Imprinting defects at human 14q32 locus alters gene expression and is associated with the pathobiology of osteosarcoma

Osteosarcoma is the most common primary bone malignancy affecting children and adolescents. Although several genetic predisposing conditions have been associated with osteosarcoma, our understanding of its pathobiology is rather limited. Here we show that, first, an imprinting defect at human 14q32-locus is highly prevalent (87%) and specifically associated with osteosarcoma patients < 30 years of age. Second, the average demethylation at differentially methylated regions (DMRs) in the 14q32-locus varied significantly compared to genome-wide demethylation. Third, the 14q32-locus was enriched in both H3K4-me3 and H3K27-me3 histone modifications that affected expression of all imprinted genes and miRNAs in this region. Fourth, imprinting defects at 14q32 - DMRs are present in triad DNA samples from affected children and their biological parents. Finally, imprinting defects at 14q32-DMRs were also observed at higher frequencies in an Rb1/Trp53 mutation-induced osteosarcoma mouse model. Further analysis of normal and tumor tissues from a Sleeping Beauty mouse model of spontaneous osteosarcoma supported the notion that these imprinting defects may be a key factor in osteosarcoma pathobiology. In conclusion, we demonstrate that imprinting defects at the 14q32 locus significantly alter gene expression, may contribute to the pathogenesis of osteosarcoma, and could be predictive of survival outcomes.


INTRODUCTION
Osteosarcoma is an aggressive bone cancer predominantly affecting children and young adolescents with a 5-year survival rate of ~70% [1]. Despite significant advances in cancer treatment, the survival rate for osteosarcoma has remained static for the past 30 years. Genetic deficiencies such as mutations of TP53, RB, p14 and p16 have previously been reported as occurring frequently in osteosarcoma [2][3][4][5]. Silencing of Wnt inhibitory factor 1(WIF1) [6], p14ARF [7], and RASSF1A [8] by DNA hypermethylation and loss-of-imprinting in IGF2 and H19 have also been shown to occur in osteosarcoma [9,10]. Recently, we demonstrated that microRNAs (miRNAs) at the14q32 locus are significantly downregulated in osteosarcoma, but not due to a loss of DNA copy number [11]. The 14q32 imprinted region is ∼1 Mb in size and contains both paternally (DLK1, RTL1 and DIO3) and maternally (MEG3, MEG8 and DIO3OS) expressed genes [12,13]. These imprinted genes have key roles in cellular functions [14][15][16], bone differentiation [17,18], and can serve as markers for complete reprogramming of induced pluripotent stem (iPS) cells [19]. Interestingly, this imprinted locus also includes over 40 miRNAs, a subset of which cooperatively regulates cMYC expression in osteosarcoma [11]. It is therefore likely that dysregulated expression of genes and miRNAs resulting from disordered imprinting at this locus could contribute to the pathogenesis of osteosarcoma [20,21]. www.impactjournals.com/oncotarget Genomic imprinting is a type of epigenetic inheritance through which differential expression of genes depends on parental origin (i.e., maternal or paternal) [22,23]. Typically, genomic imprinting occurs in pairs of genes that are located near each other. The expression of these imprinted genes is controlled by DNA methylation of unique parent-origin-alleles in the imprinting control regions (ICRs) or differentially methylated regions (DMRs) [24]. Loss of the monoallelic DNA methylation pattern in the ICRs or DMRs, also known as loss-ofimprinting, can cause changes in expression of imprinted genes in the region. For example, loss-of-imprinting at the IGF2 locus has been found in Wilms' tumor [25,26], colorectal [27], and other cancer types [28]. The three DMRs in the 14q32 locus can regulate the imprinted genes and miRNA clusters in the region [29,30].
To determine if epigenetic alteration(s) of the 14q32 locus is associated with osteosarcoma, we evaluated the DNA methylation levels of these DMRs and studied histone modifications in normal bone tissues and osteosarcoma patient samples. We also determined the imprinting status of the 14q32 locus in buccal DNA samples from osteosarcoma patients and their biological parents. Further, we studied the methylation patterns at DMRs of the homologous regions in two mouse models of spontaneous osteosarcoma. Our results suggest that imprinting defects in humans is associated with the pathobiology of osteosarcoma.

Imprinting defects at 14q32 DMRs in osteosarcoma
There are three DMRs in the 14q32-imprinted region, which can regulate the entire imprinted gene cluster [29]. Two of these DMRs, DMR-1 and DMR-2, are intergenic (IG-DMRs) and are located in close proximity to each other between the DLK1 and MEG3 genes. The third, DMR-3, is located in the promoter region of the MEG3 gene (MEG3-DMR). By bisulfite TA cloning and sequencing we determined the DNA methylation pattern in genomic regions designated site A (which includes DMR-2) and two CpG islands at sites B and C ( Figure 1A). DNA methylation analysis indicated that in normal bone tissues, the imprinting pattern (~50% DNA methylation level with one allele densely methylated and another free of methylation) was identified at site A ( Figure 1B). However in osteosarcoma SaOS2 cells, an imprinting defect of hypermethylation was found at site A compared to normal bone tissues. CpG island sites B and C were also densely methylated (~90%) at both alleles. DNA methylation analysis confirmed the locations of these DMRs in normal bone tissues and SaOS2 cells.
To determine whether imprinting defects at these DMRs are associated with osteosarcoma, we analyzed the methylation pattern of all three DMRs (DMR-1, -2 and -3) ( Figure 1C) in 8 normal bone tissues and 32 osteosarcoma samples (Supplementary Table 1)  by bisulfite-pyrosequencing. With the exception of  5 osteosarcoma samples (FT-7, -422, -426, -430 and -436), all samples were collected during local control surgeries prior to neoadjuvant chemotherapy and/or radiation. Based on our sample age distribution [31], we defined the early-and late-onset OS. 23 samples were early-onset OS (< 30 years) and 9 samples were late-onset OS (> 30 years). Our DNA methylation analysis revealed that imprinting defects, specifically hypomethylation, at DMRs were detected in most of the early-onset osteosarcoma patient samples but were not present in the late-onset samples ( Figure 1D). DMR-2 appeared to be most prone to imprinting defects.

14q32 index identifies normal/early-and late-onset osteosarcoma
To test whether imprinting defects at the 14q32 locus were the consequence of global hypomethylation, we assessed the genome-wide methylation status in osteosarcomas by measuring the methylation of LINE-1, representing 17% human genomic DNA [32] ( Figures 1E and 2A). We found that the methylation status of a single allele of LINE-1 decreased by ~5% in both early-and late-onset osteosarcoma. However, the methylation decrease at 14q32-DMRs was ~19% in earlyonset osteosarcoma and was significantly higher than that of LINE-1 (P = 0.0046). Hypomethylation at the 14q32locus was therefore not entirely the result of genome-wide demethylation in early-onset osteosarcoma. In contrast, the hypomethylation at 14q32 in late-onset osteosarcoma samples was ~6%, similar to that of LINE-1, suggesting that demethylation in late-onset osteosarcoma was likely due to global hypomethylation.
A pattern of either hyper-or hypomethylation at DMRs signifies imprinting defects. Although hypomethylation was the predominant pattern at DMR-2 in the majority of cancers compared to normal bone tissues, a few cases did present with hypermethylation at 14q32-DMRs. Thus, we calculated the extent of imprinting defects as the absolute difference of DNA methylation deviation from the normal imprinted-DMR levels (50%). Both gain and loss of methylation at DMR were defined as loss-of-imprinting and both types of changes led to similar gene expression patterns, therefore we used the absolute value of the difference between samples and the theoretical normal (50%) to define the extent. We termed this quantitative deviation of imprinting the '14q-index' (14q-I). The 14q-index was calculated as the average extent of imprinting defects at all three DMRs using the following formula: Based on the above equation, the calculated 14q-index for normal bone tissues was 7.0. It was notable that normal/early-onset osteosarcoma (12-30 years) samples showed a significantly higher 14q-index at 18.6. The average 14q-index was 8.5 for late-onset osteosarcoma (> 30 years), a non-significant change versus normal bone tissues ( Figure 2B). Based on a 95% confidence interval (CI), we defined imprinting defect as a 14q-index values > 8.5. Twenty of the 23 (87%) early-onset osteosarcomas were 14q-I(+). However, in the late-onset group only 5 of 9 (56%) were positive and the degree of 14q-index changes was not as dramatic as that of the early-onset group.

Loss of 14q32-locus imprinting is unique to osteosarcoma
Next, we investigated whether the imprinting defect was limited to DMRs in the 14q32-locus or extended to others. We tested the DNA methylation levels of three other known DMRs at the H19, MEST and PEG3 imprinted gene clusters in normal bone and osteosarcoma samples. Based on a 95% confidence interval of normal bone tissue, ~59% (19 out 32) of osteosarcoma samples showed imprinting defects at H19-DMRs ( Figure 2B). The correlation between imprinting defects at 14q32-DMRs and H19-DMR was 0.78 (R 2 = 0.61), potentially implicating both loci in tumor development ( Figures 1E, 2C). The methylation patterns of H19-DMRs, were much more complex than that of the 14q32-locus, which had both gain-and loss-of DNA methylation ( Figure 1E) and the frequency of loss-of-imprinting at H19 was much lower than that at 14q32-locus. Imprinting defects were not found in DMRs of MEST and PEG3 and there was no correlation between either of these two genes and the 14q32-locus ( Figure 2B and 2C).
We then tested whether imprinting defects at the 14q32-locus were unique to osteosarcoma or widespread among different cancer types. The DNA methylation levels of 14q32-, H19-, MEST-and PEG3-DMRs were determined using 13 cell lines representing 6 different cancer types ( Figure 2D). Hypomethylation at 14q32 IG-DMR (DMR-1), which was characteristic of human osteosarcoma tissue samples, was also observed in osteosarcoma cell lines SaOS2 and U2OS. However hypermethylation patterns were observed in osteosarcoma cell lines MG63 and HOS both characterized as having an osteoblastic phenotype. There were no consistent patterns of change in DNA methylation status in the 6 additional cancer types tested, indicating that hypomethylation of this set of DMRs may be a distinguishing feature of osteosarcoma.

Downregulation of 14q32 locus genes is associated with imprinting defects
Imprinting defects at DMRs have been shown to regulate the expression of genes within the locus. To determine the effects of imprinting defects at 14q32-DMRs on gene expression, we measured the expression levels by qRT-PCR of 6 imprinted genes in this locus including paternal (DLK1, RTL1, and DIO3) and maternal (MEG3, MEG8, and DIOAS) genes expressed in 4 normal bone tissue samples, six 14q-I(−) and eight 14q-I(+) osteosarcoma samples ( Figure 3). RTL1, DIO3, and MEG3 were relatively highly expressed in normal bone tissues, while DLK1, MEG8 and DIO3AS were expressed at lower levels. Most of the imprinted genes at the 14q32-locus (with the exception of MEG8) were downregulated in 14q-I(−) samples, and further decreased in 14q-I(+) samples.
To evaluate the co-expression of imprinted genes and miRNAs at the 14q32 locus, we analyzed the expression of DLK1-DIO3 cluster genes (DIO3, DLK1, MEG3, and MEG8) and representative 14q32 miRNAs (miR-134, miR-154, and miR-382) in an independent set of 43 osteosarcoma tumor samples. In the majority of tissues analyzed, 14q32 miRNA expression was correlated with DLK1-DIO3 cluster genes, particularly with MEG3 and MEG8 ( Figure 3C).
Because the entire imprinted region shares a similar histone modification pattern, we used levels of different histone markers at IG-DMR to represent the histone modification at this genomic region. H3K4-me3 at 14q32 IG-DMR was lower than GAPDH, but was ~80-fold higher than the IgG negative control. On the other hand, H3K27-me3 at 14q32 IG-DMR was 1.6-fold higher than at RARB and was ~180-fold higher than IgG negative control (Supplementary Figure 1E-1F). Similarly, H3K9-me2 and H3K9-me3 were detected at much higher levels at IG-DMR than IgG. Based on these observations we concluded that H3K4-me3, H3K27-me3, H3K9-me2, and H3K9-me3 were enriched at the 14q32-imprinted region.
We examined the changes to histone modification in osteosarcoma and looked for differences between 14q-I(−)  DMRs at 14q32-locus and LINE1 in osteosarcomas. (B) Determination of imprinting defects at 14q32 DMRs, H19, MEST and PEG3 in normal bone tissue (normal), osteosarcoma samples (osteosarcoma), osteosarcoma early-onset (osteosarcoma-early) and late-onset (osteosarcoma-late) samples. The imprinting defect extent is defined as the absolute methylation difference between patient samples and the proposed normal imprinted DMR levels (50% methylation). (C) Correlations between 14q-index and imprinting defects in H19, MEST, and PEG3, respectively. (D) DNA methylation patterns of 3 DMRs at 14q32-locus and 3 other imprinted loci in cancer cell lines representing 6 cancer types. *P < 0.05. Abbreviations used: OS: total OS samples, including both early-and late-onset samples; OS-early: early-onset OS samples (n = 23); OS-late: late-onset OS samples (n = 9). and 14q-I(+). We found the total H3 levels were decreased in both 14q-I(−) and 14q-I(+) samples (Supplementary Figure 2A). In 14q-I(+) samples, H3K4-me3, H3K9-me2, and H3K9-me3 were increased, and H3K27-me3 was decreased. But in 14q-I(−) samples, all of the histone markers showed a decrease. After normalizing to the level of H3, the majority of histone markers were increased in 14q-I(+) osteosarcoma group ( Figure 4B and Supplementary Figure 2), but were decreased or unchanged at most of imprinted gene loci in 14q-I(−) osteosarcoma. We deduced that all of the imprinted genes were regulated by histone modification as a single unit and silenced by the increase of three silencing markers in the 14q-I(+) region.
To confirm that the sonication process did not preferentially disrupt either allele, we used total input as a control. In all three groups, the DNA methylation levels of the input samples were consistent with the results we obtained directly from patient DNA, suggesting that sample preparation during the ChIP assay did not affect the measurements.
DNA methylation levels for both H3K4-me3 and H3K27-me3 ChIP samples shared a similar pattern with the input control in all samples ( Figure 4C), confirming that the H3K4-me3 and H3K27-me3 were equally present at both maternal and paternal alleles.

Proteins encoded by the 14q32 locus act as tumor suppressors
It has previously been reported that MEG3 is a tumor suppressor gene [15], but RTL1 and DIO3 remain candidate genes of interest and have not been well studied in osteosarcoma. Transient overexpression of DLK1, RTL1 or DIO3 individually in SaOS2 cells decreased cell viability by ~30%, compared to cells transfected with control plasmids ( Figure 4D). Individual transfection of DLK1, RTL1 or DIO3 also significantly increased apoptosis in SaOS2 cells ( Figure 4E). Our previously published work has demonstrated that several miRNAs in this area can also function as tumor suppressors in the context of osteosarcoma [11].

Imprinting instability at 14q32 locus
It is likely that osteosarcoma patients inherit, at least in part, imprinting instability from their parents, which predisposes them to the early-onset oncogenesis typical of osteosarcoma. To test this hypothesis, we measured DNA methylation levels of two IG-DMRs at the 14q32 locus in buccal samples from 12 unrelated normal subjects (median age 26 years), and 10 sets of triad samples (10 osteosarcoma patients plus their biological parents). Due to the rapid turnover of the buccal mucosa, epithelial cells are subject to an increased risk, compared to normal bone tissue, of changes in DNA methylation status and of development of loss-of-imprinting resulting from age-related methylation mechanisms [34]. In unrelated normal individuals, we found the average methylation level of DMR-1 was 53% and DMR-2 was 64%, likely due to normal age-related increases in DNA methylation. The normal range of DNA methylation in buccal tissues, was therefore defined as 51-55% at DMR-1 and 61-67% at DMR-2 (using a 95% CI) (Supplementary Table 3A).
Hypermethylation imprinting defects at IG-DMRs, especially DMR-1, were detected in all of the analyzed osteosarcoma patient samples and, importantly, in all parent samples. There was a significant difference in DMR-1 (P < 0.0001) and DMR-2 (P < 0.05) DNA methylation levels between unrelated normal healthy samples and those in the osteosarcoma patient or parent samples ( Figure 5A, 5B and Supplementary Table 3B). Patient DMR methylation patterns appear to be correlated with their respective parents' patterns ( Figure 5C-5D). The Pearson's correlation coefficient (Pearson's r) is 0.5 between osteosarcoma patients and their fathers at DMR-1, and is 0.63 between osteosarcoma patients and their mothers at DMR-2 (Supplementary Table 3C). This suggests that the imprinting defect at 14q32 might be inherited from existing parental imprinting instability, especially from the paternal side.

Imprinting defects did not result from pre-existing mutations in spontaneous osteosarcoma mouse models
To determine if imprinting defects at the 14q32 locus were a result of pre-existing mutations during osteosarcoma development, we studied tumor tissues using a spontaneous osteosarcoma model in T2/onc, p53 R270 H/+ , OSx-Cre, ROSA26-LSL-SB11 mice. These mice express a p53 R270 H point mutation and allow random insertional mutagenesis using the Sleeping Beauty (SB) transposon system. Consistent with a previous report [35], ~60% of p53 R270 H/+ mice developed osteosarcoma with an average age of onset of ~16 months. In T2/onc; p53 R270 H/+ ;OSx-Cre; ROSA26-LSL-transposase mice, 70% of mice developed osteosarcoma by ~11 months with the earliest case appearing at 8 months. SB mutagenesis led to accelerated tumor formation compared to the p53 R270 H/+ mice.
We determined the DNA methylation changes of IG-DMR at the mouse 12qF locus (homologous to human 14q32), with bone tissues from normal mice as controls. In normal bone samples, the mean DNA methylation level was 64% (95% CI: 55% -73%) ( Figure 6A and Supplementary Table 4A). In Trp53-R270 H/SB tumor samples, 44% (11 out of 25) showed imprinting defects, which was comparable to the prevalence of imprinting defects in late-onset human osteosarcoma (~56%), suggesting that late-onset osteosarcoma in humans is more likely to be caused by spontaneous somatic mutations. In contrast to human osteosarcoma samples, imprinting www.impactjournals.com/oncotarget  defects in these mice were due to hypomethylation (3 out of 25) or hypermethylation (8 out of 25). As reported earlier, it is likely that different regulatory mechanisms are involved in human and mouse at this locus [29]. In humans, DMRs are located outside CpG islands, while in mice the DMRs are located within CpG islands, which are better protected against methylation changes. The fact that 12qF imprinting defects were found in only 44% of SB-induced osteosarcoma suggested that 12qF imprinting defects were not caused by osteosarcoma formation, but might instead be one of the underlying mechanisms predisposing the tumor's development.

Imprinting defects at 12qF were related to osteosarcoma development in p53/Rb mutation mouse model
To further establish the role of imprinting defects at 12qF in osteosarcoma development, we generated an osteosarcoma mouse model with both p53 and Rb mutations ( Figure 6B-6C). 25 mice with osteosarcoma representing various tumor locations (20 limb, 3 pelvis, 1 neck, and 1 soft tumor inside the chest wall) were included in the study. We were able to determine the DNA methylation levels at the 12qF IG-DMR locus in 24 of these samples. Based on 95% CI of normal bone tissue, we defined that normal methylation range was 72~76% (Supplementary Table 5). Imprinting defects were found in all osteosarcoma samples, consistent with observations in human osteosarcoma (87%). We found both hypomethylation (66.7%; 16 out of 24) and hypermethylation (33.3%; 8 out of 24), similar to the previous mouse osteosarcoma samples ( Figure 6D).
We examined the expression levels of the homologous 14q32 locus imprinted genes in these mouse osteosarcoma samples. We found that Dio3 and Rtl1, the paternally expressed genes, were dramatically decreased in the DMR hypomethylated samples but had no significant change in the hypermethylated samples ( Figure 6E), again consistent with the human osteosarcoma samples (Figure 3). The maternally expressed genes Meg3 and Rian had no significant changes in expression although an increased trend was observed ( Figure 6E), in contrast to the human samples which saw little overall change in expression.

14q-index is a prognostic indicator in osteosarcoma
To determine whether the 14q-index can serve as a prognostic indicator in osteosarcoma, we tested an independent cohort of 15 human osteosarcoma biopsy samples obtained prior to chemotherapy/radiation with long-term clinical follow-up data. We again found significant imprinting defects at multiple loci, especially DMR-2 ( Figure 7A), validating our previous results. Hypomethylation defects, especially at DMR-2, were present in the majority of samples and 86.7% (13 out of 15) samples showed an increase in the 14q-index. Combining these 15 new samples with the initial 21 osteosarcoma samples (onset at < 30 years), we found a significant difference in the 14q-index in osteosarcoma patient samples with and without metastasis ( Figure 7B). Additionally, osteosarcoma patients with imprinting defects and a high 14-q index had poorer survival outcomes ( Figure 7C). With this limited sample size, a robust statistical analysis was not possible, however it was suggestive that patients with a normal 14q-index do have a survival advantage.

DISCUSSION
The results of our study suggest that hypomethylation at 14q32 -imprinted IG-DMRs is predominantly associated with osteosarcoma (onset < 30 years) and is a likely mechanism of downregulation of imprinted genes and miRNAs at this locus. Although hypomethylation typically leads to increased gene expression, it has been reported that hierarchical interaction of the methylation pattern in the three DMRs at the 14q32 locus does have a negative effect on gene expression [36,37]. More importantly, the observed imprinting defects in 14q32 DMRs did not appear to be due to genome-wide methylation changes. DNA methylation patterns of 14q32-DMRs in osteosarcoma appear to be significantly different than those observed in other cancers and may provide additional insights into the pathobiology of osteosarcoma.
We developed the '14q-index' (14q-I) to quantitate the degree of imprinting defects in osteosarcoma and propose that 14q-I(+) can be used as both a predictive marker of osteosarcoma and a prognostic indicator as patients with a normal index value tend to have improved survival outcomes.
We studied the effects of 14q32-locus imprinting defects on the expression of genes at this region. Notably, all the genes and miRNAs tested showed different degrees of downregulation [11]. Our data demonstrating that 14q32 locus genes (with the exception of MEG8) are significantly downregulated in osteosarcoma samples contradicts a previous report that loss-of-imprinting causes differential changes in paternal and maternal expressed genes and an inverse correlation at the IGF2-H19 locus [38]. This may be due to different regulatory mechanisms at the 14q32 and H19-IGF2 loci [39]. The imprinted genes at the 14q32-locus can be co-expressed in mesodermderived tissues and inversely expressed in lung, liver, and placenta [40]. It is likely that both the maternal and paternal imprinted genes are regulated as a single unit in bone tissue and imprinting defects can cause co-repression of both maternal and paternal expressed genes in osteosarcoma. It is interesting to note that imprinting defects have a differential impact on gene expression. In mice, the paternally and maternally expressed genes were inversely regulated with imprinting defects. However, in humans these genes were affected uniformly regardless of their parental origin, which may be due to the observed histone modifications.
There is a correlated pattern of 14q32 miRNA and DLK1-DIO3 gene cluster expression within individual patient samples. The genes at both ends of the cluster, DLK1 and DIO3, showed more variability and this may implicate additional regulatory factors. Previously, we reported that significant downregulation of miRNAs from this locus was correlated with poor prognosis in human and canine osteosarcoma [41]. Consistent with these observations, our study also revealed that DLK1, RTL1, DIO3, and miRNAs in this region are candidate proapoptotic genes [20]. MEG3 has been previously reported to function as a tumor suppressor that can regulate p53 expression [15]. Our results also suggested that imprinting defects at MEG3-DMR (DMR-3) were not a significant contributor to cancer development but could potentially hasten the disease process. This finding was consistent with previous results suggesting that IG-DMRs are germline-derived primary DMRs, while MEG3-DMR is a post-fertilization-derived secondary DMR [42]. The IG-DMR acts hierarchically to regulate the methylation status of the somatic Glt2-DMR (homolog to human MEG3-DMR) in mice. Previously, we have reported that several 14q32 miRNAs function as tumor suppressors by targeting cMYC and that blocking these miRNAs using antimiRs in normal human osteoblasts was transformative [11]. Taken together, these data suggest that the 14q32 locus contains multiple tumor suppressors.
It has previously been suggested that the 14q32-imprinted locus may share similar regulatory mechanism(s) such as the IGF2-H19 imprinting cluster [43]. However, the patterns of co-expression or co-silencing are only found in the 14q32-locus, which implies different regulatory mechanisms between the two imprinted regions. We found that both the active marker H3K4-me3 and silenced marker H3K27-me3 were enriched in this 14q32-locus. We speculate that this locus is potentially controlled as a 'bivalent domain,ʼ which is also associated with developmental and differentiation functions [44]. A bivalent domain is defined as the presence of both H3K4-me3 and H3K27-me3 at the same locus on the same allele simultaneously. We used the 14q32 IG-DMR region's DNA methylation status to define each allele's parental origin. The maternal allele is characteristically free from methylation while the paternal allele is densely methylated. Although Single Nucleotide Polymorphism (SNP) analysis would further confirm the presence of bivalent domain, we did not find suitable SNPs in this region for testing. These results suggested a possible mechanism for silencing of the entire 14q32imprinted genes and miRNAs in osteosarcoma samples, especially in 14q-I(+) osteosarcoma cases. We also found that this entire imprinted region might be regulated as a single unit in bone tissues with the boundary marked between DLK1 and DIO3.
Imprinting defects at the 14q32-locus were found in 87% of osteosarcoma samples (onset < 30 years). The specific loss of imprinting at the 14q32 and H19 loci in osteosarcoma samples suggested that these methylation changes were disease-specific and not the result of global methylation changes. Comparison of the DNA methylation patterns of 14q32 IG-DMRs in the buccal DNA samples of osteosarcoma patients, their biological parents, and unrelated normal subjects implied that imprinting instability exists in the affected child's parents but is absent in unrelated normal subjects. We speculate that inherited imprinting defects at the 14q32 locus may contribute to pediatric/young-adult development of osteosarcoma. Given the apparently 'normal' phenotype of the parents, imprinting instability itself is likely insufficient to cause cancer, but might predispose affected subjects to osteosarcoma.
It is of interest to note that osteosarcoma occurs predominantly during puberty, a period characterized by accelerated linear bone growth, and some imprinted genes in these regions have been implicated in bone differentiation. It is likely that under growth stimulation, patients with imprinting defects at 14q32 and associated silencing of multiple tumor suppressor genes are at greater risk of developing osteosarcoma. Similarly, lossof-imprinting at this locus is implicated in other cancers such as lymphoma, colon cancer, and neuroblastoma [45], which is also supported by our analysis of the imprinting status at 14q32-DMRs of 10 cancer cell lines representing six cancer types. In our study we clearly identified the differential methylation patterns in the normal (< 30 years) and late-onset osteosarcoma groups. It is possible that some of the pre-operative treatments that these patients received might also affect the epigenetic patterns although our sample collection criteria attempted to minimize this possibility. The timing of buccal sampling in patients with osteosarcoma may also have affected results, as some osteosarcoma drugs cause aggressive mouth sores [46].
Several mouse uni-parental disomy (UPD) models (pUPD12 and mUPD12) have been studied previously; these models are non-viable with skeletal and skeletal muscle defects. They suggested that imprinted genes on chromosome 12 are essential for the development of the mesodermal and neural crest-derived lineages that give rise to bone [47]. Skeletal abnormalities such as small hands, hyper-extensible joints and advanced bone age have been reported in human mUPD14 patients [48]. In osteosarcoma, imprinting defects at the 14q32-locus can cause significant downregulation of imprinted genes and miRNAs, which may further contribute to its pathobiology and disrupt normal skeletal development.
We generated mouse models of osteosarcoma by disrupting Trp53 in combination with either SB transposon-mediated mutagenesis or concomitant loss of Rb1. We found that imprinting defects at 12qF, the homolog to the human 14q32 locus, were quite uncommon in SB-induced osteosarcoma. In fact, their frequency was similar to that of human late-onset osteosarcoma, which is consistent with the hypothesis that mutations play a more causative role in the late-onset osteosarcoma development. In contrast, all Rb1/Trp53-mutant osteosarcoma mice had imprinting defects at 12qF DMRs, confirming that 12qF-imprinting defects are an important mechanism for osteosarcoma development in this model. Similar to our observations in normal/early (< 30 years) and late-onset human osteosarcoma, there were distinctive 12qF locus methylation patterns between the two mouse models that may, in part, correspond to their genetic differences. The high frequency of imprinting defects in both the Rb1/ Trp53 model and in human osteosarcoma, suggested the presence of one or more conserved mechanisms in osteosarcoma development. Rb1 and/or Trp53 mutations along with imprinting defects may cooperatively function towards development and progression of osteosarcoma. The SB mouse osteosarcomas have less genomic instability and chromothripsis, compared with Trp53 (no SB) or Rb1/Trp53 mouse osteosarcomas [49]. Further, only about 20% of SB mice showed common insertion site mutagenesis in Rb1 [50]. SB-induced tumors may represent a model for the pathogenesis of late-onset osteosarcomas, which typically have had radiation exposure, prior alkylating chemotherapy, Paget's disease, or other mutationgenerating conditions or exposures. With rampant SB mutations driving oncogenesis, epigenetic alterations may not be required to further drive the pathogenesis. Without a SB mutagenesis program, it is possible that osteosarcoma development may require epigenetic modifications to induce cancer growth and maintenance. RB1 has epigenetic functions, which have not been well sorted out in osteosarcoma but previous publications have reported on the significant role of RB-loss in osteosarcoma [51,52].
The overall change in 14q32 -DMR methylation status, quantitatively described as the 14q-index, presents a promising indictor of diagnostic and prognostic significance that will need to be validated in a larger patient set. Using osteosarcoma patient samples and two mouse osteosarcoma models, our results indicate that normal/early-onset osteosarcoma (onset < 30 years) is characterized by inherited imprinting defects that may contribute to osteosarcoma pathobiology, in contrast with late-onset osteosarcomas that appear to be driven more by different factors. These findings may provide new opportunities for improving the diagnosis and treatment options for osteosarcoma patients. Osteosarcoma tissue samples were obtained from the tissue procurement facility at the University of Minnesota or from the Cooperative human tissue network. Dr. Logan Spector provided triad buccal DNA samples. All patient samples were obtained according to the University of Minnesota institutional review board (IRB) approved protocols. Osteosarcoma tissue samples with clinical follow-up were obtained from Leiden University Medical Center, The Netherlands. Available clinical and follow-up information for these samples are provided in Supplementary Table 1.

DNA and RNA extraction and qRT-PCR
DNA was extracted with a DNA Extraction kit from Qiagen, and RNA extraction was performed with a mirVana Total RNA Isolation kit (Life Technologies). Extractions were carried out according to the manufacturer's instructions. qRT-PCR was performed with 2 µg DNase1 treated total RNA using High Capacity RNAto-cDNA kit (Applied Biosystems); and PCR amplification was performed with SYBR Green I Master kit (Roche Applied Science) in a LightCycler 480. Primers were listed in Supplementary Table 2. GAPDH and 18 s rRNA were used as internal controls. www.impactjournals.com/oncotarget mRNA and miRNA RNA-Seq expression profiling To perform high-resolution quantification of mRNA and miRNA expression using RNA-Seq, we used at least 1 μg of each pool of total RNA according to the mirVana Total RNA Isolation kit (Life Technologies). We sequenced mRNA libraries using 100 bp paired-end runs on a HiSeq 2000 (Illumina) using v3 chemistry. miRNA libraries were sequenced using 50 bp pairedend runs on a HiSeq 2500 (Illumina) using v4 chemistry.
To quality-check raw-sequence FASTQ files, we used FastQC (http://www.bioinformatics.babraham.ac.uk/ projects/fastqc/). The TopHat2 algorithm was used to align the mRNA FASTQ files to the Human UCSC hg19 Assembly genome [54]. Next, to quantify the aligned reads against the reference human genome hg19 transcript annotations, we used Cufflinks and Cuffdiff, according to the standard Fragments Per Kilobase of transcript per Million mapped reads (FPKM) methods [55]. To quantitate miRNA expression levels, mature miRNA sequences were used to search miRNA FASTQ files and normalized to Counts per Million total reads for each sample. For statistical analysis and clustering, we used the Partek Genomics Suite software package (Partek Inc). FPKM (mRNA) or Counts per Million total reads (miRNA) values were log base 2 transformed and mean centered before clustering.

DNA bisulfite pyrosequencing and TA cloning and sequencing
Bisulfate DNA treatment, pyrosequencing and TA cloning and sequencing were performed as previously described [56]. Normal buccal samples were used as control for bisulfite treatment. Two rounds of PCR were carried out to synthesize biotin-labeled specified PCR products with the primers listed in Supplementary Table 2. TA cloning and sequencing were also used in selected samples. Bisulfite-PCR products were cloned into a pCR4.0-TOPO vector and 6~10 clones were selected and sequenced. Methylation levels were calculated as the average value of the selected clones [57].

Chromatin immunoprecipitation (ChIP) assays
The ChIP assay was performed as previously described [58]. Three ~50 mg tissue samples of different groups (normal bone tissue, 14q index (−), and 14q index (+)) were treated with formaldehyde to cross-link histone to DNA. After sonication and pre-cleaning, the samples were incubated with 3~5 μl anti-H3K4me3, H3K27-me3, and IgG (Millipore) at 4°C overnight, respectively, and then with Dynal Magnetic beads from Invitrogen for 1 hour at 4°C. After washing, immunoprecipitated complexes were eluted and histone-DNA crosslinks were reversed. DNA was extracted, and real-time PCR was used to quantify the immunoprecipitated DNA.

Plasmid construction, cell proliferation and apoptosis analyses
Human DIO3 and RTL1 coding sequences were amplified and cloned into pcDNA3.1 (Invitrogen) by XbaI and XhoI. Human DLK1 overexpression plasmid was obtained from Origene. SaOS2 cells were planted in 48 well plates 24 hours before transfection. A total of 0.4 µg overexpression constructs were transfected into SaOS2 cells with lipofectamine. The same amount of pcDNA3.1 empty vector was used as control. Cell proliferation was determined after transfection for 72 hrs using Cell Titer 96 Aqueous One Solution Cell Proliferation Assay (Promega) in a Synergy 2 microplate reader. Apoptosis assays were conducted with the Vybrant apoptosis assay kit #8 (Molecular Probes, Invitrogen) on a BD FACS Canto II according to the manufacturer's instruction.

Spontaneous osteosarcoma mice models
Tumor and matched normal bone tissue samples obtained from Sleeping Beauty [50] (T2/onc; p53 R270 H/+ ;OSx-Cre; ROSA26-LSL-SB11) induced osteosarcoma mouse models were used to determine the methylation levels of 12qF DMR regions. Tumor and matched normal bone tissue samples were obtained from mice homozygous for conditionally disrupted alleles of Trp53 and Rb1, induced by Prx1Cre as described previously [59]. Schematic representation for generation of (Prx1Cre;Rb1fl/fl;Trp53-fl/fl) osteosarcoma in mice homozygous for conditionally disrupted alleles of Trp53 and Rb1 is given is Supplementary Figure 3. All animal studies were IACUC approved; and details used in the study are provided in Supplementary Tables 4 and 5.

Statistics
Statistical analysis such as mean, standard deviation, were calculated for each group analyses and the standard error was used as an index for variability. Student twotailed t-test was performed for statistical analysis. The unpaired t-test was used to compare patient samples; and the paired t-test for apoptosis, cell proliferation assays and buccal tissue of osteosarcoma family member studies. One-way ANOVA was used in the family sample studies. Kaplan-Meier curve was used in the patient survival outcome studies. www.impactjournals.com/oncotarget clinical follow-up information. We thank Dr. Xiaopin Cai for technical support.

GRANT SUPPORT
Grant funding from the American Cancer Society (RSG-13-381-01), Wyckoff Sarcoma Foundation, the Children's Cancer Research Fund and Sobiech osteosarcoma fund to SS are duly acknowledged.