Considerations and caveats at each step of the tissue processing protocol for pre-embedding immunogold labeling EM are listed here:
Initial fixation affects structural preservation and immunolabeling efficiency
Fixation typically decreases immunolabeling efficiency. In general, the classic EM fixative, glutaraldehyde, is not compatible with immunolabeling for the majority of antibodies, and paraformaldehyde (PF) is the preferred fixative for immunolabeling. Notably, structural preservation in PF-fixed samples without immunogold-labeled (Fig. 1a) is of better quality than those with labeling (Fig. 1c & d) because the latter need to go through permeabilization treatment and extensive washing between the various incubation steps.
The quality of structural preservation of PF-fixed samples can be reasonably high if samples are optimally processed. Fig 1a shows an example of dissociated hippocampal cultures fixed with 4% PF in PBS for 45 min. The membranes and various organelles were similarly well-preserved when compared to a sample fixed with 4% glutaraldehyde in 0.1 M cacodylate buffer (Fig. 1b). One conspicuous difference is that microtubules were only preserved in glutaraldehyde-fixed samples (arrows in Fig. 1b) but not in PF-fixed samples (Fig. 1a).
Typically, labeling efficiency is in reverse proportion to the degree of fixation. For some antibodies, the concentration of fixative and time of fixation need to be adjusted to a level where structural preservation will be further compromised. For example, structural preservation was better in the sample fixed at higher PF concentration and for a longer time (Fig. 1c) than the sample in Fig. 1d. However, even though membranes were not optimally preserved in this lesser-fixed sample (Fig. 1d), the resulting labeling at the postsynaptic compartment is still credible and quantification of labeling density is valid .
For perfusion-fixed brain tissues, over-fixation is a common cause for low labeling efficiency. Fig. 2a shows a brain sample that was perfusion-fixed with 4% PF for ~1 hour before the brain was vibratomed into 100 µm slices and washed in buffer. The labeling intensity with a shank 3 antibody at the postsynaptic density (PSD, arrows in Fig. 2a) was lower than that in another sample that was perfusion-fixed for 40 minutes (Fig. 2 b).
Permeabilization conditions affects membrane structure and labeling efficiency of plasma membrane proteins
Immunolabeling reagents often need to pass the biological membranes to reach their target epitopes. A common reagent to make the membranes permeable is detergent, which modifies the membranes to different degrees depending on the type of detergent, and its concentration and treatment time. Based on our empirical experience with more than 200 different antibodies, the typical permeabilization condition we chose for the first try with any antibody is a 30 min incubation in 0.1% saponin at room temperature. With this treatment, the structure of the biological membrane was generally acceptably preserved even with a very mild fixation at 15 min of 4% PF.
Saponin works preferentially on cholesterol, which is more abundant in plasma membranes (PM) than in other biological membranes , and thus, may selectively alter the configuration of epitopes of PM proteins. In turn, labeling efficiency of antibodies against PM proteins may be affected by saponin treatment. For example, on PM of skeletal muscles, labeling efficiency of a glucose transporter (GluT4) was substantially affected by the concentration of saponin . Here, we compared the labeling efficiency of PM labeling of a glutamate receptor (GluR2, an AMPA receptor of subtype 2) between samples permeabilized with saponin and samples treated with 10 min of 50% ethanol followed by several washes of water to clear ethanol off the tissue. For these ethanol-permeabilized samples, saponin was also omitted from all subsequent steps.
PM of neuronal soma was specifically labeled with GlurR2  after both permeabilization treatments (Fig. 3). However, PM appeared wavier in saponin-treated samples (Fig. 3a, b) than in ethanol-treated samples (Fig. 3c, d), and the labeling density was 40-55% lower in saponin-treated samples than in ethanol-treated samples (Fig. 3e, Additional File 1).
In summary, antibodies that worked well for immuno-fluorescence light microscopy (LM) may not necessarily work for immunogold labeling by EM. The two key factors for EM are the additional fixation required for EM samples, and milder permeabilization conditions needed to preserve membrane structures in order to interpret EM images. Our recommended preliminary testing for any antibody is to carry LM experiments with an EM-compatible fixation and permeabilization conditions. For example, samples can be fixed at 4% PF for 15, 30, 60 min initially and then permeabilized with 0.1% saponin for 30 min to see if fluorescence signals survived these conditions. If so, chances are good that the immunogold labeling would work for these antibodies at the EM level. Once it is established that a particular antibody works at the EM level, additional tweaking on fixation and permeabilization conditions can be tried to optimize between the balance of achieving better structural preservation and higher labeling efficiency.
It should be noted that Triton X, a popular detergent for immunofluorescence LM, should not be routinely used for EM studies because the biological membranes could be easily dissolved by this detergent and render the EM images uninterpretable. Triton X can only be used under specialized conditions at very low concentration and in conjunction with additional EM fixatives .
Quality control of secondary antibody
After testing a few different types of secondary antibodies conjugated with gold, our choice reagent is Nanogold-Fab’ from Nanoprobes (Yaphank, NY) due to its higher labeling efficiency  and stability over time. This secondary antibody is a Fab’ fragment of the IgG, conjugated with a covalently-bound small gold (1.4 nm) . Although the great majority of Nanogold secondary antibodies we received worked well, a quality check on any new shipment is still advised. One example is illustrated in Fig. 4 where sister cultures of hippocampal neurons were identically fixed and permeabilized, labeled with a CaMKII antibody, and subsequently incubated with two different lots of Nanogold secondary antibodies. Labeling density near the postsynaptic density (PSD) of one sample (Fig. 4a) is only 17% of the other (Fig. 4b, c). A possible reason for the low efficiency in lot 1 may be low percentage of conjugation of the gold to the secondary antibodies.
We also carried out periodic quality checks on the Nanogold secondary antibodies in our stock, and found that Nanogold antibodies were very stable, even over many years, if stored properly in the refrigerator at 4˚C. Additional File 2 indicates that Nanogold antibodies which had been stored for more than 4 years still yielded similar labeling intensity (80-92%) to recently shipped Nanogold.
Freezing Nanogold is the only handling condition in our hands that caused a substantial decrease in labeling efficiency. Thus, one reason for some lots of Nanogold to perform disappointingly (cf. Fig. 4a) could be poor handling en route, e. g. accidental freezing during shipment. The typical practice in this lab is to test any new lot of Nanogold secondary antibody with a reliable primary and secondary ab as a positive control. Once it passed the test to yield acceptable labeling efficiency, we would buy more of the same lot and keep them for several years. We always have several lots of secondary Nanogold on hand and do not wait until the supplies are very low to buy and test replacement.
Comparison of labeling efficiency by different silver or gold enhancement reagents
Since the gold particle conjugated to the secondary antibody is too small (1.4 nm) to be visible by conventional transmission EM, samples need to go through silver or gold enhancement to enlarge the small gold to become visible signals [7, 8]. The efficiency among three different types of silver or gold enhancement reagents are compared here.
The HQ silver enhancement kit from Nanoprobes needs to be handled in a darkroom under a red safety light. We found that it does not require absolute darkness and can be accommodated with a closed room with lights off, and a red safety light on the bench. The silver enhancement kit from Aurion, AURION R-Gent SE-EM, and the gold enhancement kit from Nanoprobes, Goldenhance, are used under regular bench conditions. However, even with the inconvenience of a darkened bench and the safety light requirement, the HQ kit turned out to be our choice reagent due to its higher efficiency and high specificity.
Figs. 5 a & b illustrate the difference between the HQ kit and the Aurion silver enhancement reagents in a parallel experiment. Hippocampal cultures labeled for SV2, a synaptic vesicle (SV) membrane protein , showed specific labeling on SVs in presynaptic terminals with both enhancement reagents. The Aurion enhancement kit yielded homogeneous-sized particles after 60 min of exposure at room temperature (Fig. 5a), the image of which is perfectly suitable for qualitatively illustrating the specific labeling. The Aurion kit also has the convenience that one can monitor for the progression of reaction product under a light microscope without the guess work on the treatment time. On the other hand, the HQ silver enhancement kit produced heterogeneous-sized particle after 6 min of treatment at room temperature (Fig. 5b). This particular sample in Fig. 5b was clearly over-enhanced that the large and crowded silver particles obscured the underneath structure of the SVs. However, labeling intensity is much higher by the HQ kit at 2.7 - 3.8 folds over the Aurion kit (Fig. 5d, Additional File 3). This higher detection sensitivity can be especially advantageous when detecting antigens of low concentration. Notably, the silver enhancement process progresses with time  and is temperature-dependent . Thus, the over-sized signals can be remedied by reducing time (Fig. 5c) or temperature of the silver enhancement treatment.
Labeling efficiency was also tested between the Nanoprobes HQ silver enhancement kit and the Nanoprobes Goldenhance reagents. Samples were identically treated except for the silver (Fig. 6a, c) or gold (Fig. 6b, d) enhancement steps for 8 min and 10 min each, respectively. The efficiency of the HQ kit is much higher by 2.3 - 4.1 folds (Fig. 6e, Additional File 4).
One major drawback of the Nanoprobes Goldenhance kit in our hands is that there were nonspecific, fine particles all over membranous structures (Figs. 6b, d, f). These artifacts are distinct from the real signals localized on the SVs (Fig. 6b), and were not caused by counterstaining (Fig. 6d). Fig. 6f shows a neuronal soma where membranes of different organelles, including Golgi, mitochondria and multivesicular body, were all decorated with such fine particles. One cautionary sign of the non-specificity of this artifact is that the particles in question are all over every piece of membrane, including the inner and outer membranes of the nuclear envelope and the inner and outer membranes of mitochondria where it is known that the compositions of these membranes are different [19, 20]. Although there were studies that successfully used Goldenhance without such nonspecific artifacts , we could not easy get rid of these artifacts and decided to use HQ because of its higher labeling specificity and efficiency (Fig. 6e, Additional File 4).
Recommendations on managing the HQ silver enhancement kit
Each lot of HQ silver enhancement kit needs to be characterized to determine an optimal development time. Figs. 7a & b show a pair of samples identically treated except for silver enhancement time. The sample enhanced for 5 min had smaller particles (Fig. 7a) than the sample enhanced for 8 min (Fig. 7b). It should be noted that these samples were processed 15 years ago, and we have noticed that the HQ kits in recent years required longer development time. Figs. 7c & d show a pair of parallel samples silver enhanced with two different lots of HQ kits for 12 min each. The sizes of the particles are smaller with one lot (Fig.7c) than the other lot (Fig. 7d). Thus, the silver enhancement time should be adjusted for different lots of HQ kits in order to achieve the desired size range of the particles.
Our typical practice was to buy several HQ kits of the same lot and store them at -20˚C. We thawed the HQ reagent, one kit at a time, and refrigerate the unused portion at 4˚C to be used within a month. If the thawed reagents were not expected to be used up within that time frame, we aliquoted the three components of the HQ kit into equal volumes and stored them frozen. The reagents were stable for approximately one year in our hands. Longer storage times either in the refrigerator or in the freezer, sometimes changed the optimized development time. We also noticed a trend that the quality of the silver particles tended to have more problems when the HQ reagents used were older.
Caveats on osmium treatment of silver enhanced samples
Osmium tetroxide treatment is generally required for EM samples in order to preserve membrane structures. However, it is also shown to reduce the size of the silver enhanced particles . In earlier tries, an osmium tetroxide treatment at 1% sometimes yielded irregular-shaped particles with a fuzzy border (Fig. 8a). We also found that the silver enhanced particles are poorly preserved with “reduced osmium” treatment (Fig. 8b), where potassium ferrocyanide is included with osmium tetroxide.
In order to consistently produce sharp looking silver particles, the routine in our lab now is to use fresh-made 0.2% osmium tetroxide in 0.1 M phosphate buffer at pH 7.4 for 30 min on ice. Because osmium tetroxide is very volatile, a 0.2% solution made ahead of time may become less in concentration by the time it is used. The consequence of which would be poorly preserved membranes.
One cautionary note is that on rare occasions, the osmium treatment step may bleach the silver enhancement product, resulting in samples with no silver particles at all. The fact that such samples had absolutely no silver particles anywhere in the samples raised concerns because even the control samples where primary antibody was omitted should still have some background particles. Thus, some chemical reactions must have bleached all signals between the steps of silver enhancement and the rest of the processing procedures. We now closely monitor the yellowish/brownish color of silver-enhanced samples by eye while adding the osmium tetroxide solution. We have found that if “bleaching” is to occur, the color would disappear within the first minute. It is not worth proceeding with the “bleached” samples because no signals will be preserved if bleaching occurred at this step. Also, to guard against the possibility that the silver enhancement reagents may contribute to this “bleaching” phenomenon, the repeat experiment would be silver enhanced with a different batch of the HQ kit.
Caveats on uranyl acetate (UA) en bloc staining
To increase contrast, tissues are often en bloc stained with UA following osmium treatment and prior to dehydration. This UA en bloc staining was carried out at cold temperature (4˚C) under light-tight conditions either wrapped in foil or kept in the refrigerator. Although this step is not necessary, we found the increased contrast in immunogold-labeled samples, which were typically weakly fixed with lesser structural preservation, worthy of this additional step.
Our routine was to make a stock 0.2 N acetate buffer at pH 5.0, then make the final solution of 1% UA in 0.1 N buffer, the pH of which became 5.3. An attempt to readjust the pH to 5.0 of the UA solution resulted in damaging the appearance of silver particles (small arrows in Fig. 9d). The particles became light grey, and appear to be on the verge of vanishing. Thus, the UA en bloc staining may also contribute to the disappearing of the silver particles. Adjusting the pH to 6.0 protects the integrity of the silver particles (Fig. 9b), but the contrast of the sample became so low that there is no longer any benefit of this UA en bloc staining step.
Because “greyish” silver particles were still occasionally seen in samples treated with 1% UA in 0.1N acetate buffer at pH5.3, we tried lowering the concentration and treatment time of the UA solution. Samples treated for overnight with 0.25% UA had a higher contrast and better preservation of the membranes (Fig. 10a) than samples treated for 1 hr (Fig 10b). However, the overnight treatment decreased labeling efficiency. In one experiment, labeling density for SynGAP, a Ras GTPase activating protein highly enriched in the PSD , in an overnight-treated sample was 56% of that of the 1 hr-treated sample (Fig. 10c). We suspect that some of the silver-enhanced particles disappeared during the overnight treatment. Thus, our routine treatment now is to en bloc stain with 0.25% UA for 1 hr at 4˚C.
Quality of silver particles related to sectioning and counter staining
Over the years, we noticed that the silver particles in thin sections of some samples appeared smudged as if the silver diffused out of the particles (Fig. 11a). Eventually, we traced this smudging of silver particles to the cleaning of the diamond knife used to cut thin sections. The same block cut with two different diamond knives could result in different looking silver particles. A knife that had been cleaned with ethanol prior to cutting thin sections produced smudged silver particles (Fig. 11a), and this smudging of silver can be remedied by soaking the knife in water over night (Fig. 11b).
Notably, a detergent (any kind of liquid soap)-cleaned diamond knife produced similar looking smudged silver particles, and the smudging effect left by the liquid soap is extremely difficult to remove. Even with careful management on the diamond knives, we sometimes still found smudged silver particles, especially in brain tissues. We suspect that the baskets used to hold these tissues during silver enhancement procedures might have been cleaned with detergent, thus, smudged the silver particles.
By reexamining archived grids from past years, it is clear that the quality of the silver-enhanced particles in thin sections all eventually deteriorated with time. A few examples are shown in Fig. 12. The shape of the silver enhanced particle was typically roundish with sharp edges immediately after sectioning (Fig. 12, left column), but deteriorated into irregular shaped, fuzzy particles with time (Fig. 12, right column). The degree of deterioration typically progressed with time, but both the speed and degree of deterioration were variable among different experiments, probably due to the different batches and/or age and handling of silver enhancement kits. Counterstaining of thin sections appears to somewhat protect the silver particles from deteriorating and could prolong the usable life of the thin sectioned grids.
In order to image the silver particles at their best condition, we counterstain thin sections the next day after they were cut, and examine the stained sections as soon as possible, at least within days. However, if the silver particles deteriorated in the sections, the same block can be cut further for more thin sections. The deterioration of the silver particles appears to be related to the exposed cut face of the block because the initial thin sections recut from the face of a previously cut block also contained deteriorated silver particles, yet the deeper sections of the same block yielded intact looking particles.
Considerations on Interpretation of immunogold labeling results
One of the crucial issues for any immunolabeling study is the specificity of the antibody used to label the protein of interest. Many different validation protocols are summarized in reviews [25, 26], but cross-reactive antibodies still pose problems due to lack of multiple approaches in verifications.
For immunogold labeling studies at the EM level, the simplest way to verify the specificity of an antibody is to carry a control sample where the primary antibody is omitted. In these control samples, there should only be background particles scattered randomly. If a primary antibody worked, specific signals should be localized to the expected sites with high precision. Furthermore, immunogold labeling at the EM level has an advantage of being able to use different antibodies to serve as controls for each other. For example, four different antibodies in Fig. 13 were each specifically localized to different sites which may be difficult to differentiate at the LM level, but can be unequivocally distinguished at the EM resolution level. Synaptophysin, an SV membrane protein, was localized to SVs in presynaptic terminals (Fig. 13 a, b) , while bassoon, an active zone cytomatrix protein, is localized at the active zone within 100 nm of the presynaptic membrane (Fig. 13c, d) . On the postsynaptic side, PSD95, a postsynaptic density protein, is localized immediate adjacent to the postsynaptic membrane (Fig. 13 e, f) , while shank, a PSD scaffold protein, is located to the deeper layer of the PSD (Fig. 13 g, h) . Each of these antibodies were specifically localized to their expected sites without cross-labeling other cellular elements.
Another advantage of immunogold labeling at the EM level is the precise localization of transmembrane proteins to their epitopes on either side of the membrane. For example, one antibody made against the N-terminus of the NMDA receptor (NR2B subunit) was localized to the extracellular side of the postsynaptic membrane (Fig 14 a) , and another antibody made against the C-terminus was localized to the cytosolic side of the membrane (Fig. 14b). The same extracellular NR2B antibody was correctly localized to the lumen of the endoplasmic reticulum (ER, Fig. 14c), so is another antibody against the N-terminus of an AMPA receptor, GluR2 (Fig. 14d) . Two other antibodies made against the cytosolic epitopes of ER membrane proteins, ryanodine receptor and Ip3 receptor, are localized to the cytosolic side of the ER membranes (Fig.14 e, f) .
It should be cautioned that some antibodies could cross react to additional, unintended epitopes . For example, one antibody against the homer 2 subfamily of the homer protein, a PSD scaffold protein , was correctly localized to the deeper layer of the PSD (Figs. 15a, b). However, this antibody was also localized to other locations unrelated to the expected distribution of homer proteins (Fig. 15c). Signals were seen on the cytosolic side of the nuclear envelope, associated with the nuclear pores (arrows in Fig. 15c). It was determined that this nuclear pore labeling is a cross reaction unrelated to the homer 2 protein because another pan homer antibody which labels all 3 subfamilies of the homer protein did not label the nuclear pore (Fig. 15d). Nevertheless, this Homer 2 antibody is still useful at the EM level because its labeling of the PSD is specific and undisturbed by the additional, non-specific labeling of the nuclear pores, which is distant and distinct from the PSDs.
Another example of an antibody localized to unintended organelles is shown in Fig. 16. An antibody against VGluT (vesicular glutamate transporter), an SV membrane protein specific for glutamate uptake , was correctly localized to SVs of glutamatergic synaptic terminals (Fig. 16a) and not to SVs of GABAergic inhibitory terminals (Fig. 16b). However, signals were also detected on cytosolic filaments in soma/dendrites (Fig. 16c). Since VGluT is an integral membrane protein, the label on filaments (inset of Fig. 16c) must be a cross reaction of this antibody and cannot represent VGluT localization. Although the localization on SV membranes is correct (inset of Fig. 16a), we decided not to use this antibody because the cross reaction was too widespread and can potentially obscure the correct signals.
One of the advantages of immunogold labeling is that the labeling density is quantifiable by counting individual particles . However, several issues can affect the validity of quantification. For example, immunolabeling reagents can typically penetrate throughout the monolayer cells of dissociated cultures, but brain tissues present a penetration issue due to tightly packed processes and narrow extracellular space, both of which are not conducive to penetration and diffusion of reagents. To guard against this penetration gradient issue, our routine practice is to cross section immunogold-labeled brain slices and to examine the cut edges for labeling (Fig. 17).
There is a steep downward gradient of label from the cut edge of the tissue toward the center of the brain slices (Fig. 17c-e). Although the depth of labeling and the steepness of the gradient can vary by different perfusion fixations or by different primary antibodies, the highest labeling density was consistently seen in the most superficial 1-2 µm of the brain slices. Thus, for quantification projects using perfusion-fixed brains, sampling is restricted to within 1 µm of the cut edges. It would not be a valid comparison if sampling were collected from different depth of the slices where labeling efficiencies are vastly different.
Another quantification issue is in the interpretation of aggregated signals. Tightly aggregated proteins can appear as dark material legitimately labeled with multiple signals. For example, CaMKII is shown to self-aggregate into clusters (Fig. 18 a, d) , and chromogranin A is known to aggregate in high concentrations in the dense core of dense core vesicles (Fig. 18 b, e) . However, tightly clustered particles could be artifacts, especially when there was no evidence of protein aggregates association with these clusters of particles (Fig. 18 c, f). Such clustering of particles is likely caused by clumping of secondary antibodies and could be miscounted as multiple signals. This possibility can be easily tested by using a known, good secondary antibody in parallel experiments for comparison. The lot of secondary antibody with high degree of clumping should be discarded because it cannot be used for quantification of labeling density.
Interestingly, the protein aggregates in some samples appeared denser than others, and the core of the aggregates lacked labeling (d, e). The CaMKII cluster in (d) was from a sample treated with NMDA for 3 min, where CaMKII molecules might have formed a very tight aggregation that impeded the penetration of labeling reagents. Whereas the CaMKII cluster in (a) was from a sample treated with glycine for 2 min, where the CaMKII molecules could be aggregated with a looser configuration, more conducive to penetration of reagents. On the other hand, the dense core in (e) was from a 6-day old culture, where the dense core vesicles could be more mature and the dense core was more densely packed than the ones in (b), which were from a 3-day old culture. Nevertheless, these examples illustrate that counting labels from (d, e) may not truly reflect the number of epitopes present in these tightly packed protein aggregates.