Highly Sensitive Multivariable Assay Detection of Melanocytic Differentiation Antigens and Angiogenesis Biomarkers in Sentinel Lymph Nodes With Melanoma Micrometastases

Objectives  To evaluate the prognostic value of melanocytic differentiation antigens and angiogenesis biomarkers in sentinel lymph nodes (SLNs) with melanoma micrometastases.

Design  Prognostic study of an inception cohort.

Setting  Academic research.

Patients  Between July 1, 1999, and July 31, 2002, all patients who had primary cutaneous or mucosal melanomas that have a Breslow depth of 1.5 mm or greater, ulceration, or Clark level IV or V, or had SLN biopsies.

Main Outcome Measures  By the use of quantitative reverse transcription–polymerase chain reaction, the expression of the following was analyzed in SLNs: 2 melanocytic differentiation antigens (tyrosinase [P17646] and melanoma antigen recognized by T cells [MART-1; Q16655]) and genes involved in angiogenesis (VEGF [NM_001025366] and VEGFR2 [AF035121]), lymphangiogenesis (VEGFC [NM_005429], VEGFR3 [X68203], LYVE1 [NM_016164], and PROX1 [002763]), and invasion (uPA [NM_002658], PAI1 [NM_00602], and EMMPRIN [L10240]). Outcome measures were the association of these melanocytic differentiation antigens and angiogenesis biomarkers with clinicopathologic characteristics of patients, and an evaluation of the prognostic value for relapse-free survival and overall survival.

Results  Ninety-one patients were included, with a median follow-up period of 41 months. Micrometastases were present in 15% (14 of 91) of patients. Tyrosinase (P < .001), MART-1 (P < .001), vascular endothelial growth factor 121 (VEGF₁₂₁) (P = .007), and PAI1 (P = .02) expression was significantly associated with micrometastasis. In univariate analysis, histologic findings and tyrosinase and MART-1 expression were significantly associated with relapse-free survival. Tyrosinase and MART-1 expression was associated with overall survival. A multiple Cox proportional hazards regression model identified negative histologic findings and tyrosinase expression that exceeded 27 copies/copy of TATA box-binding protein (third quartile) as significantly associated with an increased risk of relapse or death.

Conclusions  Quantitative assessment of melanocytic differentiation antigens in SLNs, which has prognostic value, is more specific than qualitative assessment. Prognosis may be more effectively predicted by the combination of quantitative assessment of melanocytic differentiation antigens in SLNs with histologic assessment. A significant association was found between the presence of micrometastases and the expression of angiogenesis biomarkers.

In most European countries, the incidence of melanoma and its mortality have risen over the past few decades.¹ In the past 10 years, the care of patients with melanoma has changed from major surgery with extensive lymph node dissection to less aggressive procedures. Sentinel lymph node (SLN) mapping avoids routine elective lymph node dissection and offers prognostic information. Patients having a positive SLN demonstrate relapse-free 5-year survival of about 53%, and those who have a negative SLN demonstrate relapse-free 5-year survival of about 83%.² Most ongoing and forthcoming adjuvant therapeutic trials rely or will rely on SLN staging.

Micrometastases in an SLN are detected after serial sectioning and immunohistochemistry by the use of HMB45 and S-100 antibodies.^3,4 Almost 20% of patients who have SLN-negative disease as staged by histopathologic examination experience recurrences.^2,5,6 More sensitive micrometastasis detection could allow more accurate staging. Results of recent studies^7-9 suggest that detection of melanocytic differentiation or of tumor-associated antigen messengers that use reverse transcription–polymerase chain reaction (RT-PCR) could improve metastasis detection in SLNs of patients with primary melanoma. In some series, these techniques raise the detection rate of histologically and immunohistochemically negative nodes to 50% and may overestimate the risk of recurrence.^8,9 The discrepancy between histologic and molecular studies vs the high sensitivity of RT-PCR can be explained by the difficulty of discriminating single bystander tumor cells from cellular aggregates with true metastatic potential. Discrimination is important because the size of any micrometastasis compared with the tumor load is a prognostic factor^3,10; therefore, available quantitative methods should be used.

Furthermore, a focus on the local microenvironment in the lymph node could help in the differentiation of bystander cells from true micrometastasis. Tumor angiogenesis and lymphangiogenesis have an important role in malignant diseases, such as melanoma, and may have a key role in the proliferation of small “dormant” micrometastases.^11,12 In addition to angiogenesis and lymphangiogenesis, extracellular matrix proteolysis is important in tumor progression. Indeed, proteases enhance the ability of tumor cells to invade surrounding tissue and to metastasize. All these processes that involve many growth factors have some significantly recognizable markers in melanoma progression.¹³ In the present study, we evaluated the SLN transcript expression profile of a gene panel involved in melanocytic differentiation, such as angiogenesis (VEGF and VEGFR2), lymphangiogenesis (VEGFC, VEGFR3, LYVE1, and PROX1), and invasion (uPA, PAI1, and EMMPRIN). The objectives of the study were as follows: (1) to evaluate in a multivariable manner the expression of tyrosinase, melanoma antigen recognized by T cells (MART-1), total vascular endothelial growth factor (VEGF), VEGF₁₂₁ (soluble VEGF isoform), and VEGFC, VEGFR2, VEGFR3, LYVE1, PROX1, EMMPRIN, uPA, and PAI1 transcripts (by the use of quantitative RT-PCR) in SLN extracts from patients with untreated melanoma to provide a descriptive analysis of gene expression; (2) to define cutoff values for these markers; (3) to correlate histologic results with molecular marker findings; (4) to assess the relationship between these markers and clinicopathologic characteristics of patients; and (5) to provide a preliminary evaluation of the prognostic value of these markers for relapse-free survival (RFS).

Between July 1, 1999, and July 31, 2002, we enrolled 98 consecutive adult patients who had primary cutaneous or mucosal melanoma with clinically negative regional lymph nodes. Patients were informed about the nature of the study and the risks and potential benefits of participation and were told that the study was in accord with the regulations of our institute and with French law. Patients gave their informed consent and had the option to withdraw from participation. This study was approved by Comité de Protection des Personnes Île de France IV. All patients had at least 1 of the following melanoma characteristics: ulceration, Clark level IV or V, or Breslow thickness of at least 1.5 mm (this cutoff represents high-risk patients with melanoma based on the American Joint Committee on Cancer staging system¹⁴ used before 2002). In all patients, an SLN biopsy was performed in combination with wide local excision of 1 to 3 cm dependent on Breslow thickness.

Lymphatic mapping techniques were performed as described by Gershenwald et al,⁶ with the SLN as the first lymph node to receive lymphatic drainage directly from the tumor site. When more than 1 lymphatic pathway was identified, a separate SLN was considered for each pathway. Lymphatic mapping was performed by the use of a handheld gamma probe. All SLNs defined by lymphoscintigraphy were subsequently mapped, with the provision that the procedure was not invasive.

All fresh SLNs were transported within 15 minutes to the pathology department. An SLN was bisected if it was 5.0 mm or larger; one half was snap frozen and kept at −80°C for molecular analysis, and the other half was paraffin embedded for histologic and immunohistochemistry analyses (3 serial sections at 200-μm intervals were cut [ie, 12 sections]). All paraffin-embedded material was cut into 4-μm sections for hematoxylin-eosin (HE) staining and for immunohistochemistry by the use of anti-S100 polyclonal antibody (1:800 dilution), HMB45 (clone PNL2 [1:50 dilution]), and anti-MelanA (clone A103 [1:50 dilution]) (all from Dako, Carpinteria, California). The frozen half yielded twenty 20-μm sections; one 5-μm section of every 10 sections was stained with HE.

Micrometastases were identified by the presence of cell clusters (≥5 cells) in the lymph node, whether subcapsular or parenchymal. Clustered cells in the lymph duct lumens were noted separately. All patients with micrometastasis subsequently underwent complete lymph node dissection of the affected nodal basin.

In 5 specimens, the small size of the SLN made it impossible to divide; therefore, molecular analysis was performed on 86 patient samples. Total RNA was extracted by the use of a guanidinium thiocyanate method.¹⁵ Three micrograms of total RNA was reverse transcribed by the use of random hexamer primers (Roche Diagnostics, Meylan, France). Transcript quantification for tyrosinase, MART-1, total VEGF, VEGF₁₂₁, VEGFC, VEGFR2, VEGFR3, LYVE1, PROX1, EMMPRIN, uPA, PAI1, and 2 control genes (B2M and TBP) was performed (LightCycler and LCFastStart DNA Master Mix kit, Roche Diagnostics) in accordance with the instructions of the manufacturer. Specific primers and hydrolyzation probes were chosen by the use of commercially available software (Primer Express Software; Applied Biosystems, Coutaboeuf, France) and are available on request. For each transcript, complementary DNA was cloned (TOPOII TA kit; Invitrogen, Cergy Pontoise, France), and a standard curve that ranged from 1 or 10 to 10⁹ molecules/μL was generated. Because there were no differences between the 2 control genes (B2M and TBP), the results are presented as copies of target gene per copy of TBP. All experiments were performed in duplicate, and coefficients of variation of all crossing point values were less than 1%.

Data are presented as medians, interquartile ranges, and ranges for quantitative variables and as counts and percentages for categorical variables. Pairwise associations between the different markers were assessed by the use of Spearman rank correlation coefficient. The association between markers and histologic findings was evaluated by the use of Somer Dxy rank correlation coefficient and was tested by the use of Wilcoxon rank sum test. The prognostic value of angiogenesis biomarkers was evaluated for RFS and for overall survival (OS). Both end points were measured at the time of study inclusion, when angiogenesis biomarker values were measured, and at the first relapse (for RFS) or the time of death (for OS). Because follow-up periods were fairly heterogeneous, both end points were measured up to 30 months, the first quartile of follow-up times. In this case, the assumption of uninformative censoring was reasonable, but this was not true for the whole follow-up period. Relapse-free survival and OS curves were generated by the use of Kaplan-Meier estimation and were compared by the use of a log-rank test. For these analyses, angiogenesis biomarkers were categorized in accordance with sample tertiles to obtain 3 equal groups, except when 2 tertiles were equal such as for tyrosinase and MART-1, in which case other percentiles (the 75th for tyrosinase and the 80th for MART-1) were selected independent of RFS or OS. This procedure avoids the use of the same data to select “optimal” cutoff points and to test their association with prognosis, without accounting for this selection. Multiple Cox proportional hazards regression models were used to adjust for potential confounders. A backward stepwise model selection procedure was then used to select a set of variables associated with prognosis. Proportional hazards assumption was then checked by the use of the Grambsch and Therneau lack-of-fit test. When relevant, models were compared by the use of their likelihood ratio and C index. A higher likelihood ratio indicates a better fit to the data, but its value has no intuitive interpretation. Conversely, the C index is a discrimination index, similar to the area under a receiver operating characteristic curve if data were binary. A value of 1.0 indicates perfect predictive discrimination, whereas a value of 0.5 indicates no ability to discriminate. All tests were 2-sided at a P < .05 significance level. Analyses were performed by the use of statistical software (R version 2.4.0; R Foundation for Statistical Computing, Vienna, Austria).

Sentinel lymph nodes were identified successfully in 91 patients, which gives a detection rate of 93%. For 7 other patients with melanoma (4 on the head, 1 on the vulva, and 2 on the trunk), the surgical procedure was unsuccessful or cancelled because access to the lymph node was judged to be technically difficult or dangerous. The median age of 91 patients was 56 years (age range, 16-89 years) (Table 1). There were 40 male patients and 51 female patients. Most melanomas were located on the extremities (n = 65; 71%), followed by the trunk (n = 201; 22%) and head or neck region (n=5; 6%). The median Breslow thickness of lesions was 2.2 mm (range, 0.9-19.0 mm). Ulceration was present in 29% (25 of 87) of patients. The median follow-up period was 41 months (range, 3-78 months).

One hundred sixty-one SLNs from 91 patients were obtained (mean, 1.8 per patient [range, 1-3 per patient]). Fourteen of 91 patients (15%) had micrometastasis detected by histologic examination or immunohistochemistry. One patient with micrometastasis had no molecular assessment because of the size of the lymph node. After regional basin dissection, 7 of 91 patients (8%) were found to have positive lymph nodes. Although adjuvant interferon alfa-2a or alfa-2b therapy (3 million IU 3 times per week) was proposed for all patients with Breslow thickness of at least 1.5 mm independent of their SLN status, it was administered to only 11 patients. After a median 41 months of follow-up, 23 patients had relapsed, and 10 patients had died.

Among 161 lymph nodes from 91 patients, 17 nodes (11%) from 13 patients (14%) had tumor involvement as indicated by HE staining. All histologically positive lymph nodes as indicated by HE staining also had tumor cells that expressed anti-S100, HMB45, and anti–MART-1 as demonstrated by the use of immunohistochemical methods. One histologically negative lymph node as indicated by HE staining showed occult metastases by immunohistochemical methods. In total, 14 of 91 patients (15%) had lymph node metastases as indicated by HE or immunohistochemistry.

Primer specificity and quantitative RT-PCR optimization for each transcript were assessed by fluorescent dye technology (SYBR Green; AnyGenes, Paris, France) and melting curve analysis. By the use of probe technology (TaqMan; Roche Diagnostics), a highly sensitive calibration curve was generated for each transcript. The threshold of detection was 1 copy for all transcripts. For all assays, intrarun and interrun variability (calculated from triplicate samples and compared by the use of the results of samples in 10 different runs) showed a mean (SD) for the crossing points of 0.15 (0.55).

Among 86 patients for whom frozen specimens were available for molecular analysis, most patients had detectable messenger RNA (mRNA) expression for angiogenic, lymphangiogenic, and invasion transcripts (uPA, PAI1, total VEGF, VEGF₁₂₁, EMMPRIN, VEGFR2, VEGFR3, LYVE1, PROX1, and VEGFC), with median values above those obtained in reactive lymph nodes and peripheral blood mononuclear cells. The distribution of tyrosinase and MART-1 expression differed: 65 of 86 patients had low tyrosinase values (0-27 copies/copy of TBP), and 68 of 86 patients had low MART-1 values (0-150 copies/copy of TBP) (Table 1). Median values of tyrosinase and MART-1 in SLNs were 1000 times lower than those observed in macrometastatic lymph nodes. All 14 histologically positive SLNs showed mRNA expression for all transcripts.

As a cutoff, we used the median transcript value of each mRNA expression in SLNs. Values at or above this cutoff were defined as mRNA-positive transcripts. Tyrosinase, MART-1, VEGF₁₂₁, and PAI1 mRNA expression was significantly associated with positive histologic findings (Somer Dxy = 0.80, P < .001; Dxy = 0.66, P < .001; Dxy = 0.49, P = .007; and Dxy = 0.41, P = .02; respectively). Little or no association was found with uPA (Dxy = 0.25, P = .17), VEGFC (Dxy = −0.03, P = .87), EMMPRIN (Dxy = −0.03, P = .86), VEGFR3 (Dxy = −0.25, P = .17), LYVE1 (Dxy = −0.20, P = .27), or PROX1 (Dxy = −0.08, P = .68).

Overall, 23 patients relapsed, and 10 patients died. Within 30 months, 20 patients had relapsed (8 locoregional and distant relapses, 9 locoregional relapses, and 3 distant relapses), 9 of whom died. Univariate analysis showed that tyrosinase and MART-1 expression was significantly associated with OS and RFS and that histologically proven micrometastases and Breslow thickness were associated with RFS (Table 2). None of the angiogenesis, lymphangiogenesis, or protease variables studied (uPA, PAI1, or VEGFs) were of prognostic value in this group of patients.

A multiple Cox proportional hazards regression model identified negative histologic findings and tyrosinase expression that exceeded 27 copies/copy of TBP (third quartile) as significantly associated with an increased risk of relapse or death. The data distribution did not allow testing for an interaction between the 2 variables, as only 2 patients with tyrosinase expression of 27 or fewer copies/copy of TBP had positive histologic findings, and neither patient died or relapsed. The best model was found by the grouping together of all patients with tyrosinase expression of 27 or fewer copies/copy of TBP, thus separating the sample into 3 prognostic groups as shown in Figure 1 and Figure 2. This predicted the outcome significantly better, with a likelihood ratio of 16.78 and a C index of 0.731 compared with a likelihood ratio of 13.18 and a C index of 0.690 obtained with histologic findings alone. The Prognostic Separation index of the model with tyrosinase expression and histologic findings reached 0.600 compared with 0.493 obtained with histologic findings alone. This index represents the difference in survival between the best and worst prognosis groups.¹⁶

Adjustment for Breslow thickness or ulceration did not significantly improve the model. Compared with a simple presence or absence of coding for tyrosinase, the use of a quantification method to define a threshold (27 copies/copy of TBP) greatly improved the prognostic value of tyrosinase expression. As a result, the likelihood ratio was 15.35 instead of 6.76, the C index was 0.713 instead of 0.656, and the Prognostic Separation was 0.422 instead of 0.235. A similar improvement was found with adjustment for histologic findings, with the likelihood ratio increasing by 4.21 and a C index of 0.725 instead of 0.698.

This study established a highly sensitive multivariable assay to detect SLNs with melanoma micrometastases. By the use of quantitative RT-PCR, the expression of 2 melanocytic differentiation antigens (tyrosinase and MART-1) was analyzed, as well as biomarkers involved in angiogenic and invasion processes (uPA, PAI1, EMMPRIN, total VEGF, VEGF₁₂₁, and VEGFR2) and genes involved in lymphangiogenesis (VEGFC, VEGFR3, LYVE1, and PROX1). Tyrosinase (P < .001), MART-1 (P < .001), VEGF₁₂₁ (P = .007), and PAI1 (P = .02) expression was significantly associated with micrometastasis (Wilcoxon rank sum test). When all biomarkers and SLN histologic findings were pooled in univariate analysis, tyrosinase and MART-1 expression was significantly associated with OS and RFS, and histologically proven micrometastases and Breslow thickness were associated with RFS. Tyrosinase expression remained an independent factor when adjusted for histologic findings.

In our series, the detection rate uncovered by the use of histologic findings or immunohistochemistry compares favorably with investigations that examined half a node⁶; however, sectioning both node halves increases the detection rate to 34% as recently validated by the European Organization for Research and the Treatment of Cancer.³ In agreement with a recent meta-analysis¹⁷ with regard to SLN molecular staging, our results show that tyrosinase and MART-1 RT-PCR expression is associated with RFS and OS. Moreover, we show that the use of 27 copies/copy of TBP as a tyrosinase expression cutoff better predicts RFS than histologic findings or RT-PCR alone. Indeed, histologic examination only (without tyrosinase RT-PCR) separates 2 groups with a 30-month RFS of 84% (negative histologic findings) and a 30-month RFS of 39% (positive histologic findings). The use of both 27 copies/copy of TBP as a tyrosinase expression cutoff for RT-PCR and histologic examination distinguishes the following 3 groups of patients after 30 months of follow-up: (1) a group with both positive histologic findings and tyrosinase expression that exceeds 27 copies/copy of TBP, which has a high rate of relapse (27% RFS); (2) a second group of patients with tyrosinase expression that exceeds 27 copies/copy of TBP and negative histologic findings (67% RFS); and (3) a third group of patients with tyrosinase expression of 27 or fewer copies/copy of TBP (all with negative histologic findings, except for 2 patients), which has an even better prognosis (87% RFS).

The use of 27 copies/copy of TBP as a tyrosinase cutoff for RT-PCR may distinguish true micrometastasis from illegitimate transcription of melanocytic differentiation antigens and isolated melanoma cells, as shown by Abrahamsen et al.¹⁸ Likewise, Satzger et al¹⁹ identified isolated HMB45-positive and MelanA-positive cells in 9.9% of SLNs from patients with melanoma and showed that their overall prognostic value was similar to that in patients with pathologically negative SLNs. Others have shown that tumor load as defined by SLN histologic findings is a significant prognostic factor.²⁰ Van Akkooi et al²¹ recently reported good prognostic value of submicrometastases (<0.1 mm). With the application of the concept of tumor load to RT-PCR analysis, quantification to define a threshold could improve the prognostic value compared with a simple presence or absence of tyrosinase mRNA. While nevus cells are detectable in 5% to 15% of melanoma-draining SLNs,^18,22-25 Abrahamsen et al¹⁸ showed intermediate values of MART-1 and tyrosinase mRNA in tumor-negative SLNs with nevus cells compared with micrometastasis in SLNs devoid of nevus cells.

Unlike differentiation markers, which directly reflect metastasis and are expressed by tumor cells, markers that characterize the tumor microenvironment can be more easily detected within lymph nodes. In our sample, PAI1 and VEGF₁₂₁ expression was significantly associated with the presence of a micrometastatic lymph node. A central role for extracellular matrix proteolysis and tumor invasion is performed by the serine protease urokinase-type plasminogen activator (uPA) and its inhibitor plasminogen activator inhibitor type 1 (PAI-1), the major physiologic regulator of plasminogen activator activity. Urokinase-type plasminogen activator catalyzes the conversion of inactive plasminogen into plasmin, a broadly acting enzyme that is able to degrade various extracellular matrix proteins and to activate metalloproteinases and growth factors.²⁶ The plasminogen activation system has been reported to be involved in melanoma progression and metastasis.²⁷ Stabuc et al²⁸ investigated the association between uPA and PAI-1 concentrations in 43 primary cutaneous melanomas (Breslow thickness <1.5 mm) and established both as prognostic factors for malignant melanoma. Significantly higher concentrations of uPA and PAI-1 were found in melanomas than in healthy surrounding skin (P < .001 for both). The PAI-1 values were significantly correlated with Breslow thickness, Clark level, and vascular invasion. Hanekom et al²⁹ suggested that plasma levels of the active fraction of plasma PAI-1 in patients with melanoma were associated with higher metastatic risk. In nude mice, McMahon et al³⁰ showed that PAI-1 can enhance and inhibit the growth of M21 human melanomas and that this seems to be owing to PAI-1 regulation of angiogenesis. In vitro investigations have shown that, in invasion of the human melanoma cell line, BLM was attenuated by an anti–PAI-1 monoclonal antibody MAI-12.³¹ We show herein for the first time (to our knowledge) that PAI1 expression is correlated with SLN metastasis, which highlights the importance of this angiogenesis biomarker in melanoma progression.

Vascular endothelial growth factor is a potent angiogenic factor that increases microvascular permeability; induces proliferation, survival, migration, and differentiation of endothelial cells; and promotes degradation of the extracellular matrix around the sprouting endothelium by the induction of the expression of proteases and interstitial collagenases. Vascular endothelial growth factor has the following 5 main isoforms produced by alternative splicing of a single gene: VEGF₁₂₁, VEGF₁₆₅, VEGF₁₈₉, VEGF₁₄₅, and VEGF_206.^14,15 The most soluble isoform is VEGF₁₂₁, which is slightly acidic because it lacks the basic amino acids responsible for heparin binding.³² Elevated levels of VEGF have been associated with poor prognosis in various solid primary tumors, including melanoma.³³ Furthermore, studies have shown that primary tumor cytosolic levels of VEGF protein not only are significantly associated with microvessel density³⁴ but also are of prognostic value in patients with node-negative and node-positive breast cancer.³⁵ Along with other researchers,³⁶ we have demonstrated that soluble VEGF isoforms (VEGF₁₂₁ and VEGF₁₆₅) have a higher angiogenic potential than cell-associated isoforms (VEGF₁₈₉). In melanoma, the overexpression of VEGF₁₂₁ and VEGF₁₆₅ by WM1341 B human melanoma cell lines resulted in aggressive tumor growth in mouse xenografts, whereas cells that overexpressed VEGF₁₈₉ remained nontumorigenic after injection into mice. Although VEGF₁₆₅ expression seems to result in the most effective tumor perfusion, VEGF₁₂₁ expression is observed during malignant melanoma progression.³⁷ In addition, it has been shown that VEGF₁₂₁ is the predominant isoform detected in melanoma cell lines.³⁷ To our knowledge, no assessment of the expression of VEGF isoforms has been reported in human melanoma specimens. In a preliminary investigation, we observed preferential expression of VEGF₁₂₁ in SLNs compared with other isoforms (VEGF₁₆₅ and VEGF₁₈₉) (data not shown).

Studies by Dadras et al^38,39 have shown that tumor lymphangiogenesis and levels of primary melanoma VEGF-C predicted lymph node metastasis. Whether the expression of lymphangiogenic factors is similar in metastatic lymph nodes compared with primary tumors has not yet (to our knowledge) been assessed in melanoma. In our study, although all of the angiogenesis biomarkers studied were expressed in SLNs, none were associated with micrometastasis or with survival. Primary tumors seem to induce SLN lymphangiogenesis before metastasis in animal models of human squamous cell tumors (skin) and mouse melanoma.^40-42 Recently, Van den Eynden et al⁴³ compared the expression of various angiogenesis and lymphangiogenesis growth factors in breast primary tumors and lymph node metastasis. In breast cancer that is metastatic to lymph nodes, VEGF-A was overexpressed compared with uninvolved lymph nodes, whereas the relative gene expression of prospero-related homeobox 1 (PROX-1) and VEGF-D was lower, and the expression of VEGF-C and VEGFR-3 was similar. Moreover, while angiogenesis and lymphangiogenesis were driven by VEGF-D and VEGF-A, respectively, in primary tumors, both processes were driven by VEGF-A in lymph node metastasis. In our study, VEGF₁₂₁ expression in SLNs was associated with tumor progression. The importance of VEGF₁₂₁ not only as an angiogenic factor but also as a lymphangiogenic factor in melanoma remains to be tested experimentally.

In our study, neither VEGF nor PAI1 expression was significantly associated with survival. However, our study was underpowered, as attested by the fact that neither Breslow thickness nor ulceration, two well-recognized prognostic factors in melanoma, was significantly associated with survival in our study.

In conclusion, our results demonstrate that quantitative assessment of melanocytic differentiation antigens has prognostic value for patients with melanoma who undergo SLN biopsy. When performed in addition to histologic examination, quantitative assessment of melanocytic differentiation antigens predicts RFS more accurately. This preliminary study requires confirmation in a larger independent sample of patients; therefore, the usefulness of this quantitative assay in clinical practice cannot be assessed yet. In addition, future research needs to elucidate a cost-benefit ratio. Finally, we show a significant association between micrometastasis and expression of the VEGF₁₂₁ isoform and PAI1. Although our study was underpowered and failed to demonstrate a link with tumor progression, the expression of these biomarkers provides a rationale for the testing of antiangiogenesis agents in patients with stage IIIA melanoma.

Correspondence: Céleste Lebbé, MD, PhD, Department of Dermatology, Hôpital Saint-Louis, 1 Ave C. Vellefaux, Paris 75010, France (celeste.lebbe@sls.aphp.fr).

Accepted for Publication: May 21, 2008.

Author Contributions: Drs Vitoux and Mourah (who share first-authorship credit), Kerob, Verola, Basset-Seguin, Baccard, Schartz, Ollivaud, Archimbaud, Servant, Revol, Toubert, Plassa, Porcher, and Lebbé and Ms Podgorniak had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Vitoux and Mourah contributed equally to this work. Study concept and design: Vitoux, Mourah, Kerob, Verola, Basset-Seguin, Baccard, Schartz, Ollivaud, Archimbaud, Servant, Revol, Toubert, Podgorniak, Plassa, Porcher, and Lebbé. Acquisition of data: Vitoux, Mourah, Kerob, Verola, Basset-Seguin, Baccard, Schartz, Ollivaud, Archimbaud, Servant, Revol, Toubert, Podgorniak, Plassa, Porcher, and Lebbé. Analysis and interpretation of data: Vitoux, Mourah, Kerob, Verola, Basset-Seguin, Baccard, Schartz, Ollivaud, Archimbaud, Servant, Revol, Toubert, Podgorniak, Plassa, Porcher, and Lebbé. Drafting of the manuscript: Vitoux, Mourah, Kerob, Toubert, Porcher, and Lebbé. Critical revision of the manuscript for important intellectual content: Vitoux, Mourah, Kerob, Verola, Porcher, and Lebbé. Statistical analysis: Porcher. Obtained funding: Mourah and Lebbé. Administrative, technical, or material support: Podgorniak and Plassa. Study supervision: Vitoux, Mourah, and Lebbé.

Financial Disclosure: None reported.

Funding/Support: This work was supported by a grant from the Société Française de Dermatologie to Drs Lebbé and Mourah.

Role of the Sponsors: The sponsors had no role in the design or conduct of the study; in the collection, analysis, or the interpretation of the data; or in the preparation, review, or approval of the manuscript.