Those who study cancer generally agree that at its most basic level, cancer is a genetic disease. This is not to mean that cancer is inherited, because although several syndromes of familial cancer susceptibility exist, most cancers occur sporadically, and therefore are not inherited in the true sense of the word. Rather, this statement implies that the biological events that initiate the malignancy represent alterations in the expression of genes. Furthermore, as cancers progress, alterations in gene expression tend to become more profound and disruptions of the chromosomes, which govern gene expression, often become quite extreme. Thus, cancer biologists are aggressively pursuing genetic studies of cancer tissues in order to identify the master genes controlling individual malignancies. As tumors become more genetically disorganized as they progress, it is critical to separate the "wheat from the chaff". In other words, one needs to ignore those genetic abnormalities that happen to be present but are not functionally critical, and to identify those critical genes which cause and/or maintain the health and viability of the cancer cell.

Although sarcomas are often grouped together on the basis of their common ancestry from primitive mesenchymal progenitors (Figure 1) and their common clinical behavior, they are at the same time a very heterogeneous group of tumors.

Not surprisingly therefore, genetic studies thus far have confirmed that while the inner workings of sarcomas with common appearance under the microscope are similar, the inner workings of the different histologic subtypes of sarcoma are highly variable. For many sarcomas, unique chromosomal translocations serve as the driving force for oncogenesis, with a different translocation and a different path by which the translocation causes cancer in each histologic subtype. For other sarcomas, complex and widespread genetic disorganization is evident in the tumor cells, and therefore one cannot even begin to guess what might be the critical genetic event that gave rise to the tumor. Coming to grips with and delineating the genetic distinctions between individual sarcoma subtypes and identifying those genes critical for some or all sarcomas is essential for new targeted therapies to be developed for these diseases.

Historically, the genetic study of tumors (a field called molecular oncology), was largely conducted by individuals or small groups which generated and tested hypotheses regarding the importance of individual genes of interest, chosen on the basis of their known mechanisms of action. Such individual genes would typically be studied at length, one by one, using approaches that allowed investigators to identify whether the gene was present, absent, increased or decreased in various conditions. Although this approach has resulted in progress in cancer biology, it is difficult to study each individual sarcoma subtype with such labor intensive techniques. Furthermore, because there are an estimated 100,000 genes in the human repertoire, it is clear that the one-by-one approach is not able to efficiently "mine the genome" for what might be a new or unexpected gene that could play an important role.

Over the last several years, a variety of "high throughput" approaches have become available which can provide a comprehensive overview of gene expression in any given cell. Together, these technologies seek to profile gene expression within the cells and the various technologies can be referred to as "expression profiling". Depending upon the exact methodology used, it may be called DNA microarray, or GeneChip analysis. In general, these technologies are complex but highly reliable, and remarkably, they can be performed with very, very small amounts of tissue obtained from a routine clinical biopsy. Not surprisingly therefore, expression profiling is a technique now commonly used by molecular oncologists to study sarcomas, in an effort to jump-start our understanding of the biology of these complex, rare and life-threatening tumors for which all too often, effective therapy is not available. In order to understand this technology, one must first have a complete understanding of the basics of gene expression.

DNA microarray analysis starts with a glass slide, upon which small stretches of DNA are affixed. Remarkably, stretches from up to 20,000 or so genes can be imprinted onto one individual slide. They are typically arranged neatly in rows so that each dot representing individual genes can be identified according to its exact coordinates in the grid. The DNAs which are chosen by the investigator to be placed upon the slide represent portions of genes which are of potential interest in cancer biology, as well as some genetic stretches which have been found in tumors, but for which the corresponding gene has not been identified. The investigator then takes the tumor of interest and extracts the RNA from the tumor using standard techniques. It is anticipated that any of the genes, which were expressed by the tumor at the time the biopsy was taken, would have RNA present in the sample. RNA tends to be unstable and therefore a reaction is performed to generate the more stable complementary DNA (cDNA) from the RNA. This is a highly standardized and commonly used approach in molecular biology and essentially allows for a stable sample of all expressed genes present in the starting sample to be generated. The cDNA is then labeled with a fluorescent dye and applied, in solution, onto the slide. If cDNA is present which is complementary to the DNA sequence for individual genes present on the slide, binding will take place. The slide is then washed and those cDNAs which did not bind are removed. A laser then scans the slide and any bound cDNA will appear green. In this way, one can get a very broad view of the genes which are expressed in a give tumor.

Although the use of cDNA from one tumor sample labeled with one color can give some information, it does not give the investigator reliable information about how this gene expression would compare to a normal cell, or to another type of cancer cell. Therefore, in order to obtain more information of gene expression relative to some control sample, investigators most commonly simultaneously apply a second source of cDNA to the slide. This could represent RNA from other tumors that appear similar when evaluated under the microscope in order to compare how similar the cells are genetically. This could also represent cDNA from normal cells that represent the tissue of origin for the individual tumor (for instance, the use of cDNA from normal muscle when performing a microarray analysis with rhabdomyosarcoma samples). The second cDNA source might also be derived from a tumor which is histologically the same, but has differing clinical characteristics such as more aggressive clinical behavior. Whatever the choice of the second cell type is, the handling of it is essentially the same as that described above with one exception. The second cDNA is labeled with a red dye and therefore will show as a red spot after laser scanning. The computer then merges the images showing red spots with that showing green spots, which results in one image which can identify genes which are expressed in both (looks yellow due to the merging of red and green), neither (black), or over/under-expressed in one versus the other (red or green). A theoretical example of how such a microarray looks is shown in Figure 2a, whereas an example of real data is shown in Figure 2b.

Although the principles involved in microarray analyses are simple and represent very straightforward applications of our current understanding of molecular biology, the technology required to accurately perform microarray analysis and to interpret the results is highly complex. Currently there are two basic types of microarray used—one involves the use of DNA imprinted slides as described above. A second, developed by the Affymetrix company, called GeneChip uses oligonucleotides rather than DNA on the physical surface to which the cDNA binds and the application of red vs. green cDNA occurs across slides rather than within one individual slide. There are multiple other commercially available kits for microarray, and each requires careful approach so that appropriate controls are used to validate the results.

Using these techniques, much has been learned about gene expression in sarcomas. A few of the results will be highlighted here, but the interested reader is referred to a recent review by Greer et al. for a more detailed discussion.

• Khan and colleagues used microarray to compare gene expression across a variety of pediatric tumors which appear very similar under the microscope (small round blue cell tumors) but which are known to have distinct derivations and distinct clinical behaviors. Using microarray and complex approaches involving "artificial neural networks" which teach a computer how to sift through masses of data for critical elements, the computerized algorithm identified 96 genes which were able to allow correct classification of these tumors 100% of the time. This provided the important proof-of-principle that tumors which are clinically classified as an individual group show their own distinct patterns of gene expression, again confirming cancer as a genetic disease. Of course, one hopes that identification of such distinct patterns of genetic expression is the first step toward identifying critical genes which could be targeted with new drugs or other therapeutics. 
• Nagayama, studied gene expression in 13 synovial sarcomas and 34 other spindle-cell sarcomas. By comparing the similarity of gene expression through a technique called "hierarchical clustering analysis", it was found that synovial sarcomas clustered closely to malignant peripheral nerve sheath tumors. This is the first data to suggest that these two tumor types, which appear to be derived from different mesenchymal tissues and which do not share a similar chromosomal translocation, might however have similar inner workings. 
• Sjogren and co-workers evaluated extraskeletal myxoid chondrosarcomas (EMCs) with differing histologies and different chromosomal translocations. Despite these differences, the microarray revealed very similar gene expression and even provided the new insight that CHI3L1, a gene which encodes a protein involved in non-malignant conditions with disruption of extracellular matrix, is highly overexpressed in these tumors. This raises the suspicion that CHI3L1 could be a target for therapy in this disease and will no doubt give rise to subsequent more focused studies to understand the biology of this molecule in this tumor.
• Khanna et al. used microarray to compare gene expression in mouse osteosarcomas which were not highly metastatic vs. those which had a propensity to metastasize. They identified that although many thousands of genes were checked, only a few were differentially expressed between these two tumor which varied in their aggressiveness. Among these, they identified ezrin as a molecule which was overexpressed in the highly metastatic type. They went on to show that lowering ezrin levels diminished the ability to metastasize and also that clinical samples of osteosarcoma which had high levels of ezrin were associated with a poor prognosis. Thus, by sifting through thousands of genes using microarray, these investigators narrowed the field of potential perpetrators of metastases to a few genes, and have gone a long way toward proving that ezrin is one of the critical players in the metastatic process.

In summary, evaluation of gene expression in the complex and heterogeneous group of cancers called sarcomas is an area of vigorous research interest. This type of an approach, which can simultaneously assess many thousands of genes from a very small amount of tumor tissue, appears to be the best way to narrow down the list of possibilities of critical targets in these tumors. The technique is complex and requires careful controls and experienced investigators to be sure that the differences observed are real, and to efficiently sift though the immense amount of data which is generated. Despite these challenges, it is anticipated that such studies carried out during the course of this decade will ultimately allow the sarcoma community to go beyond the largely descriptive categorizations which we currently have, to those defined by a genetic signature which provides insight into the "inner workings" of these cells. It is hoped that the identification of critical genetic signatures will then lead to new therapeutic targets that will ultimately improve the outcome for patients with these diseases.

Summary

Recently introduced new technologies allow for a marked increase in the speed and efficiency with which human genes can be studied to determine their role in the development of cancers of all types, including sarcomas. By being able to examine the expression of essentially all genes in individual tumor samples, researchers hope to identify new diagnostic markers, new prognostic markers and new potential targets for therapy.

Introduction

The study of human disease, including that of malignant soft tissue tumors has been enormously helped by recently developed techniques that allow for a tremendous increase in the efficiency with which the genes involved in these diseases can be studied. To put this in perspective we have to realize that each cell in the human body contains DNA encoding for approximately 25-30,000 different genes. Previously, genes were studied on an individual, gene-by-gene, basis — a researcher would be interested in one or a few genes and would study the expression of these genes in a variety of diseases. Of course researchers would make smart decisions about which genes to study, but it was impossible with those techniques to study the expression for all the genes in a reasonable, acceptable time span. The new techniques of gene microarrays allow researchers to study almost all genes in a single experiment. A variety of different platforms exist to study tumors, but most gene array technologies can examine at least 10,000, and often many more, genes. This global approach in which essentially all genes are studied in a particular tumor type is fundamentally different from the approach in which one or a few candidate genes at a time were studied (Ref. 1). As one can imagine, the amount of data that is increased by these gene array studies is extremely large and the analysis of datasets in which the expression for, say 20,000 genes in 50 tumor samples, is so large that powerful computer programs are needed to make the data interpretable to researchers.

What are gene microarrays and tissue microarrays?

An excellent review of how gene arrays work and what they can do was already presented by Dr. Crystal Mackall and Dr. Javed Khan in Gene Profiling: Unlocking the Inner Workings of Sarcoma Cells. I will just mention a few additional aspects of this technology. Figure 1 shows a schematic drawing of the type of microarray that is used at Stanford University.

In order to become functional, each gene has to be translated into a protein and it is the proteins that form the building blocks of cells. First DNA is transcribed into messenger RNA (mRNA) which leaves the nucleus and then is translated into protein in the cytoplasm of the cells. After mRNA is isolated from tumor cells it is reverse transcribed into complementary DNA (cDNA) and labeled with a fluorescent marker (red fluorescent marker for the tumor and green for the reference). The mixture of red and green fluorescent cDNA fragments is then applied directly to the arrays. These arrays consist of a glass slide on which up to 40,000 different spots are printed in neat rows and columns. Each individual spot contains DNA for a specific gene and functions as a read-out for the amount of cDNA for that particular gene that is present in the samples. The level of fluorescence for the gene spots that are present on the array gives an indication of how much RNA was present in the tumor and reference samples. Using the same principle of hybridization one can also isolate DNA (instead of mRNA) from the same tumor cells, label it with fluorescent tags and apply it to the arrays. In this case the fluorescence level at a particular spot will give an indication of whether the gene is present in higher or lower copy numbers than normal in the nucleus of the cells. This is a very powerful technique that allows researchers to search for yet another aspect (the alterations in the DNA) of the tumor cells.

Similar to gene expression profiling, the comparative genomic hybridization technique also generates tremendous amounts of data for which specialized computer programs are necessary in order to evaluate the results. Gene expression profiling is best performed on frozen tissue samples although some success has been obtained when using formalin fixed material. Comparative genomic hybridization can use both fresh frozen and formalin fixed specimens.

A final technique that should be mentioned is one that uses tissue microarrays (TMAs). This technique was developed in its current form by Kallioniemi and Sauter and has revolutionized the way in which researchers perform studies on formalin-fixed, paraffin-embedded tissue (Kononen J, Bubendorf L, Kallioniemi A, Barlund M, Schraml P, Leighton S, Torhorst J, Mihatsch MJ, Sauter G, Kallioniemi OP, Tissue microarrays for high-throughput molecular profiling of tumor specimens, Nat Med. 1998 Jul; 4(7):844-7). During surgery material is removed either for diagnostic purposes (to see what a tumor mass consists of) or for therapeutic purposes (to remove the malignancy from the patient). After each surgery tissue is fixed in formalin and embedded in paraffin blocks. Surgical pathology departments in hospitals then cut thin sections from these blocks and these are used for diagnosis. The tissue microarray technique allows hundreds of tissue cores taken from the original paraffin blocks to be combined in a new tissue microarray block. When sections are cut from these blocks immunoperoxidase studies can now be performed on hundreds of tumor samples at the same time rather than one tumor at a time as was previously the case. Like the gene array studies mentioned above this has made a tremendous change in the way researchers study paraffin-embedded material and has resulted in much bigger datasets than previously were possible. Figure 2 shows a picture of a tissue microarray paraffin block.

All the dots in the block represent cores of individual patient tumors that are present as cylinders which we see head on in this paraffin block. It is important to realize that the two techniques of gene arrays and tissue microarrays are highly complementary. As shown in Figure 3 gene microarrays study tens of thousands of markers in relatively few samples.

These gene array studies usually lead to the finding that a relatively small group of genes may be of interest for further study. It is at this point that tissue microarrays become very valuable. The gene microarray experiments are difficult to perform, are expensive and often rely on the availability of frozen tissues. Tissue microarray experiments are much cheaper to perform and can use paraffin-embedded material. Thus while gene microarray studies can look at many, many markers in relatively few samples, tissue microarray technology can be used to further investigate a relatively small number of markers (as highlighted by the gene array studies) on extremely large numbers of tumor samples. As such, it forms a terrific validation technique for the gene microarray findings.

Why study sarcomas with these techniques?

The optimal treatment of cancer patients depends on individualized treatment for each patient. As a result of studies done before the discovery of the gene microarray technology, the identification of molecularly-defined subgroups of patients has improved treatment. One example of this is found in breast carcinoma, where the expression of two genes, estrogen receptor (ER) and HER2Neu, have clear, direct implications for the prognosis of individual patients. More importantly, these results are also being used to tailor molecularly-targeted treatment. A dramatic example is also the discovery of Gleevec (Imatinib Mesylate) as an effective drug for GIST.

Of course the primary goal of sarcoma research is to find better diagnostic and prognostic markers for sarcomas and hopefully to identify new therapeutic targets that may benefit patients. Sarcomas are unique in that they are relatively rare lesions, with 8,000 new malignant cases estimated annually in the United States. The number of 8,000 cases is likely to be an underestimate as with the new Gleevec therapy for GIST patients an extremely rapid accrual of new patients occurred, leading to a new estimate of 5,000 new cases annually for this disease alone.² Previously it had been estimated that only 300-500 new cases of GIST occurred in the United States annually. It seems therefore that the actual number of sarcomas is underreported. Nevertheless, sarcomas are rare lesions and there are relatively few clinicians (including pathologists) who have extensive experience with these tumors. As a result, there is unfamiliarity with histologic diagnosis of these tumors and with their treatment options. The situation is further complicated by the fact that within the group of soft tissue tumors large numbers of diagnostic subtypes pf sarcoma exist; over 100 different tumor types can be recognized by histologic examination. Finally, relatively few diagnostic markers exist to help surgical pathologists reach the correct diagnosis. The distinction between benign and malignant lesions can also be very difficult for tumors from within a single category, such as smooth muscle tumors.

Due to the rarity of sarcomas, it can be difficult for researchers to accumulate enough specimens to perform a meaningful study. Fortunately more and more researchers are starting to pool tissue specimens. Recently, patient organizations such as LMS Research Advocates, the National Leiomyosarcoma Foundation, and Life Raft Group have greatly help with organizing a collection of tumor specimens. Of course all such activities need to be performed with respect for the patient’s privacy and in agreement with Institutional Review Boards of the institutions where the research is to be performed. These initiatives could be examples for other patient organizations to get involved in such efforts.

The power of gene microarray analysis of soft tissue tumors is that it allows for a genome-wide search for novel diagnostic markers that separate tumors in different diagnostic classes. A large number of these studies have now been performed (Refs. 3-12), but these studies involved relatively small numbers of cases from different diagnostic types. The next step will be to distinguish subsets of tumors within a particular sarcoma type. The reasons to search for novel markers that allow the separation of subsets within a tumor type are threefold:

• First, markers that are expressed in only a subset of cases with a particular diagnosis may distinguish subtypes of tumors that remain imperceptible by conventional techniques. These subtypes of tumors can then be tested for a possible sensitivity to drug therapies that, when tested on the entire tumor group, might not be detected.
• Second, genes that are differentially expressed within a diagnostic category can also be tested for their potential value as prognostic markers.
• Third, by identifying cell surface markers specific for a particular tumor, new novel potential therapeutic targets can be discovered.

In addition to the potential benefits that may derive from these studies for patients that suffer from sarcoma, there are a number of other reasons that make sarcoma research important. These can be used as valid arguments to obtain funding to study this rare disease. Previous studies of soft tissue sarcomas (STS) have yielded results that transcend the relative rarity of these lesions. For example, the enormous success of Gleevec treatment for the control of GIST has functioned as a paradigm for novel targeted therapeutics in solid tumors. As a result major efforts are underway to develop novel small molecule inhibitors for a variety of cell surface receptors in many different tumors, including carcinomas.

The study of sarcoma can also yield insight into normal connective tissue or stroma. Stroma is currently a relatively uncharacterized tissue that contains a variety of cells; these include fibroblasts, endothelial cells, dendritic cells, myofibroblasts, pericytes, perivascular smooth muscle cells and presumably many other as yet unknown cell types. The function of many of these cells is unknown, and the study of mesenchymal development is frustrated by the lack of specific markers. Recent experiments in my laboratory showed that genes that are differentially expressed in two fibroblast-derived soft tissue tumors (desmoid type fibromatosis and solitary fibrous tumor) can be used to distinguish fibroblast-like cells with identical histology in different normal tissue sites. The biological significance of these findings became clear when we noticed that breast carcinomas that express one fibroblast-like gene set versus the other in their stroma showed a significant difference in clinical outcome.

The significance of STS research is also emphasized by the fact that in recent years four meetings were organized by the NCI to support sarcoma research:

1. State of the Science meeting, June 17-18, 2002
2. Sarcoma Progress Review Group planning meeting, June 30, 2003
3. Sarcoma Progress Review Group roundtable meeting, October 8-10, 2003
4. Sarcoma-Mesenchymal Stem Cell Workshop, September 27-28, 2004.

The last meeting included researchers on mesenchymal stem cells and made it clear that there is a great need for new markers that identify subsets of mesenchymal tissue cells.

Benign Smooth Muscle Tumors and Leiomyosarcoma

Tumors derived from smooth muscle form a large subset of the STS group. They are defined by the fact that in their histology they "resemble normal smooth muscle cells." This definition is based entirely on histologic appearance and this raises problems since myofibroblasts and fibroblasts can appear similar to smooth muscle cells but give rise to different tumor types with distinct clinical behaviors. The diagnosis of smooth muscle tumors (SMT), the determination of the expected clinical behavior of these lesions, and the separation from other lesions in the differential diagnosis continues to rely heavily on morphologic features of the tumor cells and new additional markers to determine smooth muscle versus fibroblastic and myofibroblastic differentiation are needed.

SMT can originate in the dermis, from large vessels in the retroperitoneum and thorax, in the deep soft tissue and in the uterus. The clinical behavior of SMT is difficult to predict based on histologic examination. In the deep soft tissue of the thigh for example, tumors with a bland histology and minimal mitotic activity can display recurrent and even metastatic behavior. In contrast, in the uterus, some tumors with a high mitotic activity and malignant appearing pleomorphic nuclei can be cured by local excision. In women of childbearing age accurate diagnosis is obviously extremely important since a benign SMT will allow the patient to undergo local resection and retain her uterus. These smooth muscle tumors of the uterus are quite common and frequently lead to diagnostic difficulties.

Several studies have recently appeared on gene expression profiling on relatively low numbers of smooth muscle tumors, between 8 and 17 cases per study (Refs. 5, 6, 10, 13, and 14) but did not completely address these considerations. We believe that the field would benefit from a much larger study specifically addressing these questions and hope that this will form a molecular basis for the distinction between clinically aggressive and benign smooth muscle tumors.

Gastrointestinal Stromal Tumor (GIST)

GISTs are tumors that occur in the wall of the bowel and are thought to be derived from the cells of Cajal, the pacemaker cells that drive peristalsis in the intestine. Prior to the arrival of STI-571 (Gleevec® in the US, Glivec® in Europe and also know as Imatinib, and "imatinib mesylate") therapy, surgery was the only effective treatment for these lesions and many recurred, resulting in a protracted disease process that often was fatal through local recurrences or metastases to liver and lungs. A dramatic improvement in survival was noted after the discovery that GISTs expressed high levels of the tyrosine kinase receptor KIT, and that their growth could be inhibited by Imatinib, a small molecule inhibitor specific for KIT. GIST is one of the first examples of a solid tumor that responds to a therapy specifically targeted to a marker on the cell surface and its treatment is paradigmatic as a novel approach to treating malignancy. First described as a powerful inhibitor of BCR-ABL, a fusion oncoprotein in CML, Gleevec was subsequently used to treat GIST where it can cause marked growth retardation or even regression. This treatment has resulted in a response for 50% of patients with unresectable GIST and a stabilization of disease in another 28%. After the reports of successful treatment of GIST by Imatinib (Refs. 15-16), the FDA has now approved the therapy (Ref. 17). Imatinib has been used with success in a few in other tumors such as DFSP (Refs. 18-19) but is thought to act through inhibition of another tyrosine kinase receptor, in this case PDGFRb.

After the initial success of Gleevec therapy in GIST, it now has become apparent that while the majority of GISTs show an impressive response, there is a subset of tumors that fail to react. In addition, several years into the novel treatment regimen, many initially responsive tumors become resistant to Imatinib therapy. Therefore, novel drug treatments are currently being investigated and new targets are eagerly sought. Mechanisms through which GIST which initially respond to Imatinib become resistant are poorly understood and are the subject of great interest. Several researcher groups hope to contribute to the identification of the genes responsible for Gleevec resistance by combining gene expression profiling, tissue microarray studies and other high-throughput technologies such as SNP arrays.

In a second approach that indirectly addresses Gleevec resistance, we and others have used DNA microarray analysis to determine the differences in gene expression between GISTs with distinct mutations (Refs. 20-22). We hope to extend this study in the future to a larger group of GISTs with known mutations.