Splice variants of zinc finger protein 695 mRNA associated to ovarian cancer

All investigations were performed in accordance with the Declaration of Helsinki with approval by the Central Research Committee of the Mexican Institute of Social Security and The Ethics Committee of Centro Médico Siglo XXI, Mexican Institute of Social Security. After informed consent was obtained, normal ovarian tissue (HOT), borderline ovarian tumor (BOT), malignant epithelial ovarian tumors stages III and IV (MOT) tissues were collected by the clinical partners at the Oncology Hospital, National Medical Center Siglo XXI, IMSS, and at the General Hospital of Mexico SSA (Secretaría de Salud) from patients with diagnosed ovarian cancer, or healthy ovarian tissue from patients who underwent abdominal surgery for hysterectomy due to uterine myomatosis with no evidence of ovarian pathology. Routinely, during this type of procedure, in patients over 45 years old both ovaries are removed, and only one in patients under 45.

Cancer and corresponding normal tissue specimens were cut into three fragments and snapped frozen in liquid nitrogen, one of which was stored in RNA Latter^R (Qiagen, Valencia, CA, USA) at −70°C for a maximum of two months until RNA was purified, and the other two remaining fragments were formalin-fixed, paraffin-embedded, sliced, mounted on slides, and stained with HE. Only tissue samples with >80% tumor cells or normal epithelial cells (MOT or HOT, respectively), according to the histopathological examination were included for analysis.

Moreover, we also included for study ovarian cancer cell lines (OCL) NIH: OVCAR-3 [30], SK-OV-3 [31], TOV-112D [32] and TOV-21G [32], kindly provided by Dr. Laura Díaz-Cueto, Research Unit on Reproductive Medicine, Instituto Mexicano del Seguro Social (IMSS).

Samples were disrupted using a TissueLyser™ system (Qiagen, Valencia, CA, USA) for 60s at 30 Hz. Total RNA was obtained with RNeasy Mini Kit (Qiagen, Valencia, CA, USA) and total RNA concentration was quantified using a NanoDrop ND-1000 spectrophotometer and RNA quality was visualized and measured on an Agilent RNA 6000 Nano Assays in an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA).

Microarray analysis was achieved by means of CEL files of the Partek Genomics Suite 6.5 v software (Partek Incorporated, Saint Louis, MO). Probe sets were summarized by means of Median Polish and normalized by quantiles with no probe sets excluded from analysis. Background noise correction was achieved by means Robust Multi-chip Average (RMA) and data were log2 transformed. Data grouping and categorization was achieved by principal component analysis (PCA). Differentially expressed exons were detected by means of Alternative Splicing ANOVA with the healthy control samples as the baseline. Moreover, BOT, MOT and OCL were also examined against HOT by the Geometric least squares means model. Hierarchical clustering was based on the dissimilarity of samples (Euclidian method) by means of average linkage.

For linear cDNA synthesis, 1 μg total RNA was predigested with 1 U DNAse, 1 × DNAse buffer, 5 mM EDTA, after which it was incubated at 37°C for 30 min and at 65°C for an additional 10 min. Thereafter, samples were placed in master mix containing: 40 U Ribolock RNAse inhibitor, 0.2 μg random hexamer primers, 20 mM dNTP’s mix, 40 U M-Mulv reverse transcriptase (RT), and 1 × M-Mulv RT Buffer (Thermo Scientific).

Conditions for endpoint PCR amplification were: 5′ CGAATGAGAGCTGGCAAAGGCAAA 3′ Fwd., 5′ ACGCCAAGTGCCGTACAATTCATC 3′ Rev. primers (housekeeping gene RPL4) 7.5 mM, 1 × Taq buffer, 2 mM MgCl₂, 0.4 mM dNTP’s, 1.25 U Taq Pol, 2 ul cDNA; whereas ZNF695 was amplified with primers: 5′ GCCTTTGTCTCCTTGCGGC 3′ Fwd. 5′ GGCTGTCTTCTCTGTGTTCACGTT 3′ Rev. 12.5 mM, 1 × Taq buffer, 3 mM MgCl₂, 0.4 mM dNTP’s, 1.25 U Taq Pol, 2 ul cDNA. In both cases mix reactions were initially incubated at 95°C for 5 min, and then were run for 40 cycles at 94°C 45 s, 59°C 45 s, 72°C 60 s; and finally at 72°C for 5 min.

PCR products were separated by electrophoresis (2.5% agarose gels) and extracted by means of Gel extraction kit^TM (Qiagen, Valencia, CA, USA). The extracted products were ligated into pGem-T Easy Vector™ (Promega, Madinson, WI) by incubating overnight in 1.5 mL Eppendorf tubes with 2 × Rapid Ligation Buffer (T4 ligase), pGEM-T Easy Vector, PCR product and T4 DNA ligase at 4°C.

Recombinant plasmid DNA was purified with Wizard Plus Miniprep DNA Purification System™ (Promega) and selected clones were sequenced with M13 oligonucleotide and BigDye Terminator 3.1 cycle sequencing kit (Applied Biosystems), and sequenced in an Applied Biosystems Abi Prism 3130 genetic analyzer automated sequencer. Subsequently, the PCR amplicon sequences were assembled and checked against the transcript sequences annotated in the NCBI nucleotide database.

A total of 14 samples with an RNA integrity number (RIN) ≥ 8 were hybridized in GeneChip 1.0 microarrays according to the MIAME guidelines. Histopathological classification of tissues was as follows: healthy ovarian tissue (HOT) n = 4, benign ovarian tumors (BOT) n = 2, (malignant) serous epithelial ovarian tumors in stages III and IV (MOT) n = 4, and ovarian cell lines (OCL) n = 4. As a prerequisite, healthy tissue had to be free of any visible alteration, whereas all tumor tissues, benign or malignant, selected for study contained at least 90% tumor cells.

Background correction and normalization of microarrays reported no quality control errors (QC) and, as expected, the QC intergroup proportions were variable (Additional file 1), whereas gene expression histograms were similar in all samples. General gene expression was examined and visualized according to the histological groups by means of PCA. As expected, except for HOT and BOT that essentially overlapped, MOT and OCL clustered in distinct regions of the PCA plot (Figure 1). Thus, HOT and BOT clustered together in the negative end, whereas MOT and OCL clustered separately in the positive area. This indicates that our data set has the power to discriminate OC (both MOT and OCL) from normal tissue or benign tumors. Moreover, as expected, OC gene expression profiles show a wide dispersion, reflecting tumor heterogeneity. In contrast, HOT and BOT plotted in relatively close proximity, even at the level of individual samples that for the two BOT samples practically overlapped. Finally, as expected, HOT also had some individual sample variability, probably reflecting proportional differences of tissue contents, individual variability or variations in the estrous stage, neither of which was addressed.

Moreover, hierarchical clustering on the basis of relative gene expression also grouped HOT and BOT together, whereas malignancies (both MOT and OCL) clustered together but distant of HOT and BOT (Figure 2). Nonetheless, careful analysis of individual genes revealed some small intragroup differences that could reflect tumor heterogeneity and that deserves further in depth analysis. Interestingly, predominant OC-associated changes in gene expression were suppression rather than overexpression (Additional file 2).

Up to here, the results show two major ovarian tissue gene expression patterns, one specific of OC (cell lines and tumors) and the other one characteristic of borderline tumors and healthy tissue. The gross differences seen probably reflect the relatedness among the different groups. To further explore this, we compared gene expression by grouping apparently related conditions together, against each of the individual or grouped opposites (Table 1). By these means, the highest differences were MOT + OCL vs. BOT + HOT (n = 1799), followed by MOT + OCL vs. HOT (1498 differences), MOT + OCL vs. BOT (1030 differences), BOT + MOT + OCL vs. HOT (545 differences). Finally, gene expression differences between BOT and HOT were minimal (~28). On the basis of these findings, we chose to further examine the most significant differentially expressed exons.

Once we had examined the relative OC-associated gene expression profiles, it was important to examine whether some of the overexpressed genes reflected only quantitative differences or if there were also qualitative differences among them. To achieve this, we performed exon analysis of genes overexpressed in both MOT and OCL. As differential exon usage cannot be easily examined in suppressed genes, we exclusively examined overexpressed genes.

The analysis was performed by means of Alternative Splicing ANOVA and the criteria to select genes for exon analysis were: false discovery ratio (FDR) < 0.05 and fold change >3, in at least one probe set. According to these criteria, the number of overexpressed alternatively spliced genes in OC was 207 (Additional file 3). To identify OC-predominant splice variants, these genes were subjected to internal analysis by comparing the expression of each individual exon for each study group against the mean total expression of the same gene in each of the groups, and were considered only when they yielded ≥3 fold change. Moreover, we performed visual inspection of each of the individual exon expression profile graphs (MOT + OCL vs. HOT + BOT or vs. HOT or vs. BOT, all of which yielded the same genes). This procedure identified at least nine genes of potential interest (Table 2), of which, ZNF695 (encoding a zinc-finger protein) had the highest overexpression (~7 fold, FDR < 0.005) in MOT and OCL, and its mRNA had the highest significant changes in exon expression with significant suppression in one its exons when compared to HOT and BOT (Figure 3). The remaining of this study focuses on the characterization of ZNF695.

ZNF695 encode a zinc finger protein with as yet unknown functions and its gene contains six exons located in chromosome 1q cytogenetic positions 247,148,625-247,171,358. This gene has six possible transcripts of which two (ZNF695-003, and ZNF695-006) encode complete ORFs yielding a 515 and a 172 amino acid length proteins, respectively; whereas the other four transcripts encode products thought to undergo nonsense-mediated decay, a process that detects nonsense mutations and prevents the expression of truncated transcripts (http://​www.​ensembl.​org/​Human/​Search/​Results?​q=​ZNF695;site=​ensembl;facet_​species=​Human).

To characterize transcripts expressed predominantly in OC, we designed primers to identify and clone splice variants of ZNF695 most likely corresponding to the message lengths preferentially expressed in the four initial OC samples. By means of RT-PCR, in a total of 14 OC tissues (10 MOT and four OCL), expression of the three different transcripts was as follows: seven out of 10 tumor samples and three out of four cell lines expressed all three transcripts at variable degree, whereas the remaining samples (three tumors and one cell line) only expressed the two larger transcripts (Figure 4A and B). All three amplicons were gel purified, cloned into pGEM-T Easy Vector (Figure 5A) and confirmed to correspond to the aforementioned ZNF695 RNA transcripts by sequencing (Figure 5B-E). The first one appears to correspond to the full-length product (FLP), as it yields a 400 bp amplicon spanning FLP nucleotides 38 to 433 with full identity to ZNF695-003, ZNF695-006 and no other possible match (Figure 5B-C, Additional file 4). The second transcript spans 360 bp (that we name here ZNF695 transcript 4, see below) fully matching ENSEMBL nonsense-mediated decay transcript ZNF695-002, as well as NCBI peptide BAG54313.1 (Isogai, T. Helix Research Institute, Genomics Laboratory; e-mail: flj-cdna@nifty.com, http://​www.​ncbi.​nlm.​nih.​gov/​protein/​193785160?​from=​1&​to=​118) (Figure 5B, E, Additional file 5). The third transcript (named here ZNF695 transcript 5) found here aligns with both those transcripts up to nucleotide 361, but our primers cannot identify further on the 5’ direction. Although we cannot tell to which of ZNF695 splice variants it corresponds, this transcript contains a sequence partly identical to ZFN695-002 (ZNF695 transcript 4), except that it is missing a 52 bp fragment containing part of the 5’ untranslated region and misses the translation initiation signal (Figure 5D, E, Additional file 4). Although there are additional AUG codons as potential alternative translation initiation signals 3’ of the canonical AUG, these fail to yield useful ORFs. Therefore, most likely, this transcript represents a long non-coding mRNA. Finally, Figure 5F shows the ZNF695 AS model according to the transcripts found in MOT and OCL.