Population-based analysis of BAP1 germline variations in patients with uveal melanoma

Abstract

Pathogenic germline variants in the BRCA1-associated protein 1 (BAP1) gene cause the BAP1 tumor predisposition syndrome (BAP1-TPDS) with increased risk of several cancers, the most frequent of which is uveal melanoma (UM). Pathogenicity of loss-of-function (LOF) BAP1 variants is clear, as opposed to that of missense and regulatory region variants. We sequenced the coding, promoter, untranslated region (UTR) and intronic regions of BAP1 and analyzed copy number variations (CNVs). In this nationwide study, the cohort comprised UM patients diagnosed between 2010 and 2017. These included 432 of 520 consecutive Finnish UM patients, 16 of whom were familial, and one additional patient from a Finnish–Swedish family. Twenty-one different rare variants were found: seven exonic, seven intronic, four 3′ UTR and three promoter. We considered five variants likely to be pathogenic by effect on splicing, nuclear localization or deubiquitination activity. Intron 2 (c.67+1G>T) and exon 14 (c.1780_1781insT) LOF variants were presumed founder mutations, occurring in two and four families, respectively; both abolished nuclear localization in vitro. Intron 2, exons 5 (c.281A>G) and 9 (c.680G>A) missense variants markedly reduced deubiquitinating activity. A deep intronic 25 base pair deletion in intron 1 caused aberrant splicing in vitro. On the basis of functional studies and family cancer history, we classified four exon 13 missense variants as benign. No CNVs were found. The prevalence of pathogenic variants was 9/433 (2%) and 4/16 (25%) in Finnish UM families. Family cancer history and functional assays are indispensable when establishing the pathogenicity of BAP1 variants. Deep intronic variants can cause BAP1-TPDS.

Introduction

Pathogenic variants in the BRCA1-associated protein 1 (BAP1) gene cause the BAP1 tumor predisposition syndrome (BAP1-TPDS, OMIM 614327) (1). The penetrance of this syndrome is high; 85% of patients with pathogenic germline variants have at least one cancer (2). The most common tumors in BAP1-TPDS are, in descending order, uveal melanoma (UM), malignant mesothelioma, clear cell renal cell carcinoma, cutaneous melanoma and meningioma (3). UM is reported in 36% of 141 proband and in 16% of 182 non-proband variant carriers (3). We and the others have shown that ~2% of unselected patients with UM harbor a pathogenic germline variant in BAP1 (4,5). The identification of such variants helps to guide surveillance of these patients and relatives who are variant carriers, to detect second cancers as early as possible.

Loss-of-function (LOF) mutations, small insertions and deletions as well as non-synonymous single nucleotide polymorphisms (SNPs) are the most common germline variations described in BAP1 (3). In well-studied cancer predisposition syndromes such as BRCA1- and BRCA2-associated hereditary breast and ovarian cancer syndrome (OMIM 604370 and 612555) and Lynch syndrome (OMIM 120435), larger genomic arrangements contribute ~10% of the pathogenic variants (6,7). Only one whole gene deletion has been reported so far in BAP1-TPDS (3,8). This may be related to the limitations of Sanger sequencing, which is not the best method for detecting larger genomic copy number variations (CNVs). Furthermore, deeper intronic variants that affect splicing and promoter variants that alter gene expression are missed when mutation analysis is restricted to the coding regions.

Germline null variants in BAP1 are straightforward to classify as pathogenic; however, uncertainty surrounds the functional impact and pathogenicity of missense variants (3). There is a need to investigate more families with BAP1-TPDS and to perform functional analysis of mutations otherwise classified as ‘variants of unknown significance (VUS)’. These studies may reveal functions of BAP1 that are relevant to tumorigenesis. It is recommended that all VUS be evaluated beyond in silico predictions by methods such as functional assays and detecting segregation of core rare tumors in individual families (3).

BAP1 protein is a deubiquitinating (DUB) enzyme that predominantly localizes to the nucleus. It cleaves covalently attached ubiquitin from target substrates through its N-terminal ubiquitin C-terminal hydrolase (UCH) domain. BAP1 participates in the formation of multiprotein complexes and interacts with transcription factors such as HCF1, ASXL1, ASXL2, BARD1, BRCA1, YY1, FoxK1 and FoxK2 (9–13). Through these interactions, BAP1 regulates several cellular pathways, relating to cell cycle, cellular differentiation, transcription and DNA damage response (14). Deubiquitinating activity and nuclear localization are both required for BAP1-mediated tumor suppression (15). Loss of enzymatic activity has promoted tumorigenesis, and deubiquitination has been directly linked to tumor suppression through histone H2A modifications on chromatin that influence gene expression and cellular response to apoptotic signals (15–17). BAP1 also auto-deubiquitinates its own nuclear localization signal (NLS), thereby regulating its nuclear compartmentalization (18).

We have screened germline BAP1 variants in 432 consecutive, unselected Finnish patients with UM and in 1 Swedish patient from a Finnish–Swedish UM family, the proband of which had died. We also screened for CNVs using multiplex ligation-dependent probe amplification (MLPA) and sequenced the whole genomic region of BAP1 to detect regulatory region and intronic variants. Variants predicted as pathogenic by computer modeling were studied further for their deubiquitinating activity and nuclear compartmentalization, the two major functions required from functional BAP1 protein (15). The aim of our study was to determine the role of BAP1 variations in the genesis of UM.

Results

Of the 432 enrolled patients, 238 (55%) were male (Supplementary Material, Table S1). The median age of our cohort was 65 years (range, 15 to 90); 371 (86%) patients were ≥50 years. Sixteen (4%) patients had a family history of UM, and one had bilateral UM. Two (0.5%) patients had a history of cutaneous melanoma and four (1%) of renal cell carcinoma. None of our patients had a history of mesothelioma. The UM was purely choroidal in 318 (73%) patients, confined to the iris in 16 (4%) and involved the ciliary body in 99 (23%). Eleven (3%) tumors showed extraocular extension. According to the tumor–node–metastasis (TNM) classification, 130 (30%) of the tumors were stage I, 187 (43%) stage II, 92 (22%) stage III and 8 (2%) stage IV at the time of diagnosis (Supplementary Material, Table S1). The 88 eligible patients who were not enrolled were older (median, 70 vs. 65 years; P < 0.001, Kruskal–Wallis) and had a larger tumor of higher TNM stage (P = 0.002, non-parametric test for trend; Supplementary Material, Table S1). None of the non-enrolled patients had a family history of UM, but two (2%) had a history of cutaneous melanoma.

Figure 1

Schematic representation of BAP1 with identified variants, functional domains and interacting regions. BAP1 is located at the short arm of chromosome 3 (3p21.1), spans 17 exons and translates to 729 amino acids. (A) Identified variants, missing from databases or with the gnomAD frequency under 0.1% in the Finnish population, are presented. Variants studied in vitro are marked with bars. Yellow bars represent likely pathogenic intronic variations, red ones likely pathogenic missense variants, blue ones likely benign missense variants and purple one a truncating mutation. (B) N-terminal UCH domain (aa1-240) is responsible for deubiquitinating activity and has four key residues (Q82, C91, H169 and D184) that are marked with purple circles. BAP1 binds cell cycle regulator host cell factor 1 (HCF1, aa365-385) through HCF1-binding domain (HBM)-like motif (aa363-366). Interaction sites for BARD1 (aa182-365), FoxK1/K2 (aa477-526), BRCA1 (aa596-721) and YY1 (aa642-686) are indicated. Two NLSs (aa656-661 and aa717-722) and a UCH37-like domain (ULD, aa634-693) responsible of interaction with ASXL1/2 are located in C-terminus of BAP1.

Rare variants in BAP1

Sanger sequencing of all 17 exons, adjacent intronic regions, promoter, 5′ (untranslated region) UTR and 3′ UTR of germline BAP1 was performed in 430 Finnish UM patients and 1 Swedish patient from the Finnish–Swedish UM family originating in Germany. In two patients, analysis was not complete because of technical reasons. The entire intronic regions were sequenced using targeted amplicon sequencing in 272 patients. Twenty-one rare variants (minor allele frequency < 0.1% in Finns in gnomAD) were identified in 22 patients (Fig. 1 and Tables 1 and 2). Two previously described pathogenic variants (c.67+1G>T and c.1780_1781insT) (4) were found in six patients. These variants are not described in gnomAD database. Six patients had heterozygous missense variants (c.281A>G, c.680G>A, c.1339G>A, c.1526C>T, c.1689C>A and c.1727C>T). We also identified six rare intronic (Fig.1 and Table 1), three promoter and four 3′ UTR variants (Fig. 1 and Table 2).

Truncating insertion variant c.1780_1781insT

Four patients carried in exon 14 the previously reported insertion variant c.1780_1781insT (p.Gly594Valfs^*49), which alters the reading frame, leading to a premature stop codon (4). Updated pedigrees of proband UMG10-66 and UMG11-51, both with a family history of UM and one other BAP1-TPDS core cancer (renal cell carcinoma or mesothelioma), are shown in Figure 2A (FUMH-03 and FUMH-08). Two other patients, UMG15-52 and UMG16-61, carried the same insertion variant. UM and gastric cancer developed in the mother of patient UMG15-52, who was diagnosed with UM (T1a, stage I) at the young age of 35 years (Fig. 2A; FUMH-06). Patient UMG16-61 was treated for UM (T3b, stage IIIA) at the age of 79 years and gave a positive family history of mesothelioma (Fig. 2A; BUMH-02).

Splice-site variant c.67+1G>T

Two patients, UMG12-01 and UMG15-43, had the previously reported splice-site variant c.67+1G>T in intron 2 (4). Patient UMG12-01, who was reported earlier, had an updated pedigree showing family members with all four BAP1-TPDS core cancers (Fig. 2B [BUMH-01]). Patient UMG15-43, diagnosed with UM (T3b, stage IIIA) at the age 62 years, had a father who had UM (T2a, stage IIA) diagnosed at the age of 50 years and a grandfather with lung cancer (Fig. 2B; FUMH-05).

The effect of the c.67+1G>T mutation on splicing was determined using mRNA from patient UMG12-01. BAP1 was amplified and sequenced from cDNA. Polymerase chain reaction (PCR) showed two bands in gel electrophoresis (Supplementary Material, Fig. S1A). The shorter fragment corresponded to the normally spliced product (255 bp), detected also in a healthy control. Sequencing of the isolated bands confirmed the mis-splicing of intron 2 (Supplementary Material, Fig. S1B), causing it to be included in the mRNA transcript. Intron 2 of BAP1 is 105 bp long, translating to 35 amino acids; therefore, the inclusion of intron 2 stays in frame. In addition, the entire BAP1 was cloned from cDNA to plasmid and sequenced through. This confirmed the transcript to be normal except for the intron 2 inclusion (data not shown).

Non-synonymous variants

Five Finnish patients and the Swedish patient had five different non-synonymous, heterozygous missense variants in exon 5, 9 and 13 (Table 1). Two of these variants, c.281A>G in exon 5 and c.680G>A in exon 9, reside in the large catalytically active UCH-domain of BAP1.

Table 1

Exonic and rare intronic BAP1 variants

Patient	Family	Region	Chromosomal (grch37, hg19)	CDS position NM_004656.3	RS number	Protein	gnomAD (Fin)	gnomAD (all)	COSMIC	Mutation taster	SIFT	PROVEAN	CADD	Human splicing finder
UMG11-53	BUMH-03	Intron 1	chr3:52443810_52443834	c.37+24_37+48delAGGGGCT GGGGAGGCC GGATGGGCC	—	—	NA	NA		Disease causing	NA	NA	NA	Probably no impact on splicing. Potential branch point broken.
UMG15-43b	FUMH-05	Intron 2	chr3:52443729	c.67+1G>T	—	—	NA	NA		Disease causing	NA	NA	27.2	Alteration of the WT donor site, most probably affecting splicing.
UMG12-01a,b	BUMH-01a
UMG16-12	—	Intron 4	chr3:52442449	c.255+41G>C	rs200961637	—	0.00049	0.000043		Polymorphism	NA	NA	16.3	Probably no impact on splicing.
UMG16-12	—	Intron 4	chr3:52442448	c.255+42C>T	rs201679991	—	0.00049	0.000047		Polymorphism	NA	NA	18.7	Probably no impact on splicing.
UMG00-R1b	FUMH-16	Exon 5	chr3:52442068	c.281A>G	—	H94R	NA	NA	COSM6925842	Disease causing	Damaging	Deleterious	25.3	Alteration of an exonic ESE site. Potential alteration of splicing.
UMG17-35	BUMH-04	Exon 9	chr3:52440372	c.680G>A	—	R227H	NA	NA	COSM238566	Disease causing	Tolerated	Deleterious	29.3	Alteration of an exonic ESE site. Potential alteration of splicing.
UMG10-29	—	Intron 11	chr3:52439052	c.1116+74T>C	rs1171839949	—	0.00029	0.000032		Polymorphism	NA	NA	12.9	Alteration of an intronic ESS site. Probably no impact on splicing.
UMG11-28	—	Intron 12	chr3:52437944	c.1251-34G>A	rs184055262	—	0.000084	0.000028		Polymorphism	NA	NA	6.78	Probably no impact on splicing.
UMG16-46	UMH-01	Exon 13	chr3:52437822	c.1339G>A	rs762654322	V447I	NA	0.000024		Disease causing	Damaging	Neutral	24.9	Probably no impact on splicing.
UMG13-60	UMH-02	Exon 13	chr3:52437635	c.1526C>T	—	S509L	NA	NA		Disease causing	Damaging	Neutral	27.9	Alteration of an exonic ESE site. Potential alteration of splicing.
UMG11-25b	UMH-03	Exon 13	chr3:52437472	c.1689C>A	rs760711001	H563Q	0.000093	0.000008		Disease causing	Tolerated	Neutral	13.78	Alteration of an exonic ESE site. Potential alteration of splicing.
UMG10-66a,b	FUMH-03a	Exon 13	chr3:52437434	c.1727C>T	rs374920141	T576I	0.00014	0.000064		Disease causing	Tolerated	Neutral	24.6	Alteration of the WT donor site, most probably affecting splicing.
UMG16-61b	BUMH-02	Exon 14	chr3:52437263_52437264	c.1780_1781insT	—	G594Vfs*49	NA	NA		Disease causing	NA	NA	NA	Activation of an exonic cryptic donor site. Alteration of an exonic ESE site. Potential alteration of splicing.
UMG15-52b	FUMH-06
UMG10-66a,b	FUMH-03a
UMG11-51a,b	FUMH-08a
UMG12-06	—	Intron 14	chr3:52436978	c.1891-91C>T	rs781565390	—	0.00046	0.0001		Polymorphism	NA	NA	5.29	Creation of an intronic ESE site. Probably no impact on splicing.

NA, not available

Identified exonic and intronic variants with the Finnish gnomAD frequency under 0.1% are presented.

^aPreviously published.

^bPreliminary listed in Walpole et al. (3).

The Swedish patient UMG00-R1, who had undergone screening for BAP1 mutations in Sweden and had at that time been interpreted not to have any, was found to have the exon 5 c.281A>G (p.His94Arg) variant not described in gnomAD but listed in one clear cell renal cell carcinoma in the Catalog of Somatic Mutations in Cancer (COSMIC) database. The variant is predicted to be pathogenic by three prediction programs (Table 1). The patient has a strong family history of BAP1-TPDS core cancers, including UM in a maternal uncle at the age of 59 years, cutaneous melanoma and mesothelioma (Fig. 2C, FUMH-16).

Patient UMG17-35 had an exon 9 variant c.680G>A (p.Arg227His) not listed in gnomAD but with two entries in COSMIC (squamous cell lung carcinoma, prostate adenocarcinoma cell line). The majority of prediction programs predicted it to be pathogenic (Table 1). The patient was diagnosed with UM (T1a, stage I) at the age of 56 years and had been treated for breast cancer. Her sister had cutaneous melanoma, and several relatives had non-core-BAP1-TPDS cancers (Fig. 2D; BUMH-04).

We identified three different exon 13 missense variants in three patients without a family history of UM or mesothelioma. Patient UMG16-46 had a variant c.1339G>A (p.Val447Ile, rs762654322), which has an overall frequency of 0.000024 in gnomAD without representation in the Finnish population (Table 1). He had been diagnosed with UM (T2a, stage IIA, age 67; Fig. 2E, UMH-01). Patient UMG13-60 had a variant c.1526C>T (p.Ser509Leu) not listed in gnomAD. He was diagnosed with UM (T1a, stage I, age 83 years; Fig. 2F, UMH-02). Patient UMG11-25 had a variant c.1689C>A (p.His563Gln, rs760711001). This variant is multiallelic and found in gnomAD only in Finnish population, with a frequency of 0.000093. He was diagnosed with UM (T3a, stage IIB, age 65 years; Fig. 2G, UMH-03). UMG10-66, who had the c.1780_1781insT variant, carried in exon 13 an additional missense variant c.1727C>T (p.Thr576Ile) (4). The variant was considered likely to be benign because his affected son UMG08-58 carried only the insertion variant; the missense variant thus did not segregate with the disease. The prediction programs gave conflicting results for the exon 13 variants, which were not found in COSMIC or TCGA (Table 1).

Deubiquitinating activity of rare BAP1 variants

To study the effects of likely pathogenic coding region variants on deubiquitinating activity, nine vectors including two wild type controls (WT1 and WT2) were constructed and used to produce glutathione S-transferase (GST)–BAP1 fusion proteins in bacteria. The N-terminal UCH domain-located intron 2 splice-site mutant c.67+1G>T (‘intron 2 mutant’), exon 5 missense mutant c.281A>G (p.His94Arg, in the middle of the domain) and exon 9 missense mutant c.680G>A (p.Arg227His, at the far 3′ end of the domain) showed significant reduction in their activity to cleave fluorogenic ubiquitin–AMC substrate (Fig. 3A,B). The intron 2 mutant retained only 16% (P < 0.001, Welch’s t-test) of its activity when compared to WT1 (Fig. 3C). The activity of the His94Arg construct was reduced to 12% (P < 0.001). The Arg227His missense mutant reduced the activity to half (50%, P < 0.001). A minor difference in activity was seen between WT1 and WT2; the latter showed 24% (P = 0.011) higher deubiquitinating activity. Activities of exon 13 missense variants Val447Ile and His563Gln were somewhat higher when compared to WT1 (P = 0.043 and P < 0.001, respectively) but not when compared to WT2 (P = 0.94 and P = 0.24, respectively). The third exon 13 missense Ser509Leu and frameshift mutant G594Vfs^*49 activities did not differ from the wild types (Fig. 3C).

Table 2

Rare BAP1 variants in promoter and UTR regions

Patient	Region	Chromosomal (grch37, hg19)	CDS position NM_004656.3	RS number	gnomAD (Fin)	gnomAD (all)	Mutation taster	CADD
UMG10-58	promoter	chr3:52444371	c.-477delT	—	0.0	0.00009	NA	NA
UMG12-58	promoter	chr3:52444196	c.-302T>C	—	NA	NA	Polymorphism	NA
UMG12-58	promoter	chr3:52444190_52444191	c.-297_-296insCCCTCCCCTTCGCCCCCGTCCCGT	—	NA	NA	Polymorphism	4.713
UMG12-03	3′UTR	chr3:52436259	c.*45C>G	rs56898787	0.00031	0.012	Polymorphism	0.604
UMG11-44	3′UTR	chr3:52436006	c.*298G>C	rs1306377306	0.00058	0.000096	Disease causing	NA
UMG17-05	3′UTR	chr3:52435837	c.*467G*A	—	NA	NA	Polymorphism	10.53
UMG13-09	3′UTR	chr3:52435713_52435714	c.*590_*591insCAG	—	NA	NA	Polymorphism	11.58

Identified promoter and UTR variants with the Finnish gnomAD frequency under 0.1% are presented.

Figure 2

Pedigrees of patients with exonic and splice site variants. Patients affected by UM are shown as red, individuals affected by other cancers as black. BAP1-TPDS-related cancers are underlined. BAP1+ = BAP1 variant present; BAP1– = BAP1 variant absent; ca = carcinoma; CM = cutaneous melanoma; MM = malignant mesothelioma; y = years of age at diagnosis.

Figure 3

BAP1-TPDS-associated variants affect BAP1 function through deubiquitinating activity or nuclear localization. (A) GST–BAP1 fusion proteins, harboring the nine indicated coding region and splice site variants, were produced in bacteria, purified and subjected to deubiquitination assay. Empty GST vector was included as a control. A total of 100 ng of purified protein was incubated with 0.1 m fluorogenic substrate Ub-AMC. Hydrolysis of substrate by enzymatically active BAP1 resulted in increasing fluorescence signal measured at 440 nm for 40 min. BAP1-TPDS-associated mutants, the intron 2 splice site c.67+1G>T (+intron 2) and the exon 5 c.281A > G (H94R) missense, abolish enzymatic activity almost entirely. Exon 9 c.680G>A missense (R227H) reduces the activity by half whereas other variants do not have a significant effect on the activity when compared to both two wild type proteins. (B) Immunoblot of BAP1 from purified fusion proteins show equal protein loading to deubiquitination assay, except for exon 14 insertion mutant (G594Vfs^*49). (C) Bars represent the relative activity of each mutant. Endpoint fluorescence measurements were weighted according to protein loading seen in immunoblot and normalized to the wild type (WT1) (mean ± standard deviation of well mean normalized to the mean of WT1 of each experiment). Data were combined from three individual experiments each done using sample triplicates. P-values indicate the significance level of observed differences according to Welch’s t-test. (D) Collage of representative images. Transiently transfected NCI-H226 cells with different mutants were immunostained for BAP1. No endogenous staining can be seen in the empty vector control. Intron 2 mutant and exon 14 insertion (G594Vfs^*49) reduce the nuclear BAP1 localization. (E and F) Quantification of nuclear and cytoplasmic fluorescence intensities and their ratios was done using CellInSight from NCI-H226 cells transfected with constructs harboring indicated BAP1 variants. Experiment was repeated three times using sample triplicates. Statistical significance was calculated using Welch’s two sample t-test. (E) Box plots of nuclear to cytoplasmic intensity ratios. Mean ratio of each well was normalized to the mean of WT1 of each experiment. Mean ratio for a well is the mean of ratios determined on single-cell level in that well. (F) Mean nuclear and cytosolic fluorescence intensities. Mean nuclear and cytoplasmic fluorescence intensities in each well were normalized to the wild type cytoplasmic intensity (mean ± standard deviation well means normalized to the mean of WT1).

Nuclear localization of rare variants

All mutants and controls used in the deubiquitination assay were cloned to a mammalian vector, then transiently transfected and immunostained for BAP1. Cells with intron 2 and G594Vfs^*49 mutants showed loss of nuclear staining in fluorescent microscopy images (Fig. 3D). Staining of wild-type BAP1 was in concordance with previous publications with a prominent nucleus staining and diffuse faint cytoplasmic staining (15). Quantification showed significant reduction in the ratio of nuclear-to-cytoplasmic BAP1 in the catalytically inactive intron 2 mutant (P < 0.001, Welch’s t-test) and in the G594Vfs*49 mutant (P < 0.001) lacking N-terminal nuclear localization signals when compared to WT1 and WT2 (Fig. 3E). Change in the ratio occurred because of a reduction of BAP1 in the nucleus (Fig. 3F). Accumulation of BAP1 in the cytosol was not observed. The nuclear-to-cytoplasmic BAP1 ratios of the three exon 13 variants were reduced compared to WT1 (Val447Ile, P = 0.0027; Ser509Leu, P = 0.027; His563Gln, P = 0.002) but not to WT2 (Val447Ile, P = 0.59; Ser509Leu, P = 0.87; His563Gln, P = 0.36) and, therefore, were considered to be within normal limits (Fig. 3E).

Copy number variations

Specimens from all but 3 of the 433 patients were analyzed with MLPA to detect CNVs. The two excluded patients were found to have the exon 14 insertion mutation prior to MLPA screening, and the third sample did not have enough DNA. No large germline deletions or duplications were found in BAP1. Five samples gave inconclusive results because of low DNA quality.

Intronic variations

Deep intronic regions were analyzed with targeted amplicon sequencing in 272 patients. Identified rare variants (gnomAD, frequency in Finns < 0.001) are shown in Table 1 and more common ones in Supplementary Material, Table S2. Sequencing of patient UMG11-53 revealed a heterozygous 25 bp deletion (c.37+24_37+48delAGGGGCTGGGGAGGCCGGATGGGCC; Table 1, Fig. 1A and Fig. 4A) in intron 1. This deletion was not found in any of the public databases we consulted. According to MutationTaster and Human Splicing Finder, the change might alter splicing by affecting loop formation in a branch point. The variant was not identified with Sanger sequencing in the remaining 161 patients not included in the targeted sequencing. Patient UMG11-53 had UM (T3a, stage IIB, age 74 years) and died of metastases but had no family history of cancer (Fig. 2H, BUMH-03). Because no RNA was available, an in vitro splicing assay was conducted. A minigene ranging from exon 1 to exon 3 of BAP1 (Fig. 4B) was cloned from genomic DNA to an expression vector. PCR amplification confirmed aberrant splicing. This minigene produced three differently sized products, none of which corresponded to the wild type in size (Fig. 4C). Sequencing of isolated bands confirmed that no normally spliced product containing exons 1, 2 and 3 were produced (data not shown).

Figure 4

Minigene assay. (A) UMG11-53 has a heterozygous 25 bp deletion in intron 1 visualized using Integrative Genomics Viewer. (B) Minigene constructs containing the genomic region of BAP1 between first and third exon were generated. Vector with the 25 bp intron 1 deletion (c.37+24_37+48delAGGGGCTGGGGAGGCCGGATGGGCC) was cloned from genomic DNA of UMG11-53 (indicated c.37+24_37+48del). Intron 2 splice site mutant (c.67+1G>T) with established effect on splicing was cloned from UMG12-01 DNA. Control with the normal sequence was also cloned (ctrl). (C) RNA was isolated from cells transfected with minigenes (c.37+24_37+48del, c.67+1G>T, ctrl). RNA from patient UMG12-01 blood was included as splicing control, since mis-splicing of intron 2 is already established in vivo (Supplementary Material, Fig. S1). RNA from cells with WT1 construct containing full-length BAP1 cDNA sequence was used as a no splicing control. cDNA generated from RNA was subjected to PCR reaction ranging from exon 1 to exon 3. Agarose gel electrophoresis shows that no product corresponding to the wild type is generated from either minigene with an intronic mutation.

Patient UMG16-12 was homozygous for two adjacent rare variants in intron 4 (rs200961637, rs201679991). The change CG>GA is predicted in silico to be likely benign (Table 1). According to gnomAD, the frequency of both variants in the Finnish population is 0.00049. Based on the frequency of variants, clinical information and family data, this patient is not suspected of having BAP1-TPDS. Three other rare intronic variants were heterozygous and located in intron 11 (chr3:52439052 A>G) in UMG10-29, intron12 (rs184055262, chr3:52437944 C>T) in UMG11-28 and intron 14 (rs781565390, chr3:52436978G>A) in UMG12-06. They were considered likely to be benign based on in silico predictions, frequency in gnomAD and lack of core BAP1-TPDS tumors in the family (Table 1).

Variants in the predicted promoter region and UTRs

Sequencing analysis was extended to UTRs and the promoter region 400 bp upstream of 5′ UTR. Analysis revealed one base pair deletion (chr3:52444371, c.-477delT, Table 2) within the promoter region predicted by the Eukaryotic Promoter Database (EPD) for BAP1 (promoter ID: BAP1_1) in UMG10-58 found in gnomAD with a frequency of 0.00007. The family history included no BAP1-TPDS core cancers. The variant was considered likely to be benign.

Patient UMG12–58 had a T>C transition at c.-302 (chr3:52444196) and an insertion CCCTCCCCTTCGCCCCCGTCCCGT at c.-297_-296 (chr3:52444190_52444191, Table 2). The promoter fragment was cloned to a plasmid and sequenced. Both variants were in the same chromosome. The patient also had breast cancer but no family history of BAP1-TPDS core cancers. The variant was considered likely to be benign.

Rare heterozygous 3′ UTR variants were found in UMG12-03 (chr3:52436259G>A, rs56898787), UMG11-44 (chr3:52436006 C>G), UMG17-05 (chr3:52435837 C>T) and UMG13-09 (chr3:52435713_52 435714insCTG). These were considered benign based on in silico predictions or relatively high gnomAD frequencies (Table 2).

The frequency of pathogenic BAP1 variants

Eight Finnish patients with UM carried likely pathogenic variants in BAP1 (six had LOS, one had non-synonymous variant with reduced deubiquitinase activity and one had an intronic deletion that affected splicing; Fig. 1A and B, Supplementary Material, Table S3). The overall frequency of BAP1 pathogenic variants was 1.9% (8/432; 95% confidence interval [CI], 0.8–3.6). The frequency of these variants in patients <50 years of age was 3.2% (2/62; 95% CI, 0.3–11.2). A likely pathogenic variant was detected in 25% (4/16; 95% CI, 7–52) of Finnish families with a history of UM. If the Finnish–Swedish family is counted as well, this percentage was 31%.

Discussion

Our comprehensive genomic analysis of BAP1 identified 21 rare variants in 432 consecutive, consenting Finnish patients with UM and in 1 Swedish patient from a Finnish–Swedish UM family. We interpreted five of these variants to be pathogenic or likely pathogenic based on family history of BAP1-TPDS core cancers and in vitro studies. LOF variants are the most commonly reported pathogenic variants in BAP1 (3). Both variants that we identified in more than one patient caused LOF (c.1780_1781insT and c.67+1G>T, four and two patients, respectively). These are presumed founder mutations in Finland. Two other likely pathogenic variants found in one patient each were missense (c.281A>G and c.680G>A) and resided in the UCH domain of BAP1. Pathogenic variants in the promoter, UTRs or deep intronic regions of BAP1 were rare (1/432; 0.23%; 95% CI, 0.01 to 1.3); we found only one (c.37+24_37+48delAGGGGCTGGGGAGGCCGGATGGGCC). No larger CNVs were detected, and only one is reported in the previous literature (8).

A problem with cancer predisposition syndromes is that interpretation of missense variants, typically classified as VUS, is complex (19). When classifying BAP1 variants as either benign or pathogenic, a family history of core BAP1-TPDS cancers remains a key criterion. However, this approach is misleading if the variant occurs de novo, if no close relatives have a history of cancer, if the family history is incomplete or inaccurate or if there is a phenocopy in the family (20). When studying a rare cancer syndrome, such as BAP1-TPDS, data pooling is a valuable approach to interpreting variants and estimating cancer risk (3). Widely used prediction programs give informative suggestions about the nature of variants, but as our data also show, they are not always accurate. Therefore, development of reliable in vitro functional assays is necessary (3).

None of the coding region variants we identified could immediately be interpreted as benign. Although it is less likely that variants outside the UCH domain would affect the deubiquitinating activity of BAP1, e.g. through altered protein conformation, we studied all potentially pathogenic coding region and splice site variants in vitro. In addition to the enzymatic activity we assayed their nuclear localization because both functions are necessary for BAP1-mediated tumor suppression (15). Our analyses revealed four likely pathogenic variants with differing functional characteristics. One abolished both enzymatic activity and nuclear compartmentalization (intron 2 mutant, c.67+1G>T). Two others, which resided in the UCH domain (Fig. 1A and B), diminished enzymatic activity to different extent but did not significantly alter nuclear compartmentalization (missense mutation in exon 5, c.281A>G, and exon 9, c.680G>A). The fourth pathogenic variant retained normal enzymatic activity but abolished nuclear localization (insertion mutation in exon 14, c.1780_1781insT). Quantification of nuclear BAP1 staining associated with the exon 5 and 9 mutants in our overexpression in vitro system showed skewed distributions in the nuclear to cytoplasmic fluorescence ratios. It is important to note that BAP1 participates in the regulation of its nuclear compartmentalization in vivo by deubiquitinating its own nuclear localization signals (18). Loss of enzymatic activity therefore leads to reduced presence of BAP1 in the nucleus (18). UMs with monosomy 3 and somatic missense mutations in the remaining BAP1 allele residing in the UCH domain (c.660C>G or c.675A>G) have thus shown loss of nuclear BAP1 staining in surgical specimens (21).

Although DUB activity has been demonstrated also in the cytosol (17), our data in combination with family histories support the conclusion that enzymatically active BAP1 in the nucleus is needed for its function as a tumor suppressor (15). A functional protein restricted to the cytosol and a non-functioning or malfunctioning protein in the nucleus, both led to BAP1-TPDS in our UM families.

The exon 5 (c.281A>G, His94Arg) and 9 (c.680G>A, Arg227His) missense changes and other variants affecting the same amino acids are listed in the COSMIC database: His94Asp (c.280C>G) in one lung carcinoma, His94Arg (c.281A>G) in one renal cell carcinoma, Arg227Cys (c.679C>T) in ovarian and prostate carcinoma, Arg227His (c.680G>A) in lung cancer and a prostate carcinoma cell line and Arg227Pro (c.680G>C) in urinary tract and lung cancers. This supports the hypothesis that these missense variants are cancer associated.

The three rare exon 13 missense variants that gave varying in silico predictions were not confirmed as pathogenic by our assays. We cannot completely exclude, however, that these variants might affect splicing or disrupt some of the protein interactions of BAP1. For example, the variant c.1526C>T locates at a phosphorylation site (22) and within a FoxK1/K2 binding site (23). Further studies are needed to address these possibilities.

We detected one deeper intronic variant (c.37+24_37+48delAGGGGCTGGGGAGGCCGGATGGGCC) using targeted amplicon sequencing and showed altered splicing in vitro using a minigene assay. Similarly, CNVs do not seem to be prevalent in BAP1-TPDS. We did not discover any such variations, and only one whole gene deletion is reported in the literature (3,8); however, most published studies did not address larger deletions or duplications.

A limitation of our study is that we measured only two, albeit essential, functions of BAP1. We did not address tumorigenesis directly, because no established model has been developed to reliably evaluate the effect of BAP1 variants in vitro. Because no primary tumor samples were available from patients carrying the potentially pathogenic variants, we could not evaluate loss of heterozygosity of BAP1 in tumor cells.

The overall frequency of likely pathogenic variants in BAP1 among Finnish patients with primary UM remains 1.9%, and their frequency in familial UM was ~25%, consistent with our earlier report (4), in spite of comprehensive analysis that now also included other than coding variants. Although deep intronic and other non-coding variants, as well as CNVs, seem to be rare in BAP1, these aberrations should be examined if the family history suggests BAP1-TPDS and if no coding pathogenic variants are detected. Interpretation of BAP1 missense variants cannot rely only on in silico prediction programs; experimental validation is also needed. Finally, to reach a consensus on mandatory experimental assays to assess VUS, further research into BAP1 functions is needed.

Materials and Methods

Patients

Patients were eligible for inclusion in this study if diagnosed with UM between January 1, 2010, and December 31, 2017, at the Ocular Oncology Service, Helsinki University Hospital, a national referral center for eye cancer in Finland. In addition, we studied one Swedish patient (UMG00-R1) whose deceased Finnish relative had UM. The project was approved by the institutional review board of the Hospital Region of Helsinki and Uusimaa on June 11, 2012, and followed the tenets of the Declaration of Helsinki. Patients diagnosed before November 5, 2012, were invited to participate retrospectively. Patients were excluded if they were not Finnish or with Finnish ancestry. We obtained written informed consent from all participants.

We diagnosed 533 patients with primary UM and excluded 12 patients because they had no Finnish ancestry. Sixteen patients did not respond to our invitation, whereas 17 elected did not to participate. Another 50 patients died either before they received our invitation or before they donated their sample, 33 of these metastases from UM and 4 from a second cancer, confirmed histologically (colorectal carcinoma, thyroid carcinoma, oat cell lung carcinoma and malignant glioma). We collected blood from the remaining 433 patients. The inclusion rate was 83% in all 521 eligible patients and 86% in the 504 patients who could be invited to participate beginning November 5, 2012.

Statistical analyses

Comparisons between the enrolled and excluded eligible patients were performed with STATA (v15.1; Stata Corp., College Station, TX). We used Fisher’s exact test and nonparametric test for trend (24) to compare unordered and singly ordered contingency tables, respectively, and the Kruskal–Wallis test to compare continuous variables. All tests were two tailed, and P < 0.05 was taken as statistically significant. Welch’s t-test, used for determining the statistical significance of observed differences in nuclear fluorescence intensities and enzymatic activities between wild type and mutants, was performed using RStudio (v.3.3.2.; RStudio, Boston, MA), and P < 0.001 was taken as statistically significant.

DNA extraction and Sanger sequencing

Genomic DNA was extracted from whole blood using Blood DNA Isolation Kit (Geneaid Biotech, New Taipei City, Taiwan) according to manufacturer’s instructions. In addition to all 17 exons of BAP1, its predicted promoter region, located approximately 400 bp upstream from 5′ UTR, and the 3′ and 5′ UTR were sequenced by Sanger sequencing (reference sequences: ENSG00000163930, ENST00000460680.5; available at: www.ensembl.org; +1 represents the start of translation). Briefly, PCR-amplified fragments were purified (ExoSAP-IT, Applied Biosystems, Foster City, CA) then treated with BigDye Terminator version 3.1 Cycle Sequencing Kit (Applied Biosystems) and run with ABI3730xl DNA Analyzer (Applied Biosystems; detailed reaction conditions and primer sequences available upon request). Sequence analysis of the capillary sequenced fragments was performed with Sequencher 5.3 (Genes Codes, Ann Arbor, MI).

The Genome Reference Consortium Human Build 37 (GRCh37) was used for reference (University of California Santa Cruz Genome Browser; http://genome.ucsc.edu). The EPD was used to predict promoter regions (25,26). Variant allele frequencies were determined using the Genome Aggregation Database (gnomAD) v2.1 (27). In silico predictions for possible effect on protein function were done using MutationTaster (28), SIFT (Sorting Intolerant from Tolerant) (29), PROVEAN (Protein Variation Effect Analyzer) (30) and CADD (Combined Annotation Dependent Depletion) (31). Effects on splicing was modeled using the Human Splice Finder 3.1 (32). The COSMIC database v86 was used to access the somatic coding region BAP1 variations (33). The American College of Medical Genetics and Genomics guidelines were used for the interpretation of sequence variants (34).

Targeted amplicon sequencing

Of the 433 samples, 272 (63%) were sequenced by targeted amplicon sequencing at Institute for Molecular Medicine Finland (FIMM, Helsinki, Finland) to cover intronic regions of BAP1. Probes were designed with DesignStudio Sequencing Assay Designer (Illumina, San Diego, CA) and library preparation was performed using TruSeq Custom Amplicon v1.5 Library Prep (Illumina) kit. Sequencing was performed using Illumina HiSeq 2500 system (further information available on request). The analysis was performed as described previously (35) with the following modification: after genome alignment, reads were filtered by removing read pairs failing to span a single amplicon target. The pairs were trimmed from possible primer sequence that might have sequenced (35). Variants with <30% of the nucleotides, or sequencing depth < 10 reads, were considered artifacts and excluded. The variants were annotated using Annotate Variation (ANNOVAR) (36).

Patient RNA isolation and analysis

To analyze an intron 2 (c.67+1G>T) variant at RNA level, whole blood was collected from patient UMG12-01 in PAXgene Blood RNA Tube IVD and RNA was isolated using PAXgene Blood RNA Kit IVD (PreAnalytix, Hombrechtikon, Switzerland). RNA was extracted also from a healthy control. Following cDNA synthesis (iScript cDNA synthesis kit, Bio-Rad Laboratories, Hercules, CA) was done according to standard protocol. Reaction products done both with and without reverse transcriptase were subjected to PCR targeting S15 gene and pseudogene to exclude DNA contamination in RNA sample. Sequence analysis was done as described above. Detailed reaction conditions and primer sequences are available upon request.

DNA constructs

The likely pathogenic variants were introduced to a commercial mammalian expression vector pCMV6-entry with MYC/DDK-tagged wild-type BAP1 (NM_004656, Human cDNA ORF Clone, RC200378, OriGene, Rockville, MD) using site-directed mutagenesis or PCR cloning. In site-directed mutagenesis inverse PCR primers with desired point mutations were used to amplify the RC200378 plasmid. The BAP1 cDNA in RC200378 has a pre-existing synonymous SNP rs17849500 (c.366C>T, p.F122=) not present in our patients (designated WT1). Additional wild-type BAP1 cDNA construct without the SNP was also cloned (designated WT2). Standard PCR cloning generated c.67+1G>T mutant from UMG12-01 cDNA, and minigene constructs from genomic DNA of UMG11-53, UMG12-01 and a control with a sequence corresponding to the wild type BAP1. The minigene included the genomic region of BAP1 from exon 1 to 3. TA Cloning Kit (Invitrogen, Thermo Fisher Scientific, Waltham, MA) was used to clone promoter fragments from UMG12-58 to pCRII vector. Purified and ligated plasmids were transformed to competent Escherichia coli (NEB 5-alpha, New England BioLabs, Ipswich, MA) and sequenced to verify the mutation. Variants studied in enzymatic activity assay were subcloned further into bacterial expression vector pGEX-4 T-1 with the standard PCR-cloning protocol and transformed to BL21 (DE3; Invitrogen) bacteria.

Enzymatic activity assay and immunoblotting

BAP1 was produced as GST gene fusion protein in BL21 bacteria and purified according to standard GST purification protocol. Briefly, bacteria were grown to OD600 of 0.6–0.7 before induced with 0.1 μM isopropyl β-D-1-thiogalactopyranoside (IPTG) at 22°C for 4 h. Cells were collected by centrifugation at 6000g for 15 min at 4°C. Pellets were resuspended in 1/20 volume (original culture) phosphate buffered saline supplemented with protease inhibitors (cOmplete, Mini, EDTA-free Protease Inhibitor Cocktail, Roche Diagnostics, Risch-Rotkreuz, Switzerland) and 0.25 mg/ml lysozyme and sonicated. After adding 0.5% Triton-X, lysates were incubated for 30 min and centrifuged at 13 000g for 30 min at 4°C. Supernatants were incubated for 16 h with washed Protino gluthatione agarose 4B-beads (Macherey-Nagel, Düren, Germany). Fusion proteins were eluted in 50 mm Tris 10 mm gluthatione pH 8.0. Buffer exchange to Tris-buffered salinewas done using 100 K Amicon Ultra Centrifugal Filter Devices (Millipore, Burlington, MA) and protein concentration quantified using Bradford Protein Assay (Bio-Rad) or Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). A total of 0.1 μg of protein was incubated with 0.09 μm Ub-AMC-substrate (Boston Biochem, Cambridge, MA) diluted in reaction buffer 50 mm Tris pH 7.6, 1 mm dithiothreitol and 10 μg/ml BSA. Reactions were done in 100 μl volume in 96-well plates. Enzymatic activity was monitored by the release of fluorescence measured with Synergy H1 (BioTek, Winooski, VT) plate reader at 37°C for 40 min with excitation at 340 nm and emission at 440 nm. Aliquots of purified protein were used for Western blotting according to standard protocol to determine protein loading in the assay. Antibodies used were monoclonal anti-BAP1 (clone C4, Santa Cruz Biotechnology, Dallas, TX; 1:500) and goat anti-mouse IgG conjugated with horseradish peroxidase (1:5000) (DAKO, Agilent Technologies, Santa Clara, CA).

Cell culture and nuclear localization

Human mesothelioma cell line NCI-H226 (ATCC, LGC Standards, Middlesex, UK) without endogenous BAP1 was maintained in RPMI-1640 (Gibco, Thermo Fisher Scientific, Waltham, MA), supplied with 10% fetal bovine serum, penicillin and streptomycin, in a humidified chamber at 37°C in 5% CO₂. Transient transfections were performed on 80–90% confluent NCI-H226 cells grown in antibiotic-free media using standard FuGENE HD (Promega Biosciences, Madisson, WI) protocol and the recommended DNA:FuGENE HD ratio of 1:3. After 48 h incubation cells were fixed in 4% PFA for 15 min. Cells were labeled with the mouse anti-BAP1 monoclonal antibody followed by fluorescein-conjugated goat anti-mouse Alexa Fluor 488 (1:500) secondary antibody and Alexa Fluor 568 phalloidin F-actin probe (1:300; Invitrogen). Cells seeded on coverslips were mounted on glass slides using GelMount (Sigma-Aldrich, St. Louis, MO) and imaged with Axio Imager M2 (Zeiss, Oberkochen, Germany). CellInSight CX7 (Thermo Fischer Scientific) high content imaging system was used for quantification of cellular compartmentalization. Images (49 fields/well) were captured using a 10×/0.3NA objective and analyzed with Cellomics Cytoplasm to Nucleus Translocation V4 BioApplication (Thermo Fisher Scientific).

Multiplex ligation-dependent probe amplification

MLPA was used to detect larger size CNVs in BAP1, not detectable by capillary sequencing, according to manufacturer’s instructions (SALSA MLPA P417-B2 BAP1 probemix, MRC-Holland, Amsterdam, The Netherlands). Produced fragments were run with a size standard GeneScan 500 LIZ in ABI3730xl DNA Analyzer (Applied Biosystems), and data were analyzed using Coffalyser.Net software (MRC-Holland).

Splicing assay

Minigene expression constructs (see above) were transfected to NCI-H226 as described under nuclear localization. After 24 h incubation, total RNA was extracted with NucleoSpin RNA Plus kit (Macherey-Nagel) and complementary DNA was synthetized using iScript cDNA synthesis kit (Bio-Rad). cDNA was subjected to PCR amplification with the primers targeting the region from exon 1 to 3 of BAP1. Reactions were run in 1.5% agarose gel to detect splicing events. Bands were extracted (NucleoSpin Gel, and PCR Clean-up, Macherey-Nagel) and their sequence analyzed as described above (detailed reaction conditions and primer sequences available upon request).

Acknowledgements

CellInSight high content screening was performed at the Biomedicum Imaging Unit, University of Helsinki, Finland. Sequencing was performed at the Institute for Molecular Medicine Finland (FIMM). We thank Ms Annika Lipponen and Ms Sinikka Lindh for laboratory assistance and all patients for participation.

Conflict of Interest statement. J.A.T. received lecture fees from Thea Finland and Blueprint Genetics, Finland. He has served in the advisory board of Novartis Finland. T.T.K. received lecture fees from Santen Finland. The other authors declare no conflict of interest.

Funding

Helsinki University Hospital Research Fund (TYH2017218 to T.T.K.), Cancer Foundation Finland (to T.T.K.), Sigrid Jusélius Foundation (to T.T.K.), Eye and Tissue Bank Foundation (to J.A.T.), Folkhälsan Research Foundation (to J.A.T.), Eye Foundation (to J.A.T.) and Mary and Georg C. Ehrnrooth Foundation (to J.A.T.).

Author notes

These authors contributed equally as first authors.