These subtype classifications have been determined in conjunction with many gene microarray expression studies for breast cancers, particularly as applied to improving prognostic and therapeutic prediction (Kim and Paik, 2010; Turaga et al., 2010). From these approaches, multiple prognostic assays to determine risk of recurrence or treatment responsiveness, typified by Oncotype DX (Paik et al., 2005) (Genomic Health, Inc., Redwood City, CA), Mammaprint (van de Vijver et al., 2002) (Agendia Inc., Irvine, CA), and other genetic assays (Turaga et al., 2010), have been evolving into wider clinical use. Larger clinical trials are ongoing to determine the power of these gene expression signatures in reclassifying patients after conventional risk classification (Turaga et al., 2010). The cumulative genetic-based research into the different breast cancer subtypes, evaluation of additional associated immunohistochemical targets (reviewed in Choo and Nielsen, 2010), continued commercial assay development, and larger clinical studies assessing risk classification and treatment responsiveness issues will continue to accelerate all aspects of related breast cancer research in these areas.

One model serum collection cohort that could be used to design protein-based diagnostic assays is as follows. For the past 5 years, we have collected over 1,000 serum samples at two sites in a study design focused on women given a BIRADS 4 mammogram report. The Breast Imaging-Reporting and Data System (BIRADS) is used by radiologists to make treatment or follow-up recommendations by categorizing mammographic findings based on the likelihood of malignancy (Obenauer et al., 2005) across seven categories ranging from 0 (“incomplete assessment”) to 6 (“confirmed malignant”). A BIRADS 4 lesion is categorized as a “suspicious abnormality,” even though the majority (80%) will be benign conditions. Because a malignancy cannot be ruled out, a biopsy will be performed, even for those lesions further stratified into BIRADS 4a, 4b, and 4c (Lazarus et al., 2006), with a 4a lesion designation as a “low level of suspicion.” Thus, each year, a large number of women undergo a biopsy that will yield a benign diagnosis. Beyond pathological cytology, biomarkers of benign disease from any sample type, genetic, metabolite or protein, are lacking.

Our study was designed to collect prebiopsy serum from women with a BIRADS 4 mammogram who were about to have a breast biopsy procedure. The ultimate goal is to develop a proteomic-based blood test that would be used prior to biopsy that could better discriminate those BIRADS 4 designated women that are more likely to have benign disease, or confirm cancer suspicions, and therefore better justify a tissue biopsy decision. The specifics regarding sample collection have been reported (Shin et al., 2006), and are summarized as follows. Patients underwent either stereotactic or ultrasound-guided core biopsy, or needle localized excisional breast biopsy. After informed consent, which included a short questionnaire for assessment of each subject's risk factors for breast cancer, serum was collected immediately prior to the biopsy procedure. The questionnaire data included age, race, BMI, age at menarche, parity, age at first live birth, and age at menopause. Each serum specimen was assigned a diagnosis retrospectively after final pathology review. As would be expected, final pathologies for the benign cohort included fibroadenomas, fibrocystic disease, papillomas, atypical ductal hyperplasia without associated carcinoma or carcinoma in situ, lobular carcinoma in situ, and benign breast tissue (Shin et al., 2006). The cancer cohort included ductal carcinoma in situ and primarily early stage invasive ductal and lobular carcinomas (Shin et al., 2006). An initial cohort was used for extensive MALDI-TOF serum protein profiling following weak cation exchange affinity bead enrichment. Expression of a panel of 12 low molecular weight peptides could be used with a genetic algorithm to correctly classify 88% of the cancer samples and 86% of the benign disease samples (Shin et al., 2006). Despite evidence of protein differences between the two disease classifications, use of this single affinity method alone has proven to be not sensitive enough to routinely distinguish cancer from benign disease classifications.

The advantages of using this type of serum cohort center on the study design, as several known biases related to sample collection are minimized. All of the samples are collected in the same way in a similar clinic environment, from patients with the same BIRADS 4 designation. Because it is done just prior to biopsy, there is no diagnosis at the time of collection, so psychological and stress issues for each subject are similar. All associated risk factor information for breast cancer is collected, and can readily be combined with subsequent pathology and genetic properties associated with each sample. The ease and breadth of collection also facilitate their use in any developed high throughput detection platform. The clinical decisions made that brought each subject to the point of biopsy are also very similar. Because the BIRADS 4 diagnosis is based on a radiological test, samples are collected from all possible types of malignant and benign breast diseases. The collection protocol itself is potentially transportable to any breast cancer-related surgical facility that routinely performs breast biopsy procedures. Overall, collection protocols can be standardized and readily integrated into current clinical decision-making workflow associated with a benign or malignant diagnosis.

A wealth of new strategies targeting the characterization of serum glycoproteins and their carbohydrate constituents are emerging (Drake et al., 2006; Miura et al., 2008; Wada et al., 2007; Wuhrer et al., 2006). Structural elucidation of glycoproteins has long relied upon the use of chemical derivation and cleavage reactions, and differential capture by lectins, a class of proteins known to bind specific oligosaccharide moieties (Drake et al., 2006; Hirabayashi, 2004). The most common approaches for characterizing the protein constituents of N-linked glycoproteins is to use some combination of the following: peptides are generated by digestion with trypsin, the glycopeptides are isolated with one or more lectins or hydrazine linked to a support resin, the bound peptides are treated with protein N-Glycanase F (PNGaseF) to release glycans, then sequencing of the peptides is done by tandem mass spectrometry (Liu et al., 2005; Wang et al., 2006; Zhang et al., 2003; Zielinska et al., 2010). New strategies that probe different lectins bound on multiple array platforms are also emerging (Chen and Haab, 2009; Kuno et al., 2008; Mao et al., 2004; Zhao et al., 2007).

Because of the abundance of glycoproteins in serum, and the relative ease of collecting large serum sets associated with breast cancers, we propose that glycoproteomic targeted strategies could be the most effective approach to identify diagnostic biomarkers. As illustrated in the next two figures, use of lectins address some of the previously stated challenges in performing serum proteomic analyses. In Figure 1, a broad specificity lectin that binds most glycoproteins due to its affinity for mannose residues, Concanavalin A (Con A), was used with and without upfront albumin depletion strategies to isolate serum glycoproteins. Depletion of albumin by either an antibody column or affinity column prior is shown in the left panel of Figure 1, and these approaches are effective at reducing the levels of the abundant 65-kDa serum albumin. Use of just Con A alone with human serum resulted in significant enrichment of glycoproteins, and minimal amounts of bound albumin. Depletion of albumin prior to Con A binding yielded similar results. Thus, lectins like Con A, and both fucose targeting lectins used in Figure 2, enrich for serum glycoproteins while essentially minimizing the amount of albumin carried into the next phase of proteomic analyses. In our experience, many lectins have this low affinity for serum albumin binding, but not all of them have this property, which must be determined empirically.

In Figure 2, pooled sera (n=10 per groups) representative of healthy clinic control subjects, and sera from ductal carcinoma in situ, stage 1, stage 2, and stage 3 breast cancer diagnoses (Laronga et al., 2004), were incubated with two different fucose binding lectins. As seen in the gel separations of the bound and eluted proteins, there is a clear enrichment of different serum glycoproteins associated with different stages of disease, particularly for the more severe stage 3 samples. We have used this type of approach with multiple lectins like Con A, the two fucose lectins, and sialic acid binding lectins, followed by protease digestion and determination of protein identities by LC-electrospray mass spectrometry (Drake et al., 2006). A cumulative list of target serum glycoproteins that we have determined to be differentially expressed in breast cancer-related serum across many conditions is provided in Table 1. Not surprisingly, these proteins are all relatively abundant serum proteins and are generally associated with the stress or “acute-phase” response to disease. From what is known about the individual proteins, these are not necessarily breast cancer- or even cancer-specific proteins and may be more associated with the immune system. However, what has not been determined or most of these proteins is whether there are subpopulations of these abundant glycoproteins that carry cancer specific glycan signatures. This theme is slowly emerging, as for example cancer-specific glycoforms of serum haptoglobin and alpha-1-antitrypsin have been reported (Abbott et al., 2009; Comunale et al., 2010; Fujimari et al., 2008; Zeng et al., 2010). In particular, for our BIRADS4 serum cohort that contains many examples of benign breast disease and other patient demographics (race, BMI, age), targeting of individual serum glycoproteins listed in Table 1 for site-specific glycan analysis could yield distinct disease state biomarker candidates. Specific proteotypic peptides of most of these glycoproteins listed in Table 1 have also been reported in a multiple reaction monitoring (MRM) panel of 45 plasma proteins (Kuzyk et al., 2009). The relative abundance of these serum glycorproteins is also an advantage for mass spectrometry based assays, particularly those involving direct glycopeptide-based strategies using a triple quadropole mass spectrometer, which we have reported previously (White et al., 2009). Continued improvements in mass spectrometry instrumentation, and in software applications like SimGlycan (Premier Biosoft, Palo Alto, CA) for analysis of complex glycopeptide fragmentation patterns, will facilitate the characterization of these potential serum glycoprotein biomarker candidates. Using this type of mass spectrometry platform and software capability, we are currently exploring development of specific MRM assays targeting different glycopeptide species reflective of a given disease state.

Although some of these limitations can be overcome, the amounts of breast tissue available for in-depth and large-scale proteomic analyses is frequently limiting. A new proteomic approach that is relatively conservative for sample sparing is MALDI-based mass spectrometry imaging approaches (Cornett et al., 2006; Schwamborn and Caprioli, 2010). In this approach, breast cancer tissues mounted on slides are directly analyzed by MALDI-TOF mass spectrometry. Two recent papers highlight its application in relation to genetic and histologic approaches for breast cancers. In one study (Rauser et al., 2010), the MALDI tissue profiles of Her2+ versus Her2− breast cancers were compared. A definitive marker of 8,404 m/z was detected in most Her2+ samples, but not in the most Her2− tissues. It was identified as CRIP1 (cysteine-rich intestinal protein 1), and mRNA levels of CRIP1 correlated with Her2 mRNA levels (Rauser et al., 2010). In another study (Bauer et al., 2010), MALDI imaging was applied to pretreatment biopsy samples obtained from a cohort of women treated by taxane/radiation and surgery of high risk breast cancers. The pretreatment biopsies were classified based on complete pathologic response following treatments versus those with residual disease. It was found that expression of alpha-defensins in the prebiopsy tissues as detected by MALDI imaging correlated with taxane responsiveness (Bauer et al., 2010).

An example MALDI imaging workflow as applied to a breast tissue containing regions of invasive ductal carcinoma cells is described as follows, and a representative image is shown in Figure 3. An automated matrix sprayer was used to deposit an even layer of micron-sized droplets of matrix directly across the tissue containing normal epithelial cells, stromal areas, and invasive cancer cells. The spectra generated from each defined laser shot area were compared to each other, as well as the underlying pathology of the cell type present at each spot. Instead of being represented as peak heights, an individual m/z value is assigned a pixel intensity and color. Therefore, the more abundant a protein detected at a given m/z is in a given region, the brighter the pixel color will be in the image at that location. An example of the expression of a tumor associated peptide mass at 4,588 m/z in the invasive ductal carcinoma sample regions is shown in Figure 3B. In contrast, the expression of an abundant peptide not associated with the invasive ductal region is shown in blue pixel with the invasive ductal associated m/z peptide ion in red pixel (Fig. 3C).

As presented, we believe that targeting of serum (or plasma) glycoproteins and use of tissue for MALDI mass spectrometry imaging studies. Knowing the limitations of both samples, study design issues should be addressed at the beginning of the process. The many lessons learned from the development of the genetic assays currently being used clinically are also applicable to the proteomic study design strategy. The goal of the proteomic assays should be to develop biomarkers that complement the existing assays. For these reasons, tissue and serum specimens are most likely to provide the best sources for developing proteomic assays, as these can be collected to best reflect the many variables associated with breast cancers. For both approaches, there are multiple studies already published that provide optimism for pursuit of this strategy. However, there remains an inherent tumor biology limitation for both genetic and protein biomarker-centric approaches. Just because mRNA or protein can be detected, this does not necessarily inform on the cellular and physiologic function of the biomarker, nor whether it could be a potential therapeutic target. This type of functional analysis for putative biomarkers will continue to need to be assessed in cell line or animal model systems that are amenable to direct manipulation and experimentation.

The serum and tissue systems are also not mutually exclusive to each approach, as clearly tissue-derived glycoproteins can be isolated for similar types of characterization. Identification of changes in glycan constituents for either serum or tissue glycoproteins also provides structural biology data that can be further evaluated for gene expression changes of glycosyltransferases or glycosidases involved in the biosynthesis of the glycoconjugate. The glycoproteins listed in Table 1 are all relatively high abundance serum constituents, which makes them attractive targets for lectin, tissue, or antibody array platforms to provide clinical throughput capabilities. Analogous to the protein MRM assays described for plasma proteins (Kuzyk et al., 2009), the abundance of the serum glycoproteins makes them candidates for developing glycopeptide specific MRM panels, which is an ongoing project in our laboratories. These same strategies could also be applied be applied to tissue glycoproteins, or those identified in ductal fluids like NAF. Conversely, we are pursuing the design of antibodies or small molecules targeting the many known carbohydrate antigens associated with cancers (i.e., CA 15.3, CA 19.9, sialyl Lewis X) linked with a MALDI-reactive reporter group (Thiery et al., 2008) to identify glycoprotein targets by MALDI imaging analysis. MALDI imaging approaches can also target differential expression of lipids and, distribution of cellular or drug metabolites (Cornett et al., 2008).