A functional variant in HOXA11-AS, a novel long non-coding RNA, inhibits the oncogenic phenotype of epithelial ovarian cancer.

The homeobox A (HOXA) region of protein-coding genes impacts female reproductive system embryogenesis and ovarian carcinogenesis. The 5-prime end of HOXA includes three long non-coding RNAs (lncRNAs) (HOXA10-AS, HOXA11-AS, and HOTTIP) that are underexplored in epithelial ovarian cancer (EOC). We evaluated whether common genetic variants in these lncRNAs are associated with EOC risk and/or have functional roles in EOC development. Using genome-wide association study data from 1,201 serous EOC cases and 2,009 controls, an exonic variant within HOXA11-AS, rs17427875 (A>T), was marginally associated with reduced serous EOC risk (OR = 0.88 (95% CI: 0.78-1.01, p = 0.06). Functional studies of ectopic expression of HOXA11-AS minor allele T in EOC cells showed decreased survival, proliferation, migration, and invasion compared to common allele A expression. Additionally, stable expression of HOXA11-AS minor allele T reduced primary tumor growth in mouse xenograft models to a greater extent than common allele A. Furthermore, HOXA11-AS expression levels were significantly lower in human EOC tumors than normal ovarian tissues (p < 0.05), suggesting that HOXA11-AS has a tumor suppressor function in EOC which may be enhanced by the T allele. These findings demonstrate for the first time a role for HOXA11-AS in EOC with effects that could be modified by germline variants.


INTRODUCTION
The homeobox (HOX) family of genes are transcription factors that contribute to embryogenesis and carcinogenesis [1]. They are characterized by highly conserved homeodomains which enable HOX proteins to bind to specific DNA regions and activate or repress transcription of their target genes [2]. In the human genome, HOX genes are organized into four clusters (A, B, C, and D) located on four different chromosomes [3]. During the development of the female reproductive system, several 'HOXA' cluster genes that map to 7p15.2 (HOXA9, HOXA10, HOXA11, and HOXA13) are expressed uniformly along the Mϋllerian duct axis, though in adults their expression is confined to specific female organs [2,4]. Since HOXA genes promote aberrant epithelial differentiation [4], their expression has been evaluated in epithelial ovarian cancer (EOC), a malignancy that accounts for more deaths in North America than any other cancer of the female reproductive system [5]. Increased expression of several HOXA genes has been reported in human EOCs compared to normal ovarian surface epithelial (OSE) precursor tissues [4,[6][7][8][9]. The HOXA gene cluster is organized into a sense strand containing protein-coding genes and an antisense strand containing non-coding RNA (ncRNA) genes (Supplemental Figure  1A). The 5-prime region of the HOXA locus refers to the direction of the sense strand with respect to protein coding genes, with HOXA13 being the most 5-prime protein-coding gene (Supplemental Figure 1B). The 5-prime region includes three additional protein-coding genes (HOXA11, HOXA10, and HOXA9) and 3 lncRNAs, HOTTIP, HOXA11-AS, and HOXA10-AS. While several investigations support a role for the HOXA cluster of protein-coding genes [2,4] in ovarian embryogenesis and carcinogenesis, less is known regarding the locally residing lncRNAs and how they may contribute to these processes [10]. Based on recent reports which suggest that lncRNAs in the 5-prime distal region have a functional role in promoting malignant phenotypes [11][12][13], a rationale exists for their investigation in EOC.
Common germline genetic variants, or single nucleotide polymorphisms (SNPs), affecting lncRNAs have been shown to contribute to the development of multiple cancer types [14][15][16][17][18][19]. The objective of this investigation was to comprehensively examine inherited genetic variation in the three lncRNAs in the 5' end of the HOXA cluster region (HOXA10-AS, HOXA11-AS, and HOTTIP) in EOC. In particular, we tested the hypotheses that these variants associated with EOC risk, in vitro cell survival or proliferation, in vitro migration or invasion, EOC growth in vivo, and tumor expression. To examine risk, we evaluated genotype and clinical data from 1,201 invasive serous EOC cases and 2,009 controls from a North American genome-wide association study (GWAS) of EOC (Table 1). To evaluate functional aspects, we tested the most promising candidate SNP for functional impact using in vitro and in vivo assays. Our results suggest that although germline variants in lncRNA sequences within the HOXA cluster are not convincingly associated with EOC risk, these variants could be functional in driving malignant phenotypes associated with cancer. This line of research provides a new opportunity to advance our understanding of HOXA cluster-mediated regulation of EOC development.

Association of lnRNA SNPs in the HOXA cluster with EOC risk
We evaluated associations between 21 individual variants in 3 unique lncRNAs in the HOXA cluster region and serous EOC susceptibility using data from our GWAS. Figure 1 presents the regional association plot for the SNP-level p-values for the 21 variants, and for reference, 669 variants mapping to the 150 kb flanking regions. No SNPs within the 3 lncRNAs were associated with serous EOC risk at a significance threshold of p < 0.05 (Table 2). Only HOXA11-AS SNP rs17427875 (A > T; minor allele frequency (MAF) = 0.20) was marginally associated with a reduced risk for serous EOC (OR (95% CI) = 0.88 (0.78-1.01), p = 0.060) (Table 2; Figure 1). SNP rs17427875 is not in linkage disequilibrium (r 2 < 0.01) with rs11564004, the top-ranked SNP in the 150kb region downstream of the HOXA cluster (OR (95%CI) = 0.78 (0.63-0.96), p = 0.02) ( Figure 1A). Per the UCSC Genome Browser, rs17427875 is highly conserved across 100 vertebrates (majority species have only A allele), and is in a conserved peak. It also falls within an H3K27 ChIP region, a DNaseI hypersensitivity cluster, and several transcription factor ChIP regions including EZH2 and POLR2A ( Figure 1B).

HOXA11-AS rs17427875 minor allele inhibits cell survival and proliferation more significantly than common allele
HOXA11-AS is a highly conserved lncRNA across several species ( Figure 1B) [20], suggesting that this gene was retained through selective evolutionary pressures. Since the top-ranked candidate lncRNA SNP rs17427875 (A > T) resides within a likely regulatory region within the first exon we tested for allele-specific effects on the cellular phenotypes of in vitro models of EOC ( Figure  2). We cloned the full-length HOXA11-AS common allele construct, and then performed site-directed mutagenesis to generate a plasmid expressing the minor allele. We then transfected OVCA-433 and C13 ovarian cancer cell lines with either the full-length HOXA11-AS common allele www.impactjournals.com/oncotarget  or the full-length minor allele constructs and assessed phenotypic changes. Forty-eight hours after transfection we observed by real-time PCR that both common and minor allele expression was similar, at ~169 and 171fold over-expression in OVCA-433, respectively, when compared to vector control ( Figure 2A). In both common and minor allele constructs, ectopic expression of the HOXA11-AS resulted in a significant reduction of cell survival and proliferation, two major cellular processes associated with EOC development (Figures 2B-2C). Notably, the minor allele resulted in significantly less survival and proliferation compared to the common allele (p = 0.0022 and p = 0.021). Similar findings were observed in C13 cells ( Figure 2D).

The effects of HOXA11-AS rs17427875 on EOC migration and invasion
Epithelial-mesenchymal transition (EMT), migration, and invasion are important cellular phenotypes that are associated with early EOC development, progression and metastasis [21]. We therefore investigated if HOXA11-AS SNP rs17427875 regulates cell migration and invasion. Following transfection of minor allele and common allele constructs of HOXA11-AS for 48 hours, OVCA-433 and C13 cells were subjected to a twochamber assay. In OVCA-433, expression of both common allele and minor allele reduced cell migration (p = 0.029 and 0.0023, respectively) and invasion (p = 0.0082 and 0.0004, respectively). However, the minor allele exhibited a more significant inhibitory effect in both assays than the common allele (p = 0.019 and 0.0043, respectively) ( Figure 3A-3B). Similar results were obtained in C13 cells ( Figure 3C-3D).
Collectively, these findings suggest that the minor allele functions to inhibit oncogenic phenotypes to a more significant extent than the common allele in EOC cells. This is consistent with the epidemiologic trend towards reduced EOC risk among women carrying the HOXA11-AS rs17427875 T allele (OR = 0.88, p = 0.060).

HOXA11-AS rs17427875 SNP inhibits EOC growth in vivo
Having demonstrated HOXA11-AS rs17427875 SNP function in cell culture, we next investigated the Table 2: Polymorphisms associated with the risk of invasive serous epithelial ovarian cancer effect of this SNP on EOC tumorigenicity in a xenograft model. C13 cells were transfected with vector, common allele, and minor allele constructs. After G418 selection, stably transfected cells were subcutaneously injected into nude mice (4x10 6 /mouse, 8 mice/group). Tumors were engrafted and allowed to grow for 8 weeks at which point the vector control group reached tumor burden endpoint. We observed a reduced tumor size and weight in common allele and minor allele treated mice as compared to the vector control group. Notably, tumor size and weight were significantly less in the minor allele group as compared to the common allele group ( Figure 4A and 4B). Furthermore, histological analysis and immunostaining of tumor sections revealed that expression of the minor allele T reduced mitosis and significantly induced apoptosis when compared to vector control (p = 0.003) and common allele (p = 0.02, Figure 4C-4D). In contrast, there was no effect of the minor allele on angiogenesis (data not shown). This denotes that the HOXA11-AS SNP functions to inhibit tumor growth more significantly than the common allele in vivo.

HOXA11-AS expression in EOC
In addition, we evaluated HOXA11-AS expression and the effect of rs17427875 on HOXA11-AS levels in 18 human EOC tumor-normal pairs using semi-quantitative PCR. Using densitometry analysis, expression of HOXA11-AS was decreased on average more than 60% in EOC tumor tissue versus normal ovarian tissue (p = 6.5x10 -3 ; Figure 5A and 5B). The decreased HOXA11-AS expression was independent of genotype (Supplemental Figure 2). Taken together, these findings suggest a tumor suppressor role for HOXA11-AS in EOC and minor allele T having no effect on HOXA11-AS expression.

Ectopic expression and knockdown of HOXA11-AS has no effect on HOXA11 and HOXA13 expression in EOC
We also investigated the mechanism by which HOXA11-AS inhibits oncogenic phenotypes in EOC. The HOXA11-AS is a natural antisense transcript, which is also called antisense lncRNA, located at the 5' region but not overlapping with the protein coding gene HOXA11 (Supplemental Figure 1). Previous studies have shown that a number of antisense lncRNAs regulate their neighboring genes through a cis mechanism [22]. Thus, we assessed HOXA11 protein and mRNA levels following enforced expression and knockdown of HOXA11-AS in EOC cell lines. Transfection of common and minor alleles of HOXA11-AS into A2780CP and C13 cells had no effects on HOXA11 expression at both protein ( Figure 6A) and mRNA ( Figure 6C) levels. Furthermore, expression of HOXA11 was not changed after knockdown of HOXA11-AS in C13 and OV2008 as well as A2780CP cells ( Figure  6B-6C). Since HOXA13 is also close to HOXA11-AS in an opposite orientation (Supplemental Figure 1), we next examined if HOXA11-AS regulates HOXA13. RT-qPCR analysis revealed no significant changes of HOXA13 mRNA level after expression or knockdown of HOXA11-AS ( Figure 6D). These findings suggest that HOXA11-AS does not regulate HOXA11 and HOXA13.

DISCUSSION
Dysregulation of the HOX cluster of genes has been shown in many human diseases, including EOC [1,2,23]. The current study represents the first investigation to focus on the association between genetic variants involving lncRNAs in the HOXA gene cluster region and EOC risk and development. Our GWAS identified rs17427875,   Previous studies have shown that distal HOXA protein-coding genes are important in normal ovarian development and that changes in expression occur in ovarian tumors of epithelial origin [2,4], but the functional role and expression of lncRNAs residing within this region are largely unknown. Through experimental investigation we have shown that ectopic expression of HOXA11-AS rs17427875 common and minor allele results in tumor suppressive phenotypes in vitro. Notably, expression of minor allele exhibited more significant tumor suppressive activity than common allele. Mouse xenograft studies recapitulate in vivo the significant phenotypes associated with expression of the minor allele. These findings are consistent with the observation that rs17427875 was associated with a trend towards decreased risk for serous EOC in the epidemiologic study. Moreover, in 18 RNA normal-tumor pairs from EOC patients, we observed significant down-regulation of HOXA11-AS in serous EOC tumors versus normal ovarian tissue. This observation further reinforces our experimental data that HOXA11-AS functions as a tumor suppressor.
Antisense lncRNAs can be involved in regulating of the expression of their neighboring genes in cis or more distant genes via a trans mechanism [22]. Accumulating studies show that a number of antisense lncRNAs demonstrate cis regulation of their neighboring genes. For instance, ANRIL (antisense non-coding RNA in the INK4 locus) was shown to silence the INK4b-ARF-INK4A locus by recruiting the PRC1 and PRC2 complex [24]. ANRASSF1 (antisense intronic non-coding RASSF1) recruits PRC2 to the RASSF1A promoter, which leads to the accumulation of the repressive mark H3K27me3 and reduction in the RASSF1A transcription [25]. However, our data show that knockdown or overexpression of HOXA11-AS does not affect expression of its neighboring genes HOXA11 and HOXA13 ( Figure 6). Since some lncRNAs have been shown to function as miRNA sponges, we queried the RNA22 database with HOXA11-AS sequence and did not find miRNA binding sites within the rs17427875 region (data not shown). In addition, the RNAsnp database analysis shows that minor allele does not cause secondary structure change of HOXA11-AS when compared to the common allele (Supplemental Figure 3). These findings suggest that HOXA11-AS exerts its cellular function possibly through trans regulation of Figure 6: HOXA11-AS has no effects on HOXA11 and HOXA13 expression. A. and B. Indicated cell lines were transfected with common, minor allele and siRNAs of HOXA11-AS as well as vector and control siRNA. Following incubation for 48 hours, the cells were subjected to RT-qPCR analysis (upper panels) and western blot with antibodies against HOAX11, GAPDH and β-actin (bottom panels). C. and D. Indicated cell lines were treated as panels A and B and then were analyzed for HOXA11 and HOXA13 mRNA levels by RT-qPCR. distant genes.
With the rapid growth of identified lncRNAs and disease-associated SNPs, there is a great interest in studying SNPs in lncRNAs [26]. For instance, the rs2839698 TC genotype of lncRNA H19 was associated with a reduced risk of developing non-muscle-invasive bladder cancer [27]. The SNP rs6983267 within lncRNA CCAT2 has been implicated in predisposition to colorectal and prostate cancer [28]. A recent study has shown that this SNP affects the expression of CCAT2, thereby impacting tumor growth and metastasis in colorectal cancer [29]. Moreover, SNP rs920778, which locates in intron 2 of lncRNA HOTAIR, has recently been shown to be associated with esophageal squamous cell carcinoma risk and regulation of HOTAIR expression [30]. However, although HOXA11-AS was down-regulated in EOC ( Figure 5), rs17427875 had no effect on HOXA11-AS expression (Supplemental Figure 2). This could potentially be attributed to the small number of evaluated samples, or may indicate a different mechanism of action. It is important to note that while we were performing our in vitro and in vivo experiments, a larger dataset from the COGS genotyping initiative [31] became available. The association we observed with rs17427875 (A > T) in HOXA11-AS was not replicated among an independent set of cases with serous histology (OR (95% CI) = 1.00 (0.95-1.06), p = 0.98), and no associations were suggested for other lncRNA SNPs in the region. The basis for lack of consistency in the epidemiologic data is not immediately apparent but may be due to heterogeneity between study populations included in our North American GWAS and those included in COGS, an international genotyping initiative. It is also possible that the initial association observed in the North American GWAS was a falsepositive finding. This is a challenge inherent in the detection of weak effects with EOC as an endpoint. Nonetheless, the functional impact of the rs17427875 T minor allele was strong for several oncogenic phenotypes and our data demonstrate that non risk-associated polymorphisms can have allele-specific functionality, which is an important consideration for investigators performing post-GWAS functional studies. Regardless of genotype, this study provides strong evidence for the role of HOXA11-AS in EOC.

Study design and population
Details of the study design and population have been previously described [32]. Briefly, we conducted a GWAS of North American EOC case-control studies. Characteristics of the participating studies are summarized in Table 1. Cases were women diagnosed with pathologically-confirmed primary invasive EOC fallopian tube cancer, or primary peritoneal cancer ascertained from clinic-, population-, and hospital-based studies and cancer registries. To increase homogeneity, our primary analysis focused on serous adenocarcinomas, the most common histologic subtype of EOCs [33]. Moreover, most of the cases had high-grade disease. The vast majority of the cases do not have a family history of ovarian or breast cancer in a first-degree relative, and most have not been tested for BRCA1 or BRCA2 mutations. Controls were women without a current or prior history of EOC with at least one ovary intact at the reference date. All studies had data on disease status, age at diagnosis/interview, selfreported racial group, and histologic subtype. Most studies frequency-matched cases and controls on age group and race. For the present analysis, we focused on subjects of European ancestry since they represent the vast majority of subjects and the burden of EOC is the highest in this racial group [33]. European ancestry was confirmed using principal component analysis [34] with HapMap CEU populations.

Informed consent
All participants provided written informed consent for their samples and data to be used. The protocol was approved by the institutional review board at each study site.

Genotyping, quality control, and imputation
Genotypes were generated using different versions of the Illumina Infinium arrays (Table 1). Genotyping methods and sample and SNP quality control procedures have been described previously [35]. Briefly, samples were excluded if they had a call rate < 95%, > 1% discordance, < 80% European ancestry, or ambiguous gender. SNPs were excluded if they had call rates < 95% or if they had minor allele frequencies (MAF) less than 5%. To account for differ ent marker sets and improve genome coverage, genotype data was imputed for GWAS participants based on data from all 14 populations in the 1000 genomes project (1KGP, version 3, March 2012 release)) as the reference using IMPUTE2 version 2 after pre-phasing with SHAPEIT [36][37][38]. Before imputation, we excluded poor performing SNPs according to the genotyping success rates, deviation from Hardy Weinberg Equilibrium (HWE) with P < 1.0x10 -5 , replicate errors, and rare MAF.
To ensure the quality of imputed genotypes, maximum likelihood genotype imputation was carried out and an estimate of the squared correlation between imputed and true genotypes was calculated. Imputed SNPs with an R-square < 0.25 were excluded. www.impactjournals.com/oncotarget

Identification of variants in candidate HOXA lncRNA cluster genes
The coordinates for each of 3 candidate human HOXA lncRNA genes (HOXA10-AS, HOXA11-AS, and HOTTIP) were used to identify SNPs falling within the designated gene sequences. To annotate the coordinates for these lncRNAs, we used GENCODEdb (http://www. gencodegenes.org/), the most comprehensive lncRNA annotation available [39]. A total of 72 variants in these 3 genes were represented in the imputed GWAS dataset. After filtering out the variants whose distribution deviated from HWE or had a MAF < 5%, a total of 21 non-coding SNPs falling within the 3 HOXA lncRNA genes remained.

Association testing
We performed SNP-level association tests for the 21 identified SNPs using the pooled GWAS dataset. For each variant, unconditional logistic regression treating alternate alleles as an ordinal variable (a log-additive model) was used to evaluate individual SNP-serous EOC risk associations. Models were adjusted for study site and the first principal component (PC) representing European ancestry. To adjust for multiple comparisons, we estimated the false discovery rate (FDR) and used a corresponding q-value [40] of 10% to declare statistical significance. Statistical analyses were carried out using SAS (Version 9.3; SAS Inc.), PLINK (Version 1.07) [41], Matlab (R2011b; the Mathwork Inc.), and R (Version 3.0.2).

Quantitative and semi-quantitative RT-PCR
Total RNA was isolated using Trizol reagent following manufacturer's protocol and was used to generate cDNA with High Capacity cDNA Reverse Transcription Kit (Life Technologies, Carlsbad, CA, USA). Quantitative (real-time) PCR was performed with SYBR Green 2x Master Mix (Life Technologies Carlsbad, CA, USA) on ABI HT9600 from Applied Biosystems (Foster City, CA, USA). Cycle threshold (CT) values were generated and analyzed using ABI SDS version 2.3. Delta CTs were normalized to GAPDH reference gene, and ΔΔCT analysis was performed to calculate relative expression of RNA. Patient RNA was extracted from archived human primary EOC tissue and matched adjacent "normal" ovarian surface epithelial cells from each of 18 patients from the Moffitt Cancer Center Total Cancer Care Biorepository [42]. Briefly, 500 ng samples of total RNA were used for RT reaction, and 2ul of the RT product was used for subsequent semi-qPCR reaction. Quantification of PCR was performed using ImageJ software and plotted as a ratio of HOXA11-AS/GAPDH. Real-time PCR primers for HOXA11-AS and GAPDH are listed in Supplementary   Table 1. After quantifying mRNA expression levels, we compared the expression levels between matched pairs of samples using the Wilcoxon sign rank test.
HOXA11-AS common and minor allele expression plasmids Full-length insert of the common allele of HOXA11-AS was amplified using reverse transcription (RT) product as template and cloned to pcDNA3.1 vector (Life Technologies, Carlsbad, CA USA) at EcoRI/XbaI sites. After common allele expression construct was confirmed by sequencing, the minor allele expression plasmid was generated using QuikChange ® Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA USA) and introducing SNP as point mutation in insert. The primers used for HOXA11-AS site directed mutagenesis are listed in Supplementary Table 1.

Proliferation and survival assays
OVCA-433 and C13 cell lines were generous gifts from Xianghong Wang (University of Hong Kong) and Benjamin Tsang (Ottawa Hospital and Research Institute), respectively. Both cell lines were cultured under normal conditions using RPMI-1640 supplemented with 10% fetal bovine serum. OVCA-433 and C13 cells were seeded into chamber slides and transfected with vector, common, and minor allele constructs and cultured under G418 selection. EdU nucleoside analogue was allowed to incorporate during DNA synthesis for 1 hour. Cells were processed using Click-iT EdU Imaging Kit (Life Technologies) according to manufacturer's protocol. Relative proliferation was quantified by taking mean and standard deviations of 4 non-biased image fields. To assess overall cell survival, 4x10 4 cells per well were seeded in 12-well plates and transfected with constructs using 4 replicates. After 6 days selection, overall survival was assayed using 0.5mg/ml MTT reagent (Sigma, St.Louis, MO USA), solubilized in DMSO, and read at 590nm.

Migration and invasion assays
EOC cells were transfected for 48 hours with vector, common allele, or minor allele HOXA11-AS constructs. Cells were washed, trypsinized, and seeded into the upper chamber of Boyden Chambers coated with (invasion) and without (migration) Matrigel. All top chambers contained serum free media while lower chambers had RPMI media containing 10% fetal bovine serum. After 16 hours, top chambers were cleared with cotton swab, washed in PBS, and bottom chamber was stained in crystal violet containing 10% methanol. Migration and invasion were quantified by taking mean and standard deviations of 4 non-biased image fields. www.impactjournals.com/oncotarget

Mouse xenograft model
Animal studies were approved by the Institutional Animal Care and Use Committee at University of South Florida. Six weeks old female Nu/Nu mice were purchased from Charles River Laboratories (Wilmington, MA USA). C13 cells were transfected with vector, common allele, or minor allele constructs and stably selected with G418 for 2 weeks. Cells were washed in PBS, trypsinized, and resuspended in PBS. Eight mice from each group received 4x10 6 cells/100ul injection subcutaneously. Tumor volume was measured via standard caliper measurements and tumor weight was calculated at endpoint. Tumor sections were histologically analyzed and were examined with Tunel assay.

Western blot analysis of HOXA11-AS effect on its neighboring gene expression
In addition C13, A2780CP and OV2008 cells, which were gifts from Benjamin Tsang (Ottawa Hospital and Research Institute), were transfected with the common and minor allele plasmids and 2 siRNAs of HOXA11-AS. Following incubation for 48 hours, the cells were subjected to western blot analysis with antibody against HOXA11. The siRNA sequences are listed in Supplementary Table 1.