The RON Receptor Tyrosine Kinase Promotes Metastasis by Triggering MBD4-Dependent DNA Methylation Reprogramming

RON/MSP Signaling Promotes Widespread Metastasis of Human Breast Cancer In Vivo

We examined RON expression in six breast epithelial cell lines, including immortalized normal mammary epithelial cells, MCF-10A; two nonmetastatic cancer cell lines, MCF7 and T47D; and three metastatic cancer cell lines, MDA-MB-453, HCC1143, and HCC1806 (Figure 1A). RON was overexpressed in metastatic cancer cell lines compared to normal and nonmetastatic cancer cell lines. To determine whether RON/MSP drives metastasis, we engineered the nonmetastatic MCF7 and T47D breast cancer cell lines to stably express firefly luciferase, and infected them with retroviruses carrying RON and MSP cDNAs (referred to as MCF7-RON/MSP and T47D-RON/MSP; Figure 1A). Over-expressing cells had comparable RON levels with the metastatic breast cancer cell lines.

The effects of RON/MSP gain of function on tumor growth and/or metastasis of MCF7 and T47D tumors were assessed in immune-compromised mice by orthotopic implantation into the cleared inguinal mammary fat pads of 3-week-old female NOD/SCID mice. The growth of tumors was monitored weekly, and tumor metastasis was measured by bioluminescence imaging and histology upon necropsy. The MCF7-RON/MSP cells formed tumors faster compared to parental MCF7 cells, but this was not the case in the T47D model (Figure S1A). To avoid confounding effects of tumor size on metastasis, the mice were euthanized for analysis of metastasis when the tumors reached the same size endpoint for each group. RON/MSP expression was sufficient to induce high rates of spontaneous lung, bone, liver and brain metastasis from MCF7-RON/MSP and T47D-RON/MSP tumors, compared with little or no metastasis from the parental cell xenografts (Figures 1B and 1C). The metastases were verified by histological analysis and immunostaining for human cytokeratin proteins (Figure S1B). These results indicated that RON/MSP expression was sufficient to facilitate the metastatic spread of MCF7 and T47D breast cancers in vivo to sites that are highly relevant for breast cancer metastasis. These results are consistent with our previous findings, which showed a significant correlation between RON/ MSP expression and metastasis in patients with breast cancer (Welm et al., 2007).

To determine the molecular mechanisms by which RON/ MSP drives breast cancer metastasis, we performed RNA sequencing (RNA-seq) and microarray gene expression profiling of MCF7 and MCF7-RON/MSP cells. Differentially expressed genes (DEGs), based on a false discovery rate threshold of 1% and a fold change of 2, were highly correlated between the two methods. We identified 220 significant DEGs in common with both platforms (Table S1). We used the DAVID bioinformatics database to identify the gene ontology (GO) terms for the DEGs from RNA-seq and micro-array and found significant enrichment for genes involved in plasma membrane, developmental processes, receptor activity and cell signaling/communication. The GO term analysis was not used in our subsequent analysis, but indicated that RON/MSP signaling induces expression of genes whose functions are consistent with biological pathways relevant to metastasis. However, the data did not immediately implicate a clear mechanistic pathway to explain how RON/MSP drives metastasis.

RON/MSP Signaling Upregulates the Thymine Glycosylase MBD4 and Drives Altered DNA Methylation

Regulation of DNA methylation is a major mechanism involved in cell differentiation and neoplastic transformation. The epigenetic modifications contributing to metastasis, however, are much less understood. In order to determine whether alterations in DNA methylation contributed to the selective regulation of RON/MSP-regulated genes, we performed whole-genome bisulfite sequencing (WGBS) on MCF7 and MCF7-RON/MSP cells. We found 1232 differentially-methylated regions (DMRs) (Figure 2A) that occurred in both gene-body and intergenic regions, both inside and outside CpG islands (Figure 2B). DMRs regulated by RON/MSP were often intergenic, and there was a significant enrichment of hypomethylated versus hypermethylated regions (Figures 2A–2C).

Next, we wanted to determine to what degree the DEGs determined by RNA-seq between cell lines were affected by DNA methylation changes. Eleven percent of the DEGs were shown to have significant methylation differences between MCF7 and MCF7-RON/MSP cells. These observations indicated that RON/MSP expression in MCF7 cells induced massive changes in DNA methylation, which also affected expression of particular genes. These data suggested a potential role for altered DNA methylation in the mechanism of RON/MSP-mediated metastasis.

To assess if the aberrant DNA methylation in RON/MSP cells was due to differential expression levels or activities of DNA methyltransferases (DNMTs), we examined mRNA levels of DNMT3A, DNMT3B and DNMT1 in MCF7 and MCF7-RON/ MSP cells. We found no statistically significant differences (Figure S2A). We also measured DNMT activity in these cells and found no significant difference between parental cells and those expressing RON/MSP (Figure S2B).

The increased representation of hypomethylated versus hypermethylated DMRs in RON/MSP-expressing cells led us to consider DNA demethylation as a potential effect of RON/MSP. Rai et al. (2008) recently discovered a mechanism for active DNA demethylation in zebrafish that involves the cooperative actions of proteins from the cytidine deaminase family (AID and Apobecs), a G:T mismatch-specific glycosylase (Mbd4) and a DNA repair protein (Gadd45). We examined components of this DNA demethylase complex in MCF7-RON/MSP and T47D-RON/MSP cells. We noted robust upregulation of the thymine DNA glycosylase MBD4 in cells expressing RON/MSP (Figure 2D). In contrast, another thymine DNA glycosylase (TDG), which is implicated in a different mechanism of active DNA demethylation (Cortellino et al., 2011), was not significantly regulated by RON/MSP (Figure 2D). It is also important to note that the TET family, which mediates DNA hydroxymethylation was not regulated by RON/MSP (Figure S2C). Upregulation of MBD4 protein by RON/MSP was validated by western analysis of protein extracts from MCF7 and T47D cell lines (Figure 2E).

To determine whether RON and MBD4 are coordinately expressed in actual human breast tumors, we analyzed RON and MBD4 RNA expression between breast cancer and tumor-adjacent normal samples in the Cancer Genome Atlas (TCGA) data set. RON and MBD4 were both upregulated in breast tumors compared to the normal tissue (Figure 3A). We also examined RON and MBD4 protein expression in three breast reduction tissues and 13 human primary breast tumors by western analysis. RON and MBD4 were highly expressed in all seven poorly differentiated carcinomas. Lower RON and MBD4 levels were found in well-differentiated tumor samples, and little to no MBD4 or RON was detected in control samples taken from breast reduction tissue (Figure 3B). Together with our data showing that MBD4 is expressed downstream of RON/MSP in breast cancer cell lines, these data suggested that RON/MSP might promote aberrant DNA methylation through the up-regulation of MBD4. However, the necessity of MBD4 or DNA methylation for epigenetic regulation of RON/MSP downstream effectors, and any role in metastasis, had never been described.

Knockdown of MBD4 Blocks RON/MSP-Mediated Breast Cancer Metastasis

To investigate the possible role of MBD4 downstream of RON/ MSP in breast cancer metastasis in vivo, MCF7-RON/MSP and T47D-RON/MSP cells were infected with lentiviruses carrying two different MBD4-specific small hairpin RNAs (shRNAs) directed to either the 3’ UTR or the coding region (shMBD4_3UTR and shMBD4_CDS, respectively) or, as controls, anMBD2-specific shRNA (shMBD2) or a scrambled shRNA (shScr). MBD4 knockdown was validated by western blotting (Figure 3C). The effect of MBD4 knockdown on spontaneous metastasis was assessed for both cell lines in NOD/SCID mice. RON/MSP tumors expressing shMBD4 and shScr from both cell lines grew similarly to the tumors derived from RON/MSP cells; therefore, MBD4 loss of function had no effect on primary tumor growth (Figure S3A). However, spontaneous lung, liver, bone and brain metastasis was significantly inhibited in animals carrying MCF7-RON/MSP-shMBD4 and T47D-RON/MSP-shMBD4 tumors relative to the control tumors expressing RON/MSP (Figures 3D and 3E). In fact, metastasis frequencies of RON/MSP-shMBD4 cells were not significantly different from the original parental cell lines, approaching 0%. The RON/MSP-shMBD2 and the RON/MSP-shScr control tumors showed a similar frequency of metastasis as the RON/MSP tumors (Figures 3D, 3E, S3B, and S3C). To rule out off-target effects of the MBD4 shRNAs, we generated MBD4 cDNA rescue constructs (MBD4R-3UTR and MBD4R-CDS) that were resistant to their respective shRNAs. We then conducted rescue experiments in MCF7-RON/MSP-shMBD4 and T47D-RON/MSP-shMBD4 cells where re-expression ofMBD4 was at similar levels to that found in RON/MSP cells (Figure 3C). Expression of MBD4R reversed the inhibition of RON/MSP-mediated metastasis by shMBD4 (Figures 3D and 3E). These data confirm that RON/MSP drives metastasis through upregulation of MBD4, at least in these two models.

To determine whether MBD4 requires its glycosylase activity to drive metastasis, we engineered an MBD4 rescue cDNA containing a single amino acid mutation of the catalytic residue, D560A. This mutant has previously been shown to be catalytically dead (Hendrich et al., 1999, Petronzelli et al., 2000). We transplanted MCF7 and T47D cells expressing RON/MSP-shMBD4R (D560A) into mammary fad pads of NOD/SCID mice and monitored metastasis. The mutation prevented the rescue of the metastasis phenotype observed with the wild-type MBD4 rescue constructs, suggesting that MBD4 is driving metastasis through its glycosylase activity (Figures 3D and 3E). We next sought to determine if MBD4 was also required for reprogramming of DNA methylation downstream of RON/MSP.

Knockdown of MBD4 Reverses Abnormal DNA Methylation Patterns and Reverses Expression of RON/MSP-Regulated Genes

To investigate the requirement for MBD4 in aberrant DNA methylation driven by RON/MSP, we analyzed WGBS data from MCF7-RON/MSP-shMBD4 compared to MCF7-RON/ MSP. MBD4 knockdown reversed the methylation status of a specific collection of loci in MCF7-RON/MSP cells. These regions of aberrant DNA methylation were reprogrammed following the knockdown of MBD4, to levels that were comparable with parental MCF7 cells (Figures 4A and 4B). It is important to note that knocking down MBD4 also caused changes in methylation of some loci that were independent of RON/MSP (Figure 4C).

To further understand the role of MBD4-mediated aberrant DNA methylation in gene expression downstream of RON/ MSP, we also performed RNA-sequencing and microarray gene expression profiling of MCF7-RON/MSP-shMBD4 cells. We found that a statistically significant 25% of genes that were regulated by RON/MSP in MCF7 cells became reversed by MBD4 knockdown (Figure 4D). Reversal of this gene expression pattern correlated with lack of metastasis upon MBD4 knockdown. It is important to note that there were other gene expression changes induced by RON/MSP that were independent of MBD4 and vice versa. Thus, as expected, RON signaling does contribute other changes to cancer cells. Likewise, knockdown of MBD4 also affected expression of genes other than those regulated by RON/MSP, suggesting that MBD4 can also regulate expression of genes outside of the RON/MSP pathway. To dissect specific effects of the RON/MSP-MBD4 pathway, we focused on the gene set that was both RON/MSP-dependent and MBD4-dependent (reversed in RON/MSP cells when MBD4 was knocked down).

Using our WGBS and RNaseq data, we identified all of the genes that were regulated by RON/MSP in MCF7 cells that were also reversed by MBD4 knockdown. We identified 192 genes (Figure 5A), of which 64 were differentially methylated at the same locus. Because we could not definitively determine whether the remaining 128 genes were also regulated by DNA methylation in a distant region (e.g., enhancer region or inter-chromosomal interacting regions) (Bert et al., 2013, Hsu et al., 2013), we grouped all 192 RON/MBD4-dependent DEGs collectively and refer to them as the “RON/MBD4 epigenetic signature” (Figure 5A).

Specific Subclasses of Human Breast Tumors Possess the RON/MBD4 Epigenetic Signature, which Correlates with Poor Prognosis

Although experimentation on human cancer cell lines is essential for dissection of mechanistic pathways in cancer biology, it is an extrapolation from bona fide tumors. Therefore, we asked whether the RON/MBD4 epigenetic reprogramming pathway that we identified and functionally characterized in cell line models exists in human breast tumors and, if so, whether it is associated with survival outcome. To first determine whether the RON/MBD4 epigenetic signature is present in human breast tumors, we used a gene module map (Segal et al., 2004) to examine the expression of 116 genes of the 192 above-mentioned genes in 997 primary human breast cancers from the Metabric discovery cohort (the remainder of the genes were not annotated in the Metabric data set; Curtis et al., 2012). The RON/MBD4 epigenetic metastasis signature was present approximately in 25% of the breast cancers (Figure 5B).

Using the published clinical annotations for each tumor (tumor grade, expression of estrogen and progesterone receptors (ER/ PR), and HER2 status), we found that the RON/MBD4 epigenetic signature was significantly associated with grade 3, ER, and PR, tumors (Figure 5B). Several “intrinsic subtypes” of breast cancer have been previously defined on the basis of more comprehensive gene expression profiles: normal-like, luminal type A, luminal type B, HER2 like, and basal like (Sørlie et al., 2001). Indeed, the RON/MBD4 epigenetic signature was significantly enriched in basal-like tumors (Figure 5B).

We next performed Kaplan-Meier analysis examining overall survival for individuals with breast cancer across the 997 patient METABRIC data set, using the annotated 116 gene RON/MBD4 epigenetic signature. The expression of these genes was significantly associated with poor survival (p < 0.05, hazard ratio [HR] = 1.30, Figure 5C). These data strongly suggest that the RON/MBD4 pathway exists in a fairly large subset of human breast cancers (25%), and that detection of the RON/MBD4 epigenetic pathway may significantly contribute to risk assessment in breast cancer patients, even though the signature was generated a priori from all genes in the RON/MBD4 pathway. An important corollary to this work was to next ask how RON regulates MBD4 and whether the presence of a RON/MBD4 epigenetic signature would predict response to anti-Ron therapy.

MBD4 Is Upregulated by RON/MSP through PI3K Signaling

In order to further characterize MBD4 regulation by RON, we engineered an MCF7-TRE-RON/MSCV-tTA cell line, where RON expression is under the control of the tetracycline (Tet) response element (TRE) and is repressed by binding of doxycycline (dox) to the Tet transactivator protein (tTA). As expected, in the absence of dox the cells expressed RON, which began to be repressed 4 hr after the addition of dox to the culture medium and was maximally repressed 48 hr after adding dox (Figure 6A). Consistent with data in Figures 2D and 2E showing that RON regulates MBD4, we found that repression of RON also causes downregulation of MBD4 (Figure 6A). To determine whether repression of endogenous RON causes downregulation of MBD4 in other model systems, RON was knocked down using shRNA via lentiviral infection in both the RON-positive DU4475 breast cancer cell line and in primary culture from a RON-positive patient-derived breast tumor graft (DeRose et al., 2011). Scrambled shRNA was used as control. RON knockdown, although not complete, caused a reduction in MBD4 expression in both models (Figure 6B). Thus, we concluded that, in all five different models we examined, RON upregulates MBD4, which can be reversed with RON downregulation. Therefore, the pathway has potential to be blocked.

RON is known to transduce a variety of signals that regulate different downstream pathways, most notably the mitogen-activated protein kinase (MAPK) and phosphatidylinositol 3-kinase (PI3K) pathways (Li et al., 1995; Wang et al., 1996). To elucidate which signaling pathway(s) is required for regulation of MBD4 by RON, we treated the cell lines with varying doses of U0126 (a MEK/ERK inhibitor), or BKM-120 (a PI3K inhibitor) for 0.5 hr or 1 hr, followed by protein extraction and MBD4 western blotting. Examination of activated (phosphorylated) Erk and PRAS proteins in each experiment verified that the drugs were effective in inhibiting their respective targets (Figure 6C). Treatment with BKM-120, but not U0126, blocked RON/MSP-mediated MBD4 upregulation (Figure 6C). To further validate that MBD4 upregulation by RON/MSP requires PI3K/Akt signaling, the PI3K p110 alpha catalytic subunit was knocked down in MCF7-RON/MSP cells by shRNA via lentiviral infection. Knockdown of PI3K activity was determined using phosphorylation of PRAS as a surrogate measure (Figure 6D). This experiment showed that PI3K knockdown led to a reduction in MBD4 expression, similar to the results with BKM-120 (Figure 6C), demonstrating that RON/MSP regulates MBD4 through activation of the PI3K pathway.

To determine whether MBD4 is regulated specifically by RON/ MSP-mediated activation of PI3K, we engineered MCF7 cells to overexpress the other member of the MET receptor tyrosine kinase family, MET, which also activated PI3K (Figure S4A). We found that MET increased MBD4 expression, showing that MBD4 can be regulated by another, related tyrosine kinase (Figure S4A). To determine whether any activator of PI3K signaling could increase MBD4 expression, we treated EGFR-positive MDA-MB-231 cells with EGF (15 nM) for 30 min, which was sufficient to activate PI3K as determined by phosphorylation of AKT (Figure S4B). We observed no upregulation of MBD4, indicating that MBD4 is not universally regulated by activation of the PI3K pathway (Figure S4B). It is also important to note that MBD4 phosphorylation was not altered as a result of RON/MSP pathway activation (Figure S4C), suggesting that MBD4 is not a direct target of RON kinase or kinases in the PI3K pathway.

Treatment with a RON Kinase Inhibitor Prevents Metastasis of Primary Patient-Derived Tumor Grafts

Ron kinase activity, and its resulting signaling, can be blocked by small molecule kinase inhibitors. To determine if the RON/ MBD4 pathway can be blocked in patient tumors using RON kinase inhibitors, and if there is a resulting effect on metastasis, we utilized four independent experimental systems that are highly relevant to bona fide breast tumors. First, we chose two patient-derived tumor grafts (DeRose et al., 2011) that survive primary culture conditions and treated them with the RON/ MET dual inhibitor OSI-296 (Steinig et al., 2013) in vitro. Inhibition of RON (verified by reduction of phosphorylated RON protein) caused downregulation of MBD4 (Figure 7A). Next, we chose two human tumor graft lines for in vivo studies based on their RON/MBD4 signature (not shown) and on their relatively fast growth and robust metastasis (DeRose et al., 2011), which allows for assessment of tumor growth and metastasis within a timeline that is practical for drug treatment. The tumor grafts were implanted orthotopically into cleared mammary fat pads of NOD/SCID mice and allowed to grow to 100 mm³ before starting treatment with OSI-296. Although there was no significant effect on tumor growth (data not shown), inhibition of RON in the in vivo setting caused complete blockade of lymph node (Figures 7B and 7C) and lung (Figure 7D) metastasis in these two models.

To determine whether the RON/MBD4 pathway that we previously defined in MCF7 cells was affected in the tumor grafts treated with the Ron kinase inhibitor in vivo, we analyzed gene expression in tumors from mice treated with the vehicle (Trappsol) or OSI-296. Indeed, a set of nine genes (CSGALNACT1, SIGLEC6, SHC4, ABCA1, PLD1, RNF144A, SLC44A4, SLC2A13, and AXIN2) from the RON/ MBD4 epigenetic signature was significantly deregulated following treatment with the RON inhibitor in vivo. These data indicated that short-term inhibition of RON kinase with OSI-296 could not only block metastasis, but also reverse expression of some genes within the RON/MBD4 epigenetic metastasis signature.

To assess whether the expression of these nine RON/MBD4-dependent genes was associated with patient outcome in breast cancer cohorts, we performed Kaplan-Meier analysis examining overall survival for individuals with breast cancer across the 997 patient METABRIC data set (Curtis et al., 2012). The expression of these nine genes (defined as the subset of the RON/MBD4 epigenetic signature genes that were reversed in tumor grafts following RON inhibitor treatment) was significantly associated with poor survival (p = 0.04, HR = 1.55, Figure 7E). To validate our findings using additional, independent data, we generated a meta collection of gene expression data from an additional 977 patients from five independently published studies of breast cancer (van ‘t Veer et al., 2002; Miller et al., 2005; Chin et al., 2006; Desmedt et al., 2007; Heikkinen et al., 2011). The method we utilized to normalize data across the different data sets was previously described (Segal et al., 2004; Ben-Porath et al., 2008). Using these data, we confirmed that the nine gene RON/MBD4 epigenetic signature was significantly associated with poor survival (p = 0.008, HR = 1.55, Figure 7E).

Interestingly, multivariate analysis of the nine gene RON/ MBD4 epigenetic signature revealed that it was not a significant independent prognostic factor for survival in this data set and may depict similar risk as ER− status (Figure 7F). ER− tumors have a poor prognosis, in part because there are no targeted therapy options due to lack of defined pathways driving these tumors. Together, these data highlight the promising potential for (1) identifying breast cancer patients that might benefit from a RON inhibitor (approximately 25%); (2) identifying a set of biomarkers for RON inhibition, and (3) blocking RON to inhibit breast cancer metastasis.

In Vivo Metastasis

All animal procedures were reviewed and approved by the University of Utah Institutional Animal Care and Use Committee. For bioluminescence imaging, mice were anesthetized and given 150 mg/kg of D-luciferin in PBS by i.p. injection. Five minutes after injection, mice were killed and the organs were extracted and imaged ex vivo to detect metastasis foci. A detailed description of analysis is provided in Supplemental Experimental Procedures.

Expression Array Hybridization and Data Analysis

Triplicate biological replicates of hybridization were performed for each cell line. Differential gene expression was evaluated using the t test (p < 0.01) and Benjamini and Hochberg correction. The threshold was set at 2-fold change for both upregulated and downregulated genes. A detailed description of analysis is provided in Supplemental Experimental Procedures.

RNA Sequencing

Alignments were generated from Illumina Fastq files to the hg19 human genome with all known and theoretical splice junctions using Novocraft’s novoalign aligner. A detailed description of analysis is provided in Supplemental Experimental Procedures.

Computational Analytical Method for WGBS

The software packages used in this analysis are open source and available from the USeq (http://useq.sourceforge.net/usageBisSeq.html) project website (Nix et al., 2010). A detailed description of analysis is provided in Supplemental Experimental Procedures.

Analysis of Gene-Set Enrichment Patterns

To identify gene-set enrichment patterns, we used methods described previously (Segal et al., 2004) applied with Genomica. A detailed description of analysis is provided in Supplemental Experimental Procedures.

Survival Analysis

All survival data were extracted from original publications. p values were calculated using the log rank test comparing the group of individuals with tumors showing the RON/MBD4 signature to all other individuals. Univariate and multivariate analyses with Cox proportional-hazards regression were done on the individual clinical variables.