Opinion: Independent third-party entities as a model for validation of commercial antibodies

Abstract

A vast array of commercial antibodies covers a large percentage of human gene products, but determining which among them is most appropriate for any given application is challenging. This leads to use of non-specific antibodies that contributes to issues with reproducibility. It is our opinion that the community of scientists who use commercial antibodies in their biomedical research would benefit from third-party antibody characterization entities that use standardized operating procedures to assess and compare antibody performance. Ideally, such entities would follow the principles of open science, such that all antibodies against any given protein target would be tested in parallel, and all data generated released to the public domain without bias. Furthermore, there should be no financial incentive for the entity beyond cost-recovery. Such non-profit organizations, combined with other scientific efforts, could catalyse new discoveries by providing scientists with better validated antibody tools.

Introduction

High-quality antibodies that recognize single proteins from complex mixtures are essential tools for biomedical research. Hundreds of commercial organizations generate and/or distribute research antibodies, with estimates ranging from between 2.5 million (antibodyregistry.org) to 4.8 million (citeab.com) such reagents. Unfortunately, proper characterization of commercial antibodies is often lacking and approximately $1.75B research dollars are wasted yearly on non-specific antibodies [1–4]. Moreover, the use of non-specific antibodies leads to serious problems with reproducibility [5], with many specific examples recently highlighted [4]. The problem of antibody performance could be alleviated and the literature improved if there were universally accepted and transparent processes to distinguish commercial antibodies that are specific from those that do not recognize their target, or those that do recognize their target but also recognize non-intended targets [3–7]. In many respects, the onus should be on the publishing scientist to ensure that the reagents they purchase are properly characterized, but this may not be realistic: the required technologies are often beyond the scope of an academic laboratory and the cost too high. Alternatively, the onus could be on companies that generate and/or distribute antibodies, but in reality it is too expensive to characterize all but a few antibodies, because most products do not generate significant revenue. A possible solution is for third parties to characterize antibodies on behalf of the antibody manufacturers and the users and distribute costs among participants [8]. In this Opinion article, we describe how we believe such third party entities could function.

Validation strategies

In 2016, the International Working Group for Antibody Validation recommended five validation methods and suggested that at least one be applied by the user to all antibodies [3]. These are: 1) genetic strategies using knockout (KO) or knockdown (KD) samples as controls; 2) orthogonal strategies correlating antibody signal to known information about the target of interest; 3) the ability of two independent antibodies recognizing different epitopes in the same target protein to generate a similar signal; 4) monitoring the signal from an exogenously expressed protein; and 5) immunoprecipitation followed by mass spectrometry to determine if the predicted protein target is a major signal in the sample.

It is our view that the validation methods have unequal scientific value. Commercial suppliers most commonly use orthogonal and overexpression strategies, which are the most cost-effective, but the least useful. The most useful are genetic strategies, which are used to create negative controls for the various immunoassays and can provide near absolute certainty about the selective detection of the target. Genetic strategies are increasingly used to identify high-quality antibodies for targets including those involved in neurodegenerative diseases [9–11]. For example, a systematic analysis of antibodies was performed for the Parkinson’s disease gene LRRK2, using a KO approach. Out of 10 antibodies tested, all detected human recombinant LRRK2 protein, but only two generated a signal on immunoblot (IB) of brain lysates that was lost when the IB was performed on lysates from KO brains [9]. Moreover, only a single antibody generated a signal on immunohistochemistry (IHC) of brain sections that was lost in parallel sections from KO mice [9]. In an example from amyotrophic lateral sclerosis (ALS), we compared the genetic to orthogonal and overexpression methods while analyzing the performance of 16 commercial antibodies from 7 independent suppliers against C9ORF72, a major ALS disease gene. Five commercial antibodies validated for IB by their suppliers using orthogonal and/or overexpression approaches had no specific signal when validated using KO cells [12]. A similar discrepancy was observed in immunofluorescence (IF) as only one of 8 previously validated antibodies was functional for IF [12]. Interestingly and encouragingly, following publication of these data, the vendors responded positively. Five sub-performing C9ORF72 antibodies from two different producers were removed from the market. One of the vendors that had high-quality C9ORF72 monoclonal antibodies generated recombinant versions of them, and another manufacturer asked to test a pre-commercialized C9ORF72 antibody using the KO approach and decided not to go forward with commercialization after assessing the characterization data.

There are many important ongoing efforts to address antibody performance issues as recently reviewed by the Antibody Society [4]. These include: (i) online search tools that help scientists select the most promising commercial antibodies, such as The Antibody Registry (https://antibodyregistry.org/), CiteAb [13] (https://www.citeab.com/), Antibodypedia (https://www.antibodypedia.com/) and Benchsci (https://www.benchsci.com/), each of which uses their own unique filter sets; (ii) large scale initiatives that use innovative approaches to antibody validation [14,15]; and (iii) various validation efforts, including KO and KD, from several high-quality antibody manufacturers. However, most global antibody validation initiatives do not use genetic strategies, and it is also difficult to compare results from experiments done in different labs under a range of different conditions. Thus, we strongly believe that there is a need for independent efforts to characterize antibodies using KO controls when possible.

Applying the KO strategy: bridging manufacturers and end users

Third-party entities using KO strategies could provide a cost-effective mechanism to characterize commercial antibodies for end-users. To remove any conflict of interest and to secure the trust of antibody producers and users, such entities should be open, not for profit, and have no commercialization intent – a public good. A scientific advantage of this model is the capacity to compare the performance of multiple antibodies from multiple manufacturers simultaneously. An economic advantage is that such entities will be more cost effective in toto; in a single experiment they can characterize antibodies from multiple manufacturers who would otherwise have to perform, and pay for, the experiment themselves. The deposition of the data into public databases allows for end users to make informed decisions to identify high-quality reagents, with enormous time and cost savings to the system.

Reduction to practice: piloting an open access experimental hub

To test the viability of a third-party hub, seven industry-leading antibody manufacturers, the Structural Genomics Consortium and scientists from the Tanenbaum Open Science Institute at the Montreal Neurological Institute, McGill University have established a company called YCharOS (Antibody Characterization through Open Science) to work towards this purpose. YCharOS is wholly owned by a charity formed to promote open science, which provides assurance that all data is placed immediately into the public domain with no restriction on use. YCharOS has no investors, is not for profit, and will never seek patents or other forms of intellectual property. The founders have no financial stake and will not profit from the activities, and thus there is no personal financial conflict of interest in this article. Companies, charities, and individuals pay YCharOS for independent, antibody characterization, with all profits put back into the function of the organization. Note that we do not believe or state that YCharOS is “the” solution to all antibody validation issues, but rather is serving as a pilot for the concept. In the following sections we discuss procedures by which YCharOS is operating as means to provide principles by which independent antibody characterization entities could operate.

Operationally, YCharOS works with senior scientific and executive leaders from the seven companies to help understand what the major issues are that the companies face, and how best to position YCharOS between the companies and academic antibody users.

YCharOS will make KO cells in lines that express the targets of interest

The genetic/KO strategy relies on the identification of cell line(s) with meaningful endogenous expression of relevant target proteins. Proteomic databases such as PAXdb (pax-db.org) [16] and DepMap (depmap.org/portal/) [17,18] use normalized peptide counts from shotgun proteomics of tissues and cell lines to approximate protein expression levels, which vary over ~7 orders of magnitude. To explore the utility of these databases, we searched for lines that express 14 newly discovered Alzheimer’s disease (AD)-related proteins and 29 ALS-disease-related proteins (Table 1). The ALS genes were selected from the literature and the AD genes were nominated by researchers within the AMP-AD Target Discovery and Preclinical Validation Consortium [19] based on high dimensional computational studies of human brain samples. These and other novel targets and the supporting evidence for their selection are shared through the Agora platform (https://agora.adknowledgeportal.org/).

Table 1.

Search engine results for the AMP-AD nominated protein targets and ALS-linked proteins.

	selected targets		CiteAb	The Antibody Registry		PAXdb			DepMap
	Protein name	Uniprot ID	number of antibodies	number of antibodies	number of renewable antibodies	high-expessing cell lines	abundance (ppm)	Rank	high-expressing cell line
AMP-AD nominated protein targets	CD44	P16070	5449 from 92 suppliers	3164	2712	HeLa	453	441. out of 3916 [top 25 %]
	SMOC1	Q9H4F8	127 from 25 suppliers	49	22	HeLa	113	5687. out of 8116
	PRDX6	P30041	692 from 58 suppliers	333	164	HeLa	4555	18. out of 8116 [top 5 %]
	MSN	P26038	1917 from 70 suppliers	431	211	HeLa	5459	16. out of 8817 [top 5 %]
	DNM1	Q05193	1240 from 66 suppliers	208	82	U2OS	161	2584. out of 10,225
	PLEC	Q15149	229 from 44 suppliers	20	10	U2OS	9862	7. out of 5387 [top 5 %]
	PRDX1	Q06830	643 from 64 suppliers	314	172	U2OS	19,248	11. out of 7202 [top 5 %]
	GSN	P06396	862 from 64 suppliers	383	163	U2OS	263	464. out of 7202 [top 10 %]
	SYNGAP1	Q96PV0	198 from 46 suppliers	57	7	Poorly expressed in U2OS/HeLa	<2	–	U937
	SLC25A11	Q02978	234 from 47 suppliers	85	21	HeLa	356	533. out of 8116 [top 10 %]
	SYT1	P21579	1007 from 69 suppliers	1549	351	Poorly expressed in U2OS/HeLa	0	–	SHP77
	TYROBP	O43914	219 from 44 suppliers	137	16	Poorly expressed in U2OS/HeLa	0	–	U937
	VGF	O15240	355 from 51 suppliers	71	3	Poorly expressed in U2OS/HeLa	<38	–	SHP77
	RGMa	Q96B86	229 from 40 suppliers	76	17	Poorly expressed in U2OS/HeLa	0	–	COLO320
ALS-linked proteins	C9ORF72	Q96LT7	218 from 41 suppliers	29	7	U2OS	130	4529. out of 10,225
	SOD1	P00441	1707 from 77 suppliers	603	168	HeLa	3179	16. out of 3857 [top 5%]
	FUS	P35637	1830 from 70 suppliers	162	36	HeLa	1914	130. out of 3916 [top 5 %]
	NEK1	Q96PY6	138 from 32 suppliers	35	1	HeLa	113	5719. out of 8116
	TBK1	Q9UHD2	646 from 71 suppliers	292	107	HeLa	160	3091. out of 10,225
	OPTN	Q96CV9	336 from 49 suppliers	172	15	U2OS	153	3222. out of 10,225
	ATXN2	Q99700	353 from 48 suppliers	65	5	HeLa	159	3186. out of 10,225
	CHCHD10	Q8WYQ3	90 from 20 suppliers	15	4	Poorly expressed in U2OS/HeLa	<52	–	SKCO1
	VCP	P55072	1132 from 66 suppliers	226	106	HeLa	2249	91. out of 5379 [top 5 %]
	UBQLN2	Q9UHD9	228 from 50 suppliers	53	26	U2OS	246	683. out of 4013 [top 25 %]
	SQSTM1	Q13501	1049 from 57 suppliers	56	11	HeLa	328	776. out of 3857 [top 25 %]
	Kif5a	Q12840	248 from 42 suppliers	92	3	U2OS	98	5214. out of 10,225
	SETX	Q7Z333	271 from 26 suppliers	37	18	HeLa	156	3341. out of 10,225
	MATR3	P43243	215 from 38 suppliers	84	21	HeLa	1057	236. out of 3857 [top 10 %]
	VAPB	O95292	299 from 48 suppliers	108	17	HeLa	576	368. out of 8116 [top 5 %]
	TUBA4A	P68366	537 from 55 suppliers	151	95	HeLa	773	353. out of 3857 [top 10 %]
	SPAST	Q9UBP0	221 from 40 suppliers	125	44	HeLa	159	3142. out of 10,225
	FIG4	Q92562	277 from 40 suppliers	44	26	HeLa	114	5642. out of 8116
	GRN	P28799	557 from 53 suppliers	144	58	HeLa	184	1178. out of 3857
	PFN1	P07737	601 from 59 suppliers	117	33	U2OS	17,843	13. out of 7202 [top 5 %]
	SPG11	Q96JI7	98 from 32 suppliers	32	3	U2OS	115	4875. out of 10,225
	ALS2	Q96Q42	239 from 41 suppliers	120	28	HeLa	80	7519. out of 8116 [bottom 10 %]
	ANG	P03950	901 from 63 suppliers	177	75	Poorly expressed in U2OS/HeLa	0	–	SHP77
	ANXA11	P50995	267 from 47 suppliers	148	15	HeLa	276	745. out of 8817 [top 10 %]
	CHMP2B	Q9UQN3	262 from 43 suppliers	69	28	HeLa	235	829. out of 8817 [top 10 %]
	CCNF	P41002	186 from 44 suppliers	34	14	Poorly expressed in U2OS/HeLa	<2	–	ln18
	GLE1	Q53GS7	183 from 36 suppliers	81	7	HeLa	152	3550. out of 10,225
	SIGMAR1	Q99720	200 from 42 suppliers	81	7	HeLa	296	846. out of 3857 [top 25 %]
	TIA1	P31483	531 from 54 suppliers	130	29	HeLa	316	63. out of 10,225 [top 5 %]

To obtain an overview of the total number of available antibodies for each protein target, each protein was searched using CiteAb (citeab.com) and The Antibody Registry (antibodyregistry.org). Then, each protein was searched in PAXdb (pax-db.org) to evaluate whether it is present to an adequate level in either Hela or U2OS cell lines for antibody characterization studies. DepMap was used to search protein targets poorly expressed in HeLa or U2OS lines.

We first determined the expression of these proteins with PAXdb. If no cell line with satisfactory expression was identified, we searched DepMap, which provides data for additional cancer cell lines. Although the proteins in Table 1 are involved in neurodegenerative disease, cell lines with clearly defined expression were identified for most targets. It is important to emphasize that the cell line and its KO derivative are merely vehicles to identify selective antibodies, and they do not necessarily provide insight into the normal role(s) of the protein. Interestingly, HeLa and U2OS cells, both highly amenable to rapid CRISPR/Cas9 editing to generate KO lines, appear suitable for 35 of the 43 disease targets. It is therefore anticipated that a small set of immortal cell lines could cover most of the human proteome. However, it is clear that protein abundance databases based on shotgun proteomics are not exhaustive and can be inaccurate. For example, cells in the database that express the target may have been overlooked because the specific proteomics method used did not detect that protein (e.g. lack of trypsin sites, highly modified proteins). Alternatively, cell lines with high expression may not been included in the database; this may be particularly true for proteins most highly expressed in the nervous system or other specialized cell lines or tissues. In some instances, the abundance of any given target in these databases is over-estimated, such that specific and selective antibodies are missed because the levels of the protein are actually low in that cell line. Of course using a KO-based approach with a limited subset of cell lines will be inappropriate for proteins with highly restricted tissue expression or essential genes for which KO is not possible. However, with these caveats as given, we believe this approach can be used to characterize antibodies for a large number of human targets in the most cost-effective manner possible.

Antibody testing

With appropriate KO cell lines for any given protein target in place, the next step is to screen all commercial antibodies advertised for that target for use in IB, immunoprecipitation (IP), and IF (Fig. 1). While there are multiple applications that are required by users, the KO approach is most amenable to these three. For human proteins, IHC is an important applications but one that is not readily suited to this approach, since it is not possible to access human tissue from gene KOs. One might envisage using mouse KO tissues to characterize antibodies to the human homologue, but this assumes that the epitope is conserved across species and the “off-target” pattern is conserved – which is risky.

Fig. 1.

Overview of YCharOS pipeline.

1. All nominated antibodies are tested by immunoblot using WT and KO cells.
2. All nominated antibodies are tested by immunoprecipitation using WT cells.
3. All nominated antibodies are tested by immunofluorescence using WT and KO cells.
4. An open-access antibody characterization report is prepared and presented on the public domain.

In the KO-based process, all antibodies against any given protein target are first screened in side-by-side comparisons with wild-type cell lysates next to lysates in which CRISPR/Cas9 is used to KO the target gene to 1) confirm the KO status of the cell line and 2) monitor the loss of the signal from the KO lysate indicating antibody specificity. It is conceivable that antibodies could be first tested against overexpressed or purified proteins but it is our opinion that if the antibody does not detect the protein at an endogenous level it has limited value. A strength of the approach is that generally 5–15 antibodies are tested against each target. If no antibodies reveal a specific band then this forces a re-evaluation of 1) the KO status of the line and 2) the protein abundance in that line. If even a single antibody recognizes a specific band (a positive control) then the KO status of the cells is confirmed and concerns regarding expression levels in that cell are lessened.

All antibodies, even those that lack specificity by IB, are then tested by IP and IF. Antibody capture of the target by IP is monitored by IB with a KO-validated antibody. Antibody performance in IF is monitored by comparing the signal in wild-type and KO cells. In the IF approach, wild-type and KO cells are labeled with different colored fluorescent dyes and the two cell lines are plated on single coverslips in order to compare wild-type and KO cells in the same microscopic field.

Although identification of high performing antibodies for IB, IP or IF does not necessarily mean that they can be automatically translated for use in other applications, the data provide researchers with a subset of high-quality antibodies promising for other applications. It is the onus of the individual researcher to test antibodies for other applications using their own KO models. Any entity attempting to validate antibodies will need to be part of a larger community-based initiative. Antibodies targeting post-translational modifications will not be tested by YCharOS but other initiatives may incorporate appropriate characterization methods.

Which antibodies to characterize?

Any antibody characterization entity should ideally focus on renewable antibodies. Another option is to prioritize highly-cited antibodies (Table 1).

More recently, YCharOS has been working with its antibody manufacturing partners. As new protein targets enter the pipeline, the antibody manufacturers themselves select their most promising antibodies for each target, thus removing this laborious step in selecting antibodies to test. The manufacturers provide the antibodies without cost, thus significantly reducing the cost to YCharOS. All antibodies from all companies are then compared side-by-side and the data is placed in a public repository, where it can be accessed by all, and the companies can use it freely. The participating companies have agreed to this approach, even when it means the public deposition of data on sub-performing antibodies from their organization. The value they take is the enhanced sales of antibodies that pass validation and the cleaning of their own catalogues of antibodies.

Antibody redundancy

A major problem in selecting antibodies comes from a process called original equipment manufacturer (OEM) supply. OEM supply allows multiple companies to sell an antibody generated by a single manufacturer under their own brand without divulging the source of the antibody [20]. This has the advantage of providing commercial access to products in jurisdictions that have smaller markets, but researchers can unknowingly buy the same core antibody from various suppliers. This is particularly problematic if the antibody does not specifically recognize the target.

OEM supply can make it very difficult to determine if products are equivalent. In many cases the only mechanism to identify a core antibody sold by different suppliers is to compare the product datasheets, which may show the same validation images and antibody information. To estimate the degree of redundancy we examined 36 entries from the Antibody Registry that came from a search for C9ORF72 and found: (1) one C9ORF72 antibody is sold by 4 different suppliers, (2) another C9ORF72 antibody is sold by 2 different suppliers, and (3) a supplier sells four C9ORF72 antibodies without mentioning their origin and without providing validation figures on their datasheets, leaving us unable to evaluate if they are core antibodies. Of 36 C9ORF72 entries, there may be only 16–20 distinct antibodies.

A bioinformatic initiative has been created to assign a research resource identifier (RRID) to all commercial antibodies, with a focus on bringing together identical antibodies with the same RRID [21,22]. After performing the antibody characterization for C9ORF72, we realized that HPA023873 and ab121779 were the same antibody when the antibody datasheet and our data were compared, and both antibodies have now been assigned with the same RRID. For more information about RRIDs and their advantages we refer readers to the following papers [4,21–23]. These issues are also being eased by the collaboration with the antibody manufacturers.

Open science engenders trust

Antibody characterization data for all antibodies tested needs to be shared openly, free from influence from the antibody provider. YCharOS follows a standard process from the time a protein target is nominated until the antibody characterization report is placed into the public domain at Zenodo, a general purpose open access data repository operated by the European Organization for Nuclear Research, CERN (see https://zenodo.org/communities/ycharos/ for examples of antibody characterization reports). The antibody reports include detailed and partner company-endorsed standard operating procedures (SOPs) for each of the antibody-based applications. YCharOS’ reports related to AD targets are already indexed to the NIH Agora platform (https://adknowledgeportal.synapse.org/). YCharOS’ industry partners have started to upload YCharOS characterization figures on their marketing materials. Two search engines, CiteAb and The Antibody Registry, have agreed to link YCharOS antibody characterization reports published on Zenodo to their websites. Moreover, YCharOS may approach other organisations (Antibodypedia, Benchsci, NeuromAbs and the Human Protein Atlas, and scientific journals) to reach the broadest audience. Information about high quality antibodies will allow any academic scientist to select the most appropriate antibodies for their various applications and if highly-cited antibodies prove inadequate, the community will be alerted. Openly shared, and widely agreed protocols will allow any company or individual engaging an independent entity to appreciate how their products were tested. Antibody users will benefit from these shared protocols in their own experimental procedures, as subtle modifications in methodology can strongly influence antibody performance, e.g. the choice of the blocking reagent used during IB can lead to significantly different background levels of signal.

Can a KO-based form of antibody validation be taken to a proteome-wide scale?

With properly characterized and renewable commercial antibodies for most human proteins, money would be saved and reproducibility increased [24], and many antibody manufacturers, including several of those that support YCharOS, are making broad efforts to use KO cells in antibody validation [25]. Are the reagents available, and if so, how much would it cost to characterize commercial antibodies for the human proteome? The Italian physicist Enrico Fermi excelled in estimating the results of complicated calculations with only limited information. A Fermi Problem is a complex problem solved through a series of estimates. He famously calculated the energy of the first atomic bomb by pacing off the distance travelled by torn pieces of paper he threw in the air during the blast. We use this logic to estimate the cost of a proteome-wide commercial antibody characterization effort.

Are there already antibodies for all members of the human proteome?

With 2.5–4.8 million commercial antibodies, there is a high likelihood that many of the protein products of the ~20,000 human genes are covered by one or more antibodies (without considering splice variants, post-translational modifications, or conformational variants). However, some gene products are overrepresented. For example, 5166 commercial antibodies target only three proteins, actin, caspase-3 and GAPDH (Antibody Registry). To approximate the number of commercially available and unique antibodies for each human protein, we used the Antibody Registry Database, which describes millions of antibodies along with an RRID. We matched the RRIDs to a list of 16,280 human proteins with evidence of expression at the protein level (Uniprot) and found that only ~1 % have no antibodies whereas ~97 % are covered by more than 5 commercial antibodies (Fig. 2). This suggests that multiple commercial antibodies already exist for a high percentage of all human gene products.

Fig. 2.

Percentage of human proteins covered by commercial antibodies.

The Antibody Registry database was matched to a list of 16,280 human proteins with evidence of expression at the protein level (from Uniprot). Data were organized as a pie chart representing the percentage of human protein covered by 0 (light green), 1–5 (light blue), 6–20 (yellow), 21–50 (dark green) and more than 51 (dark blue) commercial antibodies.

What might be the costs for a proteome wide effort?

To estimate the costs of an effort to characterize antibodies against the entire human proteome, we made multiple assumptions, summarized in Table 2. We do not suggest that this costing exercise is specifically for YCharOS or any other single entity. We also do not claim that this is a precise costing document, it is simply a Fermi style question to think about costs.

Table 2.

Cost estimate to characterize antibodies for the entire human proteome.

		staff number
Application	Staff salary	432 targets/ year	4320 targets/year	total cost for 4320 targets/year
IB	USD 44,037.00	3	30	USD 1,321,110.00	yearly cost
IP	USD 44,037.00	3	30	USD 1,321,110.00
IF	USD 44,037.00	2	20	USD 880,740.00
cell culture	USD 44,037.00	3	30	USD 1,321,110.00
lab manager	USD 44,037.00	2	20	USD 880,740.00
total staff number at the bench		13	130
secretary	USD 44,037.00	1	10	USD 440,370.00
senior	USD 61,873.00	1	10	USD 618,730.00
data analysis specialists	USD 61,873.00	2	20	USD 1,237,460.00
chief executive officer	USD 109,046.00	1	1	USD 109,046.00
total staff number and salary		18	171	USD 8,130,416.00
consumable	for 130 people at bench			USD 4,200,000.00
overhead	represent 40 % of total annual cost			USD 4,932,166.40
total above cost for one year				USD 17,262,582.40
total above cost for 5 years				USD 86,312,912.00	cost for 5 years
Purchase KO cells	USD 1000/protein KO * 20,000 proteins			USD 20,000,000.00
Purchase antibodies	15 antibodies/protein * USD 231.5 * 20,000 proteins			USD 69,450,000.00
Purchase lab equipment for 130 people at the bench	high content confocal imaging system			USD 1,960,722.05
	lab equipement			USD 1,597,579.19
	two IB detection system			USD 127,900.00
	Total for 5 years for 20,000 proteins	Scenario A		USD 179,449,113.24
		Scenario B	Scenario A minus the antibody cost	USD 109,999,113.24
		Scenario C	Scenario B minus the revenue of $400/antibody	USD 4,999,113.24
		Scenario D	Scenario C minus quarter the cost of KO clones	-USD 886.76

We estimated the cost related to salaries, consumables, equipment and overhead required to characterize antibodies for every human protein.

Assumptions.

1. There are ~20,000 human gene products. Per protein, up to 15 most promising, and ideally renewable, commercial antibodies would be screened.

2. KO cells could be purchased from a specialized CRISPR genome engineering company at an initial cost of $1000/clone, and less as time goes on.

3. The project would require technical staff since it is not yet automated. Salaries will be roughly; Research Assistants and Administrative Assistants ($44,037), Senior Scientists, data analysis specialists, and management personnel ($61,873), a CEO for the charitable company ($109,046).

4. The number of Research Assistants required to characterize commercial antibodies for nine protein targets every week, and thus 432 targets per year is as follows:

4a. Three research assistants can screen commercial antibodies by IB.

4b. Three research assistants can screen commercial antibodies by IP using an automated system already available on the market.

4c. Two research assistants could screen the commercial antibodies for IF using a high-content confocal system.

4d. Three research assistants would be sufficient to maintain both parental and KO cells as well as to prepare and plate cells required for IB/IP/IF.

5. Two lab managers would assist these eleven research assistants, with one administrative assistant for human resources, as well as one senior supervisor and two data analysis specialists for monitoring quality control.

6. The total staff required to characterize commercial antibodies for 432 targets per year would thus be eighteen. Scaling the effort linearly would mean that 171 people could characterize ~4320 targets per year; in less than five years YCharOS could characterize antibodies for the whole human proteome.

7. The consumable costs were estimated based on our experience. With an annual cost for thirteen people at ~$420,000 per year, for 130 people working at the bench, $4,200,000 would be necessary.

8. The costs of facilities and other administration would be ~40 % of the science budget, and the equipment ~10 % of the cost.

9. The cost of a single commercial antibody is calculated by averaging the cost of all available C9ORF72 antibodies ($231.50).

10. The lab equipment required for 130 people working at the bench is listed in Table S3.

The sum of all these expenses is ~$179 M (scenario A). If participating commercial antibody suppliers agree to provide their antibodies without charge, as has been the experience of YCharOS, this would lead to a total cost of ~$110 M (scenario B). If an experimental hub charged $400.00 per antibody to be tested, this would provide ~$120 M revenues and the income and expense would be balanced (scenario C). Thus, we believe the economic analysis reveals that this initiative is financially viable, and actually represents a small fraction of the research money consumed every year in purchasing and testing poor antibodies. Moreover, an effort to characterize existing antibodies leverages the billions of dollars already spent to create them.

We stress that it is not possible for YCharOS or any other individual entity to perform such a large scale project on its own. The principles outlined can provide a framework for many independent entities to share these large-scale goals.

Conclusions and implications

There is a massive economic cost to the use of poorly characterized antibodies, with even greater consequence for confidence in science. One or many independent antibody characterization entities, following the principles outlined in this opinion piece could alleviate these problems representing a sustainable public good.

Abbreviations:

AD

Alzheimer’s disease

ALS

amyotrophic lateral sclerosis

IB

immunoblot

IHC

immunohistochemistry

IF

immunofluorescence

IP

immunoprecipitation

KD

knockdown

KO

knockout

OEM

original equipment manufacturer

RRID

research resource identifier

YCharOS

Antibody Characterization through Open Science.