Multiplexed Cell Signaling Analysis of Human Breast Cancer Applications for Personalized Therapy Julia D. Wulfkuhle,*,† Runa Speer,‡,§ Mariaelena Pierobon,† Julie Laird,4 Virginia Espina,† Jianghong Deng,† Enzo Mammano,⊥ Sherry X. Yang,# Sandra M. Swain,# Donato Nitti,⊥ Laura J. Esserman,4 Claudio Belluco,∇ Lance A. Liotta,† and Emanuel F. Petricoin III*,† Center for Applied Proteomics and Molecular Medicine, George Mason University, Manassas, Virginia 20110, Laboratory of Pathology, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, Maryland 20892, Department of Obstetrics and Gynecology, University of Tübingen, 72076 Tübingen, Germany, Department of Surgery, University of California San Francisco Medical School, San Francisco, California 94115, Department of Oncological and Surgical Sciences, Unversity of Padua, 35128 Padua, Italy, Cancer Therapeutics Branch, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, Maryland 20892, and CRO-IRCCS, National Cancer Institute, 33081 Aviano, Italy Received December 4, 2007
Phosphoprotein driven cellular signaling events represent most of the new molecular targets for cancer treatment. Application of reverse-phase protein microarray technology for the study of ongoing signaling activity within breast tumor specimens holds great potential for elucidating and profiling signaling activity in real-time for patient-tailored therapy. Analysis of laser capture microdissection primary human breast tumors and metastatic lesions reveals pathway specific profiles and a new way to classify cancer based on functional signaling portraits. Moreover, the data demonstrate the requirement of laser capture microdissection for analysis and reveal the metastasis-specific changes that occur within a new microenvironment. Analysis of biopsy material from clinical trials for targeted therapeutics demonstrates the feasibility and utility of comprehensive signal pathway activation profiling for molecular analysis. Keywords: Breast cancer • Proteomics • Cell signaling • Phosphoproteomics • Tailored therapy • Protein microarray
Introduction While individualized treatments have been used in medicine for years,1 development and implementation of new molecularly targeted agents have generated a need to more precisely define and identify those patients who will derive the most benefit from these new drugs. Current pathological parameters for breast cancer including tumor size, degree of tumor cell differentiation, presence or absence of metastases, cytogenetic analysis, and immunohistochemical classification of proteins such as HER2/neu and estrogen and progesterone receptor (ER and PR) status play an important role in therapeutic decisionmaking; however, they do not truly begin to address the * To whom correspondence should be addressed. Emanuel F. Petricoin III, Ph.D., Center for Applied Proteomics and Molecular Medicine, George Mason University, 10900 University Boulevard MS 4E3, Manassas, VA 20110. E-mail:
[email protected]. Julia D. Wulfkuhle, Ph.D., Center for Applied Proteomics and Molecular Medicine, George Mason University, 10900 University Boulevard, MS 1A9 Manassas, VA 20110. Phone: 703-993-4114. E-mail:
[email protected]. † George Mason University. ‡ Laboratory of Pathology, Center for Cancer Research, National Cancer Institute, NIH. § University of Tübingen. 4 University of California San Francisco Medical School. ⊥ Unversity of Padua. # Cancer Therapeutics Branch, Center for Cancer Research, National Cancer Institute, NIH. ∇ CRO-IRCCS.
1508 Journal of Proteome Research 2008, 7, 1508–1517 Published on Web 02/08/2008
complexity and heterogeneity of individual tumors that can lead to success or failure of a targeted therapeutic agent. The past decade has seen an explosion in the use of gene expression profiling for a new classification schema for neoplasia.2–7 While this type of analysis has generated many new hypotheses regarding tumorigenesis, and produced a suite of potentially important prognostic and predictive signatures,6 the use of gene expression analysis to direct therapeutic decision-making has some technical limitations. First, gene expression often does not correlate with protein expression or the functional (e.g., phosphorylated) forms of the encoded proteins.8–10 Second, because signal transduction is a posttranslationally driven process, it is necessary to directly investigate the protein-driven signaling cascades using a proteomicbased approach if one wants to truly elucidate ongoing cellular signaling processes. Finally, most current therapeutics are directed at protein targets, and these targets are often protein kinases, their substrates, or both. The activation states of these proteins and networks fluctuate constantly depending on the cellular microenvironment. Consequently, the application of molecular profiling to provide individually tailored therapy should include direct proteomic pathway analysis of patient material. Moreover, because the cellular kinome represents a rich source of new targets for molecular therapeutics, technologies that can profile and assess the activity of these molecules 10.1021/pr7008127 CCC: $40.75
2008 American Chemical Society
research articles
Multiplexed Cell Signaling Analysis of Human Breast Cancer in human tissues will be critical for the realization of patienttailored therapy.11,12 Breast cancer remains the second most common cause of cancer death for women, behind lung cancer, with over 40 000 women estimated to succumb to the disease in 2007.13 While adjuvant therapy for early stage disease has made significant strides,14 the treatment of advanced and metastatic disease remains difficult.15–20 Our constantly evolving understanding of the molecular underpinnings of breast cancer development continues to provide promising leads for new therapies. The development and use of trastuzumab (Herceptin), a monoclonal antibody therapeutic agent directed against the extracellular domain of the human epidermal growth factor receptor (EGFR)-2 (HER2, ErbB2), represents the first successful example of a rationally designed agent for breast cancer therapy that has come into widespread use.21–24 Trastuzumab has been used successfully both as a single agent and in combination chemotherapy in the metastatic and adjuvant setting; however, a significant number of HER2 overexpressing tumors do not respond to treatment. These results have led to the development of a number of other targeted inhibitors of the EGFR family of molecules, such as gefitinib (Iressa) and erlotinib (Tarceva), both of which are small molecule inhibitors of EGFR, and the dual specificity kinase inhibitor, lapatinib, which inhibits signaling from both the EGFR and HER2 molecules. These molecules have had modest successes as monotherapies in breast cancer25 but could have improved efficacy when used in combination with other targeted inhibitors of downstream signaling molecules.26–38 These therapeutic strategies are also currently being tested in phase II and III clinical trials in hopes of improving progression free survival in the locally advanced and metatstatic settings.26 While this current and growing cadre of targeted inhibitors represent an accelerating effort in the effective treatment of breast cancer, an effective and direct diagnostic assay format that can provide direct information about drug target activity is critically missing. Since these therapeutics target protein function, ongoing kinase activity, and receptor function, proteomic technologies are especially well-positioned to fill-in this knowledge gap. Platforms that can provide a quantitative, multiplexed read-out for cellular signaling and that can utilize microscopic quantities of tissue specimens for upfront analysis would be the most attractive systems. Protein microarrays represent an emerging technology that is rapidly becoming a powerful tool for drug discovery and signal transduction profiling of cellular material. The advantage of protein microarrays lies in their ability to provide a ‘map’ of known cellular signaling proteins that generally reflect the state of information flow through protein networks in individual specimens. Protein microarrays can be used to monitor changes in protein phosphorylation over time, before and after treatment, between disease and nondisease states, and between responders and nonresponders.39–42 Identification of critical nodes or interactions within the network is a potential starting point for drug development and the design of individual therapeutic regimens.43–47 The reverse-phase protein microarray (RPPA) format is uniquely suited to signal transduction profiling of small samples (e.g., biopsy specimens) such that rational selection and monitoring of patients for targeted medicine is now technically possible. The RPPA format immobilizes an individual test sample in each array spot, such that an array comprises hundreds of different patient samples or cellular lysates. The arrays can then be probed with
phosphospecific and total protein antibodies to assess the activation state of key signaling molecules.11,48,49 When coupled with technologies such as laser capture microdissection (LCM), the RPPA format provides the opportunity to screen clinical samples that are available in very limited quantities, such as biopsy specimens.39,50–55 Tissues are complicated 3D structures composed of large numbers of different types of interacting cell populations. In many cases, the cell subpopulation of interest in a tissue may constitute only a tiny fraction of the total tissue volume. This issue of tissue heterogeneity has been a significant barrier to the molecular analysis of normal and diseased tissue. If the extract of a complex tissue is to be analyzed using a sophisticated and sensitive technology such as gene expression or protein microarrays, the output data could be severely compromised if undesired cells contaminate the starting material. LCM and other laser microdissection technologies have been developed to provide scientists with a quick and dependable method of capturing and isolating specific cells from tissue under direct microscopic visualization.56,57 With the use of LCM, the approach to molecular analysis of pathologic processes has been significantly enhanced.58 However, since many solid human tumors are identified at a late stage when metastasis has occurred or the disease has recurred following therapy, it is not enough to develop pathwaydriven clusters and signaling portraits based on primary tissue. Because metastasis is the lethal aspect of the disease, it is crucial that we understand the molecular profile of the tumor within the metastatic microenvironment, and begin to understand the impact of the specific microenvironment on the signaling phenotype. Currently, analysis of breast cancer metastasis has focused mostly on the identification of genomic signatures of primary tumors that can discriminate metastatic from nonmetastatic tumors and, thus, prognose outcome and aggressiveness.59 Conversely, comparatively little work has been performed on the molecular changes that arise within the metastatic lesion itself, either as a consequence or a functional causal component of the metastatic process. In this report, we analyze breast cancer liver metastasis, a common site of breast tumor metastasis, and compare the signatures to the liver metastasis from other solid tumors as a beginning step to understand the impact on ongoing signaling when the tumor encounters the new metastatic environment. In this study, we employed RPPA analysis for a more in-depth analysis of cellular signaling of breast cancer tissue than has ever been conducted before in order to begin the exploration and mapping of signaling portraits of human breast cancer. We employ this technology to investigate the impact of LCM on cell signaling analysis, analyze signaling activation in the setting of metastasis, and finally illustrate the potential for cell signaling profiling in a clinical trial setting with a targeted inhibitor. This analysis will serve as the basis for further ongoing and larger studies of breast cancer signaling to begin a functional classification of the disease based on the activation states of the drug targets themselves.
Material and Methods Clinical Sample Collection and Handling. Patient eligibility criteria, treatment plans, and clinical and toxicity evaluation were approved by the appropriate institutional review boards. All patients gave written informed consent. Core biopsies and surgical specimens were anonymized and frozen at -80 °C within 5 min of surgical removal in order to minimize any Journal of Proteome Research • Vol. 7, No. 4, 2008 1509
research articles postsurgical changes in protein expression. A board-certified pathologist confirmed the presence of tumor in each sample. Laser Capture Microdissection. Frozen tissue specimens were processed for laser capture microdissection as described in Espina, et al.56,57 Between 1500 and 9000 LCM shots (15–30 µm diameter, 5–15 cells/shot) of breast carcinoma tissue were procured for each sample. Reverse-Phase Protein Microarray Construction and Data Analysis. Printing of the lysates onto nitrocellulose arrays was carried out as described previously.11,54 Cases were printed in duplicate, in 5-point dilution curves, thus, assuring that the linear detection range was encompassed for the chosen antibody concentration. A high and low control lysate (A431 squamous carcinoma cells +/- EGF for 30 min) was printed on every slide and its performance assessed as described previously.54 Array staining with antibodies was carried out as described previously.53–55 All antibodies used in these studies were validated for specificity by immunoblotting prior to use on the arrays. All end points tested in these studies are listed in Supporting Information. Total protein values were assessed by staining one or more slides with Sypro Ruby Blot Stain (Molecular Probes, Eugene, OR). Stained slides were scanned on a standard flatbed scanner as 16-bit images/600dpi, and spot intensities were quantitated and normalized to total protein values using MicroVigene software (Vigenetech, Carlisle, MA). Final normalized signal intensities were incorporated into JMP 5.0 software (SAS, Cary, NC), and an unsupervised hierarchical clustering analysis was performed (Ward method), as described previously.54 Statistical analyses were performed using the SAS statistical package (SAS). A paired t test was used for the matched samples, and t tests or nonparametric tests were used for other comparisons.
Results and Discussion Necessity and Impact of LCM on Signal Pathway Profiling in Breast Tumors. Common problems involved in both genomic and proteomic tissue profiling analyses arise from heterogeneity of tissues. Variations in cellular composition between samples in a study and also within different regions of an individual specimen can make comparisons of molecular profiles from undissected material of questionable value. An important question that continuously arises when these molecular studies are performed is: Is it necessary to perform a labor-intensive procedure such as LCM for molecular profiling, or can we rely on analysis of whole undissected tumor-enriched tissue specimens as the basis for our data? To begin to understand and evaluate the impact of LCM on signal pathway profiling, we undertook a direct comparison of protein signaling profiles in lysates from patient-matched whole crysostat sections and contiguous tissue sections that were subjected to laser capture microdissection that procured pure population of tumor epithelium. A study set of 22 core biopsy specimens was selected from a larger tissue study set based on the presence of a high relative percentage of tumor epithelium. These samples were analyzed for signaling pathway activation by RPPA using a series of 35 phospho-specific and total protein antibodies against key signaling molecules regulating cell proliferation, motility, apoptosis, and survival. Results of two-way, unsupervised clustering analysis of the resulting data are shown in Figure 1. Unsupervised analysis was chosen to evaluate similarities in overarching profiles obtained. Overall, there was no significant and clear clustering 1510
Journal of Proteome Research • Vol. 7, No. 4, 2008
Wulfkuhle et al. of samples based on procurement methodology (LCM vs undissected) (Figure 1, left panel). Direct clustering of matched microdissected and whole tissue lysates was observed for 7 of the 22 pairs analyzed, which indicates a general similarity between the undissected and microdissected signaling portraits in those tissue samples. Comparisons of the relative signal intensities between the matched samples for 4 patients from the group of 7 pairs that did cluster together (Figure 1, patients A-D) demonstrated significant, but not comprehensive, concordance of relative signaling levels between these matched samples. This clustering was sometimes driven by high levels of activation for a few end points (Figure 1, patients B and C) or by more generalized midrange to low relative levels of signaling as observed in patients A and D (Figure 1). Direct pairing of these 7 matched samples was maintained in clustering analyses of the larger study set of 153 microdissected specimens that included the undissected versus microdissected pairs for these specific samples (data not shown), which indicates that the similarity is substantial. However, matched pairing was the exception rather than the rule, with the majority of matched samples revealing significantly divergent signaling activation patterns (Figure 1, patients E-H). Many of the end points varied by 3- or 4-fold between the undissected and microdissected matched pair within any given patient sample. Past work has revealed that microdissection of more than 1000 cells produces consistent results between cells populations within the same specimen, so these differences are not expected to be due to cellular variability.39,50–55 Moreover, we used contiguous cryostat sections for the microdissected versus undissected analysis in order to minimize any impact from cellspecific heterogeneity. We next sought to understand if the impact on results obtained by microdissection were so significant as to transcend any specific patient sample. Indeed, statistical analysis revealed significant differences (p < 0.05) in signaling levels between the entire study set of LCM and whole tissue samples for 16 of the 35 end points tested (Figure 2 and Table 1). These proteins were involved in control of cell proliferation, apoptosis, and cell survival. LCM tissues had statistically significant higher phosphorylation levels than their matched whole tissue lysates for EGFR (Y992) and (Y1173) and Her2 (Y1248), as well as higher levels of ERR phosphorylation (Table 1), which could have significant impact on the use of the data from undissected tissue samples for clinical decisionmaking. Proteins involved in cell survival signaling, such as AKT and its substrate mTOR, also exhibited higher relative levels of phosphorylation in LCM tissue compared to whole tissue lysates (Figure 2; Table 1). These data demonstrate that the results obtained for any given patient or set of tissue where class prediction/prognosis biomarkers are being explored would be greatly influenced by not microdissecting the sample set, and not effectively recapitulating the signaling of the cancer epithelial cells. This initial result will be followed by larger comprehensive studies of both tumor and stromal cellular compartments in order to further understand the global impact of microdissection on proteomic analysis and the results obtained, and emphasizes the critical role that LCM technology should play in the analysis of complex tissues. Indeed, the molecular analysis of stromal compartments of tumor tissue is an important aspect to understand changes in the tumor microenvironment that can lead to pathological changes and metastasis as well as provide prognostic information. Recent results in colorectal cancer tissue indicate that cell signaling
research articles
Multiplexed Cell Signaling Analysis of Human Breast Cancer
Figure 1. Unsupervised hierarchical clustering of patient-matched undissected breast tissues and LCM-procured breast tumor epithelial cells. Red color indicates higher relative levels of a given analyte and green represents the lowest levels. Left panel: Patient samples are oriented on the vertical-axis, with LCM samples labeled in blue and undissected tissues in black. The kinase substrate end points tested are oriented on the horizontal axis. Right panel: A selection of matched tissue pairs that represent either strong or weak concordance of staining levels between the LCM and undissected sample. Table 1. P-Values for Statistically Different Signaling End Points between LCM-Procured Breast Tumor Epithelial Cells and Matched Undissected Breast Tissue
Figure 2. Comparisons of staining intensities for end points that were statistically different between LCM tissues and undissected samples. Normalized intensity values represent averages of the two groups for each end point.
architecture of the stroma near tumor epithelium are significantly altered compared to stroma that is at a distance from the tumor mass.60 Further work in this regard is necessary and ongoing for the analysis within the breast cancer microenvironment. Past work has shown that the LCM procedure itself
end points
P-value
AKT T308 Beta Actin Cleaved Caspase-3 D175 E Cadherin EGFR L858R EGFR Y992 EGFR Y1173 Estrogen receptor S118 FADD S194 GSKalfa-beta S21–9 Herb2 Y1248 IKBa ser32 IRS1 S612 mTOR S2448 PDGFR-beta Y716
0.04a
