A systems biology approach to defining metastatic biomarkers and signaling pathways

Abstract

Metastasis is the final stage of cancer and the primary cause of mortality for most solid malignancies. This terminal phase of cancer progression has been investigated using a variety of high-throughput technologies (i.e. gene expression arrays, array comparative genomic hybridization (aCGH), and proteomics) to identify prognostic expression profiles and better characterize the metastatic process. For decades, the predominant model for the metastatic process has been the “progression model,” yet recent microarray results tend to support an inherent metastatic capability within primary tumors. Moreover, studies using a highly metastatic transgenic mammary tumor model suggest germline polymorphisms are significant determinants of metastatic efficiency. Likewise, a strong concordance of survival has been observed between family members with cancer, further supporting the link between genetic inheritance and survival. In addition, chromosomal aberrations and signaling pathways related to metastatic capacity have been identified by array-chromosomal genomic hybridization (aCGH) and proteomic studies, respectively. Lastly, carcinoma enzyme activity profiles were observed using activity-based proteomics (ABPP), which may be more clinically useful than expression-based proteomics for certain cancers. Most importantly, the application of these high-throughput techniques should expedite the search for additional biomarkers, germline polymorphisms, and expression signatures with greater prognostic value.

Metastasis

Metastasis is the final stage of cancer and the primary cause of mortality for most solid malignancies. The metastatic process is extremely complex and only partially understood. A metastatic tumor encompasses not only the complexity of primary oncogenesis, but must also survive a series of events and environments before lodging in a secondary site and forming a productively proliferative mass. These events include escape from the primary tumor and invasion through the surrounding stroma, penetration into the vasculature or lymphatics, surviving anchorage-independence and hemodynamic shear forces, arrest at the secondary site either by receptor-mediated adhesion or physical trapping within the capillary beds, and finally initiation of growth in a foreign microenvironment either within the vasculature or after penetration into the surrounding parenchyma before proliferating [1].

With such complexity it is no surprise that this terminal phase of cancer progression has increasingly become the focus of a variety of genomic investigations. The first major efforts revolved around whole genome gene expression technologies. Beginning in 2002, a number of studies explored the gene expression patterns of metastatic cancer and demonstrated the ability to discriminate between highly malignant metastatic cancers and more benign neoplasias [2, 3], thereby suggesting this genomic technology may reveal molecular mechanisms underlying the metastatic process. Moreover, a variety of high-throughput technologies (i.e. gene expression arrays, array comparative genomic hybridization (aCGH), and proteomics) are currently being pursued to identify more clinically useful biomarkers to predict metastatic potential. Ideal biomarkers would, most importantly, be used to rationally select the most appropriate therapeutic strategy for individuals, thus avoiding unnecessary adjuvant therapy for individuals with a low-risk of developing metastasis and more aggressive therapy for individuals with a high-risk of developing metastasis. Ultimately, these novel biomarkers should provide patient-specific treatment options that would, in turn, improve patient outcome.

Challenging the Clonal Evolution Model of Metastasis

Over the past thirty years, the predominant model for the metastatic process has been the progression model, or “clonal evolution” model [4–7]. In this model, tumor cells within the primary tumor accumulate a series of somatic alterations during the evolution of the tumor that provide subsets of cells with the abilities needed to complete the metastatic cascade. This model, however, is not necessarily consistent with recent microarray results which have detected pro-metastatic expression profiles in bulk primary tumor tissue. Importantly, these results have led some investigators to suggest the malignancy of the tumor may be encoded early in the oncogenic process, perhaps even by the primary oncogenic lesions themselves.

Initial support for this new metastatic model was generated when a 70-gene prognosis signature, which classified patients as having either a ‘good prognosis’ or ‘poor prognosis’ signature, was identified during cDNA microarray analysis of primary breast tumors [2]. Consequently, patients with the ‘poor prognosis’ signature had a 28-fold odds ratio (OR) (95% confidence interval, CI 7–107, P = 1.0 × 10⁻⁸) of developing distant metastasis within 5 years compared to individuals with the ‘good prognosis’ signature [2]. Further, multivariable Cox regression analysis showed the 70-gene signature was a more powerful predictor of disease outcome than standard systems based on clinical and histological criteria [8]. Additionally, computational analysis using the 70-gene signature revealed primary breast tumors and their metastases displayed similar gene expression profiles [9]. Taken together, these results were interpreted to support the inherent metastatic capability of primary tumors as opposed to the clonal selection of metastatic capability.

Inherited Susceptibility Model of Metastasis

Concurrent with these studies, investigations into inherited susceptibility to metastatic progression has led to an alternative hypothesis to tumor progression that is compatible with both the clonal selection and initiating oncogenesis models. Using a highly metastatic transgenic mammary tumor model it was demonstrated that a significant differences in metastatic efficiency was observed in the F1 progeny between the transgenic mouse and other inbred strains[10], suggesting that inherited polymorphism is a significant determinant of the metastatic efficiency. Further analysis involving quantitative trait mapping in standard backcrosses and recombinant inbred (RI) backcrosses identified metastatic efficiency loci on multiple chromosomes [11], leading to the discovery of the signal-induced proliferation-associated gene1, Sipa1 [12], as the first candidate metastasis efficiency modifier locus [11]. Furthermore, germline polymorphisms in human Sipa1 have been associated with poor outcome in breast cancer [13] in at least one small pilot epidemiology study. The link between germline genetic variations and cancer outcome is further supported by genetic inheritance studies documenting a strong concordance of survival between family members with cancer. Specifically, when mother-daughter pairs with breast cancer were studied, a concordance of prognosis was observed: the HR was 0.65 in daughters whose mothers had survived >= 120 months compared to daughters whose mothers had survived less than 36 months (P-value for trend 0.02) [14]. Most importantly, a concordance of survival in family members with prostate cancer [15, 16], bladder cancer [17], renal cell cancer [17], colorectal cancer [16], and lung cancer [16] has also been reported, thereby supporting a noteworthy correlation between germline variations and survival within multiple cancer types.

In addition to the effect on the physiological manifestation of metastases, these studies also employed genomic tools to assess the impact of polymorphism on the transcriptome. Intriguingly, it was demonstrated that the newly identified metastasis efficiency modifiers not only had the capacity to affect the development of overt secondary tumors, but could also generate predictive gene signatures. These results suggest that the discrepancy between the clonal evolution and initiating oncogenesis models might be reconciled by ascribing the origin of at least some significant fraction of the predictive gene expression pattern to inherited polymorphism rather than acquired mutation. Thus, the bulk of the tumor would express a predictive gene signature, upon which additional somatic events could occur resulting in further induction of additional gene expression components and acquisition of metastatic capacity.

Target Tropism and Biological Signatures of Metastasis

The fact that somatic changes do in fact influence the metastatic prognostic signatures has been most strongly supported by recent studies examining metastatic target tropism. To characterize lung-specific metastasis, parental MDA-MB-231 cells and the 1834 sub-line (an in vivo isolate with no enhancement in bone metastatic behavior)were injected into the tail vein of immunodeficient mice [18]. The second-generation populations (denoted LM2) were rapidly and efficiently metastatic to the lung. Interestingly, transcriptomic microarray analysis comparing the LM2 and parental MDA-MB-231 cell populations revealed a 54 gene-expression signature that could effectively distinguish cell lines with a high risk and those with a low risk for developing lung metastasis (P = 0.0018). Similar results were observed in cells with enhanced metastatic abilities to bone [19].

Further investigation motifs within gene signatures has also been used to identify the biological mechanisms underlying tumor progression. For example, a link between wound healing and cancer progression was established when an “activated” and “quiescent” wound-response signature was identified in early breast cancer patients [20]. Additional signatures that reflect activated gene pathways include a hypoxia-associated transcriptional response signature [21], a signature derived from good- versus poor-outcome fibroblastic tumors (solitary fibrous tumor versus desmoid-type fibromatosis; SFT/DTF)[22], and an interferon (IFN) response signature from cocultured fibroblasts [23].

Recently, microRNAs (miRNAs) expression profiles have also been used to classify human cancers and distinquish tumors from normal tissue [24]. Briefly, miRNAs are small non-coding RNA molecules that can regulate gene expression by targeting mRNAs for cleavage or translational repression [25]. miRNAs have received a great deal of attention lately since cancer-specific miRNA fingerprints have been identified in B cell chronic lymphocytic leukemias [26], breast carcinoma [27], lung cancer [28, 29] and colon carcinoma [30]. In addition, tissue-specific miRNA expression signatures have been documented in a variety of normal tissue types [31]. Importantly, a number of miRNAs can function as either stimulators or inhibitors of breast cancer metastasis [32]. For instance, overexpression of miR-200 family members repressed epithelial-mesenchymal transition (EMT) [33] and miR-335 expression suppressed metastatic cell invasion [34]. In contrast, individual expression of miR-21 [35], miR-10b [36], miR-373 or miR-520C [37] promoted tumor invasion and metastasis. Lastly, miRNA DNA copy number abnormalities have been documented in ovarian cancer, breast cancer, melanoma [38] and chronic lymphocytic leukemia [39]. Hence further exploration of cancer-specific miRNA expression signatures and miRNA DNA copy number abnormalities should help better characterize the relationship between miRNA expression patterns and metastasis.

Genetic Aberrations in Metastatic Lesions

DNA copy number alterations can directly affect gene expression patterns to promote cancer progression [40], hence genomic analysis using array-Comparative Genome Hybridization (aCGH) has been utilized to produce high resolution views of chromosomal loss and gain within cancer genomes. Importantly, detection of biologically relevant chromosomal aberrations can be applied to identify genes driving tumorigenesis. The melanoma metastatic gene NEDD9 was recently identified using genome-wide aCGH analysis of metastatic variants from an inducible mouse model of melanoma [41]. In addition, array-CGH has been used to identify the origin of certain lesions. For instance, aCGH was successfully applied to determine whether a pulmonary nodule was primary from the lung or metastatic from the neuroendocrine carcinoma (NEC) component [42].

Numerous studies have also highlighted the potential application of aCGH as a prognostic tool during cancer diagnosis. BAC-array analysis of primary breast cancers revealed patients with tumors displaying less than 5% (median value) total copy number changes had a better overall survival (log-rank test: P = 0.0417) than patients displaying greater than 5% (median value) total copy number changes [43]. Likewise, analysis of aggressively treated early-stage breast tumors by aCGH identified four regions of recurrent amplification associated with poor outcome, and nine high-level amplifications that were strongly associated with survival duration and distant recurrence [44]. Furthermore, analysis of primary breast tumors of relatively small size and low Nottingham Prognostic Index (NPI) revealed genomic alternations that were frequently associated with a significantly worse prognosis [45]. Lastly, aCGH analysis of prostate cancer patients revealed chromosomal aberrations associated with advance stage disease, predictive of post-operative recurrence, and metastatic potential [46]. These studies illustrate the potential application of aCGH to identify clinically relevant gene-driven chromosomal alterations.

Whole-genome allelic imbalance scans were performed on eight sets of primary and metastatic lung cancers using high-resolution single nucleotide polymorphism arrays [47]. The majority (> 67%) of the genetic alterations were detected in both the metastatic tumors and their corresponding primary tumors. Some allelic imbalances, however, were detected in 50% of the metastatic tumors, but none of the primary tumors. This suggests even though a slightly greater level of genomic instability is observed in metastatic lesions compared to primary lesions, most metastatic capability is predetermined by the genomic variations present in the primary tumor. Further support for this metastatic model was generated in the mammary intraepithelial neoplasia outgrowths (MIN-O) mouse model when molecular profiles correlated with metastasis were identified in the early premalignant lesions [48].

Genetic Aberrations in Disseminating Cancer Cells and Stromal Tissue

Additional aCGH studies have explored the genetic variations present in disseminating cancer cells and neighboring stromal tissue. Analysis of single micrometastatic tumor cells isolated from the bone marrow of breast cancer patients revealed amplifications as small as 4.4 and 5 Mb [49]. Interestingly, when the genetic aberrations within single disseminating cancer cells from the bone marrow of breast cancer patients with manifest primary tumor (stage M0) or metastasis (stage M1) were examined by CGH, fewer chromosomal aberrations were detected in the single disseminating cancer cells compared to the primary tumors and metastasis (P < 0.008 and P < 0.0001, respectively) [50]. Most of all, these results support an alternative “parallel evolution” model of metastasis in which tumors cells disseminate very early during tumorigenesis, and evolve independent from the primary tumor during the metastatic process. If this is the case, then it argues that adjuvant therapies should target disseminating cells instead of primary tumors, since they contain the genetic variations which will eventually cause metastasis. Lastly, numerous amplifications and deletions were identified in host stromal cells during tumor progression using representational oligonucleotide microarray analysis (ROMA) [51]. In contrast, a more recent report observed virtually no genetic aberrations in carcinoma-associated fibroblasts (CAFs) [52]. Clearly, debate remains on whether tumor-promoting genetic aberrations exist within the tumor microenvironment [53–55]. Overall, these studies suggest that cancer treatments should target the stromal tissue surrounding each lesion and the disseminating cancer cells, in addition to the primary and metastatic lesions.

Proteomic Analysis of Metastatic Cell Lines

There are 30,000 genes in human cells with over 100,000 expected protein isoforms as a result of alternative splicing; hence, a significant amount of information is lost by sole analysis of gene expression levels. Further, since it has been shown that a relatively poor correlation exits between mRNA and protein levels [56], mRNA expression level measurements alone should not be used to estimate protein expression levels. Hence, multiple high-throughput proteomic techniques (i.e. 2D gel electrophoresis, flow cytometry, mass spectroscopy (MS), protein microarrays, and tissue microarrays) are being applied to analyze protein expression levels.

For high-throughput analysis of protein expression patterns between separate tumor samples, two-dimensional electrophoresis (2-D gels) in combination with mass spectroscopy (MS) has been quite successful. For instance, this technique revealed greater similarity between the protein expression signatures for the adriamycin resistant breast cancer cell line (MCF-7/ADR) and the ductal infiltrating breast carcinoma cell line 8701, compared to the metastatic breast cancer cell lines MDA-MB 231 and MDA-MB 435 MDA-MB cell lines [57]. Likewise, a total of 43 differentially expressed proteins were identified between melanoma cells that either overexpressed (Mel-BRMS1) or silenced (sh635) the breast cancer metastasis suppressor 1 (BRMS1) gene [58]. Overall, these studies illustrate the constructive application of 2-D gels in combination with MS to identify differentially expressed proteins related to metastatic capacity.

Although 2-D gels in combination with MS analysis is useful for identifying differentially expressed proteins, isotope coded affinity tag (ICAT) may provide more quantitative protein expression level measurements. Briefly, the ICAT labeling technique creates “heavy” and “light” proteins which are subsequently detected by tandem mass spectrometry. For example, ICAT analysis using MDA-MB-435 tumor cells transfected with either a control vector (C-100) or the Nm23-H1 metastasis suppressor gene (H1–177) revealed 129 differentially expressed proteins [59]. An additional ICAT study revealed 240 differentially expressed proteins between the metastatic gastric cancer TMC-1 cell line and the noninvasive gastric cancer SC-M1 cell line [60]. Further, when ICAT and microarray gene expression analysis were combined, Stat signal transduction pathways and differential expression of many basement membrane proteins were linked to metastasis inhibition [61]. Taken together, these studies support the application of ICAT to identify differentially expressed proteins related to metastasis.

Tissue Microarrays for Biomarker Validation and Identification

Lastly, tissue microarrays [62] have become a useful tool for high-throughput cancer profiling [63] and for the subsequent validation of gene and protein expression results. For instance, tissue microarrays, using a novel method of automated quantitative analysis (AQUA) [64–66], confirmed the gene expression changes detected by DNA microarray between the poorly metastatic GI101A human breast cancer cell line and a highly metastatic GILM2 variant. In brief, a total of 106 differentially expressed genes were detected in the highly metastatic GILM2 variant, and subsequent tissue microarray analysis later revealed high expression of HSP-70 and high nuclear CXCL-1 expression were both associated with decreased survival (P = 0.05 and 0.027, respectively) [67]. In addition, when tissue microarray analysis and high-throughput immunoblot analysis were combined with transcriptomic data, signatures of metastatic progression were identified in prostate cancer [68]. Surprisingly, integrative analysis of proteomic and transcriptomic data revealed only a 48%–64% concordance between the two data sets [68], thereby stressing the importance of validating gene expression data at the protein level before initiating comprehensive candidate gene investigations.

Promising Areas of Proteomic Research

Proteomic analysis of single cells and minute tissue biopsies is an extremely promising research avenue since it could potentially be applied during cancer screening and diagnosis when limited sample is available. In fact, direct tissue proteomics (DTP) identified 428 prostate-expressed proteins directly from minute prostate biopsies [69]. Further bioinformatic analysis of the DTP data revealed metabolic pathways potentially important during distinct stages of prostate carcinogenesis [69]. In addition, phospho-specific flow cytometry has also been used to analyze pathological signaling networks at the single-cell level [70]. The promise associated with flow cytometry is linked to its ability to identify relevant signaling networks at the single cell level which would otherwise be masked during bulk tumor analysis.

A second promising research field is the application of activity-based proteomics (ABPP) to monitor the functional activity of protein classes during tumorigenesis. For example, active site directed probes have been used to visualize the function activity of proteins in the serine hydrolase enzyme family [71]. Briefly, a biotinylated fluorophosphonate, referred to as FP-biotin, can covalently label proteins in an activity-dependent manner to monitor the kinetics of protein function. Importantly, since protease activity has been linked to cancer progression [72] and because the therapeutic potential of protease inhibitors has been investigated [73], a functional assay to assess protease activity within primary tumors might be an effective tool to evaluate whether treatment with protease inhibitors would be clinically useful. The clinical usefulness of function assays as a prognostic and therapeutic tool during cancer treatment was further illustrated when the human breast cancer cell line MDA-MB-231 was grown in culture and as orthotopic xenograft tumors in the mammary fat pad of immunodeficient mice. A suite of activity-based chemical probes identified carcinoma enzyme activity profiles that were associated with greater tumor growth rates and metastases [74]. In summary, ABPP would be a more accurate tool for clinical diagnosis and prognosis when the activity levels of specific enzymes is more tightly correlated to clinical outcome than their protein expression levels.

Conclusion

Concluding Remarks and Clinical Implications

In closing, because a number of relatively novel high-throughput techniques are currently available, views regarding the metastatic process are constantly evolving and adapting to the results from the latest high-throughput investigation. Most influential was the application of gene expression arrays to analyze the expression patterns between matched primary and metastatic lesions. These results unexpectedly challenged the long-standing “clonal evolution” model since they demonstrated metastatic capacity could be predicted within the expression profile of the primary lesions. Furthermore, investigations using metastatic mouse models, germline polymorphism screening and population genetics have uncovered a surprising link between genetic background and cancer predisposition, requiring further examination. In addition, the genomic aberrations observed in disseminating cancer cells suggest tumors evolve independently from the primary tumor during the metastatic process, thereby supporting the “parallel evolution” model. Consequently, although a number of metastatic models (i.e. clonal evolution, parallel evolution, transient compartment, germline efficiency) have been introduced to explain the metastatic process, the amount of contradictory data obtained using these technologies suggests the “correct” metastatic model has not yet been defined. The “correct” model will likely incorporate elements from each of these models to illustrate a metastasis process that relies on a complex exchange between germline polymorphisms and somatic mutations to maximize metastatic capabilities within both clonally evolved and predetermined cancer cells. Most importantly, merging the field of oncology with many of these high-throughput techniques should be extremely useful for both defining the mechanisms and pathways involved in metastasis as well as identifying both genomic and proteomic prognosis indicators. Hopefully these prognostic indicators will predict metastatic propensity with such a high degree of certainly that patient-specific treatments can be pursued based on each patient’s individual risk of metastasis.