α-Methylacyl-CoA Racemase Protein Expression Is Associated with the Degree of Differentiation in Breast Cancer Using Quantitative Image Analysis

Epidemiologic studies show that red meat and diary products, which are both sources of branched chain fatty acids, are associated with breast cancer risk. Case control studies from geographically disparate areas have found a significant positive association between the intake of meat, red meat, and high-fat meat and the risk of developing breast cancer (1-3). Breast cancer incidence is associated with per capita fat consumption and, interestingly, these rates change among migrating populations and provide additional evidence implicating a high-fat diet in cancer development (4).

α-Methylacyl-CoA racemase (AMACR) is a mitochondrial and peroxisomal enzyme that plays an important role in bile acid biosynthesis and β-oxidation of branched-chain fatty acids (5). Elevated levels of branched-chain fatty acids in the diet may contribute to induction and an increase in AMACR activity. AMACR levels increase in response to branched-chain fatty acids in dairy and beef products. However, the link of AMACR expression and neoplasia has only recently been made. Using high-throughput molecular and tissue technologies, it has been shown that AMACR is an important biological marker for prostate cancer, being overexpressed at the transcript and protein levels (6). AMACR is used in surgical pathology practice as a marker to distinguish benign prostatic glands form prostate cancer (7-12). In contrast with prostate cancer, very little is known about AMACR expression in normal breast and in breast cancer.

At this time, pathologist-based analysis is the current standard for the evaluation of immunohistochemistry; however, because of its semiquantitative nature, it is at times difficult to reproduce by different observers. For example, the interpretation of HER-2/neu by immunohistochemistry using the Hecept test varies among pathologists, raising doubts about the reproducibility of the results. Image analysis offers more reproducible results of target signal expression with a continuous, rather than nominal scale (13-22). Quantitative image analysis has not been fully evaluated in the assessment to tissue biomakers, especially when using high-density tissue microarrays.

The goals of our study were two-fold. First, we set out to characterize for the first time the expression of AMACR in normal breast, invasive carcinoma and its precursor lesion, ductal carcinoma in situ using tissue microarrays. The second aim of our work was to determine AMACR protein expression throughout a continuous range in a wide variety of breast samples and compare it to the pathologist-based, semiquantitative evaluation.

Breast tissues for tissue microarray construction were obtained from the surgical pathology files at the University of Michigan with Institutional Review Board approval. Two tissue microarrays were constructed. The first tissue microarray contained a wide spectrum of breast tissues, including normal breast and fibrocystic changes (15 cases), ductal carcinoma in situ (4 cases), invasive carcinomas (14 cases), and distant breast cancer metastases (10 cases). The second tissue microarray contained tissues derived from 160 consecutive patients with invasive carcinomas of the breast, with follow-up information treated at the University of Michigan from 1987 to 1991. Clinical and pathologic variables were determined following well-established criteria. All invasive carcinomas were graded according to the method described by Elston and Ellis (23); angiolymphatic invasion was classified as either present or absent. The tissue microarrays were constructed as previously described using a tissue arrayer (Beecher Instruments, Silver Spring, MD). Three tissue cores (0.6 mm diameter) were sampled from each block to account for tumor and tissue heterogeneity and transferred to the recipient block. Clinical and treatment information was extracted by chart review, done by the surgeon (M.S. Sabel) with Institutional Review Board approval.

A section of the tissue microarray containing normal breast, fibrocystic changes, ductal carcinoma in situ, invasive carcinomas, and breast cancer metastases was used for immunofluorescence analysis of AMACR. In addition, we also used five whole sections of normal breast derived from mammoplasty procedures and of breast cancer. The slides were soaked in xylene for 15 minutes to remove paraffin and then hydrated with graded ethanol. Pretreatment included placing the slide in a citrate buffer (pH 6.0) and heating in pressure cooker for 15 minutes. After blocking the slide in PBS containing 5% normal donkey serum and 0.1% Tween 20, the rabbit antibody against AMACR at 1:2,000 (kind gift of Prof. Ronald Wanders, Department of Clinical Chemistry, University of Amsterdam, the Netherlands), and mouse monoclonal E-cadherin (BD Biosciences, San Jose, CA) at 1:400 were applied (in PBST with 5% donkey serum) and incubated overnight at 4°C. After washing the slides, secondary donkey anti-rabbit Alexa 488 and donkey anti-mouse Alexa 555 (Molecular Probes, Eugene, OR) mixture was applied at 1:500 dilution and incubated in the dark for 1 hour. After washing the slides, anti-fade with 4′,6-diamidino-2-phenylindole was applied (Vectashield) and covered with a glass coverslip. Regular fluorescence images were taken using Zeiss Axioplan 2 microscope and confocal images were taken using Ziess LSM 510 META confocal microscope.

To test the expression of AMACR in relation to clinical and pathologic features of breast cancers, a 4-μm-thick paraffin-embedded tissue section of the tissue microarray was immunostained using a primary rabbit monoclonal antibody (p504S, dilution 1:25, Zeta Corporation, Sierra Madre, CA). Subsequently, slides were incubated sequentially with biotinylated secondary antibody, avidin-biotin complex, and chromogenic substrate 3,3′-diaminobenzidine. Slides were evaluated for adequacy using a standard bright field microscope. The majority of array spots contained tissue sufficient for the evaluation. Digital images were then acquired using the BLISS Imaging System (Bacus Lab, Lombard, IL). The tissue microarray was immunostained for estrogen and progesterone receptors and for HER-2/neu by using well-described and validated procedures (24, 25). For estrogen receptor staining, we used estrogen receptor antibody clone 6F11 (prediluted, Ventana Medical Systems, Tucson, AZ) and for progesterone receptor antibody 636 (DAKO, 1:400 dilution). HER2/neu immunostaining was done using CB11 antibody (1:40 dilution, NovoCastra, Burlingame, CA). Hormone receptor status was reported as positive or negative when >10% of the neoplastic cells exhibited nuclear staining (26). HER-2/neu status was reported as 0 to 3+ (27).

AMACR protein expression was scored using a standard, pathologist-based 4-tiered scoring system previously validated as negative (score = 1), weak (score = 2) when there was faint cytoplasmic staining or granular apical staining, moderate (score = 3) when there was diffuse granular cytoplasmic stain, and strong (score = 4) when there was diffuse intense cytoplasmic stain (6). Moderate and strong staining was considered as positive staining based on prior work. Because of tissue heterogeneity, the pathologist assigned a score only to the invasive carcinoma in each tissue microarray core. This ensured that only invasive carcinomas were scored, and not the surrounding benign glands or stroma.

The same tissue section, previously analyzed qualitatively and semiquantitatively by the pathologist, was scanned, and the tissue images were converted to digital files using the ACIS (ChromaVision Medical Systems, Inc., San Juan Capistrano, CA). This system consists of an automated robotic bright-field microscope module that is linked to a computer through a Microsoft Windows NT-based software interface. Proprietary software is used to detect the brown stain intensity of the chromogen used for the immunohistochemical analysis and compares this value to blue counterstain used as background. The ACIS II system is highly sensitive and can analyze stain intensity levels in a precise manner (range of 0-255 units). This system has been used previously to analyze immunohistochemistry on whole tissue sections and on tissue microarrays (13-22). We have tested the reproducibility of the ACIS II in pilot experiments before this study by scoring several tissue microarrays on separate occasions. The correlation coefficient for these experiments was r² = 0.973. For the current study, the desired areas of each core were characterized for precise intensity scoring. Taking into account tissue heterogeneity, we encircled the area of invasive carcinoma under the guidance of a pathologist with special interest and experience in breast pathology using the ACIS II software. Thus, only invasive carcinomas were assigned an intensity measurement; the surrounding benign glands or stroma were not scored. AMACR staining intensity was evaluated using ChromaVision on a scale of 0 to 255. The score assigned to each tissue microarray sample was automatically transferred by the ACIS II software to an excel spreadsheet. The biostatistician in the study (R. Shen) then analyzed the mean, minimum, and median of those values for each tumor. Investigators were blinded to all clinical outcome data.

The association between AMACR protein expression and the clinical and pathologic characteristics was assessed using the general estimating equation. The ordinal expression categories for AMACR were modeled using the multinomial distribution with the cumulative logit link. Tissue microarray elements were grouped by specimen (patient). The model calculates the odds of a higher expression score versus a lower score, with the odds ratio and 95% confidence intervals reported. To identify an optimal dichotomization of AMACR intensity that best differentiates survival outcome, sensitivity analysis on various cut points of the normalized intensity was applied. In particular, a series of equal-spaced threshold points ranging from the lower quartile to the upper quartile of the normalized intensity was used to dichotomize AMACR into low- and high-expressing categories. Using death or disease recurrence as patient outcome parameter, Kaplan-Meier estimates of the survival probabilities were computed and P values from log-rank tests were obtained to assess the discriminative power of the dichotomized marker. The optimal threshold point of dichotomization was then chosen to minimize the log-rank test P values.

By immunofluorescence, AMACR was found in the cytoplasm of normal and cancerous epithelial cells, including ductal carcinoma in situ and invasive carcinoma, in a granular pattern consistent with a mitochondrial and peroxisomal localization. As shown in Fig. 1, AMACR is present in the cytoplasm of normal ductal and myoepithelial cells and not in the stromal fibroblasts. AMACR is expressed in ductal carcinoma in situ and in invasive carcinomas.

Using high-density tissue microarrays, we next evaluated the expression of AMACR protein in 160 consecutive invasive breast carcinomas with long follow-up information by immunohistochemistry (Fig. 2). Table 1 shows the clinicopathologic characteristics of the patients. Of the 160 invasive carcinomas, 149 had available cores for evaluation. The association between AMACR protein levels and clinical and pathologic characteristics is shown in Table 2. Using both categorical pathologist evaluation and quantitative image analysis, increased expression of AMACR was associated with higher histologic grade, a well-known feature of aggressive breast cancer (ANOVA, P = 0.04; Table 2). AMACR expression increased with the grade of the invasive carcinoma. Pathologist-based evaluation of the immunohistochemical staining showed mean AMACR protein staining intensity of 1.75 (SE = 0.31) for grade 1, 2.24 (SE = 0.12) for grade 2, and 2.65 (SE = 0.13) for grade 3 tumors (ANOVA, P = 0.01). In agreement, image analyzer–assisted immunohistochemical quantitation revealed a mean staining intensity of 144.1 (SE = 9.6) for grade 1, 149.4 (SE = 16) for grade 2, and 154.1 (SE = 14.8) for grade 3 invasive carcinomas (ANOVA, P = 0.04), suggesting a relationship between levels of AMACR protein and tumor differentiation.

Pathologist-based and image analyses had a linear trend of association and overall good correlation (correlation coefficient 0.34, P < 0.0001). The pathologist could reliably distinguish between low and high intensity of AMACR expression (r² = 0.47), but was less accurate in discerning differences at low intensity of staining (Fig. 3A). The quantitative analysis system, because it provides continuous rather than categorical data, allowed for wider range of staining detection and more accurate evaluation of low-intensity staining (Fig. 3A). Using the Kaplan-Meier estimates of the survival probabilities, image analysis was able to better discern between patients with better and worse disease-free survival when compared with pathologist-based analysis. Image analysis evaluation suggested a trend for tumors with high AMACR expression to have a worse clinical outcome, but did not reach statistical significance (Fig. 3B). To further evaluate whether AMACR expression levels add prognostic value over the already established significance of the histologic grade of the invasive carcinomas, we calculated the c-index (concordance index). We found that when adding AMACR levels determined by image analysis, the prediction accuracy of the model, reflected by the magnitude of the c-index, improved slightly (c = 0.13) over the univariate models (grade: c = 0.11, AMACR: c = 0.04).

In the present study, we characterized the expression of AMACR in breast tissues, which was unknown. Although AMACR has been extensively studied in the context of prostate cancer, there is little data on its expression and significance in breast adenocarcinomas, and a previous report consisted of too few cases to analyze its relationship to clinical and pathologic characteristics and prognostic factors (28). In prostate cancer, detection of AMACR by immunohistochemistry is useful to distinguish benign prostate glands from prostatic adenocarcinoma (6, 8-11, 29). AMACR is also expressed in early prostate cancer precursor lesions, such as high-grade prostatic intraepithelial neoplasia, and has been suggested to be a harbinger alteration in the earliest phases of prostate cancer development (6, 8-11, 29). In contrast to prostate glands, we found, by immunofluorescence, that normal breast epithelium also expresses AMACR protein.

To further characterize AMACR expression in situ, we evaluated its expression in tissues from a large group of breast cancer patients and explored the associations between AMACR and histologic and clinical characteristics. AMACR was expressed in the cytoplasm of the invasive carcinoma cells with a granular and punctate staining. This observation was confirmed by immunofluorescence and is consistent with peroxisomal and mitochondrial localization.

In invasive carcinomas of the breast, AMACR protein levels increased with increasing histologic tumor grade, a measure of tumor differentiation and an indicator of a biologically aggressive phenotype (23, 30, 31). Numerous studies have found that patients with high-grade (poorly differentiated) carcinomas have higher rates of distant metastasis and worse survival than patients with low-grade (well-differentiated) carcinomas (23, 30, 31). Histologic grade was found to be an independent predictor or survival (30). A role for AMACR in tumor differentiation has been suggested by previous studies on prostate and colon carcinomas (12, 32). In prostate cancer, AMACR is moderately to highly expressed in well-differentiated tumors, and its expression decreased in hormone-refractory prostate cancer (12). Similarly, well and moderately differentiated colonic adenocarcinomas expressed moderate to high levels of AMACR protein, whereas anaplastic carcinomas of the colon had weak or no expression (12, 32). Notably, we found the opposite effect in breast cancer, as AMACR expression was inversely related to the degree of tumor differentiation. Poorly differentiated (grade 3) invasive carcinomas had the highest AMACR expression and well-differentiated (grade 1) tumors had the lowest. AMACR is not the only protein involved in lipid metabolism that has been found to be associated with, and perhaps to play a role in tumor differentiation. Several investigators found that the peroxisome proliferator-activated receptor-γ controls differentiation in breast, prostate, and colon cancer (33-35). Thus, our results strengthen the possible link between lipid metabolism and the process of tumor differentiation and support the hypothesis that this intriguing role of AMACR may be tissue specific. This hypothesis warrants further investigation.

Finally, we used two methods of scoring the immunohistochemical results: a standard pathologist-based, qualitative, and semiquantitative assessment, and an image-assisted quantitative system. The main difference between these methods is that the image analysis provides continuous data whereas the pathologist scoring results in categorical data. The continuous staining intensity values provided by the quantitative image analysis allows for better discrimination of subtle protein expression differences, which may not be apparent in the pathologist categorical evaluation. Our results show that whereas there was a good correlation between the two systems, the semiautomated quantitative analysis system allowed for a wider range of staining detection and more accurate evaluation of low-intensity staining. It was confirmed by numerous published studies that automated image analysis allows for detection of low expression staining not definable by manual scoring (13-22). Thus, we believe that it may be useful in the automated screening and evaluation of novel biomarkers in breast cancer.

In summary, we have characterized the expression of AMACR protein in normal breast and in breast cancer, and have explored the associations between AMACR expression and clinical and pathologic characteristics in a large group of invasive carcinomas of the breast using tissue microarrays, and two independent scoring systems: a standard pathologist-based approach and a quantitative analysis system. We found that AMACR is expressed in normal breast epithelium, in invasive carcinomas, and their precursor lesion ductal carcinoma in situ. In the group of invasive carcinomas, although AMACR does not have independent prognostic significance, its expression levels are associated with the degree of tumor differentiation. We provide further evidence for a link between proteins involved in lipid metabolism and tumor differentiation in breast cancer.Lastly, quantitative image analysis is a novel way to accurately and reproducibly evaluate immunohistochemistry in breast tissue samples using high-density tissue microarrays.

We thank Robin Kunkel for artwork and Karilynn Schneider for secretarial assistance.