ONECUT2 overexpression promotes RAS-driven lung adenocarcinoma progression

Aberrant differentiation, driven by activation of normally silent tissue-specific genes, results in a switch of cell identity and often leads to cancer progression. The underlying genetic and epigenetic events are largely unexplored. Here, we report ectopic activation of the hepatobiliary-, intestinal- and neural-specific gene one cut homeobox 2 (ONECUT2) in various subtypes of lung cancer. ONECUT2 expression was associated with poor prognosis of RAS-driven lung adenocarcinoma. ONECUT2 overexpression promoted the malignant growth and invasion of A549 lung cancer cells in vitro, as well as xenograft tumorigenesis and bone metastases of these cells in vivo. Integrative transcriptomics and epigenomics analyses suggested that ONECUT2 promoted the trans-differentiation of lung cancer cells by preferentially targeting and regulating the activity of bivalent chromatin domains through modulating Polycomb Repressive Complex 2 (PRC2) occupancy. Our findings demonstrate that ONECUT2 is a lineage-specific and context-dependent oncogene in lung adenocarcinoma and suggest that ONECUT2 is a potential therapeutic target for these tumors.

Here, we report that ONECUT2 is a lineage-specific and context-dependent oncogene in lung adenocarcinoma. Targeted manipulation of histone methylation orchestrated by the oncogenic transcriptional regulation of ONECUT2 contributes to the epigenetic reprogramming, aggressive behavior and metastasis of RAS-driven lung cancer.

Results
ONECUT2 is aberrantly activated in lung cancer. We first analyzed ONECUT2 expression across various subtypes of lung cancer using publicly available dataset GSE30219 20 . ONECUT2 was highly expressed in small cell lung cancer (SCLC), large cell neuroendocrine lung cancer (LCNEC) and lung carcinoid tumor (Carcinoid), a subtype with low-grade malignancy (Fig. 1A). Since these three subtypes are all manifested by neuroendocrine differentiation, we suspected that ONECUT2 was a neuroendocrine differentiation gene in lung tumors. Indeed, ONECUT2 expression correlated with ASCL1, a well-defined master regulator of neuroendocrine differentiation, in SCLC and LCNEC (Fig. 1B,C, Supplementary Fig. 1A). Moreover, both ONECUT2 and ASCL1 are significantly upregulated in a mouse model of small cell lung cancer where Rb1 and Trp53 are both knocked out (Fig. 1D), indicating a shared regulation. Considering ASCL1 had been proved to be a lineage-specific oncogene in high-grade neuroendocrine lung cancer and high expression of ONECUT2 in lung carcinoid tumor, we believed that ONECUT2 was unlikely a driver oncogene in lung tumor with neuroendocrine differentiation.
Besides SCLC, LCNEC and carcinoid, ONECUT2 was also highly expressed in a subset of lung adenocarcinoma (ADC) (Fig. 1A), the most frequent subtype of lung cancer. Analysis of TCGA-LUAD dataset indicated that ONECUT2 expression was elevated in cancerous tissue compared with paired adjacent normal tissues (Fig. 1E). Moreover, tissue microarray of lung adenocarcinoma and matched adjacent normal specimen showed that ONECUT2 was stained in 21 out of 75 tumor samples (nuclear positivity of 28%) ( Supplementary Fig. 1B). While ASCL1 was also overexpressed in a subset of lung adenocarcinoma, the expression of ONECUT2 and ASCL1 were barely correlated (Fig. 1F, Supplementary Fig. 1C). Furthermore, neither ASCL1 nor ONECUT2 overexpression was associated with poor prognosis in ADC (Fig. 1G, Supplementary Fig. 1D-F). By integrated analysis of gene expression and genomic alteration profiles of TCGA ADC samples, we discovered that ONECUT2 overexpression was significantly associated with shorter overall survival in RAS-driven ADC patients (Fig. 1H) or patients with oncogenic alterations in the RAS signaling pathway (RAS activated) (Fig. 1I) in the presence of TP53 loss of function alterations (See the definition of RAS-driven and RAS-activated tumor in the Methods section). However, ASCL1 overexpression was not associated with overall survival in the corresponding genomic alteration context ( Supplementary Fig. 1G,H). These data implied that ONECUT2 was a context-dependent prognostic factor and oncogene in lung adenocarcinoma.

Overexpression of ONECUT2 promotes malignant growth and invasion of A549 cells in vitro.
We hypothesized that ONECUT2 promoted RAS-driven lung adenocarcinoma malignancy. We chose A549, a human lung adenocarcinoma cell line with homozygous KRAS G12S mutation, to investigate the role of ONECUT2 in lung cancer progression. Retroviral transduction of the ONECUT2 gene resulted in A549 cells with stable overexpression of ONECUT2 (A549-ONECUT2) (Fig. 2A). A549 cells showed profound morphological changes upon ONECUT2 overexpression, with elongated cell bodies, extended dendrites and loosened cell-cell contacts   2B). Compared with A549 cells transduced with the empty vector (A549-E.V. cells), A549-ONECUT2 cells showed a lower rate of proliferation in low density two-dimensional cell culture, but maintained a high proliferation rate even when confluent, suggesting a loss of contact inhibition (Fig. 2C). Indeed, A549-ONECUT2 cells formed more colonies in soft agar (Fig. 2D,E), indicating a gain of malignant growth.
Next, we examined the effect of ONECUT2 overexpression on the migratory behavior of A549 cells. As expected, overexpression of ONECUT2 significantly promoted the migration and invasion of A549 cells, as demonstrated by a wound healing assay (Supplementary Fig. 2A) and an in vitro two-chamber Matrigel transwell invasion assay (Fig. 2F,G), respectively. Because loss of cell-cell contact and establishment of cell-substrate interaction are critical steps in the invasion of cancer cells, we further tested the changes in cell-matrix contact in response to ONECUT2 overexpression. As illustrated in Supplementary Fig. 2B, A549-ONECUT2 cells adhered much more readily to type I collagen, type IV collagen, and Matrigel than A549-E.V. cells, suggesting a remodeling of the cell-matrix interaction. Overall, we concluded that overexpression of ONECUT2 promoted the malignant growth, migration, invasion and adhesion of A549 lung cancer cells.

Overexpression of ONECUT2 promotes xenograft tumorigenesis and bone metastasis of A549 cells in vivo.
We further investigated the function of ONECUT2 overexpression in vivo. To this end, we made a truncated mutant of ONECUT2 lacking both the Cut and Homeobox DNA-binding domainsdesignated ONECUT2(ΔDBD) -and transduced A549 cells (Fig. 3A,B). We inoculated A549-ONECUT2, A549-ONECUT2(ΔDBD) and A549-E.V. cells subcutaneously at the base of the left forelimb in nude mice. www.nature.com/scientificreports www.nature.com/scientificreports/ A549-ONECUT2 cells, compared with A549-E.V. cells and A549-ONECUT2(ΔDBD) cells, gave rise to much larger tumors (as measured by both tumor mass and volume) over a period of 40 days (Fig. 3C-E). The ΔDBD mutant did not confer A549 with greater tumorigenesis potential, suggesting that DNA binding was essential for ONECUT2 function.
Next, we analyzed the role of ONECUT2 in metastasis by injecting A549-ONECUT2, A549-ONECUT2(ΔDBD) and control A549-E.V. cells intracardially into nude mice. Overexpression of ONECUT2 but not its ΔDBD mutant significantly enhanced bone metastasis of A549 cells as revealed by bioluminescent imaging (BLI), X-ray analysis, micro-CT scan and histological examination (Fig. 3F,G, Supplementary Fig. 3). Tumor cells colonized exclusively the spine and hind limbs. The metastases manifested as severe osteolytic bone lesions and massive bone destruction, as demonstrated by X-ray and CT scans (Fig. 3G). Hematoxylin and eosin staining of the metastatic sites showed massive invasion of tumor cells and distortion of normal bone structure. Importantly, mice bearing tumors composed of A549-ONECUT2 cells following intracardial injection had a significantly shorter lifespan than those bearing tumors composed of A549-ONECUT2(ΔDBD) or control cells (Fig. 3H). These data collectively indicated that ONECUT2 overexpression enhanced osteolytic bone metastasis in vivo and that this effect was dependent on ONECUT2's DNA binding domains.
Overexpression of ONECUT2 activates oncogenic pathways and lineage-specific genes. To understand the molecular events underlying ONECUT2, we set out to investigate the transcriptional targets modulated by ONECUT2 using A549-ONECUT2 and A549-ONECUT2(ΔDBD) cells. By RNA-Seq, we identified 667 genes upregulated and 618 genes downregulated in A549-ONECUT2 cells relative to A549-ONECUT2(ΔDBD) cells (Supplementary Table 1). We used DAVID to analyze the Gene Ontology biological pathways significantly enriched in the differentially expressed genes (DEGs). Enrichment map categorized enriched pathways with significant overlapping genes into several functional clusters (Fig. 4A). Specifically, genes in neuron development, cell locomotion and migration, angiogenesis, proliferation and extracellular matrix organization pathways were enriched in upregulated genes. By contrast, downregulated genes were mainly associated with inflammatory responses and chemotaxis.
We further investigated the protein-protein interaction networks of the DEG products using the STRING database. The subnetworks of ONECUT2 downstream targets are shown in Supplementary Fig. 4A-E. We found predominant and dense interaction networks of genes engaged in neural differentiation, including the well-characterized neuroendocrine regulator somatostatin (SST) and its receptor, as well as CXCR4 and CXCR7, two chemokine receptors well-implicated in cancer metastasis ( Supplementary Fig. 4A). We also discovered a subnetwork with repressed E-cadherin (CDH1) and activated OB-cadherin (CDH11) at the center, indicating a "cadherin switch" in response to ONECUT2 overexpression ( Supplementary Fig. 4B). Another noteworthy network was characterized by increased matrix metalloproteinase 1 (MMP1) expression and modulation of several extracellular matrix proteins, thus indicating a potential role of ONECUT2 in cell-matrix interaction and remodeling ( Supplementary Fig. 4C). Wnt signaling pathways were also significantly perturbed, as evidenced by altered expression of several WNT modulators, such as CER1, GLI3, HEY1 and DKK2 ( Supplementary Fig. 4D). Finally, the tyrosine kinase receptor gene PDGFRA was significantly upregulated along with its genetic and physical neighbors, further underscoring ONECUT2's oncogenicity ( Supplementary Fig. 4C). These data suggested that ONECUT2 overexpression initiated transcriptional activation programs enhancing the malignancy of lung adenocarcinoma cancer cells.
By comparing DEGs with gene sets in MSigDB, we found genes upregulated by ONECUT2 overexpression significantly overlapped with genes bearing bivalent histone marks, PRC2 targets and hallmark genes of epithelial-to-mesenchymal transition (Fig. 4B-D). Moreover, consistent with the lineage-restricted expression pattern of ONECUT2, ONECUT2 activated genes specifically expressed in neural, hepatobiliary and intestinal tissues, suggesting the activation of trans-differentiation programs in ONECUT2-overexpressed lung cancer cells (Fig. 4E).

ONECUT2 preferentially occupies and modulates the activity of bivalent promoters.
To study the mechanism of transcription regulation by ONECUT2, we used an anti-hemagglutinin (anti-HA) antibody to isolate protein-DNA complex in A549-ONECUT2 cells and high-throughput sequencing (ChIP-Seq) to reveal its DNA binding pattern. We identified over 10800 high confidence binding peaks (p < 1e-8 and fold enrichment >4) of ONECUT2 in A549 cells. Motif analysis showed a significant enrichment of hexanucleotide ATC(G/A)AT in the vicinity of the high confidence peaks (Fig. 5A). ATC(G/A)AT has been reported to be the binding motif for ONECUT1 and ONECUT2 21,22 , validating the efficacy and specificity of our assay.
We assigned ONECUT2 peaks to genes by using GREAT (v2.0). We found that a significant portion of peaks were located more than 50 kb away from transcription start sites (TSS) (Fig. 5B). We defined a gene as an ONECUT2-activated or ONECUT2-repressed target when it was associated with at least one high confident ONECUT2 binding peak within 100 kb around its TSS and when the gene was significantly upregulated or downregulated (log 2 of fold change (FC) >2 or <−2). We found 301 activated and 107 repressed target genes.
Next, we annotated the chromatin states around TSS of ONECUT2 target genes, by using the chromatin state model of A549 cells from ENCODE and Roadmap. We found that a significant portion (37.6%) of the TSS of activated target genes but not those of repressed genes (10.1%) showed chromatin marks typical of bivalent chromatin domains (Biv) and Polycomb-repressed regions (ReprPC) (Fig. 5C). Bivalent chromatin regions are characterized by H3K4me3 and H3K27me3 at the same location, where Polycomb-repressed regions are marked mainly with repressive H3K27me3 modification. At the peak level, activating peaks were enriched for Polycomb-repressed regions (ReprPC).
Given the evidence that ONECUT2 binding sites were preferentially associated with H3K27me3 modification and ONECUT2 activated genes overlapped with PRC2 targets (Fig. 4B), We speculated that ONECUT2 (2019) 9:20021 | https://doi.org/10.1038/s41598-019-56277-2 www.nature.com/scientificreports www.nature.com/scientificreports/ might activate gene expression by modulating the balance of H3K4me3 and H3K27me3 around the TSS of activated target genes. Thus, we profiled H3K4me3 and H3K27me3 modifications in A549-ONECUT2 and A549-ONECUT2(ΔDBD) cells by ChIP-Seq. We found a significant gain of H3K4me3 and loss of H3K27me3 around activating ONECUT2 peaks and TSS of activated targets, and no alterations in either modification around repressive ONECUT2 peaks, (Fig. 5D-G). TSS of ONECUT2 activated target genes were marked with ONECUT2 binding, loss of H3K27me3 and gain of H3K4me3 (Fig. 5H). These results suggest that ONECUT2 activates gene expression by increasing H3K4me3 and decreasing H3K27me3 around the TSS of its targets. ONECUT2 activates gene expression by modulating PRC2 occupancy. We next set out to determine the epigenetic machinery responsible for the changes in histone modification around ONECUT2-activated targets. Repressive H3K27me3 mark is maintained by PRC2 complex. We found EZH2 and SUZ12, two core components of PRC2, was upregulated upon ONECUT2 overexpression. EED, another component of the complex, was slightly downregulated (Fig. 6A). While abundance of PRC2 components were altered, the overall H3k27me3 level was not changed (data not shown), indicating a site-specific rather than systemic chromatin modification change.
We modulated PRC2 activity by treating A549 cells with GSK126 (5 µM for 5 days), a specific inhibitor of the EZH2. We found that genes upregulated by GSK126 were significantly overrepresented by ONECUT2-activated targets or ONECUT2-upregulated genes (Fig. 6D,E). Importantly, genes upregulated by GSK126 treatment www.nature.com/scientificreports www.nature.com/scientificreports/ significantly overlapped with genes upregulated by the genetic ablation of Eed in a mouse model of Kras-driven lung adenocarcinoma 9 (Fig. 6F). Similarly, genes upregulated by Eed loss in Kras activated and Tp53 deleted cells were significantly overrepresented by ONECUT2-activated targets or ONECUT2-upregulated genes (Fig. 6G,H). These data strongly support the context-dependent oncogenicity of PRC2 and ONECUT2 in lung adenocarcinoma.

Discussion
In this study, we reveal a context-dependent oncogenic role and epigenetic basis for the transcriptional regulation of ONECUT2 in lung adenocarcinoma. To the best of our knowledge, our work is the first systematic investigation of ONECUT2 in lung cancer progression. www.nature.com/scientificreports www.nature.com/scientificreports/ ONECUT2 has early been implicated in EMT and cancer cell invasion in colorectal cancer 23 . Recent studies revealed ONECUT2 is a driver of neuroendocrine prostate cancer and a targetable master regulator of lethal prostate cancer 18,19 . However, High level of heterogeneity of lung cancer put the investigation of a new oncogene in a much more complicated situation. ASCL1 is a master regulator for neuroendocrine lung cancer 24 , but not for neuroendocrine prostate cancer. The fact that ASCL1 and ONECUT2 are both upregulated in Rb1/TP53 knockout mice and high correlation of their expression in neuroendocrine lung cancer make it difficult to discern their functional relationship without complex genetic mouse models. However, in lung adenocarcinoma, ASCL1 expression seldom overlaps with ONECUT2. After in-depth investigation, we found ONECUT2, but not ASCL1, was a prognostic gene in lung adenocarcinoma with both RAS activated and TP53 loss of function. RAS signaling pathway is the most frequently activated oncogenic pathway in lung adenocarcinoma, accounting for over one thirds of patients. Given the previous findings that Eed loss promoted lung adenocarcinoma progression and was associated with poor prognosis in KRas G12D/+ ; Trp53 fl/fl , but not in KRas G12D/+ mice, and the potential regulation of PRC2 by ONECUT2 revealed by our study, we believe ONECUT2 is one of the genetic programs that confer PRC2 tumor-suppressive roles in RAS-driven lung cancer. Moreover, since PRC2 binds to chromatin without sequence specificity, our data indicated a site-specific modulation of PRC2 by ONECUT2. The context-dependent oncogenic role of ONECUT2 should be further investigated using genetic engineered mouse models. www.nature.com/scientificreports www.nature.com/scientificreports/ Our transcriptome and epigenetic analysis both indicated ONECUT2 activated lineage-specific genes which correlated with ONECUT2's tissue specificity. Of note, Nkx2-1 haploinsufficiency promotes mucinous LUAD and activation of enteric lineage markers that are normally repressed in alveolar cells 6 . Eed and Trp53 loss, together with Kras activation, collectively promote formation of an aggressive mucinous adenocarcinoma 9 . Given the significant overlap of genes upregulated by Eed loss in the context of Kras activation and Trp53 inactivation and genes upregulated by ONECUT2 overexpression in A549 cells, we speculate that ONECUT2 overexpression induces aberrant mucinous differentiation, which, in combination with KRAS and TP53 mutations, promotes aggressive behaviors of lung adenocarcinoma. This speculation warrants further studies.
One limitation of our study is the use of single cell line and lack of loss of function study. We chose A549 as a model cell line to investigate the role of ONECUT2. We found that overexpression of ONECUT2 in immortalized human bronchial epithelial cell did not result in growth advantage or increased invasion (data not shown). We have also tested overexpression in other RAS-driven lung cancer cell lines. Of all, A549 is the only cell line which can be efficiently transdifferentiated. This implies a high level of heterogeneity in terms of genetic context and barrier to trans-differentiation in lung cancer cell lines. Unfortunately, we could not find a suitable cell line with both oncogenic KRAS mutation and intrinsically high ONECUT2 expression to carry out loss of function study. Further studies should validate the context-dependent oncogenicity of ONECUT2 in a broader system.
In summary, our study demonstrates ONCUT2 promotes RAS-driven lung cancer progression by modulating PRC2 occupancy. ONECUT2 may be an important therapeutic target in lung adenocarcinoma.

Materials and Methods
Cell line culture and reagents. A549 cells were obtained from the ATCC in 2012. A549 and human embryonic kidney HEK293T cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM, Invitrogen) containing 10% fetal bovine serum (FBS, Gibco) supplemented with penicillin (100 U/ml) and streptomycin (100 mg/ml) (Invitrogen). For bioluminescence imaging, A549 cells were transduced with a fusion protein constructs encoding firefly luciferase. GSK126 were purchased from Selleck.

Plasmids and primers.
Human ONECUT2 cDNA was amplified from a mix of human lung cancer cell line cDNA with primers 5′-GAAGATCTAAGGCTGCCTACACCG-3′ (forward) and 5′-CCCTCGAGTCATGC TTTGGTACACGTG-3′ (reverse), and was cloned into pMIP-3 × HA (modified from pMIG, in which GFP was substituted with a puromycin resistance gene and 3 tandem repeats of the Hemagglutinin(HA) coding sequence were added at the N-terminus). ONECUT2(ΔDBD) was subcloned from ONECUT2 cDNA by using the reverse primer 5′-CCCTCGAGTCACGTGGCCACCTGCGAG-3′ into pMIP-3XHA. All sequences were verified by Sanger sequencing.
Retrovirus-mediated gene expression. The pMIP retroviral vectors, together with help vectors were transfected to 293FT cells for viral packaging. 72 h after transfection, viral supernatant was collected from transfected packaging cells and concentrated using Polyethylene Glycol (PEG8000). Target cells were infected in the presence of 8 µg/ml polybrene (Sigma) and selected with 1 µg/ml puromycin (Sigma).
Proliferation assay. Cell proliferation assays were performed using the CellTiter 96 AQueous One Solution Cell Proliferation Assay System (G3582, Promega) according to the manufacturer's instructions.
Soft agar colony formation assay. Cells (5 × 10 3 ) were suspended in 1.5 ml of 0.35% low melting point agarose (Sigma) in DMEM containing 10% FBS, and this suspension was overlaid on pre-solidified 0.5% agarose in the same medium in ultra-low 6-well plates (Invitrogen). Normal growth medium was gently layered over the cultures every 4 days for 14 days. The colonies were stained with 0.005% crystal violet for 1 h at room temperature and counted under an inverted microscope. Two-chamber transwell assay. Serum-starved A549 cells (2.5 × 10 4 ) were re-suspended in serum-free medium and seeded in the upper chamber of inserts (Costar) with 8 μm pore size coated with 0.2 mg/mL Matrigel (BD). FBS was used in the bottom chamber as the attractant. After 12-16 h, the cells in the upper chamber were removed using a cotton swab, and the invading cells was stained with crystal violet and counted.
Adhesion assay. For adhesion assay, 96-well plates were coated with 5 μg/cm 2 fibronectin (Santa Cruz), Collagen I (354231, BD), Collagen IV (354233, BD), and Matrigel (354234, BD) at room temperature for 1 h, and blocked with 10 mg/ml BSA at room temp for 1 h. A549 cells (2 × 10 4 ) were seeded for 15 mins. Then, the plates were shaken for 15 s, and the detached cells were aspirated. The remaining cells were fixed, stained with crystal violet, and lysed with 2% SDS. Absorbance were measured at the wavelength of 562 nm using a microplate reader (Thermo Fisher Scientific).
Wound healing assay. Cells were seeded in 6-well plates. When the cells reached 90% confluence, they were starved in DMEM containing 1% FBS. The cell monolayer was scraped with a 200 μl pipette tip from the center of each well and marked at the injury line. Photomicrographs of the wounds were taken at 0, 22, and 42 h.
Immunoblotting. Whole cell lysates were prepared using RIPA buffer (Cell Signaling). Proteins were separated by 10% SDS/polyacrylamide gel electrophoresis and electro-transferred to PVDF membranes (Millipore). The membrane was blocked for 1 hour at room temperature (RT) then incubated with a primary antibody overnight at 4 °C, followed by incubation with a horseradish peroxidase-conjugated secondary antibody (Bio-rad) for 1 hour at RT. Proteins were detected by enhanced chemiluminescence (Millipore). Primary antibodies against High-throughput sequencing data analysis. RNA-Seq fastq files were trimmed to remove adapter sequences and low-quality bases with Trimmomatic 26 , and then mapped to the hg19 genome and transcriptome with Tophat2 27 or Hisat2 28 . Raw count tables were generated using featureCounts 29 . FPKM and differential expression analyses were performed using gfold 30 or DESeq2 31 . ChIP-Seq fastq files were trimmed to remove adapter sequences and low-quality bases and then mapped to the hg19 genome with Bowtie2 32 or Hisat2 28 . Peak calling was performed and a bigwig coverage file for ONECUT2 was generated with MACS2 33 . Motif discovery was done with HOMER 34 . The differential binding signal tracks were generated by using deepTools 35  Two-tailed Student's t-test was performed to compare in vitro data. Two-sided Wilcoxon rank test was used to compare BLI data at each time point. Tumor growth and BLI curves were compared by using ANOVA. Log-rank test was used to compare animal and patient survival. A P-value less than 0.05 was considered significant.

Data availability
RNA-Seq and ChIP-Seq data have been deposited in the NCBI Gene Expression Omnibus (GEO) database under Accession Number GSE102599.