Chromosome 3 Status Combined With BAP1 and EIF1AX Mutation Profiles Are Associated With Metastasis in Uveal Melanoma

Abstract

Purpose.: Somatic mutations in GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 have been identified in uveal melanoma (UM). The aim of this study was to determine whether mutations in these genes in primary tumors were associated with metastases in individuals diagnosed with UM.

Methods.: A total of 63 UM cases who developed a metastasis within 48 months of primary treatment and 53 UM controls who were metastasis-free over a similar time period were selected for the study. Primary UM cases were screened for mutations in GNAQ, GNA11, SF3B1, EIF1AX, and BAP1. The association of these mutations with tumor characteristics, chromosome 3 copy number, and metastatic status was analyzed by logistic regression to estimate the odds of developing metastasis within 48 months.

Results.: As expected, tumor diameter, thickness, cilio-choroidal location, and chromosome 3 monosomy were all significantly (P < 0.02) associated with the presence of metastasis. In univariate analysis, GNA11 (odds ratio [OR] 2.5, 95% confidence interval [CI] 1.1–5.5) and BAP1 (OR 6.3, 95% CI 2.7–14.4) mutations were positively associated and EIF1AX mutation (OR 0.13, 95% CI 0.034–0.47) was inversely associated with metastatic status at 48 months after UM treatment. After adjustment for covariates, a chromosome 3 monosomy/BAP1-mutation/EIF1AX–wild-type (WT) mutation profile was strongly associated (OR 37.5, 95% CI 4.3–414) with the presence of metastasis compared with a chromosome 3 disomy/BAP1-WT/EIF1AX mutation profile.

Conclusions.: The results suggest that knowledge of mutations in BAP1 and EIF1AX can enhance prognostication of UM beyond that determined by chromosome 3 and tumor characteristics. Tumors with chromosome 3 disomy/BAP1-WT/EIF1AX-WT have a 10-fold increased risk of metastasis at 48 months compared with disomy-3/BAP1-WT/EIF1AX mutant tumors.

Introduction

Uveal melanoma (UM) is the most common adult-onset intraocular tumor and is associated with a high incidence of mortality due to metastasis, most often to the liver. ^1–3 The primary risk factors for tumor metastasis in UM include large tumor size, location, extraocular extension, ciliary body involvement, cell type, and chromosomal abnormalities, including loss of chromosome 3, gain of 8q, and loss of 1p, 6q, and 8p. ^4–17 Recent studies have focused the search for better prognostic indicators of UM on gene expression ^18–22 and gene mutation analyses. ^23–28  

Recently, somatic mutations were identified in five genes in UM: two members of the q class of the G-protein α-subunits, GNAQ ^24,25 and GNA11, ²⁶ a tumor suppressor gene located on chromosome 3p21.1, BRCA1-associated protein1, BAP1, ²³ the splicing factor SF3B1, ^27–29 and the X-linked translation initiation factor EIF1AX. ²⁸ Mutations unique to UM have been identified in up to 50% of tumors at position Q209 in exon 5 and R183 in exon 4 of GNAQ. ^24,25 These same amino acids are mutated in GNA11 in approximately 35% of tumors. ²⁶ The mutations in GNAQ or GNA11 occurred in approximately 80% of tumors in a mutually exclusive manner, ^26,30,31 and are predicted to be early, initiating events in tumorigenesis. ²⁴ Mutations at position R625 of SF3B1 have been identified in both UM ²⁷ and cutaneous melanoma, ³² as well as in breast and hematological cancers. ^33–35 Thus far, mutations in exons 1 and 2 of EIF1AX have been described only in UM. ²⁸  

Whole-exome sequencing of metastatic UM identified inactivating somatic mutations in BAP1 in 84% of metastasizing tumors, suggesting that inactivation of BAP1 is a key event occurring later in UM progression and coinciding with the onset of metastatic behavior. ²³ Somatic BAP1 mutations also have been identified in a variety of other cancers, including cutaneous melanoma, mesothelioma, and colorectal, renal cell, and lung cancers. ^36–46  

This study used a case-control design to investigate whether mutations in GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 in primary UM were associated with the development of metastasis within 48 months after primary treatment. The principal aim was to determine whether somatic mutations at these loci were independently associated with the presence of UM metastases after taking into consideration known risk factors, including tumor size, location, and chromosome 3 copy number. The ultimate goal of this work was to inform the development of enhanced prediction models that can distinguish patients at the time of initial cancer diagnosis who are at high risk for poor UM outcomes. 

Methods

Patients

All 116 patients enrolled into this study were managed by the Ocular Oncology Service at Wills Eye Hospital, Philadelphia, PA, USA, between 1990 and 2013. In total, 63 UM cases who developed a metastasis within 48 months of primary treatment and 53 UM controls who had not developed a metastasis within this same period were selected for study. We choose 48 months as a cutoff point because our previous study showed that 93% of patients developed metastasis within 48 months. ¹⁵ Information on patient demographics and metastatic status and clinical and pathological data on tumors was obtained by a retrospective review of medical charts. Chromosome and mutation analysis of primary UM tumor samples were carried out by the Genetic Diagnostic Laboratory, University of Pennsylvania, Philadelphia, Pennsylvania, United States. 

All studies adhered to the tenets of the Declaration of Helsinki. The institutional review boards of University of Pennsylvania and Wills Eye Hospital approved this research. Written informed consent for use of tissues and data for research was obtained from all patients who participated in genetic testing. 

DNA Extraction and Determination of Chromosome 3 Copy Number

Seventy-nine tumor samples were obtained by fine-needle aspiration biopsy (FNAB) and 37 by solid open biopsy of enucleated tumors. Genomic DNA was isolated as previously described. ¹⁵ Chromosome 3 copy number was analyzed in 54 tumors by microsatellite analysis ⁴⁷ and in 62 tumors using Affymetrix Human 100K, single-nucleotide polymorphism (SNP)-5.0, or SNP-6.0 genotyping arrays (Affymetrix, Santa Clara, CA, USA). ¹⁵  

Mutation Screening of GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 Genes

Screening for somatic mutations in the GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 genes was performed using the following three different methods. 

Taqman Genotyping of GNAQ, GNA11, and SF3B1.

Custom assays were designed for five previously identified mutations in GNAQ (Gln209Leu and Gln209Pro), ^24,25 GNA11 (Gln209Leu), ²⁶ and SF3B1 (Arg625Cys and Arg625His). ²⁷ The assays were performed using the StepOnePlus Real-Time PCR System following the manufacture's recommendations (Life Technologies, Carlsbad, CA, USA). 

Sanger Sequencing of BAP1 and EIF1AX.

Sanger sequencing of exons 1 to 17 of BAP1 was performed for 44 UM samples and for exons 1 and 2 of EIF1AX for all 116 samples following standard protocols. 

Next-Generation Sequencing (NGS).

Screening for mutations in the coding exons of BAP1, GNAQ, and GNA11 of an additional 72 UM samples was done using multiplex-PCR followed by NGS on the Ion Torrent PGM (Personal Genome Machine) platform. ⁴⁸ Multiplex PCR was performed using a primer mix containing primers for exons 1 to 17 of BAP1 and exons 4 and 5 of GNAQ and GNA11. The sequencing libraries were barcoded and sequenced using the Ion PGM Sequencing 200 Kit (Invitrogen, Grand Island, NY, USA) following the manufacturer's instructions. For initial validation, two UM samples with known mutations were analyzed following the protocol detailed above. The sequencing data from these two tumors allowed assessment of background noise and filtering of false-positive variant calls due to homopolymer issues. 

The sequence alignment and variant calling of the Ion Torrent PGM sequencing data were processed against the human hg19 reference sequence using Ion Torrent Suite v2.0 (Life Technologies). All variant calls were confirmed using NextGENe v2.3.0 software (SoftGenetics, State College, PA, USA). The variant output was annotated using SeattleSeq Annotation 137. ⁴⁹ Variants were called if they passed the quality control filter, which required at least ×400 coverage for the reference sequence and ×80 (20%) coverage for the variant reads. Known SNPs were filtered out and variants excluded from further analysis if the P value was greater than 0.05, the variant frequency less than 20%, or it was intronic and more than 5 bp away from the exon–intron boundary. Sanger sequence validation of 12 variants identified on the Ion Torrent platform was performed; all variants were confirmed on Sanger sequencing when the P value was less than 0.05. These criteria preclude detection of low-level variation that may be present due to tumor heterogeneity. 

Statistical Analyses

Logistic regression was used to assess the association between metastatic status and patient characteristics (sex and age), tumor characteristics (thickness, basal diameter, location), TNM (Tumor size and extent, lymph Node involvement, and presence of Metastasis) staging, ¹⁶ chromosome 3 status, and presence of somatic mutations in GNAQ, GNA11, SF3B1, EIF1AX, and BAP1. Tumor location was coded as choroidal (1), ciliary body (2), iris (3), and choroidal with ciliary body involvement (4). TNM tumor staging was done according to the American Joint Committee on Cancer TNM system. ¹⁶ Logistic regression was used to determine associations between somatic mutations in GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 and patient and tumor characteristics. Here, the presence of a somatic mutation was considered the outcome (dependent) variable; to adjust for metastatic status, it was included in all models as an indicator variable (0 = absent; 1 = present). 

Multivariate logistic regression modeling was used to determine the independent associations of mutations in GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 with metastatic status after adjusting for clinicopathological covariates shown to be statistically significantly associated with outcome in the univariate analyses. In the analysis of the joint effects of chromosome 3, EIF1AX and BAP1 mutation status, disomy-3, and BAP-wild type (WT) were coded as 0 and monosomy-3 and the presence of BAP1 mutations, both of which carry a risk of metastasis, were coded as 1. Because the presence of EIF1AX mutations was found to be protective, it was coded as 0 and the WT allele as 1. For the combined analysis, we defined the reference multilocus genotype as disomy-3/BAP1-WT/EIF1AX-mutant. Statistical analyses were carried out using SPSS 20 (IBM, New York, NY, USA), and P values less than or equal to 0.05 were considered statistically significant. 

Results

Demographic and UM Characteristics

A description of patient and UM characteristics and their association with metastatic status is presented in Table 1. Comparing cases with controls, tumor diameter (P < 0.001), thickness (P < 0.001), location (P = 0.018), and TNM stage (P < 0.001) were significantly associated with metastatic status, but sex and age were not (Table 1). Monosomy-3 was present in 75 UM samples (65%) and was significantly associated with metastasis (P < 0.001, Table 1). 

Table 1

 

Association of Patient and Tumor Characteristics With Metastasis Assessed by Univariate Logistic Regression

Tumor Variable	All Tumors, n = 116 (Frequency)	Case/Control Status		Logistic Regression
		Controls: No Metastasis, n = 53 (Frequency)	Cases: Metastasis, n = 63, (Frequency)	OR	95% CI Lower–Upper	P Value*
Sex						0.182
Female	47 (0.40)	25 (0.47)	22 (0.35)	Reference
Male	69 (0.60)	28 (0.53)	41 (0.65)	1.7	0.79–3.5
Age, y, median, range†	61, 22–88	57, 22–83	62, 22–88	1.8	0.50–6.5	0.374
Basal diameter, mm, median, range	13.0, 5.0–22	10, 5.0–19	16, 8–22	1.5	1.3–1.7	<0.001
Thickness, mm, median, range	6.6, 1–16.5	1.0, 1–13.5	8.7, 2–16.5	1.5	1.3–1.8	<0.001
Location of tumor						0.018
Choroid	79 (0.68)	43 (0.81)	36 (0.57)	Reference
Ciliary body	4 (0.034)	2 (0.038)	2 (0.032)	1.2	0.16–8.9
Cilio-choroid	31 (0.27)	6 (0.11)	25 (0.40)	5.0	1.8–13.5
Iris	2 (0.017)	2 (0.038)	0 (0)	Not estimable
TNM tumor stage‡						<0.001
Stage I	21 (0.18)	19 (0.37)	2 (0.03)	Reference
Stage IIA	26 (0.23)	20 (0.39)	6 (0.10)	2.8	0.51–15.9
Stage IIB	32 (0.28)	7 (0.14)	25 (0.40)	33.9	6.3–182
Stage IIIA	25 (0.22)	4 (0.078)	21 (0.33)	49.9	8.2–304
Stage IIIB	6 (0.053)	1 (0.020)	5 (0.08)	47.5	3.5–636
Stage IV	4 (0.035)	0	4 (0.06)	Not estimable
Chromosome 3
Disomy	41 (0.35)	30 (0.57)	11 (0.18)	Reference		<0.001
Monosomy	75 (0.65)	23 (0.43)	52 (0.82)	6.2	2.6–14.4

  P values in bold indicate statistically significant association.

*  Overall P value from Wald test.

†  Odds ratio for age is based on natural log-transformed values.

‡  Tumor staging according to American Joint Committee on Cancer TNM system. ¹⁶ In this analysis, iris melanoma was not included in the TNM staging and, because information on lymph node involvement was not available for any sample, it could not be used in determining tumor stage.

Table 2 shows associations between patient and tumor characteristics and GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 mutation status after adjusting for metastatic status. GNAQ mutations were significantly associated with basal diameter (odds ratio [OR] 1.1, 95% confidence interval [CI] 1.0–1.3), whereas both GNAQ and GNA11 mutations were associated with cilio-choroidal location (OR 0.39, 95%CI 0.15–0.99 and OR 3.5, 95% CI 1.4–8.9, respectively). BAP1 mutations were associated with tumor thickness (OR 1.2, 95% CI 1.0–1.4). SF3B1, EIA1AX, and BAP1 mutations were all significantly associated with chromosome 3 status (OR 0.20, 95% CI 0.045–0.87; OR 0.26, 95% CI 0.078–0.87; and OR 23.6, 95% CI 6.3–88.2, respectively). 

Table 2

 

Association of Patient Demographics and Uveal Melanoma Characteristics With GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 Mutations Assessed by Logistic Regression and Adjusted for Metastatic Status

Tumor Variable	GNAQ, n = 113			GNA11, n = 116
	WT, n = 61 (0.54)	Mutant, n = 52 (0.46)	OR (95% CI)*	WT, n = 75 (0.65)	Mutant, n = 41 (0.35)	OR (95% CI)*
Patient sex
Female	26 (0.43)	21 (0.40)	Reference	31 (0.41)	16 (0.39)	Reference
Male	35 (0.57)	31 (0.60)	1.1 (0.52–2.3)	44 (0.59)	25 (0.61)	0.99 (0.44–2.2)
Patient age, mo, median, range†	60, 22–84	61, 22–88	0.65 (0.17–2.4)	60, 22–88	61, 22–84	1.5 (0.37–6.1)
Basal diameter, mm, median, range	12, 5–22	14, 6–22	1.1 (1.0–1.3)	12, 5–22	13, 5–22	0.93 (0.82–1.0)
Thickness, mm, median, range	7.0, 1.0–15.0	6.4, 2.0–16.5	1.0 (0.90–1.1)	5.7, 1.0–16.5	8.0, 1.5–15.0	1.1 (0.95–1.2)
Location of tumor			P = 0.20‡			P = 0.014‡
Choroid	36 (0.59)	40 (0.77)	Reference	60 (0.80)	19 (0.46)	Reference
Ciliary body	3 (0.05)	1 (0.019)	0.29 (0.029–3.0)	1 (0.01)	3 (0.07)	9.7 (0.93–101)
Cilio-choroid	21 (0.34)	10 (0.20)	0.39 (0.15–0.99)	13 (0.17)	18 (0.44)	3.5 (1.4–8.9)
Iris	1 (0.02)	1 (0.019)	1.0 (0.60–17.2)	1 (0.01)	1 (0.03)	4.4 (0.25–77.1)
TNM tumor stage			P = 0.12‡			P = 0.14‡
Stage I	14 (0.23)	7 (0.14)	Reference	17 (0.23)	4 (0.10)	Reference
Stage IIA	12 (0.20)	13 (0.26)	2.2 (0.66–7.5)	19 (0.26)	7 (0.18)	1.4 (0.35–5.9)
Stage IIB	12 (0.20)	19 (0.37)	3.6 (0.92–14.0)	23 (0.31)	9 (0.22)	1.1 (0.24–5.2)
Stage IIIA	17 (0.28)	7 (0.14)	0.94 (0.22–4.1)	9 (0.12)	16 (0.40)	5.0 (1.0–23.9)
Stage IIIB	2 (0.03)	4 (0.078)	4.6 (0.58–36.4)	4 (0.05)	2 (0.05)	1.4 (0.16–12.3)
Stage IV	3 (0.05)	1 (0.020)	0.78 (0.058–10.7)	2 (0.03)	2 (005)	2.6 (0.22–23.3)
Chromosome 3			P = 0.96			P = 0.053
Disomy	22 (0.36)	19 (0.36)	Reference	33 (0.44)	8 (0.20)	Reference
Monosomy	39 (0.64)	33 (0.64)	0.98 (0.44–2.3)	42 (0.56)	33 (0.80)	2.6 (0.99–6.7)

Table 2

 

Continued

Tumor Variable	SF3B1, n = 110			EIF1AX, n = 111
	WT, n = 99 (0.90)	Mutant, n =11 (0.10)	OR (95% CI)*	WT, n = 93 (0.84)	Mutant, n = 18 (0.16)	OR (95% CI)*
Patient sex
Female	38 (0.38)	6 (0.54)	Reference	39 (0.42)	7 (0.39)	Reference
Male	61 (0.62)	5 (0.46)	0.51 (0.14–1.8)	54 (0.58)	11 (0.61)	1.5 (0.51–4.7)
Patient age, mo, median, range†	61, 22–88	52, 24–78	0.22 (0.037–1.3)	61,22–88	53.5, 22–84	0.90 (0.15–5.6)
Basal diameter, mm, median, range	12, 5–22	15, 6–22	1.2 (0.98–1.4)	14, 5–22	11, 6–18	1.0 (0.87–1.2)
Thickness, mm, median, range	6.6, 1.0–16.5	6.5, 2–11.7	1.0 (0.83–1.2)	7.3, 1.0–16.5	5.6, 2.0–12.5	1.0 (0.86–1.3)
Location of tumor			P = 1.0‡			P = 0.55‡
Choroid	66 (0.67)	8 (0.73)	Reference	58 (0.62)	17 (0.94)	Reference
Ciliary body	4 (0.040)	0	Not estimable	4 (0.043)	0	Not estimable
Cilio-choroid	27(0.27)	3 (0.27)	0.91 (0.21–4.0)	29 (0.31)	1 (0.06)	0.20 (0.024–1.7)
Iris	2 (0.020)	0	Not estimable	2 (0.022)	0	Not estimable
TNM tumor stage			P = 0.95‡			P = 0.34‡
Stage I	19 (0.20)	1 (0.09)	Reference	17 (0.19)	3 (0.17)	Reference
Stage IIA	22 (0.23)	3 (0.27)	2.7 (0.26–28.3)	15 (0.16)	10 (0.56)	4.8 (1.1–21.7)
Stage IIB	27 (0.28)	3 (0.27)	2.6 (0.20–32.6)	26 (0.29)	4 (0.22)	2.5 (0.41–15.5)
Stage IIIA	21 (0.22)	3 (0.27)	3.4 (0.25–45.4)	23 (0.25)	1 (0.056)	0.77 (0.064–9.4)
Stage IIIB	5 (0.052)	1 (0.09)	4.7 (0.20–110)	5 (0.066)	0	Not estimable
Stage IV	3 (0.031)	0	Not estimable	4 (0.044)	0	Not estimable
Chromosome 3			P = 0.032‡			P = 0.029‡
Disomy	33 (0.33)	7 (0.64)	Reference	27 (0.29)	13 (0.72)	Reference
Monosomy	66 (0.67)	4 (0.36)	0.20 (0.045–0.87)	66 (0.71)	5 (0.28)	0.26 (0.078–0.87)

Table 2

 

Continued

Tumor Variable	BAP1, n = 111
	WT, n = 55 (0.50)	Mutant, n = 56 (0.50)	OR (95% CI)*
Patient sex
Female	24 (0.44)	20 (0.36)	Reference
Male	31 (0.56)	36 (0.64)	1.1 (0.49–2.6)
Patient age, mo, median, range†	58, 24–84	61, 22-88	0.71 (0.17–3.0)
Basal diameter, mm, median, range	11, 5–19	14, 5–22	1.1 (0.94–1.2)
Thickness, mm, median, range	5.1, 1.0–12.7	8.2, 1.5–16.5	1.2 (1.0–1.4)
Location of tumor			P = 0.64‡
Choroid	45 (0.82)	30 (0.54)	Reference
Ciliary body	0	3 (0.05)	Not estimable
Cilio-choroid	10 (0.18)	21 (0.38)	1.9 (0.72–5.1)
Iris	0	2 (0.04)	Not estimable
TNM tumor stage			P = 0.14‡
Stage I	14 (0.26)	5 (0.09)	Reference
Stage IIA	20 (0.36)	4 (0.07)	0.42 (0.90–2.0)
Stage IIB	12 (0.22)	19 (0.35)	1.9 (0.45–8.1)
Stage IIIA	6 (0.11)	19 (0.35)	3.6 (0.76–17.2)
Stage IIIB	2 (0.04)	4 (0.07)	2.2 (2.6–19.4)
Stage IV	1 (0.02)	3 (0.06)	2.6 (0.18–37.2)
Chromosome 3			P < 0.001‡
Disomy	36 (0.66)	3 (0.05)	Reference
Monosomy	19 (0.34)	53 (0.95)	23.6 (6.3–88.2)

 Values in bold indicate statistically significant association.

*  OR and upper and lower 95%CI for association of gene mutation status (wild-type or mutant) with each variable after adjusting for metastatic status.

†  OR for age is based on natural log-transformed values.

‡  Overall P value from Wald test.

GNAQ and GNA11

A total of 52 (46%) of 113 tumors carried GNAQ mutations, with the two common GNAQ mutations (Gln209Leu/Pro) accounting for 46 of these mutations. Three additional GNAQ mutations were found in six tumors: Gln209Arg, Arg183Gln, and Arg183Tyr (Supplementary Table S1). The presence of GNAQ mutations was not associated with metastatic status (OR 0.99, 95% CI 0.47–2.1; Table 3). 

Table 3

 

Logistic Regression Analysis of Association of GNAQ, GNA11, SF3B1, and EIF1AX and BAP1 Mutations With Metastatic Status.

Gene	Mutation Status*	No. (Frequency)	Case/Control Status		Logistic Regression
			Controls: No Metastasis (Frequency)	Cases: Metastasis (Frequency)	OR (95% CI)	OR Adjusted (95% CI)†
GNAQ	WT	61 (0.54)	28 (0.54)	33 (0.54)	0.99 (0.47–2.1)	0.67 (0.24–1.9)
	Mutant	52 (0.46)	24 (046)	28 (0.46)
GNA11	WT	75 (0.65)	40 (0.75)	35 (0.56)	2.5 (1.1–5.5)	2.3 (0.75–7.3)
	Mutant	41 (0.35)	13 (0.25)	28 (0.44)
SF3B1	WT	99 (0.90)	46 (0.90)	53 (0.90)	1.0 (0.30–3.6)	0.56 (0.11–2.9)
	Mutant	11 (0.10)	5 (0.10)	6 (0.10)
EIF1AX	WT	93 (0.84)	36 (0.71)	57 (0.95)	0.13‡ (0.034–0.47)	0.13‡ (0.024–0.68)
	Mutant	18 (0.16)	15 (0.29)	3 (0.05)
BAP1	WT	55 (0.50)	36 (0.73)	19 (0.31)	6.3 (2.7–14.4)	3.6 (1.2–10.2)
	Mutant	56 (0.50)	13 (0.27)	43 (0.69)

 Values in bold indicate statistically significant association.

*  See Supplementary Tables S1 through S3 for detailed list of mutations.

†  OR adjusted for tumor diameter, thickness, and location shown to be significant in Table 1. TNM staging was not included in the adjusted model because the staging criterion was based solely on tumor size and location, which were already included in the model as separate variables.

‡  Odds ratio corresponds to a 7.9-fold decreased risk of metastasis in both the unadjusted and adjusted models.

The GNA11 mutations were present in 41 (35%) of 116 UMs and all but one of these mutations were Gln209Leu (Supplementary Table S1). There was a significant association of GNA11 mutation status with metastatic status (OR 2.5, 95% CI 1.1–5.5), but after adjustment for tumor characteristics, there was insufficient evidence to demonstrate statistical significance (OR 2.3, 95% CI 0.75–7.3; Table 3). The GNAQ and GNA11 mutations were present in a mutually exclusive pattern in a total of 93 (82%) of 113 tumors. 

SF3B1 and EIF1AX

Genotypes for two previously identified mutations in SF3B1, Arg625Cys, and Arg625His ²⁷ were determined in 110 tumors. Eight tumors carried the Arg625Cys and three carried the Arg625His mutation (Supplementary Table S1). All three Arg625His mutations were present in tumors with monosomy-3. Mutations at Arg625 were not significantly associated with metastatic status (OR 1.0, 95% CI 0.30–3.6; Table 3). 

Sixteen different mutations in exons 1 and 2 of EIF1AX were identified in 18 of 111 tumors (Supplementary Table S2). Fourteen missense changes, one splice site mutation, and one variant of unknown significance in the 5′ untranslated region were identified. Seven, including the splice site mutation, were in novel sites not previously described by Martin et al. ²⁸  

Among the 93 tumors without EIF1AX mutations, 57 metastasized and 36 did not (Table 3). A comparison of the frequency of tumors carrying an EIF1AX mutation that did, or did not, metastasize (3 [12%] of 18 vs 15 [84%] of 18) suggested that the presence of an EIF1AX mutation is protective. The effect of an EIF1AX mutation is estimated to be a decreased odds of metastasis (OR 0.13, 95% CI 0.034–0.47; Table 3), which remained significant in the model adjusted for tumor characteristics (OR 0.13, 95% CI 0.024–0.68). 

BAP1

A total of 52 unique mutations were identified in 56 (50%) of 111 tumors. These mutations included 13 missense changes, 8 nonsense, 12 5′- or 3′-splice site mutations located within 5 bp of the exon–intron boundary, 15 frameshift deletions leading to premature termination, 2 in-frame deletions, 1 complex in-frame in/del, and 1 tumor with deletion of exons 12 and 17 (Supplementary Table S3). To the best of our knowledge, the same mutations (or a different substitution at the same site) have been previously reported for only 6 of these 52 mutations. BAP1 mutations were associated with significantly increased odds of metastasis (OR 6.3, 95% CI 2.7–14.4; Table 3), which remained significant after adjustment for tumor characteristics (OR 3.6, 95% CI 1.2–10.2). 

Three tumors with mutations in BAP1 were disomic for chromosome 3. The mutations in UM11 (intron 6, c.438-2A>G abolished a 3′-splice acceptor site) and UM42 (exon 12, c.1147C>T, Arg383Cys, rs201844078) were present in heterozygous states, implying that a knockout of the WT allele had not occurred. These patients were free of metastasis at 48 months (and remained metastasis-free for 67 and 78 months, respectively). However UM13 developed metastases to the spleen within a year of initial treatment. The NGS reads of UM13 were reviewed and indicated 86% of the reads represented the mutant allele of the Arg60X mutation, implicating a copy neutral loss of heterozygosity event in the tumor. Nineteen of the 72 tumors with monosomy 3 did not carry a BAP1 sequence mutation. Of these, 10 developed metastasis and 9 did not. 

Chromosome-3 Interactions With EIF1AX and BAP1

We examined the joint effects of chromosome 3 status with EIF1AX or BAP1. We found that the combination of monosomy-3 with the EIF1AX-WT (risk) allele was significantly associated with metastasis (OR 29.7, 95% CI 3.6–244, Table 4) and remained significant after adjustment for tumor variables (OR 31.4, 95% CI 3.0–332). Tumors with monosomy-3 and either BAP1-WT or mutant alleles had significantly increased odds of metastasis (OR 3.3, 95% CI 1.0–10.8 and OR 11.5, 95% CI 4.2–31.3, respectively). 

Table 4

 

Multivariate Logistic Regression Analysis of Association of Chromosome (Chr) 3 Status and EIF1AX or BAP1 Mutations With Metastatic Status

Variables	Total (Frequency)	Case/Control Status		Logistic Regression
		Controls:	Cases:	OR (95% CI)	OR Adjusted (95% CI)*
		No Metastasis (Frequency)	Metastasis (Frequency)
Chr3-EIF1AX				P = 0.001	P = 0.021
Disomy, mutant†	13 (0.12)	12 (0.24)	1 (0.02)	Reference	Reference
Disomy, WT	27 (0.24)	17 (0.33)	10 (0.17)	7.1 (0.79–62.7)	10.1 (0.90–114)
Monosomy, mutant	5 (0.04)	3 (0.06)	2 (0.03)	8.0 (0.53–121)	38.2 (1.2–1225)
Monosomy, WT	66 (0.60)	19 (0.37)	47 (0.78)	29.7 (3.6–244)	31.4 (3.0–332)
Chr3-BAP1				P < 0.001	P = 0.015
Disomy, WT	36 (0.32)	27 (0.55)	9 (0.15)	Reference	Reference
Disomy, mutant	3 (0.03)	2 (0.04)	1 (0.02)	1.5 (0.12–18.6)	1.0 (0.072–14.3)
Monosomy, WT	19 (0.17)	9 (0.18)	10 (0.16)	3.3 (1.0–10.8)	4.1 (0.91–18.4)
Monosomy, mutant	53 (0.48)	11 (0.22)	42 (0.68)	11.5 (4.2–31.3)	7.6 (2.1–27.3)

 Values in bold indicate statistically significant association.

*  Odds ratio for chromosome 3 status and gene mutation combination adjusted for tumor diameter, thickness, and location.

†  Reference category is Chr3-disomy, EIF1AX-mutant corresponding to the two lower-risk alleles.

Finally, we investigated the presence of metastasis at 48 months when the effects of chromosome 3, BAP1, and EIF1AX status were considered jointly (Table 5). Tumors with monosomy-3/BAP1-WT/EIF1AX-WT, or monosomy-3/BAP1-mutant/EIF1AX-WT had significantly increased odds of developing a metastasis (OR 13.5, 95% CI 1.4–128 and OR 45.6, 95% CI 5.3–394, respectively) compared with tumors with the reference multilocus genotype of disomy-3/BAP1-WT/EIF1AX-mutant. Even after adjustment for tumor thickness, diameter, and location, these three loci combinations remained significant with high risk of metastasis (OR 19.1, 95% CI 1.5–251 and OR 37.5, 95% CI 3.4–414, respectively). The lowest risk of metastasis is associated with the reference variable, disomy-3/BAP1-WT/EIF1AX-mutant tumors. 

Table 5

 

Logistic Regression Analysis of Association of the Joint Effects of Chromosome 3 Status Combined With BAP1 and EIF1AX Allele Status With Metastasis

	Chromosome 3 Disomy				Chromosome 3 Monosomy
	BAP1 WT		BAP1 Mutant		BAP1 WT		BAP1 Mutant
	EIF1AX Mutant	EIF1AX WT	EIF1AX Mutant	EIF1AX WT	EIF1AX Mutant	EIF1AX WT	EIF1AX Mutant	EIF1AX WT
No.	13	23	0	3	2	17	2	48
Metastasis no/yes	12/1	15/8	—	2/1	1/1	8/9	1/1	10/38
Logistic regression
OR (95% CI)*	Reference	6.4 (0.70–58.5)	—	6.0 (0.26–140)	12.0 (0.38–375)	13.5 (1.4–128)	12.0 (0.38–375)	45.6 (5.3–394)
OR (95% CI)†	Reference	10.6 (0.92–123)	—	5.4 (0.19–153)	49.6 (1.2–2042)	19.1 (1.5–251)	13.4 (0.004–42892)	37.5 (3.4–414)

 Values in bold indicate statistically significant association.

*  Odds ratio and 95% CI intervals for joint effects model.

†  Odds ratio and 95% CI intervals adjusted for tumor thickness, diameter, and location.

Discussion

Uveal melanoma is a rare ocular tumor associated with significant morbidity. Hence, many affected individuals wish to be tested for the most sensitive prognostic marker(s). There are several traditional markers, including larger tumor size, tumor location, and chromosome 3 monosomy, that are well-established poor prognosticators for UM metastasis. ^4–15 The aim of this article was to evaluate the association of GNAQ, GNA11, SF3B1, EIF1AX, and BAP1 mutation profiles with metastasis while taking the prior known tumor characteristics into account with the goal of providing new genomic information that can enhance patient prognostic characterization. 

With this background, we used a case-control study design that included 63 UMs from patients with documented metastasis within 48 months of diagnosis and 53 UMs from patients whose tumors had not metastasized within the same time period. The distribution of tumor stages in our collection representing TNM stages I to IIIC was 18%, 23%, 28%, 22%, 5.3%, and 0%, respectively, which is similar to that found by Kujala et al. ¹⁷ In our sample set, the proportion of patients with metastases at diagnosis (TNM stage IV) was 3.5% (Table 1), which is within the generally accepted range of 1% to 4%. ¹⁶ Furthermore, the thickness and basal diameter of the tumors in our cohort and the percentage of tumors with ciliary body involvement are comparable with that found in a study of 452 UMs by Damato et al. ⁷ Thus, these data for clinical tumor characteristics provide substantial evidence that our sample set was representative of UM in general. 

This study was not designed to address the problem of potential tumor heterogeneity, which can be a function of tumor size. Maat et al. ⁵⁰ have shown for large tumors that often require enucleation, there is intratumor heterogeneity that can lead to imprecise molecular classification. In contrast, Onken et al. ⁵¹ have recently shown by gene expression profiling that there was little or no intratumor heterogeneity in either FNAB or enucleated tumors. This question remains unresolved and it could not be addressed in this study due to the lack of availability of multiple samples from the same tumor. Finally, there was a limitation in the number of available UMs with adequate follow-up time, complete information on metastatic status, and mutation status for the five genes. This led, in some instances, to a broad range in the 95% CIs, resulting in imprecise estimates of odds of metastasis, especially in the case of EIF1AX, where the frequency for the replace with mutated allele was small. 

In our study, logistic regression analysis indicated that the presence of mutations in GNA11, but not GNAQ, was associated with the development of metastasis at 48 months (Table 3). This finding of metastasis in many tumors carrying GNA11 mutations is relevant in light of the finding of Griewank et al., ⁵² where three times as many metastatic UMs carried GNA11 mutations compared with GNAQ. This led the authors to suggest that GNA11-mutant tumors may have a higher tendency to metastasize than GNAQ-mutant tumors. In our study of primary UM, the ratio of GNA11 to GNAQ mutations was 1.3. When combined with the previous observation, it suggests that cells carrying the GNA11 mutation are selected for in the metastatic process. This is supported in part by the finding that, although there was no significant difference in basal diameter between GNA11-WT and GNA11-mutant tumors, those carrying the GNA11 mutant were significantly thicker than GNA11-WT tumors (P = 0.032, data not shown). 

Harbour et al. ²⁷ and Furney et al. ²⁹ both reported a significant association between the presence of SF3B1 mutations and significantly improved survival in Kaplan-Meier survival analysis, and Martin et al. ²⁸ also reported a nonsignificant trend toward better prognosis. However, this association was not observed in our study, possibly due to differences in study design and methodology. 

Mutations in EIF1AX were shown to play a protective role in UM metastasis and, even after adjusting for the effect of other known risk factors, there was an 8-fold decreased risk of metastasis, which confirmed the trend noted by Martin et al. ²⁸  

Mutations in the chromosome-3p–linked tumor suppressor protein, BAP1 have been shown to be associated with metastasizing UM. ²³ Our study identified 52 different BAP1 mutations in 56 tumors. Forty-three (77%) of the tumors carrying BAP1 mutations metastasized, and of these, all but one was chromosome 3 monosomy, which is consistent with recessive BAP1 mutations being uncovered by the loss of one copy of chromosome 3. Nineteen of the 72 tumors with monosomy-3 did not carry a BAP1 sequence mutation, of which 10 developed metastases. For the nine UMs that did not metastasize within 48 months, it remains to be determined whether there are other, as yet unidentified, genes on chromosome 3 critical to the metastatic process that need to be lost in monosomy-3/BAP1-WT UM. 

When we evaluated the joint effects of chromosome 3 status with mutations in BAP1 and/or EIF1AX, it was difficult to provide precise estimates for the various multilocus genotype combinations, in part because the protective EIF1AX mutant allele was relatively rare. The OR for monsomy-3 tumors, without accounting for any other tumor characteristic of gene mutation, was 6.2 (95% CI 2.6–14.4, Table 1). In the joint effects model combining chromosome 3, EIF1AX, and BAP1 status, considerably more informative ORs were provided. We showed that the lowest risk of metastasis is associated with disomy-3/BAP1-WT/EIF1AX-mutant tumors, which served as the reference category. The tumors with disomy-3 and no mutations in EIF1AX or BAP1 had a 10-fold increased risk of metastasis at 48 months with respect to the reference category. Tumors with chromosome 3 monosomy and EIF1AX-WT alleles, irrespective of BAP1 status, were at a greater than 13-fold risk of developing metastases at 48 months, and if a BAP1 mutation was present, the risk increased to greater than 35-fold. 

In conclusion, the aim of this study was to determine the contribution of gene mutation profiles to metastatic outcome in UM. We have shown that although the most significant prognostic indicators remain the classic predictors, chromosome 3 status and tumor size and location, combining these factors with BAP1 and EIF1AX mutation status, adds considerably more information and significance to the stratification of individuals with respect to prognostic outcome. 

Supplementary Materials

pdf S1 

Acknowledgments

The authors thank Kim Moran for her technical assistance and Lindsey Mighion for her service in reviewing patient charts. 

Supported by development funds of Genetic Diagnostic Laboratory, Perelman School of Medicine, University of Pennsylvania (AG) and the Eye Tumor Research Foundation, Philadelphia, Pennsylvania, United States (CLS). 

Disclosure: K.G. Ewens, None; P.A. Kanetsky, None; J. Richards-Yutz, None; J. Purrazzella, None; C.L. Shields, None; T. Ganguly, None; A. Ganguly, None