Why Don’t More Researchers Validate Antibodies with WB Before Doing IHC?

As we have discussed previously in this blog, there is widespread concern about antibody validation. Of course some forms of validation using multiple antibodies or knockouts are not available for the majority of targets of interest to researchers.  But western blots are a excellent form of validation that can be readily performed by almost any lab.  Thus it is puzzling why so many immunostaining papers fail to use this straightforward antibody validation tool. I would suggest that the failure to use western blots (WBs) as validation tool before immunostaining most likely results from a widespread misunderstanding in the research community about the relevance of WB data to immunostaining. Some have suggested that the detection environment in WB with the denaturing effects of SDS is so different from that in immunostaining as to make WB data irrelevant to immunostaining. However, as emphasized by Forsström et al. (1) it is important to understand that both WB and IHC use denaturing conditions. While exceptions exist (2-4), a large body of evidence points to a high correlation between positive WB data for an antibody and good data for the same antibody in IHC, ICC and IF. (5-8). As argued by Kurien et al (8) “immunoblotting is a must to determine specificity of antibodies used for Immunohistochemirty (IHC).” The Journal of Endocrinology and the Journal of Histochemistry and Cytochemistry both editorialize in favor of using WB as an initial antibody screen (9;10). Both argue that any antibody that yields multiple bands in WBs raises a critical red flag and that the antibody should not be used in IHC unless some other test can be used to validate the antibody.

After making over 500 antibodies over the past few decades, we have found that more than 90% of the antibodies that gave a single band signal in WB also gave a good signal immunostaining. See for example Fig.1 where WB and IHC staining of an antibody to synapsin I, a neuron specific synaptic vesicle associated protein, is shown.   As shown clearly in the figure, the synapsin antibody specifically labels only the synapsin I doublet in the WB.  Similarly the IF image shows the same synapsin antibody exhibiting specific punctuate labeling characteristic of the localization of the synaptic vesicle associated protein 

So in summary, antibody validation by WB is certainly not perfect.  However, it is important to realize that WBs provide very important validation tool particularly given the fact that no other validation method is available for most targets. The ideal antibody validation tool is of course the use of knock out animals in the immunological methods of interest (however see (11;12) for some limitations on the use of knockouts in antibody validation).  Thus when knockouts are available they should almost always be used in preference over WB. Unfortunately, knockouts are available for only a very small percentage of the protein targets of interest.  Consequently, it seems illogical to let the fact that WB validation is not a perfect validation tool to limit its use as a very good antibody validation tool. This is particularly true at a time when the results of many antibody studies being published use antibodies with little or no validation, leading to data that is flawed and cannot be reproduced.  Having said that, if WBs are going to be used as a validation tool it is essential that best practices be utilized in the WB assay.  In the next few days we will post suggestions for these best practices and how to avoid pitfalls in using WB for antibody validation.

Reference List

      1.   Forsstrom B, Axnas BB, Rockberg J, Danielsson H, Bohlin A, Uhlen M. Dissecting antibodies with regards to linear and conformational epitopes. PLoS.One. 2015;10:e0121673.

      2.   Herrera M, Sparks MA, Alfonso-Pecchio AR, Harrison-Bernard LM, Coffman TM. Lack of specificity of commercial antibodies leads to misidentification of angiotensin type 1 receptor protein. Hypertension 2013;61:253-8.

      3.   Yu W, Hill WG. Lack of specificity shown by P2Y6 receptor antibodies. Naunyn Schmiedebergs Arch.Pharmacol. 2013;386:885-91.

      4.   Baek JH, Darlington CL, Smith PF, Ashton JC. Antibody testing for brain immunohistochemistry: brain immunolabeling for the cannabinoid CB(2) receptor. J.Neurosci.Methods 2013;216:87-95.

      5.   Schuster C, Malinowsky K, Liebmann S et al. Antibody validation by combining immunohistochemistry and protein extraction from formalin-fixed paraffin-embedded tissues. Histopathology 2012;60:E37-E50.

      6.   Egelhofer TA, Minoda A, Klugman S et al. An assessment of histone-modification antibody quality. Nat.Struct.Mol.Biol. 2011;18:91-3.

      7.   Sawicka M, Pawlikowski J, Wilson S et al. The specificity and patterns of staining in human cells and tissues of p16INK4a antibodies demonstrate variant antigen binding. PLoS.One. 2013;8:e53313.

      8.   Kurien BT, Dorri Y, Dillon S, Dsouza A, Scofield RH. An overview of Western blotting for determining antibody specificities for immunohistochemistry. Methods Mol.Biol. 2011;717:55-67.

      9.   Saper CB. A guide to the perplexed on the specificity of antibodies. J.Histochem.Cytochem. 2009;57:1-5.

   10.   Gore AC. Editorial: antibody validation requirements for articles published in endocrinology. E 2013;154:579-80.

   11.   Lorincz A, Nusser Z. Specificity of immunoreactions: the importance of testing specificity in each method. J.Neurosci. 2008;28:9083-6.

   12.   Watanabe M, Fukaya M, Sakimura K, Manabe T, Mishina M, Inoue Y. Selective scarcity of NMDA receptor channel subunits in the stratum lucidum (mossy fibre-recipient layer) of the mouse hippocampal CA3 subfield. Eur.J.Neurosci. 1998;10:478-87.

The Antibody Two-Step Solution

In the past year there have been a number of articles in Nature and other major journals that discuss the antibody problem in somewhat apocalyptic terms.  In my mind there are only two main issues in the antibody problem: antibody validation and antibody variability.  Both of these issues have straightforward solutions that do not require any massive influx of cash or massive restructuring of antibody production. 

            Antibody validation is the hardest nut to crack and causes the most confusion.  There is no consensus on what constitutes suitable validation and this is complicated by the different methods for antibody use.   However, antibody validation is a process like all science where knowledge increases as more and more work is done with the antibody.  As long as the data are clear and the methods used adequately described, progress in validation will occur with time.  Nature’s insistence in its instructions to authors that antibody validation data be provided is exactly the right path to be taken. However, none of this progress will matter unless we deal with the antibody variability problem.  

There are two main reasons for the variability in an antibody’s performance.  The first is that once an antibody is found to have a high demand, many different antibody manufacturers will try to make their own version of antibody so they can sell it. But all these new antibodies will differ in unknown and unpredictable ways from the original antibody.  Thus validation done on the original antibody may or not be true for the new antibodies.  One way to deal with this problem was recently suggested by Andrew Chalmers and his colleagues http://f1000research.com/articles/2-153/v2 .  They argue that all publications using commercial antibodies should all report the name of the supplier and the catalog number of the antibody used.  That way even if a supplier sells many varieties of the antibody a researcher will be able to order the same antibody that was used in the publication.  This suggestion is being incorporated into the instructions to authors in more and more journals.

 Even though this action would greatly improve the value of antibody validation, an additional source of antibody variability would remain.  This variability occurs because even if one buys the same antibody with the same catalog number, one still often encounters large variability in different lots of the same antibody obtained from different bleeds of the same animal. There is a very straightforward fix to this type of variability. The solution is to pool all the serum collected from the animals. Virtually all lot–to-lot variability can be eliminated for polyclonal antibodies if this procedure is used. It will no longer be necessary to reinvent the antibody validation wheel each time an antibody is used. Thus science can build upon itself as it is supposed to do.

Some may argue that one should use monoclonal antibodies to eliminate variability. This is unnecessary and also unwise.  It is unnecessary because for most antibodies a single rabbit can produce a 20-30 year supply of antibody.  Only small percentage of all antibodies sold ever sell more than can be produced by a single rabbit.  It is unwise because monoclonals cost at least 3X what polyclonals cost and we are unlikely to see a time in the near future when cost will be irrelevant.  Only antibodies with a known, large market are likely to justify the monoclonal cost and maintenance expense.

Scientists and journals can fix the validation problem if antibody suppliers will fix the variability problem. We call this the Antibody Two-Step Solution. 

Why Recombinant Proteins Make Poor Antibody Validation Tools

Antibody validation is a topic that has garnered a great deal of attention lately in discussions of the problem with the lack of reproducibility in science.  One type of antibody validation that should be avoided uses purified recombinant proteins.  In a typical study of this type a purified recombinant protein is run on a western blot and then the labeling of the new antibody is examined.  Given that only a single protein is run on the blot, this “validation” study offers very little information about the specificity or sensitivity of the antibody.  The blot obviously has no information about whether the antibody recognizes other proteins since no other protein is present in the blot.  Moreover, absent any information about the relevance of the amount of recombinant protein used compared to the amount of endogenous protein in situ, the experiment does not even validate the ability of the antibody to bind to the protein of interest. 

Thus a western blot with recombinant protein does little to answer the two key questions about antibody quality: 1. ) Does the antibody possess the sensitivity to recognize the antigen in situ in a tissue of interest? and 2. ) Is the antibody binding specific for the antigen of interest in situ?  Nevertheless it is not unusual to see such data used to validate an antibody in product data sheets or even in refereed publications. So it is important to look carefully at any antibody validation blots to be sure that a cell lysate and not simply a purified protein is being analyzed.

In the example above I described a situation in which a recombinant protein based assay was used to give a false positive validation.  It is also not uncommon to see a recombinant protein based assay provide a false negative result i.e. to falsely invalidate an antibody.  Such experiments are most commonly seen with phosphospecific antibodies.  Such antibodies can be extremely valuable tools as they permit one to evaluate the phosphorylation state of a single phosphorylation site on a specific protein.  A critical question in the validation of such antibodies is whether they are indeed phosphospecific.  A mutant recombinant protein with the phosphorylation site of interested mutated to a non-phosphorylatable amino acid is run along side the recombinant protein in a western blot.  The antibody on interest is then tested for binding to this assay.  Binding to the non-phosphorylatable mutant in such an assay has been used by some as evidence that the antibody’s phosphospecificity has been invalidated.  Such data do not provide such evidence.  A phospho-specific antibody will always have at least a finite affinity for the non phospho-site.  Thus when, as is typical in such studies, micrograms or hundreds of nanograms of the mutant protein are run on the blot, some binding is highly likely.  In order to use such an assay for validation it is necessary to do a very detailed dose response with multiple concentrations of both the mutant non-phospho and the phosphoprotein.  Attention must also be paid to the concentration of the protein of interest in situ and also its level of phosphorylation at the site of interest.  Determining these values is always quite problematic.  Consequently the use of such a validation technique in not recommended particularly when other much more relevant validation assays are available.  The most common such assay is the western blot performed on lysates of the tissue of interest that had been incubated in the absence or presence of a phosphatase.  In such an assay the antibody is validated if it labels a single band in the control lysate and if the labeling is absent in the lysate that had been incubated in the presence of the phosphatase. 

One issue raised in the preceding paragraph was the fact that there is always a finite affinity of a phosphospecific antibody to the non-phosphoprotein.  Such binding of the phospho-antibody to the non-phosphosite can be quite problematic in negative affinity column selection of the phospho-antibody.  In our experience we have often found that very good phosphospecific antibodies may sometime fail to flow through a non-phospho column.  To avoid failures in such negative selection experiments it is very important to optimize the antibody to peptide ratio in using such a column.   This issue will be discussed in more detail in a subsequent blog.

Antibody Variability

In a recent article in Nature Methods, Vivien Marx discussed some of the key issues relating to the use of antibodies http://tinyurl.com/osygswn . Variability in the quality of the antibodies available commercially was one of the principal problems encountered by scientists in her article.   This problem is closely related to the larger issue addressed in numerous editorials in Nature and other journals regarding the irreproducibility of research. 

There are a number of reasons for the variability in a commercial antibody's performance.  The first is that once an antibody is found to have a high demand, then many different antibody manufacturers will try to make the antibody so they can sell it. As an example take the antibody our colleague and co-founder John Haycock made to tyrosine hydroxylase (TH).  This is our best selling antibody and we have been selling it for almost 25 years. When John first started selling this antibody it was virtually the only TH antibody on the market.  However today if one examines any large antibody company’s offerings one can find hundreds of sources of this TH antibody.  So this is one huge contributor to antibody variability; namely that many different groups have made many different TH antibodies. So simply saying you used a TH antibody can mean you used any one of a huge variety of different antibodies. One way to deal with this problem was recently suggested by Andrew Chalmers and his colleagues http://f1000research.com/articles/2-153/v2 .  They argue that all publications using commercial antibodies should all report the name of the supplier and the catalog number of the antibody used.  That way even if a supplier sells many varieties of the antibody a researcher will be able to order the same antibody that was used in the publication. 

However, as Ms. Marx points out in her article, even if one buys the same antibody with the same catalog number one still often encounters large variability in different lots of the same antibody.  She quotes some antibody sellers who argue that this variability is unavoidable and simply the nature of the beast given that antibody quality can often vary across different bleeds from the same animal(s).  An example of how the antibody signal strength and specificity can vary across bleeds from the same rabbit is shown in Figure 1.

Figure 1. Western blot of serum from bleeds 1-10 collected from a single rabbit.  The level of immunoreactivity for the 210 kD band (see arrowhead) varies dramatically across the various bleeds.  Moreover, the level of cross reactivity with bands of ~100 and ~50 kD also varies across bleeds albeit in a very different pattern than that for the 210 kD protein.

We would argue that there is a very straightforward fix to this problem of variability.  All it requires is some forethought and planning before an antibody is ever released.  The solution is to pool all the serum collected from the animals before the first antibody is ever released. Then all serum sold subsequently for that antibody catalog number will by definition be identical as it all comes from the same pool.  Obviously in the serum shown in the figure, the antibody will have to be affinity purified.  In such a case, all purifications will be performed identically with identical aliquots from the same pooled starting material.  Thus virtually all lot to lot variability can be eliminated for polyclonal antibodies if this procedure is used.  It should also go without saying that once the pool of antibody has been used up then that catalog number must be discontinued and not simply continued with antibodies from another animal.

Now of course the next issue is whether or not the antibody with a given catalog number is actually specific for the target of interest.  This is a very interesting area and will be the topic for some of my next posts.

Antibodies That DO NOT WORK in Western Blots

What does it mean when an antibody company says an antibody “does not work” in western blots?

I started wondering about this question the other day after a colleague told me about a company that “validates” all of their antibodies by immunostaining in no less than 5 different tissues!  We examined the company’s website and in some cases western blots (WB) showing labeling of a single band were also presented.   But for the vast majority of products on the site only the apparently innocuous phrase “does not work in WB” was seen.  What does this mean?  I think for many antibody users (especially people who do not do WB and use various immunostaining protocols like IF or IHC) this failure to “work” in WB is often interpreted as meaning something akin to “don’t worry about the WB data, see if the antibody ‘works’ in immunostaining.”  However, I think any user of an antibody that “does not work in WB” should be very worried indeed.  This is because in the overwhelming majority of cases “does not work in WB” means the antibody labels many different proteins in a WB.  It is extremely rare to get an antibody that labels nothing in a WB. 

 Examples of antibodies that do and do not work in WB are shown in Fig. 1.  The lane at the right shows staining of a brain lysate with an antibody raised against tyrosine hydroxylase (THRAB) and it shows labeling of a single band of Mr 60,000 which is the approximate Mr of TH.  The lane at the left shows staining of an antibody that does not work in WB.  The antibody shown was raised against a protein Sap 102.  As can be seen in the image of the WB, the antibody does recognize a band at the appropriate Mr for Sap 102, but the antibody also recognizes many other bands as well.  This antibody ”did not work in WB” and it was thrown in the trash where it belonged.

So remember when you read the words “does not work in WB” you should translate that as “this antibody probably cross reacts with a number of different proteins.”

Are Monoclonal Antibodies Really More Specific?

It is common to see references to the “specificity” or “greater specificity” of monoclonal antibodies compared to polyclonal antibodies.  Is that claim justified?  No one would dispute the fact that a monoclonal is directed toward a single site or epitope of the target protein, while a polyclonal may contain antibodies directed toward multiple epitopes on the protein.  However, in my opinion, that fact alone does not make monoclonal antibodies more specific.  Indeed the single epitope toward which a monoclonal is directed may in fact be shared by many different proteins in addition to the protein of interest.   Such a monoclonal would not be specific even though it recognizes only a single epitope.  In contrast, a polyclonal antibody raised against the same protein may contain antibodies directed toward that same non-specific epitope as the monoclonal, as well as other epitopes that are more specific.  In such a situation the serum of the polyclonal would at least contain some antibodies that are specific and thus it would be “more specific” than the monoclonal.  Moreover, it may be possible using affinity purification to isolate the specific antibodies. This may seem to some as a trivial issue but it can be extremely important in IHC where it is quite difficult to control for cross reactivity.  Thus simply opting for a monoclonal is no guarantee of specificity and one must still utilize a full range of specificity controls. An example of such a non-specific monoclonal is shown in the figure 1.  Note that the monoclonal antibody recognizes the NET protein at ~50 kD but it also recognizes proteins at 75 and 95 kD.

Figure 1. The monoclonal raised against the NET protein labels three 

prominent bands at 50, 75 and 95 kD in a lysate of rat cortex.

One additional issue relating to this question is the difference in the purification/selection of monoclonal and polyclonal antibodies.  The process of purification and selection of monoclonal antibodies rarely involves screening for specificity.  Thus the hybridoma screens of hundreds or thousands of clones typically rely solely on the ability to recognize a target in a plate assay.  There is typically no selection for affinity for the target and or specificity.  This contrasts with polyclonal antibody affinity purification which, as its name suggests, can preferentially yield higher affinity and higher specificity antibodies. 

Taken together these comments are not at all meant to minimize the very real importance of monoclonal antibodies in clinical environments.  Rather these comments are meant to underscore the importance of specificity controls in using all antibodies and to show that, in some cases, polyclonal antibodies may be more specific than monoclonal antibodies.

Antibody specificity:  The use of a blocking control has only limited value

Antibody specificity is one of the key issues in determining whether you have an antibody that works.  How does one determine that the antibody specifically recognizes only the target of interest?  There are a number of control procedures one can use to be sure that the signal generated in the antibody based assay truly and quantitatively represents the presence of the target of interest. 

In western blots one can at least partially address this issue by determining that the relative molecular weight of the antibody signal matches that of the target. However in most other antibody based imaging assays (e.g. IHC and IF) no such information is available and thus determining specificity in such assays is even more critical.  One of the most common controls for antibody specificity utilizes the antigen that was used to make the antibody as a blocking control.

 Unfortunately the value of this control is often greatly overestimated.  For example take a case where an antibody raised against a protein antigen recognizes only a single epitope in the protein.  Assume for example that this antibody is non-specific and its epitope is also found in a number of other proteins. The antibody will thus recognize its epitope in all of those other proteins as well as in the target protein and thus in IHC it may give a very strong signal as it is detecting many proteins in the tissue.  When one adds the immunizing antigen (which contains the epitope) to the antibody labeling assay, the antigen blocks the antibody labeling of all the proteins which contain the epitope.  Thus it gives a complete block of all IHC signal.  Normally that is interpreted as indicating that the antibody is specific.   Clearly in this hypothetical case the blocking control failed because in fact the antibody was NOT specific.  

This effect can be seen in the Figure at right.  In this western blot as shown in lane 1, an antibody raised against synaptotagmin labels three unknown protein bands in addition to the 60k band representing synaptotagmin.  When the blocking control is used (lane 2) the labeling of the specific 60k band and all three non-specific bands is blocked.  So the blocking control eliminated all of the antibody signal but the antibody was clearly not specific for synaptotagmin.

 Thus anytime an antibody is non-specific and recognizes an epitope that is present in more than one target, the antigen blocking control is virtually useless. Since this type of cross reactivity or non-specificity is the one of the most troublesome types of antibody non-specificity, I would argue that antigen block is only one control to be used and that it is a relatively weak control for antibody specificity.

            One of the best controls for antibody specificity is recombinant tissue that has been engineered to lack the target antigen.  When using such tissue one should see no antibody signal in contrast to wild type tissue.  Phosphatase treated tissue is another one of the best controls is to use when testing phospho-specific antibodies.  Provided that the phosphatase can dephosphorylate the target, the signal from a phosphospecific antibody should be eliminated from the phosphatase treated tissue with no change in the total amount of the target protein compared to untreated tissue.