Analysis of HLA antigen expression in benign and malignant melanocytic lesions reveals that upregulation of HLA-G expression correlates with malignant transformation, high inflammatory infiltration and HLA-A1 genotype
El Chérif Ibrahim,
CEA, Service de Recherches en Hémato-Immunologie, DSV-DRM, Hôpital Saint-Louis, Institut d'Hematologie, Paris, France
Department of Cell Biology, Harvard Medical School, 240 Longwood Ave., Boston, MA 02115
Melanoma, a life-threatening neoplastic disease of melanocytic origin, displays a continuously rising incidence and mortality, which afflicts a broad spectrum of age groups. Although melanoma expresses various tumor-associated antigens, immune response against the tumor often remains inefficient. Various therapeutic trials aim at triggering the antitumor immune response by bypassing immune escape mechanisms.1 Therefore, multiple studies were devoted toward better understanding of the mechanisms allowing tumor escape from host immune surveillance in melanoma recipients. Downregulation of HLA class I antigens, as well as induction of HLA class II antigen expression, are often reported in the course of malignant transformation. Briefly, HLA class I antigen defects on tumor cells is thought to favor their escape from cytotoxic T lymphocytes (CTL) attack, whereas HLA class II antigen induction in the absence of costimulatory molecule expression may impair activation of HLA class II-restricted antitumor immune response.2 Indeed, studies have demonstrated HLA class I antigen loss in melanoma, this phenomenon being considered by many as playing a role in evasion from immune surveillance in vivo.3 Descriptions of genes included within the human major histocompatibility complex (MHC) draw new interest on the specific features of the nonclassical HLA class Ib antigens. Among them, HLA-G is characterized by a low polymorphism and a tissue-restricted pattern of protein expression, limited to extravillous cytotrophoblast cells, endothelial cells in the chorionic villi, amniotic cells and fluid and thymic medullar and subcapsular epithelial cells.4 HLA-G displays membrane-bound and soluble protein isoforms.5 Although the function(s) of HLA-G antigens remain(s) to be further elucidated, their ability to downmodulate immune functions through interaction with inhibitory receptors expressed on natural killer cells, T lymphocytes, monocytes/macrophages, B cells and dendritic cells has been demonstrated in vitro.6, 7, 8, 9, 10, 11, 12 Moreover, availability of the HLA-G leader peptide has also been shown to favor cell surface expression of HLA-E, thus favoring further immunomodulation of NK cell function.13 Finally, it has been recently reported that the soluble full-length HLA-G5 protein isoform could trigger apoptosis in activated CD8+ cells14 and that the secretion of HLA-G5 by early embryos is correlated with the success of implantation and subsequent pregnancy.15
In an initial survey of melanoma, tumor-specific upregulation of all alternative HLA-G mRNA levels as well as significant HLA-G protein expression were detected in biopsies of melanoma lesions.16 These observations were recently strengthened in 2 studies showing an increased soluble HLA-G serum concentration in melanoma patients17 and a correlation between poor prognosis of melanoma treated with interferon alpha-2b and the presence of HLA-G in biopsies of tumor lesions.18 Upregulation of HLA-G expression was also observed in renal carcinoma,19 breast carcinoma,20 lung carcinoma,21 lymphoma,22 glioblastoma23 and ovarian carcinoma.24 These studies evoke a role for HLA-G expression in the immune evasion of HLA-G-positive tumors. We here investigated whether HLA-G upregulation could represent a specific feature of melanoma cells and further examined the relevance of HLA-G expression in prediction of malignancy as compared to other alterations of HLA class Ia and class II antigen expression in melanocytic lesions.
All formalin-fixed, paraffin-embedded skin specimens surgically removed in the dermatology unit in Hôpital Tenon (Paris, France) from 1994 to the beginning of 2000 were reviewed. All cases diagnosed as naevus, lentigo or melanoma were included. In addition, randomly selected samples of lentigo and cutaneous melanoma were obtained from the dermatology department of Institut Gustave Roussy (Villejuif, France). The histopathologic examination of haematoxylin-, eosin- and safran-stained sections was blindly reviewed by 2 pathologists (Y.A. and F.B.). The 172 lesions studied were classified into 4 groups according to the Clark's classification25 corresponding to common (n = 54) and atypical (n = 16) naevocellular naevi, lentigo simplex (n = 23) and cutaneous melanoma including lentigo maligna (LM; n = 6), lentigo maligna melanoma (LMM; n = 5), superficial spreading melanoma (SSM; n = 33), acrolentiginous melanoma (ALM; n = 6), nodular melanoma (NM; n = 1), other unclassified melanoma (n = 4) and cutaneous metastasis (n = 24). In addition, 1 lymph node melanoma metastasis was included. The histologic parameters of each sample were further analyzed by histochemistry using HMB45 (anti-gp100/Pmel17) and A103 (anti-MelanA/MART-1).26, 27, 28 A control group of normal adult skin biopsies (n = 16) was also analyzed. All tissues were cut at 4 μm thickness, mounted on precleaned glass microscope slides (Menzel-Gläser, Braunshweig, Germany).
The primary monoclonal antibodies (mAbs) used in our study were the anti-MelanA/MART-1 antibody (clone A103, IgG1; Dako, Glostrup, Denmark), the anti-gp100/Pmel17 (clone HMB45, IgG1; Dako, Carpinteria, CA), the anti-Leucocyte Common Antigen/CD45RB (clone PD7/26, IgG1; Dako, Denmark), the anti-HLA-DP, -DR, -DQ antigens (clone CR3/43; Dako, Denmark), anti-native and denatured HLA-G heavy chain29, 30 (4H84, IgG1, diluted ascitic fluid; kindly provided by M. McMaster, University of California, San Francisco, CA), an anti-β2m-free HLA-B and -C heavy chains determinant31 (HC10; kindly provided by H.L. Ploegh, Harvard Medical School, Boston, MA).
Paraffin-embedded tissues were deparaffinized using toluene, rehydrated through a graded series of ethanol and rinsed in distilled water. The sections were submitted to epitope retrieval by high temperature in 10 mM sodium citrate buffer (pH 6.0) using a commercial microwave oven, prior to immunochemistry analysis to optimize immunoreactivity. Endogenous peroxidase activity was quenched by treating sections for 5 min at room temperature (rt) with 3% hydrogen peroxide in water. Prior to staining procedures, nonspecific binding was prevented by applying 30% human serum for 20 min. Primary antibodies were incubated for 30 min at rt and revealed using the Ultra-Tech HRP streptavidin-biotin universal detection system (Becton Dickinson, Immunotech, Marseille, France) according to the manufacturer's instructions. Sections were counterstained with Mayer's haematoxylin solution for 7 min, cover-slipped with permanent mounting media and viewed with a standard light microscope. The HLA-G-positive cell type was determined by careful analysis by specialized cutaneous pathologists (Y.A. and F.B.) of morphologic features of 4H84-labeled cells in comparison to serial sections labeled with various antibodies targeting melanocytes and leucocytes.
Comparisons between groups were done using χ2 test and Fisher's exact test. Various parameters were tested for an association with time to recurrence and survival in a univariate analysis using a 2-sided log-rank test.
In contrast to downregulation of classical HLA class I antigens, aberrant class II antigen expression in melanocytic cells is significantly associated with malignancy, whereas both alterations are associated with tumor thickness of primary melanoma
Analysis of HLA classical class I and class II antigen expression is detailed in Table I. All specimens of normal skin and lentigo showed a bright immunoreactivity within epidermal cells when incubated with anti-HLA-B and -C heavy chains HC-10 mAb. In contrast, naevi and melanoma displayed abnormalities of this antigen expression in 40.9% and 54.4% of cases, respectively. These abnormalities, characterized by partial or total defects in HLA class Ia antigen expression, only concerned lesional melanocytes (Fig. 1a). In our series, although defects of HLA Ia antigen expression were significantly increased in primary melanoma compared to lentigo (p < 0.0001), there was no statistical difference between naevi and primary melanoma. Moreover, occurrence of HLA class Ia antigen defects did not significantly differ whether the lesions were or were not atypical within the naevi group. However, compound and dermal naevi more often displayed HLA class Ia antigen abnormalities than junctional naevi (p < 0.0002). Interestingly, fine analysis of HLA class Ia molecule expression evidenced exclusive location of HLA class I antigen loss within the dermal component of melanocytes, whether the proliferation was benign or not. In addition, we found a significant increase of HLA Ia antigen defects within primary melanoma of Breslow index superior to 0.75 mm (p < 0.001). The higher rate of HLA class I antigen loss observed in metastatic cutaneous lesions (70.8%) compared to primary melanoma (47.3%) did not reach statistical significance (p < 0.08).
Table I. Comparative Analysis of HLA-G Antigen Expression, HLA-B and -C Antigens Alteration of Expression and HLA Class II Molecules Induction of Expression in Normal Skin, Naevi, Lentigo and Melanoma Biopsies by Immunohistochemistry
No HLA class II antigen expression could be detected on melanocytes within the 84 benign melanocytic lesions tested to the exception of a unique dermal naevus. This contrasted with the rate of activation of HLA class II antigen-positive melanocytic cells detected in primary melanoma (29.1%) and to a higher extent in cutaneous metastasis (50.0%). As observed for HLA-Ia antigen deficiencies, induction of HLA class II molecule expression was significantly increased within primary melanoma when considering the 0.75 mm Breslow barrier (p < 0.0001). Although a higher rate of HLA class II antigen expression is observed in metastatic melanoma lesions (50%) as compared to primary melanoma (29.1%), this was not statistically significant (p < 0.1).
Differential expression of HLA-G antigen in melanocytic lesions correlates with malignancy
HLA-G antigen expression was tested by immunohistochemistry in a series of 172 surgically removed melanocytic lesions as well as in control healthy skins (Table I). HLA-G expression was observed at different rates in melanocytic skin lesions. HLA-G expression, as analyzed without consideration of the type and number of cells immunostained, was significantly higher in melanoma (35.4%) as compared to whole nonmalignant melanocytic lesions (7.53%, p < 0.0001), naevi (7.14%, p < 0.0001), lentigo (8.70%, p < 0.025) or normal skin (0%, p < 0.01). No differences were noted between the rate of HLA-G expression observed in naevi whether these displayed atypia (12.5%) or not (5.56%). Interestingly, HLA-G expression was evidenced in the 2 congenital naevi that were available for testing. The nature of HLA-G-positive cells was variable (Table I) and included melanocytic cells (Fig. 1b,d–f) but also lymphocytes (Fig. 1c–g), macrophages (Fig. 1c–g) or to a lesser extent endothelial cells (Fig. 1e) and keratinocytes (Fig. 1f). Of the 11 lentigo analyzed, 1 located on the sole was characterized histologically by a high density of melanophages. These cells as well as melanocytes, inflammatory infiltrating cells and capillary endothelial cells expressed HLA-G (Fig. 1e). Finally, no correlation between age, sex or other past medical history of patients and HLA-G expression could be established.
Within primary melanoma, the rate of HLA-G-positive lesions seemed to increase with the Breslow index (25% when Breslow is < 0.75 mm and 44.4% when Breslow is > 0.75 mm), but this difference did not reach significance. Of note, HLA-G was never detected in lentigo maligna, the melanocytic malignancy associated with the best prognosis. The rates of HLA-G antigen detection in primary and metastatic cutaneous melanoma were not found to be significantly different (34.6% and 37.5%, respectively). Finally, 24 of the 25 specimens that exhibited HLA-G-positive staining of melanoma cells were also analyzed for classical HLA class I antigens loss using HC-10 mAb. In all cases, no HLA class Ia antigen defect could be evidenced in HLA-G-positive melanoma cells, demonstrating that there was no correlation between HLA class Ia antigen defects and HLA-G activation in melanoma cells in our series (Fig. 1a,g). Similarly, no correlation between HLA-G and HLA class II antigen induction in melanoma cells (Fig. 1f,g) could be established in this series.
Correlation between HLA antigen expression in primary melanoma and clinical course of the disease
HLA class II, class Ia and -G antigen expression alterations were tested in a univariate analysis in order to search for an association with time to recurrence and survival of melanoma patients whose primary tumor was analyzed by immunohistochemistry (Table II). As expected, the Breslow index, the Clark level of biopsies studied and the clinical stage of patients were confirmed to be prognostic markers of disease. Neither HLA-G nor HLA class II antigen induction of expression was correlated to recurrence-free interval or overall survival. On the other hand, HLA-B and -C antigen loss of expression were significantly associated with progression of the disease. This result highlights the probable involvement of HLA class Ia antigen downregulation in tumor escape from immune surveillance mechanisms.
Table II. Association Between the Delay Before Relapse and The Time of Survival After the excision of the Primary Melanoma Studied (Significance Degree, Log-Rank Test)
Delay before relapse
Time of survival
Parameters significantly correlated with the progression of the disease are in bold.
Association between the degree of inflammatory infiltrate and HLA antigen expression in primary melanoma
We have also evaluated whether alteration of HLA class I and II antigen expression in melanoma cells might be related to the intensity of the lesional inflammatory infiltrate in primary melanoma. For this purpose, inflammatory infiltrate were scored as 1 (low level), 2 (moderate level) and 3 (high level). As represented on Figure 2, the increase of inflammatory infiltrate correlated with defects in HLA-B and -C antigen expression. The intensity of the inflammatory infiltrate was better correlated with induction of HLA class II molecule and to a higher degree with induction of HLA-G antigen in melanoma cells. In light of these results, we reexamined the level of inflammatory infiltrate found in the benign lesions analyzed in our series. Interestingly, of the 3 benign lesions in which melanocytes were stained by anti-HLA-G antibodies, 2 exhibited a strong inflammatory infiltrate within the lesion. Moreover, of the 93 benign melanocytic lesions analyzed, only 4 were highly infiltrated by inflammatory cells, and HLA-G antigen-positive cells were evidenced in 3 of 4 of these highly inflammatory benign lesions.
HLA-G antigen expression is correlated with that of HLA-A1 in melanoma patients
The HLA genotype of patients studied was available in 28 cases and HLA-G expression was detected in 12 of 28 of these genotyped tumors. All HLA-A1 patients (9 of 28) were also positive for HLA-G. We thus evidenced a strong correlation between HLA-G upregulation in melanoma and HLA-A1 typing of the patients (Table III, p < 0.0001). Association with HLA-B8 or with the HLA-A1, -B8 haplotype was also noted, as 3 of the 3 HLA-B8 melanoma tested contained HLA-G antigen-positive cells (p < 0.07), all corresponding to HLA-A1 patients.
Table III. Correspondence Between HLA-G Antigen Detection and the HLA-A Allele Typing of 27 Melanoma Specimens
No. of tumors
Rate of HLA-G-positive tumor (%)
HLA-G antigen detection is significantly more frequent in HLA-A1 patients than in the other patients, p < 0.0001.
The hypothesis that HLA-G upregulation may be linked to the presence of both high intralesional inflammatory infiltrate and to HLA-A1 typing was further supported in 2 additional melanocytic lesions obtained from 2 HLA-A1 patients whose primary melanoma contained HLA-G antigen-positive melanoma cells and inflammatory infiltrating cells. The first one was a naevus with both junctional and dermal naevoid components, surgically removed 7 months after the primary tumor excision. This highly inflammatory benign lesion contained both HLA-G antigen-positive melanocytes and inflammatory infiltrating cells (Fig. 1f). The second additional specimen analyzed was a lymph node metastasis that presented HLA-G and leukocyte common antigen-positive cells facing HMB45-positive melanocytic tumor cells lacking both HLA-B, -C and -G antigens (Fig. 1g).
Development of melanoma occurs in stages involving various regulatory cascades that alter gene expression.32 Indeed, in addition to genetic predisposition such as mutations in p16 and p53 genes, environmental factors, such as UV light or bone marrow transplantation, are recognized as risk factors.33, 34 To establish a better diagnosis and explore new therapeutic strategies, markers designed to characterize each step of melanoma development are constantly evaluated.35, 36, 37, 38, 39, 40 Among these, the nonclassical HLA class I gene, HLA-G, was reported to be abnormally expressed in melanoma biopsies and inducible by interferon treatment.16, 18 To better evaluate the link between HLA-G protein expression and tumorigenesis of melanocytic cells, HLA-G expression was assessed in both benign and tumoral paraffin-embedded melanocytic lesions. Alteration of other HLA antigens, namely defective HLA class Ia antigen expression and induction of HLA class II antigen expression, were also reevaluated in our study, taking into account specific detection of HLA-G.41 In accordance to previous studies,16, 42, we confirmed that HLA-G protein was not expressed in normal skin. Upregulation of HLA-G expression was detected either in melanocytic cells and/or in inflammatory infiltrating cells (predominantly in monocytes/macrophages but also in lymphocytes). The finding that HLA-G expression is significantly higher in primary melanoma than in naevi (p < 0.0003) suggests that activation of expression is not a common feature of melanocytic cells but rather associated with malignant transformation of this cell type.
It has been postulated that HLA class II antigens were expressed in fetal/newborn melanocytes, repressed in adult normal melanocytes and could be reactivated in subsets of melanoma.43 In our study, almost 34% of melanoma displayed HLA class II antigen expression within various proportions of melanocytic cells. This level is slightly lower than the 40–70% rates of expression previously reported in melanoma by several authors.44 Such discrepancies may rely on the fact that previous studies were conducted using different mAbs to access MHC class II antigen expression and on frozen sections that frequently do not allow precise discrimination of cell morphology. Nevertheless, our results confirm that, as for HLA-G, activation of HLA class II antigen expression in melanocytes is significantly higher in primary melanoma that in naevi. In contrast, alterations of HLA-B and -C antigen expression were not found to be significantly different in naevi and primary melanoma.45 Our observations indicate that HLA-G and HLA class II antigen detection in melanocytic cells better correlates with malignancy diagnosis than HLA Ia antigen loss of expression. Concerning the prognosis of malignant lesions, there were no statistical correlation between HLA-G or HLA class II antigen expression and reduced time to disease progression and reduced survival in an univariate analysis. Nevertheless, we confirmed that HLA Ia antigen downregulation in primary lesions is significantly associated with reduced time to disease progression.46 We also confirmed that HLA class Ia antigen downregulation and HLA class II antigen induction in melanoma cells were correlated with tumor thickness, a marker of poor prognosis in primary melanoma.44, 46, 47 In contrast, although the frequency of HLA-G antigen upregulation tends to be higher in melanoma of Breslow index > 0.75 mm, its correlation with tumor thickness could not be firmly established. It is of note that melanoma that have a good prognosis, lentigo maligna, were always found to be devoid of HLA-G activation in this series.
Since most of the few benign lesions showing HLA-G antigen in melanocytes were characterized by a high density of the inflammatory infiltrate, we asked whether the intensity of the intralesional inflammatory infiltrate might be correlated with HLA-G antigen upregulation. We observed a tight correlation between HLA-G expression and the level of inflammatory infiltration in primary melanoma. Similar correlation was observed for HLA class II antigen induction and to a lesser extent for HLA Ia antigen downregulation. Therefore, various alterations of MHC expression in tumor cells were found to occur in melanoma in association with the high inflammatory status of the tumor. Whether these events rely on common soluble mediators or cell interactions provided by infiltrating cells remains to be defined. Alternatively, specific alterations of MHC at the surface of tumor cells or production of soluble MHC molecules by tumor cells may also modify cytokine production21, 48, 49 or effector function of infiltrating cells14, 50 and locally favor expansion of the infiltrate.
Our study is the first to our knowledge to describe a correlation between upregulation of HLA-G expression and expression of the HLA-A1 allele in melanoma patients. The HLA-G locus has been shown to be in genetic disequilibria with the HLA-A gene.51, 52 Several studies have described an association between high production of soluble HLA class I antigens and some HLA-A alleles.53, 54, 55, 56, 57 Recently, Rebmann et al. also reported an association between HLA-G polymorphism and the level of both HLA-G and HLA Ia soluble antigens in the serum of healthy individuals.58 Moreover, a genetic disequilibria was evidenced between “high secretor” HLA-A24 and -A33 alleles and HLA-G*0104 allele, suggesting that coordinated regulatory mechanisms may be involved in the control of allele-specific HLA-A and HLA-G secretion levels. In our series, association of HLA-G expression with HLA-A24 and -A33 alleles could not be evidenced as HLA-G expression was mainly found in HLA-A1 patients, which are reported as “intermediate secretor genotype”.58 Such discrepancy may be explained by the fact that the antibody we used in our study does not allow discrimination between membrane-bound and soluble HLA-G protein, knowing that expression of these isoforms may be differentially regulated in melanoma.16 Considering the functional features of HLA-G in modulating NK and CTL cytotoxic activity, this information may prove to be valuable for better prediction of patients who may benefit from vaccination trials based on eliciting an HLA-A-restricted T-cell response directed toward tumor antigens. Genotyping of the HLA-A1 allele that restricts presentation of widely expressed tumor-specific antigens such as MAGE-3 to cytotoxic T lymphocytes is commonly considered before inclusion of patients in vaccination trials with MAGE-3 peptides.59
HLA-G antigen upregulation has been recently evidenced in various diseases including viral infections,60, 61 cancers16, 17, 19–24 and may be beneficial to allogenic graft or fetal tolerance.62 Our results suggest that upregulation of HLA-G and alteration of other MHC antigens may result from several particularities of tumor cells and be influenced by infiltrating cells of the tumor microenvironment that provide signaling for this activation and may in turn be affected by HLA-G expression. Various cytokines that are often produced within the tumor microenvironment, such as IL-10 and interferons, have been shown to induce HLA-G in vitro.19, 24, 63–67 IL-10 expression has been shown be associated with HLA-G expression in lung carcinoma21 and may exert adverse effects on HLA class Ia and b expression.
The potential role of HLA-G as a molecule interfering with immune elimination of melanoma cells was recently emphasized by Wagner et al., who found that the presence of HLA-G, together with partial or total loss of HLA class Ia antigens, was associated with a poorer prognosis of the malignancy.18 These authors suggested that a combination of both abnormalities provides these tumor cells with efficient ways for escape from host immune response. However, our present study is in contradiction with the frequent occurrence of such tumors and shows that although both HLA-G and HLA class I antigen loss are associated with the malignant phenotype of melanocytes or to parameters of tumor progression, respectively, combined alteration of HLA-G and HLA class Ia antigen expression is not frequently detected on the same tumor.
The fact that nearly all in vitro functional studies demonstrated inhibiting functions of HLA-G strongly suggests that the presence of this molecule in melanoma lesions may contribute to the escape from host immune response. Although HLA-G expression is the best MHC predictor of the malignant nature of melanocytic cells, failure to correlate expression of this antigen with a worse clinical outcome or with patient survival does not allow further speculation on its exact role in favoring tumor progression. Identification of means to indirectly predict HLA-G expression, through HLA-A typing commonly documented in vaccination trials, may allow further studies interpreting the role of HLA-G in the downregulation of immune responses during the course of peptide, exosomes or dendritic cell vaccination.
In summary, we here provide evidence that activation of HLA-G expression occurs during the process of melanocyte malignant transformation and is correlated with the presence of a high density of inflammatory infiltrate and HLA-A1 genetic background. This expression is probably the consequence of various triggering factors found within the tumor microenvironment, which remain to be defined more precisely. Previously reported inhibitory functions of HLA-G toward cytotoxic immune effector cells suggest that this molecule may participate in escape from host immune response and thus favor tumor progression. Future studies designed to define the precise impact of HLA-G expression on TILs or dendritic cell function in melanoma could thus give us new insights on how to improve targeting of immunotherapies against tumors. Alternatively, evaluation of HLA-G upregulation during tumor progression, vaccine and IFN therapies could favor a better selection or monitoring of patients who may fail to respond to treatment.
We are grateful to Dr. M. McMaster and Dr. H.L. Ploegh for the kind gift of antibodies. We thank Ms. I. Krawice for technical assistance. We are grateful to Mr. E. Angevin for HLA typing of patients at the Institut Gustave Roussy. We acknowledge Mrs. J.-A. Gavigan for editing the manuscript. We thank Mrs. E. Savariou, Mr. R. Nancel and Mr. B. Boursin for photographic work. E.C.I. is a recipient of a grant from the Association pour la Recherche sur le Cancer.