Cyclin A Is a Proliferative Marker with Good Prognostic Value in Node-Negative Breast Cancer

Approximately 80% to 90% of women with node-negative breast cancer are expected to be alive and free from disease 10 years after surgery (1, 2). However, a majority will be treated with adjuvant locoregional and systemic therapy according to empirical evidence from the Early Breast Cancer Trialists' Collaborative Group. Adjuvant chemotherapy reduces the annual breast cancer death by ∼25% to 40% in node-negative patients, corresponding to an absolute improvement in 15 years survival of 5% to 10% (3). Thus, overtreatment with systemic chemotherapy is a common problem in node-negative patients. The prognostic factors most commonly used to select patients that are likely to benefit from adjuvant therapy are age, lymph node status, tumor size, histologic grade, and human epidermal growth factor receptor 2 (HER2), whereas estrogen receptor (ER) and progesterone receptors (PgR) are considered as predictive factors for endocrine responsiveness (4, 5). Proliferative markers such as Ki67 or cyclin A are not recommended for clinical use as prognostic factors neither by the American Society of Clinical Oncology Guidelines for Tumor Markers in Breast Cancer nor by St. Gallen Consensus Conference 2007 (4, 5). Studies have identified Ki67 as a prognostic factor associated with high risk for relapse and worse survival, but because of lack of consensus concerning methodology and definition of cutoff value, it is not considered for clinical practice (5). However, the multiparameter gene expression analysis Oncotype Dx was recommended for use in node-negative ER-positive patients, both as a prognostic and predictive marker by American Society of Clinical Oncology Guidelines for Tumor Markers (5). Interestingly, it has been shown that the prognostic ability of gene expression signatures is due mostly to the detection of proliferation activity; removing nonproliferative genes from the signatures did not decrease their prognostic ability (6). The importance of proliferation is further highlighted given the suggestion that proliferation genes are better indicators of tumor grade than histologic grade (7).

Ki67 is a nuclear antigen that is present in the mid-G₁, S, and G₂, and the entire M phase of the cell cycle (8). Several studies have investigated the prognostic significance in breast cancer. Most of them have shown that overexpression of Ki67 correlates statistically with poor metastases-free survival and/or death (9). Cyclins are proteins that vary in a cyclical fashion during the cell cycle. Cyclin A increases in early S phase and decreases in mid–M phase and is consequently a proliferative marker (10). The prognostic role of cyclin A in early breast cancer has been studied in several studies, of which most agree that cyclin A can predict for recurrence or breast cancer death (11-15). We have previously shown that cyclin A can be correctly assessed using tissue microarray (TMA) techniques and that reproducibility between independent readers' results is satisfactory (16). Furthermore, we have investigated and suggested optimal cutoff values for Ki67 and cyclin A (15). In this study, we wanted to further examine the role of cyclin A as a prognostic factor in node-negative patients and to standardize scoring by validating previously suggested cutoff values.

The source population of the study was a cohort of women diagnosed with breast cancer in six counties in the Uppsala-Örebro region from 1993 to 2004. Information about the patients was derived from the Uppsala-Örebro Breast Cancer Register, which is a population-based clinical database with a coverage of >98% (17). Inclusion criteria were tumor size of ≤50 mm, no lymph node metastases, and no adjuvant chemotherapy. The number of women that met the inclusion criteria during the time period in question was 900. Within this cohort, eligible cases were defined as women who died from breast cancer. All eligible cases were selected. Women that were alive at the time of the corresponding case's death were eligible as controls. Two-hundred and forty cases were identified using the regional quality register for breast cancer and the national register for causes of death. For each identified case, one control was used. Fifty patients (10%) did not fulfill the inclusion criteria after reviewing data from patient files and pathology reports or because of missing tumor blocks: 26 patients (5.5%) had new/contralateral or locally advanced breast cancer, no paraffin blocks were found in 12 patients (2.5%), 6 patients (1.5%) had non–breast cancer deaths, 4 patients (0.8%) had distant metastases at diagnosis, 1 patient (0.1%) had received adjuvant chemotherapy, and no breast surgery was done 1 patient (0.1%). These patients and their corresponding cases/controls were not included in the study. The average age was 66 years for cases and 61 years for controls. The average tumor size was 20 mm for cases and 16 mm for controls. All patients underwent either modified radical mastectomy with axillary dissection, or conservative breast surgery with axillary dissection and postoperative irradiation of the breast. Fifty-three cases (28%) and 48 controls (25%) received endocrine therapy. Patients characteristics' including grade, hormone receptors, and HER2 are shown in Table 1. The study was approved by the local ethics committee in Uppsala, Sweden.

Paraffin blocks from the patients' primary tumors were collected. H&E sections were reviewed and areas with invasive tumor were selected. Each tumor was re-evaluated and reclassified according to the Elston and Ellis grading system (R.M. Amini), see Table 2 (18). Representative areas from each tumor were punched and brought into recipient paraffin blocks to construct TMAs consisting of two cores (1 mm diameter) from each tumor. Then, 3-μm-thick to 4-μm-thick sections were cut from array blocks and transferred to glass slides.

TMA slides were deparaffinized in xylene and rehydrated through a ladder of graded ethanol (absolute ethanol, 95%, 80%, and distilled water). Antigen retrieval was done in Tris-EDTA buffer (pH 9) in a microwave oven for 10 min (750 W) + 15 min (350 W) before being processed in an automatic immunohistochemistry staining machine according to standard procedures (Autostainer; Dako, Sweden). All antibodies were applied for 30 min at room temperature. The following monoclonal antibodies were used: cyclin A (NCL-Cyclin A, 1:100; NovoCastra Laboratories), Ki67 (1:200, M7240; Dako), ER (NCL-ER-6F11, 1:150; NovoCastra Laboratories), and PgR (NCL-PGR, 1:100; NovoCastra Laboratories). Immunostainings were detected via DAKO Cytomation envision/HRP kit K5007. For cyclin A and Ki67 stainings, tonsil samples were used as positive controls; and for ER and PgR stainings, breast cancer tissue was used. The primary antibody was omitted from negative controls.

HER2 status was determined using immunohistochemical staining with HercepTest (DAKO). Cases with a negative or weak and noncontinuous membraneous positivity were considered as negative (0 and 1+). A positive HER2 result was immunohistochemical staining of 3+ (a uniform and intense membranous positivity in more than 30% of the tumor cells) and cases with continuous membranous, but less intensive staining were considered as 2+. All cases considered as 2+ were further analyzed by fluorescent in situ hybridization, as well as cases with a cytoplasmic and/or doubtful immunohistochemical staining pattern. Fluorescent in situ hybridization ratios (HER2 gene signals to chromosome 17 signals) of more than 2.2 were considered as HER2-positive cases.

In situ hybridization was done on deparaffinized 4 to 5 μm tissue sections. Pretreatment was done with sodium sulfocyanate (Vysis) for 30 min at 80°C and protease treatment for 25 to 60 min at 37°C. Hybridization was done in DAKO Cytomation Hybridizer with denaturation 73°C for 5 min and 20 h overnight incubation at 37°C.

All cyclin A and Ki67 TMAs were analyzed by one investigator (C. Ahlin). See Fig. 1. Hormone receptor analyses were done by a pathologist (W. Zhou). All scoring was supervised by a board-certified breast pathologist (R.M. Amini). The percentage of cyclin A, Ki67, ER-positive and PgR-positive breast cancer cells was counted in high-power fields (40× objective) in both tissue cores on TMA. Only unequivocal nuclear staining was accepted as a positive reaction. In most cases, 1,000 cells were counted in each tumor, and for all tumors, a minimum of 200 cells. All cyclin A and Ki67 statistical analyses were done using both average and maximal values for each of the patients. When calculating for cyclin A and Ki67 maximum value in percentages, we counted the high-power field that had the largest number of positively stained cells out of the two biopsies and divided by the entire number of cells from that particular high-power field. To get the average percentage value, we divided all positive cells from the two biopsies with the entire number of cells from the same biopsies.

To obtain unbiased estimates of relative risk, controls were selected by incidence density sampling, which involves matching each case to a sample of those who were at risk at the time of the case occurrence. Because as many women as ∼20% of the entire cohort was chosen as controls and all eligible women with an event were studied as cases, the loss in power in comparison with a complete analysis of all cohort members was small. A study based on 190 cases yields a power of 90% to detect an odds ratio (OR) of 2.5 on a statistical significance level of 5% if we assume a prevalence of 30% cyclin A overexpression in the cohort.

Conditional logistic regression analysis was done to estimate ORs and confidence intervals (CI) using the proportional hazard regression procedure in statistical analysis software (SAS Institute, Inc.). Established and potential prognostic factors such as age, tumor size, hormone receptors, histologic grade, mitotic count, tubuli, nuclear atypia, Ki67, and cyclin A were analyzed in univariate models. Because histologic grade, mitotic count, nuclear atypia, Ki67, and cyclin A were highly correlated in this study, multivariate analysis was considered to be not appropriate. In addition, models adjusted for age and tumor size were done. Correlations of Ki67 and cyclin A to other clinicopathologic parameters were assessed with Spearman's correlation test for completeness. The cutoff values used in the study for routine stainings or clinicopathologic parameters are shown as reference values in Table 3. Cutoff values used for cyclin A and Ki67 were determined in a previous article in which metastases-free survival and overall survival were calculated for each decile. The most optimal cutoff value was defined as the decile that resulted in the highest risk ratios for metastases-free survival and overall survival (the 7th decile; ref. 15).

Results from routine stainings for hormone receptors and HER2 are described in the patients' characteristics (Table 1). For cases, the median for cyclin A_ave was 8% (range, 0-39%) and for cyclin A_max it was 15% (0-48%). The corresponding values for controls were cyclin A_ave 4% (0-42%) and cyclin A_max 8% (0-88%). For cases, the median for Ki67 average value (Ki67_ave) was 10% (range 0-85%) and for Ki67 maximum value (Ki67_max) was 20% (0-93%). The corresponding values for controls were Ki67_ave 6% (0-88%) and Ki67_max 13% (0-96%). Staining results were missing from 12 tumors (3%) regarding cyclin A, and 17 tumors (4%) regarding Ki67 stainings due to lack of invasive tumor on the TMA.

Ki67 and cyclin A were highly correlated to grade and to each other (see Table 2).

A statistically significant association was observed between expression of cyclin A_ave, cyclin A_max, Ki67_ave, Ki67_max, grade, mitotic count, tumor size, ER, PgR, and breast cancer death using conditional logistic regression in a univariate model. Overexpression of HER2 was not statistically associated with breast cancer death. The numerically highest ORs for breast cancer death were predicted by grade, tubuli, and cyclin A_max (OR > 3). OR for breast cancer death using Ki67 was numerically lower (OR < 2), but the difference was not statistically significant. When adjusting the statistical model for age and tumor size, no significant changes were seen (for overview, see Table 3).

Within the case group studied on its own, there was a significant difference in time to distant metastases comparing patients with low cyclin A to high cyclin A (40 and 28 months, respectively). Patients with low cyclin A had a longer survival from diagnosis than patients with high cyclin A (53 and 43 months, respectively). However, survival from distant metastases to death was longer for patients with high cyclin A compared with low cyclin A (median 8 and 5 months, respectively).

We examined the prognostic value of cyclin A in node-negative breast cancer using standardized methodology regarding scoring and cutoff values. To validate our patient material, we analyzed common prognostic factors such as tumor size, age, hormone receptors, and grade which all turned out to be prognostic factors in our material with an OR of ∼2 to 3 in a univariate analysis. In our study, HER2 was not a prognostic factor, probably due to few patients with HER2 overexpression (10% in cases and 7% in controls) and consequently low statistical power.

Using cutoff values defined in a previous article for Ki67 (Ki67_ave 15% and Ki67_max 22%) and cyclin A (cyclin A_ave 8% and cyclin A_max 11%), we found that both Ki67 and cyclin A could predict early relapse and breast cancer death (15). We examined cyclin A and Ki67 stainings considering both average values of positive cells and maximum values (hotspots). In previous studies, average values are most often used for survival prediction (11-13). However, in the clinical setting, pathologists commonly refer to maximum values (hotspots) of Ki67 which seems reasonable because the most dedifferentiated part of the tumor otherwise is considered when evaluating prognostic factors. We found that both average and maximum values regarding Ki67 and cyclin A could be used as prognostic factors. Numerically, cyclin A_max value was the best predictor of breast cancer death even though statistically no significant difference could be detected between either Ki67 and cyclin A or between average and maximum values.

Another goal of the study was to compare cyclin A to the already established prognostic factor grade and to Ki67. However, Spearman's correlation test showed a very high correlation between grade, Ki67, and cyclin A, which produces unacceptable uncertainty (large variance) in regression coefficient estimates (multicollinearity). Therefore, it was not statistically appropriate to perform a multivariate analysis including the mentioned parameters. The only way we could compare grade, Ki67, and cyclin A in this study was by looking at univariate analysis and compare the different risk estimates obtained by the different markers. Univariate analysis shows that grade, Ki67, and cyclin A are equally strong as prognostic factors with an OR of 2 to 4, even though grade and cyclin A numerically are stronger than Ki67. Grade is the only one of these that is an established prognostic factor. However, because grade is the sum of three different morphologic features, it suffers from a much higher subjectivity than immunohistochemical markers such as Ki67 or cyclin A (19). This has been shown in studies looking at reproducibility with achieved κ values of 0.5 for grade, 0.6 to 0.8 for Ki67, and 0.7 to 0.9 for cyclin A (16, 20, 21). Determination of grade is more time-consuming than scoring for either Ki67 or cyclin A. Furthermore, a recent study suggests that proliferation genes are better at indicating tumor grade than histologic grade (7).

Having shown that cyclin A is a prognostic factor in node-negative breast cancer, another very important issue arises: is cyclin A a predictor of chemotherapy response? A previous study on soft tissue sarcoma suggested that cyclin A, but not S phase fraction or Ki67, could predict for chemotherapy sensitivity (22). Studies on breast cancer and proliferation markers as predictors for chemotherapy response have suggested that the S phase fraction might predict chemotherapy response, at least in the neoadjuvant setting (23). As of today, no proliferation marker can be suggested as a predictor of response to chemotherapy, even though some data points in the favor of using S phase fraction which is of particular interest to this study because cyclin A is an immunohistochemical marker for the S phase. In this study, patients with low cyclin A had longer metastases-free survival compared with patients with high cyclin A which is to be expected from a good proliferative marker. Also, survival from diagnosis was longer for patients with low cyclin A compared with high cyclin A. However, survival from distant metastases to death was significantly longer for patients with high cyclin A compared with low cyclin A. Because the majority of chemotherapy-naïve patients with distant relapse are offered chemotherapy, this could imply that cyclin A is a predictive factor for chemotherapy response. Furthermore, a previous study showed that cyclin A lost its predictive value in patients receiving adjuvant chemotherapy, again suggesting cyclin A as a predictive factor for chemotherapy (15). However, further studies on this issue are warranted.

One limitation of this study is that women receiving adjuvant endocrine treatment were also included in the study. Hence, some women had received adjuvant systemic treatment. However, a univariate analysis showed that endocrine treatment was a weak and not statistically significant prognostic factor for breast cancer death in this cohort with OR, 0.8 (95% CI, 0.5-1.30). Additionally, an analysis on cyclin A adjusted for endocrine treatment showed no change in the prognostic power of cyclin A with an OR of 3.5 (95% CI, 2.1-5.6) in the adjusted analysis as compared with 3.4 (95% CI, 2.1-5.5) in the nonadjusted analysis. Thus, endocrine treatment did not appreciably affect the prognostic value of cyclin A.

Stratifying the analyses for patients with ER-positive tumors (ER+) and ER-negative tumors (ER−) suggests that cyclin A is more important as a prognostic factor in ER+ compared with ER− with an OR, 3.8 (95% CI, 2.2-6.6) and OR, 1.6 (95% CI, 0.7-3.6), respectively, which is well in line with previously published data (6). However, this stratification was not preplanned and should be interpreted with caution.

The source population of the study was defined from a population-based register with near-complete coverage. Hence, the generalizability of the results is not threatened by referral bias. The study design was a nested case-control study with control selection according to incidence density–based sampling and the exclusion from the case and the control series was equally large and due to similar reasons in both groups (24). Thus, our data should, in a valid way, reflect results that had been produced if the whole cohort had been studied. The power was appropriate to detect clinically relevant differences. We caution that analyses of ORs may not be the best way to build prognostic models. However, the purpose of this study was to analyze cyclin A in context with other, previously published markers. A further development would be to include cyclin A in decision-making type of analyses.

To summarize, we suggest that cyclin A is at least as good a prognostic factor as grade and Ki67 value in node-negative breast cancer. Additionally, reproducibility for cyclin A is better than for histologic grade and validated cutoff values are now available for cyclin A. Adding cyclin A as a proliferative marker to already established clinicopathologic factors may improve the separation of low-risk and high-risk breast cancer.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We are grateful to the Foundation at the Clinic of Oncology in Uppsala, Research Fund, the Gunvor and Ivan Svenssons Foundation, the Regional Research fund for Uppsala/Örebro, The Swedish Breast Cancer Association, and Lion's Cancer Foundation for financial support. We thank Ulrika Larsson for skilful technical assistance, Dr. Manuel de la Torre for excellent help in tumor scoring, and the Uppsala/Örebro Breast Cancer Group for scientific support. Lastly, we are indebted to our regional pathology departments for providing important tumor specimens.