INTRODUCTION

Since its description by Kononen et al. (1), the tissue micro-array technique has proven to enable high-throughput molecular analysis of hundreds of tissue specimens or cells in a single experiment. As many as 1000 cylindrical tissue cores from individual tumor blocks can be precisely arrayed into a new “recipient” paraffin block. Consecutive sections from the arrays then allow DNA, mRNA, or protein targets to be analyzed from defined, morphologically almost-identical regions of the tumor. Previous studies demonstrate that such arrays provide powerful tools to analyze molecular mechanisms important in carcinogenesis (1, 2, 3).

The Radiation Therapy Oncology Group (RTOG) recognizes the importance of tissue resources for present and future correlative trials. Since 1996, the RTOG tissue bank has acted as a core resource for numerous laboratory correlative studies associated with phase III clinical trials.

We currently have a total of 12,261 blocks of prostate, brain, head and neck, gastrointestinal, gynecological, bladder and lung tumors, stored in optimal conditions, thus providing a significant tissue resource to qualified investigators. Many valuable tissue blocks are loaned to our group from various institutions, although some of them refuse to provide blocks for studies because of concern about retention. The tissue micro-array technique would allow entire archives of thousands of existing formalin fixed tissues from pathology laboratories to be placed in a few dozen high-density tissue micro-arrays to survey a variety of different tumor types as well as different stages of tumor progression (1, 2). It would also have the potential to resolve the problem of tissue block retention and block depletion.

Before adopting the micro-array technique for routine use at the RTOG tissue bank, we sought to evaluate whether such arrays could substitute for whole-section analyses. To do this we compared the percentage of agreement between replicate arrays and whole section assessment. We also evaluated the technique itself, in terms of time and cost effectiveness. Procedural improvements and interpretive correlations were very important secondary considerations.

MATERIALS AND METHODS

Case Selection

A total of 150 samples consisting of formalin-fixed, paraffin-embedded banked tissue from prostate cancer, gliomas, and head and neck carcinomas were chosen from RTOG Protocols 7401, 7611, 7903, 7918, 8302, 8007, 8201, 8531, 8610 and 9501. Samples were chosen from those cases in which more than one large block of tumor was available so that the availability of tissue for correlative studies would not be compromised. One hematoxylin and eosin section was made from each block to define representative tumor regions. Discrete tumor regions that were not necrotic or fibrotic were marked with black ink to guide the histologist in the construction of the tissue array. Another section was cut at 4 μm for routine immunohistochemistry preparation.

Tissue cylinders, 0.6 mm in diameter, were randomly punched from the marked tumor regions of each donor tissue block. These cylinders were placed into four recipient paraffin blocks, using a custom-made precision instrument (Beecher instruments, Silver Spring, MD; 1). Tissue core punches were organized in three groups of 50 samples each (Fig. 1), approximately 1 mm apart. A sample template designating the sample location by case number (identical to the number of the parent slide) was created. The location was designated using a number–letter grid for ease of correlating position to case numbers. Slides prepared after punching of the blocks document the slight tissue damage and distortion that results (Fig. 2).

FIGURE 1
figure 1

Low-magnification photograph of group of 50 head and neck cancers representing 1/3 micro-array samples in the array. Marker punch is seen at lower left to direct pathologist.

Full size image
FIGURE 2
figure 2

Photograph of head and neck cancer slide stained with hematoxylin and eosin after punching for tissue array. Note lack of distortion of surrounding tumor.

Full size image

Preparation of Tissue Micro-Arrays

Four identically prepared tissue arrays (for descriptive purposes, designated 1–4, see Statistical Analysis, below) were prepared from each parent block.

Micro-array blocks and parent blocks were cut using an automated Micron 355S microtome and standard histological techniques. After some experimentation, it was decided to abandon the use of adhesive-coated tape-sectioning system to prepare the micro-array slides. Our experience with this alternate technique has been described (4). Much less tissue damage, less background staining, and fewer missing samples occurred using conventional histologic technique than using the adhesive-coated tape-sectioning system (1).

Immunohistochemical Analysis of p53

One 5-μm section of each tissue micro-array block and one section of each original tissue block was cut using conventional histological techniques and transferred to S/P® Brand Superfrost plus slides (Allegiance Healthcare corporation).

Immunohistochemical evaluation for p53 protein overexpression was done by standard methods (5). DO7 monoclonal antibodies directed against the p53 protein, antigen retrieval using citrate buffer at pH 6.0, 25 minutes of microwave heating and 25 minutes of cool down, LSAB II detection system (DAKO K0675), and diaminobenzidine as a chromogen were used. All staining was done using a DAKO Autostainer (DAKO Corporation, Carpinteria, CA)

Interpretation of p53 Staining

Parent slides of each case were separately scored for their p53 expression. Each tissue array was also scored. Threshold for abnormal p53 expression was >20% of tumor cells showing positive nuclei. We and others have used this threshold in previous studies (5, 6).

Two methods of evaluation of data were used:

1. The original slide and each separate array were visually estimated for total percentage of cells with p53 expression. Concordance was determined between the results obtained in each separate array and the whole section.

2. The number of positive and total cells in the central strips of a 10 mm × 10 mm grid (Kfarmannruling reticle) from each micro-array was counted under 200 × magnification. The percentage of p53-positive cells were calculated. Percentages of the four micro-arrays were averaged and correlation established with the results from similar cell counts from the parent slides. Estimate of positivity and count of percentage of positive cells were then compared.

Statistical Analysis

There were four comparisons to be done for each disease site. Initially, the results of the first array were compared with the parent slide. Agreement was defined as positive for p53 on the array and positive for p53 on the slide, or, negative for p53 on the array and negative for p53 on the slide, using threshold value of 20% of positive cells. The percentage of the 50 cases in agreement was determined for each disease site. The next analysis combined the results of the first and second arrays. The number of cases in which the results of both arrays agreed was determined, and then the percentage of these cases that agreed with the parent slide was determined. Similarly, the next analysis determined the number of cases that agreed on three arrays and then determined the percentage of these cases that agreed with the parent slide. The final analysis combined the results of four arrays in the same manner.

RESULTS

A total of 149 cases were available for analysis in each one of the four multitumor micro-arrays. There was representation of 88 to 100% of the cases in each array. Missing samples in some arrays and too few tumor cells present in the area to be counted were responsible for the noninformative cases. Only one case was absent in all four arrays. Seventy sections were cut from the four-array blocks without loss of any other cases. We have found it routinely possible to cut ≤100 slides from array blocks of ≥2 mm thickness, based on our experience with 20 different array preparations, each with three replicate arrays (unpublished observations). The average number of cells in each strip varied from 40–584 cells. Up to 2240 cells were present in a single 0.6-mm punch sample. The four arrays contained a total of 7781cells.

Table 1 presents a summary of the correlation obtained between the visual estimate in each separate array and the visual estimate of the whole section. The table shows 86–90% of agreement when using only one array and a progressive increase in the probability of matching the results of the parent section using the combination of two, three and four arrays. The use of four arrays gave us the ability to match the parent section in 97–100% of the cases.

TABLE 1 Agreement between Slides and Tissue Microarrays 1–4 (Visual Estimate) Using Positivity Threshold of 20%
Full size table

When using the cell count method, which calculated the percentage of positive cells, we obtained matching results in 84% of brain tumor cases, 89.7% of prostate tumor cases, and 90% of head and neck tumor cases (Table 2). Discrepancies were present in 19 out of 149 cases (12.75%). Reviewing the discrepancies, two cases scored 9% and 15% positive, thirteen cases varied between 21% and 30% positive, and four cases varied from 31% to 38% positive. In other words, the further the values were from the selected threshold of 20%, the more likely the values were to agree (Table 3). In addition, many discrepant cases had arrays selected from contiguous areas in blocks with significant heterogeneity.

TABLE 2 Agreement between Slides and Tissue Micro-Arrays 1–4 (cell count), Using Positivity Threshold of 20%
Full size table
TABLE 3 Discrepancies between Whole-Section Slide Results and Array Results (Cell Count)
Full size table

Cost Comparison

Evaluation of cost of immunohistochemical studies of p53 in four arrays compared with cost of examining 150 slides showed the following:

Immunohistochemistry materials and reagents costs were $61.34 for the 4 array slides and $2276.46 for the 150 slides. However, the creation of the tissue array was very labor intensive. The histologist spent 32 hours to prepare, build, and cut the arrays but only 8 hours to cut the slides from the 150 blocks. The time taken to perform this assay on the arrays was 3 hours, in contrast with 16 hours for the 150 slides. Total cost of the procedure, including technical time and materials but excluding the time to create the arrays, was $761.32 for the arrays and $2,756.46 for the 150 conventional whole-section slides. We estimate that the cost to produce an array, including the time to select tumor regions and prepare the array blocks, is ≥ $5000 for 4 replicate arrays. This significant cost is balanced against the reality that 4 arrays can generate data for ≤200 different tests, depending on the block thickness.

The time to interpret the slides was decreased for arrays because arrays consisted only of tumor regions that were optimal for scoring (Fig. 3). Whole-slide sections vary in tumor content and other features, which requires time to select a region for counting (Fig. 4). We have found that evaluation of a whole-slide preparation for p53 expression requires about 5–10 minutes per case. The evaluation of the array case requires only 1–2 minutes per case (unpublished observations). In our experience, the evaluation of array preparations is much easier when the number of cases is limited to <400. Larger numbers of array spots on a single slide produce significant fatigue in viewing. Such fatigue promotes errors in interpretation because of the need for interruption of effort and potential for distraction. This is the principal reason that we chose to position each case as a single spot in each of three array blocks rather than attempting to position all of the cases in triplicate on a single slide.

FIGURE 3
figure 3

Photomicrograph of head and neck cancer specimen stained for p53. Percentage of positive nuclei is easily interpreted.

Full size image
FIGURE 4
figure 4

Photomicrograph of head and neck cancer specimen from which area shown in Figure 3 was removed. Slide has been stained with p53. Heterogeneity of staining is evident.

Full size image

DISCUSSION

The tumor tissue array technique has been shown to be a powerful and rapid tool for the simultaneous analysis of hundreds of tumor specimens (1, 2, 3, 7, 8, 9, 10, 11). In several recent studies, results were available in 1 to 2 weeks (3, 11). Furthermore, using this technique in combination with fluorescence in situ hybridization (FISH), complementary DNA (cDNA), and comparative genomic hybridization (CGH), the tissue micro-array technique is considered not only to be a complement and enhancer for these genome-screening tools but also to be a powerful strategy to explore the molecular basis of cancer progression (1, 2, 7, 8).

Notwithstanding its power as a high-throughput method, a critical issue is whether minute tissue samples are representative of their donor tumors (1, 2, 3, 9). Several studies have demonstrated that the arrays are representative by comparing array results and existing data from the literature (1, 2, 3). Recent studies in fibroblastic tumors, bladder cancer, and prostate cancer have shown that the expression of Ki-67 by immunohistochemistry is highly correlated with the results in whole-slide preparations if triplicate or quadruplicate arrays are used (12, 13, 14). In the prostate cancer study, increasing the number of replicate arrays beyond 4 did not result in increased correlation with whole-slide results, suggesting that 3 or 4 arrays are indeed the optimal number (14). In one recent study, tissue micro-array results of estrogen receptor, progesterone receptor, and p53 protein expression in breast cancer showed more significant association with clinical survival than the results of whole-slide preparations (15).

To demonstrate whether the 0.6-mm cores of the tissue arrays are representative of the donor tissue blocks, we analyzed the expression of p53 in whole sections of the donor tumor and their respective tissue arrays, using methods used in previous laboratory correlative studies of prostate, anal canal, and head and neck cancer (5, 6). Using standardized immunohistochemical protocols, p53 can be detected as a generalized distribution in tumor tissue. This characteristic allows effective correlation and evaluation between micro-array technique and whole-section analysis. The 20% cutoff point was selected because it offered the greatest prognostic power in multiple studies (5, 10). Counting or eyeballing the percentage of positive cells in four different high-power fields of a whole section can be extrapolated to the tissue micro-array slide because the area contained in a 200 × field of a whole section is approximately the same as a 0.6-mm core seen under the same magnification.

By making visual estimates or counting the percentage of positive cells in the arrays, we obtained excellent correlation (84–100% agreement) between the conventional studies and the tissue micro-array technique. Using the combined score of three or four arrays, there is a very high probability of correlation, confirming the results of others.

In this pilot study, we have demonstrated and corroborated the utility of the tissue micro-array technique. Using this technique, a large number of specimens were efficiently evaluated on a single slide. The original tissue specimen (paraffin block) was retained, with only slight distortion of four punch holes and considerable remaining tissue for additional studies, despite the predominance of biopsy specimens in our tissue bank (3, 4, 8, 11). For tissue banking, this slight damage supports the main objective of the tissue bank—that is, to preserve the maximum amount of tissue in each tissue block so that larger numbers of correlative trials can be done with the patient material. More institutions should be willing to provide tissue blocks for 0.6-mm core removal, particularly if the paraffin block will be quickly returned to them in useable condition.

Laboratory correlative studies are highly valued in clinical trial investigations because such studies will ultimately help to stratify patients to most optimal therapy. Such studies are often statistically compromised by the lack of available tissue on all patients in the trial (16). We are hopeful that that application of this method will allow us to accrue more complete patient materials on our trials. Another important benefit of application of tissue array technology in clinical-trials tissue banks is the opportunity for multiple investigators to simultaneously test different markers on the same patient population and then compare and correlate their results. Unexpected associations of marker expression by investigators pursuing varying hypotheses may then be found. For example, in our head and neck trials, we have investigators pursuing antiangiogenesis, EGFR antagonism, and the role of tumor proliferation in clinical trials and associated laboratory correlative studies. If promising interactions are suggested, arrays will allow for more rapid testing of additional patient cohorts. Finally, use of tissue arrays, consisting of defined and carefully selected tumor regions, also enables more standardized interpretation of tumor markers than does examination of whole-slide preparations (9). Such standardized interpretation will also promote more reliable statistical evaluation (16).

Up to 100–200 sections can be cut from each array block (1, 3, 4). One challenge to this approach is that further consecutive sections may contain missing samples. This issue depends on the tissue depth of the original specimen provided to us. Some blocks have been almost depleted by the time we receive them. In this study, we were able to obtain up to 70 sections from each original block before starting to miss a significant number of cores. Depletion of blocks most commonly occurs because of repeated sectioning of blocks at different times, requiring the histotechnologist to “reface” the block before cutting on a different microtome and potentially causing loss of tissue (4). We hope that promotion of providing cores of tissue instead of unstained sections will alleviate this serious problem to laboratory correlative studies using clinical-trials tissue banks.

Discrepancies in observations were observed; we attribute these to tumor heterogeneity. In discrepant cases, several punches were closely clustered, which did not adequately sample the heterogeneous tumor. Tumor heterogeneity is a significant issue in choosing the sample for micro-array immunohistochemical studies. Several investigators have mentioned the possibility of improving the representativeness of tissue micro-array data by including random samples from different sites of a tumor in each array (1, 8, 9, 14).

When taking into account time expenditures and cost issues, we found that the histologist and pathologist used more time in selecting blocks and constructing the array but that technologist time in performing the immunohistochemical testing and pathologist time for interpretation of array results was significantly less.

We conclude that the tissue micro-array technique offers important advantages for tissue banks associated with clinical-trials groups such as RTOG. The potential advantages are the accrual of more complete patient samples, opportunity for increased numbers of correlative studies at significantly reduced incremental costs, and the potential for improved standardization of testing and resulting. Because the results of tissue arrays are highly congruent with findings obtained from analysis of much larger tissue specimens, this technique is preferable for many of our proposed tissue bank investigations. Use of at least three or four replicate arrays compensates for tumor heterogeneity. For markers with highly specific localization within tumors, tissue micro-arrays may not provide such an effective alternative to whole-slide staining.