An MGRN1-Based Biomarker Combination Accurately Predicts Melanoma Patient Survival

Abstract

With ever-increasing incidence and high metastatic potential, cutaneous melanoma is the deadliest skin cancer. Risk prediction based on the Tumor-Node-Metastasis (TNM) staging system has medium accuracy with intermediate IIB-IIIB stages, as roughly 25% of patients with low-medium-grade TNM, and hence a favorable prognostic, undergo an aggressive disease with short survival and around 15% of deaths arise from metastases of thin, low-risk lesions. Therefore, reliable prognostic biomarkers are required. We used genomic and clinical information of melanoma patients from the TCGA-SKCM cohort and two GEO studies for discovery and validation of potential biomarkers, respectively. Neither mutation nor overexpression of major melanoma driver genes provided significant prognostic information. Conversely, expression of MGRN1 and the melanocyte-specific genes MLANA, PMEL, and TYRP1 provided a simple 4-gene signature identifying with high-sensitivity (>80%), low-medium TNM patients with adverse outcomes. Transcriptomic analysis of tumors with this signature, or from low-medium-grade TNM patients with poor outcomes, revealed comparable dysregulation of an inflammatory response, cell cycle progression, and DNA damage/repair programs. A functional analysis of MGRN1-knockout cells confirmed these molecular features. Therefore, the simple MGRN1-MLANA-PMEL-TYRP1 combination of biomarkers complemented TNM staging prognostic accuracy and pointed to the dysregulation of immunological responses and genomic stability as determinants of a melanoma outcome.

1. Introduction

Cutaneous melanoma (MM), the deadliest type of skin cancer, is one of the human cancers with higher metastatic potential and mutational burden due to the genotoxicity of solar ultraviolet radiation [1,2,3,4,5]. Its high mutational rate likely contributes to tumor heterogeneity [6] and the generation of neoantigens that confer high immunogenicity to the lesion [7,8,9,10], enabling immune-mediated disease control [11,12]. The main MM driver mutations have been identified, leading to targeted treatment opportunities. Frequent and mutually exclusive oncogenic mutations on the BRAF, NRAS, or NF1 genes account for roughly 75% of MMs [13,14,15,16]. Accordingly, MM has been classified into four molecular subtypes, namely BRAF-, NRAS-, or NF1-mutated and triple wildtype (WT) [13].

Immunohistochemical diagnosis of MM is mostly based on melanocyte-specific markers such as PMEL, MLANA, or TYR, along with relatively unspecific and pleiotropic cancer biomarkers like S100B, and proliferation-related markers like MKI67 [17,18]. Unfortunately, the prognostic value of these biomarkers, except maybe PMEL [19], remains uncertain. Moreover, in most cancers, neither the presence of mutations in major drivers nor their level of expression within the tumor provides significant prognostic insight [20].

Currently, the prediction of MM patients’ overall survival (OS) and the choice of treatment options are mostly based on the Tumor-Node-Metastasis (TNM) staging system of the American Joint Committee on Cancer (AJCC) [21], relying on tumor features such as Breslow depth, ulceration, and presence of in-transit, satellite, and/or microsatellite metastases. Low-medium TNM stage (0–IIIB) MMs have good prognostic with high 5-year OS rates, and the exeresis of the lesion is often curative. Accordingly, to avoid overtreatment, MM patients with a priori favorable low-medium-grade TNM are not usually treated with adjuvant therapies. Conversely, patients with advanced, high-grade TNM (IIIC–IV) MMs still have a grim prognosis and are eligible for targeted therapies or immunotherapy [22,23]. However, roughly 25% of low-medium-grade TNM patients undergo an aggressive disease with OS comparable with high-grade TNM patients [24]. Thus, the TNM-based prognostic of stage IIB–IIIB MM patients remains uncertain. In this ambiguous situation, MM patients in stages IIIA–IIIB are not eligible for immunotherapy in many countries [25]. To better stratify patient risk, several methods [26,27,28] and promising emerging biomarkers such as PRAME [29], as well as prognostic genetic signatures, have been proposed [30,31,32], but to our knowledge, none of them is currently widely accepted or included in clinical trials. Thus, reliable MM prognostic biomarkers are still required to improve the choice of treatment options and to avoid the problems of under- or overtreatment.

Mahogunin Ring Finger-1 (MGRN1) is a RING finger domain-containing E3 ubiquitin ligase with key functions in melanocytes and MM cells [33,34]. A natural complete loss-of-function mutation at the md locus of mice leads to dark coat color and abnormal body weight [35], abnormal left-right axis patterning, congenital heart defects, and high embryonic lethality. Viable adults develop spongiform-like neurodegeneration [35]. We previously reported that Mgrn1-null mouse melanocytes displayed a differentiated phenotype with numerous highly melanized melanosomes [36]. MGRN1-deficient melanoma cells also displayed differentiated and less aggressive phenotypes, with induction of intercellular contacts and matrix adhesion, reduced motility, migration, and invasion capabilities, and aberrant cell cycle progression with significant accumulation of cells in the S and G2/M phases [37,38,39]. Based on these observations and on the recent report that genes whose expressions correlate with adverse outcomes across cancer types often encode for housekeeping genes with roles in cell cycle progression [20], we hypothesized that the level of expression of MGRN1 in MM might provide useful prognostic information. This might be particularly relevant in ambiguous clinical situations where the accuracy of TNM-based risk prediction is insufficient.

To identify suitable predictive biomarkers that could complement TNM staging, here we used the expression data from the GDC-SKCM cohort of The Cancer Genome Atlas (TCGA) project [40] and other MM datasets to analyze the relationship between MGRN1 expression, alone or combined with several known MM biomarkers, and patient OS. We also investigated the effects of MGRN1 expression on the transcriptomic landscape across the SKCM cohort. We found that the level of MGRN1 expression, alone or combined with the differentiation markers PMEL, MLANA, and TYRP1, (i) provided prognostic information on the OS of MM patients with unprecedented accuracy, (ii) was associated with specific transcriptomic landscapes across MM, with differential expression co-occurring with dysregulation of gene sets involved in immunological responses, cell cycle, and DNA damage/repair, and (iii) identified a subset of patients with unexpectedly short OS based on their TNM stage. Moreover, we validated our in silico transcriptomic analyses of MM tumors using MGRN1 knockout in cultured cells. Our results showed that information based on MGRN1 expression in combination with PMEL, MLANA, and TYRP1 reliably complemented the prognostic information of TNM staging, underlining the potential of this small set of genes as an accurate and sensitive prognostic biomarker for MM.

2. Results

2.1. TNM-F Patients with Short OS Exhibit a Specific Transcriptomic Profile

The prediction of the outcome of MM patients is currently based on TNM staging according to the AJCC [21]. Using clinical information of over 470 patients in the GDC-SKCM dataset from TCGA, we analyzed the impact of TNM at diagnosis on patients’ OS (Figure 1).

As expected, the median survival of SKCM dataset patients decreased as the TNM stage increased (Figure 1a). A highly significant difference in survival rates was also observed when we grouped all patients into two categories: patients with a priori favorable stages (TNM-F, stages 0–IIIB, median OS 1419 days), or patients with unfavorable prognostic (TNM-NF, stages IIIC–IV, median OS 590 days, HR = 3.3) (Figure 1b,c), confirming the association of lower TNM at diagnosis and longer OS. Importantly, these results also underlined the medium-low accuracy of TNM-based risk prediction, since roughly 25% of TNM-F patients with a priori favorable prognosis had shorter OS than the median of TNM-NF patients (590 days, p < 0.0001) (Figure 1b,d). We looked for specific molecular patterns associated with better OS, using gene set enrichment analysis (GSEA) [41] to compare the transcriptomes of TNM-F patients with long OS (>3000 days, ~2-fold the median for TNM-F patients, green box in Figure 1b) or short OS (<590 days, pink box in Figure 1b). Volcano plots showed disparate transcriptomic landscapes in these groups, with upregulation of EMT, inflammatory, and interferon response gene sets and downregulation of cell cycle and DNA damage-related gene sets, as well as several melanocyte-specific genes and known cancer biomarkers in TNM-F patients with longer survival (Figure 1e, left and right panel). Such differences were not found when we compared the transcriptomes of MM from TNM-NF on one hand, and TNM-F patients with short OS on the other, indicating similar transcriptomic landscapes in these adverse conditions, independently of their TNM at diagnosis. This suggested the possibility of finding molecular biomarkers for the subset of TNM-F patients with unexpectedly short OS.

To look for such markers, we first analyzed the impact of mutation or overexpression of the major drivers that define the molecular subtypes of MM (BRAF, NRAS, NF1) [13,14,15,16], using information from the GDC-SKCM cohort. We also considered TP53-mutated tumors in this analysis, since TP53 encodes for a tumor suppressor frequently mutated in SKCM [15,16] whose mutation correlates with worse outcomes in many cancers [20]. Kaplan–Meier curves for patients of the different subtypes were similar, either when we analyzed the complete SKCM cohort of TCGA (Figure S1a) or the subset of TNM-F patients (Figure S1b). We also obtained similar results when we compared carriers of mutated and WT forms of the BRAF, NRAS, NF1, or TP53 genes (Figure S1b). Moreover, no significant differences were observed between the normalized expression of BRAF, NRAS, NF1, and TP53 in SKCM tumors and normal skin (Figure S1c). Finally, Kaplan–Meier curves for patients with high or low expression of these genes yielded similar OS rates except for NF1 and BRAF. In this case, we found a roughly 2-fold increase in median OS for patients bearing tumors with high BRAF expression compared with low BRAF expression (Figure S1d) for the complete SKCM cohort, and a similar, although insignificant trend for the TNM-F subset of patients. This might be related to the effect of BRAF-targeted therapies.

2.2. The Level of Expression of MGRN1 and Genes Involved in Melanocyte Differentiation Predicts Patient Survival Better than Driver Genes

The previous results indicated that the major MM drivers provided little prognostic information identifying high-risk patients with a priori favorable TNM. Thus, we focused on the genes differentially expressed in shorter or longer survival patients highlighted in Figure 1e. MKI67 and S100B are low-specificity biomarkers used in several cancer types. PMEL, MLANA, and MITF are also employed as diagnostic markers by pathologists [17,18,19], and there is some evidence pointing to the prognostic value of PMEL expression [19,42]. TYR, TYRP1, and DCT encode for melanogenic enzymes [43] determining the amount and type of melanin pigments formed within melanocytes. MC1R, encoding for α-MSH receptor 1, is a well-known genetic determinant of MM and non-melanoma skin cancer risk [43,44]. Finally, MGRN1 encodes for an E3 ubiquitin ligase interacting with MC1R [33], which was previously suggested as a potential prognostic MM biomarker [37] based on strong effects on the phenotype of melanocytes and MM cells [36,38,39].

All these genes, except TYRP1, were overexpressed in MM vs. normal skin (Figure 2a–j, box plots on the left of each panel). Except for S100B, survival analysis using the complete SKCM cohort showed a significant association of lower expression of the potential markers with favorable outcomes and longer OS (Figure 2a–i, graphs on the right of the panels).

The increase in median OS of patients from the favorable lower expression tercile ranged from 1.6- to 2.0-fold. Among these potential markers, MGRN1 was particularly attractive in light of its strong effects on the differentiation and genomic stability of MM cells [36,38,39]. To further explore the association of MGRN1 expression and MM outcome, we divided the cohort into subgroups of patients with longer or shorter survival than the median OS of the cohort (1070 days, Figure 3a). We observed an association of high MGRN1 expression with adverse outcomes.

Next, we compared the OS of patients in the higher and lower terciles of MGRN1 expression (MGRN1-High and MGRN1-Low), and we found again a significant adverse effect of high MGRN1 expression (Figure 3b). GSEA of these subgroups of tumors with low or high MGRN1 expression revealed a markedly different transcriptomic landscape (Figure 3c). Compared with high MGRN1 expression MM, low MGRN1-expressing tumors displayed positive normalized enrichment scores (NES) for the epithelial-mesenchymal transition (EMT), ultraviolet response down, and inflammatory response gene sets, and negative enrichment for gene sets related to cell cycle control, and DNA damage repair. Since the impaired expression of genes involved in cell cycle regulation and DNA repair may lead to genomic instability, we interrogated the transcriptomic data for features of DNA damage [4]. We found higher mutational burden and chromosome number alterations in MGRN1-Low compared with MGRN1-High tumors, and a similar trend for the weighted genome instability index (wGII) (Figure 3d). These data were consistent with preliminary previous data suggesting a role of MGRN1 in genomic stability in mouse [39] and human MM cells [37].

We next validated the transcriptomic data using functional assays and explored a causal relationship between MGRN1 deficit and the transcriptomic differences in MGRN1-Low and -High MM (Figure 4). To this end, we obtained independent MGRN1-KO clones of HBL human MM cells knocked out for MGRN1 with CRISPR-Cas9, as previously described [37] (clones 2.1, 3.7, and 4.9). MGRN1 knockout was confirmed by analysis of the edited sequences and Western blot [38]. All clones displayed early truncation of the MGRN1 protein with loss of all relevant functional domains, thus ensuring complete loss-of-function of the truncated proteins. One of these clones (4.9) was analyzed by RNAseq to compare its transcriptome with control cells and with MGRN1-Low MM. GSEA revealed significant upregulation of inflammatory response-related gene sets and repression of cell cycle-related sets in MGRN1-KO cells compared with control cells (Figure 4a), as previously observed in MGRN1-Low compared with MGRN1-High MM. Next, we identified the specific genes differentially expressed ([FC] > 1.25) in MGRN1-KO cells or MGRN1-Low MM that contributed to NES of the inflammatory response, p53 pathway, and cell cycle-related gene sets, to build the corresponding heat maps. A strong and consistent differential expression of these genes was observed in MGRN1-KO cells compared with control cells (Figure 4b) as well as in MGRN1-Low compared with -High MMs. Therefore, transcriptomic analysis revealed a comparable dysregulation of inflammatory response and DNA metabolism genes in MGRN1-Low MM and MGRN1-KO HBL cells. Importantly, to validate our transcriptomic analysis, we compared the secreted cytokines in conditioned media from control and MGRN1-KO cells using protein arrays and multiplex analysis (Figure 4c). We observed a 4-fold lower secretion of the master anti-inflammatory cytokine IL10 in media conditioned by MGRN1-KO cells, consistent with the upregulation of the inflammatory response gene set (Figure 3c and Figure 4a).

Finally, to prove a causal relationship between MGRN1 deficiency and genome instability, we compared the burden of DNA breaks in control and MGRN1-KO cells using γH2AX labeling (Figure 4d) and comet assays (Figure 4e). Cells from all MGRN1-KO clones exhibited increased γH2AX labeling and comet tail moments, confirming an increased burden of DNA strand breaks following the knockout of MGRN1.

2.3. Comparable Gene Expression Patterns in Patients with Longer OS and Low MGRN1, PMEL, MLANA, or TYRP1 Expression

Since the level of expression of MGRN1 and the melanocytic genes analyzed above were significantly correlated with the median survival of patients of the complete SKCM cohort, we next interrogated the TNM-F subgroup for a possible association of any of these genes with unexpectedly short OS rates. We obtained plots of OS for TNM-F patients with 33% higher or lower expression of the genes and the corresponding Kaplan–Meier curves (Figure 5, panels a–j). Both analyses showed significantly shorter OS of TNM-F patients expressing high levels of MGRN1, PMEL, MLANA, and TYRP1. We confirmed a significantly higher expression of these genes in the tumors from TNM-F patients with short OS (<590 days, Figure 5, panel k). For the other potential biomarkers, results were less clear-cut in that at least one of the three analyses performed (median OS of patients with high or low gene expression, Kaplan–Meier analysis, and median gene expression in long or short OS patients) did not yield a statistically significant result. Accordingly, special attention was paid to MGRN1, PMEL, MLANA, and TYRP1.

Interestingly, for each of these genes, the GSEA of the low vs. high expression tumors revealed comparable profiles. These profiles were consistent with those of MGRN1-KO vs. control cells, with positive enrichment of EMT and inflammatory response gene sets, and negative enrichment of p53 pathway, UV response, and gene sets involved in cell cycle and DNA damage or repair in the low expression group (Figure S2, panels a–d). Given the similar results obtained for MGRN1, PMEL, MLANA, and TYRP1, we asked whether the expression of these four genes might be regulated by common mechanisms so that the information that they may provide would be redundant rather than additive. Therefore, we plotted the expression of each of these genes against all others (Figure S2, panel e). The only significant correlation was between MLANA and PMEL, consistent with the major role of MITF as a transcriptional regulator of both genes [45].

2.4. An MGRN1-Based Expression Panel Complements TNM-Based Prediction of Outcome

The results summarized in Figure 5 and Figure S2 showed that high expression of MGRN1, PMEL, MLANA, and TYRP1 was associated with an adverse outcome in TNM-F stage patients. To confirm additivity of the information provided by these biomarkers, we constructed datasets of TNM-F patients with low or high expression of MGRN1 plus one (PMEL, MLANA, or TYRP1), two (PMEL + TYRP1; MLANA + PMEL; TYRP1 + MLANA), or all three (PMEL + MLANA + TYRP1) individual biomarkers and obtained plots for the OS of patients in these subgroups (Figure 6). Compared with the subsets of patients stratified according to MGRN1 expression alone, all possible combinations of MGRN1 and one or two markers (2g or 3g combinations, respectively) yielded a bigger difference in median values of OS for the high and low expression subsets of patients (Figure 6a). Moreover, the Kaplan–Meier analysis for all combinations, except MGRN1 + MLANA, confirmed higher significant differences in median survival and improved HR values between the high and low expression groups of TNM-F patients (Figure 6b,c). Furthermore, the analysis of Receiver Operating Characteristic (ROC) curves indicated better performance for 3g signatures (Figure 6d) compared with 2g analysis, with increased areas under the curves (AUC) that reached values > 0.70 for all the 3g combinations.

Finally, we assessed the prognostic performance of a 4-gene signature (4g) by comparing survival rates for two subsets of low-medium TNM patients corresponding to the simultaneous 33% upper or lower expression of MGRN1, PMEL, MLANA, and TYRP1 (4g-H³³ or 4g-L³³, respectively). The high expression group (n = 26) had a median OS roughly 4-fold shorter than the low expression group (n = 40) (Figure 7a, left), with a highly significant difference in their Kaplan–Meier survival curves (HR = 5.7, Figure 7a, middle). ROC analysis confirmed the robustness of the test (AUC > 0.80, Figure 7a, right). Importantly, we replicated these results using independent datasets available on the GEO platform as validation cohorts. After normalization of data from the GSE19234 [46] and GSE65904 [47] studies, survival analysis demonstrated that the 4g signature predicted the OS of patients (Figure 7b), confirming the result obtained with the SKCM cohort. Overall, these results proved that a 4g signature combining MGRN1 and 3 MM biomarkers identified a subset of patients with favorable (0–IIIB) TNM stage but adverse outcomes (OS comparable with the median of patients with advanced IIIC–IV TNM stage). However, because most melanomas from TNM-F patients did not show simultaneously the high or low expression of the four genes, the previous analyses considered only 20% of the TNM-F patients in TCGA and identified 10% of the low-grade TNM patients that underwent unexpected adverse outcomes. Therefore, we analyzed the performance of the 4g signature under less stringent conditions compatible with the complete TNM-F subset of TCGA patients (n = 332). To this end, we considered three groups of TNM-F patients: MM with higher or lower expression than the median for each of the four genes simultaneously (4g-H⁵⁰, n = 58 and 4g-L⁵⁰, n = 79, respectively), and the rest of TNM-F patients (mixed expression levels of genes in the signature, 4g-M⁵⁰ group, n = 195). The results are shown in Figure 7c.

Patients in the 4g-L⁵⁰ group had significantly longer OS than 4g-H⁵⁰ or 4g-M⁵⁰ patients (Figure 7c, left). Kaplan–Meier curves for 4g-L⁵⁰ and 4g-H⁵⁰ patients confirmed this difference (OS 5107 days vs. 2005 days, AUC for ROC curve 0,73), and the outcome was comparable for the 4g-M⁵⁰ (OS 2192 days) and 4g-H⁵⁰ groups (Figure 7c, middle, and right graphs). As shown by Venn diagrams, the 4g signature defined as 4g-H⁵⁰ + 4g-M⁵⁰ comprising all MM except those with expression of MGRN1, PMEL, MLANA, and TYRP1 simultaneously below the median (4g-L⁵⁰), identified 84% of TNM-F patients with OS shorter than 5 years, with a 34% false discovery rate (FDR) (4g-H⁵⁰ + 4g-M⁵⁰ patients with long OS) (Figure 7d). Therefore, this signature complemented TNM staging by identifying with high-sensitivity 0-IIIB, low-medium TNM stage patients with adverse outcomes.

2.5. Increased Genomic Stability and Infiltration of M2 Macrophages as Molecular Features Associated with Poor Prognosis in TNM-F Patients

Gene enrichment analysis of 4g-L³³ compared with 4g-H³³ TNM-F patients highlighted the differential expression of gene sets involved in inflammatory and immunological responses, as well as regulation of cell cycle progression and DNA damage/repair (Figure 8a). Importantly, the transcriptomic profile of 4g-L³³ tumors was comparable with MM of low MGRN1 expression (Figure 3c), or from low-medium TNM patients with long OS (Figure 1e), or MGRN1-KO MM cells (Figure 4a). Moreover, the transcriptional fingerprint of 4g-L³³ and 4g-H³³ melanomas on one hand (Figure 8a) and tumors from patients with long (>3000 days) or short OS (<590 days) on the other (Figure 1e) yielded similar heatmaps, where most genes associated with EMT, immune responses, cell cycle, and DNA damage contributing to enrichment with individual [FC] > 1.5 were common (Figure 8b). Thus, the transcriptomic fingerprint associated with the 4g-L³³ signature is comparable to the one associated with patients with the best OS.

Finally, we validated the inflammatory profile observed in our transcriptomic analysis. Macrophage activation and function in vivo lie between two extreme situations defined by in vitro activation with specific cytokines. M1 macrophages are induced by treating resting M0 macrophages with interferon-γ and other TH1-type cytokines to perform pro-inflammatory and anti-tumoral functions, whereas M2 macrophages stimulated by TH2 cytokines are anti-inflammatory and pro-tumoral (reviewed by [48,49]). On the other hand, MMs are strongly immunogenic tumors, in part because of their high mutation burden [3,14,15,16] that may provide an independent predictor of immune infiltration, prognosis, and response to immunotherapy [50,51,52]. Based on these data, we compared the relative infiltration by M0, M1, or M2 macrophages as well as several indexes of genomic damage in TNM-F tumors from patients with long or short OS, and 4g-L³³ or 4g-H³³ signatures (data from [4,51]). We found significantly higher infiltration of M2 tumor-associated macrophages (TAMs) with pro-tumoral and anti-inflammatory activity in tumors with 4g-H³³ signature or short OS compared with tumors with 4g-L³³ signature or long OS, respectively (Figure 8c). Moreover, 4g-H³³ or shorter OS tumors displayed significantly lower genomic scar indexes than tumors with 4g-L³³ signatures or longer OS (Figure 8d).

Overall, these data pointed to higher genomic stability and infiltration of M2 TAMs as distinctive features of tumors from TNM-F patients undergoing an unexpectedly aggressive course of the disease. They also indicated that the MGRN1-, PMEL-, MLANA-, and TYRP1-based 4g signature accurately identified this subset of patients.

3. Discussion

In situ, SKCM diagnosed at early TNM stages can be cured by surgically removing the lesion. Unfortunately, owing to its high metastatic potential, MM is diagnosed in disseminated stages not amenable to exeresis more often than other cancers [5]. Despite recent advances in targeted therapies and immunological treatments, the prognostic of these metastatic MMs remains obscure due to high rates of non-responders and extremely frequent relapse in responders. The lack of reliable prognostic biomarkers further complicates the clinical handling of MM patients [18].

Currently, prognostic assessment and treatment decisions of MM are based primarily on the TNM staging system of the AJCC [21]. TNM staging considers tumor size and ulceration, the presence of MM cells in regional lymph nodes and in-transit, satellite, microsatellite, and/or distant metastases, as well as LDH levels. TNM stages range from 0 to IV, where IIIC, IIID, and IV are the stages with the worst five-year survival rates. Conversely, low-medium TNM stages (0–IIIB) are considered predictive of favorable outcomes, and adjuvant therapy is not usually indicated in stage IIIB to avoid overtreatment with poor risk-benefit ratios [22,23]. Nevertheless, it has been reported that around 15% of MM deaths are caused by metastases of thin lesions classified as low-risk according to their TNM [24]. Indeed, TNM staging does not allow for distinguishing indolent tumors with less aggressive phenotypes from aggressive tumors diagnosed early after the onset of the disease. The inability to discriminate less aggressive MM against early diagnosed aggressive tumors is a quantitatively important limitation since approximately 75–80% of MMs are diagnosed at primary, localized (stages 0–II), or regional (stages III) stages. In agreement with these data, our analysis of the SKCM cohort indicated that around 25% of low-medium TNM stage patients underwent an aggressive disease with OS shorter than the median of high TNM-grade patients. Therefore, substantial uncertainty remains regarding the benefit of adjuvant treatment in the initial phases of the disease, which is in fact not supported by many national administrations for stage IIB and even IIIB. Moreover, the choice of the optimal treatment option in stages IIB and IIC is a difficult matter, as not only relapse rates but also other clinical factors such as age and comorbidity must be considered. Accordingly, tools to inform clinical decisions in these ambiguous situations are needed to minimize the costs and morbidity associated with under- or overtreatment.

Efforts have been made to identify predictive biomarkers complementing the TNM staging in MM. Studies looking for prognostic genomic features of many cancer types, including MM, identified groups of genes predicting patient survival [32,53,54,55,56,57,58,59], sometimes with better accuracy than TNM staging [46]. The genomic features potentially useful for diagnostic, predictive, and therapy selection purposes in MM have been recently reviewed [17,60]. However, these studies have not yet been translated into clinical practice, due to factors including the limited amount of material available for analysis after mandatory histopathological examination, technical requirements, and frequently complex interpretation of results. Therefore, independent predictive biomarkers are still required.

Given the well-known heterogeneity of MM, we anticipated that no individual biomarker would adequately capture the complexity of the tumors in vivo. Therefore, we looked for as simple as possible combinations of markers. On the other hand, although the analysis of protein expression by IHC is a widely accepted diagnostic method, it presents several drawbacks compared with gene expression determinations, particularly when the estimation of the level of the marker, or rather, its presence or absence, is required for predictive purposes (reviewed in [30]). These drawbacks include difficult normalization, observer variability, incomplete reproducibility, and sample heterogeneity. Conversely, genomic analyses are sensitive and easier to normalize and provide information on both the sequence and the level of expression of targets. Moreover, a recent report showed that mRNA expression data obtained via RNA-seq and protein levels measured by reverse-phase protein assays yielded similar results and led to comparable survival profiles [20]. Therefore, we used gene expression data in our attempt to identify a combination of a few biomarkers complementing the predictive potential of TNM staging. Moreover, we prioritized potential targets already employed in IHC to allow for the use of the panel at the gene or protein expression levels.

We first used extensive genomic, transcriptomic, and clinical data available from pan-cancer studies as The Cancer Tumor Atlas project [40]. These data have been used in successful pan-cancer studies of prognostic markers [20]. Analysis of the GDC-SKCM cohort indicated that neither the mutational status nor the level of expression of the major validated MM drivers BRAF, NRAS, or NF1 provided useful prognostic information, in agreement with a recent report showing that, counterintuitively, the mutation or overexpression of verified oncogenes is rarely associated with poor prognosis in human cancers [20]. Accordingly, we focused on genes involved in biological functions likely related to the phenotype of MM cells, such as proliferation, differentiation, motility, and genomic stability. To foster the possible clinical application of our results, we performed a targeted search aiming at a set of candidates as limited as compatible with sensitive and accurate risk estimation, ideally repurposing markers already in use. Moreover, to increase the chances of identifying genes yielding additive, rather than redundant information, we targeted genes with roles in different cellular processes and whose expression levels in MM do not show a significant correlation, likely because of independent regulatory mechanisms. We found that a 4-gene combination consisting of MGRN1, PMEL, MLANA, and TYRP1 identified low-medium-grade TNM patients with unexpectedly adverse outcomes with a high 84% sensitivity and reasonable 34% FDR, yielding an ROC curve indicative of accurate performance (AUC 0.73).

The molecular bases of the association of MGRN1, PMEL, MLANA, and TYRP1 expression within the tumor and patient survival remain speculative. PMEL and MLANA encode melanocyte-specific proteins that contribute to melanosome architecture [61] and are widely used as diagnostic MM biomarkers [18]. Since PMEL and MLANA act as differentiation antigens recognized in most metastatic MM [26,62], the finding that their lower expression was significantly associated with longer survival may appear counterintuitive. However, PMEL and MLANA expression in melanocytes is upregulated by the MITF transcription factor, and their low expression likely indicates low MITF transcriptional activity, which has been associated with senescence and, eventually, cell death (reviewed in [63]). It could be argued that the expression of MITF itself might provide a better predictor of outcome than PMEL or MLANA. However, the widely accepted rheostat model for the biological role of MITF posits that MITF activity, rather than abundance, determines the regulation of MITF targets and the resulting phenotypes of melanocytes and MM cells [64,65]. Since MITF activity is tightly regulated at the post-transcriptional level by a variety of posttranslational modifications, the availability of cofactors [63], and maybe also by epigenetic modification of the target genes promoters [66], the level of expression of MITF targets might provide a better indication of MITF activity than the expression of MITF. On the other hand, TYRP1 encodes for a melanocyte-specific melanogenic enzyme [67]. Although its best-known biological function in pigmentation does not hint at a connection with patient survival, a correlation of high TYRP1 mRNA expression with adverse clinical outcomes of MM patients has already been noticed by others [42,68], and it was shown that TYRP1 mRNA sequesters miR-16, thereby repressing its ability to downregulate the expression of target mRNAs involved in MM cell proliferation and tumor growth [69].

Finally, knockout or knockdown of MGRN1 has been shown to increase differentiation and cell adhesion to a variety of extracellular matrices and to impair the proliferation of melanocytes and MM cells [36,38,39]. Here, we showed that the knockout of MGRN1 in human MM cells led to genomic instability, as demonstrated by the increased burden of DNA breaks. Moreover, transcriptomic analysis of MGRN1-KO cells indicated the downregulation of genes involved in cell cycle progression and upregulation of gene sets of inflammatory and immunologic responses. Importantly, an analysis of the secretome of MGRN1-KO cells confirmed strong downregulation of the master anti-inflammatory cytokine IL10. Thus, MGRN1 appeared important for genomic stability and its downregulation was sufficient to trigger a cell-autonomous transcriptional program strongly influencing the immunologic behavior of MM cells. Interestingly, a recent study identified MGRN1 in an 8-gene signature predicting the response to immunotherapy in MM, further supporting its role in modulating the immune microenvironment of the tumor [70]. Also of note, the GSEA of the SKCM cohort indicated a similar transcriptional landscape in tumors with low expression of MGRN1, suggesting comparable transcriptional effects of MGRN1 in cultured MM cells and the tumors. More importantly, MM with low expression of the 4g signature displayed a consistent transcriptomic landscape, with upregulation of gene sets related to inflammatory and interferon-γ responses and downregulation of cell cycle and DNA repair genes. These 4g-L tumors also displayed signs of genomic instability such as higher mutational burden, ploidy, and wGII, as well as decreased presence of pro-tumoral M2 macrophages. Accordingly, it is tempting to speculate that the association of the 4g signature with patient outcome could be accounted for primarily by a significant genomic scar in 4g-L tumors, associated with increased inflammatory and interferon-γ responses and downregulation of genes involved in cell cycle progression. In turn, these phenotypes could result from decreased MGRN1 expression, high miR-16 activity due to low expression of TYRP1 mRNA, and low MITF activity revealed by low PMEL and MLANA expression. In this scenario, most genes within the signature would hold a different mechanism of association with outcome, thus affording maximal accuracy to the prediction based on their combined scores.

One limitation of our study is that it mostly relied on the TCGA MM cohort, which may be somewhat heterogeneous because it contains data from patients treated with different regimes. To minimize this limitation, we used data from two other studies as a validation cohort [46,47], with results that supported the predictive potential of the 4g biomarkers. Other limitations are the lack of a retrospective validation study with actual surgical specimens from MM patients, and the fact that transcriptomic analysis of MGRN1-KO cells was performed using a triple WT cell line, i.e., belonging to one out of four molecular subtypes. In any case, we believe that the sensitive detection of patients with low-medium-grade TNM and adverse outcomes reported here underlines the potential of the proposed 4-biomarker combination and warrants future studies of its clinical application. It will also be interesting to see whether the 4g biomarkers are indicative of a response to immunotherapy, given the identification of mutational burden as an independent predictor of immunotherapy response across different types of cancer [52,71,72].

4. Materials and Methods

4.1. Cell Cultures

The HBL human MM cell line (triple WT molecular subtype) was obtained in 1979 from a Caucasian female donor at the Laboratory of Oncology and Experimental Surgery, LOCE-Institut J. Bordet, Université Libre de Bruxelles (ULB), Belgium. The cell line was a kind gift from Prof. G. Ghanem. Cells were grown in DMEM-GlutaMAX enriched with 10% fetal bovine serum (FBS), 100 U/mL penicillin, and 100 μg/mL streptomycin sulfate, using cell culture reagents from Gibco (Gaithersburg, MD, USA).

4.2. Generation of MGRN1-KO Cells

Permanent knockout of MGRN1 was performed as previously described using target sequences for CRISPR-RNA from Dharmacon [38]. Efficiencies and potential off-targets were estimated with Breaking-Cas (http://bioinfogp.cnb.csic.es/tools/breakingcas). Cells were transfected with 0.5 μg of CRISPR plasmid (pSpCas9(sgRNA2, sgRNA3 or sgRNA4)) and 1 mg/mL Lipofectamine 2000, using the empty plasmid (pSpCas9) for control cells (CTR). After incubation for 72 h at 37 °C, puromycin-resistant clones were selected. The confirmation and selection of KO clones were performed by automated sequencing after PCR amplification of a 500 pb fragment on exon 1 using genomic DNA as a template, and Western blot for MGRN1 detection in all selected clones.

4.3. RNAseq Analysis

Total RNA from MGRN1-KO (clone 4.9) and control cells (n = 4 for each condition) were isolated and purified with RNeasy^® Mini kit (QIAGEN, Germantown, MD, USA) as per instructions. A total of 400 ng of total RNA samples (concentration ≥ 20 ng/µL) were analyzed by Novogene Co., Ltd. (Cambridge, UK). This company performed quality control evaluations (all samples showed an RNA integrity number, RIN = 10), as well as cDNA synthesis, library construction (poly A enrichment), and transcriptomic analysis (Illumina Sequencing PE150, Illumina Inc., San Diego, CA, USA) yielding 15 G of raw data per sample.

4.4. Differential Expression and Enrichment Analysis

Raw RNAseq data counts from TCGA and our RNAseq studies were processed and normalized with RStudio (version 4.3.1), using the DESeq2 R package [73]. The GSEA (version 4.3.2; https://www.gsea-msigdb.org/gsea/index.jsp) was used for functional enrichment analysis, collapsing the pre-ranked gene datasets into the Hallmark gene sets collection and the Ensembl Gene ID chip platform (Molecular Signature Database human collection, MSigDB version 2023.2; [74,75]), with classic as the enrichment statistics method. For the construction of pre-ranked datasets, we only used genes with p-value (pVal) and adjusted p-value (pAdj) ≤ 0.05, ordered by fold-change. Pre-ranked datasets will be provided by the corresponding author upon reasonable request. After collapsing native features from the pre-ranked datasets into gene symbols, enough genes were identified in all cases (from 5000 to ≥16,500). For GSEA results, only gene sets with a normalized enrichment score (NES) lower than −2.0 or higher than 2.0, p-value ≤ 0.05, and false discovery rate (FDR) ≤ 0.25 were considered. Heatmaps of custom gene sets were generated using the Pretty Heatmaps R package. All links used are provided in Table S1.

4.5. Data Processing

Unless otherwise indicated, all data were obtained from the GDC-TCGA-SKCM cohort, before September 2024, using the UCSC Xena browser (https://xenabrowser.net/) and from GEPIA (http://gepia2.cancer-pku.cn/) [76]. Data were processed as described below and will be provided under reasonable request. For optimal operation during differential expression analysis using DESeq2, the values expressed as “Log₂(count+1)” were transformed into “counts” using RStudio. GEO data (https://www.ncbi.nlm.nih.gov/geo/) were used for validation studies [77]. Transcriptomic raw data from GEO: GSE19234 were handled and normalized with R packages for Affymetrix [78]. In the case of GEO: GSE65904, quantile normalization was used. The genomic scar features analyzed were the weighted genome instability index (wGII, defined as the fraction of the genome with aberrant copy number, weighed on a per chromosome basis) and ploidy (defined as the number of sets of chromosomes in the cells). Values of these features in melanomas from patients with known OS were obtained from Marquard and coworkers [4]. Tumor mutational burden (TMB) was calculated from GDC-TCGA-SKCM data, considering all indels and SNPs in each melanoma sample compared with the reference genome. Finally, immunological data (M0, M1, and M2 macrophage infiltration) were from [51]. Only data from patients with registered survival time were used, and the corresponding resources, links, and identifiers are specified in Table S1.

4.6. Immunoblotting

Control and MGRN1-KO cells were lysed in RIPA solubilization buffer (50 mM Tris-HCl pH 8, 150 mM NaCl, 0.5% sodium deoxycholate, 0.1% SDS, 1% Triton X-100) supplemented with 0.1 mM phenylmethylsulfonyl fluoride (PMSF), 10 mM iodoacetamide (IAA), 10 mM N-ethylmaleimide (NEM), and phosphatase inhibitors (200 mM imidazole, 100 mM NaF, 100 mM Na₃VO₄, and 1 M β-Glycerol phosphate). Lysates were collected, gently shacked, and centrifuged for 25 min at 13,500 RPM at 4 °C. Supernatants were prepared with 4X Loading Buffer (250 mM Tris pH 6.8, 8% SDS, 20% glycerol, 0.08% bromophenol blue and 3.2 M β-mercaptoethanol). Electrophoresis was performed on polyacrylamide gels (PAGE) in the presence of SDS (sodium dodecyl sulfate). Resolved proteins were transferred onto a PVDF membrane (Immobilon-P). Anti-MGRN1 (rabbit polyclonal, 1:4000 dilution, #11285 Proteintech) and GAPDH (rabbit polyclonal, 1:5000 dilution, sc-25778) were used for overnight incubation at room temperature, followed by incubation with secondary antibody for 1 h. The ECL Plus detection System (GE Healthcare Technologies, Inc., Chicago, IL, USA) was used for detection.

4.7. Analysis of Secreted Cytokines in Cell Culture Media

The Human Cytokine Antibody Array C5 from the RayBiotech kit was used for the analysis of secreted cytokines, according to the manufacturer’s instructions and using the provided reagents. Briefly, culture media from MGRN1-KO or control cells grown in serum-free medium were collected and 10-fold concentrated using Amicon Ultra filters. Membranes were gently blocked, washed, and incubated overnight with a biotinylated antibody cocktail, followed by HRP-streptavidin conjugate and extensive washing. Finally, images were acquired and analyzed with a FUSION Solo S imaging system (Vilber) and ImajeJ software, version 1.54f (National Institutes of Health, Bethesda, MA, USA), respectively. Quantitative expression of IL10 in the same samples was validated via multiplex analysis (ProcartaPlex™ Human Cytokine Panel 1B, Thermofisher, Waltham, MA, USA), as per instructions.

4.8. Immunofluorescence, Confocal Microscopy, and Image Quantification

Cells were seeded in 24-well plates containing sterile coverslips and grown to 60% confluency. Cells were fixed in 4% formaldehyde, with 0.5% Triton-X 100 (v/v), and blocked with 5% BSA in PBS (1 h, room temperature). For γH2AX staining, cells were labeled with a monoclonal antibody (anti γH2AX, ab2893 Abcam, Cambridge, UK, 1:1000), followed by an Alexa 488-conjugated secondary antibody. After immunostaining, samples were mounted with a medium from Dako (Glostrup, Denmark) and examined with a Leica laser scanning confocal microscope with a 63× objective lens and Qwin software, version 3.1.0, from Leica (Leica Microsystems GmbH, Wetzlar, Germany). Single plane images corresponding to Z positions of the maximal DAPI signal were acquired. Laser power, gain, and offset were established to minimize background, sample saturation, and photobleaching. Nuclear signals were quantified by calculating the pixel intensity in single-cell nuclei relative to the nucleus area. This analysis was performed using the same Qwin software as above. At least 200 randomly selected cells per condition were quantified.

4.9. Comet Assays

Alkaline Comet assays were performed according to the manufacturer’s protocol (Trevigen, Gaithersburg, MD, USA). Briefly, cells were mixed with low melting point agarose (at 37 °C) at a ratio of 1:10 (v/v) and then pipetted onto microscope slides. After adhesion at 4 °C for 30 min in the dark, slides were immersed in precooled lysis buffer overnight at 4 °C. Slides were next immersed in electrophoresis solution (200 mM NaOH, 1 mM EDTA, pH > 13) at room temperature for 20 min in the dark, placed into an electrophoresis chamber and electrophoresed in the same buffer at 25 V (adjusting the current to 300 mA) for 30–45 min. After electrophoresis, slides were immersed twice in distilled water and 70% ethanol, and dried (37 °C, 30 min) to bring all the cells to a single plane. Finally, DNA was stained with SYBR™ Green I Nucleic Acid Gel Stain, and images were taken in a Nikon Eclipse TS2 microscope with a 10× lens. Quantitative analysis of the tail moments was performed using CASPLAB, version 1.2 software. At least 100 randomly selected comets were analyzed per sample.

4.10. Quantification and Statistical Analysis

All analyses were conducted using GraphPad Prism, versions 8 or 9. A normality test was performed for two group comparisons, and an unpaired two-tailed Student’s t-test was applied in case of non-statistical variance differences among groups. For non-normal distribution, the Mann–Whitney test was performed. When comparing more than two groups, a normality test was performed and one-way Anova with Dunnett’s post-test was applied when the variance among groups was not statistically different. When the variance was different, a Kruskal–Wallis test with Dunn’s post-test was performed. Kaplan–Meier survival curves for defined groups were compared using a Log-rank (Mantel–Cox) test, obtaining Hazard Ratios (HRs) with the Mantel–Haenszel method with a 95% confidence interval. Correlation analysis was performed using the Pearson test. For Receiver Operating Characteristic curves (ROC curves), the Wilson/Brown method was used with a 95% confidence interval. Unless otherwise indicated, results were expressed as median and are indicated in each figure. p-values of equal or lower than 0.05 were considered statistically significant: * indicates ≤ 0.05; ** ≤ 0.01; *** ≤ 0.001; and **** ≤ 0.0001.

5. Conclusions

We confirmed that the TNM staging system of the AJCC, universally used for stratification and prognostic assessment in MM, predicts the outcomes of patients with low-medium accuracy, particularly in intermediate IIB–IIIB stages. Thus, reliable MM prognostic biomarkers are required to improve risk prediction and inform the choice of therapeutic options, avoiding the substantial morbidity associated with under- or overtreatment. We found that neither mutation nor overexpression of TP53 or the major driver genes in MM (BRAF, NRAS, NF1) provided significant prognostic information. Conversely, the level of expression of a small subset of four genes (MGRN1, PMEL, MLANA, and TYRP1) accurately predicted the outcome of patients with low-medium-grade MM and identified with high sensitivity the subgroup of these patients undergoing an aggressive disease. Moreover, the expression of this 4-gene panel was associated with a specific transcriptomic landscape indicative of dysregulation of gene sets involved in immunological responses, and genomic stability. Notably, PMEL and MLANA are currently used as diagnostic biomarkers, and we confirmed that the knockout of MGRN1 in MM cells increased genomic damage and led to a comparable transcriptomic profile. Therefore, our results may foster the use of the MGRN1-PMEL-MLANA-TYRP1 panel to complement the prognostic information of TNM staging.