PRDM16 suppresses HIF-targeted gene expression in kidney cancer

Renal cell carcinoma (RCC) is one of the top 10 most common malignancies affecting both men and women (Siegel et al., 2018). The most frequent histology is clear cell RCC (ccRCC). The most common tumor-initiating event in this malignancy is alteration of the von Hippel-Lindau (VHL) gene, which encodes an E3 ubiquitin ligase. The VHL complex targets proteins for proteosomal degradation. The most well-characterized substrates of VHL are the hypoxia inducible factors (HIF-1α and HIF-2α; Ivan et al., 2001; Maxwell et al., 1999). VHL loss, silencing, or mutation can result in the aberrant stabilization of HIFs. Many HIF-α target genes encode proteins that can promote renal carcinogenesis. These include factors critical to fundamental processes such as angiogenesis (vascular endothelial growth factor), cell proliferation and/or survival (TGF-α), and extracellular matrix modulation (matrix metalloproteinase 1).

While VHL loss is critical for HIF signaling, recent evidence indicates that additional mechanisms operate in RCC to promote the HIF axis. Moreover, as there are many HIF target genes, identification of those critical to RCC progression remains poorly characterized. The molecular events that fine-tune this signaling axis may provide novel insight into tumor growth and progression. Recent studies indicate a role for epigenetics in the modulation of HIF signaling in RCC. Improved understanding of the epigenetic mechanisms that modulate the HIF axis and the target genes amplified by these additional mechanisms may lead to novel insights with biomarker and/or therapeutic implications. Here, we identify epigenetic silencing of the transcription factor PR (PRD1-BF1-RIZ1 homologous) domain–containing 16 (PRDM16) in RCC. Loss of PRDM16 leads to the enhanced expression of the HIF-responsive gene semaphorin 5B (SEMA5B), which supports RCC growth in vivo. Collectively, our studies support a novel mechanism by which renal cancer cells modulate HIF-dependent signaling to promote tumor growth.

We recently reported an integrative genomic analysis of DNA methylation and gene expression landscapes of kidney cancer that included three sample groups: normal kidney (n = 9), primary RCC (n = 9), and metastatic RCC tissue deposits (n = 26; Nam et al., 2019). We analyzed the transcriptomes of the deposited data (series GSE105261) using GEO2R (National Center for Biotechnology Information [NCBI]) and identified the top 250 most differentially expressed genes (regardless of directionality) based on the F-statistic, an analysis that combines t-statistics for all pairwise comparisons when more than two sample groups are present (Table S1). Genes previously described to have altered expression in RCC were also altered in this dataset, including SFRP1 (decreased expression in tumors compared with normal) and NDUFA4L2 (increased expression in tumors compared with normal; Gumz et al., 2007; Minton et al., 2016). Based on rank ordering of the F-statistic, we noted that PRDM16 was among the genes with the most significant differential expression in this dataset. We observed reduced expression with tumor progression (Fig. S1 A). PRDM16 encodes for the transcription factor PRDM16. PRDM16 has a prominent role in brown fat physiology (Kajimura et al., 2008; Seale et al., 2007, 2008). Consistent with prior studies in mouse tissues, we identified expression of PRDM16 in brown adipose tissue (with relatively low expression in white adipose tissue (Fig. 1 A). Among the tissues with the highest expression of this factor are the kidney, heart, and small intestine. Consistent with these data, transcriptomic analysis of human tissues also demonstrates relatively high expression of PRDM16 in the kidney (Fig. S1 B; Uhlen et al., 2017). Using real-time quantitative PCR, we confirmed the loss of PRDM16 mRNA expression in tumor samples of a series of patient-matched tumor/normal pairs (Fig. 1 B). Moreover, we observed reduced PRDM16 mRNA in a panel of RCC lines relative to RPTEC (renal proximal tubule epithelial cells) untransformed renal epithelial cells (Fig. S1 C). Consistent with the mRNA data, immunoblotting of patient-matched samples also demonstrated reduced PRDM16 protein expression in tumor (Fig. 1 C). Furthermore, immunohistochemical staining of normal kidney demonstrated nuclear PRDM16 expression in renal proximal tubular epithelium, the cellular origin of ccRCC (Fig. 1 D and Fig. S1 D). In contrast, PRDM16 expression was not detected in tumor cells of ccRCC. PRDM16 is a member of the PR domain–containing family. We therefore examined the relative expression of PRDM16 as well as other PR domain–containing family members in The Cancer Genome Atlas (TCGA) data using the UALCAN analysis portal (http://ualcan.path.uab.edu; Chandrashekar et al., 2017). Consistent with our observations, analysis of TCGA RNA sequencing (RNaseq) data demonstrates a significant loss of expression of PRDM16 in ccRCC (Fig. 1 E). We also analyzed proteomics data on clear cell renal cancer from the National Cancer Institute (NCI)–sponsored Clinical Proteomic Tumor Analysis Consortium, which was recently reported using the UALCAN analysis portal (Chen et al., 2019; Clark et al., 2019). These data clearly demonstrate reduced PRDM16 protein in RCC and therefore validate both our immunoblot and immunohistochemistry analyses (Fig. 1 F). In comparison to the prominent down-regulation of PRDM16, only modest (∼1.3-fold) reductions in PRDM2 and PRDM4 expression were found in RCC (Fig. S1 E). Notably, among tumor samples, lower PRDM16 expression is associated with worsened prognosis (Fig. 1 G). However, no significant association with outcome was observed for PRDM2 or PRDM4 (data not shown).

The robust down-regulation of PRDM16 in RCC tissues and cell lines led us to consider possible mechanisms by which it is silenced. The most common tumor-initiating event in ccRCC is alteration of the VHL gene (Gnarra et al., 1994). Loss of VHL function results in the aberrant stabilization of the HIFs. Analysis of TCGA data demonstrates that PRDM16 is silenced in both VHL mutant and WT ccRCC relative to normal kidney (Fig. 1 H). Prior studies indicate that PRDM family members may be a subject to epigenetic regulation in cancer (Tan et al., 2014; Watanabe et al., 2007). We therefore evaluated whether DNA methylation could promote PRDM16 silencing. We first analyzed methylome (450K array) and gene expression data on tumor samples from the TCGA dataset. Based on analysis using the University of California, Santa Cruz (UCSC) genome browser, there are three CpG-rich regions around the PRDM16 transcription start site (depicted in Fig. 1 I). We identified the 10 CpG loci within this region whose methylation has the strongest anticorrelation with PRDM16mRNA expression. Half of these sites (include the top four most anticorrelated loci) reside within CpG island no. 1 upstream of PRDM16 (shown in yellow; Fig. 1 I and Fig. S2 A). A plot demonstrating the inverse relationship between CpG site methylation and PRDM16 mRNA expression for the CpG locus (cg01514538) with the strongest negative correlation is shown in Fig. 1 J. TCGA methylome data were generated with standard bisulfite sequencing, which cannot resolve between 5-methylcystone (5mC) and 5-hydroxymethylcytosine (5hmC). While 5mC accumulation in gene promoters is associated with silencing, 5hmC accumulation has been shown to accumulate at enhancer regions and may be associated with increased expression (Tsagaratou et al., 2014). We therefore assessed DNA 5mC levels at CpG sites within this island via chromatin immunoprecipitation–quantitative PCR (ChIP-qPCR) using an antibody specific to 5mC. We found significant enrichment of 5mC within the region encompassing cg01514538 in RCC cell lines RXF-393 and RCC4 (Fig. 1 K). Moreover, 5mC levels were higher in multiple regions encompassing CpG sites within this island (cg05346286, cg03969902, and cg01514538) in tumor relative to normal kidney in multiple patient-matched samples (Fig. 1 L). As a positive control, we identified increased 5mC enrichment in RCC (as compared with matched normal kidney) of the promoter of ESSRG, a gene that we recently reported to be methylated in RCC (Fig. S2, B and C; Nam et al., 2019). Collectively, these data demonstrate that the PRDM16 promoter region is methylated in RCC. Analysis of TCGA data demonstrates reduced PRDM16expression in other tumors such as lung adenocarcinoma with promoter methylation (Fig. S2, D–F).

The prominent methylation and silencing of PRDM16 in RCC led us to consider the biological significance of this finding. As noted previously, PRDM16 has a prominent role in brown fat metabolism and has previously been shown to induce expression of the mitochondrial metabolism-related transcription factors in adipocytes including PGC-1α and ERR-γ (encoded by PPARGC1A and ESRRG) as well as uncoupling protein 1 (UCP1; Kajimura et al., 2008; Seale et al., 2007, 2008). Notably, the expression of PPARGC1A and ESRRG has previously been shown to be reduced in RCC (LaGory et al., 2015; Nam et al., 2019). We therefore assessed the effects of PRDM16 on the expression of these factors in RCC cells via stable transduction in RCC cells. Immunoblotting demonstrates that the level of PRDM16 protein expression achieved by transduction is comparable to that in normal kidney (Fig. 2 A). Ectopic expression of PRDM16 in multiple RCC lines examined failed to show a consistent effect on the expression of these factors (Fig. 2 B). In line with these data, we did not observe an increase in oxygen consumption (Fig. 2 C). Prior studies have demonstrated that agents such as forskolin and rosiglitazone can enhance PRDM16 function in mouse brown fat cells (Ohno et al., 2012; Seale et al., 2007). Both agents were able to promote PRDM16’s induction of ESRRG in multiple lines (Fig. S3, A and B). However, no consistent effects were observed on the expression of PPARGC1A or UCP1 or on oxygen consumption (Fig. S3, A–C). We did observe that PRDM16 restoration could suppress proliferation in OSRC-2 cells and to a lesser extent in Caki-1 and 786-O RCC cells (Fig. 2, D–F). Moreover, PRDM16 reduced transwell migration, wound healing, and in vitro invasion of RCC cells (Fig. S4, A–C). These data prompted us to assess the effect of PRDM16 on in vivo tumor growth. In both OSRC-2 and Caki-1 cells, restoration of PRDM16 expression significantly suppressed RCC xenograft growth (Fig. 2, G and H).

Our observation that PRDM16 can suppress in vitro proliferation and in vivo tumor growth despite the lack of induction of PPARGC1A and ESRRG (under basal conditions in the absence of rosiglitazone and forskolin) led us to consider if there were effects of PRDM16 outside of the established role in cell metabolism. We performed RNaseq analysis in 786-O RCC cells plus or minus exogenous expression of PRDM16. Analysis demonstrated that PRDM16 exerts a repressive effect on the transcriptional landscape (Fig. 3, A and B). Of the genes altered with PRDM16 restoration, two thirds of genes were down-regulated, whereas only one third of genes were up-regulated. Pathway analysis revealed that PRDM16 suppresses the expression of genes involved in axonal guidance and signaling, including members of the semaphorin (SEMA) family of transmembrane proteins (Fig. 3, B and C; and Table S2). SEMAs are a class of signaling molecules whose function has been primarily studied in the nervous system. SEMAs bind to their receptors, referred to as plexins, which are known to associate with and activate tyrosine kinases (Artigiani et al., 2004; Giordano et al., 2002; Oinuma et al., 2004). Notably, increased SEMA5B in RCC was also identified in our initial GEO2R analysis as a differentially expressed gene (Table S1 and Fig. S5 A). We assessed the relative expression of SEMA family members repressed by PRDM16 in the TCGA dataset on ccRCC and found that SEMA5B had the highest expression in ccRCC (Fig. 3 D). Furthermore, analysis of TCGA data across all tumor types demonstrates that SEMA5B expression is highest in ccRCC among all other tissues, benign or malignant (Fig. 3 E). RT-qPCR analysis of matched tumor/normal pairs from ccRCC patients confirmed increased SEMA5B (Fig. 3 F). We validated the suppression of SEMA5B by PRDM16 in OSRC-2 and RXF-393 RCC cells (Fig. 3 G). Additionally, this finding was validated in vivo as SEMA5B mRNA expression was lower in PRDM16 expressing xenograft tumors relative to control tumors (Fig. 3 H).

The high expression of SEMA5B in ccRCC led us to consider if VHL has any role on the regulation of SEMA5B expression. We first characterized SEMA5B mRNA expression via RT-qPCR analysis and observed that SEMA5B expression was low in VHL WT lines. VHL mutant lines had variable expression (Fig. 4 A). We therefore examined the expression of SEMA5B as a function of VHL expression in paired RCC lines (786-O and RCC4) plus or minus VHL. Parental RCC4 cells express both HIF-1α and HIF-2α whereas parental 786–0 cells express only HIF-2α. As VHL is known to promote the degradation of HIF, reconstitution of VHL led to an expected reduction in the expression of the HIF target gene GLUT1 in both RCC4 and 786-O cells (Fig. 4 B). VHL restoration in RCC4 cells resulted in reduced expression of PDK1, a known HIF-1α target gene (Fig. 4 B; Kim et al., 2006; Papandreou et al., 2006). Consistent with the absence of HIF-1α expression in 786-O cells, restoration of VHL in these cells had no effect on PDK1. Restoration of VHL led to a dramatic reduction in SEMA5B mRNA (Fig. 4 C). Consistent with VHL’s role in HIF regulation, restoration of VHL in OSRC-2 cells led to reduced HIF-1α and HIF-2α protein (Fig. 4 D). In concert with our transcript data, VHL restoration also led to reduced SEMA5B protein (Figs. 4 D and S5, D and E). As further evidence for a role of HIF in the regulation of SEMA5B, we examined the effects of hypoxia mimetics dimethyloxalylglycine (DMOG) and CoCl₂ in VHL-expressing RCC cells (RCC4/VHL and Caki-1). Both DMOG and CoCl₂ promoted induction of the known HIF target genes (GLUT1 and PDK1) along with SEMA5B (Fig. 4, E and F). The marked induction of SEMA5B expression under hypoxia mimetics led us to consider if the relatively hypoxic environment in vivo could also induce SEMA5B. We therefore examined expression of canonical HIF target genes (Fig. 4 G) and SEMA5B (Fig. 4 H) in Caki-1 cells (transduced with a control vector) grown in vivo compared with cells grown under standard in vitro culture conditions. Caki-1 cells are VHL WT and therefore do not demonstrate increased HIF protein expression under normoxic conditions. While we observed a modest but significant increase in PDK1 and GLUT1 expression in vivo (Fig. 4 G), we found a marked increased (>40-fold) of SEMA5B expression in vivo relative to in vitro (Fig. 4 H). We next determined if PRDM16 could suppress the HIF-mediated induction of SEMA5B. Consistent with parental cells, DMOG treatment of Caki-1 cells stably transduced with control vector led to a significant increase in SEMA5B expression. However, the expression of PRDM16 blunted the ability of DMOG to induce SEMA5B (Fig. 4 I) but not other HIF target genes (Fig. 4 J).

We next assessed the biological significance of elevated SEMA5B expression in RCC cells. Prior studies examining the functional significance of increased SEMA5B in RCC were limited in scope (Hirota et al., 2006). We therefore knocked down SEMA5B expression via stable introduction of shRNA with two different nonoverlapping constructs in the OS-RC-2 cell line. We confirmed target gene knockdown via qPCR and immunoblotting (Fig. 5 A). Both knockdown clones demonstrated reduced proliferation relative to control vector cells (Fig. 5 B). Similar results were obtained in RCC4 cells (Fig. 5, C and D). We next performed a gain-of-function analysis by expressing SEMA5B (hemagglutinin [HA]-tagged) in nontransformed HK2 renal epithelial cells. SEMA5B expression was confirmed by immunoblotting (Fig. 5 E). Increased SEMA5B expression promoted proliferation in HK2 cells (Fig. 5 F). Similar results were obtained when SEMA5B was expressed in 769-P cells (Fig. S5, B and C). Based on these data, we examined the role of SEMA5B in vivo. Notably, we found that SEMA5B knockdown significantly reduced OSRC-2 xenograft growth (Fig. 5 G). We also examine SEMA5B’s functional significance in the setting of intact VHL but with HIF activation. VHL WT Caki-1 cells were cultured in the presence of the hypoxia mimetic DMOG, which results in increased SEMA5B and GLUT1 expression (see sh-Control in Fig. 5 H). As expected, SEMA5B shRNA reduced SEMA5B expression without effects on GLUT1expression (see sh-1 in Fig. 5 H). Knockdown of SEMA5B in cells cultured with DMOG resulted in reduced proliferation (Fig. 5 I). In contrast, SEMA5B knockdown has no effect on proliferation in vehicle-treated cells that have low SEMA5B expression.

Given the functional significance of SEMA5B in promoting RCC proliferation and in vivo tumor growth, we next investigated the mechanism by which PRDM16 repressed SEMA5B expression. Prior studies have demonstrated that PRDM16’s effect on transcription, either promoting or suppressing, are mediated in part by interacting proteins (Kajimura et al., 2008; Seale et al., 2007). As noted before, PRDM16 is expressed at high levels in the kidney, but its function has not been characterized. To gain insight into PRDM16 biology in the kidney, we assessed for PRDM16-interacting proteins in HEK293T cells. We selected these cells because they are renal in origin. To facilitate our studies, we ectopically expressed an N-terminal FLAG-tagged version of PRDM16 in these cells followed by immunoprecipitation (IP) of cell lysates with anti-FLAG antibody and IgG antibody control. Immunoprecipitated fractions were electrophoresed in an SDS-PAGE gel followed by in-gel trypsin digestion of the lanes with subsequent analysis by liquid chromatography–mass spectrometry (LC-MS). Notably, CtBP-1/2 were among the most enriched proteins (Table S3). Prior studies indicate that PRDM16 interaction with CtBPs, which function as corepressors, is critical to the suppression of the white fat gene expression program in brown adipose tissue (Kajimura et al., 2008). We confirmed this interaction by immunoprecipitation in HEK293T and OS-RC-2 cells (Fig. 6, A and B; compare lanes 1 and 2 of immunoprecipitates). Prior studies have identified that PRDM16 interaction with CtBP1 and CtBP2 is dependent on the PLDLS motif in PRDM16 (amino acids 804–808) motif (Kajimura et al., 2008). Mutation of this motif to PLASS disrupts PRDM16/CtBP interaction in adipocytes. In agreement with these data, mutation of this motif in PRDM16 disrupts CtBP interaction (Fig. 6, A and B; compare lanes 1 and 3 of immunoprecipitates) in HEK293T and OS-RC-2 cells. Immunoblotting of inputs from both cell lines demonstrated no significant effects on CtBP protein expression. We next examined the ability of WT and mutant PRDM16 to suppress SEMA5B mRNA and protein expression (Fig. 6, C and D). Consistent with prior data, WT PRDM16 significantly reduced SEMA5B expression, whereas mutant PRDM16 had no effect on SEMA5B. Analysis of ChIP sequencing (ChIP-seq) tracks deposited by the Encyclopedia of DNA Elements (ENCODE) consortium using the UCSC genome browser demonstrates the presence of a CtBP binding site within 2 kb of the SEMA5B transcription start site (Fig. 6 E). We therefore assessed if PRDM16 or CtBP can bind to this region via ChIP-qPCR. Both WT and mutant PRDM16 demonstrated similar binding to this region relative to IgG control. In contrast, CtBP binding to this region was significantly reduced in RCC cells expressing mutant PRDM16 as compared with WT PRDM16 (Fig. 6 F). Collectively, these data indicate that CtBPs’ interaction with PRDM16 promotes CtBPs’ binding to the SEMA5B promoter.

Based on these data, we next assessed whether PRDM16’s interaction with CtBPs is critical for their effects on tumor phenotypes. Whereas WT PRDM16 transduced RCC cells demonstrated reduced proliferation compared with control cells, RCC cells transduced with mutant PRDM16 failed to demonstrate any decrease in proliferation (Fig. 6 G). Furthermore, PRDM16 mutant transduced cells readily grew in vivo in contrast with WT PRDM16-expressing cells, which demonstrated markedly reduced tumor growth (Fig. 6 H). We next assessed the functional significance of PRDM16’s repression of SEMA5B expression. We therefore reintroduced both PRDM16 and SEMA5B in addition to control vector and PRDM16 alone (please see Western blot, Fig. 6 I). OSRC-2 RCC cells with PRDM16 restored demonstrate reduced SEMA5B as well as reduced proliferation and colony formation (Fig. 6, J and K). However, restoration of SEMA5B can partially rescue PRDM16’s effects on proliferation and colony formation (Fig. 6, J and K). These data indicate that PRDM16’s effects are, at least in part, SEMA5B dependent. Collectively, these data demonstrate that PRDM16 antagonizes the proliferative effects of the HIF/SEMA5B axis in RCC. Furthermore, these data indicate that PRDM16’s effects on both the HIF/SEMA5B axis and on tumor growth are dependent on its interaction with CtBPs.

Here, we report that PRDM16 is silenced in the most common type of RCC and that this factor can suppress xenograft tumor growth. We demonstrate a novel role for PRDM16 in the suppression of the HIF-responsive gene SEMA5B. Moreover, our data suggest a role for SEMA5B in tumor growth. Additionally, these are the first data to reveal that PRDM16’s transcriptional repressive properties can promote tumor-suppressive effects. Specifically, PRDM16’s interaction with the corepressor proteins CtBP1/2 is critical for its suppression of SEMA5B expression, proliferation, and in vivo tumor growth. As outlined in Fig. 7, our data have relevance to both VHL WT tumors, which are hypoxic, and VHL mutant tumors, which are pseudohypoxic. In either case, HIF stabilization ensues with promotion of target gene expression.

These are the first data, to our knowledge, to demonstrate a role for PRDM16 in the suppression of solid tumor growth. Prior studies have implicated loss of PRDM family members including PRDM16 in lung cancer (Tan et al., 2014). However, the functional significance of this loss was not fully elucidated as in vivo studies are lacking. Moreover, we would like to point out that prior studies examined PRDM16 methylation through methodologies that use standard bisulfite sequencing. However, bisulfite sequencing cannot resolve between 5mC and 5hmC. This is notable as recent studies indicate that 5hmC enrichment may be associated with activation, as opposed to suppression, of gene expression. Hence, resolution of these marks is critical to determine the precise role of DNA methylation in gene expression.

The majority of data linking PRDM16 with malignancy is in the context of leukemia. In leukemia, gene arrangements of PRDM16 result in loss of the PR domain, which is associated with methyltransferase activity (Lahortiga et al., 2004). Zhou et al. (2016) recently demonstrated that this domain is critical for suppressing leukemia development in a murine mixed lineage leukemia model. In particular, PRDM16’s H3K4 methyltransferase activity promoted expression of the of the transcription factor GFI1b, which in turn suppressed the expression of Hox genes. Of note, they observed progressively increased PRDM16 gene methylation with malignant transformation in this model. These data suggest that PRDM16 methylation, and therefore its silencing, in cancer is not simply a random occurrence. Mechanistic insight of this silencing warrants further study.

Our data indicate that the mechanism of PRDM16’s role in tumor biology is context dependent and that PRDM16’s transcriptional repressive properties are responsible for its suppression of tumor phenotypes in the context of kidney cancer. PRDM16’s repressive effects have been well-characterized in the context of adipocyte biology, where it has been shown to selectively repress white fat gene expression. Our data indicate that SEMA5B is one of the targets of the PRDM16/CtBP repressive complex in renal cancer. SEMAs have recently been found to affect tumor progression by various mechanisms, including modulation of tumor angiogenesis (Basile et al., 2007; Zhou et al., 2017). In addition, SEMA–plexin interaction can induce phosphorylation of tyrosine kinases including Met and Ron to promote invasiveness and tumor metastasis (Conrotto et al., 2005; Giordano et al., 2002). SEMA family members can be broadly divided into eight groups. Class 5 SEMAs, consisting of SEMA5A and SEMA5B, are expressed in vertebrates. There may be some functional redundancy between SEMA5A and 5B. Both can bind plexins A1 and A3, suggesting the possibility of compensation if one or the other is inhibited (Matsuoka et al., 2011). However, the low expression of SEMA5A in RCC (Fig. 3 D) would argue against a compensatory effect upon SEMA5B knockdown. The high expression of SEMA5B in ccRCC coupled with our loss-of-function studies in RCC cells suggest opportunities for intervention. Recent studies indicate that SEMA interaction with their plexin receptor is targetable (Matsunaga et al., 2016). Alternatively, the intracellular signaling cascades activated by SEMA5B might also be targetable. Although ccRCC is refractory to traditional chemotherapeutic agents as well as radiation, the efficacy of small molecule inhibitors has led to the approval of several tyrosine kinase inhibitors for advanced RCC. Hence, delineation of the events of the downstream signaling events of SEMA5B–plexin interaction warrants further investigation.

Our data add to the increasing complexity of HIF signaling in ccRCC. Mutations of VHL are highly prevalent in ccRCC. Our data add to the growing body of evidence of the role of epigenetics in the amplification of HIF signaling in RCC. PBRM1 is commonly mutated in ccRCC (Varela et al., 2011). This is a component of the SWI/SNF (SWItch/Sucrose Non-Fermentable) chromatin remodeling complex. Recent studies indicate that alteration of PBRM1 promotes the expression of HIF-responsive genes (Gao et al., 2017). Recent studies have also implicated DNA hypomethylation in promoting HIF signaling. We recently demonstrated evidence of promoter hypomethylation in RCC tissues of several HIF target genes, including those encoding enzymes such as HK2 and aldolase C which are involved in glycolysis (Nam et al., 2019). Experimental studies of RCC by Vanharanta et al. (2013) demonstrate that DNA demethylation promotes the expression of the HIF target gene CYTIP in an in vivo model of ccRCC lung colonization. Our studies support a novel mechanism for the promotion of HIF signaling through DNA hypermethylation of the PRDM16 gene, which results in enhanced expression of the HIF target SEMA5B gene.

As there are a multitude of HIF target genes, a challenge in the field has been deciphering which of these targets are critical to RCC progression. HIFs as well as their target genes may have contrasting properties, i.e., tumor-promoting or -suppressing. In RCC, HIF-1α is thought to be tumor suppressive whereas HIF-2α is thought to be tumor-promoting (Kondo et al., 2003; Shen et al., 2011). While on balance HIF-1α is tumor suppressive, there is a subset of HIF-1α responsive genes that has experimentally been shown to promote phenotypes such as proliferation including NDUFA4L2 (Minton et al., 2016). Our findings herein, as well as previously published data, indicate that epigenetic alterations can amplify HIF signaling in RCC. Hence, identification of HIF-1α and/or HIF-2α targets, amplified through various epigenetic mechanisms, could pinpoint those genes most contributory to renal carcinogenesis and tumor progression.

There are limitations of our study. While we demonstrate a role for PRDM16 in suppressing SEMA5B, there are likely other targets responsible for PRDM16’s effect on RCC cells. These may include other SEMA family members based on the results of our transcriptomic studies. In addition, we may need to consider the transcriptional activating properties of PRDM16. Prior studies in adipocytes demonstrate that PRDM16 induces the expression of PPARGCA and ESRRG genes. Work by our group and others demonstrates reduced expression of these factors in ccRCC (LaGory et al., 2015; Nam et al., 2019). Additionally, both factors have a known role in the transcription of nuclear encoded genes involved in mitochondrial metabolism, e.g., TCA cycle enzymes and nuclear encoded respiratory chain subunits. Although we saw only modest effects on these factors in vitro that were variable between RCC lines, we cannot exclude effects in vivo.

In summary, our data demonstrate a role for PRDM16 silencing in ccRCC. Our studies suggest that this factor is involved in modulating HIF-mediated SEMA5B expression. These data provide novel insight into the contribution of epigenetics in regulating HIF signaling in ccRCC. Moreover, they provide new lines of study with the potential for therapeutic interventions in tumors driven by the HIF-SEMA5B axis.

786-O-VHL or control (empty) vector was obtained from W.G. Kaelin Jr. (Dana-Farber Cancer Institute, Boston, MA) and were cultured in DMEM. RCC4 (kindly provided by P. Ratcliffe, University of Oxford, Oxford, UK) and HK2 cells were also cultured in DMEM. The Caki-1 cell line was cultured in MEM, and RPMI 1640 media was used for OS-RC-2 and RXF393 cells. HEK-293T cells were grown in DMEM containing 4.5 g/liter glucose. Media for these cell lines were supplemented with 10% fetal bovine serum and 1× penicillin streptomycin. 786-O, Caki-1, 769-P, and HK2 cells were acquired from the American Type Culture Collection. RXF-393 cells were acquired from the NCI. RPTEC cells were obtained from Lonza, and cells were cultured in renal epithelial cell growth basal media. Cell lines were periodically screened for Mycoplasma.

Human PRDM16 cDNA was purchased from Genecopoeia. Human SEMA5B cDNA was acquired from TRANSOMIC technologies. HA-VHL plasmid was acquired from W.G. Kaelin Jr. via Addgene. shRNA constructs (from Sigma-Aldrich) are as follows: pLKO.1-shControl (SHC002), pLKO.1-shSEMA5B1 (TRCN0000060474), and pLKO.1-shSEMA5B2 (TRCN0000060473). Sources of antibodies used in this study are as follows: PRDM16 from Thermo Fisher Scientific (720206); CtBP-1/2 from Active Motif (61261); Flag (for immunoblotting and immunoprecipitation; F1804) SEMA5B (HPA066548) from Sigma-Aldrich; Flag (for ChIP; 14793), HA (3724), and HIF-1α (14179) from Cell Signaling Technology; HIF-2α (GTX632015) from GeneTex; normal IgG (12-371) from Millipore; and β-actin (ab49900) from Abcam.

Gene expression array data from RCC tissues and normal kidney were generated following our previously described protocol with the Illumina Human HT-12 V4.0 expression beadchip (Nam et al., 2019). The raw data and processed data have been uploaded in the Gene Expression Omnibus under accession no. GSE105261. The RNaseq of WT and PRDM16-overexpressing 786-O cells was done in replicates. Adapter sequences and low-quality reads were trimmed from fastq files using trim_galore (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). With TopHat v2.1.0 (Kim et al., 2013), trimmed raw read sequences were mapped to Homo sapiens reference genome (hg38). The aligned reads were assembled into genes, and their abundance was estimated as fragments per kilobase of exon per million fragments mapped using Cufflinks v2.2.1 (Trapnell et al., 2012). Differential expression analysis was performed using the Cuffdiff module of Cufflinks. Genes with absolute fold change of greater than or equal to ≥1.5 and P value <0.05 were considered as differentially expressed genes. The raw and processed data have been uploaded to the Gene Expression Omnibus under accession no. GSE130049. The Database for Annotation, Visualization and Integrated Discovery v6.7 (Huang et al., 2009) was used for gene ontology enrichment analysis on differentially expressed genes.

Plasmid DNAs were transfected into HEK293T cells for production of lentiviral particles. 4 µg of lentiviral vector plasmid, 2 µg of the gag-pol packaging plasmid, and 1 µg of pMD.G (VSV-G) plasmid were transfected with FuGENE6 (Promega). Supernatants were harvest 48 h after transfection and filtered with a 0.45 µm pore size filter. Viral extract was then used to transduce target cells. The lentivirus-infected cells were selected with puromycin.

To measure oxygen consumption rate (OCR) and extracellular acidification rate, Seahorse XFe96 Analyzer (Agilent Technologies) was used as previously described (Isono et al., 2016). The seeding density of cells was 30,000–50,000 cells per well. The following day, cells were washed with extracellular flux media (pH 7.4) and allowed to equilibrate for 1 h before measuring basal OCR/ECAR.

For bioinformatic analyses of PRDM16, TCGA KIRC Kidney Clear Cell Carcinoma HumanMethylation450 methylation values, HiSeqV2 gene expression values, and clinical data were downloaded from the UCSC Cancer Genomics browser. Statistical tests were performed using the Partek genomics suite. To identify cg loci that are hypermethylated in primary tumor tissue versus normal tissue, a t test of differences in the tumor vs. normal mean β values was performed for all cg loci, and resulting t-statistic and unadjusted P values were calculated. To identify cg loci that were inversely associated with PRDM16 gene expression in primary tumors, the PRDM16 gene expression values were compared with the methylation values of cg loci, and Pearson correlation coefficients (r) and unadjusted P values were calculated. TCGA gene expression analysis and survival analysis were performed using the UALCAN web portal (Chandrashekar et al., 2017) and KM-plotter web portal (Nagy et al., 2018), respectively. Pancancer analyses across multiple TCGA datasets were performed with the Gene Expression Profiling Interactive Analysis web server (Tang et al., 2017).

Cell proliferation, wound healing, Boyden chamber migration, and matrigel-based invasion assays were performed to investigate in vitro tumor phenotypes as described earlier (Shelar et al., 2018; Shim et al., 2014). For in vivo xenograft tumor studies, immunodeficient nude (Nu/Nu) mice were obtained from Jackson Laboratory or Charles River and fed a standard chow diet. Cells in sterile PBS were mixed with an equal volume of matrigel and injected (2 × 10⁶ cells/injection) subcutaneously in the flanks of 5–6-wk-old nude mice. Caliper measurements of growing tumors were taken periodically, and the tumor volumes were calculated using the formula (L × W²/2), where “W” is the smallest diameter and “L” is the largest perpendicular diameter.

Total RNA from human tissues was isolated with the RNeasy Mini Kit (Qiagen). RNA from culture cells were extracted with Direct-zol RNA miniprep kit (Zymo Research). cDNA was generated using a High-Capacity cDNA Reverse Transcription kit (Applied Biosystems). qPCR analysis was performed using the Taqman Gene Expression Master reagent mixed with Taqman primers and analyzed with the QuantStudioTM 6K Flex Real-Time PCR System (Applied Biosystems). mRNA expression levels were normalized to human TATA-binding protein (TBP) or human large ribosomal protein RPLPO1, and the normalized cycle threshold (Ct) values were quantified using the double delta Ct analysis. qPCR data represent relative expression. In general, controls in experimental data are normalized to a value of 1. Indicated Taqman primers were predesigned from Applied Biosystems as follows: PRDM16 (Hs00922682_m1), SEMA5B (Hs00400720_m1), PPARGC1A (Hs00173304_m1), ESRRG (Hs00976243_m1), UCP1 (Hs01084772_m1), GLUT1 (Hs00892681_m1), PDK1 (Hs01561847_m1), RPLPO1 (Hs99999902_m1), and TBP (Hs00427620_m1).

All immunoprecipitation steps were performed at 4°C. Cells expressing Flag-tagged PRDM16 WT or PRDM16 mutant were lysed in buffer X (50 mM Hepes, pH 7.5, 150 mM NaCl, 0.1% NP-40, and 1 mM EDTA) containing 1× protease/phosphatase inhibitor (Thermo Fisher Scientific). After clearing, 50 μl slurry of a protein A/G-agarose (Santa Cruz Biotechnology, Inc.) and 2 µg of anti-Flag or control IgG antibody was added to 2 mg of the extracted proteins and rotated for 16 h. Next, the beads were spun down at 600 g, and the immunoprecipitates were washed thrice with buffer Y (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 2 mM EDTA) for 10 min each with rotation. Finally, the beads were boiled in 2× Laemmli buffer to elute the immunoprecipitated proteins. Protein samples prepared in Laemmli buffer were resolved in 4–15% SDS-PAGE (Biorad) followed by transfer onto polyvinylidene difluoride membrane (Immobilon P, Millipore) using standard procedures. The membranes were blocked with either 3% BSA or 5% skim milk and probed with specific antibodies.

To validate hypermethylation of CpG islands located in the upstream of PRDM16 transcription start site, genomic DNA were isolated from RCC cells (RXF-393, RCC4) and patient-matched tumor/normal kidney samples. The 5mC ChIP experiment was performed using the manufacturer’s protocol (Cell Signaling Technology, 76853). To study PRDM16-CtBP binding site near the SEMA5B transcription start site, stably transfected OS-RC-2 cells expressing either Flag-PRDM16 WT or Flag-PRDM16 mutant were cultured in 150 mm tissue culture dishes to 90% confluency (∼6 × 10⁶ cells per plate). The ChIP experiment was done following the manufacturer’s protocol (Millipore, 17-10085). Input DNA and immunoprecipitated DNA were purified (Zymo Research D4003) and analyzed by quantitative PCR using SYBR green fluorescent dye (Applied Biosystems). The protein-bound DNA was calculated as a ratio to input DNA. Primer sequences used in the ChIP assays are as follows: F1/R1 (5′-CAC​ACG​GCT​GAA​GGT​CAT​AG-3′/5′-TTT​CAC​ACG​CTT​TCC​CTC​TT-3′), F2/R2 (5′-CTG​TGG​GTA​ACG​AAG​TTG​CT-3′/5′-ACC​TTC​AGC​CGT​GTG​TTC-3′), F3/R3 (5′-CGG​CCG​AAT​TGG​GAT​CT-3′/5′-GGA​AGG​TGG​CAG​AGC​GA-3′), and F4/R4 (5′-GGG​AAG​GGA​CCT​CGT​GTA​AA-3′/5′-TTA​ACC​CTA​ATC​CGG​CCA​GT-3′).

Experimental results are displayed as either by median or by the mean ± SD. One way ANOVA or two-tailed nonparametric Student’s t test was performed using GraphPad Prism v7.03 to determine significant differences between control and experimental groups as mentioned specifically in the figure legends. P values of <0.05 were considered statistically significant.

All animal studies were conducted in accordance with the National Institutes of Health guidelines for humane treatment of animals and were approved by the Institutional Animal Care and Use Committee. Deidentified human tissue samples were obtained and used in accordance with a protocol approved by the Institutional Review Board at the University of Alabama at Birmingham. Tissues used for transcriptomic analyses were acquired from Cooperative Human Tissue Network, an NCI-supported resource. Per Cooperative Human Tissue Network protocol, patient consent is obtained for the use of samples for research purposes.

Fig. S1 characterizes PRDM16 expression in normal/tumor tissues and RCC lines. Fig. S2 characterizes PRDM16 expression and methylation in TCGA data. Fig. S3 characterizes the effect of PRDM16 on metabolic factors. Fig. S4 characterizes the effect of PRDM16 on in vitro phenotypes in RCC cells. Fig. S5 characterizes SEMA5B expression in ccRCC, the SEMA5B effect on proliferation in RCC cells, and SEMA5B expression by VHL. Table S1 is a gene list of differentially expressed genes upon analysis of transcriptomic data from normal kidney, primary tissue, and metastatic tissues deposits. Table S2 summarizes RNaseq analysis of 786-O cells plus or minus PRDM16. Table S3 summarizes PRDM16 (FLAG-tagged) interacting protein analysis by LC-MS.

The research reported in this article was supported by National Institutes of Health/NCI grant R01CA200653 and Department of Veterans Affairs grant I01BX002930 (S. Sudarshan) and in part by the University of Alabama at Birmingham O’Neal Comprehensive Cancer Center (P30CA013148).

Author contributions: A. Kundu, H. Nam, S. Shelar, G. Brinkley, S. Karki, T. Mitchell, R. Kirkman, Y. Tang, G.C. Rowe, and S. Wei acquired data. A. Kundu, D.S. Chandrashekar, C.B. Livi, T. Mitchell, P. Buckhaults, S. Varambally, and S. Sudarshan analyzed data. A. Kundu and S. Sudarshan contributed to the conception, design, and writing.