Referencing cross-reactivity of detection antibodies for protein array experiments

Protein arrays are frequently used to profile antibody repertoires in humans and animals. High-throughput protein array characterisation of complex antibody repertoires necessitates the use of extensively validated secondary detection antibodies. This article details the validation of an affinity-isolated anti-chicken IgY antibody produced in rabbit and a goat anti-rabbit IgG antibody conjugated with alkaline phosphatase using protein arrays consisting of 7,390 distinct human proteins. Probing protein arrays with secondary antibodies in absence of chicken serum revealed non-specific binding to 61 distinct human proteins. Despite the identified non-specific binding, the tested antibodies are well suited for use in protein array experiments as the cross-reactive binding partners can be readily excluded from further analysis. The evident cross-reactivity of the tested secondary detection antibodies points towards the necessity of platform-specific antibody characterisation studies for all secondary immunoreagents. Furthermore, secondary antibody characterisation using protein arrays enables the generation of reference lists of cross-reactive proteins, which can be then marked as potential false positives in follow-up experiments. Providing such cross-reactivity reference lists accessible to the wider research community may help to interpret data generated with the same antibodies in applications not only related to protein arrays such as immunoprecipitation, Western blots or other immunoassays.


REVISED Amendments from Version 1 Introduction
Secondary label-conjugated and non-conjugated detection antibodies are frequently used in a wide range of research applications. However, they are often affinity-isolated, polyclonal reagents that may lack the highest standard of antibody validation. The antibodies characterised in this study are a polyclonal anti-chicken IgY antibody produced in rabbit (31104, Thermo Fisher) and a polyclonal goat anti-rabbit IgG antibody conjugated with alkaline phosphatase (AP) (A3687, Sigma-Aldrich). Although the use of the rabbit anti-IgY antibody in the literature is limited, the goat antirabbit IgG AP has been extensively utilised for over 15 years 1,2 .
The research conducted in this laboratory examines complex antibody repertoires in humans and animals by means of protein arrays. Protein arrays are frequently used to profile antibody binding to human proteins in autoimmune disease 3 , cancer 4 and in healthy individuals 5 . Other protein array applications include recombinant 6 and hybridoma-derived 7 antibody characterisation studies. This article investigates the cross-reactivity of a rabbit anti-chicken IgY and an alkaline phosphatase-conjugated goat anti-rabbit IgG, which were used for the profiling of IgY antibody responses to human antigens in chickens immunised with human cancer cells. The protein array technology applied here, developed by Büssow and colleagues 8 , is comprised in its current version of a fully annotated set of 7,390 distinct human proteins, that may serve as potential antigens. The aim of this study is to define a cross-reactivity reference list for the two described secondary antibodies, which can then be used to eliminate non-specific binders from ongoing chicken IgY profiling studies. Furthermore, publication of the crossreactivity reference list provides a valuable resource of potential false-positive binders to researchers using the same antibodies.

Antibody details
Rabbit anti-chicken IgY (H+L) secondary antibody (Thermo Fisher Scientific, Product number 31104, Lot code PK19380211) is a polyclonal antibody that targets the variable heavy and light chains of chicken IgY immunoglobulins (Table 1). The antibody was isolated from the serum of the antigen-immunised rabbit through immunoaffinity chromatography using antigen coupled to agarose beads. The antibody was added to the protein array at a 1/1,000 dilution in 2% (w/v) bovine serum albumin (BSA, Sigma-Aldrich, A2153) in tris-buffered saline (TBS, Trizma ® Base, Sigma-Aldrich, T6066 and sodium chloride, Fisher Scientific, S/3160/68) with 0.1%, v/v, Tween 20 (Sigma-Aldrich, P1379).
Alkaline phosphatase-conjugated goat anti-rabbit IgG (whole molecule) (Sigma-Aldrich, Product number A3687, Lot code SLBJ6146V) is a polyclonal antibody that targets all rabbit IgGs ( Table 1). The antibody was isolated through immunospecific purification of antisera from a rabbit IgG-immunised goat. Following isolation, the anti-rabbit IgG was conjugated to alkaline phosphatase using glutaraldehyde-based cross-linkage. The antibody was added to the protein array at a 1/1,000 dilution in 2% (w/v) BSA in tris-buffered saline (TBS) with 0.1%, v/v, Tween 20.

Protein arrays
Unipex protein arrays were obtained from Source Bioscience Life Sciences (Nottingham, UK). The Unipex arrays comprise of 15,300 fully annotated E. coli clones expressing a total of 7,390 distinct in-frame ORF human recombinant proteins. The Unipex proteins are immobilized under denaturing conditions directly on the PVDF membrane surfaces exposing linear sequence epitopes ideally suited for epitope mapping, antibody profiling and antibody crossreactivity analyses. The details of protein arrays utilised in this study are provided in Table 2. For general information on Unipex protein arrays please refer to: (http://www.lifesciences.sourcebioscience.com/media/290406/sbs_ig_manual_proteinarray_v1.pdf). Cross-reactivity assessment Antibody cross-reactivity was assessed using Unipex protein arrays. The detailed experimental protocol is provided in Table 3. Briefly, secondary rabbit anti-chicken IgY and goat anti-rabbit IgG AP were validated in preparation for a chicken IgY antibody profiling experiment of a chicken immunised with human cancer cells. Protein arrays were probed with secondary antibodies in the absence of IgYcontaining chicken serum, as described in Table 3. Signal generation for array-bound secondary antibodies was obtained using AttoPhos AP fluorescent substrate system (Promega, S1001) diluted 1 in 8 in AP buffer (1mM MgCl2, Sigma-Aldrich, M4880 and 100mM Tris base, pH 9.5). Protein array image acquisition was conducted using a Fuji scanner Fla5100. Positive signals were localized according to the manufacturer's protocol. Briefly, array proteins were spotted in duplicate in a 3×3 square pattern. The centre spot of each square being a guide dot surrounded by eight flanking protein spots. Each protein was spotted around the navigation dot in one of four predetermined patterns (see Figure 1b). Varying background intensities were controlled by adjusting brightness and contrast of the image using Visual Grid software (GPC Biotech) to allow best possible scoring conditions. The degree of signal intensity was evaluated for each protein pair with the value 1 corresponding to a weak signal, value 2 corresponding to a moderate signal and value 3 corresponding to a strong signal. The x-y-coordinates of each positive pattern were merged with the Unipex protein database provided by the manufacturer (Source Bioscience) resulting in identification of GenBank and UniGene ID's for each positive signal. This is a commonly used method for scoring signal intensities as previously shown by this group 4 and others 9,10 . Protein annotations were retrieved from the Unipex database provided by the manufacturer and updated using the National Cancer Institute's UniGene CGAP Gene Finder tool (http://cgap.nci.nih.gov/Genes/ GeneFinder).

Epitope analysis
To investigate whether antibody binding to protein arrays was due to epitope similarities between the animal immunogens used to produce the secondary antibodies and the human proteins on the arrays we performed a comparative analysis as follows. Sequences of human antigens on the array bound by the secondary antibodies were obtained from the PubMed website (http://www.ncbi.nlm. nih.gov/protein/) using IDs present in the Unipex protein database and compared to chicken immunoglobulin proteins [Ig lambda chain C region (NCBI Accession: P20763.1), Ig lambda chain V-1 region (NCBI Accession: P04210.1), immunoglobulin Y heavy chain constant region (NCBI Accession: XP_015130394.1) and immunoglobulin Y heavy chain variable region (NCBI Accession: ADF29959.1)], as well as to rabbit immunoglobulin proteins [Ig gamma chain C region (UniProtKB: P01870), immunoglobulin heavy chain VDJ region, partial (NCBI Accession: AAA51320.1), Ig lambda chain C region (UniProtKB: P01847.2) and Ig lambda chain variable region, partial (NCBI Accession: AAA31364.1)]. For antigen similarity comparisons, sequence similarities were analysed using BLAST. Non-intersecting protein sequence alignments were analysed using the local similarity program SIM adjusted to the BLOSUM62 comparison matrix to ensure amino  acid complementarity of linear B-cell epitopes as previously shown 11 . The threshold for sequence similarity was set to BLAST E-values below 1 × 10 -10 and SIM score values above 50.

Results
Probing protein arrays with antibodies allows the assessment of their specificity and cross-reactivity across a large numbers of potential antigens in parallel 12,13 . Here we investigated the crossreactivity of a secondary rabbit anti-chicken IgY and a goat antirabbit IgG labelled with AP, using a single set of human protein arrays in the absence of chicken serum. We identified a total of 63 binding events, of which 61 corresponded to unique proteins ( Table 4). The identified positive signals varied in strength, as shown in Figure 1, with intensity 3 being the strongest and 1 the weakest. Five of the identified signals were scored as intensity 3, twelve signals were scored as intensity 2 and remainder scored as intensity 1. The original protein array images are shown in Figure S1 and Figure S2 (Supplementary material) and protein array images with highlighted positive signals, which correspond the cross-reactive proteins listed in Table 4, are shown in Figure S3 and Figure S4 (Supplementary material).
The 61 identified proteins comprised of a wide range of human proteins, including immunoglobulins, as well as a variety of nuclear, cytoplasmic and cell-membrane proteins with a diverse range of functions (Table 4). In order to identify shared epitopes that could explain the observed antibody cross-reactivity, and to deduce the origin of the non-specific binding to either of the two tested antibodies, we investigated sequence similarities between the human proteins and the immunogens used to produce the antibodies. We conducted a linear (BLAST) and a segmented (SIM) in silico sequence analysis of chicken IgY and rabbit IgG immunoglobulins against 61 arrayidentified human proteins as detailed in Supplementary Table 1. In total, 5 proteins met the BLAST threshold criteria of E-values below 1 × 10 -10 , as well as the SIM threshold criteria of scores above 50. A further 9 proteins met the SIM threshold, but they did not meet the BLAST criteria (Table 5).
All 5 proteins that met both threshold criteria belong to the immunoglobulin class of proteins, four being variable regions of Ig heavy chains and one a heavy constant gamma chain. The in silico sequence analysis revealed the highest sequence similarity to Ig heavy variable and constant chains, respectively, of both, the chicken IgY and rabbit IgG in all cases (Supplementary Table 1) making it impossible with this approach to deduce the origin of the cross-reactivity.
The 9 proteins that met only the SIM threshold criteria belong to a wide range of protein classes, however, none of those proteins belongs to the immunoglobulin class of proteins. In silico sequence analysis revealed that 8 of those proteins have a high local sequence similarity to the chicken immunoglobulin Y heavy chain constant region, but not to any other chicken and rabbit Ig regions. The analysis revealed further, that Inverted Formin, FH2 and WH2 Domain Containing (INF2) showed high local similarity exclusively to the rabbit Ig gamma chain constant region (Supplementary Table 1).

Conclusion
This work illustrates the cross-reactivity of an antibody-based detection system for IgY binding. The polyclonal anti-IgY rabbit antibody in combination with an anti-rabbit IgG alkaline phosphatase-conjugated antibody was shown to bind to 61 human proteins present on Unipex protein arrays comprising of 7,390 human proteins. Characterisation of this cross-reactivity provides a 'false-positive' database for future chicken antisera characterisation on protein array systems not limited to the Unipex protein array used here. These results, in combination with 'false-positives' from earlier research investigating antibody cross-reactivity by this group 12 and others 13 may provide valuable information for future protein array-based experiments. Reference lists provided by such experiments would be further strengthened by arrays that include additional portions of the human proteome and/or post-translational modifications. Using antibodies that have been extensively characterised on protein arrays will reduce the risk of identifying irrelevant cross-reactive secondary antibody binding to the array as a host-antigen response.
It is important to note that the current study was a one-off experiment and repeat experiments may increase the reliability of the data. The reproducibility of the binding events identified in this study was further warranted by evaluating each protein in two discrete positions on the array. Of the 63 binding events, five were scored as intensity 3, twelve were scored as intensity 2 and the remainder were intensity 1. While the assay is unable to conclusively distinguish the precise cause of the differences in signal intensities, it can be assumed to be due to variations in antibody affinity and avidity, the availability of the epitope for binding, and protein concentrations on the array. A follow-up quantitative Western blot analyses and titration experiments would help further to shed more light into differences in antigen-binding kinetics.
The secondary antibodies utilized in this study are polyclonal, isolated by immunoaffinity chromatography. The presented crossreactivity reference list may, therefore, show some variation when a different lot of the antibody is used. We have previously shown that conditions applied during affinity chromatography may affect specificity 14 . When assessing protein array images, we found a considerable discrepancy in background intensity of array part 1 and 2. It is important to highlight that the part 1 and part 2 of the array are generated from distinct clone libraries of different tissue origin. Part 1 of the array was generated from human brain tissue using a pQE30NST vector, whereas part 2 of the array was generated from different sources of tissue, including T cells and lung tissue, using a pQE80LSN vector. The tissue origin and the utilised bacterial vector are potential contributing factors for the variances in background noise.
Since both antibodies were used as a pair in this study, it was not possible to directly deduce the exact cross-reactivity profile for each individual antibody. We have therefore taken an in silico sequence analysis approach and we found that five of the identified proteins were of the immunoglobulin class of proteins with very high sequence similarities to both, the chicken IgY and the rabbit IgG immunoglobulins. Such cross-reactivity is not surprising considering that the antibodies are polyclonal and the immunogens were immunoglobulins of both hosts. In addition, the data sheet provided with the anti-chicken IgY antibody produced in rabbit (31104, Thermo Fisher) has specified that this antibody may crossreact with immunoglobulins from other species. The data sheet for the goat anti-rabbit IgG AP antibody (A3687, Sigma-Aldrich) has specified binding to all rabbit immunoglobulins. The in silico sequence analysis revealed furthermore 8 proteins with high sequence similarity to chicken IgY heavy chain constant region and one protein with high sequence similarity to rabbit Ig gamma chain constant region. In order to tackle this issue experimentally, a single labelled antibody should be tested on its own in future experiments. Furthermore, if a non-labelled antibody is to be tested, two experiments should be performed, one with a labelled and non-labelled antibody pair such as demonstrated in this study, and one additional experiment with the labelled antibody alone, thereby allowing allocation of exact cross-reactivates by simply subtracting 'falsepositives' from both sets of results.
In conclusion, the antibodies tested in this study showed crossreactivity to unrelated human proteins as well as to human immunoglobulin proteins, which are homologous to the original immunogens. Despite the identified non-specific binding, the tested antibodies are suitable for use in protein array experiments as the cross-reactive binding partners can be readily excluded from further analysis. As both antibodies were used as a pair in this study, the possibility to deduce the exact cross-reactivity profile for each individual antibody may be limited. However, the crossreactivity reference list provided in this paper can be further utilised to validate research using those antibodies in applications other than protein arrays.
Author contributions ROK and GSK designed the study, DL performed the protein array experiments and GSK conducted data analysis. GSK conceived and DL performed in silico sequence analyses. GSK wrote and DL and ROK critically reviewed and edited the article. All authors have agreed to the final content of the manuscript.

Competing interests
The authors do not declare any competing interests.

Grant information
This material is based upon works supported by the Irish Cancer Society Research Fellowship Award CRF10KIJ (GSK), the Science Foundation Ireland under CSET Grant no. 10/CE/B1821 and the Enterprise Ireland Dairy Processing Technology Centre award.
I confirm that the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.     Table 4 are highlighted corresponding to their intensity as red (intensity 3 = strong), green (intensity 2 = intermediate) and yellow (intensity 1 = weak) circles.  Table 4 are highlighted corresponding to their intensity as red (intensity 3 = strong), green (intensity 2 = intermediate) and yellow (intensity 1 = weak) circles.
1. In the present work, the authors have tested direct binding of secondary antibodies to arrays of human proteins.

Open Peer Review
Readers who use array technology may benefit from the present work, since they will become aware of the problem of signals caused by secondary antibodies and not by the primary antibody. It appears that human immunoglobulins are frequently detected by secondary antibodies, which is a useful finding that would likely also be relevant for other secondary antibodies.

4.
It would be interesting how strong the signals caused by the secondary antibodies are in comparison to signals obtained in the presence of a primary antibody.
In comparison, the part 1 image has a much higher background than part 2. It appears that very clear signals were obtained from part 2, but not from part 1. In the part 1 image, there is considerable background and almost all positions have been slightly stained. I would recommend repeating the experiment to verify whether the weak signals obtained on part 1 can be reproduced.
Two secondary antibodies were used in the same experiment. Therefore, it cannot be determined which of the two antibodies gave rise to the signals on the array. This problem should be discussed.
No competing interests were disclosed.

Competing Interests:
I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 16 May 2017 , Dublin City University, Ireland

Gregor Kijanka
The authors would like to thank Dr. Konrad Büssow for his thorough review of this article and his helpful comments. Dr. Büssow points out that the authors should stress that both secondary antibodies were used in the same experiment using one single set of protein arrays. This experimental design issue entails that it cannot be determined which signals are caused by which antibody. We have highlighted and discussed both issues throughout the text and we performed an additional sequence analysis in an approach to clarify the origin of the signals on the in silico protein array. The results of these analyses are presented in the new Table 5 and Supplementary  table 1 and are further discussed in the text.
Dr. Büssow has furthermore highlighted the differences in background signal between the two arrays of the protein array set. The authors have encountered similar background differences when using other sets of antibodies and serum samples and find similar discrepancies in background noise being likely due to different tissues and vectors used for the generation of the distinct expression clone libraries utilized for array 1 and 2. This issue is now specifically highlighted in the article.
This article describes an experiment performed to characterize the background signals in a particular combination of three commercially available research tools: a protein macroarray on PVDF membranes used in conjunction with two antibodies used for detection. As no first antibody or serum was used in this experiment, all the signals could be attributed to unwanted, unspecific reactivity of the detection antibody combination used. A list of genes was generated from these signals that is proposed as a reference database of for other researchers.
In general, the approach is scientifically sound and feasible. Background reactivities may limit antibody-based assays and need to be accounted for. So performing a control experiment without serum or first antibody on a protein array and with just the detection antibodies makes perfectly sense to control for unspecific binding. The title of the paper is appropriate, the abstract gives enough information on the setting. The background information about the antibodies is described in enough detail. However, the narrow focus of the paper and a number of technical issues limit the quality of the paper and its utility for the readership.

Major issues
The experiment was performed only once. Consequently, the reliability of the results will be limited.
Only one specific combination of a protein macroarray with two consecutive detection antibodies was analyzed. It remains unclear, whether the results obtained would apply to other lots of the antibodies or whether they are specific for a certain preparation, limiting the benefit of this protein list as a reference database and also limiting the replication of results by other groups.
The authors suggest that their results may also apply to other protein array systems. This claim needs substantiation, especially in the case of proteins derived from high-throughput cloning E.coli that do not show authentic posttranslational modification patterns and often contain extra amino acid sequences that may cause unspecific binding. The paper discusses cross-reactivity with human Ig genes. A sequence analysis of the other cross-reactive proteins with IgY and rabbit Ig sequences may provide evidence for the mechanisms behind this phenomenon, expanding scope and depth of this so far rather descriptive study.

Minor issues
Antibody concentrations should be given explicitly, e.g. as µg/ml rather than as dilutions.
The procedure of signal quantification and scoring needs to be described in more detail. The description states "Positive signals were localized according to the manufacturer's protocol" -what exactly was done to identify positive signals? The pictures provided show varying background intensities as well as a number of very dark spots that do not appear in the analysis. Which algorithm was used to include or exclude signals? How were the different signal intensities attributed to the score values 1, 2 and 3?
It would be interesting to know why this specific combination of two detection antibodies was used here: a polyclonal anti-chicken IgY antibody produced in rabbit and then a polyclonal goat anti-rabbit IgG antibody conjugated with alkaline phosphatase. Was there no conjugated anti-chicken antibody available? Every additional antibody will add to the number of unspecific reactions, so using just one instead of two may help reduce background.
The abstract does not provide a conclusion on whether the antibodies should be used in a This study presents data concerning the issue of secondary antibody cross-reactivity towards antigens other than desired immunoglobulins. By screening a high-throughput protein array, the authors establish the amount and identity of proteins detected by commercially available secondary antibodies, a rabbit anti-chicken antibody combined with an AP-conjugated goat anti-rabbit antibody. : The title might contain the information that detection antibodies were used. The Title and Abstract two abstract represents a sound summary of the work performed.
The methods used are described clearly, especially by showing a concise work flow as seen in Article: table 3.
Results are described appropriate and sufficiently. Supplementary Figures S1 and S2 have very Data: huge size and are dispensable. The sentence about signal intensity differences due to varying protein amounts should be part of the conclusion section and also discussed more extensively.
: The conclusions drawn are appropriate and concise. Briefly can some information be drawn Conclusion from the kind / category of proteins falsely detected?
No competing interests were disclosed.

Competing Interests:
I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.