Marginal and Joint Distributions of S100, HMB-45, and Melan-A Across a Large Series of Cutaneous Melanomas

Data from 3 nonoverlapping series of melanoma patients were included in the analysis (Yale Human Investigations Committee protocol No. 8219), which consisted of 212 primary and 237 metastatic cutaneous melanomas surgically removed at Yale-New Haven Hospital during 1959–1994, and 104 additional metastatic melanomas surgically removed during 1995–2005, for which the formalin-fixed, paraffin-embedded (FFPE) tissue blocks were not exhausted during the diagnostic process and for which clinical information was available. Demographic data, clinical course, and follow-up through March 1, 2011, were obtained through comprehensive review of the participant's medical record, the archives of the Connecticut Tumor Registry, the Social Security Death Index, and the State of Connecticut Vital Records. Incomplete medical records resulted in missing clinical and demographic annotations.

A tissue microarray (TMA) representing single 0.6-mm-diameter cores from each of the eligible specimens was constructed by using the standard method.²²  For internal quality control, 4 primary melanomas and 61 metastases were cored in duplicate. Duplicate cores of FFPE pellets constructed from each of 15 melanoma cell lines^23,24  were added as additional controls. Histopathologic annotation was conducted by pathologic review of included cases as previously described,²⁵  with incomplete data fields arising from cases with missing tissue blocks/slides, which prevented correct readjudication of the case. Assessment of melanin levels in each histospot was conducted by a single observer (W.R.B.) on a hematoxylin-eosin (H&E)–stained cut of the TMA and graded on a semiquantitative, 4-level scale with 0 representing no observable melanin and 3 indicating deep pigment in greater than 75% of the melanocytes.

Two serial sections of the TMA were dewaxed in 2 exchanges of xylene and rehydrated with an ethanol gradient. Following antigen retrieval at supra-atmospheric pressure in boiling 6.5 mM sodium citrate (pH 6.0) for 10 minutes, endogenous peroxidase activity was blocked with 0.75% hydrogen peroxide, and nonspecific antigens were neutralized with 0.3% bovine serum albumin. Each of prediluted, neat HMB-45 or anti-Melan-A A103 mouse monoclonal antibodies (Biogenex, Fremont, California) were multiplexed with rabbit anti-S100 polyclonal antibody (1:200; Dako, Carpinteria, California), added to the TMA slide, and incubated overnight at 4°C. Secondary antibodies, AlexaFluor-546–conjugated goat anti-rabbit (1:100; Life Technologies, Carlsbad, California) diluted into anti-mouse Envision (neat, Dako), were then added for 1 hour at room temperature, followed by a 10-minute incubation with Cy5-tyramide (Perkin-Elmer Life Sciences, Waltham, Massachusetts) to label the anti-mouse Envision. Slides were coverslipped by using Prolong Gold with 4′,6-diamidino-2-phenylindole (DAPI, Life Technologies), the latter to visualize nuclei. Negative controls were obtained through omission of the primary antibody.

Automated image acquisition was done as described previously.^23,26  Briefly, sets of monochromatic, high-resolution (1024 × 1024 pixel, 0.5 μm) images were captured for each histospot in each of the DAPI, AlexaFluor-546, and Cy5 fluorescent channels by using a modified, computer-controlled epifluorescence microscope (Olympus BX-51 with xy stage and z controller, Olympus America Inc, Center Valley, Pennsylvania) illuminated by a high-pressure mercury bulb (Photonics Solutions, Edinburgh, United Kingdom) coupled with a high-resolution monochromatic camera (PCO-Tech, Romulus, Michigan).

Histospots containing less than 3% of tumor tissue were excluded from further analysis. Photomicrographs representing S100, HMB-45, and Melan-A immunostaining were visually compared to a referent H&E section for melanoma-specific immunostaining by 2 independent observers (H.V., B.E.G.R.). A case was scored “positive” for the selected antigen if at least 25% of the tumor cells, as defined by the corresponding area on the referent H&E section, displayed immunofluorescence. For cases represented by more than 1 histospot, the case was designated as “positive” if all histospots demonstrated at least 25% immunoreactivity. Cases with immunofluorescence covering less than 25% of the tumor region were classified as “negative.” Discrepancies in staining evaluation were resolved through consultation with a third investigator (K.A.S.). Marginal, as well as pairwise and 3-way joint distributions across the assayed melanoma antigens, were determined by using standard univariate statistics. Bivariate analyses comparing marginal and joint immunostaining patterns with individual clinicopathologic parameters were evaluated by using χ²  analysis or analysis of variance, as appropriate. Associations with melanoma-specific survival were performed by using Kaplan-Meier product-limit and Cox proportional hazards survival analyses. All statistical analyses were conducted with the StatView statistical package (SAS Institute, Cary, North Carolina).

To assess the marginal and joint distributions of S100, HMB-45, and Melan-A/MART-1 in our large series of cutaneous melanomas, we stained 2 serial sections of our TMA, multiplexing either HMB-45 or the anti–Melan-A/MART-1 A103 mouse monoclonal antibody with the Dako anti-S100 rabbit polyclonal antibody on a single slide. The 322 surgical specimens (121 primary tumors and 201 metastases) with evaluable data across all 3 markers were included in this analysis. S100 staining that persisted over at least 25% of the tumor area was detected in 295 of 322 of the included histospots (91.6%), with immunofluorescent signal noted in both the nucleus and cytoplasm of positive-staining specimens. By contrast, the remaining 27 melanomas were S100 negative, with evidence of S100 expression in 5% or fewer of the spotted cells, consistent with the clinical definition of S100 negativity; focal S100 expression was not observed in our included histospots. Similarly, HMB-45 stained at least 25% of the tumor area in 203 of 322 melanomas (63.0%) (73 primary tumors and 130 metastases), and Melan-A/MART-1 stained 236 melanomas (73.3%) (84 primary tumors and 152 metastases). For HMB-45 and Melan-A, focal expression in fewer than 25% of the assayed melanocytes was scored as a negative histospot.

A total of 27 melanomas, including 21 of 201 metastases and 6 of 121 primary tumors (10.4% versus 5.0%, P = .10) were S100 negative. The clinicopathologic characteristics of the S100-negative melanomas are displayed in Table 1. Seventeen (63.0%) of the S100-negative melanomas developed in male patients (P = .61). S100 negativity was observed across a broad spectrum of histopathologic subtypes among the primary tumors and in locoregional as well as visceral disease among the metastases. Other than a higher but nonsignificant count of pan-MAA–negative melanomas noted among the metastases (5 of 201 [2.5%] versus 1 of 121 [0.8%]); Fisher exact, P = .42), these melanomas were unremarkable with respect to clinicopathologic criteria including growth pattern, overall morphology, and pigmentation. Review of their associated H&E sections revealed melanomas with unremarkable histology. S100-negative primary melanomas displayed solid-nested growth, whereas S100-negative metastases were all solid, with cell morphology that included epithelioid, spindled, or hybrid morphology, typical of the disease (Figure 1, A through F).

Next, we considered the joint distributions across all 3 antigens (Table 2, Figure 2, A through P). Cellular immunoreactivity of 25% or greater for all 3 antigens assayed was observed in only 160 melanomas (49.7%). By contrast, 156 melanomas expressed only 1 or 2 of the assayed antigens, and all combinations of the 3 assayed antigens were observed among our data set. Additional commonly occurring categories included S100+/HMB-45−/Melan-A+ melanomas (n = 61; 18.9%) and melanomas expressing only S100 (n = 49; 15.2%). HMB-45 or Melan-A immunoreactivity was also observed among the S100-negative melanomas, with each antigen present in a similar number of assayed lesions (18 versus 15). Yet, 10 S100-negative melanomas demonstrated discordant staining patterns between HMB-45 and Melan-A/MART-1. Finally, 1 primary melanoma and 5 metastatic melanomas did not react with any assayed antibody. The 2-way cross-tabular joint analysis of HMB-45 and Melan-A expression identified 172 (53.4%) melanomas positive for both antigens, 55 (17.1%) negative for both antigens, and 95 (29.5%) discordant for the assayed MAAs, with similar distributions noted among primary tumors and metastases (P = .65). Among melanomas discordant across HMB-45 and Melan-A expression, a significantly larger number were Melan-A+ (n = 64) versus HMB-45+ (n = 31; P < .001).

We also assessed the association between antigen marginal and joint distributions and the semiquantitative melanin pigment score. Whereas HMB-45 staining was independent of pigmentation (P = .13), Melan-A staining was associated with pigmentation, as a larger percentage of moderately or highly pigmented melanomas were Melan-A negative (melanin = 2 or 3; n = 17 of 42, 40.5%) when compared to mildly or unpigmented lesions (melanin = 1 or 0; n = 62 of 260, 23.8%) (P = .01, Table 3). Of the 42 moderately or highly pigmented melanomas with S100 immunostaining data, only 5 were S100 negative, whereas, by contrast, 22 S100-negative melanomas arose in unpigmented lesions. The association between joint HMB-45/Melan-A distribution and degree of pigmentation yielded a highly significant result (P = .007; Table 3), with a smaller percentage of HMB-45+/Melan-A+ lesions (35.7% versus 51.7%–60.7%) and a larger percentage of doubly negative lesions (28.6% versus 9.8%–17.7%) among the highly pigmented lesions.

For the subset of primary melanomas (n = 121), we examined the associations between HMB-45 and Melan-A marginal and joint expression and other clinicopathologic features as well as melanoma-specific survival. Among single-marker bivariate comparisons, the only significant association revealed that HMB-45− melanomas were more likely to possess in-transit metastases at the time of diagnosis (P = .02); all of the remaining comparisons yielded null results (Table 4). We also noted no significant differences in the distribution of clinicopathologic criteria across the HMB-45/Melan-A joint expression categories (Table 5). Lack of expression of either HMB-45 (hazard ratio [HR] = 1.20, 95% confidence interval [CI]: 0.65–2.21; P = .56) or Melan-A (HR = 1.09, 95% CI: 0.56–2.12; P = .81) was not associated with melanoma-specific mortality after adjusting for Breslow thickness (mm), stage at diagnosis, and age at diagnosis. Survival analysis across the 4 joint distribution categories yielded overlapping product-limit survival curves on univariate analysis (P > .99; Figure 3) and no significant difference in a multivariable proportional hazards model after adjusting for Breslow thickness and stage at diagnosis (P = .88; Table 6).

The differential expression of S100, HMB-45, and Melan-A/MART-1 has been extensively studied in cutaneous malignant melanoma. Yet, while trends for their marginal distributions are well known, the overlap in their expression is less well characterized. To the best of our knowledge, no study has evaluated the joint distributions of these antigens in samples of greater than 100 melanomas, and no study has considered the distribution of established melanoma clinicopathologic parameters in HMB-45/Melan-A null or discordant lesions compared with the referent HMB-45+/Melan-A+ tumor.^14–17  To address this issue, we considered the marginal and joint expression of S100, HMB-45, and Melan-A/MART-1 across a large TMA containing representative histospots from 213 primary melanomas and 342 melanoma metastases. Data from the 322 lesions (121 primary tumors and 201 metastases) with evaluable data for all 3 markers were included in our analysis.

With immunofluorescence-based immunohistochemistry to assay our targeted antigens, 91.6% of our samples were S100+, 63.0% were HMB-45+, and 73.3% were Melan-A/MART-1+, each registering slightly below the recognized sensitivities for each of these respective antigens,⁹  with similar rates of immunoreactivity for primary and metastatic lesions. As we selected a cutoff of at least 25% for immunolabeling across eligible melanocytes in each arrayed histospot, our observed decrease in sensitivity might be due to a more stringent criteria than the clinical standard of greater-than-5% positivity,²⁷  such that lesions with patchy, focal staining would be alternatively classified as negative in our analysis. We also did not observe any association between HMB-45 or Melan-A immunoreactivity with any assayed clinicopathologic factor except for an increase in the lack of HMB-45 immunoreactivity among melanoma primary tumors with known microsatellitosis, which is consistent with the observation of less frequent HMB-45 antigenicity among melanoma metastases.^9,21  These data could suggest that lack of HMB-45 immunoreactivity might be a property of primary melanomas that are more likely to metastasize as opposed to a phenotype acquired after metastasis. We also did not detect an association with melanoma-specific mortality with either HMB-45 (P = .56) or Melan-A (P = .81), among our subset of primary melanomas, after adjusting for Breslow thickness, stage at diagnosis, or age at diagnosis. Although our data are consistent with previously published null results for Melan-A,^28,29  the data for HMB-45 are less straightforward, with reports of significantly improved,²⁹  significantly impaired,³⁰  and no association with survival²⁸  for HMB-45 expression, all reported by using cutoffs of either 50% or 90% staining efficiency.³¹ 

Interestingly, 27 of our assayed melanomas (8.4%) were S100 negative, representing, to the best of our knowledge, the largest single collection of S100-negative melanomas reported in the literature. To confirm this, we conducted a systematic search of the PubMed (US National Library of Medicine) database through November 18, 2011, by using the following keyword search: (“S100-negative” AND “melanoma”). The search returned 4 articles of which two^32,33  addressed melanoma clinical samples. Nine additional articles were identified through review of cited references,^7,8,34–40  including 1 article³⁵  that contained a meta-analysis of 12 additional articles all published before 1992. Although the meta-analysis published cumulative data describing S100 negativity in 13 of 230 primary tumors and 9 of 166 melanoma metastases, no single article identified in our search reported on more than 20 S100-negative melanomas with all but 2 studies^32,36  describing fewer than 5 cases.

S100-negative lesions have been described more frequently among junctional nests on sun-damaged skin³³  and among metastatic lesions, even in the setting of a previously documented S100-positive primary tumor.³²  Lack of S100 expression may occur more frequently among ocular or acral melanomas^32,34  but has been observed across all histologic subtypes.³⁵  Six of our S100-negative melanomas arose in primary lesions with the remaining 21 occurring in metastases. One study reporting on 11 S100-negative melanoma metastases with matched primary blocks suggested that loss of S100 antigenicity might be acquired along with metastatic competency, as 9 of 11 had S100-positive primary lesions³² ; a second study with 12 S100-negative melanomas reported equal rates between primary tumors and metastases.³⁶  While a higher percentage of our assayed metastases (10.4%) were S100 negative than were our primary tumors (5.0%), this difference trended toward but did not achieve statistical significance (P = .10), most likely owing to the small size of our S100-negative sample.

We also did not detect any significant associations between S100 immunoreactivity and other clinicopathologic criteria. Melanomas arising in males were as likely to be S100-negative as were melanomas arising in females. Among primary lesions, S100-negative melanomas occurred across a broad range of Breslow thicknesses and histopathologic subtypes. While 1 study noted a possibly increased rate of S100-negative lesions among acral lentiginous melanomas³⁴  and our study reports 2 S100-negative lesions arising in acral lentiginous or amelanotic melanomas, most of our S100-negative lesions arose in superficial spreading or nodular melanomas. Among metastases, lack of S100 antigenicity was observed for cutaneous, soft tissue, lymph node, and visceral lesions.

Six of our S100-negative melanomas, 1 primary and 5 metastases, also lacked expression of HMB-45 and Melan-A. This “triple-negative” pattern was also noted among 7 of 17 S100-negative metastases reported by Aisner et al,³²  with 5 of these 7 lesions arising from the back or shoulder. Interestingly, 5 of 6 of our metastatic “triple negatives” arose in cutaneous or subcutaneous soft tissue; however, our lesions derived from a more diverse set of primary tumor locations including the face, leg, and abdomen as well as the back. Application of next-generation sequencing to the exomes of S100-negative melanomas might be useful for identifying underlying molecular changes associated with both overall lack of S100 antigenicity and concordant lack of S100, HMB-45, and Melan-A/MART-1 immunoreactivity.

While the marginal distributions for S100, HMB-45, and Melan-A positivity and their associations with recognized melanoma clinicopathologic parameters are well established,⁹  the literature describing their joint distributions is much sparser. While MAA expression discordance can be expected in light of the well-documented differing sensitivities of the commonly used MAAs,⁹  there are very few published studies that describe the prevalence and clinical significance of MAA-concordant and MAA-discordant lesions. The largest of these studies evaluated overlapping HMB-45/Melan-A immunoreactivity in 65 melanoma metastases and 10 cutaneous primary tumors.¹⁴  While the authors¹⁴  explored the distribution across all pairwise antigen combinations, correlations with clinicopathologic criteria other than histologic subtype were not reported. Among their sample of 30 melanomas, Xu et al¹⁵  observed some HMB-45/Melan-A discordance, with positive Melan-A expression occurring only in 9 of 14 S100+/HMB45− spindled or epithelioid melanomas. In their study of 17 S100-negative melanomas, Aisner et al³²  noted that all lesions were concordant for HMB-45 and Melan-A, and HMB-45/Melan-A discordant lesions were not observed. To the best of our knowledge, our study of 322 assayable melanomas represents the largest series to date for which the joint distribution of HMB-45 and Melan-A is considered and the only study to consider the relationship between joint HMB-45/Melan-A expression and commonly reported clinicopathologic criteria among eligible primary lesions. Most of our lesions were concordant for HMB-45 and Melan-A expression, with 53.4% expressing both antigens and 17.1% lacking expression in 25% or more of the arrayed melanomas. We also observed melanomas discordant for MAA expression, with 31 melanomas (9.6%) expressing HMB-45 only and 64 melanomas (19.9%) expressing Melan-A only. Further bivariate analyses for the 121 assayed primary melanomas revealed no significant associations with any of the recognized clinicopathologic criteria, and survival analyses revealed virtually overlapping survival curves across the 4 joint distribution categories, suggesting that the prognostic impact of MAA discordance may be small. While the significantly larger number (P < .001) of discordant lesions that express only Melan-A can be explained by Melan-A's comparatively more diffuse and intense staining of melanomas—which persists into the dermal layers of the assayed lesions^14,41  and possibly produced fewer false negatives among our sample of representative 0.6-mm histospots—we cannot exclude the possible role of genetic or epigenetic factors, which underlie the development and distribution of the 4 classes of HMB-45/Melan-A expression–defined melanomas; moreover, a possible relationship with the levels of Microphthalmia-associated transcription factor (MITF) expression⁴²  may yield compelling insight into mechanisms relevant for melanocytic lesion development and progression.

While our study includes numerous strengths, including our large sample size, use of TMAs and automated image capture that eliminate the potential for laboratory drift by assaying all samples simultaneously in the same experimental batch, and comprehensive specimen annotation to enable robust clinicopathologic associations, we also recognize several limitations with our experimental approach. First, although quantitative immunofluorescence is a well-established, unbiased measure of antigen expression across a broad spectrum of cancers,^43,44  the presence of photoreactive melanin might confound immunofluorescent readouts in melanoma.⁴⁵  Melanin not only exhibits broad spectral absorption that decreases monotonically with increasing wavelengths from 300 to 1100 nm,^46,47  but also has recently been shown to display autofluorescence with separate excitations in the ranges of 370 to 470 nm and 785 nm, and corresponding emissions at 540 nm and 890 to 900 nm, respectively,^47,48  with the former in close proximity to the emission wavelength of our mask fluorophore (546 nm). While HMB-45 expression was independent of melanin distribution (with similar proportion of HMB-45–negative lesions among all 4 categories of pigmentation), Melan-A immunostaining was associated with melanin levels; and despite similar proportions of Melan-A–negative lesions observed in each of the unpigmented and the mildly and moderately pigmented melanomas, an excess of Melan-A negative lesions was observed among the highly pigmented melanomas. This effect also translated into a similar significant association among the HMB-45/Melan-A joint distribution categories. Because the association with melanin expression did not extend to all assayed MAAs, we cannot rule out an underlying biologic mechanism that requires further elucidation.

A second limitation of our analysis is that our data were collected from a TMA that represented each melanoma with a single 0.6-mm histospot and were not validated on whole sections. Although current clinical diagnostic standards typically stipulate adjudication of whole sections and define MAA positivity as any immunoreactivity in 5% of cells therein,²⁷  use of TMAs circumvents the risk for batch-to-batch variation, which could be incurred during a whole-slide approach. However, recent TMA validation experiments have demonstrated that intratumoral heterogeneity can yield significant core-to-core variability of protein expression. The observed variance was driven by both the lability of the target protein and the antibody selected for its measurement such that, in a marker-dependent fashion, as many as 11 independent histospots may be required to adequately capture potential heterogeneity.⁴⁹  Specific TMA validation experiments conducted with a small series of antigens with prognostic potential suggest that, for melanoma, a minimum of 3 independent cores are needed to achieve 90% concordance for overall positive staining with whole sections.^50–53  In the context of recognized focality of HMB-45 immunostaining and the association of decreased HMB-45 and Melan-A staining in deeper dermal regions of primary melanomas, we cannot rule out some degree of measurement error, which may have increased our false-negative rate through our strategy of sampling only 1 single 0.6-mm histospot from each index lesion. Future work would include validating our findings across additional, redundant builds of the TMA.

Our experimental design is further limited by the need to sample HMB-45 and Melan-A independently of each other on serial sections of our TMA and then to reconstruct joint distributions through the overlay of corresponding images across the serial sections. Although we forgo some precision by not being able to multiplex both HMB-45 and Melan-A on the same section, our method for assessing joint distribution matches current clinical practice for evaluating occult micrometastases and malignancies of unknown primary tumor in the surgical pathology suite.²⁷ 

In summary, we have comprehensively characterized the marginal and joint distributions, clinicopathologic correlates, and prognostic potential for 3 clinically relevant MAAs in a large series of primary and metastatic melanomas. Our study also describes the largest single series to date of S100-negative melanomas. Future directions include transcriptome profiling and whole-exome sequencing of lesions representative of each joint distribution category, as well as analysis of MITF and other MAA expression to further elucidate discriminating molecular characteristics that define the individual MAA-based melanoma subclasses.