Novel genetic variants of PIP5K1C and MVB12B of the endosome-related pathway predict cutaneous melanoma-specific survival

Abstract

Endosomes regulate cell polarity, adhesion, signaling, immunity, and tumor progression, which may influence cancer outcomes. Here we evaluated associations between 36,068 genetic variants of 228 endosome-related pathway genes and cutaneous melanoma disease-specific survival (CMSS) using genotyping data from two previously published genome-wide association studies. The discovery dataset included 858 CM patients with 95 deaths from The University of Texas MD Anderson Cancer Center, and the replication dataset included 409 CM patients with 48 deaths from the Nurses’ Health Study (NHS) and the Health Professionals Follow-up Study (HPFS). In multivariate Cox proportional hazards regression analysis, we found that two novel SNPs (PIP5K1C rs11666894 A>C and MVB12B rs12376285 C>T) predicted CMSS, with adjusted hazards ratios of 1.47 (95% confidence interval = 1.15-1.89 and P = 0.002) and 1.73 (1.30-2.31 and 0.0002), respectively. Combined analysis of risk genotypes of these two SNPs revealed a dose-dependent decrease in CMSS associated with an increased number of risk genotypes (P _trend = 0.0002). Subsequent expression quantitative trait loci (eQTL) analysis revealed that PIP5K1C rs11666894 was associated with mRNA expression levels in lymphoblastoid cell lines from 373 European descendants (P<0.0001) and that MVB12B rs12376285 was associated with mRNA expression levels in cultured fibroblasts from 605 European-Americans (P<0.0001). Our findings suggest that novel genetic variants of PIP5K1C and MVB12B in the endosome-related pathway genes may be promising prognostic biomarkers for CMSS, but these results need to be validated in future larger studies.

Introduction

Cutaneous melanoma (CM) is the most aggressive skin cancer, ranking the fifth most common fatal malignancy among males and the sixth among females in the United States [1]. In 2019, an estimated 96,480 new CM cases were diagnosed in the United States, accounting for about 5.5% of all new cancer cases, of whom 7,230 patients died of the disease [2]. Although targeted therapy and immunotherapy have remarkably improved outcomes in patients with CM over the past decade, prognosis for patients with advanced or metastatic CM is still very poor, with five-year overall survival (OS) of 30% to 40% [3,4]. Several clinical characteristics have been reported to affect the prognosis of CM patients, such as tumor stage, Breslow thickness, and ulceration [5], but the prognosis based on these variables still has limited accuracy. Recently, genetic factors such as single nucleotide polymorphisms (SNPs) have been shown to affect CM disease-specific survival (CMSS) [6]. Therefore, identifying the roles of these genetic factors in CM survival may improve personalized management and treatment of CM patients.

Studies have suggested that some SNPs affect protein functions by various mechanisms, including gene expression regulation, and play an important role in molecular pathogenesis [7]. Genome-wide association studies (GWASs) have advanced the investigation of complex disease genetics and identified thousands of disease-associated SNPs [8]. However, associations detected by GWASs do not yield specific gene targets or provide biological mechanisms, because the SNPs identified and reported are often located in intragenic regions. However, there is a growing realization that the majority of phenotypic variation remains unaccounted for, and much of the heritability remains unexplained [9]. From identification of an association to a functional discovery, incorporating biological insights via explicit modeling of underlying biological pathways is essential and has become the focus of recent research [10]. To date, few novel and functional SNPs have been identified to be associated with CM prognosis in the reported GWASs. In addition, most of the identified SNPs failed to reach the rigorous genome-wide significance level, and >90% of disease-associated SNPs are located in non-protein-coding regions of the genome, of which many are far away from the nearest known gene [11]. However, in the post-GWAS era, one can use a hypothesis-driven strategy to identify functional genetic variants in targeted biological pathway genes through their associations with CMSS at a pathway level, and this can be achieved with publicly available genotyping data from multiple previously-published GWAS datasets [10].

The endosomal system is a complex network of membrane-bound compartments, which regulates cell polarity, adhesion, signaling, immunity, and tumor progression, and may influence tumor progression and prognosis of cancer patients. Studies have suggested that cancer development and tumor progression involve endosomal upregulation of oncogenic signalling, likely through defective trafficking of growth factor receptors, increased recycling, or decreased degradation [12,13]. For example, the endosome plays an important role in carcinogenesis of melanocytes through the interaction between melanoma oncogene MITF and macrophage [14]. A multiepitope tissue analysis revealed that endosomal translocation and hence metalloproteinase ADAM10 activation were key steps in the transformation of melanocytes and melanoma development [15]. On the other hand, some animal studies have indicated that Rab7 GTPase, a late endosome/lysosome-associated small GTPase, regulates tumorigenesis and mediates growth and metastasis of melanoma under the control of the endosomal/lysosomal system in lysosomal acid lipase-deficient mice [16]. Furthermore, the endosome participates in hypoxia-induced melanoma metastasis through the upregulated expression of Rab5 in another mouse model [17].

In humans, the endosomal system as the pathogen entry-point is equipped to process and display the foreign fragments to elicit an effective immune response [18]. In counteracting with deleterious effects of tumor agents, antigen-presenting cells (APCs) rewire their endosomal pathway to optimize the presentation of antigens to the immune system, while canonical presentation of exogenous antigens by major histocompatibility complex (MHC) class II (residing in the late endosomal compartment termed the MIIC) to CD4+ T cells is transiently supplemented by cross-presentation on MHC class I to CD8+ T cells [19,20]. These indicate that the endosomes are involved in CM carcinogenesis, metastasis, and immune response. Therefore, we hypothesize that genetic variants of endosome-related pathway genes are associated with CM survival. We tested this hypothesis using genotyping data from publicly available melanoma GWAS datasets.

Materials and methods

Study populations

The discovery dataset included genotyping and survival data on 858 non-Hispanic white patients with CM recruited at The University of Texas MD Anderson Cancer Center (MDACC) GWAS study derived from a hospital-based case-control study. The available demongraphic and clinical information included age, sex, Breslow thickness, metastasis, ulceration of tumor, mitotic rate, and survival outcome. The replication dataset comprised genotyping and survival data on an additional 409 participants with invasive CM from two cohort studies: the Nurses’ Health Study (NHS) and the Health Professionals Follow-up Study (HPFS). There were some differences in age and sex distributions between the two datasets, e.g., the proportion of young patients (≤50 years) was 43.24% in the MDACC dataset and 17.60% in the Harvard dataset. Also, the percentage of female participants was 42.19% and 66.26% in the MDACC and Harvard datasets, respectively. Subject selection and data collection for both discovery and replication datasets have been published in detail elsewhere [21,22]. All patients in both studies provided written informed consent under a research protocol approved by the Institutional Review Boards of the MDACC, Brigham and Women’s Hospital, and the Harvard T.H. Chan School of Public Health, and those of participating registries as required.

Gene selection and SNP genotyping

The genes involved in the endosome-related pathway were selected from the Molecular Signatures Database of Gene Set Enrichment Analysis (GSEA) website (http://software.broadinstitute.org/gsea/msigdb/search.jsp), using the keyword “endosome”. After the removal of 62 duplicated genes, six genes in the X chromosome and one pseudogene, we used the 228 endosome-related pathway genes located on the autosomes as candidate genes for further analysis (Table S1). In the MDACC GWAS dataset, genomic DNA was extracted from whole blood cells and used for genotyping in the Illumina HumanOmni-Quad_v1_0_B array. The genotyping data are available at the National Center for Biotechnology Information Database of Genotypes and Phenotypes (dbGaP Study Accession: phs000187.v1.p1). Genome-wide imputation was performed based on the CEU data from 1000 Genomes Project (phase I v2), utilizing the MACH software (March 2010 release). We used both typed (with a genotyping success rate of 95% and a Hardy-Weinberg equilibrium P value of 10^-5) and imputed (r²>0.8) common SNPs (with a minor allele frequency of 0.05) within ± 2 kilobase flanking regions of these endosome-related pathway genes. In the validation GWAS dataset, whole blood DNA samples were used for genotyping with the HumanHap610, Affymetrix 6.0, and Illumina HumanHap550 arrays, and imputation was based on haplotype information from 1000 Genomes project CEU data (2012 release), by applying MACH with quality control similar to that used for the MDACC GWAS dataset.

Statistical methods

We first assessed associations between all available SNPs in endosome-related pathway genes and CMSS in a single-locus Cox proportional hazards regression analysis. Then we performed multivariate Cox proportional hazards regression analyses with adjustment for available covariates. We used the less stringent Bayesian false-discovery probability (BFDP) for multiple testing correction, because the vast majority of the SNPs under investigation are in high linkage disequilibrium (LD), and we intended to identify all possible functional SNPs with a relaxed correction method. We subsequently evaluated cumulative effects of identified SNPs and also showed the associations between CM survival probability and combined risk genotypes via Kaplan-Meier (KM) survival curves. The receiver operating characteristic (ROC) curve and time-dependent area under the curve (AUC) were used to compare the prediction accuracy of models including both clinical and genetic variables on CMSS. To evaluate the correlation between the identified SNPs and mRNA expression levels of their genes, we performed expression quantitative trait loci (eQTL) analyses using the mRNA expression data of lymphoblastoid cell-lines derived from the 373 European-descendants in the 1,000 Genomes Project, and 483 cultured fibroblasts obtained from the genotype-tissue expression (GTEx) project database. Other details are presented in Supplementary Methods.

Results

Patient features

Baseline characteristics of CM patients from both GWAS datasets have been described elsewhere [21,22]. The MDACC dataset had demographic and clinical information about age, sex, tumor stage, Breslow thickness, ulceration and mitotic rate, while the NHS/HPFS dataset had only age, sex, and survival outcome. Compared with the NHS/HPFS dataset, patients diagnosed in the MDACC dataset were younger (82.4% vs. 56.8% for >50 years) and more likely to be male (33.8 vs. 57.8%). The MDACC dataset included CM cases of different stages diagnosed at an outpatient clinic, while the NHS/HPFS dataset included CM cases likely at an earlier stage I/II from the active follow-up of healthy populations. For these reasons, the median follow-up was shorter among MDACC patients, compared with NHS/HPFS patients (81.1 months vs. 179.0 months); however, death rates during follow-up were similar: (95/858 or 11.1% and 48/409 or 11.7%), respectively (Table S2). Because none of the principal components based on the GWAS genotyping data was significantly associated with CM survival, suggesting no significant population stratification in either the MDACC or NHS/HPFS datasets, we did not adjust for the principal components in either the discovery or replication analyses.

Associations between SNPs in the endosome-related pathway genes and CMSS

After removal of 62 duplicated genes, one pseudogene (AC034102.1), and six genes on the X chromosome (i.e., MSN, SNX12, IL2RG, IQSEC2, SH3KBP1, MTM1 because no standard statistics have been established for sex-specific analysis), 228 endosome-related genes remained for further analysis. As shown in Figure 1, a total of 4,941 genotyped and 31,127 imputed SNPs were extracted for 228 endosome-related pathway genes from the MDACC discovery dataset. The associations between these SNPs and CMSS were assessed in a single-locus analysis using Cox regression; 2,527 SNPs were found to be associated with CMSS (P<0.05) in an additive genetic model; and 2,091 SNPs remained noteworthy after multiple testing correction by BFDP <0.8. With subsequent replication in the NHS/HPFS dataset, only 38 SNPs remained significantly associated with CMSS, which are located in five genes (33 SNPs in MVB12B, two in PIP5K1C, and one each in AGAP1, IQSEC3, and IQSEC1) (Table S3) and remained significant in the meta-analysis of the two datasets without obvious heterogeneity (Table S4). In the LD analysis of these 38 significant SNPs, we found two SNPs in PIP5K1C and 33 SNPs in MVB12B to be in high LD (Figure S1). The remaining SNPs were further tested for their independence in predicting CMSS.

Figure 1.

The flowchart of the study. Abbreviations: SNP, single-nucleotide polymorphism; MDACC, The University of Texas MD Anderson Cancer Center; GWAS, genome-wide association study; MAF, minor allele frequency; HWE, Hardy Weinberg equilibrium; BFDP, Bayesian false-discovery probability; NHS/HPFS, the Nurses’ Health Study/Health Professionals Follow-up Study; AUC, area under curve; ROC, receiver operating characteristic.

Two independent SNPs predict CMSS

To identify SNPs with independent effects, we conducted stepwise multivariate Cox regression analyses to assess the effects of 38 validated SNPs on CMSS in the MDACC dataset, but not in the NHS/HPFS dataset that did not contain the same clinical covariates as the MDACC dataset did. Three SNPs (rs11666894, rs12376285, and rs77278014) in three genes (PIP5K1C, FAM125B, and IQSEC1, respectively) remained significantly associated with CMSS (P<0.05) in the presence of clinical covariates. After we expanded this prediction model with adjustment for an additional 40 previously reported survival-associated SNPs in the same MDACC GWAS dataset, we found that two SNPs (PIP5K1C rs11666894 A>C and MVB12B rs12376285 C>T) remained significantly associated with CMSS (P = 0.003 and 0.013, respectively) (Table 1). The results of meta-analysis of these two independent SNPs in each dataset are shown in Table 2 without heterogeneity across datasets. Specifically, as shown in Table 3, we found that PIP5K1C rs11666894 C and MVB12B rs12376285 T alleles were both associated with a poor CMSS in the MDACC dataset (P _trend = 0.033 and 0.004, respectively), with similar results in the NHS/HPFS dataset (P _trend = 0.022 and 0.017, respectively) and in the combined MDACC and NHS/HPFS dataset (P _trend = 0.002 and 0.014, respectively) (Table 3). For visual presentation, all of the identified SNPs in the present study (including the two independent SNPs) are depicted in a Manhattan plot (Figure S2), and additional regional association plots for these two SNPs are displayed in Figure S3.

Table 1.

Two independent SNPs identified by multivariate Cox proportional hazards regression analysis in the MDACC melanoma GWAS dataset

Parameter	Category1	Frequency	HR (95% CI)2	P 2	HR (95% CI)3	P 3
Age	≤50/>50	371/487	1.40 (0.89-2.20)	0.145	2.36 (1.35-4.13)	0.003
Sex	Female/Male	362/496	1.62 (1.01-2.60)	0.045	1.35 (0.80-2.28)	0.263
Regional/distantmetastasis	No/Yes	709/149	3.86 (2.51-5.92)	<0.0001	12.41 (6.77-22.76)	<0.0001
Breslow thickness (mm)	≤1/>1	347/511	1.19 (1.12-1.25)	<0.0001	1.26 (1.17-1.36)	<0.0001
Ulceration	No/Yes	681/155	2.40 (1.56-3.71)	<0.0001	3.40 (2.00-5.78)	<0.0001
Mitotic rate (/mm2)	≤1/>1	275/583	2.53 (1.24-5.15)	0.011	2.68 (1.19-6.03)	0.017
PIP5K1C rs11666894 A>C	AA/AC/CC	240/447/171	1.50 (1.09-2.06)	0.013	1.76 (1.21-2.56)	0.003
MVB12B rs12376285 C>T	CC/CT/TT	597/237/24	1.67 (1.13-2.46)	0.010	1.78 (1.13-2.81)	0.013

Abbreviations: SNP, single-nucleotide polymorphism; MDACC, The University of Texas MD Anderson Cancer Center; HR, hazards ratio; CI, confidence interval.

¹The “category” was used as the reference;

²Stepwise analysis included age, sex, regional/distant metastasis, Breslow thickness, ulceration, mitotic rate and SNPs;

³Forty published SNPs were used for post-stepwise adjustment.

Four SNPs were reported in previous publication (PMID: 25953768); Three SNPs was reported in the previous publication (PMID: 25628125); Four SNPs were reported in the previous publication (PMID: 25243787); Two SNPs were reported in the previous publication (PMID: 26575331); Two SNPs were reported in the previous publication (PMID: 30734280); Three SNPs were reported in the previous publication (PMID: 30596980); Three SNPs were reported in the previous publication (PMID: 29313974); Two SNPs were reported in the previous publication (PMID: 29088810); Four SNPs were reported in the previous publication (PMID: 28796414); One SNPs were reported in the previous publication (PMID: 28542949); Two SNPs were reported in the previous publication (PMID: 28499756); Three SNPs were reported in the previous publication (PMID: 27914105); Three SNPs were reported in the previous publication (PMID: 27578485).

Table 2.

Meta-analysis of the two independent SNPs in endosome-related pathway genes identified from two previously published melanoma GWAS datasets

SNP	Allele1	Gene	Discovery-MDACC (n = 858)				Validation-NHS/HPFS (n = 409)				Combined-analysis (n = 1267)
			EAF	HR (95% CI)	P 2	BFDP	EAF	HR (95% CI)	P 3	BFDP	P het	I 2	HR (95% CI)	P 4
rs12376285$	C>T	MVM12B	0.17	1.75 (1.19-2.56)	0.004	0.434	0.19	1.71 (1.10-2.66)	0.017	0.691	0.938	0	1.73 (1.30-2.31)	1.87×104
rs11666894$	A>C	PIP5K1C	0.46	1.40(1.03-1.91)	0.033	0.771	0.49	1.61(1.08-2.44)	0.022	0.709	0.593	0	1.47 (1.15-1.89)	0.002

¹Reference allele>effect allele;

²Adjusted for age, sex, Breslow thickness, distant/regional metastasis, ulceration and mitotic rate in the additive model;

³Adjusted for age and sex in the additive model;

⁴Meta-analysis in the fix-effect model;

^$Imputed SNP.

Abbreviations: SNP, single-nucleotide polymorphism; GWAS, genome-wide association study; MDACC, MD Anderson Cancer Center; EAF, effect allele frequency; HR, hazards ratio; CI, confidence interval; BFDP, Bayesian false-discovery probability; P_het, P value for heterogeneity by Cochrane’s Q test; MVB12B, multivesicular body subunit 12B; PIP5K1C, phosphatidylinositol-4-phosphate 5-kinase type 1 gamma.

Table 3.

Joint-analysis of the two identified independent SNPs in endosome-related pathway genes and CMSS of patients in the MDACC dataset, the NHS/HPFS dataset, and the MDACC and NHS/HPFS pooled dataset

Genotype	MDACC (n = 858)				NHS/HPFS (n = 409)				MDACC+NHS/HPFS (n = 1267)
	Frequency		Multivariate analysis1		Frequency		Multivariate analysis2		Frequency		Multivariate analysis3
	All	Death (%)	HR (95% CI)	P	All	Death (%)	HR (95% CI)	P	All	Death (%)	HR (95% CI)	P
PIP5K1C rs11666894 A>C
AA	240	17 (7.08)	1.00		98	8 (8.16)	1.00		338	25 (7.40)	1.00
AC	447	54 (12.08)	1.70 (0.96-2.99)	0.068	208	21 (10.10)	1.28 (0.56-2.90)	0.556	655	75 (11.45)	1.59 (1.01-2.51)	0.044
CC	171	24 (14.04)	2.03 (1.05-3.90)	0.035	103	19 (18.45)	2.42 (1.06-5.54)	0.037	274	43 (15.69)	2.18 (1.33-3.57)	0.002
Trend test				0.033				0.022				0.002
AC+CC	618	78 (12.62)	1.79 (1.03-3.08)	0.040	311	40 (12.86)	1.65 (0.77-3.54)	0.197	929	118 (12.70)	1.77 (1.15-2.72)	0.010
MVB12B rs12376285 C>T
CC	597	61 (10.22)	1.00		265	21 (7.92)	1.00		862	82 (9.51)	1.00
CT	237	29 (12.24)	1.48 (0.94-2.32)	0.093	129	26 (20.16)	2.60 (1.46-4.63)	0.001	366	55 (15.03)	1.55 (1.10-2.18)	0.012
TT	24	5 (20.83)	5.28 (2.00-13.95)	0.001	15	1 (6.67)	0.81 (0.11-6.00)	0.833	39	6 (15.38)	1.55 (0.67-3.56)	0.302
Trend test				0.004				0.017				0.014
CT+TT	261	34 (13.03)	1.63 (1.06-2.51)	0.027	144	27 (18.75)	2.41 (1.36-4.26)	0.003	405	61 (15.06)	1.55 (1.11-2.16)	0.010
Number of risk genotypes4
0	165	8 (4.85)	1.00		60	4 (6.67)	1.00		225	12 (5.33)	1.00
1	507	62 (12.23)	2.58 (1.16-5.72)	0.020	243	21 (8.64)	1.31 (0.45-3.84)	0.618	750	83 (11.07)	2.04 (1.11-3.74)	0.021
2	186	25 (13.44)	3.77 (1.61-8.87)	0.002	106	23 (21.70)	3.45 (1.19-10.03)	0.023	292	48 (16.44)	3.09 (1.64-5.82)	5E-04
Trend test				0.002				0.001				2E-04
0-1	672	70 (10.42)	1.00		303	25 (8.25)	1.00		975	95 (9.74)	1.00
2	186	25 (13.44)	1.69 (1.06-2.70)	0.027	106	23 (21.70)	2.76 (1.56-4.88)	0.001	292	48 (16.44)	1.71 (1.21-2.43)	0.002

¹Adjusted for age, sex, Breslow thickness, distant/regional metastasis, ulceration and mitotic rate in the MDACC dataset;

²Adjusted for age and sex in the NHS/HPFS dataset;

³Adjusted for age and sex in the combined MDACC and NHS/HPFS dataset;

⁴Risk genotypes include PIP5K1C rs11666894 AC+CC and MVB12B rs12376285 CT+TT.

Survival of CM patients with combined risk genotypes

To investigate the joint effect of the two independent SNPs on CMSS, we combined the risk genotypes of PIP5K1C rs11666894 AC+CC and MVB12B rs12376285 CT+TT. We divided the patients into three groups according to their number of risk genotypes (NRG), and the trend test in each dataset revealed a statistically significant dose-response effect of NRG on survival. That is, the higher NRG, the worse survival in the MDACC dataset (P _trend = 0.002), the NHS/HPFS dataset (P _trend = 0.001), and the MDACC and NHS/HPFS combined dataset (P _trend = 0.0002) after adjustment for covariates, wherever appropriate (Table 3).

We also dichotomized the patients into 0-1 and 2 NRG groups. As illustrated in Table 3, compared with the 0-1 NRG group, the 2-NRG group had a significantly worse CMSS in all datasets: for MDACC (HR = 1.69; 95% CI = 1.06-2.70, P = 0.027), NHS/HPFS (HR = 2.76; 95% CI = 1.56-4.88, P = 0.001) and the combined MDACC and NHS/HPFS dataset (HR = 1.71; 95% CI = 1.21-2.43, P = 0.002). We used Kaplan-Meier survival curves to illustrate the association between NRG and CMSS (Figure 2A-C).

Figure 2.

Two independent SNPs in endosome-related pathway genes predict cutaneous melanoma survival and eQTL analysis for them. Kaplan-Meier survival curves of CM patients by the combined risk genotypes of PIP5K1C rs11666894 and MVB12B rs12376285: dichotomized 0-1 risk genotype group and 2 risk genotypes group in the MDACC dataset (A), the NHS/HPFS dataset (B) and combined MDACC and NHS/HPFS dataset (C). The correlation of rs11666894 and PIP5K1C mRNA expression from the 1000 Genomes Project in an additive model (D) and an dominant model (E). The correlation of rs12376285 and MVB12B mRNA expression in cultured fibroblasts from the GTEx (F). Abbreviations: SNP, single-nucleotide polymorphism; eQTL, expression quantitative trait loci; CMSS, cutaneous melanoma-specific survival; MDACC, The University of Texas MD Anderson Cancer Center; NHS, the Nurses’ Health Study; HPFS, the Health Professionals Follow-up Study; GTEx, Genotype-Tissue Expression project.

Stratified analyses for the effect of combined risk genotypes on CMSS

Next we performed a stratification analysis to evaluate whether the combined effect of risk genotypes on CMSS was modified by other covariables, including age, sex, tumor stage, Breslow thickness, ulceration, and mitotic rate in the MDACC dataset, but only age and sex in the NHS/HPFS dataset. Compared with those with 0-1 risk genotypes, patients with 2 risk genotypes had a significantly worse survival, particularly evident in the subgroups of those aged >60 years, males, no regional/distant metastasis, Breslow thickness >1 mm, no ulceration, and mitotic rate >1 in the MDACC dataset, and the subgroup aged >60 and female in the NHS/HPFS dataset. Furthermore, there was a significant interaction between risk genotypes and regional/distant metastasis (P = 0.010), but not among the other subgroups (Table S5).

ROC curves and time-dependent AUC for CMSS prediction

To further evaluate the predictive value of the two independent SNPs, we used time-dependent AUC of the ROC curves for CM patients in the presence of available covariates (i.e., age, sex, tumor stage, Breslow thickness, ulceration, and mitotic rate). As shown in Figure S4A-F for the ROC curves, compared with clinical pathologic variables alone, we found that the combination of clinical pathologic variables and risk genotypes improved prediction performance of the five-year CMSS by 1.40% (AUC = 63.60% to 65.00%, P = 0.500) in the combined MDACC and NHS/HPFS dataset, without reaching statistical significance.

In silico functional validation of the two independent SNPs

To further explore the molecular mechanisms underlying the associations between the two independent SNPs and CMSS, we performed additional eQTL analysis for correlations between risk genotypes of the two independent SNPs and their corresponding mRNA expression levels in 373 normal lymphoblastoid cell lines from the 1000 Genomes Project database. We found that the rs11666894 C allele showed a significant correlation with decreased mRNA expression levels of PIP5K1C in both additive and dominant models (P = 3.12×10^-4 and P = 6.16×10^-5, respectively) (Figure 2D-E) but not in a recessive model (P = 0.076) (Figure S5D). Meanwhile, we found no evidence of a correlation between MVB12B mRNA expression levels and rs12376285 genotypes from the 1000 Genomes Project (Figure S5A-C). Additionally, we performed eQTL analysis by extracting data from the GTEx Project, which has data on both genotypes and mRNA expression for PIP5K1C rs11666894 and MVB12B rs12376285 obtained from cultured donor fibroblasts. The results showed that the rs12376285 T allele was significantly correlated with lower mRNA expression levels of MVB12B (P = 9.100×10^-4) (Figure 2F) but not the PIP5K1C rs11666894 allele (P = 0.880) (Figure S5E).

Finally, we evaluated the associations of mRNA expression levels of PIP5K1C and MVB12B in different cancers with survival, based on The Human Protein Atlas database (www.proteinatlas.org). As shown in KM survival curves (Figure S6A-E), higher expression levels of PIP5K1C mRNA were significantly associated with longer survival in patients with cancers of endometrium, cervix, pancreas, or kidney, indicating that PIP5K1C expression is a prognostic marker for survival in other cancer patients as well. Similarly, a higher expression level of MVB12B mRNA was significantly associated with a longer survival of patients with pancreatic cancer.

Discussion

To date, a limited number of susceptibility loci in a few pathways have been identified for predicting CM survival. To our knowledge, the present study is the first post GWAS-based pathway analysis to evaluate the associations between genetic variants in the endosome-related pathway genes and CM survival. We found that two novel independent SNPs, PIP5K1C rs11666894 A>C and MVB12B rs12376285 C>T, independently or jointly modulated survival of Caucasian CM patients. Specifically, PIP5K1C rs11666894 C and MVB12B rs12376285 T alleles were both identified as prognostic risk alleles for CMSS, and their combined NRG was significantly associated with worse CMSS. Furthermore, the prognostic risk alleles were significantly associated with decreased mRNA expression in lymphoblastoid cell lines or cultured fibroblasts. These results suggest potential biological roles of endosome-related pathway genes in CM survival.

PIP5K1C, located on chromosome 19p13.3, encodes phosphatidylinositol-4-phosphate 5-kinase type-1 gamma; this kinase has been reported in the formation of membrane contact sites among various organelles, including the plasma membrane, endoplasmic reticulum, and endolysosomal system. This kinase also performs spatial and temporal control of phosphatidylinositol (4,5)-bisphosphate (PIP2) synthesis as well as regulates E-cadherin cell-cell contacts and growth factor-stimulated dendritic cell (DC) migration; these findings suggest that PIP5K1C regulates key steps in metastasis [23-25]. PIP5K1C has been known to have immune-regulatory effects via the modulation of neutrophil polarization and infiltration [26]. PIP5K1C may also be a critical regulator of T-cell activation through PIP2, which in turn regulates T-cell receptor (TCR) signaling [27,28], and E-cadherin, which plays a critical role in the action of cytotoxic T lymphocytes (CTLs) in TCR-mediated killing by interacting with and directing exocytosis of lytic granules to the cancer cell surface at the immunological synapse [29,30].

To date, no studies on associations of PIP5K1C variants with cancer risk have been reported. However, a case-control association study showed that eight genetic variants in PIP5K1C were associated with alcohol use disorder in a cohort of African ancestry and that acute alcohol exposure led to up-regulation of PIP5K1C in a mouse model, indicating the potential biological significance of the PIP5K1C gene [31], but there was no report about the role of PIP5K1C in tumor progression in melanoma. Based on the current results, however, we speculate that PIP5K1C is more likely to be a suppressor gene in CM, because a higher mRNA expression of PIP5K1C predicted a better survival. Therefore, the molecular mechanisms of genetic variants in PIP5K1C underlying the observed association with survival are worthy of further investigation.

MVB12B, located on chromosome 9q 33.3, encodes a protein known as multivesicular body subunit 12B, also called a family with sequence similarity 125, member B. The protein product of MVB12B forms a component of the endosomal sorting complex required for transport (ESCRT)-I, a highly conserved complex [32,33] that is involved in vesicular trafficking (transcytosis); together, these molecules may exert either cancer-promoting or anti-tumoral effects [34,35] and are also presented either for the establishment or suppression of host immune responses [36,37]. It is known that MVBs are involved in a variety of diseases, include cancers of the breast, thyroid, ovaries, and colon as well as melanoma [38,39]. Specifically, it has been reported that the immunological activity of T cells regulated by the phosphorylation of S222 in MVB12B is essential for stimulation of immune STING signaling in recipient cells, resulting in inhibited T-cell proliferation and primed T cells for apoptosis [40].

Recently, a GWAS-based study of intra-ocular pressure (IOP) showed that the SNP rs2286885 in MVB12B was associated with IOP in the TwinsUK cohort, indicating a potential functional role of MVB12B in glaucoma [41]. However, no association between MVB12B and CM risk or CM survival has been reported. In the present study, we showed that MVB12B rs12376285 C>T, as a CMSS-associated variant, was associated with reduced MVB12B mRNA expression levels in skin fibroblasts, which supports a potential protective role of MVB12B in CM. The observation that the rs12376285 T allele may significantly decrease MVB12B mRNA expression levels, a possible molecular mechanism underlying the observed association, needs to be validated in additional in-depth mechanistic studies. It has been reported that other micro-environmental factors, such as immune cells, endothelial cells, the extracellular matrix, and soluble molecules, can interact with host fibroblasts to drive tumor progression [42]. Therefore, the role of MVB12B in regulating the melanoma microenvironment also warrants additional investigation.

Although no studies on the associations between PIP5K1C or MVB12B and CMSS have been published, the relevant correlations between the endosome and immunotherapies for CM have been well studied [43]. It is well known that both natural and therapeutically enhanced cytotoxic T lymphocyte (CTL) responses in melanoma patients are of limited efficacy. A recent report demonstrated that the mechanisms underlying the CTL effector-phase failure in melanoma were related to the endosome system, where human melanoma cells upon conjugation with CTL underwent active late endosome/lysosome trafficking at the lytic synapse, paralleled by cathepsin-mediated perforin degradation and deficient granzyme B penetration [44]. Another recent report showed that a novel peptide antigen-loaded nanoplex vaccine based on the endosome-destabilization inhibited tumor growth and metastasis in murine melanoma models, indicating an immune role of the endosome in the melanoma vaccine [45]. Furthermore, an effective antitumor immune response was induced using a mannose-functionalized endosomolytic nanocomposite(s)-based approach in a C57BL/6 mouse melanoma model, suggesting that prompt endosomal release and escape could be a promising approach to efficient CM immunotherapy [46]. Overall, immune functions, such as antigen presentation, strongly depend on a fully functional endosomal system that delivers exogenous antigens to both MHC class I and MHC class II pathways through cross-presentation of dendritic cells [47,48]. In addition, human dendritic cells take up the cell-penetrating melanoma antigen peptides through the endosome [49]. Therefore, given the poor prognosis of metastatic CM, genetic variants in endosome-related genes may explain individual variation in response to immunotherapy and thus in prognosis of melanoma patients [50].

Several limitations in the present study should be noted. Compared with the MDACC discovery dataset, the NHS/HPFS validation dataset had fewer participants with fewer variables for further analysis, which could reduce statistical power in validating the effects of other SNPs identified in the discovery. Furthermore, some clinical covariates of the study populations were not available in the NHS/HPFS dataset, further limiting the validation of the findings; however, the consistent survival associations across different datasets in stratification analysis suggest that our results might not be seriously biased by the absence of certain covariates in the replication population. Finally, although several genetic variants in the endosome-related genes backed by in silico functional evidence were found to be associated with CMSS, the exact molecular mechanisms through which these SNPs underlie the observed associations remain unclear.

In conclusion, two independent SNPs (i.e., PIP5K1C rs11666894 A>C and MVB12B rs12376285 C>T) were found to be significantly associated with CMSS in both MDACC discovery and NHS/HPFS replication datasets. The combined analysis revealed that these two SNPs had a significant association with CM survival and that patients with more risk genotypes had a worse prognosis, possibly through the allelic effects on their gene expression. Our findings provide new insights for additional functional studies to further support these genetic variants of the endosome-related pathway genes as new prognostic biomarkers or as clinical decision-making indicators for CM patients and their caregivers, once these findings are validated by additional large patient studies.

Disclosure of conflict of interest

None.

Abbreviations

APCs

antigen-presenting cells

AUC

area under curve

BFDP

Bayesian false-discovery probability

CI

confidence interval

CM

cutaneous melanoma

CMSS

cutaneous melanoma-specific survival

CTLs

cytotoxic T lymphocytes

DCs

dendritic cells

eQTL

expression quantitative trait loci

ESCRT

endosomal sorting complex required for transport

FAMB25

family with sequence similarity 125, member B

GSEA

Gene Set Enrichment Analysis

GTEx

Genotype-Tissue Expression

GWAS

genome-wide association study

HPAs

The Human Protein Altas

HR

hazards ratio

HR_adj

adjusted hazards ratio

LD

linkage disequilibrium

MDACC

The University of Texas MD Anderson Cancer Center

MHC

major histocompatibility complex

MHC I

MHC class I molecules

MHC II

MHC class II molecules

MVB12B

multivesicular body subunit 12B

NHS/HPFS

the Nurses’ Health Study/Health Professionals Follow-up Study

NRGs

number of risk genotypes

PIP2

phosphatidylinositol (4,5)-bisphosphate

PIP5K1C

Phosphatidylinositol-4-phosphate 5-kinase type-1 gamma

ROC

receiver operating characteristics curve

SNP

single-nucleotide polymorphism

TCGA

the Cancer Genome Atlas

TCR

T cell receptor