Ki67 Staining Protocol: Fixation, Retrieval, and Scoring

Ki67 staining depends on a chain of pre-analytical and analytical decisions, each of which can shift the proliferation index enough to change a clinical classification. From the moment tissue leaves the patient to the moment a pathologist or algorithm counts positive nuclei, variables like cold ischemia time, fixation duration, retrieval buffer pH, and antibody clone all influence the final number. Getting reliable, reproducible Ki67 immunohistochemistry requires controlling every link in that chain, and understanding where the biggest sources of error hide.

What Ki67 Actually Tells You

Ki67 is present in cells that are actively cycling through division, but its behavior is more nuanced than a simple on-or-off signal. Live-cell tracking studies have shown that Ki67 accumulates during the S, G2, and M phases of the cell cycle, and then degrades continuously once a cell enters G1 or exits the cycle entirely into a resting state. The amount of Ki67 left in a resting cell depends on how long it has been out of the cycle, making Ki67 a graded marker of proliferation rather than a binary switch.

This graded nature matters for staining protocols because weakly positive cells, those recently exited from cycling, sit right at the detection threshold. A protocol that is slightly too aggressive with fixation or slightly too gentle with antigen retrieval can push those borderline cells from positive to negative, artificially lowering the proliferation index. The practical consequence is that protocol optimization is not just about getting “a stain” but about getting consistent staining across the full intensity range.

Cold Ischemia and Tissue Handling Before Fixation

The clock starts ticking the moment tissue is removed from its blood supply. Cold ischemia time, the gap between excision and placement in fixative, is one of the largest and most underappreciated sources of Ki67 signal loss. A controlled study of breast cancer specimens found that Ki67 expression declined sharply with increasing delays before fixation. At half an hour to one hour of delay, Ki67 values were still in the high 30s (percent positive), but by six hours they had dropped into the high 20s, and by eight hours some specimens showed values as low as single digits.

An even more dramatic demonstration comes from animal studies simulating human autopsy conditions. When mouse tissue was left at room temperature for three hours before cooling, Ki67 staining of pancreatic beta cells dropped from about 7% to roughly 4%, and continued falling over the next several hours. The researchers concluded that past studies reporting low Ki67 in human autopsy tissue may have been measuring degradation artifacts rather than genuine biology.

The practical takeaway is straightforward: get tissue into fixative as quickly as possible. For surgical specimens in clinical settings, current guidelines for breast cancer biomarkers recommend minimizing cold ischemia time and using core needle biopsies when feasible, since their small size allows rapid and uniform fixation.

⁴

Fixation Duration and Its Interaction with Retrieval

Neutral buffered formalin remains the standard fixative for Ki67 immunohistochemistry, but how long tissue sits in it matters. Short fixation times tend to preserve Ki67 antigenicity well. One study using tonsil tissue found that good Ki67 staining was obtained at all microwave antigen retrieval times when tissue had been fixed for only four hours. But tissue fixed for 48 hours showed poor staining unless the retrieval step was extended to at least 14 minutes of microwave heating.

⁵

Fixative pH also plays a role. A comprehensive preanalytical study found that Ki67 positivity was enhanced by low-pH fixatives and short fixation times, and that the choice of primary antibody clone interacted with these variables. Whether the staining was done manually or on an automated platform, however, did not significantly affect the result.

⁶

The key lesson here is that fixation and antigen retrieval are not independent steps. A lab using overnight fixation protocols (roughly 24 hours) needs to compensate with a more aggressive retrieval step, while a lab processing biopsies with shorter fixation can use milder retrieval and still get clean staining. If you change one without adjusting the other, you risk either signal loss or excessive background.

Veterinary pathology offers a useful parallel. A study of canine mast cell tumors found that extending fixation to 48 or 72 hours caused some tumors to lose Ki67 immunostaining entirely, and others to shift from high to low growth-fraction classifications. The effect was not uniform across cases, meaning prolonged fixation does not just reduce signal intensity evenly; it can selectively eliminate staining in some tumors while leaving others relatively intact, which is worse than a consistent undercount.

⁷

Antigen Retrieval Buffers and pH

Heat-induced epitope retrieval is standard for Ki67, and the choice of buffer pH is one of the single most impactful analytical variables. A head-to-head comparison of retrieval solutions found that pH 9 EDTA was far superior to pH 6 citrate for Ki67 staining of tonsil surface epithelium, which is commonly used as a quality-control tissue.

⁸

A separate study testing three retrieval buffers at different pH levels (citric acid at pH 6.0, EDTA at pH 8.0, and EDTA-Tris at pH 9.0) confirmed that Ki67 retrieved best in the high-pH EDTA-Tris buffer. The same was true for P504S and P63 antigens, while CD10 was less sensitive to pH differences.

⁹

If your lab is currently using citrate buffer at pH 6 for Ki67, switching to an EDTA-based buffer at pH 8 or 9 is likely the single easiest change you can make to improve staining quality. Just be aware that higher-pH retrieval can also increase background in some tissue types, so any switch should be validated on your own control tissues before going into routine use.

Which Antibody Clone to Use

MIB-1 has been the default Ki67 antibody for decades, largely because it was the first clone validated for use on formalin-fixed, paraffin-embedded tissue. But it is no longer the only serious option, and external quality-assurance data suggest it may not even be the best one. UK NEQAS proficiency testing data showed that the 30-9 clone (Ventana) and K2 clone (Leica Biosystems) were associated with the highest mean staining-quality scores, and both were significantly better than MIB-1 (Dako), MM1 (Leica), and SP6 from various suppliers.

¹⁰

That said, “better staining quality” in a proficiency-testing context does not automatically translate into better clinical performance. A comparison of five Ki67 antibodies in breast cancer found that while most clones could separate high- and low-risk patient groups, there were considerable differences in the actual Ki67 labeling index scores they produced. The concordance across clones was only moderate. Some clones, including immunofluorescent-labeled MIB-1 and B56, failed to separate risk groups at certain cutoff thresholds.

¹¹

A more focused head-to-head comparison of SP6 and MIB-1 in primary breast cancer found that roughly 89% of samples were classified into the same proliferation group regardless of which antibody was used. Both antibodies were associated with worse outcomes when Ki67 was high, and inter-observer agreement was somewhat better for MIB-1 than for SP6.

¹²

The bottom line on clone selection is that it matters less which clone you pick than that you stick with one and validate your cutoffs against it. A lab switching from MIB-1 to 30-9 should not assume the old cutoff values still apply. Whenever you change clones, re-validate on your control tissue and, ideally, re-establish your clinically relevant thresholds.

Clinical Cutoffs and Scoring Recommendations

For breast cancer, the most common clinical application of Ki67, the International Ki67 Working Group published updated recommendations in 2021. When following their analytic and scoring guidelines, Ki67 cutpoints of 5% or below and 30% or above have sufficient clinical utility for patients with ER-positive, HER2-negative stage I or II disease. Tumors at 5% or below are considered clearly low-proliferation, while those at 30% or above are high-proliferation, and both categories carry enough certainty to inform chemotherapy decisions. The gray zone between 6% and 29% remains more ambiguous, and the working group has not endorsed a single cutpoint within that range.

¹³

The working group also recommended that specimens be processed in line with existing ASCO/CAP guidelines for HER2 and hormone receptors, and that core needle biopsies be preferred over excision specimens because their smaller size reduces fixation variability. These processing recommendations directly affect how reliable the Ki67 number will be once it reaches the pathologist’s microscope.

AI-Assisted and Digital Scoring

Manual Ki67 scoring is slow and variable. Different pathologists looking at the same slide can arrive at different proliferation indices, and even the same pathologist may score differently on different days. Digital image analysis and AI tools are increasingly used to address this.

A study that had pathologists score Ki67 both with and without an AI assistance tool found that AI reduced the average scoring error from about 6 percentage points to about 2 percentage points. Inter-observer reproducibility improved substantially as well, with the intraclass correlation coefficient rising from 0.70 without AI to 0.92 with AI. Scoring was also faster with the tool, dropping from an average of about 23 seconds per tissue microarray core to about 19 seconds.

¹⁴

An open-source deep learning model tested in a resource-limited setting achieved a diagnostic accuracy of 95% when compared against an expert pathologist’s scores, with sensitivity of 94% and specificity of about 98%.

¹⁵

Another study combining a standardized reference card with AI found that when pathologists used both tools together, their agreement with a gold-standard count was very strong, with intraclass correlation coefficients above 0.95, and inter-observer agreement exceeded 0.90.

¹⁶

AI scoring is not magic, though. The algorithm is only as good as the staining it receives. All the pre-analytical and analytical variables discussed above still apply. A poorly fixed, under-retrieved slide will produce an inaccurate AI count just as it would produce an inaccurate human count. What AI does well is eliminate the human variability layer once you have a well-stained slide in front of it.

Multiplex Panels That Include Ki67

Ki67 increasingly appears not just as a standalone stain but as one marker in multiplex immunofluorescence panels designed to characterize the tumor microenvironment. One nine-color automated panel included Ki67 alongside PD-L1, PD-1, CD3, CD8, Foxp3, CD68, and pancytokeratin, allowing researchers to assess which immune cells in and around the tumor are actively proliferating.

¹⁷

Quantitative single-cell imaging studies have combined Ki67 with ER, PR, HER2, TP53, P21, P16, CD8, and CD20 on tissue microarrays to map the spatial heterogeneity of breast cancer subtypes at single-cell resolution.

¹⁸

Multiplexing with Ki67 introduces extra optimization challenges. Each antibody in the panel requires its own retrieval and stripping conditions, and the order in which antibodies are applied can affect the signal of markers that come later. Ki67, being a nuclear antigen, tends to be relatively robust to the repeated heating cycles required by tyramide-based multiplex workflows, but it is still worth validating Ki67 signal intensity in the full panel against a single-plex control to make sure the multiplex process has not cost you signal.

When Ki67 Protein and mRNA Disagree

Researchers sometimes compare Ki67 immunohistochemistry results to MKI67 mRNA levels measured by RT-qPCR. These two measurements do not always line up well. One study found only a moderate correlation between quantitative immunohistochemistry and RT-qPCR for Ki67 (Spearman’s r of 0.50), which was notably weaker than the correlation for ER or PR.

¹⁹

A separate study comparing Ki67 protein expression in whole tissue sections versus tissue microarrays found that MKI67 mRNA correlated with protein expression at r = 0.52 for whole sections but only r = 0.34 for tissue microarray cores.

²⁰

Several factors explain this gap. Ki67 protein has a half-life that varies depending on where a cell is in its cycle, as described earlier, so protein levels at any given moment do not simply mirror ongoing transcription. Tumor heterogeneity also plays a role: a tissue microarray core samples a tiny area that may not represent the tumor as a whole, while mRNA extraction typically homogenizes a larger tissue volume. If your research design involves correlating protein-level Ki67 data with gene-expression assays, expect disagreement in the intermediate range. The two measurements are answering slightly different biological questions.

Tonsil as a Control Tissue

Tonsil tissue is the most widely used positive control for Ki67 staining, and for good reason. Its germinal centers contain a high proportion of proliferating cells, giving a strong positive signal, while the surrounding mantle zone and surface epithelium provide built-in low-proliferation comparators. Research has specifically identified the tonsil surface epithelium as an ideal monitoring site for Ki67 staining quality, because its expected staining pattern is well characterized and sensitive to changes in retrieval conditions.

²¹

If you are setting up a Ki67 protocol from scratch or troubleshooting an existing one, run your tonsil control first and evaluate not just the germinal centers (which will be positive under almost any reasonable conditions) but the surface epithelium. If the epithelial cells are not staining with the expected basal-layer pattern, your retrieval is probably insufficient. If you see diffuse cytoplasmic haze across the section, your blocking or detection system needs attention. The tonsil’s layered architecture gives you multiple internal checkpoints in a single slide, which is hard to beat as a quality-control tool.