Purpose of review: Cutaneous melanoma still represents a paradox among all solid tumors. It is the cancer for which the best prognostic markers ever identified in solid tumors are available, yet there is very little understanding of their biological significance. This review focuses on recent biological data that shed light on the clinico-biological correlations that support the 2010 AJCC melanoma staging system.
Recent findings: E-cadherin is a keratinocyte–melanoma adhesion molecule whose loss is required for the acquisition of an invasive phenotype. Recent data showed that this loss is mediated by the transcription factor Tbx3 which is also involved in suppressing melanocytes senescence. CCN3 is present in melanoma cells close to the epidermal–dermal interface, but not in melanoma cells that have invaded deep into the dermis. It has been recently demonstrated that CCN3 decreases the transcription and activation of matrix metalloproteinases and suppresses the invasion of melanoma cells. These results suggest that the absence of CCN3 in advanced melanoma cells contributes to their invasive phenotype and that ulceration modifies the microenvironment allowing CCN3-depleted melanoma cells to invade.
Summary: A major challenge is to replace outcome clustering based on artificial biomarker breakpoints by a continuous multidimensional prognostic model. Major improvement will come from shared computerized tools allowing to generate continuous likelihood scores for diagnosis, prognosis and response prediction. This will lead to the development of platforms which can be used by scientists from different fields to integrate and share high-quality data in the precompetitive setting and generate new probabilistic causal models.
aSegal Cancer Centre/Jewish General Hospital, Departments of Pathology & Oncology, McGill University, Montreal, Canada
bDaniel Den Hoed Cancer Center & Erasmus University Medical Center, Department of Surgical Oncology, Rotterdam, The Netherlands
Correspondence to Alan Spatz, MD, Segal Cancer Center/Jewish General Hospital, Department of Pathology, 3755 Cote Ste-Catherine, Montreal QC H3T 1E2, Canada Tel: +1 514 825 577; e-mail: email@example.com
Although recent data have shed light on extremely promising antimelanoma targeted therapies, cutaneous melanoma still represents a paradox among all solid tumors. It is the cancer for which the best prognostic markers ever identified in solid tumors are available, but with very little understanding of their biological significance. It can be expected that the strongest prognostic biomarkers, such as the ones that were included in the last 2010 AJCC staging system [1••], are surrogates of key biological events. Therefore understanding the correlations between the prognostic factors and the biology of the disease is a major objective of melanoma translational research. This could lead to the identification of new potentially drugable pathways and new predictive factors.
Primary tumor thickness
In 1970, Alexander Breslow  reported that thickness, cross-sectional areas and depth of invasion were prognosis of cutaneous melanoma recurrence or metastasis rate at 5 years. It is notable that although it identified the most robust and reliable prognostic feature among all histological prognostic features ever described in cancer, this article is based on an error. Breslow considered maximal thickness as an indicator of tumor burden and cross-sectional area as the other important prognostic feature. We now know that the prognostic significance of Breslow's index is actually not related with tumor burden and that cross-sectional area does not predict clinical outcome. Nevertheless, the last 2010 AJCC melanoma staging confirmed the prognostic significance of Breslow's thickness as the 10-year survival is 92% among the patients with T1 melanomas (≤1.00 mm), and is 50% in patients with T4 melanomas (≥4.01 mm) [1••].
Several studies have attempted to identify an expression signature associated with Breslow's index progression [3,4]. These studies identified only a few genes whose expression decreases as the thickness increases, and they include E-cadherin, cadherin-19, bcl2a1 as well as proto-cadherin 7 (pcdh7), the regulator of G-protein signaling 20 (rgs20), and E-cadherin is a keratinocyte–melanoma adhesion molecule whose loss is required for the acquisition of an invasive phenotype. Interestingly, this loss is mediated by the transcription factor Tbx3 that is also involved in suppressing melanocytes senescence through repressing the cyclin-dependent kinase inhibitors p19(ARF) and p21(WAF1/CIP1/SDII) [5••]. The cadherin switch is an early phenomenon during melanoma progression, which is associated with increased motility and invasiveness of the tumor and altered signaling, leading to decreased apoptosis and senescence blockage [6,7•]. There are no data to support the view that melanoma thickness would directly promote melanoma metastasis, for instance in increasing the likelihood for the melanoma cells to encounter vessels. On the contrary, the Breslow's thickness is likely to be the phenotypic indicator of the biological cascade involving the melanoma cells at the leading edge of the tumor. The importance of the tumor microenvironment that exists at the bottom of the invasive component, at the interface with the tumor cells, is illustrated by urokinase-like plasminogen activator (uPA). The expression and activation of uPA and its receptor (uPAR) are increased at the bottom of the melanoma, at the interface between the invasive front and the host tissue, promoting melanoma invasion in collaboration with matrix metalloproteinase 9 (MMP-9) [8,9]. It has been recently demonstrated that macrophages co-cultivated with melanoma cells express higher levels of uPAR and MMP-9 compared to macrophage cultured alone [10••]. The enhanced uPAR and MMP-9 expression in macrophages co-cultivated with tumor cells seems a rather specific phenomenon, generated through a cell-to-cell contact mechanism. This points to a cooperation between tumor cells and macrophages, which is elicited by tumor cells themselves, in generating key enzymes essential in the promotion of tumor invasiveness, such as uPAR and MMP-9. Therefore, one may regard Breslow's index as a quantitative surrogate of the multifactorial biological machinery that drives melanoma progression and invasion.
Primary melanoma ulceration
Survival rates of patients with an ulcerated melanoma are proportionately lower than those of patients with a nonulcerated melanoma of equivalent T category, but are remarkably similar to those of patients with a nonulcerated melanoma of the next highest T category [1••]. For example, 5-year survival is 79% for a T3a nonulcerated melanoma and is 82% for a T2b ulcerated melanoma. A T4a nonulcerated melanoma has a 5-year survival of 71%, similar to that of a T3b ulcerated melanoma with a 68% rate.
The biological significance of melanoma ulceration is almost completely unknown. There are two possible explanations for the adverse prognostic value of ulceration in melanoma: ulceration is a surrogate of an intrinsic biological attribute of the tumor or the host that favors its dissemination, or ulceration directly favors the dissemination of the tumor, for example, by modifying the local environment. Among the intrinsic properties of the melanoma that might favor both ulceration and dissemination, the most convincing evidence is for the role of the proliferative activity of the tumor and the recent data on the dual role of osteopontin. Proliferation of the tumor in the vicinity of the epidermis may erode it by contact and thus favor tumor expansion. However, in the AJCC database, ulceration remains a prognostic factor in all T categories even when the mitotic rate is entered into the model. Recent studies have shed light on the role of the matrix proteins in melanoma progression [11,12,13••,14,15]. Osteopontin is a glycophosphoprotein cytokine with pleiotropic effects [12,16,17]. In normal tissues, it plays a role in inflammation, vascular and bone remodeling, and wound repair. It also has a role in cell adhesion, chemotaxis, prevention of apoptosis, invasion, migration, and anchorage-independent growth of tumor cells. Osteopontin expression is predictive of reduced relapse-free survival [18,19••]. CCN3 is another matrix protein that modulates cell–matrix interactions. CCN3 is known to inhibit melanocyte proliferation and stimulate adhesion to collagen type IV, the main component of the basement membrane [13••,14,15]. CCN3 has a unique role in securing adhesion of melanocytes to the basement membrane which is distinct from other melanoma-produced matrix proteins, and which acts as a de-adhesive molecule and antagonist of focal adhesion. CCN3 is present in melanoma cells close to the epidermal–dermal interface, but not in melanoma cells that had invaded deep into the dermis or had metastasized to lymph nodes . CCN3 decreases the transcription and activation of MMPs and suppresses the invasion of melanoma cells. These results suggest that the lack of CCN3 in advanced melanoma cells contributes to their invasive phenotype and that ulceration modifies the microenvironment allowing CCN3-depleted melanoma cells to invade.
The hypothesis of a direct influence of ulceration on the local environment that may favor melanoma progression has been reinforced by studies on the interactions between melanocytes and keratinocytes. The penetration of melanoma cells through the dermal–epidermal junction is associated with the dissolution of the native epidermal basement membrane collagens type IV and VII . The ability of melanoma cells to cross the dermal–epidermal junction correlates with their metastatic potential. Normal melanocytes are coupled with keratinocytes by gap junction formation, whereas melanoma cells do not form these heterotypic gap junctions [21,22,23•]. Instead melanoma cells communicate among themselves and with fibroblasts. This switch in communication partners coincides with the shift from E-cadherin to N-cadherin expression during melanoma development. Forced expression of E-cadherin by adenoviral gene transfer in N-cadherin-expressing melanoma cells restores gap junctional compatibility with keratinocytes [24,25]. Therefore melanocyte transformation is associated with loss of the pre-existing gap junctional activity with keratinocytes but a concomitant gain of communication with a newly juxtaposed cell type, the fibroblasts. This abnormal relation between melanocytes and fibroblasts might be the basis of an imbalance of growth factor production, especially βFGF, in the immediate area of the melanocyte. Once the homeostatic balance is lost and malignant transformation has occurred, microenvironmental factors, such as degradation of matrix components and host–tumor interactions, are essential for survival and growth of malignant cells [26,27]. Keratinocytes control cell growth and dendricity, as well as expression of melanoma-associated cell surface molecules of normal melanocytes . Therefore ulceration might represent for the melanoma cells a very effective way to evade the keratinocyte-mediated control.
Another possibility to explain the prognostic significance of ulceration with biological events is that ulceration could be a surrogate for unknown host factors which influence the clinical course of melanoma. The post-hoc analyses of the EORTC1892 and EORTC18991 adjuvant interferon (IFN) therapy trials indicate that patients with an ulcerated primary could be more sensitive to IFN than patients with nonulcerated primaries [28••]. Among 2644 patients randomized, 849 had ulcerated primaries, 1336 nonulcerated primaries, and 459 with unknown ulceration . In the ulceration group the impact of IFN treatment was much greater than in the nonulceration group for relapse-free survival (test for interaction: P = 0.02), distant metastasis-free survival (P < 0.001), and overall survival (OS) (P < 0.001). The greatest reductions occurred in patients with stages IIB/III-N1 with ulceration. In a retrospective analysis of 537 consecutive micro-metastatic sentinel lymph nodes with melanoma, we reported that ulceration in the primary tumor is associated with a lower density of mature dendritic cells in the sentinel node as compared when primary melanoma ulceration is absent (P = 0.0005) . Therefore, melanoma ulceration is associated with a defect in Th1 response. Whether this defect is due to constitutional host characteristics that also favor melanoma ulceration needs further investigations. It is possible that exogenous IFN during adjuvant therapy palliates an insufficient Th1 response.
Proliferation of the primary melanomas, defined by the mitotic rate, is a powerful and independent predictor of survival. As a result, primary tumor mitotic rate is now a required element for the 2010 edition of the melanoma staging system [1••]. Data from the AJCC Melanoma Staging Database demonstrate a highly significant correlation between increasing mitotic rate and declining survival rates (P = 0001). In a multifactorial analysis of 10 233 patients with clinically localized melanoma, mitotic rate was the second most powerful predictor of survival, after tumor thickness [1••].
Genes identified in a validated and reproducible signature that predicts for metastases or death was mainly associated with replication or DNA repair . For replication, two pathways were over-represented: the replication origins firing (ROF) genes and the separation of sister chromatids by securin. It is critical that chromosomal DNA is precisely duplicated during S phase of the eukaryotic cell cycle, with no sections of DNA left unreplicated or replicated more than once. There is a considerable plasticity in this process because cells license many potential replication origins, of which only a small percentage are used in any one cell cycle, with the others remaining ‘dormant’. This means that the usage of replication origins can change under different circumstances. For example, dormant replication origins can be activated when replication forks are inhibited to allow timely completion of the replication program. It is also crucial that each replication origin does not fire more than once in a single S phase, as this would lead to local amplification of the DNA . During late mitosis and early G1, the cell licenses replication origins for use in the upcoming S phase by loading protein complexes composed of Mcm proteins (Mcm2–7 complexes) onto the origin DNA . During S phase, Mcm2–7 at licensed origins can initiate replication forks. The Mcm2–7 complex moves with the replication forks, providing the essential DNA helicase activity that unwinds the DNA. This means that when an origin initiates a pair of forks, it is converted into the unlicensed state and cannot fire again. Poor prognostic melanomas are characterized by a global overexpression of ROF-related genes . Mcm-4 and Mcm-6 overexpression is strongly correlated with metastasis-free survival and OS. This prognostic value is maintained when age, sex, location of the primary tumor, thickness and ulceration are introduced in the multivariate model. The entire ROF system is locked by geminin that complexes CDT1 and CDC6 [33,34]. When CDT1 and CDC6 are released, they can recruit MCMs at the replication origins. When this interaction is altered, the helicases cascade becomes overactive leading to replication increase. Aggressive melanomas need a fast and effective replication, and need to repair mistakes induced during replication. Overexpression of DNA repair genes is associated with metastases or death. Increase in postreplicative DNA repair capacity associated with topoisomerase II alpha could explain spontaneous resistance of most melanomas toward radiotherapy and alkylating agents .
Tumor-infiltrating lymphocytes (TILs) in melanoma are responsible for tumor killing and may induce spontaneous regression. Brisk TILs in the melanoma vertical growth phase (VGP) is a strong, albeit not independent, prognostic factor associated with superior survival. In a study of 1171 patients with cutaneous VGP melanoma with at least 3 years of follow-up, we investigated whether more detailed classes of TIL patterns, based on topography and intensity, are independent prognostic factors and identify patients with good prognosis . TIL infiltrate was assessed for pattern (absent, nonbrisk, brisk), intensity (scanty, moderate, dense), and topography (peripheral, central, both). Pattern was assessed qualitatively: brisk TIL infiltrate was defined as a continuous band of lymphocytes at the base of the melanoma, or throughout the tumor. Other studied variables were thickness, mitotic count, ulceration, sex, age, anatomic site, melanoma-related death (MRD). A brisk infiltrate was observed in 21.2% of the cases. Adjusted hazard ratio for MRD as compared to the absent category was 0.82 [95% confidence interval (CI) = 0.64–1.06] in the nonbrisk category, and 0.43 (95% CI = 0.28–0.68) in the brisk category. On the basis of the Cox model, brisk TILs was an independent prognostic factor for MRD (P < 0.001) controlling for thickness, mitotic count, ulceration, sex, age and site. Intensity was significantly associated with MRD for melanomas with peripheral and central brisk TILs (P = 0.027) but not for other melanomas. Remarkably, no death was observed in the 5.5% of melanomas with dense, brisk TIL infiltrate at 10 years. Therefore, it is likely that there is a small population of ‘super-responder’ patients whose tumor is characterized by a continuous and dense TIL infiltrate and who do not display signs of tumor progression.
The sex effect on survival is another unresolved mystery in the melanoma field. The male sex is associated with an adverse outcome that persists even after adjustment for other prognostic variables [37•,38]. After adjustment the relative excess risk to die from melanoma is 1.85 (95% CI = 1.65–2.10) [37•]. This sex impact on mortality risk is observed at all stages. No biological explanation is identified so far. In particular, it is difficult to evoke a hormonal influence as the adjusted risk estimates are similar among patients below 45 or above 60 years of age. As most of the cancer-testis antigens (CTAs) genes are located on the X-chromosome, one possibility would be that CTA expression differs between females and males, but in fact no difference has been observed. Moreover, data regarding the prognostic impact of CTA expression are conflicting. A possible confounding factor would have been differences in behavior toward UV exposure. The sex effect is unchanged when body site is introduced in the model strongly suggesting that the survival difference among sexes is not due to behavorial differences. Integrative studies correlating DNA changes and expression data are powerful to identify new genetic defects involved in tumor progression. In order to identify new pathways involved in melanoma progression and to better understand the sex effect on melanoma prognosis, we correlated micro-array comparative genomic hybridization with expression levels of several genes in frozen melanoma samples. A significant correlation between DNA copy number and mRNA level was detected for 851 genes (FDR-adjusted P < 0.05). Among the 32 females, losses in X chr were significantly associated with DMFS (log-rank P = 0.009). The affected X chr was always the inactive X. These results point to the existence of one or several key metastasis suppressor genes located on the X chromosome. Whether this would be sufficient to explain the sex difference is still unclear and will need population studies.
There is a strong need to refine prognostication in melanoma. The main reason is that we need to replace outcome clustering, based on artificial biomarker breakpoints, by a continuous multidimensional prognostic model. The pace of new biomarkers development will quickly make it impossible to update the list of prognostic variables to assess each time a new biomarker is identified. Major improvement will come from shared computerized tools which will help us in generating continuous likelihood scores for diagnosis, prognosis and response to treatment predictions. This will lead to the development of platforms which can be used by scientists from different fields to integrate and share high-quality data in the precompetitive setting and generate new probabilistic causal models.
References and recommended reading
Papers of particular interest, published within the annual period of review, have been highlighted as:
• of special interest
•• of outstanding interest
Additional references related to this topic can also be found in the Current World Literature section in this issue (p. 281).
1•• Balch CM, Gershenwald JE, Soong SJ, et al
. Final version of 2009 AJCC melanoma staging and classification. J Clin Oncol 2009; 27:6199–6206.
2 Breslow A. Thickness, cross-sectional areas and depth of invasion in the prognosis of cutaneous melanoma. Ann Surg 1970; 172:902–908.
3 Winnepenninckx V, Lazar V, Michiels S, et al
. Gene expression profiling of primary cutaneous melanoma and clinical outcome. J Natl Cancer Inst 2006; 98:472–482.
4 Jaeger J, Koczan D, Thiesen HJ, et al
. Gene expression signatures for tumor progression, tumor subtype, and tumor thickness in laser-microdissected melanoma tissues. Clin Cancer Res 2007; 13:806–815.
5•• Rodriguez M, Aladowicz E, Lanfrancone L, Goding CR. Tbx3 represses E-cadherin expression and enhances melanoma invasiveness. Cancer Res 2008; 68:7872–7881.
6 Watson-Hurst K, Becker D. The role of N-cadherin, MCAM and beta3 integrin in melanoma progression, proliferation, migration and invasion. Cancer Biol Ther 2006; 5:1375–1382.
7• Wu Y, Lin Y, Liu H, Li J. Inhibition of invasion and up-regulation of E-cadherin expression in human malignant melanoma cell line A375 by (−)-epigallocatechin-3-gallate. J Huazhong Univ Sci Technolog Med Sci 2008; 28:356–359.
8 De Vries TJ, De Wit PE, Clemmensen I, et al
. Tetranectin and plasmin/plasminogen are similarly distributed at the invasive front of cutaneous melanoma lesions. J Pathol 1996; 179:260–265.
9 Bianchini F, D’Alessio S, Fibbi G, et al
. Cytokine-dependent invasiveness in B16 murine melanoma cells: role of uPA system and MMP-9. Oncol Rep 2006; 15:709–714.
10•• Marconi C, Bianchini F, Mannini A, et al
. Tumoral and macrophage uPAR and MMP-9 contribute to the invasiveness of B16 murine melanoma cells. Clin Exp Metastasis 2008; 25:225–231.
11 Robert G, Gaggioli C, Bailet O, et al
. SPARC represses E-cadherin and induces mesenchymal transition during melanoma development. Cancer Res 2006; 66:7516–7523.
12 Chang PL, Harkins L, Hsieh YH, et al
. Osteopontin expression in normal skin and nonmelanoma skin tumors. J Histochem Cytochem 2008; 56:57–66.
13•• Fukunaga-Kalabis M, Martinez G, Telson SM, et al
. Downregulation of CCN3 expression as a potential mechanism for melanoma progression. Oncogene 2008; 27:2552–2560.
14 Fukunaga-Kalabis M, Santiago-Walker A, Herlyn M. Matricellular proteins produced by melanocytes and melanomas: in search for functions. Cancer Microenviron 2008; 1:93–102.
15 Vallacchi V, Rodolfo M. Regulatory role of CCN3 in melanoma cell interaction with the extracellular matrix. Cell Adh Migr 2009; 3.
16 Smit DJ, Gardiner BB, Sturm RA. Osteonectin downregulates E-cadherin, induces osteopontin and focal adhesion kinase activity stimulating an invasive melanoma phenotype. Int J Cancer 2007; 121:2653–2660.
17 Buback F, Renkl AC, Schulz G, Weiss JM. Osteopontin and the skin: multiple emerging roles in cutaneous biology and pathology. Exp Dermatol 2009; 18:750–759.
18 Rangel J, Nosrati M, Torabian S, et al
. Osteopontin as a molecular prognostic marker for melanoma. Cancer 2008; 112:144–150.
19•• Conway C, Mitra A, Jewell R, et al
. Gene expression profiling of paraffin-embedded primary melanoma using the DASL assay identifies increased osteopontin expression as predictive of reduced relapse-free survival. Clin Cancer Res 2009; 15:6939–6946. Expression data were obtained from formalin-fixed paraffin-embedded primary melanomas and provided evidence that osteopontin expression is a prognostic biomarker.
20 Bechetoille N, Haftek M, Staquet MJ, et al
. Penetration of human metastatic melanoma cells through an authentic dermal-epidermal junction is associated with dissolution of native collagen types IV and VII. Melanoma Res 2000; 10:427–434.
21 McGary EC, Lev DC, Bar-Eli M. Cellular adhesion pathways and metastatic potential of human melanoma. Cancer Biol Ther 2002; 1:459–465.
22 Daniel-Wojcik A, Misztal K, Bechyne I, et al
. Cell motility affects the intensity of gap junctional coupling in prostate carcinoma and melanoma cell populations. Int J Oncol 2008; 33:309–315.
23• Villares GJ, Dobroff AS, Wang H, et al
. Overexpression of protease-activated receptor-1 contributes to melanoma metastasis via regulation of connexin 43. Cancer Res 2009; 69:6730–6737.
24 Hsu M, Andl T, Li G, et al
. Cadherin repertoire determines partner-specific gap junctional communication during melanoma progression. J Cell Sci 2000; 113(Pt 9):1535–1542.
25 Hsu MY, Meier FE, Nesbit M, et al
. E-cadherin expression in melanoma cells restores keratinocyte-mediated growth control and down-regulates expression of invasion-related adhesion receptors. Am J Pathol 2000; 156:1515–1525.
26 Haass NK, Smalley KS, Li L, Herlyn M. Adhesion, migration and communication in melanocytes and melanoma. Pigment Cell Res 2005; 18:150–159.
27 Tucci MG, Lucarini G, Brancorsini D, et al
. Involvement of E-cadherin, beta-catenin, Cdc42 and CXCR4 in the progression and prognosis of cutaneous melanoma. Br J Dermatol 2007; 157:1212–1216.
28•• Eggermont AM, Suciu S, Santinami M, et al
. Adjuvant therapy with pegylated interferon alfa-2b versus observation alone in resected stage III melanoma: final results of EORTC 18991, a randomised phase III trial. Lancet 2008; 372:117–126.
29 Eggermont A, Suciu S, Testori A, et al
. Ulceration of primary melanoma and responsiveness to adjuvant interferon therapy: analysis of the adjuvant trials EORTC18952 and EORTC18991 in 2,644 patients. J Clin Oncol 2009; 18 [Abstract 9007].
30 Elliott B, Scolyer RA, Suciu S, et al
. Long-term protective effect of mature DC-LAMP+ dendritic cell accumulation in sentinel lymph nodes containing micrometastatic melanoma. Clin Cancer Res 2007; 13:3825–3830.
31 Blow JJ, Dutta A. Preventing re-replication of chromosomal DNA. Nat Rev Mol Cell Biol 2005; 6:476–486.
32 Courbet S, Gay S, Arnoult N, et al
. Replication fork movement sets chromatin loop size and origin choice in mammalian cells. Nature 2008; 455:557–560.
33 Gonzalez MA, Tachibana KE, Laskey RA, Coleman N. Control of DNA replication and its potential clinical exploitation. Nat Rev Cancer 2005; 5:135–141.
34 Gonzalez MA, Tachibana KE, Adams DJ, et al
. Geminin is essential to prevent endoreduplication and to form pluripotent cells during mammalian development. Genes Dev 2006; 20:1880–1884.
35 Kauffmann A, Rosselli F, Lazar V, et al
. High expression of DNA repair pathways is associated with metastasis in melanoma patients. Oncogene 2008; 27:565–573.
36 Spatz A, Gimotty PA, Cook MG, et al.
, editors. Protective effect of a brisk tumor infiltrating lymphocyte infiltrate in melanoma: an EORTC melanoma group study. J Clin Oncol 2007; 16 [Abstract 8517].
37• de Vries E, Nijsten TE, Visser O, et al
. Superior survival of females among 10,538 Dutch melanoma patients is independent of Breslow thickness, histologic type and tumor site. Ann Oncol 2008; 19:583–589.
38 Lasithiotakis K, Leiter U, Meier F, et al
. Age and gender are significant independent predictors of survival in primary cutaneous melanoma. Cancer 2008; 112:1795–1804.