Urinary Metalloproteinases: Noninvasive Biomarkers for Breast Cancer Risk Assessment

The metalloprotease family of enzymes is integral to the process of tumor progression, from angiogenesis and cell migration to remodeling of the tumor microenvironment, invasion, and metastasis. As part of this process, matrix metalloproteinases (MMPs) enter the circulation and can potentially serve as biomarkers for disease stage and progression. Elevated levels of MMPs and a disintegrin and metalloprotease 12 (ADAM 12) have been shown in breast tumors (1-4) and the serum and plasma of breast cancer patients (5, 6). We were the first to report that urinary MMPs and ADAM 12 could be also be detected in the urine of breast cancer patients and that these urinary MMPs are predictive of disease status (7). This finding confirmed that overproduction of MMPs by a tumor communicating with the vascular and lymphatic systems results in increased levels of MMP activity in urine as well as blood. It has also been reported that levels of other regulatory molecules overproduced by tumors, such as the angiogenic peptide basic fibroblast growth factor, have been measured in body fluids of cancer patients and have been shown to be independent predictors of disease status (8, 9). Since our original report, there are now several studies that support our findings that urinary MMPs predict neoplastic disease status (10-15).

Subsequently, we and others have shown that MMPs are required for the switch to the angiogenic phenotype, an early and critical event in cancer growth and progression (16, 17). We have isolated and identified ADAM 12 as being present in the urine of breast cancer patients (18). ADAM 12 is a member of a glycoprotein family that is related to MMPs. As with MMPs, increased expression of ADAM 12 has been reported previously in breast, colon, and lung carcinoma tissues (4). We have also shown that urinary ADAM 12 levels also significantly increase with disease progression in breast cancer patients and correlate with stage of disease (18).

The goal of identifying women at high risk of developing breast cancer and providing safe effective risk reduction to this group is compelling. Breast cancer remains the most common cancer among women and the second leading cause of cancer deaths in women today (19, 20). The American Cancer Society estimates that ∼182,460 women in the United States will be diagnosed with invasive breast cancer and 40,480 women will die of the disease in 2008 (21). Earlier detection and treatment are thought to improve survival, yet breast cancer can be “an unpredictable disease” because even very small lesions at the limit of detection by mammography, magnetic resonance imaging, or palpation can progress to metastatic disease (22-24). Given the current limitations of early detection, identifying women at risk for the disease and providing risk reduction strategies is a critically important goal. As a result, breast cancer risk assessment is becoming an increasingly significant part of the process of counseling women about their health. This has become especially important as genetic testing and risk reduction continue to evolve.

Accordingly, atypical hyperplasia and lobular carcinoma in situ (LCIS), recognized as markers of an increased risk of breast cancer, are coming under increased scrutiny. Elevated levels of MMPs have been documented in the plasma of women at high risk for breast cancer (5, 6). Additionally, angiogenesis has been documented previously in atypical hyperplasia and LCIS (25). Our group has reported previously that MMPs and ADAM 12 can easily be measured in urine and that elevated levels of these urinary enzymes accompany the angiogenic switch process. Given that MMP activity and angiogenesis are among the earliest events in tumor progression, we hypothesized that the study of these urinary proteins would provide useful information regarding atypical hyperplasia and LCIS, proliferative lesions of the breast that signal increased risk of breast cancer development. Here, in this study, we show that urinary biomarker levels are significantly elevated in women at increased risk for breast cancer and that this information may potentially be used in conjunction with the Gail model as a novel breast cancer risk assessment tool.

The study population included 148 women: 44 women with biopsy-proven atypical hyperplasia, 24 women with biopsy-proven LCIS, and 80 normal controls. Women undergoing evaluation and treatment for breast complaints were enrolled at Surgery, Radiation Oncology, and Medical Oncology clinics at Beth Israel Deaconess Medical Center, Mount Auburn Hospital, and Dana-Farber Cancer Institute. Normal healthy controls were enrolled from the population of women who came in for routine screening mammograms at Brigham and Women's Hospital and reported no chronic medical problems and no breast complaints, had a normal mammogram reading, and were on no medications.

All participants completed a detailed medical history form at the time of urine donation and gave fully informed consent to join the study. Risk scores for normal controls and women with atypical hyperplasia were calculated using the modified Gail model (26, 27). Gail scores were not calculated for women with LCIS, as this model is not valid for women with this diagnosis. Institutional review board approval for the study was obtained at each institution. Pregnant and breast-feeding women were excluded from the study. Characteristics of the study population are detailed in Table 1.

Urine was collected according to the institutional bioethical guidelines pertaining to discarded clinical material. Patients were seen in an ambulatory care setting and provided a voided specimen. Urine samples were kept on ice for no longer than 2 h and then frozen at -20°C, as described previously by us, and then stored at-80°C (7). Protein concentration of urine was determined by the Bradford method using bovine serum albumin as the standard. Before analysis, urine samples were tested for the presence of blood using Ames Multistix 7 reagent strips (Miles), and specimens containing blood were excluded.

Urine samples were analyzed by substrate gel electrophoresis (zymography) as reported previously (7). Thirty microliters of each urine sample were subjected to electrophoresis using gelatin as the substrate. Zymograms were processed and evaluated independently without knowledge of the clinical status of the individuals from whom the urine specimens were obtained as reported previously (7). MMP identity was verified by immunoblot analyses using monospecific antibodies (7).

Equal amounts of urinary protein (20 μg) were separated by SDS-PAGE under reducing conditions as described previously (18). Resolved proteins were electrophoretically transferred to nitrocellulose membranes (Schleicher & Schuell) and treated as described previously (28, 29). A polyclonal antibody against human ADAM 12, rb122, was used at a concentration of 1 μg/mL (18, 30). Labeled proteins were visualized with enhanced chemiluminescence (Pierce Chemical). Band intensities were analyzed with UN-SCAN-IT (Silk Scientific) software digitizer technology.

Mammograms were evaluated using the standard American College of Radiology Breast Imaging Reporting and Data System (31).

Urinary MMPs were compared among atypical hyperplasia, LCIS, and normal controls using χ² analysis. ANOVA with Bonferroni-adjusted comparisons was used to evaluate differences in ADAM 12 levels among the three groups (32). Multiple stepwise logistic regression analysis using a backward selection procedure was applied to determine predictors that differentiate atypical hyperplasia and LCIS from controls by considering four MMPs, ADAM 12 as a continuous variable, age, and Gail scores with the likelihood ratio test (LRT) used to assess statistical significance (33). Odds ratios and 95% confidence intervals for significant predictors were determined using exact methods and probability curves for estimating the likelihood of atypical hyperplasia as a function of ADAM 12 levels and Gail 5-year risk scores were derived using regression variables (slope and intercept coefficients) from the final multivariate model (34). Receiver operating characteristic curve analysis was applied to assess diagnostic accuracy of ADAM 12, Gail scores, and the combination for differentiating atypical hyperplasia from normal (35). Statistical analysis was done using the SPSS software package (version 15.0; SPSS). Two-tailed values of P < 0.05 were considered statistically significant. Power analysis was conducted a priori and indicated that a minimum sample size of 24 patients in each of the atypical hyperplasia and LCIS groups and 80 controls would provide 90% power (α = 0.05; β = 0.20) to detect a significant difference of 20% in the positive expression of each MMP between patients and controls and using a binomial Z-test for independent proportions (36) and a 30% difference in mean ADAM 12 levels between the study groups using ANOVA (version 6.0, nQuery Advisor; Statistical Solutions).

The study population consisted of 148 women: 44 women with atypical hyperplasia, 24 women with LCIS, and 80 normal controls. All diagnoses were biopsy proven. Gail score calculations were consistent with an average risk of breast cancer in the normal controls having a mean 5-year risk of 1.0 and were elevated in the patients with atypia with a mean risk of 3.8. There was no significant difference in smoking or drinking habits between the groups. Mammograms were read as normal with Breast Imaging Reporting and Data System scores 1 or 2 in 96% of the normal controls, 60% of the women with LCIS, and 36% of the women with atypia. Fifty-two percent of the women with atypia and 36% of the women with LCIS had mammograms classified as Breast Imaging Reporting and Data System 4 or 5, which are scores suspicious for, or highly suggestive of, malignancy. The study group was slightly older than the control group; appropriate adjustments were made in the statistical analyses (Table 1).

MMP-9, MMP-2, the MMP-9/NGAL complex, and ADAM 12 were consistently detected in the urine of the majority of the patients studied. Representative zymograms for MMP-9 and Western blots for ADAM 12 are shown in Fig. 1.

With respect to ADAM 12, univariate analysis indicated that women with atypical hyperplasia and LCIS had mean levels of 20.7 ± 16.8 and 14.7 ± 6.9 densitometric units (DU), respectively, which were significantly higher than normal controls (2.1 ± 2.8 DU) as determined by ANOVA with Bonferroni adjustment (both P < 0.001). The median ADAM 12 level for normal controls was 0. Atypical hyperplasia and LCIS groups did not differ significantly from each other in mean or median ADAM 12 level (P > 0.20; Table 2A). There were also significant differences in the percentage of individuals with positive MMP-9 between normal controls and women diagnosed with atypical hyperplasia or LCIS (Pearson χ² = 6.17 on 2 df; P < 0.05).

Stepwise multiple logistic regression analysis revealed that continuous ADAM 12 level (P < 0.0001), positive MMP-9 (P = 0.02), and age (P = 0.04) were independently predictive in differentiating women diagnosed with atypical hyperplasia from controls (Table 2B). The adjusted odds ratio for ADAM 12 in differentiating atypical hyperplasia from control is 1.4, implying that each 10-unit increase is associated with an increased odds of 28 times (1.4¹⁰) that the individual has atypical hyperplasia rather than being a normal healthy control. This is equivalent to an increased probability of 97%, where probability = odds / (1 + odds) = 28 / 29. For binary MMP-9 analysis, an individual who is positive has an estimated risk five times higher to have atypical hyperplasia compared with testing negative for MMP-9 (odds ratio, 5.1; 95% confidence interval, 1.4-17.9). Other variables tested, including MMP-2, MMP-9/NGAL, and MMP > 150 kDa, were not predictive of atypical hyperplasia (all P > 0.05).

When evaluating the predictors of LCIS, logistic regression indicated significant multivariate predictors identical to those for atypical hyperplasia, including ADAM 12 (P < 0.0001), MMP-9 (P = 0.014), and age (P = 0.05; Table 2B). These variables provide independent information in differentiating women with LCIS from normal controls, where the adjusted odds ratio for ADAM 12 is 1.6, implying that each 10-unit increase is associated with an increased odds 110 times (1.6¹⁰) that the individual has LCIS rather than being a normal control. This is equivalent to an increased probability of >99%. A woman who is positive for MMP-9 has a risk of LCIS over 13 times higher compared with an individual testing negative for MMP-9 (odds ratio, 13.8; 95% confidence interval, 1.7-110.7). Other variables, including MMP-2, MMP-9/NGAL, and MMP > 150 kDa, were not significant predictors of LCIS (all P > 0.05).

By means of logistic regression analysis, nonlinear equations were derived to estimate the probability of atypical hyperplasia and LCIS based on different intervals ADAM 12 and were highly significant for atypical hyperplasia (LRT, 58.4; P < 0.0001) and for LCIS (LRT, 53.3; P < 0.0001). Empirical data for controls and patients with a diagnosis of atypical hyperplasia (Fig. 2A) or LCIS (Fig. 2B) are represented by bars for each group, reflecting the percentage of women with ADAM 12 levels in each interval. Theoretical curves illustrating the probability of atypical hyperplasia or LCIS diagnosis compared with normal are shown according to ADAM 12 interval in each figure and clearly show the separation between the patients and controls.

As shown in Fig. 2A, 57% of controls and only 7% of patients with atypical hyperplasia had ADAM 12 levels of 0 DU, whereas 75% of patients with atypical hyperplasia and only 4% of controls had levels over 10 DU. The predicted probability of atypical hyperplasia is 7% for individuals who have ADAM 12 levels of 0, 40% for those with levels between 5 and 10 DU, and 85% for women with ADAM 12 levels 10 to 20 DU and 95% for levels over 20 DU. Comparatively, as depicted in Fig. 2B, 57% of controls and only 4% of patients with LCIS had ADAM 12 levels of 0, whereas almost 80% of patients with LCIS and only 4% of controls had levels over 10 DU. The probability of LCIS is <5% for individuals with positive ADAM 12 levels <2 DU, 52% for those with levels 5 to 10 DU, 85% for women with ADAM 12 levels of 10 to 20 DU, and 97% for women with levels >20 DU.

Urinary ADAM 12 levels were then multiplexed with Gail 5-year risk scores, which reflect clinical information (27). Gail scores less than 1.67% are considered low risk, whereas scores equal to or over 1.67% are high risk for the development of breast cancer. Multiple logistic regression analysis identified ADAM 12 level (LRT, 19.92; P < 0.0001) as a significant predictor for differentiating atypical hyperplasia from controls. The results of this modeling approach can be seen as the increasing probability of atypical hyperplasia with increasing ADAM 12 levels separately according to high or low Gail 5-year risk (Fig. 3). For example, the probability of atypical hyperplasia for an ADAM 12 level of 2 DU is 90% for women who have Gail scores ≥1.67% and essentially 0% for those with low-risk Gail scores <1.67%. On the other hand, the probability of atypical hyperplasia in women with low-risk Gail scores <1.67% (bottom curve) starts to increase with moderately high ADAM 12 levels (e.g., levels of ≥12 DU). For example, ADAM 12 levels of 14 and 15 DU are associated with probabilities of 50% and 75%, respectively, in this subgroup of women with low-risk Gail 5-year scores. There is an estimated ≥90% probability of atypical hyperplasia in individuals with Gail 5-year risk of <1.67% and ADAM 12 levels of ≥16 DU. As Gail scores are not appropriate for use in women with LCIS, this model applies only to women with atypical hyperplasia.

Similarly, receiver operating characteristic analysis of continuous ADAM 12 levels alone shows excellent discrimination in differentiating women with atypical hyperplasia from normal controls, with area under the receiver operating characteristic curve of 0.914. Urinary ADAM 12 levels also provide exceptional discrimination in differentiating women with LCIS from normal controls, with area under the curve of 0.950. As the maximum area under the curve is 1.0 for any test, these scores show the enormous predictive accuracy of urinary ADAM 12 levels.

When urinary ADAM 12 levels are multiplexed with the clinical information that Gail 5-year risk scores provide, further receiver operating characteristic analysis indicates that the optimal combination is Gail + 0.15 × ADAM 12 (area under the curve = 0.996). Therefore, the best performance is obtained when both ADAM 12 and the Gail risk score are used together. The optimal cutoff for this combination index is 2.8. Using the cutoff of 2.8, the sensitivity for the combination is 0.976 (41 of 42 atypical hyperplasia cases were classified correctly) and specificity is 0.977 (43 of 44 controls were classified correctly). Consequently, use of this ADAM 12-Gail combination index in our population yields only one false positive (one control has a combination index of 2.95, scoring above 2.8) and one false negative (one woman with atypical hyperplasia has a combination of 2.3, scoring below the 2.8 cutoff).

Because the atypical hyperplasia and LCIS study groups were older than the controls and had a higher percentage of menopausal women, we examined five potential confounders to assess their association with ADAM 12 and MMP-9. Age, timing of urine collection, menopausal status, alcohol usage, and smoking were each examined to determine whether any of them correlated with these biomarkers and thus could be confounding variables in this study. None of these five variables were significantly associated with urinary ADAM 12 or MMP-9 results.

Age. There were no significant correlations between age and ADAM 12 levels in normal controls (Pearson r = 0.09; P = 0.54), atypical hyperplasia (Pearson r = -0.03; P = 0.83), or LCIS (Pearson r = -0.04; P = 0.83). Comparing MMP-9 positive versus MMP-9 negative, there were no significant differences in age based on the Student's t test: normal controls (48.1 ± 8.5 versus 47.6 ± 7.9; P = 0.80), atypical hyperplasia (54.6 ± 7.4 versus 53.4 ± 8.9; P = 0.60), and LCIS (56.2 ± 9.8 versus 52.3 ± 7.2; P = 0.39).

Timing of Urine Collection. ADAM 12 levels (DU) collected before or after diagnosis were not significantly different according to timing of urine collection based on Kruskal-Wallis test (atypical hyperplasia, P = 0.32; LCIS, P = 0.07). With respect to MMP-9, χ² analysis revealed no relationship between the timing of urine collection and the percentage of MMP-9 positive expression for atypical hyperplasia (P = 0.83) or LCIS (P = 0.48).

Menopausal Status. Menopausal women were compared with premenopausal women with regard to urinary ADAM 12 and MMP levels. ADAM 12 levels (DU) were not significantly different according to menopausal status based on the Mann-Whitney U test (controls, P = 0.81; atypical hyperplasia, P = 0.37; LCIS, P = 0.08). MMP-9-positive rates according to menopausal status indicated no differences based on Fisher's exact tests (controls, P = 0.47; atypical hyperplasia, P = 0.11; LCIS, P = 0.99).

Alcohol Usage. Women who did not drink alcohol were compared with those who consumed less than five alcoholic beverages a week and those who drank five or more alcoholic beverages a week. ADAM 12 levels (DU) were not significantly different according to alcohol usage based on the Kruskal-Wallis test (controls, P = 0.28; atypical hyperplasia, P = 0.15; LCIS, P = 0.50). The MMP-9-positive rates also showed no differences in alcohol usage with χ² analysis (controls, P = 0.57; atypical hyperplasia, P = 0.61; LCIS, P = 0.63).

Smoking. Women who never smoked were compared with those who smoked in the past but stopped smoking, those who continue to smoke less than a pack a week, and those who smoked a pack or more a week. ADAM 12 levels (DU) were not significantly different according to smoking status based on the Kruskal-Wallis test (controls, P = 0.82; atypical hyperplasia, P = 0.62; LCIS, P = 0.55). There was also no difference in MMP-9-positive rates according to smoking status based on χ² analysis (controls, P = 0.47; atypical hyperplasia, P = 0.78; LCIS, P = 0.09).

As the options for breast cancer risk reduction improve, the identification of women with an elevated risk for developing breast cancer is becoming increasingly important (37). Current approaches to risk assessment include a complete clinical evaluation with careful physical exam, radiologic studies, family history, and risk profiling using mathematical models, such as BRCAPRO and Gail, to aid in decision making (38).

High-risk patients are encouraged to comply with a regimen of close surveillance, including mammograms and breast exams. If the lifetime risk is 25% to 30%, more intensive screening with breast magnetic resonance imaging alternating with mammograms as well as genetic counseling and testing can also be considered (39, 40). Although mammography remains the “gold standard” for breast cancer detection, mammogram screening yields a false-negative rate of 10% to 30% and is compromised in women with high breast density (41). False positives are also a substantial problem given that 10% of all screens are read as abnormal and almost all of these are false positives (42, 43). Lifestyle changes, such as decreasing fat and alcohol intake and exercise benefits, can also be encouraged (44). Finally, medical risk reduction can be approached by avoidance of exogenous estrogens and consideration of Tamoxifen, Raloxifene, or an aromatase inhibitor while acknowledging that the side effects of these medications preclude use in many women (45). Ultimately, more reliable, less invasive, and less expensive approaches for breast cancer risk assessment and early detection, as well as better alternatives for medical intervention, are needed.

In this study, we have analyzed the urinary expression of MMP-9 and ADAM 12 in patients with biopsy-proven atypical hyperplasia and LCIS, indicators of an increased risk of developing breast cancer. We have also carefully examined potential confounding variables, including age, timing of urine collection, menopausal status, alcohol usage, and smoking, to determine whether any of them were correlated with these biomarkers. We found that none of these five variables were significantly associated with the urinary ADAM 12 or MMP-9 results. However, in keeping with current guidelines for careful analysis of new biomarkers (46), our results require further validation in a much larger population before clinical application.

It is important to note that MMPs are involved in the modulation of modulate normal cellular behavior and cell-cell communication as well as tumor angiogenesis and progression (47). For this reason, MMPs might be elevated in conditions that are not specifically breast cancer related (48-50) and would never be used as a sole determinant of risk assessment or diagnosis. Therefore, in clinical context, and in conjunction with the Gail model and a full assessment of the patient, these noninvasive biomarkers could be a useful adjunct to currently available risk assessment tools.

Our data show that urinary ADAM 12 and MMP-9 are highly significant predictors of breast cancer risk markers, atypical hyperplasia and LCIS. ADAM 12 levels in particular were found to provide excellent discrimination in differentiating women with atypical hyperplasia or LCIS from normal controls. Moreover, when ADAM 12 levels are multiplexed with the Gail risk score, the resultant index appears to provide an accurate tool to distinguish normal controls from women with atypical hyperplasia (sensitivity, 0.976; specificity, 0.977). Presently, the Gail model is not appropriate for use in women with LCIS. The potential of an accurate, noninvasive urine test for assessing breast cancer risk with is enormously appealing. Urine tests would be less invasive, less costly than current screening modalities, and easily tolerated and would encourage higher compliance with screening and therapeutic monitoring. Once validated in larger studies, such a test could potentially provide a useful adjunct for breast cancer risk assessment.

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 thank Barbara Yee for helpful advice and Susan Kilroy for technical assistance and the dedicated students who have worked with us: Joyce Alencherril, Amy Curdie, Gina Dapul, Robert Doiron, Emily Edenburn-MacQueen, Megan Fitzgerald, Lilit Garibyan, Rebecca Harrington, Lia Moriguchi, Carolyn Nolan, Elizabeth Palazzo, Shreya Patel, Linden Spital, Perrine Terry, Dorothy Weiss, and Erica Zarolnick.