Background
Ovarian malignancies account for 4% of cancer in women and are the most frequent cause of death due to gynecological cancer in Western countries [
1]. Carcinomas are the most common subtype, with the serous histotype being particularly prevalent [
1]. Most serous ovarian carcinomas are genomically unstable. Approximately 50% of the tumors have defects in the homologous recombination DNA repair pathway, with
BRCA1 and
BRCA2 alterations as the most frequent, while the remaining half show less characteristic aberration patterns [
2‐
4].
Genomic imbalances are widespread in serous ovarian carcinomas and in a large The Cancer Genome Atlas (TCGA) study of ovarian carcinomas, 113 significant focal DNA copy number alterations were identified [
4]. Structural chromosomal aberrations are also frequently seen, with involvement of chromosome 19 being particularly common [
4‐
10]. Chromosome 11 has been reported to be one of the recurrent partners in such rearrangements [
9,
11,
12]. The functional consequences of these aberrations are not understood.
Genomic rearrangements may lead to transcriptional deregulation, gene truncations, or gene fusions that encode fusion proteins [
13]. Abnormal transcription events can also result in chimeric RNAs, e.g., by RNA polymerase read-through of adjacent genes [
14]. Fusion genes have been identified in several epithelial cancers [
15], but none have so far been validated as recurrent in independent cohorts of ovarian cancer [
13]. In a recent study of 92 serous ovarian carcinomas using DNA and RNA sequencing, gene breakage was found to be a common mechanism for inactivating tumor suppressor genes but no fusion gene was identified as a recurrent, biologically plausible driver in tumorigenesis [
10]. Still, for subgroups of patients, gene fusions might play such a role.
In the present study, we looked for fusion gene candidates and aberrantly expressed genes as well as the possible mechanisms behind such expression changes in a series of ovarian carcinomas with cytogenetically identified chromosome 19 changes. We used two different genome-scale approaches: exon-level microarrays and transcriptome sequencing (RNA-seq).
Discussion
Large cohort studies of serous ovarian carcinomas have not found recurrent gene fusions [
4,
10]. However, isolated events in single patients can still be important steps in tumorigenesis and/or progression of individual tumors.
We report here a combination of transcriptome analyses of ovarian carcinomas selected on the basis of known presence of structural rearrangements of chromosome 19 in their karyotypes. The
DPP9-
PPP6R3 fusion rearrangement was identified by RNA-seq and exon-level microarray analyses and validated by RT-PCR in one ovarian serous carcinoma. The fusion leads to disruption and subsequent deregulation of
DPP9 gene expression.
DPP9 was found rearranged with
PLIN3 in another serous carcinoma, and also in this case the
DPP9 expression was disrupted and lowered toward the 3′ end. The lost fragment of the
DPP9 transcript includes the code for the functional peptidase and esterase-lipase domains of the DPP9 protein. Previously, Hoogstraat et al. [
30] found rearrangement of
DPP9 in a high-grade serous ovarian carcinoma by means of whole-genome mate-pair sequencing, a
DPP9-
PAX2 in-frame rearrangement with the breakpoint after exon 11 in
DPP9, i.e., close to the breakpoint identified in the present
DPP9-
PPP6R3 fusion. No information was reported on
DPP9 expression. Despite the presence of three different partner genes (
PPP6R3,
PLIN3, and
PAX2) involved in the
DPP9- rearrangements seen until now, it seems that they all lead to loss of the 3′ part of the
DPP9 transcript (Additional file
1: Table S2). The effect of the disruption of
DPP9 expression was evaluated by BLAT; despite different fusion breakpoint positions, both the
DPP9-
PPP6R3 and the
DPP9-
PLIN3 rearrangements would lead to loss of the same functional domains at protein level, namely the peptidase and esterase-lipase domains.
In the TCGA dataset of high-grade serous ovarian carcinomas, 7% of the samples had downregulated gene expression of
DPP9 without DNA copy number alteration (i.e., 23 of 316 samples, data available at cBio Cancer Genomics Portal, Memorial Sloan-Kettering Cancer Center (MSKCC),
www.cbioportal.org, default settings [
4,
31,
32]). This was also the case in our samples where
DPP9 was found downregulated without DNA copy number change. Patch et al. [
10] showed that, in ovarian carcinomas, gene breakage was a common mechanism for inactivating tumor suppressor genes (
RB1,
NF1,
RAD51B, and
PTEN); thus, loss-of-function gene changes could explain the observed
DPP9 downregulation.
We have found no studies describing fusions involving
DPP9 in the Mitelman database or in the TCGA fusion gene portal [
33,
34]; thus, this is the first report showing that
DPP9-rearrangements occur in serous ovarian carcinoma.
The
DPP9 gene encodes a serine protease that belongs to the DPPIV subfamily and is ubiquitously expressed (
www.uniprot.org, [
35]). Proteases may act as tumor suppressors [
36] as exemplified by
DPP4 which is homologous to
DPP9, both encode proteins that harbor the DPPIV-domain as well as hydrolase and peptidase conserved domains. Loss of
DPP4 contributes to tumorigenesis in several cancers [
36], including ovarian carcinoma [
37], whereas forced expression has shown growth inhibitory effect on glioma cells [
38]. The DPP9 protein participates in cell signaling and has several tumor suppressing abilities such as inducing apoptosis, suppressing proliferation, and attenuating activation of the oncogene AKT (protein kinase B) [
39]. The
DPP9-rearrangements resulted in loss of the active sites of DPP9; it is well known that gene fusions may represent loss-of-function events which play a role in carcinogenesis, as reported in colorectal [
40] and prostate cancer [
41].
LHX2 was among the genes showing differential expression of either the 5′ or 3′ end of the transcript (Fig.
4a). Different gene expression profiles among samples can sometimes be due to expression of alternative transcripts. This does not seem to be the case for
LHX2, however, which has several annotated transcript isoforms according to the ENSEMBL data base [
42], but none that can explain the observed gene expression breakpoint
. The gene encodes a transcriptional activator that has been reported to promote tumor growth and metastasis in breast cancer [
43]. Its involvement in three gene fusions (
IGH-LHX2,
ADAMTS13-LHX2, and
AAK1-LHX2) has been described in chronic myelogenous leukemia, breast cancer, and uterine carcinosarcoma, respectively [
34,
44].
Three of our samples showed strong expression of the 3′ of
LYNX1. The exon-level gene expression results could possibly be explained by alternative transcripts since the increased expression matches the starting point of two transcript isoforms (transcripts ENST00000521396 and ENST00000317543). No fusions involving
LYNX1 have been reported in the literature so far, but the suggested fusion partner
FCF1 was involved in a
TIMM9-
FCF1 fusion in an astrocytoma [
34]. Another example of a transcript breakpoint that corresponds to the start of a transcript isoform is provided by
PRKAR2A, but in that case a true fusion event seems more likely given the RNA-seq data. A
CDC25A-
PRKAR2A fusion was previously reported in an ovarian carcinoma [
34].
Deregulation of
MMP27 was consonant with the findings made by cytogenetics-based genomic analysis, RNA-sequencing, and studies of the exon-level gene expression profile. The RNA-seq listed
TMEM123 as the 5’partner. A fusion of the same two genes has been reported in a breast carcinoma, while in one ovarian carcinoma
TMEM123 was found fused with
MMP7 [
34]. In all these three reports,
TMEM123 had the same breakpoint position. Since
MMP27 and
TMEM123 map closely on the same DNA strand (in 11q22.2), an interstitial deletion could have caused the fusion, and imbalances in this chromosomal region have been seen by both HR-CGH and karyotyping [
12]. Upregulation of matrix metalloproteinases are known to play a role in cancer and metastasis [
45].
Previously performed karyotypic analyses gave information about the possible chromosomal rearrangements behind several of the identified fusion gene candidates. One example is
ZBTB46-
WFDC13, where both genes map to chromosome 20 and where the karyotype included the following highly rearranged chromosome: der(19)(20p13 → 20q13::19p13 → 19q13::8q22 → 8qter) [
18]; it is reasonable to assume that additional submicroscopic alterations were also present leading to the gene rearrangement. Interestingly,
WFDC13 is closely related to the clinically important gene
WFDC2 which is also known as Human Epididymis Protein 4.
WFDC2 is known to be overexpressed in ovarian carcinomas and encodes the HE4 protein, one of very few biomarkers used to monitor the disease in ovarian cancer patients [
46].
PRKAR2A-rearrangements have been reported as an in-frame
CDC25A-
PRKAR2A fusion in the TCGA ovarian cancer data set [
34]. Interestingly, the breakpoint thus identified corresponds to the exon-level breakpoint in our sample 14. This sample was not RNA-sequenced, so we do not know if this rearrangement caused the gene expression profile.
The chromosome 19 sub-analysis highlighted the
FAM129C gene which by RNA-seq was found to participate in the generation of the fusion gene candidate
DDA1-
FAM129C. Both the genomic findings and gene expression analyses gave results pointing in the same direction. The two genes are located only 211 kbp apart on the same strand (19p13.11); thus, an interstitial deletion could have caused the fusion. Another fusion involving this gene, namely the
CLTC-FAM129C in breast cancer, was reported before [
34]. Little is known about the function of
FAM129C or the possible consequences of its upregulated expression.
Some notes of caution are warranted on the limitations of the approach we have used to identify potential fusion gene partners by looking for genes with different expression of their 3′ and 5′ ends. Gene fusions involving the promoter region of one gene and the coding sequence of another will not be detected and would constitute “false negatives”. Furthermore, the algorithm might misevaluate expression of different transcript isoforms from a single gene and nominate it as a fusion gene candidate, i.e., a “false positive”. The algorithm is also sensitive to technical noise in the data. Some of the breakpoints identified did not seem to be due to abnormal transcription but to noise within a limited sample set. By using different methods for gene expression analysis and comparing the results, we have tried to identify gene expression alterations that arise due to fusion events rather than due to expression of different transcript isoforms or alternative splicing.