Epigenetic reprogramming and aberrant expression of PRAME are associated with increased metastatic risk in Class 1 and Class 2 uveal melanomas

Abstract

Background:
We previously identified PRAME as a biomarker for metastatic risk in Class 1 uveal melanomas. In this study, we sought to define a threshold value for positive PRAME expression (PRAME+) in a large dataset, identify factors associated with PRAME expression, evaluate the prognostic value of PRAME in Class 2 uveal melanomas, and determine whether PRAME expression is associated with aberrant hypomethylation of the PRAME promoter.

Results:
Among 678 samples analyzed by qPCR, 498 (73.5%) were PRAME- and 180 (26.5%) were PRAME+. Class 1 tumors were more likely to be PRAME-, whereas Class 2 tumors were more likely to be PRAME+ (P < 0.0001). PRAME expression was associated with shorter time to metastasis and melanoma specific mortality in Class 2 tumors (P = 0.01 and P = 0.02, respectively). In Class 1 tumors, PRAME expression was directly associated with SF3B1 mutations (P < 0.0001) and inversely associated with EIF1AX mutations (P = 0.004). PRAME expression was strongly associated with hypomethylation at 12 CpG sites near the PRAME promoter.

Materials and methods:
Analyses included PRAME mRNA expression, Class 1 versus Class 2 status, chromosomal copy number, mutation status of BAP1, EIF1AX, GNA11, GNAQ and SF3B1, and genomic DNA methylation status. Analyses were performed on 555 de-identified samples from Castle Biosciences, 123 samples from our center, and 80 samples from the TCGA.

Conclusions:
PRAME is aberrantly hypomethylated and activated in Class 1 and Class 2 uveal melanomas and is associated with increased metastatic risk in both classes. Since PRAME has been successfully targeted for immunotherapy, it may prove to be a companion prognostic biomarker.

Full text

Oncotarget1
www.impactjournals.com/oncotarget
www.impactjournals.com/oncotarget/ Oncotarget, Advance Publications 2016
Epigenetic reprogramming and aberrant expression of PRAME
are associated with increased metastatic risk in Class 1 and
Class 2 uveal melanomas
Matthew G. Field1, Michael A. Durante1, Christina L. Decatur1, Bercin Tarlan1,
Kristen M. Oelschlager2, John F. Stone2, Jefm Kuznetsov1, Anne M. Bowcock3,
Stefan Kurtenbach1, J. William Harbour1
1Bascom Palmer Eye Institute, Sylvester Comprehensive Cancer Center and Interdisciplinary Stem Cell Institute, University
of Miami Miller School of Medicine, Miami, FL, USA
2Castle Biosciences, Inc., Friendswood, TX, USA
3National Heart and Lung Institute, Imperial College London, London, UK
Correspondence to: J. William Harbour, email: harbour@miami.edu
Keywords: PRAME, preferentially expressed antigen in melanoma, uveal melanoma, DNA methylation, chromosomal
instability
Received: June 07, 2016 Accepted: July 13, 2016 Published: July 30, 2016
ABSTRACT
Background: We previously identied PRAME as a biomarker for metastatic risk
in Class 1 uveal melanomas. In this study, we sought to dene a threshold value for
positive PRAME expression (PRAME+) in a large dataset, identify factors associated
with PRAME expression, evaluate the prognostic value of PRAME in Class 2 uveal
melanomas, and determine whether PRAME expression is associated with aberrant
hypomethylation of the PRAME promoter.
Results: Among 678 samples analyzed by qPCR, 498 (73.5%) were PRAME- and
180 (26.5%) were PRAME+. Class 1 tumors were more likely to be PRAME-, whereas
Class 2 tumors were more likely to be PRAME+ (P < 0.0001). PRAME expression was
associated with shorter time to metastasis and melanoma specic mortality in Class
2 tumors (P = 0.01 and P = 0.02, respectively). In Class 1 tumors, PRAME expression
was directly associated with SF3B1 mutations (P < 0.0001) and inversely associated
with EIF1AX mutations (P = 0.004). PRAME expression was strongly associated with
hypomethylation at 12 CpG sites near the PRAME promoter.
Materials and methods: Analyses included PRAME mRNA expression, Class 1
versus Class 2 status, chromosomal copy number, mutation status of BAP1, EIF1AX,
GNA11, GNAQ and SF3B1, and genomic DNA methylation status. Analyses were
performed on 555 de-identied samples from Castle Biosciences, 123 samples from
our center, and 80 samples from the TCGA.
Conclusions: PRAME is aberrantly hypomethylated and activated in Class 1 and
Class 2 uveal melanomas and is associated with increased metastatic risk in both
classes. Since PRAME has been successfully targeted for immunotherapy, it may prove
to be a companion prognostic biomarker.
INTRODUCTION
Uveal melanoma is the most common primary
cancer of the eye and the second most common form
of melanoma. Due to a high rate of metastasis, much
research has focused on the development of biomarkers
to predict metastatic risk. Previously, we described a gene
expression prole that could be performed on a ne needle
biopsy of the primary tumor that accurately predicted
metastasis [1]. Tumors with the Class 1 prole have a
low metastatic risk, whereas those with the Class 2 prole
have a high metastatic risk. Consequently, a 15 gene
array (12 discriminating genes and 3 control genes) was
developed and prospectively validated [2, 3]. This assay

Oncotarget2
www.impactjournals.com/oncotarget
is now available commercially as the DecisionDx-UMTM
test (Castle Biosciences), which has been independently
validated [4] and is widely used to stratify patients for
metastatic surveillance and to identify high risk patients
for adjuvant therapy trials [5].
While the vast majority of metastatic events in
uveal melanoma arise from Class 2 tumors, a small
subset of Class 1 tumors also give rise to metastasis. We
found that the expression of two of the 12 discriminating
genes on the array (CDH1 and RAB31) could be used
to identify Class 1 tumors that may have increased
metastatic risk. Class 1 tumors with low expression
of these genes and very low predicted metastatic risk
were called Class “1A,” whereas those with high
expression and higher predicted metastatic risk were
called Class “1B.” In our efforts to further improve
the prognostic accuracy of the gene array platform, we
conducted a genome wide search for new biomarkers
and found that mRNA expression of the cancer-testis
antigen Preferentially Expressed Antigen in Melanoma
(PRAME) was an accurate biomarker for metastasis in
Class 1 tumors [6]. In that initial study, we found that
any detectable mRNA expression of PRAME above
baseline was associated with increased metastatic risk.
Limitations of that study included a relatively small
number of tumor samples that were biased towards
larger tumor size.
To date, there have been ve common driver
mutations identied in uveal melanoma: BAP1, EIF1AX,
GNA11, GNAQ and SF3B1 [7–11]. Mutations in BAP1,
SF3B1 and EIF1AX are almost mutually exclusive and are
associated with high, intermediate and low metastatic risk,
respectively [6, 12]. Also, SF3B1 mutations were found to
be associated with PRAME expression [6].
The purpose of the present study was to study
PRAME expression in a much larger number of Class 1
and, for the rst time, in Class 2 uveal melanomas spanning
the true range of tumor sizes encountered in clinical
practice. We sought to dene a threshold value for calling a
tumor sample positive for PRAME expression (PRAME+),
compare PRAME expression to the 1A/1B designation
in Class 1 tumors, identify clinical and molecular
factors associated with PRAME expression, evaluate the
prognostic value of PRAME expression in Class 2 tumors,
and determine whether PRAME expression in uveal
melanoma is correlated with promoter hypomethylation.
RESULTS
To evaluate the spectrum of PRAME mRNA
expression and to establish a threshold for positive
PRAME expression in primary uveal melanoma,
we analyzed qPCR data from 678 tumor samples,
including 123 of our samples and 555 de-identied
samples submitted from a large number of ocular
oncology centers to Castle Biosciences. These samples
included 454 (67.0%) Class 1 tumors and 224 (33.0%)
Class 2 tumors. Class 1 tumors included 317 (69.8%)
Class 1A tumors, 131 (28.9%) Class 1B tumors, and 6
(1.3%) tumors for which 1A/1B information was not
available. Whereas most samples showed negligible
PRAME expression, a subset of samples showed a broad
range of PRAME expression (Figure 1A). We previously
showed that any PRAME expression above baseline was
associated with increased metastasis in Class 1 tumors
and consequently dened any expression above baseline
as positive PRAME expression (PRAME+) [6]. In this
study, we used the same methodology to establish a
broadly applicable PRAME+ threshold from qPCR data
using a much larger dataset that included both Class 1
and Class 2 tumors, with a majority derived from ne
needle biopsy of small and medium sized tumors and
a smaller number from large, enucleated specimens
that is representative of the actual distribution of tumor
sizes encountered in clinical practice (Figure 1B–1C).
A similar method was used to determine a PRAME+
threshold using RNA-Seq data from The Cancer Genome
Atlas (TCGA) dataset (Figure 1D).
Overall, 498 (73.5%) tumors were PRAME− and 180
(26.5%) were PRAME+. Class 1 tumors were more likely
to be PRAME−, whereas Class 2 tumors were more likely
to be PRAME+ (Fisher exact test, P < 0.0001) (Figure 2A).
Among Class 1 tumors, 357 (78.6%) were PRAME− and
97 (21.4%) were PRAME+. Among Class 1A tumors, 261
(82.3%) were PRAME− and 56 (17.7%) were PRAME+.
Among Class 1B tumors, 93 (71.0%) were PRAME− and
38 (29.0%) were PRAME+. Class 1A tumors were more
likely to be PRAME−, whereas Class 1B tumors were
more likely to be PRAME+ (Fisher exact test, P = 0.01)
(Figure 2B). Among Class 2 tumors, 141 (62.9%) were
PRAME− and 83 (37.1%) were PRAME+. Additionally,
we determined PRAME mRNA status in commonly used
UM cell lines: Mel202 and MP41 are PRAME+, whereas
92.1, Mel270, Mel290, and MP46 are PRAME−.
Association between PRAME and clinical
features
Clinical annotations and PRAME expression
were available for 123 of our uveal melanoma samples
(Supplementary Table S1). The only features that were
signicantly associated with PRAME+ status were larger
tumor diameter and thickness (Mann-Whitney test, P = 0.01
and P = 0.02, respectively). To expand this analysis, we
examined the TCGA Research Network dataset which
consists of an independent cohort of 80 primary uveal
melanoma samples (http://cancergenome.nih.gov/)
(Supplementary Table S2). Since PRAME expression
data were obtained from RNA-Seq analysis in the
TCGA dataset, we established a threshold for PRAME+
expression using the same procedure as for qPCR data
(Figure 1D). Consistent with our original dataset, PRAME+

Oncotarget3
www.impactjournals.com/oncotarget
status in the TCGA dataset was associated with larger
tumor diameter (P = 0.02). In both datasets, well-known
risk factors for metastasis such as increased patient age,
ciliary body involvement, and extrascleral tumor extension
were not associated with PRAME expression status.
We previously showed that PRAME expression
was associated with increased metastatic risk in Class 1
uveal melanomas [6]. Here, we extend that analysis
to determine whether PRAME expression is also a
biomarker for metastasis in Class 2 tumors. For this
analysis, we combined our 123 cases with the 80 cases
from the TCGA. Using Kaplan-Meier survival analysis,
PRAME+ status was associated with shorter time to
metastasis for both Class 1 and Class 2 tumors together
(P = 0.0002; 22 metastatic events; median follow-up of
19 months, range 0–125 months) and for Class 2 tumors
alone (P = 0.01; 15 metastatic events; median follow-up
of 18 months, range 0–89 months)(Figure 3A–3B).
Similarly, PRAME+ status was associated with shorter
time to melanoma-specic mortality for both Class 1 and
Class 2 tumors together (P = 0.001; 32 melanoma-specic
mortality events; median follow-up of 19 months, range
0–142 months) and for Class 2 tumors alone (P = 0.02;
28 melanoma-specic mortality events, median follow-up
of 18 months, range 0–89 months) (Figure 3C–3D).
Association between PRAME and chromosomal
alterations
To identify chromosomal copy number changes
that may be associated with PRAME expression, we
analyzed 26 of our cases and 80 TCGA cases for which
chromosomal copy number and PRAME expression
data were available (Supplementary Table S3). Overall,
PRAME+ tumors were strongly associated with 6q loss
(P < 0.0001), 8p loss (P = 0.04), 8q gain (P < 0.0001)
and 16q loss (P < 0.0001) (Figure 4). Notably, PRAME
expression status was not associated with monosomy 3
(P = 0.3), the chromosomal alteration most strongly
associated with metastasis in uveal melanoma,
highlighting the potential benet of including PRAME
expression status in a prognostic test.
We then analyzed Class 1 and Class 2 tumors
separately. Among Class 1 tumors, PRAME+ status was
associated with 1q gain (P = 0.04), 6p gain (P = 0.01),
6q loss (P < 0.0001), 8q gain (P < 0.0001), and 16q
loss (P = 0.004). PRAME+ status in Class 2 tumors was
associated with 6p gain (P = 0.04), 6q loss (P = 0.0001), 8p
loss (P = 0.05), 8q gain (P = 0.02) and 16q loss (P = 0.003).
PRAME expression was not associated with monosomy 3
in either comparison (Supplementary Table S3).
Figure 1: Dening the threshold for PRAME+ expression status. (A) PRAME mRNA expression plotted from lowest to highest
expression for 678 uveal melanoma samples measured by qPCR with a LOESS model (second degree, family = ”Gaussian”, spanning
0.4, tting by least-squares). (B) Predicted PRAME mRNA expression for an additional “hypothetical” 678 samples based on the LOESS
model. (C) Predicted slope change between each of these predicted points. (D) The same process depicted in panels A–C was repeated
separately for the RNA-Seq data from 80 TCGA uveal melanoma samples in order to generate a predicted slope change plot. For both
datasets, the threshold for PRAME+ (red) was dened as the point where the slope sustainably rose above baseline (blue).

Oncotarget4
www.impactjournals.com/oncotarget
Figure 2: Summary of PRAME expression status measured by qPCR. (A) PRAME expression status with respect to gene
expression prole classication in 678 uveal melanomas. (B) PRAME expression status in 454 Class 1 uveal melanomas with respect to
1A/1B sub-classication.
Figure 3: Prognostic signicance of PRAME expression status in uveal melanoma. (A) Kaplan-Meier survival plot showing
metastasis-free survival for Class 1 and Class 2 tumors combined, with respect to PRAME expression status. (B) Kaplan-Meier survival
plot showing metastasis-free survival for Class 2 tumors only, with respect to PRAME expression status. (C) Kaplan-Meier plot showing
melanoma-specic mortality for Class 1 and Class 2 tumors combined, with respect to PRAME expression status. (D) Kaplan-Meier
survival plot showing melanoma-specic mortality for Class 2 tumors only, with respect to PRAME expression status.

Oncotarget5
www.impactjournals.com/oncotarget
Association between PRAME and driver gene
mutations
To identify common driver mutations that may be
associated with PRAME+ status, we analyzed 59 of our
cases for which mutation data were available, as well
as the 80 TCGA cases, for mutations in EIF1AX, BAP1,
GNA11, GNAQ and SF3B1 (Supplementary Table S4).
When Class 1 and Class 2 tumors were considered
together, PRAME+ status was associated with BAP1
mutations (P = 0.02). However, this association is likely
due to BAP1 mutations occurring almost exclusively in
Class 2 tumors [9], which we show here to be associated
with PRAME+ status. When Class 1 tumors were analyzed
separately, PRAME expression was directly associated
with SF3B1 mutations (P < 0.0001) and inversely
associated with EIF1AX mutations (P = 0.004). There
were no mutations associated with PRAME expression in
Class 2 tumors when analyzed separately.
PRAME expression is associated with aberrant
promoter hypomethylation
Testes is the only normal adult tissue that expresses
PRAME mRNA at appreciable levels (Figure 5A),
which strongly suggests that the expression of PRAME
in uveal melanoma is anomalous. Consequently, we
hypothesized that PRAME may become aberrantly
activated in uveal melanoma by hypomethylation of
the promoter region. Consistent with this hypothesis,
12 CpG sites within and near the PRAME promoter
were signicantly hypomethylated (FDR < 0.05 for all
probes) in PRAME+ tumors compared to PRAME−
tumors (Figure 5B). We validated these ndings using
bisulte conversion followed by Sanger sequencing in
a subset of cases (Supplementary Table S5). Strikingly,
there was a highly signicant correlation between the
level of hypomethylation of all 12 CpG sites and the
level of mRNA expression (P < 0.0001) (Figure 5C and
Supplementary Figure S1). The most differentially
methylated CpG site (recognized by probe cg27303185)
is hypermethylated in all adult tissues except placenta and
sperm (Figure 5D). These data indicate that the PRAME
promoter region is normally hypermethylated and silenced
in virtually all normal adult tissues, but it is targeted for
hypomethylation and aberrant transcriptional activation
during uveal melanoma progression.
DISCUSSION
We previously reported that PRAME mRNA
expression is a signicant risk factor for metastasis in
Class 1 uveal melanomas, and we developed a general
method for establishing a PRAME+ threshold in various
datasets [6]. In that article, our analysis included a much
greater proportion of large tumors treated by enucleation
than are encountered in actual clinical practice. However,
since we show here that PRAME expression is strongly
associated with larger tumor size, a study composed
primarily of large tumors may not accurately reect
the true range of PRAME expression. To pursue the
development of PRAME as a clinical biomarker, we sought
here to rigorously establish a standard method for dening
the PRAME+ expression threshold using a standardized
and widely used qPCR platform. To achieve a widely
applicable threshold and avoid potential systematic biases
arising from a single center study, we analyzed a large
number of samples obtained from many different ocular
oncology centers representing a wide range of tumor
sizes and both Class 1 and Class 2 tumors in proportions
representative of actual clinical practice. From this
analysis, we established a PRAME+ threshold and used it
to identify clinical, chromosomal and mutational features
associated with PRAME expression. We also established
a PRAME+ threshold for RNA-Seq using the TCGA
dataset, but this threshold must be considered provisional
since that dataset was composed primarily of very large
tumors treated by enucleation. Indeed, 43% of the TCGA
samples were PRAME+, compared to only 27% of our
samples.
Across all samples, larger tumor size was the only
clinical feature that correlated with PRAME+ status,
Figure 4: Association of PRAME expression with chromosomal gains and losses. The bar graphs depict chromosomal gains
and losses that were signicantly associated with PRAME+ tumors when Class 1 and Class 2 tumors were analyzed together, and when
each class was analyzed separately. PRAME+ (red), PRAME− (blue).

Oncotarget6
www.impactjournals.com/oncotarget
suggesting that PRAME becomes transcriptionally
activated later during tumor progression. Interestingly,
even though PRAME+ status was a stronger predictor of
metastasis in Class 1 tumors, it was also associated with
metastasis in Class 2 tumors. In our earlier study that was
much smaller and biased towards larger tumors treated by
enucleation, we did not nd a correlation between PRAME
expression status and the “1A/1B” system that the clinical
test currently uses to indicate low (1A) versus moderate
(1B) metastatic risk [6]. In the present study that included
a much larger number of samples that better represented
the full spectrum of uveal melanomas, we observed
a highly signicant correlation between PRAME+
status and “1B” status. Nevertheless, since there were a
number of discordant cases, we are preparing to start a
multi-center prospective study to determine the relative
prognostic value of PRAME expression status versus the
1A/1B designation in Class 1 tumors in order to determine
the optimal biomarker for increased metastatic risk in
Class 1 tumors. A limitation of this analysis is the limited
follow-up, particularly from the TCGA dataset, which
results in a large number of censored data points. Our
planned prospective multicenter study with long follow-up
is the appropriate study design to validate these ndings.
Since we found that PRAME+ correlates signicantly
with tumor size, this multi-center study will also evaluate
whether there is a minimum threshold tumor size at which
point PRAME becomes prognostic.
PRAME expression was associated with specic
chromosomal gains and losses, some of which were
specic to either Class 1 or Class 2 tumors. Changes that
were associated with PRAME+ status in both Class 1 and
Class 2 tumors included 6p gain, 6q loss, 8q gain and
16q loss. 6p gain and 6q loss were frequently found in
the same tumor samples, likely representing the formation
of an isochromosome 6p [13, 14]. A previous study
identied 16q loss in 16% of uveal melanomas, but no
prognostic signicance was found [14]. Our study using
a larger number of samples and more accurate molecular
analytical methods indicates that 16q loss may indeed
have prognostic signicance. 1q gain was associated
with PRAME+ status only in Class 1 tumors, which
conrms our previous observation [6]. 1q gain has only
rarely been mentioned in the uveal melanoma literature
[15], but our ndings suggest the need for further studies
to determine whether 1q gain has pathogenic as well as
prognostic signicance. 8p loss was associated with
PRAME+ status only in Class 2 tumors, whereas 8q
gain was associated with PRAME+ status in both tumor
classes. 8q gain is prevalent in both Class 1 and Class 2
Figure 5: Transcriptional activation of PRAME is associated with hypomethylation of the PRAME promoter in uveal
melanoma. (A) The only normal adult human tissue that expresses high levels of PRAME mRNA is testis. Data were obtained through
the GTEx Portal [41]. (B) Locations of 12 CpG sites (blue bars) within or near the PRAME promoter that exhibited signicantly decreased
methylation in PRAME+ uveal melanomas (n = 41) compared to PRAME- samples (n = 39) at a signicance level of FDR < 0.05.
(C) Scatter plots showing the relationship between PRAME mRNA expression levels (obtained from TCGA RNA-Seq data) and PRAME
promoter methylation (obtained from TCGA Innium HumanMethylation450 BeadChip data) using two representative methylation
probes (cg17648213 and cg27303185). Spearman’s rank correlation coefcient was used to determine P-values. Graphs depicting the
other 10 differentially methylated probes are in Supplementary Figure S1. (D) Methylation data for the cg27303185 methylation probe
was plotted for normal tissues obtained from Marmal-aid [40]. A separate panel (right) depicts PRAME+ and PRAME- uveal melanomas
samples for comparison. RPKM, reads per kilobase of transcript per million mapped reads; CPM, counts per million.

Oncotarget7
www.impactjournals.com/oncotarget
tumors, but the mechanism leading to 8q gain tends to be
different between the two tumor classes [16]. In Class 1
tumors, 8q gain often occurs through gain of an entire
copy of chromosome 8 or by simple gain of the q arm,
whereas in Class 2 tumors, 8q gain frequently occurs
through formation of an isochromosome 8q, which is
accompanied by loss of 8p [17]. The common association
of PRAME expression and isochromosome formation on
chromosomes 6 and 8 is of interest and may provide new
insight into uveal melanoma tumorigenesis. We previously
showed that genes which become aberrantly up-regulated
in PRAME+ tumors are enriched for functions related to
chromosome maintenance, meiotic recombination and
telomere maintenance [6]. In addition, the PRAME protein
has been shown to associate at transcriptional target
sites on chromatin with the KEOPS/EKC complex [18],
which is involved in chromosome segregation, telomere
maintenance and other highly conserved functions [19].
Hence, aberrant expression of PRAME may predispose
tumor cells to isochromosome formation, as well as other
forms of aneuploidy that promote tumor progression.
Our nding that PRAME becomes aberrantly
hypomethylated and transcriptionally activated during
uveal melanoma progression is similar to ndings in other
cancers [20, 21] and may have therapeutic implications.
Since PRAME is not normally expressed in most normal
adult tissues, targeted molecular inhibition of the PRAME
protein or immunotherapy directed against PRAME−
expressing tumor cells may be well tolerated. Indeed,
there is growing evidence that PRAME may be a good
target for immunotherapy [22–25]. Since the PRAME
protein is not normally expressed on the cell surface,
one strategy is to target PRAME using a T-cell receptor
mimic (TCRm) monoclonal antibody that recognizes
the PRAME300–309 peptide presented by HLA*A02:01 on
the cell surface [26]. Others have developed PRAME−
specic cytotoxic T lymphocytes that have shown
effective responses against PRAME-expressing tumor
cells, including progenitor populations that are notoriously
resistant to current cancer therapeutic strategies [24, 27].
Furthest along in development are vaccines against
PRAME that are currently undergoing clinical trials in
cutaneous melanoma and other cancers (Trial numbers
NCT01149343, NCT01853878 and NCT00423254) [28].
Interestingly, we evaluated PRAME expression of two
matched primary and metastatic UM samples analyzed by
the Illumina HumanRef-8 v1.0 expression microarray in
our previously published dataset (GEO accession number
GSE39717) [29], and we found that both the primary and
metastatic samples were PRAME+ (data not shown),
supporting a mechanistic role for PRAME expression
in UM metastasis. Since no effective therapies currently
exist for metastatic uveal melanoma [30], our center and
others are preparing to undertake clinical trials to assess
the efcacy of PRAME-directed immunotherapy in
appropriately selected patients.
In summary, we have provided a threshold for
PRAME+ expression from qPCR data for primary uveal
melanomas across a wide spectrum of tumor sizes and
in both tumor classes representative of actual clinical
practice. We previously identied PRAME expression as a
biomarker for increased metastatic risk in Class 1 tumors
[6], and here we showed for the rst time that PRAME
expression is also associated with worse prognosis
among Class 2 tumors. We demonstrated that specic
chromosomal gains and losses, as well as specic driver
mutations, are found preferentially in PRAME+ tumors.
Finally, we showed that specic CpG sites around the
PRAME promoter are differentially hypomethylated
in PRAME+ tumors, suggesting that the aberrant
transcriptional activation of PRAME in uveal melanoma
is the result of epigenetic reprogramming during tumor
progression. In addition to its prognostic value, PRAME
expression status may potentially be useful in the future
for guiding the use of PRAME-directed immunotherapy,
which would make PRAME the rst true “companion
prognostic” biomarker in uveal melanoma.
MATERIALS AND METHODS
The sources of all uveal melanoma samples used
in this study are summarized in Supplementary Table S6.
Tumor samples were obtained from 123 primary uveal
melanomas from the practice of one of the authors
(JWH), including 64 samples that were included in a
previous publication [6]. The research was conducted in
a HIPAA-compliant manner in accordance with the tenets
of the Declaration of Helsinki. Approval was obtained
from the Institutional Review Board of the University of
Miami. Written informed consent was obtained from each
patient from our center. Baseline clinical information and
patient outcomes were recorded. De-identied PRAME
expression and GEP Class data were obtained from 555
uveal melanoma samples from Castle Biosciences that had
been collected between July 21, 2015, and March 2, 2016,
as part of internal PRAME qPCR method development.
These samples were obtained as formalin-xed parafn-
embedded tissue from enucleations in 55 (9.9%) cases and
as fresh-frozen samples from ne needle aspirate biopsies
in 500 (90.1%) cases. The data available for these cases
included GEP class, 1A versus 1B subtype for Class 1
tumors, and PRAME mRNA expression. Additionally
we analyzed clinical, whole exome sequencing, RNA
sequencing, SNP 6.0 array data, and DNA methylation
data from 80 uveal melanoma samples generated by the
TCGA Research Network: http://cancergenome.nih.gov/.
PRAME mRNA expression analysis
For the RNA samples from our center and from
Castle Biosciences, PRAME mRNA expression was
analyzed by qPCR using the Applied Biosystems 7900 HT

Oncotarget8
www.impactjournals.com/oncotarget
Real-Time PCR System with TaqMan primers and Gene
Expression Master Mix following the manufacturer’s
protocol as previously described [6]. Ct values were
calculated using the manufacturer’s software and
ΔCt values were calculated by subtracting the geometric
mean of the Ct values of the endogenous control genes
from the mean Ct values for PRAME, as previously
described [6]. Relative normal expression was calculated
using the equation 2^-ΔCT. For the 80 samples from the
TCGA, raw RNA-Seq datasets were aligned to the hg19
genome using STAR [31], which was also used to generate
count les. Count les were then normalized using
DeSeq2 [32]. Next-generation sequencing analysis was
conducted on Pegasus, the supercomputer administered
by the Advanced Computing Group of the Center for
Computational Science at the University of Miami. For
PRAME mRNA expression in normal tissues, RNA-Seq
data was obtained from the Genotype-Tissue Expression
(GTEx) project [33].
Estimating class status from RNA-sequencing
data
For research purposes of this study, we estimated the
gene expression prole class assignment for the 80 TCGA
samples, which were analyzed by RNA-Seq. Raw RNA-
Seq datasets were prepared using the pipeline described
in the previous section. The top 20% most variable genes
were selected, analyzed by principal component analysis,
and plotted using the stats, matrixstats, and rgl packages,
respectively, in R (version 3.2.3). This analysis grouped
the samples into two clusters, as we have previously
described for Class 1 and Class 2 tumors [34]. The identity
of each cluster was determined to be most consistent with
Class 1 versus Class 2 based on the expression of genes
previously known to be differentially up-regulated in each
Class. The DecisionDx-UM test results were available for
11 of these samples, and there was 100% concordance
with our class assignment. This method was used solely
for research purposes and is not meant for actual clinical
testing, as it has not been prospectively validated in a
manner analogous to the DecisionDx-UM test.
Determining PRAME+ expression threshold
qPCR and RNA-Seq samples were separately
ordered from lowest to highest relative and normalized
PRAME expression, respectively, and each was plotted
with a line representing the best-tting LOESS model
(second degree, family = ”Gaussian”, spanning 0.4 for
qPCR and 0.45 for RNA-Seq, tting by least-squares)
(Figure 1A). Based on the LOESS model, a predicted
dataset tting the LOESS model was generated
(Figure 1B) and the slope between each predicted point
was calculated and plotted (Figure 1C) to represent the
change in slope. The point of inection where the slope
sustainably rose above baseline was dened as the cut-off
for PRAME+ and PRAME− (Figure 1C–1D).
Exome sequencing and chromosomal copy
number analysis
Whole-exome sequencing was conducted on 24 of
our primary uveal melanomas and matched blood using
NimbleGen SeqCap EZ Human Exome Library v2.0
(Roche Nimblegen) and run on the Illumina Genome
Analyzer II. Exome sequencing data on 80 primary
uveal melanoma TCGA samples were downloaded from
CGHub and aligned to the hg19 reference genome using
Novoalign. Variant calling was conducted using Mutect2
[35] and Varscan2 [36]. Chromosomal copy number
analysis was obtained for 106 samples, including 26
samples from our center (15 from previously published
data and 11 newly analyzed from exome sequencing data)
and 80 from the TCGA. Chromosomal gains and losses
were called by CNVKit [37] for exome sequencing data
and by ASCAT [38] for TCGA SNP 6.0 array data.
DNA methylation analysis
The 80 TCGA uveal melanoma tumors samples
were assayed for global DNA methylation status with the
Innium HumanMethylation450K BeadChip (Illumina).
This kit interrogates ~450,000 methylation sites at single-
nucleotide resolution, including at CpG sites within
promoter, 5ʹUTR, rst exon, gene body, and 3ʹUTR
regions. Methylation data underwent quality control,
normalization, and differential analysis of PRAME+
and PRAME− samples using the ChAMP methylation
pipeline in R [39]. CpG sites that were differentially
hypomethylated at a signicance level of FDR < 0.05 were
plotted along the PRAME locus using the GViz package
in R. All 12 methylation probes targeting PRAME that
are included in the Methyl450K array were signicantly
hypomethylated in the TCGA PRAME+ samples.
For validation, primers were designed against a
region containing 3 of these 12 probes and validated in
4 PRAME+ and 3 PRAME− samples. This validation
study was small due to limited sample availability.
For primer design, 500 ng of tumor DNA was bisulte
converted using the EZ Methylation-Lightning Kit
(Zymo Research). Primers for PCR amplication
of the PRAME promoter were designed with the
Bisulte Primer Seeker (http://www.zymoresearch.
com/tools/bisulte-primer-seeker). Forward Primer:
GAAGGATTTCGTGTTTAAGGTTTTTTAAGG. Reverse
Primer: GTGTTTTTATTTTGGAAATAGAGATTTAGT
TTTTTTT. The PRAME promoter region was amplied
with the EpiMark Hot Start Taq polymerase (New England
Biolabs) at Tm = 54.5°C, and the PCR product puried by
agarose gel separation/elution before Sanger sequencing.
The status of the PRAME methylation site detected by

Oncotarget9
www.impactjournals.com/oncotarget
Innium HumanMethylation450K BeadChip probe
cg27303185 in normal tissues was obtained from Marmal-
aid [40] and plotted in a box-and-whisker with ggplot2 in
R in comparison to TCGA uveal melanoma data.
Statistical analysis
Statistical analysis was performed using Medcalc®
version 14.10.2. Fisher’s exact test was used to evaluate
discrete dichotomous variables, the Mann-Whitney test for
comparison of continuous variables, Spearman’s rho for
correlation analyses of continuous variables, and Kaplan-
Meier survival analysis for determining the association of
PRAME expression status with patient outcomes.
ACKNOWLEDGMENTS
The authors acknowledge the support of
the Oncogenomics Core and the Biostatistics and
Bioinformatics Core at the Sylvester Comprehensive
Cancer Center, and the Center for Computational Science
at the University of Miami for help and support with the
data management and analytics.
CONFLICTS OF INTEREST
Dr. Harbour is the inventor of intellectual property
related to the gene expression prole technology used in
the study. Drs. Harbour and Bowcock are the inventors
of intellectual property related to the discovery of BAP1
mutations in uveal melanoma. Dr. Harbour is a paid
consultant for Castle Biosciences, which licensed this
intellectual property, and he receives royalties from
its commercialization. Kristen Oelschlager and Dr.
John Stone are employees and stockholders of Castle
Biosciences. No other authors disclose a conict of
interest.
GRANT SUPPORT
This work was supported by National Cancer
Institute grants R01 CA125970 (J.W.H.), R01 CA161870
(J.W.H. and A.M.B.) and F30 CA206430 (M.G.F.),
Research to Prevent Blindness, Inc. Senior Scientic
Investigator Award (J.W.H.), Melanoma Research
Foundation (J.W.H., M.G.F.), Melanoma Research
Alliance (J.W.H.), Ocular Melanoma Foundation
(J.W.H.), the 2015 RRF/Kayser Global Pan-American
Award (J.W.H.), the Sylvester Comprehensive Cancer
Center, the University of Miami Sheila and David Fuente
Graduate Program in Cancer Biology (M.G.F., M.A.D.),
the Center for Computational Science Fellowship
(M.G.F.), and the AACR-Ocular Melanoma Foundation
Fellowship in honor of Robert C. Allen, MD (S.K.). The
Bascom Palmer Eye Institute also received funding from
NIH Core Grant P30EY014801, Department of Defense
Grant #W81XWH-13-1-0048, and a Research to Prevent
Blindness Unrestricted Grant.
REFERENCES
1. Onken MD, Worley LA, Ehlers JP, Harbour JW. Gene
expression proling in uveal melanoma reveals two
molecular classes and predicts metastatic death. Cancer
Res. 2004; 64:7205–7209.
2. Onken MD, Worley LA, Tuscan MD, Harbour JW. An
accurate, clinically feasible multi-gene expression assay
for predicting metastasis in uveal melanoma. J Mol Diagn.
2010; 12:461–468.
3. Onken MD, Worley LA, Char DH, Augsburger JJ,
Correa ZM, Nudleman E, Aaberg TM, Jr., Altaweel MM,
Bardenstein DS, Finger PT, Gallie BL, Harocopos GJ, et al.
Collaborative Ocular Oncology Group report number 1:
prospective validation of a multi-gene prognostic assay in
uveal melanoma. Ophthalmology. 2012; 119:1596–1603.
4. Correa ZM, Augsburger JJ. Sufciency of FNAB aspirates
of posterior uveal melanoma for cytologic versus GEP
classication in 159 patients, and relative prognostic
signicance of these classications. Graefes Arch Clin Exp
Ophthalmol. 2014; 252:131–135.
5. Aaberg TM, Jr., Cook RW, Oelschlager K, Maetzold D,
Rao PK, Mason JO, 3rd. Current clinical practice:
differential management of uveal melanoma in the era
of molecular tumor analyses. Clinical ophthalmology
(Auckland, NZ). 2014; 8:2449–2460.
6. Field MG, Decatur CL, Kurtenbach S, Gezgin G, van der
Velden PA, Jager MJ, Kozak KN, Harbour JW. PRAME
as an independent biomarker for metastasis in uveal
melanoma. Clin Cancer Res. 2016; 22:1234–1242.
7. Van Raamsdonk CD, Bezrookove V, Green G, Bauer J,
Gaugler L, O’Brien JM, Simpson EM, Barsh GS,
Bastian BC. Frequent somatic mutations of GNAQ in uveal
melanoma and blue naevi. Nature. 2009; 457:599–602.
8. Van Raamsdonk CD, Griewank KG, Crosby MB,
Garrido MC, Vemula S, Wiesner T, Obenauf AC,
Wackernagel W, Green G, Bouvier N, Sozen MM,
Baimukanova G, Roy R, et al. Mutations in GNA11 in uveal
melanoma. N Engl J Med. 2010; 363:2191–2199.
9. Harbour JW, Onken MD, Roberson ED, Duan S, Cao L,
Worley LA, Council ML, Matatall KA, Helms C,
Bowcock AM. Frequent mutation of BAP1 in metastasizing
uveal melanomas. Science. 2010; 330:1410–1413.
10. Harbour JW, Roberson ED, Anbunathan H, Onken MD,
Worley LA, Bowcock AM. Recurrent mutations at codon
625 of the splicing factor SF3B1 in uveal melanoma. Nat
Genet. 2013; 45:133–135.
11. Martin M, Masshofer L, Temming P, Rahmann S, Metz C,
Bornfeld N, van de Nes J, Klein-Hitpass L, Hinnebusch AG,
Horsthemke B, Lohmann DR, Zeschnigk M. Exome
sequencing identies recurrent somatic mutations in

Oncotarget10
www.impactjournals.com/oncotarget
EIF1AX and SF3B1 in uveal melanoma with disomy 3. Nat
Genet. 2013; 45:933–936.
12. Decatur CL, Ong E, Garg N, Anbunathan H, Bowcock AM,
Field MG, Harbour JW. Driver mutations in uveal
melanoma: associations with gene expression prole and
patient outcomes. JAMA ophthalmology. 2016.
13. Prescher G, Bornfeld N, Friedrichs W, Seeber S, Becher R.
Cytogenetics of twelve cases of uveal melanoma and
patterns of nonrandom anomalies and isochromosome
formation. Cancer Genet Cytogenet. 1995; 80:40–46.
14. Kilic E, van Gils W, Lodder E, Beverloo HB, van Til ME,
Mooy CM, Paridaens D, de Klein A, Luyten GP. Clinical
and cytogenetic analyses in uveal melanoma. Invest
Ophthalmol Vis Sci. 2006; 47:3703–3707.
15. Aalto Y, Eriksson L, Seregard S, Larsson O, Knuutila S.
Concomitant loss of chromosome 3 and whole arm losses
and gains of chromosome 1, 6, or 8 in metastasizing
primary uveal melanoma. Invest Ophthalmol Vis Sci. 2001;
42:313–317.
16. Ehlers JP, Worley L, Onken MD, Harbour JW. Integrative
genomic analysis of aneuploidy in uveal melanoma. Clin
Cancer Res. 2008; 14:115–122.
17. Onken MD, Worley L, Harbour JW. A metastasis modier
locus on human chromosome 8p in uveal melanoma
identied by integrative genomic analysis. Clin Cancer Res.
2008; 14:3737–3745.
18. Costessi A, Mahrour N, Sharma V, Stunnenberg R,
Stoel MA, Tijchon E, Conaway JW, Conaway RC,
Stunnenberg HG. The human EKC/KEOPS complex is
recruited to Cullin2 ubiquitin ligases by the human tumour
antigen PRAME. PLoS One. 2012; 7:e42822.
19. Srinivasan M, Mehta P, Yu Y, Prugar E, Koonin EV,
Karzai AW, Sternglanz R. The highly conserved
KEOPS/EKC complex is essential for a universal tRNA
modication, t6A. EMBO J. 2011; 30:873–881.
20. Qian J, Zhu ZH, Lin J, Ming Yao DM, Li Y, Yang J,
Wang CZ. Hypomethylation of PRAME promoter is
associated with poor prognosis in myelodysplastic
syndrome. Br J Haematol. 2011; 154:153–155.
21. Schenk T, Stengel S, Goellner S, Steinbach D, Saluz HP.
Hypomethylation of PRAME is responsible for its
aberrant overexpression in human malignancies. Genes
Chromosomes Cancer. 2007; 46:796–804.
22. Gerard C, Baudson N, Ory T, Segal L, Louahed J.
A comprehensive preclinical model evaluating the
recombinant PRAME antigen combined with the AS15
immunostimulant to ght against PRAME-expressing
tumors. J Immunother. 2015; 38:311–320.
23. Ikeda H, Lethe B, Lehmann F, van Baren N, Baurain JF,
de Smet C, Chambost H, Vitale M, Moretta A, Boon T,
Coulie PG. Characterization of an antigen that is recognized
on a melanoma showing partial HLA loss by CTL expressing
an NK inhibitory receptor. Immunity. 1997; 6:199–208.
24. Amir AL, van der Steen DM, van Loenen MM, Hagedoorn RS,
de Boer R, Kester MD, de Ru AH, Lugthart GJ, van Kooten C,
Hiemstra PS, Jedema I, Grifoen M, van Veelen PA,
et al. PRAME-specic Allo-HLA-restricted T cells with
potent antitumor reactivity useful for therapeutic T-cell
receptor gene transfer. Clin Cancer Res. 2011; 17:5615–5625.
25. Spel L, Boelens JJ, van der Steen DM, Blokland NJ,
van Noesel MM, Molenaar JJ, Heemskerk MH, Boes M,
Nierkens S. Natural killer cells facilitate PRAME-specic
T-cell reactivity against neuroblastoma. Oncotarget. 2015;
6:35770–35781.
26. Chang A, Dao T, Scott A, Dubrovsky L, Liu C,
Scheinberg DA. A Therapeutic TCR Mimic Monoclonal
Antibody for Intracellular PRAME Protein in
Leukemias. Blood. 2015; 126:2527–2527. doi: 10.18632/
oncotarget.5657.
27. Quintarelli C, Dotti G, Hasan ST, De Angelis B, Hoyos V,
Errichiello S, Mims M, Luciano L, Shafer J, Leen AM,
Heslop HE, Rooney CM, Pane F, et al. High-avidity
cytotoxic T lymphocytes specic for a new PRAME-
derived peptide can target leukemic and leukemic-precursor
cells. Blood. 2011; 117:3353–3362.
28. Weber JS, Vogelzang NJ, Ernstoff MS, Goodman OB,
Cranmer LD, Marshall JL, Miles S, Rosario D, Diamond DC,
Qiu Z, Obrocea M, Bot A. A phase 1 study of a vaccine
targeting preferentially expressed antigen in melanoma
and prostate-specic membrane antigen in patients with
advanced solid tumors. J Immunother. 2011; 34:556–567.
29. Worley LA, Onken MD, Person E, Robirds D, Branson J,
Char DH, Perry A, Harbour JW. Transcriptomic versus
chromosomal prognostic markers and clinical outcome in
uveal melanoma. Clin Cancer Res. 2007; 13:1466–1471.
30. Luke JJ, Triozzi PL, McKenna KC, Van Meir EG,
Gershenwald JE, Bastian BC, Gutkind JS, Bowcock AM,
Streicher HZ, Patel PM, Sato T, Sossman JA, Sznol M, et al.
Biology of advanced uveal melanoma and next steps for
clinical therapeutics. Pigment Cell Melanoma Res. 2015;
28:135–147.
31. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C,
Jha S, Batut P, Chaisson M, Gingeras TR. STAR: ultrafast
universal RNA-seq aligner. Bioinformatics. 2013; 29:15–21.
32. Anders S, McCarthy DJ, Chen Y, Okoniewski M,
Smyth GK, Huber W, Robinson MD. Count-based
differential expression analysis of RNA sequencing data
using R and Bioconductor. Nat Protoc. 2013; 8:1765–1786.
33. Consortium GT. Human genomics. The Genotype-Tissue
Expression (GTEx) pilot analysis: multitissue gene
regulation in humans. Science. 2015; 348:648–660.
34. Labrecque MP, Takhar MK, Jagdeo JM, Tam KJ, Chiu C,
Wang TY, Prefontaine GG, Cox ME, Beischlag TV. A
TRIP230-Retinoblastoma Protein Complex Regulates
Hypoxia-Inducible Factor-1alpha-Mediated Transcription
and Cancer Cell Invasion. PLoS One. 2014; 9:e99214.
35. Cibulskis K, Lawrence MS, Carter SL, Sivachenko A,
Jaffe D, Sougnez C, Gabriel S, Meyerson M, Lander ES,
Getz G. Sensitive detection of somatic point mutations
in impure and heterogeneous cancer samples. Nature
biotechnology. 2013; 31:213–219.

Oncotarget11
www.impactjournals.com/oncotarget
36. Koboldt DC, Larson DE, Wilson RK. Using VarScan 2 for
Germline Variant Calling and Somatic Mutation Detection.
Curr Protoc Bioinformatics. 2013; 44:15 14 11–17.
37. Talevich E, Shain AH, Botton T, Bastian BC. CNVkit:
Genome-Wide Copy Number Detection and Visualization
from Targeted DNA Sequencing. PLoS Comput Biol. 2016;
12:e1004873.
38. Van Loo P, Nordgard SH, Lingjaerde OC, Russnes HG,
Rye IH, Sun W, Weigman VJ, Marynen P, Zetterberg A,
Naume B, Perou CM, Borresen-Dale AL, Kristensen VN.
Allele-specic copy number analysis of tumors. Proc Natl
Acad Sci U S A. 2010; 107:16910–16915.
39. Morris TJ, Butcher LM, Feber A, Teschendorff AE,
Chakravarthy AR, Wojdacz TK, Beck S. ChAMP: 450k
Chip Analysis Methylation Pipeline. Bioinformatics. 2014;
30:428–430.
40. Lowe R, Rakyan VK. Marmal-aid—a database for Innium
HumanMethylation450. BMC Bioinformatics. 2013;
14:359.
41. Carithers LJ, Ardlie K, Barcus M, Branton PA, Britton A,
Buia SA, Compton CC, DeLuca DS, Peter-Demchok J,
Gelfand ET, Guan P, Korzeniewski GE, Lockhart NC, et al.
A Novel Approach to High-Quality Postmortem Tissue
Procurement: The GTEx Project. Biopreservation and
biobanking. 2015; 13:311–319.