Immunohistochemistry is a sensitive and versatile method widely used to investigate the cyto- and chemoarchitecture of the brain. It is based on the high affinity and selectivity of antibodies for a single epitope. However, it is now recognized that the specificity of antibodies needs to be tested in control experiments to avoid false-positive results due to non-specific binding to tissue components or recognition of epitopes shared by several molecules. This ‘Technical Spotlight’ discusses other pitfalls, which are often overlooked, although they can strongly influence the outcome of immunohistochemical experiments. It also recapitulates the minimal set of information that should be provided in scientific publications to allow proper evaluation and replication of immunohistochemical experiments. In particular, tissue fixation and processing can have a strong impact on antigenicity by producing conformational changes to the epitopes, limiting their accessibility (epitope masking) or generating high non-specific background. These effects are illustrated for an immunoperoxidase staining experiment with three antibodies differing in susceptibility to fixation, using tissue from mice processed under identical conditions, except for slight variations in tissue fixation. In these examples, specific immunostaining can be abolished depending on fixation strength, or detected only after prolonged postfixation. As a consequence, antibody characterization in immunohistochemistry should include their susceptibility towards fixation and determination of the optimal conditions for their use.
Immunohistochemistry (and immunocytochemistry) are versatile methods for the detection of specific molecules, mainly proteins, in tissue preparations and in isolated cells. While applicable to all organisms in various biomedical disciplines, these methods are widely used in neuroscience to unravel the complex cyto- and chemoarchitecture of the nervous system. They allow not only to determine whether a selected molecule is present or not in a given brain region (or cell type), but also provide precise information about its subcellular (including ultrastructural) localization. The publication of erroneous conclusions due to overlooked pitfalls of immunohistochemistry (used here as a generic term for both methods) is a major concern in view of the widespread use and great utility of this method in neuroscience. Steps have been taken to define minimal quality standards and safeguards minimizing the occurrence of false-positive findings due to non-specific antibody binding to tissue components (Saper, 2005; Rhodes & Trimmer, 2006). Just as relevant, but largely ignored, are false-positive findings due to improper localization of the antigen, as well as false-negative findings due to failure to detect an antigen for methodological reasons (e.g. fixation strength or tissue processing; Lorincz & Nusser, 2008). Furthermore, the premise of every scientific publication is to provide sufficient information in the method section to allow others to reproduce the results. Yet, with immunohistochemistry, there is little agreement about which information is necessary for this purpose. Scant description of the procedures used potentially precludes not only reproduction of experiments, but also interpretation of results.
The purpose of this review is to highlight major pitfalls that can affect the outcome of an immunohistochemical reaction and to recapitulate the minimal set of information required in a publication to allow proper evaluation and replication by others.
Limitations of immunohistochemistry
Immunohistochemistry is a robust, versatile and sensitive method. It is, however, largely empirical and requires a good understanding of its specific limitations to be applied successfully:
1 Immunohistochemistry does not ‘show’ the target of interest, but provides, at best, indirect evidence for its presence in the tissue or cell examined in the microscope. Epifluorescence microscopy detects the light signal emitted by a fluorochrome bound to the primary antibody (direct immunofluorescence) or a fluorochrome bound to a secondary antibody directed against the primary antibody (indirect immunofluorescence). When an enzymatic reaction, such as immunoperoxidase, is used, one merely sees a change in tissue coloration reflecting deposition of a substrate around the site bound by the enzyme catalysing the reaction. Interpretation of such staining patterns therefore relies on the assumption that primary and secondary antibodies are selectively bound to their targets.
2 Lack of staining for a molecule of interest cannot be interpreted as absence of the molecule; at best, one can conclude at absence of immunoreactivity, which can have multiple causes, including insufficient affinity of the antibody, suboptimal tissue processing or immunohistochemical detection procedure, or epitope masking, to name just a few. Setting the threshold between background and weak specific staining is arbitrary, so that one can never formally conclude that the molecule of interest is absent in the tissue investigated.
3 Ideally, a tissue section (or cell preparation) should remain unstained after immunohistochemical processing if it is devoid of the target antigen recognized by the antibody used. In practice, this is generally not the case; immunoglobulins (IgGs) bind with low affinity to numerous (mostly unidentified) tissue constituents. Consequently, antibodies known to produce a specific staining when their epitope is present can give rise to non-specific signals in tissues devoid of their target, such as a section from a knockout mouse (see Appendix).
4 The pitfalls of immunohistochemistry are potentially so numerous that it can become impossible to determine with certainty whether an antibody produces a specific staining pattern. Correct interpretation of immunohistochemical experiments relies on additional information about the expression and localization of the molecule of interest (e.g. in situ hybridization, biochemical isolation). This limitation applies particularly to ubiquitous molecules, such as signalling molecules, present at low concentrations in all cell types. Furthermore, interpretation can be biased by partial knowledge: if one detects staining of cell nuclei with an antibody against a protein known to be membrane associated, does it mean that the staining was non-specific or that this protein can be translocated to the cell nucleus? Immunohistochemistry alone cannot provide a definitive answer to this question; translocation would need to be demonstrated by other means to allow a proper conclusion.
What should be reported in an original publication?
There is no standard immunohistochemical procedure giving optimal results for chemically distinct molecules (see Box 1). The outcome of an immunohistochemical reaction depends on the properties of the antibody used, as well as on several physico-chemical parameters of the tissue to be processed, notably its fixation. To allow replication and assessment of the staining pattern, key parameters determinant for the outcome of the immunohistochemical experiments should be described in the method section of research reports:
• Animal species, strain or genotype, sex, age and supplier.
• Type of fixative, fixation procedure, including duration of postfixation.
• Tissue-processing procedure (including timing and temperature of reagents); if applicable, description of antigen-retrieval procedures.
• Precise and complete description of the primary antibody (host species, nature of the antigen and its preparation), provenance and characterization of its specificity (explicit description or citation of relevant manufacturer information and references to published data).
• Staining procedure: concentration of primary and secondary antibodies (in the case of affinity-purified antibodies, the absolute concentration should be given); reagents used for reducing non-specific binding (bovine serum albumin, normal serum, etc.); duration and temperature of incubation; detection procedure used.
• Description of the control experiments performed to test the specificity of the staining pattern observed. If available, tissue devoid of antigen should be tested under identical conditions, or another primary antibody directed against a different epitope should be used. Controls by omission of primary antibodies, replacement with non-immune immunoglobulins or preadsorption with antigens provide no definitive information about the specificity of a primary antibody reaction.
Factors influencing the quality of immunohistochemical staining
Immunochemical detection of a selected molecule relies on the high affinity and selectivity of antibodies (mainly IgG and IgM) for a single epitope formed by this molecule. In principle, immunohistochemistry can therefore be used to detect with great sensitivity a single molecular entity (‘target’), or even single molecules, in highly complex and heterogeneous structures. However, antibodies interact in often unpredictable ways with various tissue and cell constituents, notably after the chemical fixation that is necessary to preserve tissue structure during histological preparation. Major factors influencing binding of antibodies to their target have been reviewed recently (Lorincz & Nusser, 2008). These factors include the method of cell/tissue fixation (type of fixative, fixation procedure, postfixation), tissue processing (cryoprotection, freezing, embedding, sectioning) and detection method (immunofluorescence, immunogold, enzymatic reaction, such as peroxidase staining), as well as quality of the tissue to be processed. Because fixation changes the chemical properties of tissue constituents and alters three-dimensional protein conformation by cross-linking, it has a major impact on affinity and selectivity of antibodies. Epitope masking can also occur when fixation alters penetration of antibodies into the tissue. The synaptic cleft is a typical example of a largely inaccessible structure in fixed tissue. Ideally, a compromise between histological preservation and retention of antigenicity should be sought. However, this compromise is largely empirical and optimal conditions have to be determined for each molecule of interest (see Appendix).
Additional confounding factors can arise when the fixation is achieved through vascular perfusion, because physiological variables can affect vascular flow and therefore the speed and strength of fixation. A poor perfusion leads to variable fixation that can regionally affect the intensity of staining within tissue sections. The physiological status of the tissue at the time of fixation also influences the quality of immunohistochemical reactions. Ischaemia can cause a redistribution of the antigen of interest, for example upon presynaptic release of a neurotransmitter, whereas mechanical or chemical lesions greatly enhance the risk of non-specific binding of IgGs to histological sections, resulting in decreased signal-to-noise or even complete masking of specific signals after immunohistochemical processing. Finally, tissue processing can also have adverse effects on antigenicity, for example due to heating, dehydration, etc., but it can also be beneficial, as shown with antigen-retrieval methods, which aim to reverse some of the deleterious effects of fixation and embedding (Fritschy et al., 1998).
Evaluation of immunohistochemical staining patterns
To assess the specificity of an antibody used for immunohistochemistry, controls for false-positive staining are an absolute necessity. There are several possible sources of false-positive signals, which require independent verification:
1 Does the tissue contain fluorescent molecules (e.g. autofluorescent pigments) or enzymes (e.g. peroxidases) that might confound the detection system used? Control reactions without secondary antibodies provide a direct test for this possibility.
2 Do the secondary antibodies alone produce any staining? The issue is particularly relevant when working with mouse monoclonal antibodies in mouse tissue sections, due to the nature of the secondary antibodies. While only negligible background staining arises from non-specific binding of anti-mouse IgGs in healthy tissue, the situation changes dramatically in lesioned tissue, because the anti-mouse IgGs used as secondary antibody strongly bind to non-specific epitopes exposed upon tissue damage. Similarly, in which anti-mouse IgGs produce extensive background staining in juvenile animals due to non-specific binding to extracellular matrix molecules.
3 Does the primary antibody label the expected structures? There is a large consensus that a control staining in sections known to be devoid of the target antigen (e.g. derived from a knockout mouse) represents the best control available, because it excludes the possibility that the antibody recognizes an epitope present in multiple molecules. However, as discussed elsewhere (Lorincz & Nusser, 2008), even this control can be misleading, for example in the case of upregulation of a homologous protein in the mutant tissue. Nevertheless, the increasing availability of knockout mice justifies that this control be done whenever possible. Another ‘golden control’ is to compare the staining pattern with that produced by another antibody raised against a distinct epitope of the same target molecule (but see Lorincz & Nusser, 2008, for possible pitfalls due to epitope masking). When these controls are not available, a rigorous comparison of the immunostaining pattern with the gene expression pattern derived from in situ hybridization experiments represents an alternative, at least for molecules with a regionally restricted expression pattern.
It should be emphasized that failure to detect a target antigen by an antibody does not necessarily indicate that the antibody is not suitable due to lack of affinity and/or specificity. Lack of staining might suggest a false-negative result due to loss of antigenicity upon tissue fixation or tissue processing having a deleterious effect on antigen preservation or accessibility (see Appendix for two such examples). In this case, multiple tests should be performed aiming at improving the immunoreaction.
Quantification of immunohistochemical staining patterns
Immunohistochemistry is the method of choice to investigate the distribution of selected molecules in tissue preparations. Its intrinsic limitations preclude, however, quantifying the absolute number of molecules or their concentration in the sample investigated. In practice, the affinity of antibodies for their target cannot be determined in histological preparations. Furthermore, as noted above, the influence of fixation on the availability and conformation of epitopes is generally not known, and often not predictable. The strength of fixation also influences penetration of antibodies into tissue sections, resulting potentially in variations in staining intensity unrelated to the density of antigens in the tissue. Even applying antibodies to living cells is potentially prone to artefacts, as antibody binding might trigger non-physiological redistribution of the target, e.g. by internalization. Finally, most detection systems are not linear and cannot be calibrated accurately in the absence of information about antibody-binding affinity, whereas fluorochromes are subject to modifications (e.g. photobleaching) during image acquisition, which are very difficult to normalize across experiments.
All these factors contribute to render immunohistochemistry largely unsuitable for quantitative analyses. Comparing staining intensities between antibodies provides no relevant information about the relative amount of the different target molecules present in the tissue. Even relative comparisons of staining intensities obtained with a given antibody across brain regions or between preparations provide rough estimates at best, and should be complemented whenever possible with quantitative measurements derived from other methods.
Immunohistochemistry is an invaluable tool for characterizing the rapidly growing number of new proteins and other molecules identified in brain tissue. However, its credibility is questionable in the absence of specific safeguards minimizing the risk of erroneous conclusions due to false-positive or false-negative results. Thus, concerted efforts from producers and users of antibodies should be encouraged to test antibodies not only for their specificity against a defined antigen, but also for their susceptibility towards fixation and for determining the optimal conditions of their use in immunohistochemistry. Here, we highlight the potential impact of tissue fixation on immunohistochemical reactions (see Appendix) to incite investigators to explore which fixation procedure and tissue processing method are most appropriate for their purpose. Such choices are largely empirical, but can be guided by experience and by the properties of the molecule to be investigated. Detailed reporting of technical procedures in original publications would contribute to disseminate this knowledge and facilitate reproduction of successful protocols. Finally, the awareness of reviewers and editors for common errors occurring with immunohistochemistry should be fostered to enhance the overall quality of published results.
The research presented here was supported by the Swiss National Science Foundation.
lacking the GABAA receptor α3 subunit
Appendix. Effect of tissue fixation on the quality of immunohistochemistry: comparison of three antibodies in three differently fixed tissues
To illustrate the principles discussed in this review, tissue from four adult male mice (three C57Bl6/J mice; RCC, Füllinsdorf, Switzerland and one GABAA receptor α3 subunit-knockout (GABRA3-null) mouse;Yee et al., 2005) was prepared after perfusion–fixation under deep pentobarbital anaesthesia (50 mL/kg, i.p.). All animal experiments were approved by the local authorities (Cantonal Veterinary Office of Zurich). Except for slight variations of the perfusion and postfixation, procedures were identical for all mice. In particular, the same fixative solution [4% paraformaldehyde in 0.15 m sodium phosphate buffer, pH 7.4, containing 15% of a saturated solution of picric acid (final concentration 0.2%); Fritschy & Mohler, 1995) was used for perfusion.
• For mouse 1 (wild-type) and mouse 2 (GABRA3-null), the perfusion cannula was inserted through the left ventricle into the ascending aorta; a constant flow (10 mL/min) of phosphate-buffered saline (10 mL) followed by ice-cold fixative (50 mL) was established with a peristaltic pump, resulting in a rapid fixation monitored by the increase in body rigidity. The brain was extracted immediately after the perfusion; it had a firm consistence and was yellow due to the picric acid in the fixative. It was postfixed in the same solution for 4 h at 4°C.
• Mouse 3 (wild-type) was perfusion-fixed exactly like mice 1 and 2, but postfixation of the extracted brain lasted for 18 h.
• In mouse 4 (wild-type), the cannula was inserted into the left ventricle only and the perfusion flow was slow and irregular (3 mL/min), resulting in a weak fixation, as judged by the rigidity of the tail and body musculature. The extracted brain was pale and soft, confirming the weak fixation. It was postfixed for 4 h, like for mice 1 and 2.
Tissue cryoprotection, sectioning and the immunoperoxidase staining procedure were performed identically for the four mice, as described (Fritschy & Mohler, 1995), using the Vectastain Elite kit (Vector Laboratories, Burlingame, CA). Primary antibodies included a guinea pig antiserum against the γ-aminobutyric acid (GABA)A receptor α3 subunit (see Studer et al., 2006 for characterization), a mouse monoclonal antibody against cholecystokinin (CCK; CURE/UCLA Digestive Disease Center University of California, Los Angeles, USA; Ohning et al., 1994) and a rabbit polyclonal antibody against serotonin (coupled to bovine serum albumin with paraformaldehyde; Immunostar, Stillwater, MN, USA; product Nr. 20080). These antibodies were selected because they are, in our experience, differentially sensitive to fixation. For controlling the specificity of the α3 subunit antiserum, mouse 2 lacked the GABAAα3 subunit (a GABRA3-null mutant, see Yee et al., 2005). Antibodies were diluted in Tris-saline buffer (pH 7.4) containing 2% normal goat serum to block non-specific binding sites and 0.2% Triton X-100. Sections were processed free-floating, mounted on glass slides, coverslipped and photographed with a digital camera. For details of the staining procedure, please refer to Fritschy & Mohler (1995) and Fritschy et al. (1998).
Figure 1A–D illustrates the dependence of GABAA receptor α3 subunit-immunoreactivity on fixation and the presence of antigen in transverse sections through the amygdala. Specific staining was observed in mice 1 and 4, with the highest signal-to-noise ratio being obtained in the weakly fixed tissue (mouse 4; Fig. 1D). In contrast, variable cytoplasmic staining was evident in the section from the GABRA3-null mouse (Fig. 1B), indicating that this antiserum binds non-specifically to cytoplasmic constituents in the absence of its target antigen. A similar pattern was evident in mouse 3, notably in the basolateral nucleus (Fig. 1C), suggesting that prolonged postfixation causes a dramatic loss of immunoreactivity in the neuropil and appearance of non-specific staining of neuronal somata.
Comparing Fig. 1B and C might lead to the premature conclusion that the α3 subunit antiserum is not suitable for immunohistochemistry, because it produces mainly background staining. Novel, or unknown, antibodies should be tested routinely with various protocols to determine their sensitivity to fixation. Finally, note that the background staining seen in the GABRA3-null mouse can be completely eliminated by antigen-retrieval in citrate buffer (Fritschy et al., 1998; Yee et al., 2005).
Figure 2A–C illustrates immunoperoxidase staining for serotonin in the CA1 area of the hippocampus. It is best preserved in rapidly perfused tissue after short postfixation (mouse 1), revealing a homogeneous plexus of fine varicose axons in the stratum oriens and radiatum, being more dense in stratum lacunosum-moleculare (Fig. 2A). Prolonged postfixation (mouse 3) resulted in a different morphology and regional distribution of immunopositive axons, which appeared coarser and considerably less dense in the stratum oriens than in stratum radiatum (Fig. 2B). These alterations were accompanied by increased background staining, notably in the pyramidal cell layer. Finally, immunodetection of serotonin was precluded in the weakly fixed tissue (mouse 4), in which the background was even more intense, notably in the stratum lacunosum-moleculare (Fig. 2C). In this case, the strong background likely reflects vesicular release of serotonin during the weak perfusion–fixation procedure, emphasizing the need of a fast fixation to avoid such artefacts.
Figure 2D–F illustrates the innervation of CA1 pyramidal cell somata by CCK-positive axon terminals, known to belong to a specific population of basket cells (Freund & Katona, 2007). Most strikingly, these axons are detectable by immunoperoxidase staining only following a prolonged postfixation (mouse 3; Fig. 2E), whereas in the regularly fixed tissue (mouse 1; Fig. 2D) no staining is evident, and in the weakly fixed tissue (mouse 4) strong background staining occurs in the stratum oriens and stratum radiatum (Fig. 2F). The dependence of this monoclonal antibody on strongly fixed tissue for proper CCK immunostaining suggests that fixation might influence the conformation of its epitope. Because the perfusion used here is relatively brief, it is possible that CCK might be detectable using a different fixative or perfusion protocol. In any case, immunohistochemical detection of CCK in axon terminals of the hippocampus is delicate and we failed to obtain positive results with a commercially available polyclonal antibody.
These three examples underscore the fact that an accurate description of the distribution of an antigen in brain tissue sections might require comparison of various fixation and tissue-processing protocols. Previous knowledge of function and distribution of a molecule of interest provides crucial help for evaluating its distribution pattern and establishing whether these procedures should be improved. With regard to the three markers investigated here, the following points were considered.
• The perfusion procedure can strongly impact the strength of fixation. All mice were properly perfused, as evidenced by the complete absence of erythrocytes (which express endogenous peroxidase activity and become strongly labelled in immunoperoxidase experiments). The weak fixation of mouse 4, deduced from the soft consistency of the tissue, could have been achieved also with a fixative containing a lower concentration of paraformaldehyde.
• As a membrane protein participating to synaptic protein complexes, the GABAA receptor α3 subunit should not be detected predominantly in cell somata (as in Fig. 1C), whereas diffuse staining of the neuropil reflects its dendritic distribution. White matter is expected to remain totally unstained.
• Serotonin in the hippocampus is only present in axons, as serotonergic cells are located in the raphe nuclei; anterograde tracing studies have shown that fine varicose axons originate from the dorsal raphe nucleus and coarse axons from the median raphe nucleus (Freund et al., 1990). Both types of fibres are evident only in Fig. 2A.
• CCK is a neuropeptide expressed by a subset of GABAergic interneurons in the hippocampal formation, notably by a subpopulation of basket cells. Failure to detect it (as in Fig. 2D and F) most likely represents false-negative results, as confirmed by Fig. 2E.
A direct consequence of these observations is that multiple immunofluorescence staining experiments require in principle antibodies that have similar requirements for tissue fixation and processing. According to our results, it would be problematic, for instance, to combine staining for CCK and the GABAA receptor α3 subunit in the same section (or even in the same animal). Development of alternative fixation and tissue preparation procedures might be necessary to overcome such constraints. For example, we have shown recently that several limitations due to tissue fixation can be alleviated in tissue sections derived from living brain slices (Schneider Gasser et al., 2006).
Finally, these examples show that the effect of tissue fixation and processing on the apparent distribution of antigens in immunohistochemistry can be a major limiting factor for quantitative analyses, even for relative comparisons. The tissue sections illustrated in Fig. 2A and C are from mice that were perfusion-fixed with the same solution and processed under identical conditions, yet yielding dramatically different results. The only difference between these mice was the perfusion procedure (placement of the cannula and flow rate of the fixative). Slight variations in perfusion–fixation between animals are almost unavoidable in practice. The serotonin staining illustrates that such variations can dramatically affect the apparent density of serotonergic axons in the forebrain.