Immunohistochemical Expressions of Ki-67, Cyclin D1, β-Catenin, Cyclooxygenase-2, and Epidermal Growth Factor Receptor in Human Colorectal Adenoma: A Validation Study of Tissue Microarrays

Tissue microarray (TMA) technology was first reported by Kononen et al. in 1998 (1), and since then, has been increasingly used in cancer research. It is a potentially important technique for translational research because of its high-throughput parallel molecular profiling at the DNA, RNA, and protein levels for large numbers of samples (2, 3). In distributing hundreds of small cylindrical tissue biopsies from individual blocks into a single tissue microarray, TMA has many advantages over conventional techniques. Among these advantages are high efficiency, uniform reaction conditions, reagent conservation, multiple built-in controls, reduced damage to donor tissue blocks, easier automated imaging analysis, and rapid linking of molecular changes to clinical variables (2-4).

The major criticism of the TMA technique is that it uses only a small fraction of a tissue specimen, which may not be representative of the whole tissue section, especially for antigens with heterogeneous staining patterns in tumors. Because of this potential limitation, many validation studies have been conducted to optimize the sampling strategy for various solid cancers (5-14). However, many biomarkers are expressed with different phenotypes in tissue samples, which may alter the validity and reliability of using TMA to assess these markers. In the present study, we evaluated the validity of TMA constructed from colorectal adenoma tissue for the immunohistochemical expression of five molecular markers [Ki-67, cyclin D1, β-catenin, cyclooxygenase-2 (COX-2), and epidermal growth factor receptor (EGFR)] with different distribution patterns and quantitative levels. We further evaluated the tumor sampling strategy and imaging analysis criteria.

Included in this study were 63 patients with a first-time diagnosis of adenoma between January 1996 and December 2002 at Vanderbilt University Medical Center. These patients were recruited as part of a large study to identify predictors for adenoma recurrence among patients with either multiple adenomas or adenomas with advanced pathology defined as ≥1 cm, tubulovillous or villous, or high-grade dysplasia. Written informed consent was obtained from all participants and relevant committees for the use of human subjects in research approved the study protocol.

Before constructing a TMA block, serial 5-μm sections were cut from each donor block. One of these sections was stained with H&E for marking morphologically representative areas of the tumor. Four areas in each large polyp case (one each at the periphery and deep zones, two at the middle zone) were targeted. Using Beecher Instruments Tissue Arrayer (Silver Springs, MD), tissue cylinders with a diameter of 0.6 mm were punched from the four targeted areas in each donor block and deposited into a 15 × 17 (255 cores) TMA block, which contained 240 cores of adenoma tissues and 15 cores of normal human tissues as built-in controls. The normal human tissue blocks (colon, small intestine, stomach, esophagus, liver, kidney, lymph note, skin, muscle, pancreas, prostate, brain, spleen, heart, and lung.) were obtained from PathServe Autopsy and Tissue Bank (San Francisco, CA). The TMA blocks were heated at 36°C for 30 minutes, and multiple serial 5 μm sections were cut and placed on charged slides. One section was stained with H&E.

Before staining, the slides were equilibrated to room temperature and then baked at 55°C for 1 hour. The standard indirect immunoperoxidase protocol (horseradish peroxidase detection kit; BD PharMingen, San Diego, CA) combined with monoclonal or polyclonal antibodies was used to detect Ki-67, cyclin D1, β-catenin, COX-2, and EGFR. The primary antibodies and antigen retrievals are detailed in Table 1. After developing the reaction with diaminobenzidine as the chromogen, the sections were counterstained with hematoxylin. The known positive tissues and adenomas were used for positive controls. The primary antibody was omitted for negative controls.

The study pathologist scored the whole tissue sections and TMA slides separately. The TMA tissue cores were imaged using the BLISS slide scanner (Bacus Laboratories, Inc., Lombard, IL). These images were saved electronically and then reviewed using WebSlide Browser 3 software for manual scoring. The pathologist-based manual scoring criteria are shown in Table 1. For all markers, the four TMA cores and the whole tissue slide of each sample were reviewed to estimate the average staining intensity, which is the fraction of total positive cells to total cells counted in each sample. The average staining intensity of the four cores in TMA slides was generated as a single variable to represent the expression level of a biomarker in the sample. All cells in the negative core are included in the denominator in this calculation. Identically, the average staining intensity of the corresponding whole section of the sample was also generated as a single variable. We did not derive zone-specific staining intensity score in this study, because there was no definitive borderline between the middle zone and superficial or central zones in whole sections, and the zonal scoring would arbitrarily introduce additional variations. The stroma, hemorrhage, necrosis, and poorly stained areas were excluded.

The immunostaining quantitative analysis was done with a system composed of an Olympus BX40 microscope (Olympus American Inc., FL), a Retiga FAST 1394 color digital camera (QImaging, British Columbia, Canada) and BioQuant NOVA Prime imaging analysis software (BioQuant, Nashville, TN). The video images were captured using a 10× objective lens (low-power magnification ∼1.1 mm²) under a constant state of exposure control. The diameter of TMA core is ∼0.7 mm in slides and the area is ∼0.35 mm². Therefore, a whole TMA core is included in a low-power field. When each field was measured, region of interest tools were used to exclude blank, folding, hemorrhage, necrosis, poorly stained, and stromal areas, whereas threshold tools were used to precisely define and measure total epithelial area, total nuclear area, positively stained area, and average gray density. Manual editing of fields was used to eliminate nonspecific artifacts. The average intensity score of the four cores analyzed from each case represented the expression intensity of markers. For Ki-67 and cyclin D1, which showed a nuclear staining phenotype, the nuclear area and positive nuclei in each field were measured. After four fields of the sample were measured, the final immunoreaction indices were generated automatically by setting algorithms “total positive area / total nuclear area” (ref. 14; Fig. 1A-C). For β-catenin, COX-2, and EGFR, which showed cytoplasmic and/or membranous staining phenotypes, the epithelial area, positively stained area, and average gray density of the positive area were measured. After four fields of the sample were measured, the final immunoreaction indices were generated automatically by setting algorithms “total epithelial area / total positive area × average density” (Fig. 1D-E).

Frequencies were used to describe categorical variables, the mean and SD were used for continuous variables. A weighted kappa statistic was used to evaluate the comparability of data from the semiquantitative whole section and TMA. The intraclass correlation coefficient was used to assess the comparability of the semiquantitative and quantitative TMAs. All analyses were done using SAS version 9.1 (SAS Institute, Cary, NC).

The clinicopathologic features of the participants included in this analysis are presented in Table 2. Approximately half of the patients were male. Slightly more than half of the polyps were >1 cm and were tubular.

Two 15 × 17 (255 cores) TMA blocks were constructed (Fig. 2), with a total loss rate of 16% (84 of 510). Among the lost cores, 72 were missing from the slide and 12 were mispunched with <10% of the disc area containing tumor glands. Any lost cores were placed in the next TMA block in order to obtain quadruplicate punches from each donor block for imaging analyses. The missing rate in smaller polyps (29%; 52 of 180) was significantly higher than that in larger polyps (≥1 cm in size, 7%; 20 of 300) in two TMA blocks (P < 0.0001).

The expression phenotypes of the five markers in adenoma tissues are shown in Fig. 3. In general, colorectal adenoma exhibited zonal expression patterns for all five markers. EGFR showed the most heterogeneous and weakest staining pattern, and COX-2 showed an irregular expression pattern in colorectal adenoma tissues.

We found excellent concordance between TMAs and standard whole sections for semiquantitative results of the five markers (Table 3). The percentage of agreements not due to chance (weighted kappas) between TMAs and whole sections were 79%, 90%, 88%, 77%, and 81% for Ki-67, cyclin D1, β-catenin, COX-2, and EGFR, respectively. The majority of discordant results for β-catenin (6 of 8), COX-2 (7 of 8), and EGFR (8 of 8) were “underevaluated” or “false-negative” in comparing TMA to whole sections. In these discordant cases, the immunostaining scores in TMAs were mostly lower than those in corresponding whole sections.

The quantitative data are compared with semiquantitative data from TMA sections in Table 4. A good correlation was observed between the two scoring methods for all five of the biomarkers with intraclass correlation coefficients ranging from 0.50 to 0.80.

TMA technology is increasingly being used for the high-throughput analysis of the diagnostic, predictive, or prognostic value of candidate markers. As yet, there is no standard sampling strategy suitable for many molecular markers expressed in certain tumor tissues. One of the key issues related to sampling methods is the number of cores needed in the TMA to reflect the expression levels of biomarkers in the whole section. Currently, most investigators use three 0.6-mm cores in TMA-based cancer studies (7, 10, 14, 15). However, in tumors which have marked heterogeneity, three 0.6-mm cores may not be sufficient, particularly for biomarkers with multiple levels of phenotype classification. For example, an evaluation using three cores per fibroblastic tumor specimen resulted in concordance rates between 96% and 98% for readings distinguishing between two-category phenotypes (p52 and Ki-67, respectively), and 91% for three-category phenotypes (pRB; ref. 16). In a study of lung cancers, most discordance was due to underevaluation (8). Two studies have shown that this issue can be addressed by using more cores (10, 15). Therefore, an increased number of cores may be needed for some heterogeneous tumors to avoid significant sampling bias and to improve concordance rates. In our study, we did not validate the optimal core numbers because (a) nearly half of all adenoma specimens are small (<0.5 cm) and four cores on small samples will exhaust the entire adenoma specimen, and (b) the average intensity of four cores and the corresponding whole section were collected and compared, instead of the separate zonal scoring of each TMA core and each zone in the corresponding whole section. The latter method would arbitrarily introduce additional variations because of the obscure borderline between the middle zone and the superficial or central zones in whole sections. Therefore, the one-by-one increase of core numbers with their zonal scores for the validation of optimal core numbers was technically difficult and not conducted in this study. Up to now, most validation studies punched three to five cores of each sample (6, 7, 11-14) and found that three cores are generally representative of whole sections. Considering the potential heterogeneity of biomarkers in our study, we used four cores in our TMA strategy and found a good correlation between TMA and the corresponding whole section even for heterogeneous markers such as EGFR. Thus, although we have a strong theoretical basis for our use of four cores, additional studies are needed to confirm this supposition. Most of the discordance we observed for β-catenin, COX-2, and EGFR was due to underevaluation or false-negative results, similar to the observations of p53, p16, and Rb expression in lung cancers (8), because of the punches sampled from tumors with heterogeneous and focal staining patterns. Our loss rate was 16%, consistent with reported levels of 4.3% to 20% (6-8, 11-13, 15), although we did find that the rate of loss was substantially higher for small polyps than larger polyps. It suggested that the small and thin biopsy tissues constructed in TMA are easily lost during the sectioning of the TMA slides, and hence, decrease the efficiency of TMA technology and waste the study materials. Therefore, these multiple small polyp tissues (∼20% of total cases in this validation study) are processed with standard whole sections, instead of TMA construction.

Most published TMA validation studies have used conventional manual scoring methods with two to four categories (5-13), although they may have used different manual scoring systems and these systems may have varied by tumor site. For example, p53 positivity was defined as the percentage of positive cells: >5% in rectal cancer (6) but >10% in breast cancer (12), whereas p53 overexpression was defined as >10% in fibroblastic tumors (7) but >50% in colorectal carcinoma (15), or even evaluated on a four-point scale (0-to-3) in ovarian carcinoma (11). The different cutoff values may affect concordance between TMAs and standard whole sections. In this study, we first scored EGFR, the most heterogenous and weakest marker in colorectal adenoma tissues, with criteria described by Nielsen et al. (ref. 9; 0-5%, negative; 5-20%, weak; >20%, strong) and then with other criteria described by Deeb et al. (ref. 17; 0-10%, negative; 10-50%, weak; >50%, strong). The latter method improved the concordance substantially (weighted kappa improved from 0.41 to 0.74). Therefore, it is critical to optimize the scoring criteria for each biomarker in order for TMAs to be representative of standard whole sections. Because the obscure borderline between middle and superficial or central zones may produce additional variations of zonal scoring, the average staining intensity of each marker in whole sections was used. The evaluation of average staining intensity of whole sections (7, 11-13) or randomly selected fields covering whole sections (6, 14) is a standard scoring method and is widely used by investigators. Inter-reader variability may contribute to the inconsistency between TMAs and whole tissue sections (particularly in manual scoring). This variability is substantially smaller in quantitative image analysis of TMAs. We compared three separate quantitative readings of COX-2 staining and found that the intraclass correlation coefficients ranged from 0.91 to 0.98 in our study (data not shown).

Currently, automated image capture and analysis can be conducted conveniently with commercial automated scanning systems and compatible imaging quantification software for the high-throughput expression profiling of proteins (18). Imaging quantification has advantages over conventional pathologist-based manual scoring, such as automation and the ability to quantify rapidly using a continuous scale which may overcome inaccuracies in distinguishing between 0 and 1+, 2+, and 3+ in the conventional four-point scale (19, 20). However, only a few groups have applied quantitative analysis in TMA validation studies for a few biomarkers such as ER (19), HER2 (20), and Ki-67 (14). The quantitative data of ER and HER2 from TMAs showed a high degree of correlations with semiquantitative data from TMAs (R = 0.732-0.884 for ER; R = 0.704 for HER2) and with clinical predicting outcome (19, 20). The quantitative analysis of Ki-67 in the TMA validation study was done by comparing quantitative data from TMAs (five cores from each case) and whole sections (10 regions subdivided equally for TMA sampling) in prostate cancers (14). The results showed that three cores were required to optimally represent Ki-67 expression with respect to the standard tumor slide. Three to four cores gave the optimal predictive value in a prostate cancer outcome array. In colorectal adenomas, however, there is no TMA validation report using quantitative analyses, and it is unclear whether a good agreement between TMAs and whole tissue sections scored with conventional manual methods signifies that the same agreement would be obtained if scored quantitatively. In the present study, the continuous data from the quantitative TMAs were compared with ordinal data from the semiquantitative TMAs, and good correlations between the two scoring methods were found for all five biomarkers with intraclass correlation coefficients ranging from 0.50 to 0.80. The excellent concordance between TMAs and standard whole sections analyzed by semiquantitative methods were also found for the five markers with the kappa values ranging from 0.77 to 0.90. This indirectly suggests that quantitative TMAs with four cores from each patient may accurately represent standard whole sections and can be reliably used in colorectal adenoma studies.

β-Catenin is the most important protein in the Wingless/Wnt pathway and has complex subcellular localizations (in the nucleus, cytoplasm, and cytoplasmic membrane) as well as oncogenic properties (21). In this validation study, the total scoring of β-catenin was used to show that the heterogeneous expression pattern of β-catenin did not affect the use of TMA in the extensive colorectal adenoma study. In the full cohort study, the staining and scoring methods have to be improved to distinctively record the intensity of nuclear β-catenin from cytoplasmic/membranous staining in adenoma tissues. A newly developed quantitative system has been reported to allow the rapid, automated, quantitative analysis of TMAs, including the separation of tumor from stromal elements and the subcellular localization of signals (19). The dual fluorescent immunostaining of β-catenin with other biomarkers such as TCF-4 (a nuclear protein combining with β-catenin to regulate target genes) may also be helpful.

One of the important issues raised by the investigators in the TMA study is how to properly preserve antigenicity in sectioned TMA slides, which must be cut once for maximal sectioning and may not be used for staining until months or even years later. A previous study showed that most proteins in archival paraffin blocks retain their antigenicity for >60 years (10), but once the archival tissue blocks were sectioned, the antigenicity of proteins on the slides would quickly degrade or even become lost with storage time, resulting in false-negative results (22-25). In our study, the sections were coated with a thin layer of paraffin on the surface and placed in a vacuum desiccator (Model 55300-00; Labconco Co., Kansas City, MO) at 4°C in a cold room for long-term storage. After 6 months, we compared the immunoactivity of five markers between these specially treated sections with the conventionally stored (untreated, at room temperature) or freshly cut sections in three cases of adenoma tissues and normal human tissues (kidney and esophagus). The three groups of sections were stained simultaneously for each biomarker and scored semiquantitatively in a blind way. The staining intensity of nuclear markers (Ki-67 and cyclin D1) showed a slight decrease of color in the conventional group, and the immunostaining of the three other markers did not have any apparent changes (Supplemental file). The evaluation of longer periods of storage time will be necessary to further verify the protective effect of our method for long-term storage of unstained sections.

In summary, the protein expressions of Ki-67, cyclin D1, β-catenin, COX-2, and EGFR displayed heterogeneous and zonal expression patterns in colorectal adenomas. The sampling strategy and quantitative TMAs with four cores of each patient may accurately represent standard whole sections and can be reliably used in colorectal adenoma studies.

We thank Anthony L. Frazier for technical assistance, Mark S. Ross for technical consultation of Bliss Slide Scanner system, and Bethanie Hull for literary suggestions.