Proteomic Analysis to Identify Breast Cancer Biomarkers in Nipple Aspirate Fluid Free

Early detection is a major factor contributing to the 3.2% annual decline in breast cancer death rates over the past 5 years (1, 2). Unfortunately, currently available breast cancer screening tools such as mammography and breast examination miss up to 40% of early breast cancers and are least effective in detecting cancer in young women, whose tumors are often more aggressive. An invasive needle or surgical biopsy must be performed when an area of suspicion is identified to confirm, by cytologic or histologic evaluation, the presence of malignant disease. The development of noninvasive techniques that would distinguish between women with and women without breast cancer, as well as between different disease stages, is, therefore, of crucial importance.

Recent advances in comprehensive molecular technologies have allowed the analysis of global gene expression or protein profiles in cancerous versus normal tissues with the goal of identifying RNA or protein markers that are differentially expressed between benign and malignant tissues. One such study (3, 4) analyzed RNA by using serial analysis of gene expression to identify molecular alterations involved in breast cancer progression. The authors concluded that very few genes were up-regulated in all tumors, presumably reflecting the high degree of diversity among tumors at the molecular level. The most dramatic changes occurred at the transition from benign breast tissue to in situ carcinoma, in which many of the highly expressed genes encoded secreted proteins, mainly cytokines and chemokines, implicating abnormal paracrine and autocrine signaling in the initiation of breast cancer.

Other nucleic acid-based studies have examined samples of normal versus cancerous tissues, established cell lines, and samples from before and after treatment with chemotherapeutic agents for differences in expression of a set of >400 genes, by using hierarchical clustering to analyze the results (5, 6). Although the authors report variation among different tumor samples, these studies have shown that cancers can be classified into subgroups based on patterns of gene expression, and that this classification may prove useful for prognosis.

Several proteomic studies have compared normal and cancerous breast cells [reviewed in detail by Hondermarck et al. (7)]. A hierarchical clustering analysis of proteomes of normal and different stages of disease have shown that it is possible to distinguish between normal, benign, and cancerous breast tissues on the basis of the protein profile (8). Page et al. (9) established an extensive map of the normal human luminal and basal (myoepithelial) breast cell proteome as the basis for future comparisons with breast cancer cells. The authors observed 170 differentially displayed protein spots, and identified 51 of them. A subsequent proteomic study of breast ductal carcinoma identified 57 proteins that were differentially expressed based on disease stage, 10 of which were validated by immunohistochemical studies (10). Many were proteins involved in the regulation of intracellular trafficking, cytoskeletal architecture, chaperone function, apoptosis, and genome instability. In a more focused study, the molecular chaperone 14-3-3 was shown to be involved in the transition of breast epithelial cells to neoplasia. By virtue of its role as a tumor suppressor, it was suggested that 14-3-3 may be a useful marker to identify cells that have undergone this transition (11).

Interestingly, many proteins that have been identified by proteomic studies are different from those found by nucleic acid-based studies. This underscores the importance of performing biomarker screens at the protein level and suggests that many of the differences between normal and cancer samples are due to posttranslational modifications such as glycosylation or truncation (12). However, although promising, proteomic analyses of breast tissues, to date, suffer from the fact that the tissues are not homogeneous and more than one cell type is likely to be present in the samples. For this reason some of the above studies were done on cells in culture, although the relevance of the in vitro physiology and biochemistry in these studies to that of the normal or diseased human breast is unclear. Many of the studies that use cells have found few proteins with substantial differential expression. Moreover, studies requiring a large cellular sample require an invasive procedure to obtain the samples for analysis.

Nipple aspiration is a noninvasive, low-cost procedure that provides a relatively small set of breast-specific proteins. Because the proteins are secreted, they represent the final processed form of the protein, which makes proteomic analyses less ambiguous and can provide clues to changes in protein translational rates, posttranslational modification, sequestration, and degradation that lead to disease. Validation of these noninvasive procedures requires the demonstration that biological markers in the fluid correlate with breast tissue pathology and, thus, predict disease. These markers may or may not have been previously identified in cells from breast cancer specimens.

Several studies indicate that nipple aspirate fluid (NAF) contains potentially diagnostic or prognostic markers of breast cancer. A study that used surface enhanced laser desorption ionization (SELDI) analysis identified five differentially expressed proteins in NAF from benign versus cancerous breasts (13), although none of these proteins have been identified. Subsequent studies found that low levels of prostate-specific antigen (PSA) and high levels of insulin-like growth factor binding protein-3 (IGFBP-3) are associated with breast cancer (14), and that high levels in NAF of urinary plasminogen activator (uPA) and its receptor, uPAR, significantly contributed to a model that predicted which women had breast cancer (15).

In this study, we present (a) a partial description of the NAF proteome (41 protein identifications), (b) the identification of three candidate proteins by two-dimensional PAGE and matrix-assisted laser desorption ionization–time-of-flight mass spectrometry (MALDI-TOF MS) with different expression levels based on whether a subject did or did not have breast cancer, and (c) the subsequent validation of these differentially expressed proteins by using an ELISA on a large set of NAF specimens from women with and without breast cancer. Three candidate proteins: gross cystic disease fluid protein (GCDFP)-15 (also know as prolactin-induced protein or BRST2), lipoprotein D [apolipoprotein D (apoD)], and α1-acid glycoprotein (AAG), were differentially expressed in a small set of cancer versus benign samples analyzed by two-dimensional PAGE and subsequently examined by ELISA in a large set of NAF samples comparing women with or without breast cancer, as well as women with in situ (early stage) versus invasive (more advanced stage) disease. We demonstrate that the levels of GCDFP-15 are substantially lower and AAG are substantially higher in samples from women with breast cancer, whereas levels of apoD in NAF were not associated with disease. This stepwise approach to biomarker detection and validation suggests that proteomic analysis of bodily fluids such as NAF are appropriate for biomarker detection, can be readily validated by using high-throughput, quantitative technologies and may prove useful for early breast cancer detection and/or prognosis prediction.

NAF specimens from subjects, 17–83 years old, were collected after Institutional Review Board approval and informed consent were obtained. Some subjects had specimens analyzed from one breast only, whereas others had specimens analyzed from both breasts. The presence or absence of cancer was based on whether or not disease was present in a given breast. Fifty-three specimens were collected from 38 patients without evidence of cancer in the analyzed breast. The remaining 52 specimens were obtained from 51 subjects that had pathologically confirmed ductal carcinoma in situ (DCIS) or invasive breast cancer. For women with breast cancer, all of the specimens were collected before, or in conjunction with, definitive treatment. Ninety percent (94 of 105) of the samples were from Caucasian subjects, 9% from African Americans, and 1% other (Table 1). Forty-eight percent of NAF specimens were from premenopausal subjects. Seventy-three percent of specimens came from subjects who had taken birth control pills, and 33% came from subjects who had taken hormone replacement. The number of live births experienced by participants ranged from 0 to 9, and the number of miscarriages from 0 to 5.

Nipple aspiration was performed as described previously (16). A trained physician or nurse clinician aspirated nipple fluid noninvasively using a modified breast pump. The nipple was cleansed with alcohol. A warm, moist cloth was placed on the breast after the alcohol evaporated. The cloth was removed after 2 minutes and the subject massaged her breast with both hands while the plunger of the syringe was withdrawn to the 7-mL level or until she experienced discomfort. Aspiration was repeated on the opposite breast, if present. Fluid in the form of droplets (1–200 μL) was collected in capillary tubes, and the samples were immediately snap-frozen at −80°C. If keratin plugs rather than NAF were obtained after suctioning was completed, the plugs were removed with an alcohol swab and suctioning was repeated. Occasionally this procedure was repeated two or three times to remove all of the plugs and, routinely, fluid was then obtained.

Samples for two-dimensional PAGEanalysis were thawed by breaking the capillary tube directly into denaturing buffer (7 mol/L urea, 2 mol/L thiourea, 1% (w/v) C7 detergent (Sigma-Aldrich, St. Louis, MO), 40 mmol/L Tris), at a ratio of 80 μL buffer to 1 μL NAF. The proteins were reduced by using 5 mmol/L tri-butylphosphine (TBP) for 1 hour at room temperature and were alkylated with 15 mmol/L iodoacetamide. Pharmalytes isoelectric focusing carrier ampholytes (pH 3–10; Sigma-Aldrich) were added to a final concentration of 1.5%, the samples were centrifuged at 150,000 × g for 1 hour at 4°C in a Beckman TL-100 centrifuge, and the supernatants were kept.

Diluted sample (250 or 400 μg) was used to hydrate 11-cm or 24-cm strips, respectively, overnight at room temperature. First-dimension separation was done by focusing the proteins at a total of 80 K volt-hours, with a 6000 V programmable power supply (Proteome Systems Ltd, Boston MA). Second-dimension separation was performed by PAGE on 8-to-18% gradient gel chips (Proteome Systems Ltd), by using Tris-Tricine buffer according to the manufacturer’s instructions. Gels were stained overnight by using Colloidal Coomassie Brilliant Blue, destained in 1% acetic acid, and scanned.

Electronic images of two-dimensional gels were analyzed by using Phoretix 2D-Advance software (Nonlinear Dynamics, Newcastle, United Kingdom). Protein spots were excised and arrayed into 96-well MultiScreen model R5, glass-filled polypropylene plates (Millipore, Bedford, MA) by using 2-mm diameter pins on the GelPix robotic spot excision station (Genetix, Ltd, Hampshire, United Kingdom). Gel plugs were destained in 3× 200 μL of 50% (v/v) acetonitrile, 50 mmol/L NH₄HCO₃ for 15 minutes at room temperature and were dehydrated in 100% acetonitrile for 5 minutes at room temperature. Proteins were digested with 50 μL of sequencing grade trypsin (Promega, Madison, WI) at 4 μg/mL in 50 mmol/L NH₄HCO₃ at 37°C for 16 hours. Tryptic peptides were extracted from the gel plugs by shaking for 15 minutes in 50 μL of 60% acetonitrile, 1% formic acid, collected into the V-well collection plate under vacuum and concentrated to 5 to 15 μL by centrifugal vacuum evaporation. Tryptic peptide samples (0.5 μL) were applied to a 96 × 2 Teflon MALDI plate by using a Symbiot IV liquid handling station (Applied Biosystems, Inc., Foster City, CA). The samples were mixed with an equal volume of the matrix solution, 10 mg/mL α-cyano-4-hydroxycinnamic acid (Sigma-Aldrich) in 60% acetonitrile, and 0.3% trifluoroacetic acid on an Applied Biosystems Voyager-DE Pro MALDI TOF MS, operated in the positive-ion delayed-extraction reflector mode. Peptides were ionized/desorbed with a 337-nm laser and spectra acquired at 20 kV accelerating potential with optimized settings. The close external calibration method, which used a mixture of standard peptides (Applied Biosystems), provided a mass accuracy of 25 to 50 ppm across the mass range of 600 to 5000 Da. Peptide spectra were automatically processed for baseline correction, noise removal, and peak deisotoping, and were analyzed with Protein Prospector.⁴ Search criteria required the match of at least four peptides, with a mass error of less than 50 ppm for a protein assignment. The data were manually reexamined to insure maximum peptide coverage for the identified proteins. When there were isoforms of a protein (i.e., train of spots), all of the spots were analyzed, but only one entry was made in Table 2.

Frozen NAF samples were resuspended by crushing the tube into phosphate-buffered saline (PBS; pH 7.2) containing a protease inhibitor mixture to prevent proteolysis [1:100 dilution of 100× protease inhibitor cocktail: 20 mmol/L 4-(2-aminoethyl)benzenesulfonyl fluoride (AEBSF), 100 μg/mL peptstatin A, 10 μg/mL leupeptin)]. A constant ratio of 70 μL of buffer to 1 μL of NAF was maintained in all of the samples to keep the protein concentration comparable between samples, and to insure complete solubility of larger volume samples, because NAF samples are rich in protein (most samples contain between 50 and 300 mg protein/mL) and viscous.

ELISAs were performed by directly adsorbing the NAF samples onto 96-well plates. Five microliters of diluted NAF containing 5 to 20 μg of protein were mixed with 45 μL of coating buffer (10 mmol/L NaCl, 50 mmol/L NaPO₄, pH 7.5) in the first well, and subjected to 11× 2-fold serial dilutions into 25 μL of coating buffer. The plates were incubated overnight at 4°C to let the proteins adsorb to the plate. Excess solution was decanted, and the plates were blocked with coating buffer containing 0.5% bovine serum albumin for 2 to 4 hours at 4°C. Excess blocking buffer was decanted and 25 μL of prediluted primary monoclonal antibody in PBS were added to each well as follows: for apoD (Signet Laboratories, Dedham, MA, cat. no. 9780-02), 1:200 dilution; for GCDFP-15 (Signet, cat. no. 611-13), 1:20 dilution; and for AAG (BIODESIGN, Saco, ME, cat. no. H45190M), 1:100 dilution. The plates were incubated overnight at 4°C. Excess unbound primary antibody was removed, and the plates were washed eight times under running distilled water. Twenty-five microliters of secondary antibody, goat antimouse immunoglobulin (“Linking reagent” no. 4, Signet Labs) at 1:30 dilution in PBS were added to each well, and the plates were incubated for 2 hours at room temperature. Unbound antibody was removed, the plates were washed as above, and were incubated for 2 hours at room temperature in 25 μL of mouse antigoat tertiary monoclonal antibody conjugated to horseradish peroxidase (“Labeling reagent” no. 5, Signet Laboratories) at 1:30 dilution in PBS. The plates were washed and developed with 100 μL of developing solution [100 mmol/L sodium citrate (pH 5.0), 10 mg/mL o-phenylenediamine dihydrochloride (OPD; Sigma), and 0.006% H₂O₂] for 15 minutes at room temperature. The reaction was stopped by adding 100 μL of 2.5 mol/L H₂SO₄, and the plates were read in a microtiter plate reader at 490 nm. Antigen concentration in the sample was calculated as the reciprocal of the dilution at the 50% point of the maximum of the linear range of the dilution curve per milligram of total NAF protein.

Protein concentration was determined by using the bicinchoninic acid assay (BCA, Pierce Biotechnology, Rockford, IL), with 0.1% bovine serum albumin as a standard in 96-well plates.

All of the data are presented as median and range (minimum to maximum). Comparisons between groups (cancer versus no cancer) were conducted by using the Mann–Whitney Rank Sum Test (underlying conditions for parametric tests were not met). Comparisons of more than two groups (e.g., by disease stage-benign, DCIS, invasive cancer) were conducted by using the Kruskal–Wallis One-Way Analysis of Variance on Ranks (underlying conditions for parametric tests were not met). Analyses of categorical data were conducted by using the χ² test. Pearson Product Moment Correlation was used for correlation analyses. In all instances, significance was set at P < 0.05. All of the data were analyzed with SigmaStat for Windows, Version 2.03S (SPSS, Inc., Chicago, IL).

We tested 105 NAF samples, 53 from normal breasts and 52 from breasts with DCIS or invasive cancer. Among the clinical variables compared (age, number of live births, number of miscarriages, ever use of birth control pills or hormone replacement therapy, menopausal status, and race), two variables, age and menopausal status, were related to cancer status (Table 1). Samples from women with cancer tended to be older than their noncancer counterparts and were postmenopausal more often than their noncancer counterparts.

To optimize protein identification, two-dimensional PAGE was performed on NAF from one subject without breast cancer on two gels with different pH ranges: 3–10 and 4–7 (Fig. 1). Spots were excised from each of the two gels and were subjected to MALDI-TOF analysis. Ninety-one spots were excised from the pH 3–10 gel. Of these, 69 (73%) were identified, yielding 35 different protein identifications. Of the 22 spots that were not identified, 9 spots gave good spectra but no protein identification, and 13 spots resulted in poor spectra. A total of 94 spots were excised from the pH 4–7 gel. Of these, 53 (56%) spots were identified, yielding 28 different protein identifications. Thirty spots gave no spectra, and 11 spots gave reasonable spectra and no identification. Combining the data from the two gels led to the identification of 41 unique proteins. Positive identification was assigned when there were at least four peptides matched and the percentage linear coverage was over 15%. The list of proteins and detail about their function are presented in Table 2. Several proteins have been referred to by different names in different studies; these are shown under synonyms in Table 2.

We performed two-dimensional PAGE on 20 NAF samples (10 from subjects with, and 10 from subjects without breast cancer). Despite variation between different individuals, both in the normal and the diseased samples, we were able to pick 10 protein spots that were visually differentially expressed in cancer patients. These proteins were analyzed by MALDI-TOF MS and were identified as prolactin-induced protein; apoD; α1-acid glycoprotein; IgJ; complement C4A (two spots); α2 glycoprotein 1 and Igα (2 spots). Three of these proteins (Fig. 2)—spot 1, prolactin-induced protein (GCDFP-15 or BRST2); spot 2, apoD; and spot 3, AAG, which were up-regulated in the NAF from a cancer-containing breast compared with a normal breast from the same woman, as well as from at least two other breasts with cancer—were chosen for further analysis.

Because the preliminary association of these markers with breast cancer was made in a small sample set, the above proteins were next investigated by ELISA in 105 NAF samples (53 from benign breasts and 52 from diseased breasts). The concentration of each of these proteins relative to the total protein concentration in the sample was determined. We used a direct ELISA method on serially diluted NAF extracts. In the absence of commercial kits, this technique is a well-established method to quantitate the level of antigen in a sample. Because we used the same concentration of antibody throughout the experiment in all of the wells and we scored the results at the point at which 50% if the antibody was bound, we were able to determine how much each NAF sample needed to be diluted to bind 50% of the antibody. Table 3 shows the correlation of these proteins with the presence of breast cancer and menopausal status.

ELISAs revealed that median GCDFP-15 expression was 6-fold lower in samples from breasts with cancer than in those without. This lower expression in breasts with cancer than in samples from benign breasts was significantly different, both when considering all subjects (P < 0.001) and premenopausal subjects (P = 0.011). GCDFP-15 levels were not significantly different in NAF samples from postmenopausal women.

Apolipoprotein D levels were not associated with the presence of breast cancer, whether considering all subjects or subjects divided by menopausal status.

AAG levels in NAF samples were higher in women with breast cancer, when considering both all subjects (P = 0.001) and premenopausal subjects (P = 0.002). The levels of AAG were not related to breast cancer in postmenopausal women.

We wished to determine whether GCDFP-15, apoD, or AAG levels in NAF were related to disease stage (no cancer, DCIS, or invasive breast cancer). We found that the lowest levels of GCDFP-15 (P = 0.003) and the highest levels of AAG (P < 0.001) were in breasts from women with DCIS (Table 4). apoD levels were not significantly different based on disease stage. Women with invasive cancer had higher levels of AAG than breasts with benign disease but lower levels than in breasts with DCIS.

GCDFP-15, apoD, and AAG expression were measured in both pre- and postmenopausal women. The data presented in Table 5 includes all women, regardless of breast cancer status, as well as controlling for cancer status. GCDFP-15 levels were higher in the breasts of premenopausal women, both overall (P = 0.004) and in breasts without cancer (P = 0.049). apoD expression was not significantly influenced by menopausal status. AAG levels were significantly higher in post- than in premenopausal women, both overall (P = 0.019) and in breasts without cancer (P = 0.009).

We believe that proteomic analysis of NAF is more advantageous than proteomic analysis of cells from breast cancer tissue because the protein comparison is less complex. This argument extends to a recent study of the interstitial fluid of breast tumors, in which more than 1000 proteins were identified (17). Moreover, the secreted proteins in NAF are fully processed, glycosylated, and truncated, and are not influenced by transient protein modifications (such as phosphorylation or changes in redox state, which influence the position of a protein spot on a two-dimensional gel) that affect proteins within the cell. Two-dimensional PAGE was chosen to screen NAF because it provides a convenient and rapid method for protein identification (based on MALDI-TOF MS) that then allows subsequent high-throughput quantitative ELISA studies of the candidate proteins. Although two-dimensional PAGE has limitations in protein separation and sensitivity, it is an excellent choice for initial investigations and, in the present study, posed no problems because of the limited number of proteins in NAF samples. Indeed, the protein makeup of NAF is quite simple, allowing us to identify many of its constituent proteins, as well as easily identify a number of proteins that are differentially expressed. We predict that other proteomic techniques [e.g., isotope-coded affinity tags (ICATs)] will identify additional markers. Nonetheless, the present study strongly supports the idea that NAF will be useful in the diagnosis and prognosis of breast cancer.

Proteomic analysis of a limited number of benign and breast cancer NAF samples suggested that there is protein expression variability between women. Nevertheless, proteomic analysis by two-dimensional PAGE identified candidate proteins that are differentially displayed between benign and cancerous samples, and subsequent quantitative analysis with ELISA in a large sample of women showed that the expression of two of the markers (GCDFP-15 and AAG) were, and one (apoD) was not, altered in breast cancer. This underscores the need to validate initial observations with larger sample sets.

Using two-dimensional PAGE and MALDI-TOF MS, we identified 41 distinct proteins in the NAF proteome. We used Colloidal Coomassie Brilliant Blue to identify proteins, and we observed by using parallel gels that commercial silver stains did not appreciably improve protein detection (data not shown). Our attempts to remove albumin and immunoglobulins by using commercial kits also were not very effective. Thus, it is possible that we missed the subset of proteins that were masked by these highly abundant proteins. Additionally, the identification of highly glycosylated proteins may have been limited because, at times, there are very few unglycosylated peptides after trypsin digestion. Thus, although not all NAF proteins were identified by our approach, we think that our approach to defining the NAF proteome provides unique and important information to better characterize this important biological fluid.

A recent study by Varnum et al. (18) analyzed the NAF proteome by using liquid chromatography. More than 60 proteins were identified, many of the proteins the same as those identified in the present study, but a substantial subset of proteins (21 in our study, 35 in the other) are unique to each study. It should be noted that in the Varnum study, pooled NAF samples were used, whereas in our study individual NAF samples were run on two-dimensional gels. Clearly, both studies should be considered when assessing the NAF proteome.

Many of the proteins that we identified in the NAF proteome could potentially be markers of disease, including ras-related protein; metastasis-associated protein; BCL2, which has been implicated in the suppression of cell death; CD5, which is reported to play a role in the inhibition of apoptosis; retinol-binding protein, which has recently been shown to suppress breast cancer cell survival and has been shown to be down-regulated in a subset of breast cancer; clusterin, which has been associated with cell death and apoptosis; and transferin, which has been assigned a role in cell proliferation. (For additional references to these proteins and their functions, see http://us.expasy.org/sprot, and links to related references.)⁵ Some of the NAF proteins are known blood proteins. Whether these are blood contaminant proteins or true NAF proteins will require further investigation.

Three proteins, GCDFP-15, AAG, and apoD, were studied in detail. These proteins were chosen because they were all differentially expressed in a small number of cancer versus benign samples, and antibodies were commercially available. Additionally, there was previously existing literature connecting these proteins to breast cancer (19, 20, 21), confirming the importance of these proteins in disease prediction.

We found that GCDFP-15 levels in NAF from cancerous samples are substantially lower than in samples from breasts without cancer, findings consistent with a report in breast tissue (22). Petrakis et al. (23) demonstrated that the levels of GCDFP-15 in NAF are lower in Asian than in non-Asian women and that, under the influence of soy ingestion (24), GCDFP-15 levels are higher in NAF from normal than from hyperplastic breasts. NAF levels of GCDGP-15 analyzed by one-dimensional PAGE (25) were higher in women without breast cancer. Our findings confirm and extend these findings in a large cohort of women, and demonstrate that this association is limited to premenopausal women.

AAG is an acute phase reactant that is known to play an immunomodulatory role (26). In patients with chronic myelogenous leukemia in blastic phase, AAG levels reflect pharmacological resistance to chemotherapy (27). We observed that AAG expression was significantly higher in samples from pre- but not from postmenopausal women with breast cancer. That GCDFP-15 and AAG were substantially related to breast cancer in pre- but not in postmenopausal women suggests that these proteins may be influenced by ovarian hormones. AAG has been observed to bind tamoxifen, a known estrogen receptor modulator (28). In postmenopausal women, unopposed estrogen use was associated with modestly lower levels of AAG (29). This is consistent with our finding of higher levels in postmenopausal women without breast cancer, most of whom have very low levels of circulating estradiol, and with the fact that oral estrogen treatment lowers AAG levels in the blood (30). Positive immunoreactivity for GCDFP-15 in breast tumors was found to be highly dependent on androgen receptor status, but unrelated to estrogen or progesterone receptor (31). The identification of biomarkers of breast cancer in premenopausal women is of great importance because, compared with cancers in postmenopausal women, breast cancers that develop in these women are more likely to have a genetic cause, are less likely to respond to hormonal therapy, and are more likely to be aggressive (32). Alterations in these proteins in premenopausal women, if confirmed, may lead to therapies to treat hormone-unresponsive cancers, those most likely to lead to death.

Apolipoprotein D was first identified as a component of breast cyst fluid and was called GCDFP-24 because of its size (33). It is present in benign and malignant human breast tissues and is a major protein component in cyst fluid from women with human gross cystic breast disease (34). A prior report that evaluated NAF by using immunoblotting suggests that levels of apoD are higher in women without than in women with breast cancer (35). Our analysis of apoD in NAF by ELISA failed to demonstrate a significant relationship to breast cancer.

In summary, we screened cancerous and benign NAF samples by two-dimensional PAGE both to determine proteins present in the NAF proteome and to identify proteins that were differentially expressed between women with and without breast cancer. The proteome as identified by two-dimensional PAGE differs from, and is complementary to, a prior report (17) on the proteome with liquid chromatography, because each technique identified both a similar and a unique subset of proteins. No attempt was made in the NAF study that used liquid chromatography to quantitate and compare NAF protein levels in women with and without breast cancer. Validation in a large cohort of subjects by ELISA of the three candidate proteins identified by two-dimensional PAGE demonstrated that two of the three were linked to breast cancer, one directly (AAG) and one inversely (GCDFP-15), and that the associations were seen only in premenopausal women. Our findings extend previous observations linking GCDFP-15 and breast cancer, demonstrate a link between AAG and breast cancer, and point to a hormonal influence on the expression of these markers. We believe that a combination of proteomic approaches will result in a comprehensive NAF proteome map. Because of the variability of expression of biomarkers within the population, we suggest that all of the identified proteins be measured in a large number of NAF samples by using ELISAs to further correlate their levels with various disease stages. For proteins for which commercial antibodies are not available, antibodies can be generated against synthetic peptides derived from the protein sequence. In this way a noninvasive predictive breast cancer protein signature will ultimately be developed.

We thank the Proteomics Center at the University of Missouri-Columbia for their assistance with protein identification, and Arvan Chan for help with the preparation of the figures.