Four Immune Modulating Genes in Primary Melanoma That Predict Metastatic Potential

Abstract

Introduction:

Histologic characteristics cannot adequately predict which patients are at risk of developing metastatic disease after excision of primary cutaneous melanoma. The aim of this study was to identify immunomodulatory genes in primary tumors associated with development of distant metastases.

Materials and methods:

Thirty-seven patients with primary melanoma underwent surgical excision. RNA was extracted from the primary tumor specimens. cDNA was synthesized and used with Human Gene Expression microarray. Differential expression of 74 immunomodulatory genes was compared between patients who developed distant metastases and those who did not.

Results:

Six of 37 patients developed distant metastases during the time of the study. Differential expression of microarray data showed upregulation of four immunomodulatory genes in this group. These four genes—c-CBL, CD276, CXCL1, and CXCL2—were all significantly overexpressed in the metastatic group with differential expression fold change of 1.15 (P = 0.01), 1.16 (P = 0.04), 2.51 (P < 0.001), and 1.68 (P < 0.02), respectively. CXCL1 had particularly high predictive value with an area under the curve of 0.80. Multivariate analysis showed only expression of CXCL1 (P = 0.01) remains predictive of distant metastases in melanoma patients. This result was confirmed using quantitative real-time polymerase chain reaction.

Conclusions:

CXCL1, CXCL2, c-CBL, and CD276 are immunomodulatory genes present in primary melanoma that are strongly associated with development of metastatic disease. Identification of their presence, particularly CXCL1, in the primary tumor could be used as a predictor of future risk of metastatic disease and thereby to identify patients who might benefit early from immunotherapy.

Introduction

In the Annual Report to the Nation on the Status of Cancer 2020, melanoma incidence was reported to have increased for both men and women in the United States. It is currently the fifth most common cancer for men and sixth for women.¹ Fortunately, the majority of new cases are diagnosed at early stages, and the 5-year survival rates for American Joint Committee on Cancer (AJCC) stage I and II vary from 99% to 82%.² However, outcomes are markedly worse for patients with metastatic melanoma as the 10-year survival, which used to be 10%-15% before the era of immunotherapy, is still low but improved in the context of modern therapy at 34%-52%.^2–4 Current methods of staging primary melanoma are based on clinicopathologic features of the primary tumor and sentinel node status.⁵ Despite this, we cannot accurately predict which patients will develop distant metastases. Since disease management and outcome are so difficult by the time metastases are identified, there is a pressing need to identify the patients at risk of metastasis before distant disease develops.

Immune checkpoint inhibitors and targeted therapy drugs have become the cornerstone of therapeutic management of metastatic melanoma.⁶ Given the prominence of immunomodulation, we proposed examining immune modulating genes from the primary tumor as a method to predict patient outcome. We hypothesized that genetic analysis of the immunomodulatory genes in primary wide excision specimens would identify patterns of expression that could distinguish patients at risk of developing distant metastases.

Materials and Methods

Study population

Patients diagnosed with primary cutaneous melanoma and no distant metastases presented to a single surgeon at a tertiary referral center between 2004 and 2007 were enrolled in this study (Fig. 1). All patients had biopsy-proven melanoma prior to resection. We selected 37 consecutive patients who had residual, visible disease after biopsy and prior to wide local excision in order to procure a sufficient amount of tissue required for running the analyses. The patients were followed prospectively and similarly regardless of stage for a median of 38 mo (range, 1–144 mo) during 2004–2018. All patients were planned for indefinite follow-up per the surgeon’s practice. Thirty-two out of 37 patients attended their appointments for at least 2 y, which is the window during which most recurrences are expected. For the five patients with less than 2-year follow-up, three had distant metastasis and all passed within 2 y of surgery, one with no metastasis died of non-melanoma causes, and one was lost to follow-up. Six patients out of the 37 (16% of the cohort) developed distant metastases during the study period. This established two groups for comparison: those who developed distant metastases (n = 6) and those who did not (n = 31). Patient demographics and tumor characteristics are summarized in Table 1. All specimens were collected under institutional review board approval with patients’ consent.

Fig. 1 –

Study design.

Table 1 –

Clinical and pathologic characteristics.

Variables	Patients (MET)	Patients (No-MET)	P value
Total	6	31
Age			0.15
Under 60	0 (0%)	12 (39%)
60 and above	6 (100%)	19 (61%)
Gender, no. (%)			0.06
Male	6 (100%)	16 (52%)
Female	0 (0%)	15 (48%)
Site of specimen, no. (%)			1.00
Head and Neck	2 (33%)	9 (29%)
Trunk/Extremities	4 (67%)	22 (71%)
Thickness of primary tumor (mm), median[range]	2.1 [0.15–5]	1.05 [0.23–10]	0.66
Follow-up (mo), median[range]	38 [4–128]	26 [1–144]	0.56
Lymph node metastasis, no.			0.18
Yes	2 (33%)	3 (10%)
No	4 (67%)	28 (90%)
Mitosis, no. (%)			0.33
Low	2 (33%)	21 (68%)
High	3 (50%)	10 (32%)
Unknown	1 (17%)	0 (0%)
Ulceration, no. (%)			0.05
Yes	3 (50%)	4 (13%)
No	2 (33%)	26 (84%)
Unknown	1 (17%)	1 (3%)
Regression, no. (%)			0.56
Yes	0 (0%)	7 (23%)
No	5 (83%)	23 (74%)
Unknown	1 (17%)	1 (3%)

Data are expressed as number (%).

Low mitosis defined as 0/mm²; high mitosis equal to or greater than 1/mm²

RNA extraction and microarray assay

Tissue samples from wide excision of the primary tumor were collected, placed in RNA later (Qiagen, Hilden, Germany), and immediately preserved in liquid nitrogen. The specimens were securely de-identified through a centralized database. All samples were confirmed to contain >95% tumor cells. RNA was extracted from the frozen tissue samples using the Qiagen RNeasy Mini Kit. The RNA concentration was measured by NanoDrop Spectrophotometer (NanoDrop Products, Wilmington, DE) and the quality determined by the Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA). All RNA samples were processed for microarray assay according to the manufacturer’s protocol. Briefly, RNA was labeled using the Quick AMP Labeling Kit (Agilent Technologies), purified on RNeasy columns (Qiagen), and hybridized on Agilent Human 44K 60-mer oligonucleotide microarray chips (Agilent Technologies, Santa Clara, CA). Microarray chips were incubated at 60°C for 16–17 h, washed, covered with ozone barrier, and then run through the Agilent Scanner (G2565CA). Agilent Feature Extraction Software (version 9) was used to quantify the intensity of fluorescent images.

Quantitative real-time polymerase chain reaction

To validate the four genes, we performed quantitative real-time polymerase chain reaction (qRT-PCR) on the same primary tumor samples. Reverse transcription of 1 μg of RNA was performed using iScript cDNA synthesis (Bio-Rad) in a final volume of 40 μL, following the manufacturer’s instructions. Each qRT-PCR was done using 1 μL of cDNA, 10 μL of 2X PowerUp SYBR Green Master Mix (Thermo Fisher Scientific, Waltham, Massachusetts), and 500 nM of primers in a final volume of 20 μL. Samples were assessed in triplicates on a QuantStudio 6 PCR System (Applied Biosystems, Waltham, Massachusetts) with the following protocol: 95C for 3 min, 40 cycles at 95°C for 15 s, and 60°C for 40 s. The primer sets, described in Table 2S (Supplemental Data), were synthesized by Integrated DNA Technologies. Quantification of the expression was assessed using the ΔΔCt method with β-2-microglobulin (B2M) and succinate dehydrogenase complex flavoprotein subunit A (SDHA) expression as reference genes. Expression levels of the four genes were compared using the paired Student’s t-test with GraphPad Prism 8. A P value <0.05 was considered statistically significant: *P < 0.05.

Statistical analyses

Patient demographics and tumor characteristics were compared between the two groups using Student’s t-test, Fisher’s exact test, and Mann-Whitney U test set at a significance level of 0.05. A list of 79 immunomodulatory genes (Supplemental Data 1), identified by extensive literature review for genes involved in cancer metastases mechanism, was preselected for analysis. Search terms were entered into PubMed that included “melanoma” and “immunomodulation” or “immunotherapy” with full text articles available. A list of all genes mentioned in these articles was generated, and this list was cross-checked with the list of genes available for study in the microarray which resulted in a concordant list of 79 genes set for microarray processing before any analytical comparisons. Due to unavailable or incomplete data in the microarray set, five genes from the preliminary list were excluded. Microarray gene expression was processed with quantile normalization and log2 transformation. The differential expression of the remaining 74 genes was analyzed and compared between the metastatic and nonmetastatic groups using Student’s t-test. The significance level of differential expression was set to a false discovery rate of 0.05.

Univariate logistic regression model was performed to determine the power of each demographic factor, tumor characteristic, and expression of the four genes to predict distant metastases. Multivariable logistic regression with stepAIC was performed on the significant variables from the univariate analysis to identify significant predictors of distant metastases. The Kaplan-Meier method was used to estimate the overall survival between groups using 90% quantile separation of CXCL1 expression levels. Log-rank testing was used to compare the difference in survival. Area under the curve (AUC), sensitivity, specificity, and positive and negative predictive values were calculated for CXCL1 as a predictive biomarker.

Results

Clinical and pathologic characteristics

There was no statistically significant difference between the patients who developed distant metastases and those who did not in consideration of age, gender, site of primary tumor, Breslow thickness, sentinel lymph node metastases status, mitotic index (low = 0, high ≥1), regression status, or length of follow-up (Table 1). Seventy percent (26/37) of patients from both groups were ≥60 y old with primaries of the trunk and extremities. Ulceration was the only characteristic significantly different between groups, with the group that developed distant metastases demonstrating a higher rate of ulceration (50% versus 13%, P = 0.05). Among the six patients who developed distant metastases, two had metastases to the lungs, three had distant subcutaneous metastases, and one had diffuse metastatic disease. Distant metastases were diagnosed within 2 y of primary diagnosis for half of the patients and between 4 and 5 y after initial diagnosis for the other half.

Differential expression analysis

Differential expression analysis of the microarray data of the 74 immunomodulatory genes showed significant upregulation of CXCL1, CXCL2, CD276, and c-CBL in primary tumor specimens from patients who eventually developed distant metastases. The differential expression fold exchange (DEFC) of these four genes was 2.51 (P < 0.001), 1.68 (P < 0.02), 1.15 (P = 0.04), and 1.16 (P = 0.01), respectively (Fig. 2 A–D). All four genes have roles in the immune cascade.

Fig. 2 –

Differential expression fold change (DEFC) of (A) CXCL1, (B) CXCL2, (C) CBL, and (D) CD276 in the metastatic (MET) versus nonmetastatic (no MET) groups.

Prediction of distant metastases

Univariate analysis demonstrated that ulceration (P = 0.05) and overexpression of CXCL1 (P = 0.01), CXCL2 (P = 0.04), CD276 (P = 0.05), and c-CBL (P = 0.03) are predictive of distant metastases development (Table 2). However, after multivariable analysis with StepAIC, only CXCL1 remained significant as a predictor of distant metastases (P = 0.01; Table 2). As a predictive biomarker, CXCL1 demonstrated an AUC of 0.80. The sensitivity and specificity of CXCL1 to predict distant metastases were 0.67 and 0.97, respectively. The positive predictive value was 0.80 and the negative predictive value was 0.94 (Table 3).

Table 2 –

Factors predictive of developing metastases.

Variable	Univariate analysis P value	Multivariate analysis P value
Age	0.55	NS
Gender	0.99	NS
Thickness	0.65	NS
Ulceration	0.05	NS
Positive LN	0.15	NS
Primary site	0.83	NS
CXCL1	0.01	0.01
CXCL2	0.04	NS
CD276	0.05	NS
c-CBL	0.03	NS

Each variable was compared between the metastatic and nonmetastatic group.

Age defined as under 60 versus 60 and above. Thickness in millimeters. Primary site as Head and Neck versus trunk/extremities.

LN = lymph node; NS = not significant.

Table 3 –

CXCL1 and CXCL2 as predictors of distant metastases.

Biomarker	AUC	Sensitivity	Specificity	Accuracy	Positive predictive value (PPV)	Negative predictive value (NPV)
CXCL1	0.80	0.67	0.97	0.92	0.8	0.94
CXCL2	0.67	0	0.97	0.81	0	0.83

AUC = area under the curve.

Overall survival

Overall survival based on CXCL1 expression was calculated comparing those with high CXCL1 expression to those with low expression. The patients with high CXCL1 expression had 5-year survival of 50% compared to 97% for the patients with low expression, though this did not reach statistical significance (P = 0.1, Fig. 3).

Fig. 3 –

Overall survival based on CXCL1 expression.

Validation of the four genes

To validate our microarray results, we performed qRT-PCR on the RNA extracted from the same primary tumors. We assessed the expression level of the CXCL1, CXCL2, c-CBL, and CD276 (Fig. 4 A-D) using the ΔΔCT method with B2M as reference. These results showed that only CXCL1 is significantly upregulated in patients who will develop distant metastases (P = 0.0069). This confirmed the microarray results. Similar results were obtained with SDHA as reference (Fig. 1S) and only showed CXCL1 expression significantly upregulated in patients with distant metastases (P-value = 0.02).

Fig. 4 –

Comparison of the expression level of the four genes: (A) CXCL1, (B) CXCL2, (C) CBL, and (D) CD276, between MET and no-MET groups, obtained by quantitative real-time polymerase chain reaction.

Discussion

This study shows that there is differential expression of immunomodulatory genes in the primary melanoma tumors of patients who develop distant metastases. Four genes were significantly overexpressed in the primary specimens of those who developed distant metastases: CXCL1, CXCL2, CD276, and c-CBL. These genes have various roles in cell signaling pathways and have been described previously as important mediators in cancer, including melanoma. Of these four genes, CXCL1 showed 2.51 times greater expression in the primary specimens of the group that developed distant metastases (P < 0.001). Furthermore, this overexpression was confirmed by qRT-PCR. CXCL1 was the most robust factor in multivariable analysis at predicting metastases (P = 0.01) and demonstrated an excellent AUC at 0.80. With high specificity (0.97), positive predictive value (0.8), and negative predictive value (0.94), CXCL1 is a good potential independent biomarker in primary melanoma.

There is a growing body of research outlining the importance of these immunomodulatory genes and their oncologic influence. This study is consistent with other findings, and here we review published research focused on each of the four genes that emerged relevant, with particular emphasis on CXCL1. CXCL1 and CXCL2 are genes on chromosome four that encode cytokines involved in growth and inflammation.⁷ The chemokines they encode are secreted by macrophages for neutrophil recruitment in the presence of necrotic cells.^8,9 Both exert their signal through the same receptor, CXCR2. Studies have shown that high expression of these genes is associated with poor prognosis in a number of cancers, such as in colorectal cancer.¹⁰ A meta-analysis of 17 studies including 2265 patients diagnosed with 10 types of cancers demonstrated that high CXCL1 expression correlated with advanced TNM stage and lymph node metastasis.¹¹ Conventional wisdom has long taught that CXCL1 may promote metastasis in melanoma.¹² A series of studies by Richmond and others in the 1980s established that CXCL1, also known as melanoma growth stimulatory activity, increased growth rates when added to human melanoma cell lines.^13–15 In another contemporary study, anti-CXCL1 antibodies successfully reversed this effect and thereby slowed melanoma growth.¹⁶ Similar results have been replicated in mouse models.¹⁷ Tumorigenic effects of CXCL1 in melanoma have been attributed to both autocrine and paracrine signaling pathways.¹⁸ Recently, research has begun to define the potential role CXCL1 could play in tailoring melanoma treatment. Thurneysen et al. found that CXCL1 and CXCL2 were upregulated in the melanoma patients who responded to pazopanib treatment.⁹ A study by Taube et al., from Johns Hopkins, showed that PDL1+ melanoma had overexpression of several genes including CXCL1.¹⁹ An international collaboration led by Young et al., found that CXCR2 was upregulated in tumors of patients receiving mitogen activated protein kinase (MAPK) inhibitors.²⁰ CXCR2 antagonists have been used in animal and human trials, including the CXCR2 inhibitor SB225002 in the study by Bento et al., which resulted in significant loss of IL-1B protection from BRAF and MAPK inhibition.^20,21 Another study by Kemp et al. demonstrated that a dual CXCR1/2 small-molecule inhibitor, ladarixin, successfully inhibited motility and cell cycle progression and promoted apoptosis in human melanoma cell lines inoculated in mice.²² With the therapeutic potential of CXCR2 antagonism, tumors with CXCL1/CXCL2 overexpression would not only offer prognostic information but also an opportunity for therapeutic intervention. Furthermore, CXCR2 antagonism could work synergistically to enhance the efficacy of existing BRAF/MAPK inhibitors, which often lose their efficacy over time after tumor mutagenesis.^19,20,23

It must be noted that not all studies have come to the same conclusions. While our findings are in line with the vast majority of existing data and support the predominant theory that CXCL1 promotes growth and metastasis in melanoma, exceptions exist that claim the opposite.^24,25 This discordance stresses the importance of continuing research to clarify and confirm the details of melanoma biology and behavior and CXCL1’s role therein.

CD276, which showed higher expression in the group with distant metastases (DEFC = 1.16; P = 0.04), belongs to the B7/CD28 family of immunoregulatory proteins implicated in cancer progression and metastasis.^26,27 It is a cell surface transmembrane glycoprotein that regulates T-cell response, and it appears to have a pro-oncogenic role in cancer.^27–29 In a study by Wang et al., CD276 knockdown reduced melanoma migration and metastasis, while overexpression increased metastases.²⁷ Flem-Karlsen et al. showed that knockdown of CD276 can make melanoma tumors more sensitive to dacarbazine and MAPK/mamalian target of rapamycin inhibitors.²⁶ CD276 inhibitors have been used in animal models and a CD276 inhibitor called enoblituzumab is under investigation in a phase 1 clinical trial.³⁰ Beyond melanoma, CD276 knockdown in breast cancer cells increased the sensitivity to paclitaxel and AKT/mamalian target of rapamycin inhibitors.^30,31 In an experimental metastasis model, CD276 silencing reduced metastatic capacity and increased the symptom-free survival of mice, suggesting that CD276 bears importance for chemoresistance and metastasis and is a potential target for “antimetastasis therapy.”³²

c-CBL was also overexpressed in the distant metastatic group (DEFC = 1.15; P = 0.01). c-CBL is an E3 ubiquitin ligase that regulates nuclear B-catenin and angiogenesis via tyrosine kinase receptor (TKR) and Wnt signaling. The CBL family, which includes b-CBL and c-CBL, participates in the MAPK pathway and targets TKR for degradation. With CBL silencing, the TKR signaling increases and levels of FAK-SRC-GRB2 decrease, which translates to less proliferation.^33,34 In a study published in Nature by Paolino et al., c-CBL knockdown resulted in decreased melanoma proliferation, migration, and invasion by driving up the population of natural killer cells attacking metastases.³⁵ There is also increasing evidence that overexpression of c-CBL contributes to the pathogenesis in prostate, gastric, pancreas, lung, colon, and hematogenous malignancies.^33,36–42

Of the existing clinicopathologic predictors of distant disease, tumor thickness and ulceration are the only validated prognostic markers included in the eighth edition of the AJCC staging system for melanoma.^2,43 Consistent with other investigations, our study showed that ulceration was a significant predictor of distant disease on univariate analysis. Thickness was not significant in this study, but the trend is reflective of the AJCC findings as the metastatic group in this cohort had thicker melanomas. Interestingly, lymph node status remains a factor of unclear implications as it does not always reflect tumor biology or propensity for distant spread.⁴³ In this study, there were five patients in the cohort who had positive sentinel nodes, three of whom never developed distant metastases. Conversely, there are patients with negative sentinel nodes who develop distant disease, as occurred in this study, and also in MSLT-I. In 2014, MSLT-I reported that 12% of negative sentinel node patients with intermediate thickness primaries developed distant disease as first site.^43,44 Aside from tumor characteristics and lymph node status, serum S-100 has been used as a biomarker for following risk of recurrence in high-risk melanoma. However, it has an AUC of 0.66 for predicting metastasis, which is noticeably lower than the AUC of 0.80 calculated here for CXCL1.^45,46 With imperfect predictability by lymph node status, only ulceration and thickness are reliable as prognostic markers in staging. There remains a need for more accurate markers to identify high-risk patients and thereby allow earlier intervention for improved outcomes. The four genes overexpressed in this study, particularly CXCL1, are candidate biomarkers for the task that might be useful for risk stratification.

There are several strengths of the present study. First, all patients had visible, residual melanoma after biopsy, which is not commonly the case because preoperative biopsy often leaves little to no intact tumor before surgery. Second, data on these patients were collected prospectively and the long follow-up revealed a subset who developed distant metastases. Third, by analyzing 74 immunomodulatory genes from a preselected list of 79 immunomodulatory genes for analysis, the odds of a false discovery rate were substantially reduced to 0.7%. Had the whole genome been analyzed, then 1000 genes might emerge relevant by random chance. The greatest limitation of this study is the small number in the cohort. Theoretically, it is possible that one data point with particularly high expression could drive the overall statistical significance and influence the study conclusions; however, this was not the case on our original analysis. We repeated the analysis using two different reference genes to normalize our qRT-PCR data and we obtained similar results in both cases. This led to the conclusion that overexpression of CXCL1 in patients who develop distant metastasis is not due to one specific patient with very high expression. With small size, focusing the analysis to the selected immunomodulatory genes to minimize false discovery is the best effort that could be done to strengthen the findings. This work is a pilot study to assess feasibility and generate initial observations. Now that we have completed this work, we plan to carry the investigation forward with a validation study. The inherent constraints of messenger RNA (mRNA) analysis are a second limitation. While measuring mRNA is an indicator of gene regulation, it is an indirect technique for quantifying levels of protein expression and cannot directly reveal gene mutation. The advantage is that microarray analysis of mRNA has been more affordable historically compared to DNA sequencing. A third limit is there is not yet any direct validation of the findings, but this would be a next step along the continuum of this investigation.

Conclusions

This work adds to the existing body of literature depicting the importance of immunomodulatory genes in cancer. CXCL1, CXCL2, CD276, and c-CBL have been implicated in melanoma in multiple studies besides ours.^{7,11,26,27,32–34,40} Overexpression of these four genes in the primary tumors of patients who eventually developed distant metastases is consistent with other studies about their mechanism of action. There is mounting evidence to corroborate the impact and role that immunomodulatory genes have in melanoma progression.

To our knowledge, this study is the first to analyze the role of immune-modulating genes in predicting distant metastases using primary tumors and to uncover genes informative of the risk of distant disease, particularly CXCL1. This could open the opportunity to use the primary tumor for risk stratification and to identify candidates for immunotherapy earlier in the disease course. Currently, there is no other similar means to prediction. Castle Biosciences offers a gene panel approach for predicting lymph node metastasis which also includes clinical features into the model. Ours is independent of clinical features, focuses on a single gene, and strives to predict distant metastasis rather than lymph node metastasis. Although melanoma is often responsive to treatment, the mutagenic behavior of melanoma leads to frequent drug resistance, underscoring the importance of new drug development and the utilization of combination therapy.^{19,20,23,26,32} With ongoing investigation into new inhibitors involving these four genes and others in the inflammatory cascade, new and synergistic drug development capitalizing on these findings can advance the therapeutic frontier of immunotherapy.