This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Liu, E. T. Search for Related Content PubMed PubMed Citation Articles by Liu, E. T. Social Bookmarking                            What's this?

Molecular Oncodiagnostics: Where We Are and Where We Need to Go

Genome Institute of Singapore, Singapore

MOLECULAR MARKERS in clinical oncology can be divided into diagnostic markers, which distinguish one disease from another; prognostic markers, which are associated with the clinical behavior of a tumor; and predictive markers, which are used to predict outcome of therapy and to aid in the selection of optimal treatment. Diagnostic and prognostic markers, though important in clinical management, are deterministic in nature, in that the natural course of a cancer is not likely to be changed because of knowledge of that marker status. However, more excitement within the last decade has been centered on predictive markers, many of which are also the targets for specific therapeutics. The importance, of course, is that the ascertainment of these predictive markers may guide treatment selection that can change the course of a disease. The most obvious examples include the HER-2 status in anthracycline use in adjuvant chemotherapy of breast cancer and for the effectiveness of trastuzumab in metastatic disease. Moreover, the selectivity of all-transretinoic acid and imatinib mesylate in PML-RARA and BCR-ABL translocation–positive leukemias, respectively, highlight the importance of correct diagnostics in treatment planning. The interest in emerging predictive markers is both real and justified.

In this issue, Ambros et al,¹ however, raise a troubling aspect of molecular cancer diagnostics that has been previously recognized but is still not articulated very well. Quality control, cutoff criteria, and consistent analytic formats may seem pedestrian but are critical to achieving practical clinical impact. Unfortunately, the absence of accepted standards may lead to disturbingly high rates of false calls or, at least, uncertainty in the results. In this study, 137 coded specimens from 17 neuroblastomas across 10 European molecular diagnostic laboratories were analyzed for N-myc (MYCN) amplification, loss of heterozygosity at 1p36.3, and tumor ploidy, using a variety of accepted methods for the detection of molecular perturbations. Their results show that data discrepancies that would change clinical interpretation occurred in the range of 7% to 8%, and that the majority of the inaccuracies may be due to the clinician’s not considering something as simple as the proportion of tumor cells in the sample. If significant clinical decisions are made on any one test, then this range of technical error is of great concern.

How Exact Can We Be?

Although the initial reaction to these data may be one of alarm, it is actually surprising that the error rate is as low as documented given the different analytic platforms (fluorescent in situ hybridization [FISH], Southern blot, polymerase chain reaction [PCR]) used. Clinicians, accustomed to the reproducibility of routine laboratory tests such as serum sodium and quantitative immunoglobulin levles, are often surprised at the qualitative and relatively inexact nature of molecular diagnostics. Results from molecular technologies are method-, reagent-, and operator-sensitive. For example, Southern blot analysis for MYCN amplification detects a band of a specific molecular weight and is sensitive to DNA degradation. Although PCR is subject to PCR amplification bias and detects only the presence of a specific fragment of the gene, it is much less sensitive to DNA degradation. A result from either PCR or Southern blot hybridization is an average of the DNA in the tumor, which includes stromal and inflammatory cells. FISH, however, like immunohistochemical analysis of tissue sections, detects single-cell events in a population of cells within a tumor. Despite the fact that all techniques seek to identify gene amplification, the data outputs are sufficiently different that the results are not always readily comparable. The most rational approach would then be to standardize one technical platform and enforce its use as the gold standard. Unfortunately, sometimes even economics impede the adoption of standards. The low profit margins for some diagnostics hinder the development of standardized kits tested in rigorous (and expensive) clinical trials. Nevertheless, these issues can still be adequately addressed by organization and consensus.

However, even if standardization were simple, the practical realities are much more complicated. First, molecular technologies are not stable and are highly fluid: new, more robust, more exact, and cheaper approaches emerge frequently. The evolution of the bcr-abl translocation assay from one based on cytogenetics to Southern blot hybridization and reverse transcriptase PCR is an excellent example of this improvement cycle. This raises the thorny question of whether each technical improvement of a molecular test needs to be validated in a completely independent clinical trial. If this principle were applied to predictive markers in studies where the outcome may require 5 to 10 years of follow-up, few advances would be made. Second, even if tests could be standardized, biologic variability limits the convertibility of one analytic platform to another. For example, protein levels and gene amplification measure related, but clearly different, targets. Immunohistochemical analysis for HER-2 overexpression correlates with gene amplification 70% to 80% of the time at best. Does the 20% to 30% discrepancy nullify the utility of this test as a predictive marker? Investigators have found that, despite these discrepancies, HER-2 overexpression by immunohistochemical analysis and HER-2 amplification by either differential PCR or FISH were both able to distinguish the subset of node-positive patients benefiting from dose-intense chemotherapy.² Moreover, it is questionable what can be considered the gold standard. Biologic reality would suggest that the protein product represented by the immunohistochemical result should be more associated with tumor behavior than gene amplification and should therefore be considered the biochemical gold standard. For the HER-2 marker, however, recent data suggest that FISH analysis for gene amplification is more likely to predict response to trastuzumab than the standardized immunohistochemical test. These results seem counterintuitive but perhaps can be explained by the fact that immunohistochemistry is a less quantitative and potentially less consistent analytic test than FISH.

The many ways in which a molecular marker can be defined as abnormal as compared with normal can also confuse the clinical interpretation of molecular results. For example, P53 mutations with biologic consequences can be found as missense mutations anywhere in the gene that give rise to an abnormal protein or as deletion-insertion or splice mutants that render the transcript unstable and short-lived. In addition, the same biologic outcome can be achieved by alternative abnormalities that alter downstream p53 biology or biochemistry such as murine double minute 2 (MDM2) amplification, or the presence of human papilloma viral oncoprotein E6 that enhances the degradation of the p53 oncoprotein.³ For the p53 status of a cancer, no single molecular test will completely define the functional abnormality and will therefore always be incomplete. This will become an important issue when therapeutics directed at abnormal p53 pathways are developed.³

The conclusion that molecular tests are imprecise or technically unstable and should not be used is, however, inappropriate. Nevertheless, the examples discussed here should force us to develop more structured strategies in marker development and in informing the clinical community about how best to interpret these markers. Several groups have suggested standards in marker development that are reasonable and should be heeded.^4,5 The foundations of these recommendations are precision in the detection of a valid target, reproducibility of the test, and stable access to necessary reagents over time. The work by Ambros et al¹ also highlights the importance of standardized tissue processing and the need to assess tumor and normal-tissue content (Table 1). But given the progressive importance of tumor markers in guiding therapeutic options, we should consider different ways of interpreting marker data and new approaches to speed their development and validation.

One of the first lessons in medical school is to be suspicious of a single test result in the absence of corroborating data. This caution is based on the clinician experience that tests may be imperfect and on the knowledge that our therapies have serious consequences. In molecular diagnostics, our concerns should be the same. The associated reality, however, is that as clinicians, we routinely integrate a multitude of patient data in developing a clinical profile on which to base our decisions, such as when we assess the prognostic potential of an individual with node-negative breast cancer that is estrogen receptor-positive, progesterone receptor-positive, HER-2–positive, grade 3 with vascular invasion, and less than 1 cm in size. The data integration that follows is qualitative, experiential, and ad hoc.

Our recent experience with expression array data in clinical oncology suggests an interesting approach to data integration.⁶ Such an approach assumes that the integration of many individually imprecise data points may uncover fundamental and stable biologic truths. Clinical applications of microarray results have proved this to be true in lymphomas, breast cancers, and leukemias. Despite the variance in the results of each individual probe on an array, the clustered data that take into account the composite movement of data points are surprisingly reproducible and consistent in their associations (Figs 1 and 2). The bases of this approach are mathematical and statistical algorithms that seek to identify clusters within large data sets and to provide an estimate of their robustness as class identifiers. Although they have been commonly used for many decades in a variety of disciplines, the application of these tools in clinical medicine has been limited. However, advances in computational methods and in data visualization have now rendered the data output from these analyses interpretable to nonexperts and perhaps adaptable to clinical use.

View larger version (52K): [in this window] [in a new window]   Fig 1. Anatomy of a hierarchical cluster.

View larger version (43K): [in this window] [in a new window]   Fig 2. (A) Visual presentation of clinical data. Green indicates good prognosis; red, poor prognosis; and black, equivocal prognosis. (B) "Heat map" format of presenting clinical data on a number of breast cancer patients.

Marker Development and the New Realities

The new molecular reality is that scientists are now able to generate a large number of potential diagnostic and prognostic markers with remarkable speed. The high-throughput capabilities of new technologies, such as expression and tissue arrays and proteomic approaches, identify definitive disease markers, often without obvious mechanistic associations. In this scenario, our approach of picking one marker at a time for development is unacceptably slow. Instead, we suggest that the following model for marker development should be considered (Fig 3). Marker genes associated with disease or prognosis identified by high-throughput procedures or database searches (as in the Cancer Genome Anatomy Project)⁷ and then validated on a separate tissue set will need to be identified. First, the associated full-length cDNAs are cloned and recombinant proteins expressed to generate antibodies. A collection of these will be made available for any clinical trials group for testing on therapeutic trials, with an understanding that the raw data will be retained in a central data repository for later use in meta-analyses. We estimate that for breast cancer alone, there may be between 50 and 300 such markers, depending on the stringency of selection.⁸ Conceivably, all cancers can be studied in this fashion. The appropriate funding agencies for such an enterprise may be the National Institutes of Health, a nonprofit agency, or a consortium of companies.

View larger version (37K): [in this window] [in a new window]   Fig 3. A model for marker development.

REFERENCES

1. Ambros IM, Benard J, Boavida M, et al: Quality assessment of genetic markers used for therapy stratification. J Clin Oncol 21:2077–2084, 2003[Abstract/Free Full Text]

2. Thor AD, Berry DA, Budman DR, et al: erbB-2, p53, and efficacy of adjuvant therapy in lymph node-positive breast cancer. J Natl Cancer Inst 90:1346–1360, 1998[Abstract/Free Full Text]

3. Lane DP, Lain S: Therapeutic exploitation of the p53 pathway. Trends Mol Med 8:S38–S42, 2002 (suppl)[CrossRef][Medline]

4. Bast RC Jr, Ravdin P, Hayes DF, et al: 2000 update of recommendations for the use of tumor markers in breast and colorectal cancer: Clinical practice guidelines of the American Society of Clinical Oncology—American Society of Clinical Oncology Tumor Markers Expert Panel. J Clin Oncol 19:1865–1878, 2001[Abstract/Free Full Text]

5. Allred DC, Harvey JM, Berardo M, et al: Prognostic and predictive factors in breast cancer by immunohistochemical analysis. Mod Pathol 11:155–168, 1998[Medline]

6. Miller LD, Long PM, Wong L, et al: Optimal gene expression analysis by microarrays. Cancer Cell 2:353–361, 2002[CrossRef][Medline]

7. Strausberg RL, Greenhut SF, Grouse LH, et al: In silico analysis of cancer through the Cancer Genome Anatomy Project. Trends Cell Biol 11:S66–S71, 2001 (suppl)[Medline]

8. Liu ET, Sotiriou C: Defining the galaxy of gene expression in breast cancer. Breast Cancer Res 4:141–144, 2002[Medline]

CiteULike    Complore    Connotea    Del.icio.us    Digg    Facebook    Reddit    Technorati    Twitter    What's this?

This article has been cited by other articles:

				M. J. Ratain and R. H. Glassman Biomarkers in Phase I Oncology Trials: Signal, Noise, or Expensive Distraction? Clin. Cancer Res., November 15, 2007; 13(22): 6545 - 6548. [Full Text] [PDF]

				L. Pusztai and K. R. Hess Clinical trial design for microarray predictive marker discovery and assessment Ann. Onc., December 1, 2004; 15(12): 1731 - 1737. [Abstract] [Full Text] [PDF]

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Liu, E. T. Search for Related Content PubMed PubMed Citation Articles by Liu, E. T. Social Bookmarking                            What's this?