January 16, 2008
Molecular Diagnostic Testing in Malignant Gliomas: A Practical Update on Predictive Markers
By Yip, Stephen Iafrate, A John; Louis, David N
Abstract Advances in understanding the molecular underpinnings of cancer and in molecular diagnostic technologies have changed the clinical practice of oncologic pathology. The development of targeted therapies against specific molecular alterations in cancer further portends changes in the role of the pathology laboratory to guide such custom therapies. To reconcile the flood of scientific discoveries in this area, the promises of highly touted novel therapeutics, and the practicality of applying this knowledge to the day-to-day practice of clinical neuropathology, the present review highlights the operative differences between diagnostic, predictive, and prognostic markers, and discusses issues surrounding the transition of prospective biomarkers to routine laboratory implementation. This review focuses on 3 predictive molecular markers that are either in clinical use or are contemplated for use in the evaluation of malignant gliomas: assessment of 1p/19q loss in oligodendroglial tumors, examination of O^sub 6^-methylguanine DNA methyltransferase promoter methylation status in glioblastomas, and molecular dissection of the epidermal growth factor receptor- phosphatidylinositol 3-kinase pathway in glioblastomas. Implementation of such predictive markers is not straightforward and requires critical review of the available literature and attention to practical laboratory, compliance, financial, and clinical management issues.
Key Words: Biomarkers, Glioblastoma, Glioma, Molecular diagnostics, Neuropathology, Oligodendroglioma.
Advances in understanding the molecular basis of disease and identification of disease biomarkers have heralded the era of molecular diagnostics in pathology. This development, however, comes with unique sets of challenges that will confront the pathologist. Indeed, the pathologist will be essential to evaluate scientific evidence critically and to determine whether a candidate molecular marker can provide information to alter disease management. The pathologist will also have to consider the practicality of adopting a new test in the clinical setting. Such considerations are now typically undertaken under increasing pressure from clinicians to incorporate testing for newly discovered markers. In this regard, the pathologist must balance the desire to bring the latest research findings to the bedside with the need to observe numerous checks and balances that ensure that newly incorporated markers are ready for clinical testing.
NOVEL MARKERS: BASIC CONSIDERATIONS
Biomarkers are important diagnostics tools in oncology; they also aid in monitoring disease progression and assist in determining prognosis and predicting therapeutic response (1, 2). Biomarkers vary from specific proteins and antigens to unique genetic, epigenetic, and proteomic profiles, but common to all markers is that they provide information specific to a disease process. They function to supplement, and rarely to supplant, the histopathologic examination of tissues that is the mainstay of traditional oncologic pathology.
Diagnostic, Prognostic, and/or Predictive
Before the laboratory implementation of a novel biomarker, there must be a clear definition of whether its use lies in the diagnostic, prognostic, or predictive realm. A diagnostic marker aids in the initial recognition of cancer or its relapse (3). A prognostic marker presents information pertaining to the natural history of disease and, therefore, the probability of disease relapse and durability of remission. A predictive marker identifies probability of response to a specific therapeutic agent or modality (4). Differentiating between these different types of markers is important so that one can make informed clinical decisions (5, 6). A single marker, however, may provide various combinations of diagnostic, prognostic, and predictive information. For example, expression of the estrogen receptor by immunohistochemistry is relevant in the diagnostic workup of metastatic lesions suspicious to be derived from a breast carcinoma primary, but it is also relevant to survival (prognostic) and potential response to hormonal interventional therapies (predictive) (7).
The clinical use of a potential marker is based on the magnitude of its diagnostic, prognostic, or predictive abilities and on the reliability of such changes. The former reflects the degree of clinical benefit that results from decisions based on information provided by the marker. For example, a candidate molecule that identifies tumors with a significantly enhanced response to a therapeutic agent can have dramatic impact on patient care. This applies not only to the identification of patients who may benefit from the agent but also to the exclusion of predicted nonresponders from unnecessary treatment, sparing them the untoward side effects and the expense of the agent. The second feature is marker reliability, that is, consistent association of the marker with specific clinical behavior and outcome. Implementation of markers into the clinical laboratory must be based on proper validation of assays to ensure that the results are reliable. When markers are both reliable and relevant to clinical care, they serve an important role in identifying clinically relevant of patient subgroups. However, when these 2 parameters are not met, the use of a new marker can complicate patient management and have deleterious consequences.
Issues Surrounding the Development of Novel Biomarkers
The development of biomarkers for early screening of malignancies has been divided into 5 phases, analogous to the development of therapeutics. Although the glioma markers discussed below are not screening markers per se, but rather function in patients already diagnosed with gliomas, the same systematic approach of evaluating markers certainly can be modified and adopted for tumor markers. The 5 phases range from preclinical evaluation to validation in retrospective studies, followed by prospective studies, and the realization of its impact on cancer management (8). Phase 1 consists of preclinical exploratory studies comparing diseased with nondiseased tissues with the aim of identifying unique markers or specific patterns. This can be achieved using immunohistochemistry, enzyme-linked immunosorbent assay, or, more commonly nowadays, data from one of the "-omics" technology such as gene expression microarray analysis or proteomic studies. Phase 2, which is clinical assay validation, attempts to confirm expression of the biomarker in small numbers of patients known to have the disease, in comparison to a normal cohort, thus establishing the true-positive and false- positive rates. Phase 3 consists of retrospective longitudinal repository studies that evaluate the ability of the biomarker to detect preclinical disease. Phase 4 moves the study to prospective screening in a relevant population. Population-based cancer control studies constitute the goal of Phase 5. Substandard performance of the biomarker in this phase, which assesses the "practicality" of the candidate marker, can potentially derail its adaptation for widespread clinical use. Certainly, Phases 1 and 2 are identical for markers that are used for cancer screening and for evaluating existing tumors, and the next 3 phases can be modified accordingly for tumor markers. In fact, the same general principles apply for the evaluation of real-world applicability of both screening and tumor markers (Table 1). For example, similar systematic approaches for the evaluation and implementation of tumor markers have been proposed by breast cancer researchers (9).
TABLE 1. Factors That Can Affect the Widespread Implementation of a Candidate Biomarker
Many molecules that are initially touted to have great promise do not make it to the final stages of acceptance as clinically validated markers (Fig. 1). In addition, the initial optimism associated with the discovery and characterization of such molecules is generally dampened by the time of widespread laboratory use. This is frequently associated with the publications of follow-up reports examining and reexamining the use of the putative marker in real- world studies with large numbers of clinical specimens. As a result, the eventual adaptation of a biomarker in daily pathology practice is dependent on many factors but hinges on its ease of use and standardization, as well as robustness when facing an inherently heterogenous pool of clinical pathologic specimens and the vagaries of daily routine laboratory practice (10, 11).
Compliance, Billing, and Reimbursement
Medical laboratory testing in the United States, including molecular diagnostics, is regulated under the Clinical Laboratory Improvement Amendments (12). Established in 1988, this set of legislations set out operating guidelines of excellence for the diagnostic laboratory that include specific regulations on quality control and assurance, method validation, personnel, and proficiency testing. Adherence to these regulations depends on the complexity of the individual tests. To ensure maintenance of quality assurance in the often-changing field of molecular diagnostics, an expert panel has made recommendations that include establishment of laboratory- focused consortia and databases to facilitate collaboration and reporting of test results, development of positive controls for test validation, improved performance evaluation program, and strengthening of existing molecular pathology training programs (13). Compliance with an established standard of excellence requires ongoing evaluation and reevaluation of testing procedures and outcomes, a cycle of activity that is especially important in a rapidly changing field (14, 15). In the clinical laboratory, this is best served by having a dedicated molecular diagnostic division that can maintain the necessary volume for the different assays for both cost-efficiency and for maintenance of competency. FIGURE 1. Typical sequence of events after the discovery of a biomarker; initial excitement is often tempered by real-world performance of the marker (see text).
Financial considerations are an important issue in integrating molecular testing into clinical pathology. The rapidity of assay development and the demand for new diagnostic procedures put significant pressure on realigning existing billing and reimbursement policies and contracts with the novel tests (16, 17). In addition, many of these molecular diagnostic methods require expensive capital expenditure for specialized equipment, reagents, and dedicated technical training. These factors also argue for the creation of a dedicated service that can handle molecular diagnostic needs from the different subspecialized areas of pathology. It is intended that the initial outlay in expenditure and the cost of maintenance of molecular diagnostics will prove to be economical in the long run by integration of molecular tests into the rest of traditional disease management (17). In an era of increasingly sophisticated and "personalized" (not to mention expensive) therapeutics, similarly personalized molecular diagnostics assays are essential in ensuring that only those patients with the appropriate molecular profiles receive the targeted therapeutic agent. Molecular diagnostics should help reduce or prevent the escalation of costs in a market that is becoming increasingly permeated with potent, molecularly targeted therapies (18). Currently, however, the number of molecular assays is expanding rapidly, and this already has major financial implications. For example, at our institution, test codes for molecular diagnostics have increased from 18 in 1998 to 134 in 2002 to 360 in 2006. This has brought about major increases in the total reference laboratory budget for our hospital, with molecular diagnostic expenses approximately doubling from $1 to $2 million (19).
Cost/benefit issues and the perceived extent of benefits to the population at large also determine the eventual adaptation of the marker for clinical use. On a practical note, widespread acceptance and clinical adaptation of the marker by the scientific and medical community are usually followed by acceptance from the various financial payers such as Medicare and private insurers. In reality, payers are often slow to reimburse testing associated with new technological advances. For example, with the current procedural (CPT) codes, there are no appropriate codes for large-scale microarray-type gene expression or genomic surveys (there are 3 new codes for small-scale arrays, not for arrays interrogating thousands or hundreds of thousands of targets). Introduction and acceptance of these assays in clinical service is just a matter of time; therefore, all parties should anticipate technological trends to ensure appropriate coding and, therefore, fair financial compensations for novel molecular diagnostic procedures. Acceptance of even more standard molecular assays, however, may vary from payer to payer and from state to state; for example, in our experience in Massachusetts, the combination of brain tumor diagnosis (International Classification of Diseases, 9th Revision) codes and molecular procedural (CPT) codes are interpreted by current local Medicare coding rules as medically unnecessary, and this often results in nonpayment from Medicare, whereas payment is allowed when the same CPT codes are aligned with International Classification of Diseases, 9th Revision codes for other cancer sites. These agencies will demand proof of cost-effectiveness and accountability. Abuse and misuse of these novel tests (many of which remain relatively costly), especially in billing, can have significant financial/ legal consequences. It is the responsibility of the practicing pathologist to be up-to-date with the recommended guidelines to order these tests appropriately (20, 21). In this regard, a recent analysis proposed a hierarchic approach to the implementation of biomarker development with coevolvement of therapeutic options; they also placed special emphasis on the education of patients and clinicians on various aspects of the biomarker assay, including evidence gaps and available and pending therapeutic options (9). Similar to therapeutics, there should be collaborative postmarketing surveillance of approved biomarker assays. Equally importantly, issues of privacy and protection against discrimination based on biomarker information must be addressed (9). Such practical demands often affect biomarker adoption and sometimes necessitate more complex marker measurements being restricted to large medical centers or private national laboratories.
PREDICTIVE MOLECULAR MARKERS IN DIAGNOSTIC NEUROONCOLOGY
Improvement in our understanding of the molecular basis of brain neoplasia and advances in technology are slowly transforming the field of brain tumor diagnostics (22). There now exist several examples of molecular markers in neuropathology that serve essential roles in the evaluation of tumors with ambiguous/difficult histologic features. A particularly compelling example is INI1 testing in atypical teratoid/rhabdoid tumor, which also illustrates the continuous adaptation to evolving technology in the life of a biomarker, having shifted from fluorescent in situ hybridization (FISH) assays to gene sequencing to immunohistochemistry in a matter of a decade (23, 24, 25). In addition, the same obstacles that stand in the way of integration of molecular diagnostics into traditional pathology service (discussed above) also apply to neuropathology (26). The issues of initial capital outlay and maintenance of programs, as well as the capability to recover these costs through billing and reimbursement, will continue to present challenges to the practicing neuropathologist.
The remainder of this review summarizes the data on the 3 sets of predictive markers that have been most recommended for the clinical evaluation of malignant gliomas: 1p/19q loss, O^sub 6^- methylguanine DNA methyltransferase (MGMT) inactivation, and epidermal growth factor receptor (EGFR)-phosphatidylinositol 3- kinase (PI3K) pathway activation. We have restricted the review to predictive markers because such markers, given their potential use in directing therapies, are often rapidly requested by clinicians once their roles have been suggested. Prognostic markers, on the other hand, are far more numerous-as can be seen by a quick perusal of most pathology and oncology journals-but are not as often demanded by either oncologists or patients.
For each of these markers, consideration is given to existing evidence to support use of the marker, the current status of assays for measuring the marker, use of the marker in clinical trials, and use of the marker in routine clinical practice. By working through these parameters, a report card can be issued for each marker, which in turn enables the pathologist to appreciate at what stage the marker stands in its development.
Chromosome 1p and 19q Analysis in Oligodendroglial Tumors
Rationale and Evidence for Marker Relevance
The biologic basis of why deletions of chromosomal arms 1p and 19q are associated with markedly improved clinical courses in patients with oligodendroglial tumors is unclear; the rationale for undertaking such analysis is based solely on empirical observations published in the literature. In 1998, Cairncross et al (27) reported the clinical implications of allelic loss of 1p and 19q in patients with histologically defined anaplastic oligodendroglioma. They found that loss of 1p was a predictor of chemosensitivity to the procarbazine-lomustine-vincristine (PCV) chemotherapy regimen. Furthermore, combined 1p/19q loss was associated with enhanced chemosensitivity and longer overall survival. Although this proved to be a reproducible observation that spawned widespread use of 1p/ 19q analysis in clinical neuro-oncology, the underlying biologic reason for these associations has thwarted explanation. For example, the putative tumor suppressor genes on 1p and 19q have not been identified despite extensive work (28-32), and recent evidence suggests the deletions may reflect whole-chromosome arm loss as a result of a centromeric translocation of 1 and 19 (33, 34). Moreover, 1p/19q loss is inversely related to TP53 mutations and 2 genomic abnormalities frequently found in astrocytomas: 10q deletions and amplification of the EGFR gene (35). In addition, 1p/ 19q genotype seems related to tumor location in the brain and, therefore, possibly to cell of origin (36, 37). Thus, the favorable phenotype can result from a combination of molecular factors for which 1p/19q loss is simply a convenient marker. However, to date, genetic profiling at the DNA and RNA levels has not defined such a combination of molecular factors (32, 38). Despite the lack of a biologic mechanism, clinical 1p/19q testing is the most mature molecular diagnostic test in the evaluation of malignant gliomas.
The 1p/19q story has already generated a couple of hundred scientific articles that have ranged from analyses of individual candidate genes to moderately large clinical correlation studies. The clinical correlations have definitively shown that those anaplastic oligodendrogliomas with 1p and 19q deletions are clinically distinct from those histologically similar tumors that lack deletions (39, 40). The observations have also generated substantial discussion and debate that has centered on 2 major open questions: (1) In which clinicopathologic setting is 1p/19q testing useful? For example, should testing be extended to low-grade oligodendrogliomas and to oligoastrocytomas in addition to the anaplastic oligodendrogliomas? (2) Is the test really predictive or simply prognostic, or a combination of predictive and prognostic? Can the test also be used diagnostically, that is, to define oligodendroglioma? The answer to the initial question is emerging through published studies and clinical use in multiple large institutions: 1p/19q testing may prove useful in the setting of any "oligodendroglial" tumor, whether high-grade or low-grade, whether pure or mixed with astrocytoma. For instance, in addition to the clear situation for anaplastic oligodendrogliomas, 2 articles have suggested that the slow but steady radiologic responses of low- grade oligodendroglial tumors to temozolomide correlate with loss of 1p (41, 42). Another has demonstrated that prognosis in oligoastrocytomas relates to 1p/19q status, with those patients whose oligoastrocytomas had 1p/19q deletion having better prognoses than those with p53 expression; this article, however, did not look at therapeutic response as a parameter, and, therefore, the predictive power of 1p/19q was not assessed (43).
The answer to the second question is less clear, but is likely to involve a group of biologic features that affect both natural history (i.e. prognosis) and overall therapeutic sensitivity (i.e. prediction, although possibly not response to a few specific therapies). For example, although the initial studies addressed response to PCV chemotherapy, more recent studies have shown that 1p/ 19q status can predict response to temozolomide as well (41, 44). Furthermore, 2 recent prospective randomized Phase 3 trials validated 1p/ 19q status prediction with regard to PCV chemoresponsiveness but also showed that patients with 1p/19q- deleted tumors fared better when treated with radiation alone (45, 46). A prior study had also demonstrated that patients whose oligodendrogliomas had 1p/19q loss had longer times to progression after radiation therapy with or without chemotherapy (47). In combination, such data suggest that the predictive value of 1p/19q testing is not strictly related to chemosensitivity, but may reflect a broader therapeutic sensitivity. Finally, given the interrelated nature of patient survival with therapeutic intervention, it remains possible that 1p/19q status is also related to natural history of the disease, with 1p/19q-deleted tumors being more slowly growing lesions, but that assessment of natural history is not possible given the overwhelming likelihood of therapeutic intervention.
At the same time, it is important to realize that the 1p/ 19q story is not the whole story. For example, a small number of patients with intact 1p/19q might benefit from PCV therapy (48). Clearly, in this regard, further identification and characterization of molecularly distinct subsets of patients within the 1p/19q- defined groups-either with regard to current therapeutics (PCV, temozolomide, radiation) or future targeted therapy-are necessary.
Status of Assays
1p/19q status can be assessed by loss of heterozygosity (LOH) assays, FISH, array comparative genomic hybridization (aCGH) and an assortment of less commonly used methods (49, 50). The 3 major techniques are listed in Table 2 along with their positive and negative attributes. The first 2 methods are relatively common clinical molecular diagnostic techniques that have their unique advantages and disadvantages. In general, the choice of one over the other depends primarily on local expertise, existing laboratory capabilities, and pathologist preferences. Other techniques such as quantitative polymerase chain reaction (PCR) have also been used but have not proven to be popular for clinical use (51).
Polymerase chain reaction-based LOH assays detect allelic losses by comparing polymorphisms between tumor and normal DNA (which is most commonly isolated from peripheral blood leukocytes). In tumors with 1p or 19q loss, there is loss of 1 of the normal heterozygous alleles (if the maternal and paternal copies can be distinguished, the polymorphism is said to be heterozygous or informative) (Fig. 2A). Polymerase chain reaction primers are designed against unique microsatellite sequences at each chromosomal region of interest. Because not all patients are heterozygous at all test loci, multiple (often 3) microsatellite loci are assessed to ensure that enough informative loci will be available to detect loss. Because the technique is allele based, rather than chromosomal number based (see below), an advantage of an LOH assay is that it can detect mitotic recombination events in which there are 2 copies of a single allele in tumor cells; however, mitotic recombination is a common mechanism for TP53/17p loss in malignant gliomas but not for 1p or 19q loss (53). A significant logistic limitation to LOH assays is the requirement for constitutive DNA, typically from blood, because this requires obtaining blood from patients (sometimes after they have left the hospital) and coordination of 2 specimens and their corresponding DNA. Typically, the pathologist selects a paraffin block of formalin-fixed tissue that adequately represents the tumor, that is, preferentially containing an abundance of viable neoplastic oligodendroglial cells with minimal contamination from other tissue type and tumor necrosis. Usually, multiple scrolls of the block are collected, followed by DNA extraction. The subsequent PCR step can be automated and batched, and reading of the subsequent electrophoretic tracings is generally straightforward and can be quantitated. Polymerase chain reaction assays are easy to develop and are moderately scaleable, allowing the possibility for growth of molecular diagnostic assays in the future.
TABLE 2. Comparison of 3 Common Molecular Assays for 1p/19q Status Determination
FIGURE 2. (A) 1p and 19q loss of heterozygosity (LOH) PCR. Capillary electrophoresis chromatogram generated after PCR amplification of DNA isolated from peripheral blood (bottom) and tumor (top) from a patient with oligodendroglioma. Loss of heterozygosity is noted for 2 markers in the tumor. The normal DNA shows heterozygous peaks for both markers (blue peaks), with marker 1 (chromosome 1p) showing 112- and 114-bp alleles and marker 2 (chromosome 19q) showing 158- and 170-bp alleles. Size markers are represented as red peaks. (B) 1p fluorescence in situ hybridization (FISH). FISH Dual-color FISH of an oligodendroglioma specimen showing loss of chromosome 1p as illustrated by the single red (1p probe) in comparison to 2 green signals (1q probe) per cell. (C) Array CGH of anaplastic oligodendroglioma. Array CCH performed using DNA from an anaplastic oligodendroglioma demonstrated loss of 1p and 19q with concordance to locations of markers used for LOH PCR analysis (red arrows) and relative to bacterial artificial chromosomes used as FISH probes (green arrows). Sex chromosomes were mismatched as internal hybridization control. Figure courtesy of Dr. Gayatry Mohapatra, Massachusetts General Hospital, Boston, MA.
Fluorescent in situ hybridization uses unique fluorescent probes that are hybridized directly to tissue sections, permitting direct microscopic evaluation of chromosomal copy number in tumor cells (Fig. 2B). Fluorescent in situ hybridization can be performed on formalin-fixed, paraffin-embedded tissue (as tissue sections or as dissociated cells from thick tissue slices) and can be used in conjunction with tissue microarray to maximize efficiency (54, 55). Importantly, FISH obviates the need for constitutive DNA and is a familiar technique in most pathology departments. Another advantage of this technique is the relative preservation of histologic detail and, therefore, the ability to study cells with obvious histologic abnormalities or cells from a specific population. The disadvantages of FISH include its relatively low scalability in the setting of multiple assays, and it is relatively labor-intensive. In addition, FISH lacks the ability to detect mitotic recombination events (because 2 copies of the same allele are present); fortunately for testing of oligodendrogliomas, mitotic recombination does not underlie allelic loss in these tumors (56). On the whole, FISH seems to be the most popular technique in current use in the United States for 1p/ 19q evaluation.
Array CGH uses genome-wide differential labeling of tumor and normal DNA for a comparative hybridization to arrays of DNA (57). The comparative hybridization of the differentially labeled tumor and normal DNA provides an estimate of copy number at every chromosomal locus on the array (Fig. 2C). A variety of arrays have been used: some feature genome-wide representation with chromosomal loci as either large bacterial artificial chromosomes, cDNAs, or relatively short oligonucleotides; others feature genomic regions specific to a certain disease process or group of processes (58). Array CGH is a powerful method primarily because it provides the capability to screen the entire genome and is, thus, the ultimate in scaleable assays. Although aCGH lacks the ability to detect mitotic recombination events with standard arrays, the use of single nucleotide polymorphism arrays allows detection of such events (59, 60). At the present time, aCGH remains a predominantly experimental technology, and the current cost of implementation for widespread diagnostic use may be prohibitive for many pathology departments. However, in the future, aCGH may be the most commonly used assay for assessing chromosomal copy number changes at all different genomic sites and in all different human tumors. Recommendations for Routine and Clinical Trial Testing: Diagnostic Use
The strong association of 1p/19q loss with oligodendroglioma histology has raised the question of whether oligodendrogliomas should be defined by this genetic signature, much in the same way as some sarcomas and hematologic malignancies are now defined by specific genetic abnormalities. Although 1p/19q loss is very common in oligodendrogliomas, being found in approximately 60% to 80% of cases in most series, this genetic combination is not present in all histologically defined oligodendrogliomas (49, 61). This begs the question of what to call those tumors that microscopically qualify as oligodendrogliomas but that do not harbor 1p/19q loss-a question that cannot be answered at the present time. Furthermore, there seem to be correlations between histologic features and genotype in that oligodendrogliomas with classic histologic features are preferentially associated with 1p/19q loss (62). Such associations complicate any simplistic assessment of how to define these tumors.
One approach has been to perform 1p/19q analysis in those tumors that are difficult to classify histologically as either oligodendroglioma or oligoastrocytoma. In most neuropathology practices, such diagnostically challenging cases are not uncommon. Furthermore, in our experience reviewing tumors called oligodendroglioma by general pathologists, we have found that knowledge of 1p/19q status can provide a method for training one's eye to recognize more classical oligodendrogliomas and, therefore, to homogenize the diagnosis of oligodendroglioma somewhat (63); setting such standards can be useful and can counter the more indiscriminate use of the term "oligodendroglioma" by pathologists who do not want to miss a potentially chemosensitive tumor. In these diagnostic settings, it has been tempting to conclude that the presence of 1p/19q loss means that the tumor is an oligodendroglial neoplasm, and that the absence of 1p/19q loss means that it is not. Although we agree that assessment of 1p/19q status in such cases may provide information on prognosis and possibly on response to therapy, we do not advocate using the genetic information to change a histologic diagnosis. This may become the practice in the future if additional data supporting this approach were published, but there are no sufficient published data at the present time to do so.
Practically, if the goal of diagnosis is to stratify tumors for prognostic and predictive reasons, then one should place combined emphasis on both tumor histology and 1p/19q status. Indeed, this was the conclusion of the most recent World Health Organization Classification of Tumours of the Central Nervous System consensus meeting, which defined oligodendroglioma as "a diffusely infiltrating, well-differentiated glioma of adults, typically located in the cerebral hemispheres, composed of neoplastic cells morphologically resembling oligodendroglia and often harbouring deletions of chromosomal arms 1p and 19q." (64).
Recommendations for Routine and Clinical Trial Testing: Prognostic Significance and Prediction of Therapeutic Sensitivity
The central issue with regard to 1p/19q testing in glioma patients is how results influence clinical decision making. Because analysis of 1p/19q has only been adopted within the past 10 years, many questions remain unanswered, and practice varies from institution to institution and from neuro-oncologist to neuro- oncologist. Nonetheless, there are data on usage from a recent survey of practicing neurooncologists, primarily in North America, who treat patients with anaplastic oligodendroglioma (65). Of the 93 respondents, 75.3% reported ordering 1p/19q testing on anaplastic oligodendroglioma patients 76% to 100% of the time; only 5.4% of the respondents never ordered 1p/19q testing. Thus, the test is widely and frequently used, and it is possible that the few neuro- oncologists not ordering the test may not have ready access to testing. This survey also cataloged therapeutic choices based on 1p/ 19q status in the same cohort of patients: either temozolomide, followed by radiation therapy; concurrent temozolomide/radiation therapy, followed by temozolomide; radiation, followed by temozolomide; or temozolomide only. In those patients whose anaplastic oligodendrogliomas had intact (i.e. no loss) 1p/19q, the most common treatment regimen from the respondents was concurrent temozolomide/ radiation, followed by temozolomide. On the other hand, in those patients whose anaplastic oligodendrogliomas had 1p/ 19q loss, the most common treatment regimen from the respondents was temozolomide alone. The specific reasons behind this therapeutic shift away from early radiation were not defined in the survey, but most neuro-oncologists have rationalized this shift by citing the potential neurotoxic effects of radiation, particularly in the setting of patients who have large tumors and/or are expected to have long survivals. Interestingly, for patients with loss of 1p but intact 19q, the therapeutic choices were intermediate between the 1p/ 19q loss and 1p/19q intact cases, perhaps as a result of the data showing that tumors with 1p loss/19q intact were sensitive to PCV chemotherapy but did not have as long survival rates as combined 1p/ 19q loss cases (27, 40). Therefore, although the optimal therapeutic courses for different genetic subsets of anaplastic oligodendroglioma have not been defined, the survey demonstrates that the results of 1p/19q testing are influencing therapeutic decisions in patients with anaplastic oligodendroglioma.
For low-grade oligodendrogliomas, the situation is even less defined. In our experience, we have seen the results of 1p/ 19q testing used for both prognostic and predictive purposes in different patient settings. For example, in the setting of a patient with a Grade II oligodendroglioma in which signs and symptoms (e.g. seizures) are well controlled, the finding of 1p/ 19q loss can be used as a rationale to watch such a patient carefully over time under the assumption that a low-grade oligodendroglioma with 1p/19q loss will be a very slowly growing tumor. On the other hand, in the setting of a symptomatic Grade II oligodendroglioma, the finding of 1p/19q loss can be used as a rationale for early treatment with temozolomide under the assumption that the tumor will likely display chemosensitivity. In such scenarios, therefore, use of the test is highly dependent on individual patient issues.
For oligoastrocytic tumors, the situation is also poorly defined. Many would argue, as shown in some articles (43), that those oligoastrocytic tumors with 1p/19q loss are more likely to behave like oligodendrogliomas, and that those with p53 alterations and 1p/ 19q intact are more likely to behave like astrocytomas. The corollary for treatment would be to take a more aggressive therapeutic approach for patients whose tumors have a more astrocytoma-like genotype than those that have a more oligodendroglioma-like genotype, but the issue has not been studied in any definitive manner. Again, in such settings, use of the test is highly dependent on individual neuro-oncologists' practices and on individual patient issues.
For clinical trials of oligodendroglial tumors, 1p/19q testing is mandatory. An open question, however, is whether such testing should be used up front to guide placement into 1 treatment arm versus another or if such testing should be a retrospective analysis parameter. Clearly, stratification in any trial of oligodendroglioma needs to take into account 1p/19q status.
In summary, 1p/19q LOH testing plays a common role in the management of patients with oligodendroglial neoplasms, primarily by fine-tuning prognostic or predictive estimates. Current assays are robust, and the resulting information complements, but does not replace, histologic diagnosis. In our opinion, 1p/19q testing should be performed on all glial tumors suspected to have an oligodendroglial component, and the results linked to tumor histologic findings. Ongoing diagnostic issues include further molecular dissection of subgroups and the eventual development of consensus treatment paradigms.
MGMT Testing in Glioblastomas
Rationale and Evidence for Marker Relevance
Many chemotherapeutic agents used in the treatment of glioblastoma are alkylating agents. Although nitrosoureas such as lomustine and carmustine were the most common agents used in the past (66), there has been a marked shift in recent years to the oral imidazotetrazine derivative temozolomide. The primary mechanism of action of temozolomide is the addition of methyl groups to the O^sub 6^ position of guanine to produce lethal methylguanine adducts. The DNA repair enzyme MGMT reverses this process by repairing methylguanine adducts; MGMT effects a stoichiometric and irreversible transfer of the methyl group at the O^sub 6^ position of the modified guanine to a cysteine residue of MGMT, with restoration of the normal configuration of the nucleotide. Concurrently, the finite pool of cellular MGMT enzyme is gradually depleted in the presence of large amounts of O^sub 6^-modified guanine. As a result, inactivation of MGMT leads to a decreased ability to repair DNA and a corresponding increase in cytotoxicity. It has therefore been logically hypothesized that a decrease in MGMT expression would correlate with an increased sensitivity to such alkylating agents. In fact, O^sub 6^-benzylguanine, an irreversible inhibitor of MGMT that depletes tumor of MGMT activity, has been shown to enhance in vitro and in vivo cytotoxicity of alkylating agents against tumors (67), and this agent has been tested as a chemosensitizer in glioma patients (68, 69). Most significantly for this review, a body of literature supports the hypothesis that the level of tumor MGMT correlates with survival (70, 71) and with alkylating agent sensitivity in gliomas (72, 73). The latter predictive correlation is discussed below. In gliomas and other tumors, epigenetic silencing of MGMT gene transcription secondary to promoter hypermethylation, rather than gene mutation or deletion, has been observed as the major mechanism for MGMT inactivation (74, 75). At the protein level, however, MGMT protein expression is heterogeneous within tumors and within small regions of tumors, suggesting molecular heterogeneity as well. As discussed below, this heterogeneity complicates in situ interpretations of MGMT expression.
In 2005, a multinational collaborative trial reported that MGMT gene silencing predicted enhanced benefits in glioblastoma patients treated concurrently with radiotherapy and temozolomide (76). Conversely, patients with tumors not methylated at the MGMT promoter did not benefit as much from this therapy and showed only marginal improvement in median overall survival compared with radiotherapy alone. The authors concluded that knowledge of MGMT promoter methylation status could help guide decision making: adjuvant temozolomide would be incorporated into the standard treatment regimen for patients whose glioblastomas had methylated MGMT promoters, but potentially withheld from patients whose glioblastomas did not have methylated MGMT promoters. This would help to minimize unnecessary hematopoietic toxicity in the patient population least likely to derive therapeutic benefit from temozolomide (76). Another recent study suggested that MGMT promoter methylation status was useful in predicting temozolomide responsiveness in low-grade gliomas and highlighted the possibility of applying MGMT testing in patients with lower-grade gliomas (77).
Status of Assays
O^sub 6^-Methylguanine DNA methyltransferase promoter methylation status has been evaluated primarily via methylation-specific PCR (MSP) of sequences within the CpG-rich MGMT promoter region. Methylation-specific PCR relies on the differential susceptibility of methylated versus unmethylated cytosines to sodium bisulfite modification, which leads to selective amplification with primers specific to either the originally methylated or unmethylated sequence (Fig. 3A) (78). Methylation-specific PCR lends itself to routine laboratory testing, and can be used on DNA from formalin- fixed paraffin-embedded tissue. Nevertheless, the technique remains highly dependent on tissue quality and quantity, the specificity of the primers selected, the adequacy of bisulfite treatment, and PCR conditions. In addition, the inherent heterogeneity of glioblastomas also makes interpretation problematic; it is not uncommon to see amplification of both methylated and unmethylated MGMT promoter sequences in the same specimen, which may represent tumor cell heterogeneity or the presence of nonneoplastic cells.
For the above reasons, laboratory testing for MGMT promoter methylation is not straightforward. Nonetheless, given the use of MSP to evaluate other disease loci such as promoter hypermethylation of the mismatch repair gene MLH1 in patients with sporadic colorectal cancer (79), optimization and standardization of MGMT MSP assays should be possible (80). There are also several other bisulfite-based analytic techniques, including quantitative MSP, bisulfite sequencing, methylation-sensitive single-strand conformation analysis, and mass spectrometric-based quantitative analysis, and some have been used to analyze MGMT (78, 81, 82).
Because transcriptional silencing is the end result of promoter methylation, examination of MGMT protein expression by immunohistochemistry would seem a logical approach to MGMT analysis (Fig. 3B). This has the further advantages of being relatively inexpensive and able to be performed in any clinical diagnostic laboratory (83-85). Unfortunately, there is marked intratumoral heterogeneity for MGMT immunoreactivity in a substantial number of malignant gliomas that complicates interpretation. Although it is possible to issue a report citing the percentage of positive cells in a given area, the significance of such numeric indices is not clear. Moreover, the immunohistochemical assays for MGMT are not necessarily robust or consistent, and some laboratories have not been successful in optimizing these assays for clinical use. Finally, although one would expect concordance between MGMT promoter methylation status and MGMT protein expression, discrepancies have been reported in gliomas (80, 86) and in other tumors as well (87, 88). In fact, other variables such as methylation dosage and methylation status of other regions of the MGMT gene also seem to contribute to transcriptional control (89, 90).
FIGURE 3. (A) O^sub 6^-Methylguanine DNA methyltransferase (MGMT). Methylation-specific PCR (MSP) of primary glioblastoma cases illustrating promoter hypermethylation (Case 1) and absence of methylation (Cases 2 and 3). In Case 1, PCR reaction using methylation-specific primers for the MGMT promoter generated amplification product equivalent in size to that of the positive control (+CON). Interestingly, primers specific to the unmethylated MGMT promoter sequence generated a faint band of amplification product in Case 1. This highlights the inherent heterogeneity of the cellular material used for MGMT MSP testing with mixtures of hypermethylated and unmethylated MGMT promoters. On the other hand, Cases 2 and 3 showed clearly the absence of MGMT promoter hypermethylation by the absence of products using methylation- specific primers but generation of PCR product identical in size to that of the negative control (-CON). (B) O^sub 6^-Methylguanine DNA methyltransferase immunohistochemistry. Tissue section from a glioblastoma specimen unmethylated at the MGMT promoter (by MGMT MSP) demonstrated concordant nuclear expression of the MGMT protein in tumor cells. However, MGMT immunoreactivity is heterogeneous in the tumor tissue. O^sub 6^-Methylguanine DNA methyltransferase immunoreactivity of the vascular cells act as internal positive control (magnification x200).
Recommendations for Routine and Clinical Trial Testing
As discussed above, MSP is currently the most established technique to assess MGMT and is the technique for which the most convincing clinical correlations have been reported (76). Going forward, a better understanding of the clinical impact of abrogation of MGMT expression via promoter hypermethylation will help to accelerate adoption of this test and immunohistochemical assessment in clinical practice.
Despite concerns about assay quality and significance, MGMT status at both the gene methylation and protein levels has been correlated with prognosis in patients with glioblastoma. However, because prognostic assessments in glioblastoma are typically not of major use (i.e. only implying modest differences in survival), MGMT testing is not typically undertaken for prognostic purposes. We would further argue that predictive testing for MGMT is also not ready for routine clinical implementation at the present time. Although the data have demonstrated differences in significance between the survival benefits conferred by temozolomide in patients whose glioblastomas have MGMT methylation versus those whose tumors lack MGMT methylation, there was a trend that nearly achieved significance (p = 0.06) for temozolomide benefit in patients whose tumors lacked methylation (76). Given that temozolomide is a well- tolerated oral agent and the only agent in the last 2 decades to show a definite survival benefit (91), it seems highly unlikely that temozolomide would be withheld from a patient whose glioblastoma lacked MGMT promoter methylation. One can argue that negative testing might enable a decision to stop temozolomide therapy in a patient whose tumor demonstrated radiologic progression during therapy, but that same decision might be reached as well on the basis of clinical and radiologic data. Therefore, it seems unlikely that MGMT promoter methylation assays will play an important predictive role until the time that an alternative chemotherapy to temozolomide exists. However, for clinical trials, particularly those that involve alkylating agents such as temozolomide, MGMT testing will be an important analysis parameter.
EGFR-PI3K Pathway Evaluation in Glioblastoma
Rationale and Evidence for Marker Relevance
Aberrant expression of proteins implicated in growth pathways is common in glioblastomas. Because many of these proteins are also potential targets for therapeutic intervention using specific molecular inhibitors, their detection may have significant predictive potential. Indeed, there have been dramatic successes using this approach in systemic cancers in recent years, as shown by the use of imatinib mesylate (Gleevec; Novartis, Basel, Switzerland), a small molecule tyrosine kinase inhibitor that preferentially targets the BCR-ABL fusion protein in chronic myelogenous leukemia and the C-KIT oncogene in gastrointestinal stromal tumor (92, 93), and the use the small molecule EGFR inhibitors gefitinib (Iressa; AstraZeneca, London, UK) and erlotinib (Tarceva; Genentech/OSIP, San Francisco, CA) nonsmall cell lung cancer (NSCLC) (94, 95). In NSCLC patients, EGFR inhibitors have marked effects in those tumors that harbor specific activating mutations in the EGFR gene (96, 97). The role of identifying such key translocations and mutations increasingly resides in the molecular diagnostics laboratory, and the rapid surge in the identification and validation of molecular targets with therapeutic and diagnostic potentials will increasingly act as the impetus to adopt these markers and to develop diagnostic assays in the clinical setting (98, 99). For glioblastoma, as discussed below, EGFR represents a logical target for such inhibitors, and, therefore, molecular evaluation of the EGFR pathway correspondingly represents a logical focal point for developing predictive markers in the setting of such inhibitors. Epidermal growth factor receptor has long been implicated in glioblastoma (100-103). Up-regulation of EGFR signaling through amplification of the gene or activating mutations is common in glioblastomas, occurring in approximately one third of cases (104). The unique EGFRvIII mutant, which is identified in half of those glioblastomas with EGFR amplification, seems to be a constitutively active receptor (105). Missense mutations may also be common within the exons encoding the extracellular domains of EGFR (106). Although such genetic changes seem frequent, most studies of both wild-type EGFR and the vIII mutant have not identified a clear prognostic role for EGFR status in the setting of glioblastoma. Nonetheless, as discussed below, there may be a predictive role for assessing EGFR at the genetic or protein level.
Epidermal growth factor receptor represents an attractive target for molecular inhibition for a number of reasons. One, the EGFRvIII mutant receptor is nearly unique to glioma cells, which makes it an ideal target (107). Two, EGFR amplification and overexpression, when present in glioblastomas, are often found in extremely high levels. Three, the recently described mutations in the extracellular domains of EGFR have been demonstrated in vitro to have oncogenic effects and also to convey sensitivity to a small molecule EGFR tyrosine kinase inhibitor (106). These features suggest that analysis of EGFR, at either a genetic or protein expression level, can provide information relevant to therapeutic response.
Complicating this simplistic assessment, however, is a complex and somewhat redundant signal transduction network that bridges EGFR at the cell surface to its eventual oncogenic effects in the nucleus. At least 1 key effector cascade is the PI3K pathway that involves a variety of key molecules known to be dysregulated in glioblastoma, notably AKT, mammalian target of rapamycin (mTOR), and phosphatase and tensin homolog (PTEN) (108). The complex interplay among these molecules and the extent of pathway redundancy may confound decision making based on simple molecular assays, especially with regard to which patient population might derive benefit from small molecule tyrosine kinase inhibitors. Nonetheless, 2 studies have identified relatively similar molecular changes that may correlate with response to EGFR inhibition in glioblastomas.
In 1 study, analysis of 41 glioblastomas showed 7 of 13 patients with tumors with diffuse EGFR immunopositivity and p-AKT immunonegativity (EGFR+/p-AKT-) responded clinically to erlotinib, whereas only 1 in 5 from the EGFR-/p-AKT- group responded (109). Epidermal growth factor receptor gene amplification or EGFR expression predicted responsiveness. However, AKT activation status as measured by phosphorylation (p-AKT) seemed to be a very powerful predictor of erlotinib response. In another report on 49 glioblastoma patients, tumors with coexpression of both the mutant EGFRvIII and PTEN showed statistically significant response to the kinase inhibitors erlotinib and gefitinib (110). Interestingly, this group did not demonstrate a significant relationship between EGFR gene amplification and clinical response. The assumption underlying these studies is that expression of these molecules equates with receptor activation. However, it is not clear if overexpression of wild-type EGFR has identical effects to overexpression of the vIII mutant, and the selection pressures that facilitate vIII mutation would suggest that vIII conveys an oncogenic advantage and that, as a result, these should not be biologically identical. Both articles additionally suggest that an intact PTEN-AKT pathway is necessary for clinical response to the inhibitors; erlotinib and gefitinib seem to be beneficial primarily in patients with low levels of activated AKT (p-AKT) and preserved expression of PTEN, both indicative of an intact PTEN-AKT pathway. Of note, tumors with PTEN loss are significantly more susceptible to targeted inhibition of mTOR, a downstream signal effector that is negatively regulated by PTEN (111).
The above data suggest that glioblastoma response to EGFR inhibitors is related to overexpression of wild-type or mutant EGFRvIII and to an intact PI3K-PTEN pathway. However, other reports fail to document therapeutic benefit from EGFR inhibitors in glioblastoma patients. Several gefitinib trials have demonstrated partial responses but no change in overall survival in glioblastoma patients (104, 112), and 1 Phase 2 erlotinib trial resulted in median progression-free survival of only 12 weeks (113). In addition, little diminution of EGFR phosphorylation, a marker of receptor activation determined by Western blotting, was noted in glioblastoma tissues obtained from patients treated with erlotinib and gefitinib (114). Thus, the role of predictive markers may be minimal unless clear responses to these therapies are observed in at least a small subset of glioblastoma patients.
It is also important to note that amplification and overexpression of EGFR as demonstrated by FISH or immunohistochemistry, expression of the vIII mutant of EGFR as shown by immunohistochemistry, and increased phosphorylation of EGFR and its downstream signaling partners as shown by protein analysis do not automatically equate to sensitivity to specific small molecule kinase inhibitors such as gefitinib, erlotinib, and newer generations of these compounds. Spatial filling into the tertiary kinase domain as determined by the unique activating mutations and interplay by the various downstream signal transducers may play equally important roles in predicting clinical response.
Status of Assays
At present, immunohistochemistry is the most convenient method for examining expression of EGFR and its signaling partners (Fig. 4B) (115, 116). For most pathology laboratories, immunohistochemistry is quick and can be incorporated into the normal clinical laboratory workflow. Notably, however, a survey of 5 commercially available EGFR antibodies on a panel of soft tissue sarcomas found that variations in the frequency of EGFR immunopositivity were dependent on the type of antibody and on the scoring scheme (117). In addition, although phospho-specific antibodies are capable of detecting phosphorylated and activated isoforms of the protein, as demonstrated by their use in p-AKT detection (109), immunohistochemistry based on phospho-specific antibodies require a high degree of optimization and careful interpretation (118). Clearly, ongoing clinical trials that incorporate immunohistochemistry testing of EGFR and its signaling partners should fully disclose the pertinent details of the experimental protocols to allow replication and standardization of the assays. At this time, clinical testing of components of the EGFR signaling pathway is not mature.
FIGURE 4. (A) Epidermal growth factor receptor (EGFR) fluorescence in situ hybridation (FISH). Dual-color FISH of a glioblastoma specimen showing amplification of the EGFR gene locus (numerous green signals per cell) in comparison to normal diploid control chromosome 7p copy number control probe (2 red signals per cell). (B) Epidermal growth factor receptor immunohistochemistry. A glioblastoma with amplification of the EGFR gene demonstrating strong immunoreactivity for cell surface as well as expression of EGFR protein. Endothelial cells function as internal negative controls (magnification x 400). Tissue used for photographs courtesy of Dr. Cathy Nutt, Massachusetts General Hospital, Boston, MA.
Fluorescence in situ hybridization remains the most widely used method for assessment of EGFR gene copy number, with most laboratories scoring cases as amplified or nonamplified (115). In addition, the assay is robust and straightforward to develop and perform in any laboratory familiar with FISH (Fig. 4A). In comparison to immunohistochemistry, as outlined previously and in Table 2, FISH places greater demand on the resources of the pathology laboratory, and there are discrepancies between EGFR gene amplification as determined by FISH and increased EGFR expression by immunohistochemistry (119, 120). Reconciliation of the differential findings and further in-depth exploration of the mechanisms underlying EGFR gene amplification should help to clarify the use of both techniques. It is not inconceivable, as is current practice, to rely on results of both techniques for making treatment decisions.
Recommendations for Routine and Clinical Trial Testing
The paradigm illustrated by the EGFR story-in the setting of NSCLC response to molecular inhibitors in the presence of EGFR gene mutation and in the possible setting of glioblastoma response to molecular inhibitors in the presence of EGFR overexpression and intact downstream pathways-is perhaps the most exciting in current neurooncology. The paradigm elegantly combines a known biologic mechanism with a specific molecular inhibitor and a reasonable laboratory technique for evaluation of the relevant marker. Indeed, such studies should be a routine component of all clinical trials using molecular inhibitors. However, the marker assays evaluated as part of these trials must be as comprehensive as possible and must be as clearly reported as possible to derive the most useful translational information from such trials.
For current diagnostic applications, it is hard to recommend such testing in a routine predictive situation because practical problems prevent widespread adoption of these assays. These include availability of antibodies, interpretation of results, reproducibility of the assay, and, most importantly, how much actual therapeutic benefit can be realized from identification of these molecules in subsets of glioblastoma patients. Practically, the high cost of many of the novel therapeutics demand the accurate and precise identification of responsive patient populations, and this should propel continual development of novel and better diagnostic techniques and the search for biomarkers in these biologically important pathways. However, at the present time, evaluation of these markers should be reserved for the trial situation. SUMMARY
The potential clinical significances of 1p/19q loss, MGMT promoter hypermethylation, and aberrations in EGFR signaling pathways have raised the possibility of molecular tailoring of glioma therapy (121). We summarize our recommendations in Table 3. Briefly, 1p/19q testing is widely used to assess prognosis and manage patients, to guide clinical trials, and, less accurately, to establish diagnoses; 1p/19q assays are robust but not based on biologic knowledge of distinct pathways or molecules. Testing for MGMT promoter hypermethylation is based on a sensible hypothesis and should be incorporated into clinical trials, but optimal assays are uncertain and not standardized among different laboratories. Finally, evaluation of the EGFR-PI3K-PTEN signaling cascade is based on an elegant biologic rationale, but the inherent complexity of the signaling pathway makes it difficult to expect a single assay to serve as a simple readout, and practical issues prevent robust testing at the present time. Moreover, such approaches may only be relevant in the context of specific molecular inhibitors of this pathway. The most important feature of this latter observation, however, is that it established an important paradigm for how particular molecular parameters may reflect the likely effects of particular molecular inhibitors. In this regard, ongoing drug discovery and the rollout of new designer compounds will place additional pressure on designing new diagnostic assay to identify patient subsets who might respond to the new therapies. Nonetheless, this approach represents the future of personalized targeted medicine.
There will clearly be increasing pressure on the neuropathologist to adopt novel molecular tests. However, the neuropathologist has a duty to be the practical and critical decision point for assay implementation. The neuropathologist must be familiar with the issues surrounding molecular assays, as well as the clinical, financial, and societal ramifications if new assays are instituted. Indeed, the era of personalized molecular medicine places the neuropathologist in an exciting role, but one that demands a considerable degree of responsibility.
TABLE 3. Recommendations for Molecular Markers in Malignant Gliomas
1. Schmitt M, Harbeck N, Daidone MG, et al. Identification, validation, and clinical implementation of tumor-associated biomarkers to improve therapy concepts, survival, and quality of life of cancer patients: Tasks of the Receptor and Biomarker Group of the European Organization for Research and Treatment of Cancer. Int J Oncol 2004; 25:1397-406
2. Ludwig JA, Weinstein JN. Biomarkers in cancer staging, prognosis and treatment selection. Nat Rev Cancer 2005;5:845-56
3. Partridge M, Gaballah K, Huang X. Molecular markers for diagnosis and prognosis. Cancer Metastasis Rev 2005;24:71-85
4. Conley BA, Taube SE. Prognostic and predictive markers in cancer. Dis Markers 2004;20:35-43
5. Hayes DF. Prognostic and predictive factors revisited. Breast 2005;14: 493-99
6. Hayes DF. Prognostic and predictive factors for breast cancer: Translating technology to oncology. J Clin Oncol 2005;23:1596-97
7. Keen JC, Davidson NE. The biology of breast carcinoma. Cancer 2003;97:825-33
8. Pepe MS, Etzioni R, Feng Z, et al. Phases of biomarker development for early detection of cancer. J Natl Cancer Inst 2001;93:1054-61
9. Hinestrosa MC, Dickersin K, Klein P, et al. Shaping the future of biomarker research in breast cancer to ensure clinical relevance. Nat Rev Cancer 2007;7:309-15
10. Nelson NJ. Experts wrestle with problems developing biomarkers, search for new tests. J Natl Cancer Inst 2006;98:578-79
11. Twombly R. Identity crisis: Finding, defining, and integrating biomarkers still a challenge. J Natl Cancer Inst 2006;98:11-12
12. Seabrook JM, Hubbard RA. Achieving quality reproducible results and maintaining compliance in molecular diagnostic testing of human papillomavirus. Arch Pathol Lab Med 2003;127:978-83
13. Williams LO, Cole EC, Lubin IM, et al. Quality assurance in human molecular genetics testing: Status and recommendations. Arch Pathol Lab Med 2003;127:1353-58
14. Grody WW. Proficiency testing in diagnostic molecular pathology. Diagn Mol Pathol 1994;3:221-23
15. Joint Commission on Accreditation of Healthcare Organizations. Approved: New and revised molecular testing standards for laboratory. Jt Comm Perspect 2007;27:6-7
16. Ross JS. Economic, regulatory, and practice issues in molecular pathology and diagnostics. Am J Clin Pathol 1999;112:S7- 10
17. Ross JS. Financial determinants of outcomes in molecular testing. Arch Pathol Lab Med 1999;123:1071-75
18. Ross JS. Molecular pathology and managed care: Challenges and opportunities. Diagn Mol Pathol 1998;7:189-91
19. Ardisson L, Iafrate J, Lewandrowski K. An analysis of reference laboratory testing in a large urban academic medical center: The impact new molecular diagnostic technologies. Lab Med 2007;38:472-75
20. Hernandez JS. Cost-effectiveness of laboratory testing. Arch Pathol Lab Med 2003;127:440-45
21. McShea A, Marlatt MW, Lee HG, et al. The application of microarray technology to neuropathology: Cutting edge tool with clinical diagnostics potential or too much information? J Neuropathol Exp Neurol 2006;65:1031-39
22. Wrensch M, Fisher JL, Schwartzbaum JA, et al. The molecular epidemiology of gliomas in adults. Neurosurg Focus 2005;19:E5
23. Bruch LA, Hill DA, Cai DX, et al. A role for fluorescence in situ hybridization detection of chromosome 22q dosage in distinguishing atypical teratoid/rhabdoid tumors from medulloblastoma/central primitive neuroectodermal tumors. Hum Pathol 2001;32:156-62
24. Biegel JA, Fogelgren B, Zhou JY, et al. Mutatio