisions.comwww.biotechnology-journal.com

BiotechnologyJournal Biotechnol. J. 2014, 9

Figure 5. Results from comparison of WB and IHC-analysis on human cell lines. Examples of IHC and WB results for (A) an antibody directed to the76 kDa A-kinase anchoring protein 8 (AKAP8). (B) An antibody directed toward the CD44 protein, which is reported to have transcripts encoding proteinsof six different molecular weights, 3, 16, 39, 53, 77, and 82 kDa. (C and D) Two different antibody raised against the 31 kDa C/EBP epsilon protein, one produced within the Protein Atlas program (C) and the other provided from an external source (D).

10 © 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

ognize the aimed target protein when used in the WBapplication. A similar pattern could be seen among theexternally provided antibodies, of which 84% have beensuccessfully validated by the suppliers in specificallydesigned WB experiments, but in the standard setup only48% gave supportive results. Hence, the validity of theacquired data is very dependent on the experimental setup and confirmation by using another experimental plat-form or another biosample may not be valid.

To illustrate application-dependence of antibodies,we have shown data from the use of three antibodies forrecognition of a well-known protein, the estrogen recep-tor 1 (Fig. 3). All three antibodies were validated for cor-rect staining in the IHC, but still the data gained from thestandardized WB analysis gave conflicting results, exceptfor antibody HPA000449, which were able to recognize aprotein with the expected molecular weight. When usingthe estrogen receptor in the WB analysis, all antibodieswere able to recognize the target protein. The acquiredresults show that the origin and treatment of the biosam-ples clearly influence the possibility for the antibodies torecognize the aimed target protein selectively. Hence,careful interpretation of the data is needed to understandand validate the obtained results.

As expected, the results from a pair-wise comparisonof IHC and WB data support the conclusion that theamount of target protein in the sample very much influ-ences the results of the analysis (Fig. 4). The fact that anti-bodies might perform differently in IHC and WB applica-tions makes interpretation of the results obtained for pro-teins with unknown distribution and expression patternschallenging. Also, the inherent limitation with the WBtechnique, both regarding accuracy and resolution inmolecular weight assessment and quantification is fur-ther complicating the possibility to draw firm conclu-sions. For example, a long exposure time of the WB mem-brane or too high concentration of primary antibodymight give bands from low affinity interactions with oth-er proteins present at much higher concentrations thanthe target protein.

The benefit of using antibodies when elucidating thepresence, localization, and amount of a certain protein isobvious and many different platforms that can be used forthis purpose are available. The ability of an antibody toselectively bind to a specific protein is very hard to assesssince the epitopes that are recognized might be differ-ently exposed depending on sample preparation andchoice of technical platform. Hence, an antibody thatselectively binds to a certain epitope in one platformmight be able to recognize another protein with a similarepitope if another sample preparation or experimentalplatform is used. Moreover, the relative concentration dif-ference between the aimed target protein and a possibleoff-target epitope will influence the specificity of the anti-body. If the data achieved is to be used for quantification,a certain caution has to be taken depending on the

method for quantification used. Since, the antibody mightbe able to bind to a protein complex that includes the pro-tein of interest, an overestimation of the amount of targetprotein might be the result.

In summary, we show that the performance of an anti-body depends on both the affinity of the antibody towardthe targets as well as the relative concentration of the tar-get and other proteins in the sample (“off-target-pro-teins”). The results clearly demonstrate that even an anti-body validated as highly specific toward its target mightstill show cross-reactivity to off-target-proteins if thoseare present in much higher concentrations in the sample.This means that the outcome of antibody validation iscontext- and application-dependent. Caution is neededwhen interpreting the results, especially if the targetingantibody is directed toward low abundant proteins. More-over, depending on the exact epitope that the antibodyinteracts with, there is a risk that more than one proteinfrom the same protein family will be recognized. The reac-tivity of the antibody is also dependent on how it is used,coupled on a solid phase or in solution. By supplementingthe antibody-based analysis by measuring the expressionlevels of mRNA, the achieved data can be further verified[36]. Also, more than one antibody targeting the protein ofinterest can be used in order to ensure correctness of thedata. Two antibodies that target the same protein couldbe used in parallel, analyzing the same sample but also toconfirm a similar patter within the same experiment, forexample as in 2D-GE analyses.

The authors would like to thank all colleagues for theirinvaluable contribution to the HPA project. The HPA pro-gram is financed by grants from the Knut and Alice Wal-lenberg Foundation.

The authors declare no conflict of interest.

5 References

[1] Anderson, N. L., Anderson, N. G., The human plasma proteome: His-tory, character, and diagnostic prospects. Mol. Cell. Proteomics2002, 1, 845–867.

[2] Söderlind, E., Carlsson, R., Borrebaeck, C. A., Ohlin, M., The immunediversity in a test tube-non-immunised antibody libraries and func-tional variability in defined protein scaffolds. Comb. Chem. HighThroughput Screen. 2001, 4, 409–416.

[3] Liu, B., Marks, J. D., Applying phage antibodies to proteomics:Selecting single chain Fv antibodies to antigens blotted on nitrocel-lulose. Anal. Biochem. 2000, 286, 119–128.

[4] Nord, K., Gunneriusson, E., Ringdahl, J., Stahl, S. et al., Binding pro-teins selected from combinatorial libraries of an alpha-helical bacte-rial receptor domain. Nat. Biotechnol. 1997, 15, 772–777.

[5] Clark, S. L., Remcho, V. T., Aptamers as analytical reagents. Elec-trophoresis 2002, 23, 1335–1340.

[6] Marasco, W. A., Intrabodies: Turning the humoral immune systemoutside in for intracellular immunization. Gene Ther. 1997, 4, 11–15.

www.biotechnology-journal.com www.biotecvisions.com

BiotechnologyJournal Biotechnol. J. 2014, 9

© 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 11

[7] Forrer, P., Binz, H. K., Stumpp, M. T., Plückthun, A., Consensusdesign of repeat proteins. ChemBioChem 2004, 5, 183–189.

[8] Bjorling E., Uhlen, M., Antibodypedia, a portal for sharing antibodyand antigen validation data. Mol. Cell. Proteomics 2008, 7, 2028–2037.

[9] Burnette, W. N., “Western blotting’’: Electrophoretic transfer of pro-teins from sodium dodecyl sulfate—polyacrylamide gels to unmodi-fied nitrocellulose and radiographic detection with antibody andradioiodinated protein A. Anal. Biochem. 1981, 112, 195–203.

[10] Uhlen M., Oksvold, P., Fagerberg, L., Lundberg, E., Jonasson, K., et al., Towards a knowledge-based human protein atlas. Nat. Bio -technol. 2010, 28, 1248–1250.

[11] Nilsson, P., Paavilainen, L., Larsson, K., Odling, J. et al., Towards ahuman proteome atlas: High-throughput generation of mono-spe-cific antibodies for tissue profiling. Proteomics 2005, 5, 4327–4337.

[12] Westermark, B., Ponten, J., Hugosson, R., Determinants for theestablishment of permanent tissue culture lines from humangliomas. Acta Pathol. Microbiol. Scand. A 1973, 81, 791–805.

[13] Chen, T. R., In situ detection of mycoplasma contamination in cellcultures by fluorescent Hoechst 33258 stain. Exp. Cell Res. 1977, 104,255–262.

[14] Boutet, E., Lieberherr, D., Tognolli, M., Schneider, M. et al., UniPro-tKB/Swiss-Prot: The manually annotated section of the UniProtKnowledgeBase. Methods Mol. Biol. 2007, 406, 89–112.

[15] Flicek, P., Aken, B. L., Beal, K., Ballester, B. et al., Ensembl 2008.Nucleic Acids Res. 2007, 36, D707–D714.

[16] Mulder, N. J., Kersey, P., Pruess, M.Apweiler, R., In silico characteri-zation of proteins: UniProt, InterPro and Integr8. Mol. Biotechnol.2008, 38, 165–177.

[17] Allen, H., Kisailus, E., Glycoconjugates – Composition, Structure andFunction, Marcel Dekker, New York, USA 1992.

[18] Andersson, A. C., Stromberg, S., Backvall, H., Kampf, C. et al., Analy-sis of protein expression in cell microarrays: A tool for antibody-based proteomics. J. Histochem. Cytochem. 2006, 54, 1413–1423.

[19] Kampf, C., Olsson, I., Ryberg, U., Sjostedt, E. et al., Production of tis-sue microarrays, immunohistochemistry staining and digitalizationwithin the human protein atlas. J Vis Exp. 2012; 63, 3620.

[20] Stromberg, S., Bjorklund, M. G., Asplund, C., Skollermo, A. et al., A high-throughput strategy for protein profiling in cell microarraysusing automated image analysis. Proteomics 2007, 7, 2142–2150.

[21] Berglund, L., Bjorling, E., Oksvold, P., Fagerberg, L. et al., A gene-centric human protein atlas for expression profiles based on anti-bodies. Mol. Cell. Proteomics 2008, 7, 2019–2027.

[22] Barbe, L., Lundberg, E., Oksvold, P., Stenius, A. et al., Towards a con-focal subcellular atlas of the human proteome. Mol. Cell. Proteomics2007, 7, 499–508.

[23] Agaton, C., Falk, R., Hoiden Guthenberg, I., Gostring, L. et al., Selec-tive enrichment of monospecific polyclonal antibodies for antibody-based proteomics efforts. J. Chromatogr. A 2004, 1043, 33–40.

[24] Berglund, L., Bjorling, E., Jonasson, K., Rockberg, J. et al., A whole-genome bioinformatics approach to selection of antigens for sys-tematic antibody generation. Proteomics 2008, 8, 2832–2839.

[25] Uhlen, M., Bjorling, E., Agaton, C., Szigyarto, C. A. et al., A humanprotein atlas for normal and cancer tissues based on antibody pro-teomics. Mol. Cell. Proteomics 2005, 4, 1920–1932.

[26] Kohler, G., Milstein, C., Continuous cultures of fused cells secretingantibody of predefined specificity. Nature 1975, 256, 495–497.

[27] Heldring, N., Pike, A., Andersson, S., Matthews, J. et al., Estrogenreceptors: How do they signal and what are their targets. Physiol.Rev. 2007, 87, 905–931.

[28] Akileswaran, L., Taraska, J. W., Sayer, J. A., Gettemy, J. M. et al., A-kinase-anchoring protein AKAP95 is targeted to the nuclearmatrix and associates with p68 RNA helicase. J. Biol. Chem. 2001,276, 17448–17454.

[29] Coghlan, V. M., Langeberg, L. K., Fernandez, A., Lamb, N. J. et al.,Cloning and characterization of AKAP 95, a nuclear protein thatassociates with the regulatory subunit of type II cAMP-dependentprotein kinase. J. Biol. Chem. 1994, 269, 7658–7665.

[30] Eide, T., Coghlan, V., Orstavik, S., Holsve, C. et al., Molecular cloning,chromosomal localization, and cell cycle-dependent subcellular dis-tribution of the A-kinase anchoring protein, AKAP95. Exp. Cell Res.1998, 238, 305–316.

[31] Ponta, H., Sherman, L.Herrlich, P. A., CD44: From adhesion mole-cules to signalling regulators. Nat. Rev. Mol. Cell Biol. 2003, 4, 33–45.

[32] Lekstrom-Himes, J. A., The role of C/EBP(epsilon) in the terminalstages of granulocyte differentiation. Stem Cells 2001, 19, 125–133.

[33] Yamanaka, R., Lekstrom-Himes, J., Barlow, C., Wynshaw-Boris, A. et al., CCAAT/enhancer binding proteins are critical components ofthe transcriptional regulation of hematopoiesis (Review). Int. J. Mol.Med. 1998, 1, 213–221.

[34] Matagne, A., Joris, B., Frere, J. M., Anomalous behaviour of a pro-tein during SDS/PAGE corrected by chemical modification of car-boxylic groups. Biochem. J. 1991, 280(Pt 2), 553–556.

[35] Creamer, L. K., Richardson, T., Anomalous behavior of bovine alphas1- and beta-caseins on gel electrophoresis in sodium dodecyl sul-fate buffers. Arch. Biochem. Biophys. 1984, 234, 476–486.

[36] Danielsson, F., Wiking, M., Mahdessian, D., Skogs, M. et al., RNAdeep sequencing as a tool for selection of cell lines for systematicsubcellular localization of all human proteins. J. Proteome Res. 2013,12, 299–307.

www.biotecvisions.comwww.biotechnology-journal.com

BiotechnologyJournal Biotechnol. J. 2014, 9