Section 3 -

Microarray Analysis of Soft Tissue Tumors Torsten Nielsen University of British Columbia

Microarrays – History Microarrays are among the newest availab le tools that fit within the topic of Ancillary Techniques in the Evaluation of Soft Tissue Tumors. When most people use the term "microarray," they are referring to chips or slides containing thousands of DNA elements that are used for concurrent genome-wide measurement of mRNA levels, to generate a "gene expression profile." The first uses of DNA microarrays for expression profiling were reported 10 years ago [Shena 95][Shalon 96][Schena 96] but the technique did not really capture the attention of the biomedical research community until the release of a special issue of Nature Genetics devoted to the topic (The chipping forecast. Nature Genetics volume 21, Supplement, 1999). Some early studies on small numbers of sarcomas and sarcoma cell lines demonstrated the feasibility of applying this approach to musculoskeletal tumors [Welford 98][Khan 98][Wolf 00]. The 2001 USCAP/IAP meeting in Atlanta not only saw the inaugural companion meeting of the Association for Molecular Pathology, but also the first public presentations of the application of microarrays to large numbers of soft tissue tumor specimens [Ladanyi 01][Nielsen 01][Schofield 01]. These reports were closely followed by the first published datasets on small blue cell tumors [Khan 01], gastrointestinal stromal tumors [Allander 01], and adult soft tissue sarcomas [Nielsen 02]. Microarrays – What They Tell You Microarray data sets are exciting because of their sheer size and theoretical capacity to capture a global view of gene expression. In doing so, they not only measure the expression of every individual gene represented in the microarray, but also provide a comprehensive survey of whole pathways and banks of interacting genes (whose relationships may not have been appreciated before). This yields a readout of the molecular choreography of a tissue, and represents a new level of information not assessable with other avai lable techniques. Single biomarker approaches have found at best ancillary uses and are always applied in the context of histomorphology. The vast amounts of information generated by microarray expression profiles suggested to many that microarrays had the potential to replace histomorphology as a diagnostic gold standard. Indeed, in 1999 the National Cancer Institute issued a "director's challenge" series of major grant funding opportunities that was explicitly "intended to redefine tumor classification, moving from morphology-based to molecular-based classification schemes." Microarrays – Platforms The simplest microarray platform to understand is the spotted cDNA microarray. Essentially these are comprised of small aliquots of distinct cloned gene sequences arrayed by the thousands on glass slides. Tumor RNA is reverse-transcribed in the presence of fluorescing nucleotides and hybridized onto these arrays to give a readout of the abundance of each transcript in the tissue sample for which a probe exists. To avoid some of the problems associated with poorly-defined clones, repetitive sequences and variable hybridization kinetics, fully defined oligonucleotides can be spotted to act as the arrayed probes. Agilent arrays are a commercial product that uses ink-jet technology to synthesize such defined oligonucleotides (60-mers) directly on slides. Affymetrix microarrays are synthesized on silicon wafers using photolithography techniques (original reference [Fodor 93]) and can achieve extremely high densities (105-106 elements per chip). Genes are represented as a "probe set" of several 25mers from different parts of each gene. These arrays are particularly expensive but have the advantage of being standardized across labs purchasing the same generation of chips. It should be noted that a focused expression profile can be generated without microarrays. Quantitative PCR techniques can now be scaled up to test many score of genes concurrently, and this technique works on formalin-fixed, paraffin-embedded material [Paik 04]. Newer microarray platforms are also being rapidly developed (e.g. Illumina, NimbleGen) which have particular technical advantages. Lastly, mention should be made of other types of "microarray" techniques. Array-based comparative genomic hybridization (aCGH) uses microarrays which are designed represent genomic DNA sequences and are used to interrogate tumor DNA samples for genetic copy number changes [Atiye 05]. Tissue microarrays are quite different, containing small cores from paraffin blocks arranged such that several hundred tumor samples can fit onto a single glass slide [Kononen 98]. The morphologic and epidemiologic information gained from tissue arrays complement expression profile data quite well and so represent an excellent validation platform and translational research tool. Microarrays – Data Analysis Microarray data sets are complicated and require multiple steps in analysis (including normalization, filtering for data quality, and filtering for probable biological interest). These approaches are not standardized and are a subject of ongoing, active development in the scientific community. Comparing tissue samples to one another can be done in either an unsupervised (highlighting the patterns intrinsic to the data) or a supervised fashion (used when seeking answers to a particular question, such as the top genes associated with a particular diagnosis, prognosis or treatment response). Major challenges include false discovery rates, technical artifacts, biological and sampling variability. Results are also heavily influenced by the choice of comparison specimens included in the study. Against this background, there have been several important insights gained about musculo-skeletal tumors through microarray analysis of primary human tumor specimens, which can roughly be characterized as relating to tumor biology and diagnostics, prognostics-prediction, and target discovery. Microarrays for Sarcoma Biology and Diagnostics An early insight gleaned from unsupervised microarray analyses is that sarcoma expression profiles fall into two main camps: diagnostic entities with fairly consistent and distinctive banks of up- and down-regulated genes (e.g. gastrointestinal stromal tumor [Nielsen 02], synovial sarcoma [Nagayama 02], extraskeletal myxoid chondrosarcoma [Subramanian 05]), and entities where the specimens do not consistently group together by gene expression profile (e.g. pleomorphic sarcomas) [Nielsen 02][Segal 03][Baird 05]. These findings are consistent with what was already well-appreciated from karyotypic analyses [Borden 03], but further suggest that these complexities dominate not only the DNA organization, but also the transcriptional profile of pleomorphic sarcomas. Leiomyosarcomas may be an exception here as many express a strong smooth muscle signature including calponin, tropomyosin, myosin and leiomodin [Nielsen 02][Segal 03][Baird 05]. Small blue round cell tumors, which are difficult to diagnose by histology alone, have quite distinct gene expression profiles and are very well distinguished by microarray analysis using analytical techniques such as artificial neural networks [Khan 01]. This work demonstrates a particular strength microarrays have for diagnostics, inasmuch as the data analysis integrates all degrees of positive or negative expression and biomarker specificity across tumor classes to generate a classification. Tumor progression can be assessed by comparing malignant tumors to their precursor lesions. Malignant peripheral nerve sheath tumors, when compared to Schwann cells, exhibit a profound loss of Schwannian differentiation (SOX10, PMP22, NGFR, S100B) and concurrently upregulate a smaller set of neural crest and mesenchymal stem cells genes (SOX9, TWIST1) [Miller 06]. Published data on the most relevant comparison (plexiform neurofibroma) has unfortunately been limited to small studies using very small cDNA microarrays [Karube 06]. The published datasets do suggest that neurofibromatosis-associated and sporadic cases of MPNST are extremely similar on a molecular level [Watson 04]. Microarray analysis shows that progression of chondrosarcomas occurs with increased expression of genes facilitating anaerobic metabolism and decreased production of matrix [Rozeman 05]; few enchondromas were included. Similar analyses would also be interesting to compare lipoma, well-differentiated liposarcoma (WDLS), and dedifferentiated liposarcoma (DDLS), but some studies to date have lumped WDLS/DDLS with myxoid liposarcoma [Skubitz 05]. DDLS and round cell liposarcomas have distinct expression profiles, with DDLS showing high expression of MDM2, CDK4, and SAS, the genes located in the 12q amplicon [Segal 03]. Pathway signatures associated with sarcomas include Wnt/ β-catenin in synovial sarcoma, a consistent finding in several microarray studies [Nielsen 02][Segal 03][Baird 05] that has since been confirmed by functional molecular biology studies [Pretto 06]. Not surprisingly, Wnt signaling (including WISP-1) is a feature of desmoid-type fibromatosis [Skubitz 04][West 05], along with an extensive bank of genes (including ADAM12) involved in extracellular matrix production and remodeling. A relatively direct way of turning a gene expression profile into a diagnostic tool is to survey the lists of highly expressed genes for immunohistochemical markers. In this fashion, protein kinase C-theta, prominently expressed in gastrointestinal stromal tumors (GIST) in two published cDNA microarray studies [Allander 01, Nielsen 02] was validated as a useful diagnostic marker by other independent groups [Blay 04, Motegi 05]. PCK-theta has a clinically-useful role in that it is positive in GISTs associated with both KIT and PDGFRA mutations. Similarly, DOG1 can work as an immunohistochemical marker of both these molecular subtypes of GISTs; in this case the identified gene was largely uncharacterized and no preexisting antibody was availab le, but open reading frames predicted from a cDNA spot sequence were used to generate peptides and raise a novel antisera [West 04]. Microarrays for Sarcoma Prognosis and Prediction Comparison of gene expression profiles in primary tumors from patients with known outcome can be used to generate prognostic signatures of up- and down-regulated genes. Poor-outcome Ewing sarcoma, for example, has been linked to specific downregulated apoptosis / tumor suppressor genes and upregulated cell cycle / signal transduction genes [Ohali 04]. Upregulated cell cycle and signal transduction genes were also a feature of a gene signature prognostic of metastasis in soft tissue leiomyosarcoma [Lee 04]. Soft tissue tumor expression signatures even have prognostic significance when expressed in the stroma of other tumor types. Fibromatosis signatures (similar to a scar response) are associated with good outcome breast carcinomas, whereas the solitary fibrous tumor signature (which has a putative epithelial support role) is found in tumors with poor clinical outcomes [West 05]. Expression profiles can also be correlated with response to chemotherapy. In osteosarcoma, several papers [Mintz 05][Ochi 04][Man 05] have generated expression profiles associated with histologic response to neoadjuvant chemotherapy. However, the signatures so generated highlight different genes and pathways (osteoclast activation vs. metabolic enzymes vs. cell cycle), a somewhat disappointing result which may in part be related to differences in patient selection and treatment regimen, specimen handling, microarray platform and bioinformatic approaches, compounded by biological and technical variability and false discovery rates. A general problem impacting prognostic and predictive studies is the serious risk of generating an overfit model which cannot be generalized beyond the patient population and treatment regime upon which it was derived [Ransohoff 04]. External validation by other groups remains the most convincing evidence for a gene expression signature, a standard particularly hard to reach in sarcomas where the specimens and active research groups are few. Microarrays and Identification of Treatment Targets / Therapy Whereas the complexity of microarray data sets creates problems for gene signature model building, their comprehensiveness does make this an excellent approach for the identification of therapeutic targets. An important confirmation of the ability of microarrays to identify treatment targets in sarcomas comes from examination of GISTs [Allander 01][Nielsen 02], wherein c-kit expression was prominent in both unsupervised and supervised analyses. Lists of expressed genes often contain "low hanging fruit" – targets which are inhibited by preexisting drugs, few of which will have been ever tested against specific sarcomas. Examples of drug targets evident in sarcoma expression signatures include cyclooxygenase-2 expression in pediatric sarcomas [Dickens 02], vascular endothelial growth factor receptor-1 in hemangiopericytomas [Baird 05], colony stimulating factor-1 in pigmented villonodular synovitis [West 06] and, in synovial sarcoma, HER2 [Allander 02], epidermal growth factor receptor [Nielsen 02] and fibroblast growth factor receptor-3 [Ishibe 05]. Indeed, the plethora of new agents targeting receptor tyrosine kinases makes expression surveys of this class seem particularly worth pursuing [Baird 05]. Genes encoding surface membrane proteins can by their nature be targeted by custom-designed monoclonal antibodies (e.g. Frizzled homolog-10 in synovial sarcoma [Nagayama 05]). Targeted therapies are most likely to be successfully developed for cancers with a distinct underlying molecular biology, and are most needed for cancers for which no effective systemic therapy currently exists; the sarcoma world has many entities which fit these criteria. Nevertheless, the shortage of patients availab le for entry into trials means that supporting data should be generated to back up microarray findings. Protein-level validation is an important and relatively easy first step [Thomas 05] and is an area where tissue microarrays complement DNA microarrays particularly well [Nielsen 03] [Baird 05]. However, expression at an RNA or protein level does not guarantee that inhibitors of that biomarker will be effective [Lubieniecka 05]; indeed, if the target is expressed to compensate for an oncogenic event, an inhibitor may make the disease worse! In vitro testing of agents identified by expression profiling can provide a screen for efficacy, give information on mechanism [Ishibe 05] and even suggest related drug agents that might be better choices to advance to clinical trials [Terry 05]. Perspectives In sarcomas as for other cancers, the initial optimism for DNA microarray technology as a clinical tool, based on its exceptional comprehensiveness in surveying diagnostic biomarkers and treatment targets and its ability to identify gene signatures not easily recognized with previous techniques, has been tempered by real difficulties in application. These include not only cost and special tissue handling, but also major challenges in data interpretation and translation. Overall, DNA microarrays function extremely well as discovery tools, having already yielded new biological insights, identified diagnostic biomarkers, and highlighted potential targeted therapies for specific sarcomas. External validation of key findings is particularly important in the microarray field, but is difficult to achieve in sarcomas; however, the molecular biology of this group of diseases is such that DNA microarray technology is likely to prove particularly valuable. How DNA microarrays might end up applied to clinical practice remains an open area of development. Much of the relevant diagnostic information may be contained in a few specific biomarkers that can be detected by cheaper, established techniques such as immunohistochemistry. Limited gene expression signatures that cover the important genes for diagnosis and treatment planning can be detected more economically by methods such as quantitative PCR, which can be applied to small, formalin-fixed specimens. The microscope is not going to be replaced as the most important diagnostic tool for cancer any time soon. On a scientific discovery level, however, it may be fair to say that we are approaching the limits of useful new information that can be gained from light microscopy, and that microarray research is a particularly exciting, dynamic and rapidly-evolving field which will probably lead to further insights into the biology, diagnosis and treatment of sarcoma for some time to come. References Allander SV et al (2001) Cancer Res 61:8624-8 Allander SV et al (2002) Am J Pathol 161:1587-95 Atiye J et al (2005) Genes Chromosomes Cancer 42:158-63 Baird K et al (2005) Cancer Res 65:9226-35 Bibikova M et al (2004) Am J Pathol 165:1799-807 Blay P et al (2004) Clin Cancer Res 10:4089-95 Borden EC et al (2003) Clin Cancer Res 9:1941-56 Dickens DS et al (2002) Pediatr Dev Pathol 5:356-64 Fodor SP et al (1993) Nature 364:555-6 Ishibe T et al (2005) Clin Cancer Res 11:2702-12 Karube K et al (2006) J Clin Pathol 59:160-5 Khan J et al (1998) Cancer Res 58:5009-13 Khan J et al (2001) Nature Med 7:673-9 Kononen J et al (1998) Nature Med 4:844-7 Ladanyi M et al (2001) J Mol Diagn 3:92-7 Lee YF et al (2004) Cancer Res 64:7201-4 Lubieniecka JM & Nielsen TO (2005) J Surg Oncol 92:267-71 Man TK et al (2005) Cancer Res 65:8142-50 Miller SJ et al (2006) Cancer Res 66:2584-91 Mintz MB et al (2005) Cancer Res 65:1748-54 Motegi A et al (2005) Pathol Int 55:106-12 Nagayama S et al (2002) Cancer Res 62:5859-66 Nagayama S et al (2005) Oncogene 24:6201-12 Nielsen TO et al (2001) Mod Pathol 14(Suppl 1):16A Nielsen TO et al (2002) Lancet 359:1301-7. Nielsen TO et al (2003) Am J Pathol 163:1449-56 Ochi K et al (2004) Int J Oncol 24:647-55 Ohali A et al (2004) Oncogene 23:8997-9006 Paik S et al (2004) N Engl J Med 351:2817-26 Pretto D et al (2006) Oncogene [Epub ahead of print] PMID:16462762 Ransohoff DF (2004) Nature Rev Cancer 4:309-14 Rozeman LB et al (2005) J Pathol 207:61-71 Schena M et al (1995) Science 270:467-70 Schena M et al (1996) Proc Natl Acad Sci USA 93:10614-9 Schofield D et al (2001) Mod Pathol 14(Suppl 1):18A Segal NH et al (2003) Am J Pathol 163:691-700 Shalon D et al (1996) Genome Res 6:639-45 Skubitz KM & Skubitz AP (2004) J Lab Clin Med 143:89-98 Skubitz KM et al (2005) Cancer Invest 23:105-18 Subramanian S et al (2005) J Pathol 206:433-44 Terry J et al (2005) Clin Cancer Res 11:5631-8 Thomas DG et al (2005) Cancer 103:830-8 Watson MA et al (2004) Brain Pathol 14:297-303 Welford SM et al (1998) Nucleic Acids Res 26:3059-65 West RB et al (2004) Am J Pathol 165:107-13 West RB et al (2005) PLoS Biol 3:e187 West RB et al (2006) Proc Natl Acad Sci USA 103:690-5 Wolf M et al (2000) Cancer Genet Cytogenet 123:128-32