A novel role for the tumor suppressor gene ITF2 in lung tumorigenesis through the action of Wnt signaling pathway

Despite often leading to a platinum resistance, platinum-based chemotherapy continues to be the standard treatment for non-small cell lung cancer (NSCLC) and ovarian cancer. In this study, we used array-CGH and qRT-PCR methodologies to identify and validate cytogenetic alterations that arise after cisplatin treatment in four paired cisplatin-sensitive/resistant cell lines. We report that long-term cell exposure to platinum induces an activation of the Wnt pathway that is concomitant with ITF2 deletion. Restoration of ITF2 expression normalizes the Wnt pathway activation and re-sensitize tumor cells to platinum. Our translational approach analyzing a total of 55 lung and ovarian primary tumors and control samples, showed a frequent downregulation of ITF2 expression. As ITF2 is a negative regulator of the Wnt/β-catenin pathway activity, we used whole transcriptome sequencing (RNA-seq) and Wnt-pathway analysis in a subgroup of NSCLC patients to identify genes with a potential role in the development of this malignancy. Three exhaustive bioinformatics contrasts found three coding genes (HOXD9, RIOX1 and CLDN6) and one long non-coding RNA (XIST) with significant expression differences (FDR<0.05). Further functional assays overexpressing ITF2 in cisplatin-resistant cells suggest the regulation of HOXD9 by the Wnt pathway and its implication on NSCLC progression.


SUMMARY
Despite often leading to a platinum resistance, platinum-based chemotherapy continues to be the standard treatment for non-small cell lung cancer (NSCLC) and ovarian cancer. In this study, we used array-CGH and qRT-PCR methodologies to identify and validate cytogenetic alterations that arise after cisplatin treatment in four paired cisplatinsensitive/resistant cell lines. We report that long-term cell exposure to platinum induces an activation of the Wnt pathway that is concomitant with ITF2 deletion. Restoration of ITF2 expression normalizes the Wnt pathway activation and re-sensitize tumor cells to platinum.
Our translational approach analyzing a total of 55 lung and ovarian primary tumors and control samples, showed a frequent downregulation of ITF2 expression. As ITF2 is a negative regulator of the Wnt/β-catenin pathway activity, we used whole transcriptome sequencing (RNA-seq) and Wnt-pathway analysis in a subgroup of NSCLC patients to identify genes with a potential role in the development of this malignancy. Three exhaustive bioinformatics contrasts found three coding genes (HOXD9, RIOX1 and CLDN6) and one long non-coding RNA (XIST) with significant expression differences (FDR<0.05). Further functional assays overexpressing ITF2 in cisplatin-resistant cells suggest the regulation of HOXD9 by the Wnt pathway and its implication on NSCLC progression.

INTRODUCTION
Although platinum-based chemotherapy still plays an important role in the treatment of many solid tumors, the disease progresses to a platinum-resistant state in a high percentage of the diagnosed cases of Non-small cell lung cancer (NSCLC) and ovarian cancer [1,2] which are two of the most deadly cancers plaguing our society. The former accounting for more than 80% of primary lung-cancer cases and the latter boasting the highest mortality of the gynecological malignancies worldwide [3]. Cisplatin (CDDP) is a platinum compound widely used in the treatment of solid tumors. It induces apoptosis in cancer cells by binding to the N7 position of the guanines and crosslinking DNA [4,5]. However, CDDP also leads to cytogenetic alterations, such as deletions or amplifications of genes involved in tumor progression, metastasis and drug response [6], which contributes to the development of CDDP-resistance [7][8][9].
In this study, we performed high resolution million feature array-Comparative Genomic Hybridization (aCGH) with four NSCLC and ovarian cancer sensitive/resistant paired cell lines previously reported by our group [10] to explore the chromosomic deletions that differ in the resistant subtypes. We found a common deletion that includes the Transcription Factor 4, TCF4 (hereafter called ITF2). ITF2 is a downstream target gene of the Wnt/β-catenin pathway, negatively regulating its activity [11,12]. Wnt signaling has been identified as one of the key signaling pathways in cancer, and more recently, also involved in drug resistance of primary tumors such as colon or ovarian cancer [13,14]. However, its role driving platinum resistance in NSCLC has not been defined yet and very little is also known about how ITF2 is involved in tumorigenesis. Therefore, further study of the role of ITF2 and the Wnt signaling pathway in tumor response to chemotherapy may provide new ways to fight resistance to this popular treatment.
Here we report the frequent downregulation of ITF2 in NSCLC patients and in cisplatinresistant cancer cells. Furthermore, we present the Wnt-signaling pathway as a molecular mechanism behind the development of resistance through the action of ITF2 and affecting the expression of specific genes that might be also used as potential therapeutic target.

ITF2 is frequently downregulated by chromosomal deletion after CDDP cell exposure
Our cytogenetic study showed different genomic alterations in the CDDP resistant subtypes, using the sensitive parental cell lines as reference genome. We found two deleted regions shared by both tumor types, in at least three of the four cell lines (H23R, A2780R and OVCAR3R) located on 18q21.2-18q21.31 and 18q21.32, affecting the genes RAB27B, CCDC68, TCF4, TXNL1, WDR7 and BOD1P; and genes ZNF532, SEC11C, GRP respectively. We also observed a common deleted region on 2q22.1 that included the gene LRP1B in the NSCLC cell lines and an additional common region only shared by the ovarian cancer cell lines on 9q22.33 that included part of the gene TMOD (Supplementary Table 1).
We selected LRP1B and TCF4 (ITF2) genes that were completely deleted in the same tumor type or in at least three of the four cell lines respectively (Figures 1A and Supplementary   1A). The deletion of ITF2 in H23R and A2780R cell lines resulted in a significant loss of ITF2 expression compared to the sensitive subtypes, validating the results obtained in the arrays CGH ( Figure 1B). No changes were observed neither in OVCAR3 cells for ITF2 expression nor in H23 and H460 cells for LRPB1 ( Figure 1B and Supplementary Figure 1B).

Wnt canonical signaling pathway is increased in CDDP resistant cells with ITF2 deletions
Due to the association of ITF2 with the Wnt/Β-catenin/TCF pathway, we studied the transcriptional activity of the Wnt pathway in H23S/R and A2780S/R cells, in which the Cisplatin-resistant phenotypes harbors the ITF2 deletion. We transfected cells with the Wnt reporters (Super8xTop-Fop vectors) and induced the Β -catenin/TCF4 activity either by LiCl 4 treatment, which inhibits GSK3-b, or by co-transfection with the constitutively stable Β catenin-S33Y mutant. We observed higher luciferase activity in A2780R cells compared with the parental sensitive ones, indicating an increased transcriptional activity of Β -catenin/TCF in response to both pharmacological and functional activation of the pathway (Figure 2A).
This effect was not observed in H23R cells ( Figure 2B). We also observed an increase in the gene expression levels of DKK1 and more slightly in Cyclin D1, in A2780 cells two of the four analyzed downstream effectors genes of this pathway ( Figure 2C).

Restoration of ITF2 increases the sensitivity to CDDP by decreasing β -catenin/TCF transcriptional activity
To test the role of ITF2 in cisplatin resistance, we transiently overexpressed ITF2 cDNA in A2780 cells, in which our previous results confirmed the ability to evaluate changes in the transcriptional activity of the Wnt pathway. Overexpression of ITF2 in A2780R resulted in a significant increase in sensitivity to cisplatin from the dose of 0,5ug/ml (p<0.01), showing an intermediate phenotype between the resistant and sensitive subtypes ( Figure 3A). In addition, ITF2 restoration induced a dramatic decrease in cell viability (p<0.05) 24 hours after transfection compared with the parental resistant cells transfected with the empty vector ( Figure 3B). ITF2 overexpression at 24 and 72 hours after transfection was confirmed by qRT-PCR ( Figure 3C). Moreover, ITF2 overexpression recovered the levels of Bcatenin/TCF transcriptional activity observed in sensitive cells ( Figure 3D). In fact, the expression of downstream target DKK1 was restored from 24h and maintained 72h after ITF2 overexpression (R-ITF2), similar results were also observed at 72h for Cyclin D1.

The expression of ITF2 is frequently downregulated in NSCLC and ovarian tumor samples
To validate our in vitro results we determined the clinical implication of ITF2 and DKK1 expression in NSCLC and ovarian cancer patients. The relative expression of both genes 5 was measured in two cohorts of fresh frozen tumor samples (T) and adjacent tissue (ATT) from NSCLC (Table 1) and ovarian cancer patients (Supplementary Table 2).
We observed that ITF2 expression is frequently downregulated in NSCLC and ovarian tumor samples ( Figure 4) validating our in vitro data. Fifteen out of 25 tumor samples of NSCLC patients had lower expression of ITF2 compared to the normal lungs mean (NLM) ( Figure   4A). Furthermore, as reported in our experimental data, we observed the opposite expression profile between ITF2 and DKK1 in 60% of NSCLC samples. However, this situation was found only in approximately 10% (1 out of 9) of the ovarian cancer samples ( Figure 4B). We did not observe differences between ATT and normal lung samples (LC) for ITF2 (p=0.177) and DKK1 (p=0.693) in the NSCLC cohort ( Figure 4A).
The Kaplan Meier curves, analyzing the overall survival (OS) according to the median of ITF2 and DKK1 expressions, showed that NSCLC patients with high ITF2 expression tend to have better overall survival, p=0.1 ( Figure 4C). No differences in survival were observed for DKK1 expression p=0.6 (Supplementary Figure 2A). These results were statistically confirmed in 1 926 lung cancer patients by using the Kaplan Meier plotter online tool, noticing that those patients with high expression of ITF2 (Supplementary Figure 2B) and low expression of DKK1 (Supplementary Figure 2C) had a significantly better overall survival rate (p=0.016 and p<0.001, respectively).

Identification of candidate genes involved in the Wnt signaling pathway through the analysis of RNA-seq in NSCLC patients
To further explore the role of the Wnt-signaling pathway in lung cancer tumorigenesis, we performed a whole transcriptome analysis performing RNA-seq on 14 samples including nine NSCLC samples, six of them with an inverse expression profile between ITF2 and DKK1 (Pat3, Pat6, Pat10, Pat22, Pat25 and Pat26) and three with the same expression profile (Pat8, Pat16 and Pat18). Three ATT (Pat9, Pat21 and Pat25) and two normal lung 6 samples (LC1 and LC2), were considered as controls for comparisons. ITF2 was mainly downregulated in NSCLC patients, while DKK1 showed a more heterogeneous expression pattern. Therefore, we considered DKK1 as the best parameter to decide the bioinformatics analysis of the RNA-seq. This analysis focused on three main contrasts: contrast A, differential gene expression analysis between tumors and controls, contrast B, comparison between tumors with high and low expression of DKK1, and contrast C comparison of the tumor samples with high expression of DKK1 with the controls (Supplementary Figure 3).
We selected those genes that showed significant expression differences (FDR<0.05) in at least two of the three contrasts, prioritizing contrast B (Supplementary Figure 3 and Supplementay Table 3). The bioinformatics analysis also focused in all annotated genes related to the Wnt-pathway.
We analyzed the expression of 9 candidates, but only four of them were detected by qPCR-PCR. Validation of RNA-sequencing results was performed in three coding genes (HOXD9, RIOX1 and CLDN6) and one long non-coding RNA (XIST). An accurately correlation with the RNA-sequencing data was found for HOXD9, CLDN6 and XIST genes (r=0.83, r=0.97 and r=0.97, respectively) ( Figure 5A-C), while for RIOX1 the correlation coefficient was less marked, probably because of the sample size (r=0.58) ( Figure 5D). From all four candidates, only HOXD9 and RIOX1 expression showed correlation with ITF2 expression, (Pearson=-0.24 and Pearson=0.68, respectively) ( Figure 5E and 5F). No correlation was found for XIST or CLDN6 (Supplementary Figure 4).
In order to gain insight into the role of ITF2 regulating the expression of the selected candidates in NSCLC, we overexpressed ITF2 in H23 lung cancer cells. Transfection efficiency was confirmed by qRT-PCR at 24 and 72 hours after transfection ( Figure 6A). As expected from the primary tumors results, the overexpression of ITF2 induced a significant decrease of HOXD9 (p=0.04) and an increase of RIOX1 expressions reaching the levels of the sensitive cells 72 hours after transfection ( Figure 6B). Having identified HOXD9 and RIOX1 as potential Wnt pathway candidate genes regulated by ITF2, we studied their clinical translational application in the cohort of NSCLC patients used for the RNA-seq analysis. 7 A negative correlation for HOXD9 and a positive correlation for RIOX1 was found in terms of patients overall survival ( Figure 6C and 6D). In addition, the survival analysis performed by the Kaplan Meier method after stratifying patients according to the median of both genes expression showed that patients with lower expression of HOXD9 present a significant better overall survival rate (p=0.046), whereas patients with low RIOX1 expression tend to live less, although no statistical significance was found in this case ( Figure 6E and 6F).

DISCUSSION
Wnt signaling has been recently reported to be involved in driving platinum resistance of several tumor types [13,15]. However, the molecular mechanisms implicated are not clear, specially in NSCLC. In the present work, we have studied the involvement of the Wnt signaling pathway in tumorigenesis through a combined experimental approach by using both CGH arrays and RNA-sequencing. We have found that long term exposure to platinum induces a frequent deletion of ITF2 that is involved at least in part in the activation of the Wnt pathway.
We firstly identified a common deletion in H23, OVCAR3 and A2780 cells, induced by cisplatin treatment in chromosome 18, including two completely deleted genes, LRP1B and ITF2. A similar deletion was identified in a previous study where they analysed the cisplatin response in ovarian cancer samples, supporting our results [16]. There is however, another study analyzing the cytogenetic alterations of CDDP-resistant A2780R cells, which shows different genomic alterations. This was probably due to the specificity of the CGH-array used and the experimental design that included less representative number of probes and was performed in only one cell line [8]. We were not able to validate the LRP1B expression changes by using an alternative technique, situation that has been previously reported [17].
In our case, it could be due to the mosaicism observed in this region that occurs in less than 20% of the resistant cells. The level of mosaicism that can be detected is dependent on the sensitivity and spatial resolution of the clones and rearrangements present [18].
8 Nevertheless, LRP1B could still play a role in tumor progression as several studies link its downregulation through deletion and carcinogenesis [19][20][21]. ITF2 expression changes were confirmed in H23R and A2780R but not in OVCAR3R cells, also probably due to the level of mosaicism (36%) observed in these cells. Our results indicate that low levels of mosaicisms would make the validations of expression changes by another quantitative technique difficult, probably because the alteration at expression levels are not significant enough to be detected.
The fact that ITF2 is deleted and downregulated after platinum treatment, provides us with a new insight regarding its importance in resistance to platinum chemotherapy in lung and ovarian cancer. Supporting our results a previous study using targeted sequencing in a PDX-based modeling of breast cancer chemoresistance, identified a genomic variant of ITF2 that depicted a link between its altered expression and breast cancer chemoresistance, although no detailed mechanism was provided to connect ITF2 function to chemoresistance ITF2 is a transcription factor belonging to the basic Helix Loop Helix (bHLH) family, which can act as a transcriptional activator or repressor [23, 24] but its regulation still is unknown. It is important to distinguish ITF2, whose real name is TCF4, from the T-Cell Factor 4 (TCF7L2), also known as TCF4, which is the bcatenin transcriptional partner [25]. In fact, ITF2 expression is induced by the Β -catenin/TCF complex, but at the same time, it acts as a repressor of this complex by interfering with the binding of Β -catenin to TCF4. This causes a decreases in the expression of Wnt target genes, leading to the repression of cell proliferation [12]. Consistent with these studies, we have observed that the resistant A2780 cells have an increased activity of the Β -catenin/TCF transcription, which is concomitant with the increased expression of the downstream effector gene DKK1, probably due to the absence of ITF2. In contrast, we did not observe differences in H23 cells, which could be explained by previous observations [26], confirming that H23 has a high basal activity in the Wnt signaling pathway and therefore exogenous activation may not show a difference.
Moreover, we have observed that the overexpression of ITF2 in A2780R cells leads to a 9 decrease in cell viability, rescuing the sensitive phenotype maybe through the inhibition of the excessive proliferation and the activity levels of the Β -catenin/TCF transcription. In fact, our results indicate that resistant cells respond better to the activation of the Wnt pathway, an effect that is restored after the re-expression of ITF2. Therefore, the Wnt signaling pathway plays an important role in the resistance to cisplatin through ITF2 in cancer cell lines.
Our translational analysis, based on the expression levels of ITF2 and DKK1 genes in two different cohorts of patients was aimed to elucidate the role of this pathway in tumor progression and chemotherapy response. ITF2 expression was frequently downregulated in NSCLC and ovarian tumor samples, validating our in vitro data. The expression levels of DKK1, however, showed a more heterogeneous pattern in the NSCLC tumor samples, while no differences were observed in the ovarian tumors, suggesting an aberrant activation of the Wnt signaling pathway in lung cancer. In fact, our in silico analysis of 1 926 NSCLC patients indicates a significant increased overall survival associated with high expression levels of ITF2 and low expression of DKK1. The same findings without statistical significance were observed from our "in house" cohorts, probably due to the sample size. However, one of the strengths of our cohort is that is comprised by fresh frozen samples, enabling us to perform high quality RNA-sequencing in a group of NSCLC patients, in order to determine the involvement of the Wnt-pathway components in lung cancer development. The differential expression of DKK1 within the tumor samples allowed us to perform three different bioinformatics contrasts in order to explore all the possibilities regarding tumor development and the Wnt signaling pathway. Contrast A aimed to identify genes with a possible involvement in lung cancer development by comparing differential expression in tumors versus control samples. Contrast B was made to identify alterations in the Wnt pathway in NSCLC tumors and those that could be used as potential therapeutic targets. Finally, contrast C was able to identify genes regulated by the Wnt pathway and others involved in NSCLC development. Using this approach we were able to identify coding genes, noncoding genes and transcripts that had not been functionally characterized previously [27, 10 28]. Indeed, in this study we have identified three coding genes, HOXD9, CLDN6 and RIOX1, and one non-coding gene, XIST, which could be involved in NSCLC progression through the Wnt signaling pathway. Our data was validated by two alternative methodologies, both showing strong positive correlations of three of the candidates HOXD9, CLDN6 and XIST and a slightly weaker correlation for RIOX1.
HOXD9 and CLDN6 were significantly downregulated in tumors with high expression levels of DKK1 and upregulated in tumors compared with controls, indicating that these genes could be involved in tumor progression through an aberrant activation of the Wnt signaling pathway. In the case of HOXD9, the tumor samples had higher expression levels than the controls, as it has been previously reported [29]. We also observed a negative correlation of HOXD9 and ITF2 expression levels. In addition, patients with lower levels of HOXD9 had better overall survival than those with upregulated expression of this gene.
These results are consistent with previous studies linking a high expression of HOXD9 with glioblastoma and hepatocarcinoma [30,31]. Our functional analysis showed that ITF2 overexpression in lung cancer cells H23R decreased the expression of HOXD9, while we expected a recovery of its expression to the sensitive subtype levels. Therefore we believe that an alternative regulatory mechanism affected by ITF2 is modulating the expression of Although little is known about RIOX1, it has recently been shown to be involved in renal and colorectal carcinogenesis [39,40]. These results envision these two candidates, who undoubtedly play a role in tumorigenesis, as potential therapeutic targets. In fact, it is becoming clear in the field that a gene can exhibit a double function in cancer, being involved in both tumor development as well as in the response to an antitumor drug, such as MGMT, IGFBP-3, MAFG genes or even miRNAs like miR7 [10, [41][42][43][44] In essence, we have identified ITF2 as a frequently downregulated gene in cisplatinresistant cancer cells as well as in NSCLC and ovarian cancer patients. Moreover, we define the activation of the Wnt-signaling pathway as a molecular mechanism behind the development of cisplatin resistance in cancer cells through the action of ITF2, providing novel insights into the molecular biology and the cellular mechanisms involved in the acquired resistance to the most widely-used chemotherapy agent, cisplatin. Additionally we have suggested two potential therapeutic targets for further study, ITF2 and HOXD9.

Cell culture and cell-viability assays
The NSCLC and ovarian cancer cell lines H23, H460, OVCAR3 and A2780 were purchased from the ATCC (Manassas, Virginia, USA) and ECACC (Sigma-Aldrich, Madrid, Spain) and cultured as recommended. Their CDDP-resistant variants H23R, H460R, OVCAR3R and A2780R were previously established in our laboratory [10,45]. Cisplatin (Farma Ferrer, Barcelona, Spain) was used for CDDP-viability assays. Cells were seeded in 24-well dishes at 40,000 cells/well, treated with increasing doses of CDDP (0, 0.5, 1, 1.5, 2 and 3µg/ml) for an additional 72 or 48 hours and stained as described [46]. Cell viability comparing sensitive vs. resistant cell lines was estimated relative to the density recorded over the same experimental group without drug exposure at same period of time. Cell authentication is included in Supplementary Table 4.

Clinical sample and data collection
We selected a representative number of fresh frozen surgical specimens from University pathological and therapeutic data were recorded by an independent observer and a blind statistical analysis was performed on these data.

DNA extraction and array of Comparative Genome Hybridization
DNA from cell lines was isolated as previously described [47] and used to analyze copy  Supplementary Table 5.

NGS (RNA-seq) and Wnt signaling pathways analysis
Total RNA from nine tumor tissues, three lung adjacent normal tissue (ATT) from NSCLC samples and two tissue samples of non-neoplastic origin from autopsies were sent to Sistemas Genómicos Company (Valencia, Spain) for RNA-sequencing. Library samples were prepared and sequenced as recommended by the manufacturer (Illumina, San Diego, California, USA) deeply described in the GEO repository number GSE127559. The bioinformatic analysis was performed in the HULP. Reads were analyzed to quantify genes and isoforms through the RSEM-v1.2.3 methodology (RNA-seq by Expectation Maximization) [48] and using the hg19 versions as reference for annotation. The differential expression was carried out with edgeR, which can estimate the common and individual dispersion (CMN and TGW, respectively) to obtain the variability of the data [49]. p-values and FDR statistical analysis were performed by Cmn and Twg models and the statistical cutoff point was set as FDR<0.05. Normalization was performed by the TMM method (Trimmed mean of M-values) [50]. The bioinformatics analysis also included an efficiency analysis for every sample, considering the total efficiency as the percentage of reads annotated belonging to a transcript regarding the total fragments initially read. When using Principal Component Analysis (PCA), no differences were observed between samples from nonneoplastic autopsies and adjacent normal tissue (ATT) from NSCLC patients in terms of transcriptmic profile, therefore both types of samples were considered as a reference group for the differential expression analysis (Supplementary Figure 1). Three different bioinformatics contrasts are described in detail in the Results section.

Statistical analysis
Data          P a t 9 . A T T P a t 2 1 . A T T P a t 2 5 . A T T L C 1 L C 2 P a t 3 . T P a t 6 . T P a t 8 . T P a t 1 0 . T P a t 1 6 . T P a t 1 8 . T P a t 2 2 . T P a t 2 5 . T P a t 2 6 . T P a t 9 . A T T P a t 2 1 . A T T P a t 2 5 . A T T L C 1 L C 2 P a t 3 . T P a t 6 . T P a t 8 . T P a t 1 0 . T P a t 1 6 . T P a t 1 8 . T P a t 2 2 . T P a t 2 5 . T P a t 2 6 . T P a t 9 . A T T P a t 2 1 . A T T P a t 2 5 . A T T L C 1 L C 2 P a t 3 . T P a t 6 . T P a t 8 . T P a t 1 0 . T P a t 1 6 . T P a t 1 8 . T P a t 2 2 . T P a t 2 5 . T P a t 2 6 . T P a t 9 . A T T P a t 2 1 . A T T P a t 2 5 . A T T L C 1 L C 2 P a t 3 . T P a t 6 . T P a t 8 . T P a t 1 0 . T P a t 1 6 . T P a t 1 8 . T P a t 2 2 . T P a t 2 5 . T P a t 2 6 . T P a t 9 . A T T P a t 2 1 . A T T P a t 2 5 . A T T L C 1 L C 2 P a t 3 . T P a t 6 . T P a t 8 . T P a t 1 0 . T P a t 1 6 . T P a t 1 8 . T P a t 2 2 . T P a t 2 5 . T P a t 2 6 . T