Patients and samples
After receiving Institutional Review Board approval, we obtained informed consent, prospectively enrolled participants and collected 79 ND samples in Grand Forks, North Dakota, at the University of North Dakota, in Columbia, MO, at the Ellis Fischel Cancer Center and in London, UK, at the Royal Marsden Cancer Center. Participants were enrolled from January 2008 to February 2010. Details of TF sample collection and analysis were previously reported [9]. Among the 124 ND samples collected in the TF report there was sufficient remaining NAF to analyze both uPA and PAI-1 in 24 and uPA alone in an additional 11 samples. Criteria for enrollment were the same in the current as in the TF report. A subject was classified as postmenopausal if at least one year had passed without a menstrual period or she had undergone bilateral oophorectomy prior to enrollment. Women who had undergone hysterectomy without bilateral oophorectomy were considered postmenopausal if they were over 50 years old. If follicle stimulating hormone (FSH) levels were available, a level of 34 mIU/mL or greater was used to classify women as postmenopausal. All ND samples were collected prior to excisional biopsy or mastectomy. Comparisons of uPA, PAI-1 and TF were based on the histopathologic findings in the clinical report. For the purposes of this report, we define "cancer" as pathologic evidence of either ductal carcinoma in situ (DCIS) or invasive cancer; "no cancer" as histopathology which was normal, usual hyperplasia, and/or atypical hyperplasia; "benign" as histopathology containing normal findings and/or usual hyperplasia; and "abnormal" as histopatology demonstrating atypical hyperplasia, DCIS, and/or invasive cancer.
Sample collection
ND (1-10 μL) was obtained from the breast with a lesion prior to surgery. Lesions included women with 1) pathologic nipple discharge (PND); 2) a suspicious lesion identified on imaging, be it mammogram, ultrasound or breast magnetic resonance imaging; and/or 3) a palpable lesion that was not a simple cyst. Samples were collected as described previously [11]. Briefly, after informed consent was obtained, ND fluid was aspirated using a breast pump (NAF) or collected after the participant massaged her breast (PND). Samples were collected into capillary tubes, the volume of NAF measured and stored at -80°C until use.
Preparation
The portion of the capillary tube containing the sample was introduced into a 1.7 mL Eppendorf tube and 100 μL of a 0.1 mol/L solution of sodium bicarbonate (pH 7.8) added. The capillary tube was then crushed with a glass rod and the mixture vortexed to disperse the sample. The crushed capillary tube was left in the bicarbonate buffer overnight at 4°C, and the mixture then centrifuged (14,000 g, 5 min) and the supernatant used without further dilution.
Analysis
uPA and PAI-1. ELISA kits for uPA and PAI-1 were obtained from American Diagnostica, Inc. (Greenwich, CT). Levels of these two markers in ND samples were determined according to the manufacturer's instructions. Briefly, 100 uL of standards, samples and blanks were pipetted into microplate wells coated with monoclonal antibodies respectively specific for uPA and PAI-1 and incubated overnight at 4°C. After washing (× 4), enzyme-linked antibodies specific for each analyte were added to the wells and incubated for 1 h at room temperature. The wells were washed again, diluted enzyme conjugate (streptavidin conjugated horseradish peroxidase) was pipetted into the wells, incubated (1 h, RT), then washed again. Substrate reagent was added to each well, followed by a stop solution (0.5 M sulfuric acid). Absorbance was measured with a microplate reader. Detection limits were 10 pg/mL for uPA and 50 pg/mL for PAI-1.
TF. The evaluation of this carbohydrate was previously reported [9].
Statistical analysis
Biomarker levels for uPA and PAI-1 in the ND samples were heavily skewed and not normally distributed. We therefore described and analyzed biomarker levels using medians and log-transformed (Log10) means to achieve normal distributions. We performed a Shapiro-Wilk test to determine the goodness-of-fit of the data to a normal distribution. None of the log-transformed variables were significantly different from a normal distribution. Unpaired t-tests were calculated to determine significant differences between log10 means of the biomarkers. Logistic regression analyses are based on log10 transformed data. Three logistic regression models were used to examine the effects of uPA and PAI-1 in predicting (a) cancer vs. no cancer diagnosis; (b) cancer vs. benign diagnosis; and (c) abnormal vs. benign diagnosis. Various demographic factors, include age, menopausal status, and hormone replacement, were included in the logistic models as covariates. Because menopausal status was a near significant predictor in all three overall logistic models, follow-up logistic models were conducted separately for pre- and for post-menopausal women, controlling for age. Receiver operating characteristic (ROC) curves were calculated based on the logistic regression results for the premenopausal women's data, and area under the curve (AUC) values are presented. The optimal subset of all variants, uPA, PAI-1, TF, Tn, and covariates was selected to achieve the best AUC value.