SMAD family member 3 (SMAD3) and SMAD4 repress HIF2α-dependent iron-regulatory genes

Hypoxia-inducible factor 2α (HIF2α) directly regulates a battery of genes essential for intestinal iron absorption. Interestingly, iron deficiency and overload disorders do not result in increased intestinal expression of glycolytic or angiogenic HIF2α target genes. Similarly, inflammatory and tumor foci can induce a distinct subset of HIF2α target genes in vivo. These observations indicate that different stimuli activate distinct subsets of HIF2α target genes via mechanisms that remain unclear. Here, we conducted a high-throughput siRNA-based screen to identify genes that regulate HIF2α's transcriptional activity on the promoter of the iron transporter gene divalent metal transporter-1 (DMT1). SMAD family member 3 (SMAD3) and SMAD4 were identified as potential transcriptional repressors. Further analysis revealed that SMAD4 signaling selectively represses iron-absorptive gene promoters but not the inflammatory or glycolytic HIF2α or HIF1α target genes. Moreover, the highly homologous SMAD2 did not alter HIF2α transcriptional activity. During iron deficiency, SMAD3 and SMAD4 expression was significantly decreased via proteasomal degradation, allowing for derepression of iron target genes. Several iron-regulatory genes contain a SMAD-binding element (SBE) in their proximal promoters; however, mutation of the putative SBE on the DMT1 promoter did not alter the repressive function of SMAD3 or SMAD4. Importantly, the transcription factor forkhead box protein A1 (FOXA1) was critical in SMAD4-induced DMT1 repression, and DNA binding of SMAD4 was essential for the repression of HIF2α activity, suggesting an indirect repressive mechanism through DNA binding. These results provide mechanistic clues to how HIF signaling can be regulated by different cellular cues.

Hypoxia plays a fundamental role in the pathophysiology of common causes of mortality, including ischemic heart disease, stroke, cancer, chronic lung disease, and congestive heart fail-ure (1). In cancer, hypoxia is considered a hallmark for solid tumors. In response to hypoxia, tumor cells activate genes that are critical in angiogenesis, cell survival, cell proliferation, and glucose metabolism (2)(3)(4). Hypoxia signaling is mediated by the central transcriptional factor hypoxia-inducible factor (HIF), 3 a family of per-ARNT-Sim of basic helix-loop-helix proteins, which includes three isoforms HIF1␣, HIF2␣, and HIF3␣ (5)(6)(7)(8)(9). HIF2␣ signaling is activated rapidly and is important for tumor inflammation and colon cancer progression (10 -13). In addition to hypoxic regulation of HIF2␣, HIF2␣ is stabilized by low iron. In response to low iron, HIF2␣ regulates iron absorption (14). Several iron-absorptive genes are direct HIF2␣ target genes in vivo (15). However, iron deficiency and iron overload disorders do not up-regulate any hypoxic HIF2␣ target genes in the intestine. Similarly, inflammatory and tumor foci can induce a distinct subset of HIF2␣ targets genes that are not regulated by iron demand. These observations demonstrate that different HIF2␣ stimuli activate different subsets of HIF2␣ target genes.
Mothers against decapentaplegic homolog (SMAD) 3 and SMAD4 are ligand-stimulated transcription factors, which are similar to HIF2␣, and play essential role in inflammation, colon cancer progression, and iron regulation (16 -19). Bone morphogenetic protein (BMP) and transforming growth factor (TGF) ␤ are canonical ligands that activate SMAD signaling. The TGF␤ superfamily signals exert growth inhibition effect on normal epithelial cells, and the loss of function promotes tumorigenesis (20). Moreover, BMP signaling is essential in regulating the hepatic master iron-regulatory hormone hepcidin. Upon ligand binding to type I and type II, TGF␤ and BMP receptors lead to phosphorylation of receptor-activated SMADs (Smad2 and Smad3) at conserved C-terminal Ser-Ser-Xaa-Ser motifs (19,21). The receptor-activated SMADs partner with common SMAD (SMAD4), translocate to the nucleus and drive transcription (22)(23)(24). The structure of SMAD protein is highly homologous, consisting of an N-terminal Mad homology domain 1 (MH1), a linker region, and a C-terminal Mad homology domain 2 (MH2) (23 cro ARTICLE binding domain, which facilitate the nucleus import, whereas MH2 domain is important for protein-protein binding (25,26).
Using a high-throughput siRNA screen for genes that modulate HIF2␣ activity, SMAD3 and SMAD4 were identified as selective repressors for HIF2␣ iron-regulatory genes but not angiogenic and glycolytic genes (27). Our data demonstrate that SMAD3 and SMAD4 are iron-regulated transcription factors that are decreased following iron deficiency, leading to a derepression and optimization of HIF2␣-dependent iron absorption. Moreover, it provides a mechanistic insight into how a single transcription factor can regulate different target genes depending on the upstream stimuli.

SMAD4 was an essential repressor of HIF2␣-dependent DMT1 activation
HIF2␣ modulators were assessed using a high-throughput siRNA screen for genes that regulate the Dmt1 promoter (a HIF2␣-selective promoter) (15,27). DMT1 has four isoforms because of the combination of two 5Ј processing (transcribed from two distinct regulatory regions) and two 3Ј-UTR (presence or absence of an iron-response element (IRE)) variants: DMT1A, DMT1A-IRE, DMT1B, and DMT1B-IRE. Previous studies have shown that DMT1A is the most abundant isoform in the duodenum (28) and HIF2␣ specifically regulates DMT1A (Ϯ IRE), but not DMT1B (Ϯ IRE) (15). Therefore, to assess the role of SMADs in intestinal HIF2␣ regulation, the DMT1A promoter was utilized (a schematic representation and full promoter sequence are shown in Fig. S1). In brief, in HCT116 cells overexpressing HIF2␣, DMT1 promoter luciferase activity was assessed using siRNA-based screen with a druggable target library (Fig. 1A). Previously we identified a battery of genes that were shown to be essential activators for HIF2␣ (27). In the present work, the data were assessed for genes that repressed HIF2␣ function. Through this analysis, we identified 37 genes that repressed HIF2␣ activity on the Dmt1 promoter ( Table   Figure 1. SMAD4 is essential for the suppression of HIF2␣ iron target genes. A, schematic of the high-throughput siRNA screening assessing HIF2␣induced Dmt1 promoter luciferase assay in HCT116 cells. B, DMT1 promoter luciferase in HCT116 expressing scrambled shRNA (control, Con), or two different shRNAs for SMAD4 (S4shRNA1 or S4shRNA2) cell lines when transfected with different amounts of HIF2␣ overexpression construct. Right panel, SMAD4 Western blotting analysis. C and D, DMT1 promoter luciferase with either empty vector (EV) control or HIF2␣-overexpressing construct (HIF2␣) in HCT116 control cells or HCT116 SMAD4 knockout (C, S4KO) or SMAD3 knockout (D, S3KO) cell lines. The right panels are SMAD3 and SMAD4 Western blotting analysis. Luciferase data were normalized to ␤-galactosidase, and Western blots were normalized to GAPDH. *, p Ͻ 0.01; **, p Ͻ 0.001 compared with control or as indicated on the graph. SMAD4 Suppresses HIF2␣ activity S1). HIF2␣-induced DMT1 luciferase activity was significantly potentiated following siRNAs specific for SMAD3 or SMAD4 (Fig. 1A). To confirm the role of SMAD3 and SMAD4 in HIF2␣ regulation, SMAD4 stable knockdown in HCT116 cells were generated using shRNAs. SMAD4 protein level was confirmed with Western blotting analysis (Fig. 1B), and the cells were assessed using the Dmt1 promoter luciferase assay. Consistent with the siRNA screen, knockdown of SMAD4 potentiated HIF2␣ activity (Fig. 1B). These data were further confirmed using CRISPR/CAS9-mediated knockout of SMAD4 (Fig. 1C), but knockout of SMAD3 did not result in potentiation of HIF2␣ activity (Fig. 1D).

SMAD3 and SMAD4 were sufficient to selectively suppress HIF2␣-dependent iron-regulatory promoters
To further assess the role of SMADs, SMAD3 and SMAD4 were overexpressed, and HIF2␣ activity was evaluated. Dmt1, Dcytb, and Fpn promoter luciferase activities were significantly increased in HCT116 cells following HIF2␣ overexpression. SMAD3 and SMAD4, alone or combination, significantly inhibited HIF2␣-induced activity ( Fig. 2A). However, SMAD3 and SMAD4 exerted opposite effects on HIF2␣ glycolytic and inflammatory genes (Fig. 2B). Consistent with the data in HCT116 cells, the repressive function of SMAD3 and SMAD4 was also demonstrated in SW480 human colon cancer-derived cell line (Fig. 2C). Interestingly, the highly homologous SMAD2 did not repress HIF2␣ activity (Fig. 2D), and SMAD3 and SMAD4 did not alter HIF1␣ activity (Fig. 2E). Together, these data suggest that both SMAD3 and SMAD4 are sufficient to repress HIF2␣ activity; however, only SMAD4 is essential.

Low-iron decreased SMAD3 and SMAD4 protein in vitro and in vivo
To understand whether SMAD signaling was integrated into cellular iron content, SMAD3 and SMAD4 levels were assessed following changes in cellular iron levels. Deferoxamine (DFO), an iron chelator, significantly decreased SMAD3 and SMAD4 protein levels (Fig. 3A). The iron chelation was confirmed by assessing iron storage protein ferritin (FtnH) expression. Interestingly, this decreased SMAD3 and SMAD4 protein levels were reversed once iron was restored by adding back ferric ammonium citrate (FAC) (Fig. 3A). In vivo experiments also confirmed the negative feedback regulation between iron and SMAD protein levels. The duodenum from the mice that were on 2 weeks of iron-enriched diet (350 ppm), iron-replete (35 ppm), or low-iron (Ͻ5 ppm) diet demonstrated a dose-dependent decrease of SMAD3 and SMAD4 levels (Fig. 3B). SMAD signaling is initiated via diverse BMP and TGF␤ ligands; therefore it is unclear whether intestines have basal SMAD3 and SMAD4 signaling. To examine the distribution of SMAD3 and SMAD4, a cellular fractionation was performed. At the basal level, SMAD3 is present in cytosol, nucleus including both chromatin-bound and non-chromatin-bound fractions in HCT116 cells and mouse duodenal tissue (Fig. 3C), whereas SMAD4 was mainly localized to the nucleus and enriched in chromatin-associated fraction (Fig. 3C). These data demonstrate that SMAD3 and SMAD4 are responsive to the changes of cellular iron, and SMAD signaling is highly active in normal intestine.

Iron starvation degraded SMAD3 and SMAD4 via the proteasome
To understand the mechanism of decreased SMAD3 and SMAD4 protein levels following iron deprivation, gene expression was assessed in the small intestine from WT mice on ironenriched or low-iron diet for 2 weeks. There was no difference in either SMAD3 or SMAD4 expression in the iron-deficient mice, compared with the mice fed with iron-enriched diet (Fig.  4A). To demonstrate that the loss of SMAD3 and SMAD4 upon iron deprivation was via post-transcriptional mechanisms, expression constructs of FLAG-tagged SMAD3 and SMAD4 were overexpressed in HCT116 cell line. The cells were treated with DFO, and SMAD3, SMAD4, or FLAG Western blots were performed. A decrease in SMAD3, SMAD4, and FLAG was observed following DFO treatment (Fig. 4, B and C). To investigate whether this was mediated by proteasomal degradation, the cells were also treated with the proteasomal inhibitor Mg132. Mg132 completely rescued SMAD3 and SMAD4 expression following DFO treatment (Fig. 4, B and C). SMAD3 and SMAD4 stability during iron deprivation were assessed by cycloheximide chase assays in DFO-treated or untreated HEK293T cells. DFO treatment reduced SMAD3 half-life from 3 h (29) to 1 h and that of SMAD4 from 16 h (30) to 8 h (Fig. 4D). To further study whether iron deprivation induced the proteasomal degradation of SMAD3 and SMAD4, an immunoprecipitation assay was performed in HCT116 cells transfected with FLAG-SMAD3 or FLAG-SMAD4 plasmid, followed by 1 h of Mg132 pretreatment and overnight DFO treatment. Following immunoprecipitation with anti-FLAG magnetic beads, ubiquitin Western blotting analysis showed dramatic increase of ubiquitin conjugation with SMAD3 and SMAD4 (Fig. 4E), suggesting that the low iron-induced decrease in SMAD3 and SAMD4 was through the ubiquitin proteasome pathway (Fig.  4E). These data illustrate that low-iron levels result in SMAD3 and SMAD4 protein degradation through ubiquitin proteasome pathway.

SMAD3 and SMAD4 repressive function was not via phosphoactivation or direct protein-protein interaction
To examine the mechanism by which SMAD3 and SMAD4 repress HIF2␣, direct protein-protein interaction was assessed. Co-immunoprecipitation was performed in HEK293T cells following transfection of HIF2␣, SMAD3-FLAG, or SMAD4-FLAG or the co-transfection of HIF2␣ and SMAD3-FLAG or HIF2␣ and SMAD4-FLAG. Immunoprecipitation with FLAG and then HIF2␣ and FLAG immunoblotting were performed. HIF2␣ was not observed after FLAG pulldown, suggesting that there was no direct protein-protein interaction between HIF2␣ and SMAD3 or SMAD4 (Fig. 5A). Next, we assessed whether SMAD3 phosphorylation is required; three essential phosphoserine sites were mutated to alanine (S33A) (31). Interestingly, S33A still significantly suppressed HIF2␣-induced DMT1 luciferase activity to the same extent as WT SMAD3. This indicated that the cross-talk between HIF2␣ and SMADs was not through SMAD3 activation (Fig. 5B). Interestingly, a consensus SBE (32) in the DMT1 pro-SMAD4 Suppresses HIF2␣ activity moter region was identified. To examine whether the suppression was through direct binding of SMAD3 or SMAD4 to DMT1 promoter, the consensus SBE sequence was deleted (DMT1⌬SBE). SMAD3 and SMAD4 were still able to suppress HIF2␣-induced DMT1 activity (Fig. 5C). To further identify the region of SMADmediated repression, DMT1 promoter luciferase activity was per-  formed in 5Ј-truncated mutants of the DMT1 promoter. Both SMAD3 and SMAD4 significantly suppressed HIF2␣-induced DMT1 promoter activity that contained 0.2 kb of the proximal promoter. This suggests the possibility of novel interaction(s) with transcription factors that bind in this 0.2-kb region (Fig. 5D). Using publicly available databases JASPAR and GTRD, three overlapping transcription factor-binding sites in the 0.2-kb region were identified, namely FOXA1, KLF4, and Stat3. Upon deletion of these binding sites on DMT1 luciferase promoter (⌬FOXA1, ⌬KLF4, and ⌬Stat3), ⌬FOXA1 completely rescued the SMAD4induced suppression (Fig. 5E). Taken together, SMAD3 and SMAD4 suppressive function on DMT1 activity was indirect and possibly through the regulation of FOXA1.

SMAD4 DNA binding was critical for the suppression of HIF2␣ iron target genes
To understand whether SMAD4-mediated repression is indirect and is due to the activation of SMAD4-dependent target genes, DNA-binding mutants of SMAD4 were generated. Mutations of two critical residues (Arg-81 and Lys-88) in the MH1 dramatically reduce the ability of the SMAD4 to bind chromatin (33), which our data further confirmed (Fig. 6B). In addition, both mutations S4R81A and S4K88A did not alter SMAD4 expression (Fig. 6A). Strikingly, the suppression of DMT1 activity by SMAD4 was completely abolished by S4R81A and S4K88A mutants (Fig.  6C). These data indicate that the SMAD4 DNA binding is essential for the suppression of HIF2␣ iron target genes.

Intestine-specific SMAD4 knockout mice had increased expression of iron transporters
Mice with an intestinal SMAD4 deletion (Smad4 ⌬IE ) were generated by crossing the Smad4 F/F mice to the villin-Cre mice. SMAD4 Western blotting analysis showed successful SMAD4 disruption in the intestine of the Smad4 ⌬IE mice compared with the littermate controls (Smad4 F/F ) (Fig. 7A). Expression of two canonical HIF2␣-iron target gene Dmt1 and Dcytb were unchanged in the Smad4 ⌬IE mice on an iron-enriched diet. However, Dmt1 and Dcytb were robustly potentiated following 1 week of low-iron diet (Fig. 7B). These data were consistent with protein-level changes, DMT1 was significantly elevated in the Smad4 ⌬IE mice, and DCYTB, although not significant, demonstrated an increase in expression compared with littermate controls (Fig. 7C).

Discussion
Intestinal HIF2␣ is a master regulator for iron absorption, glycolysis, angiogenesis, and inflammation (34 -39). Interestingly, different HIF2␣ stimuli activate a distinct subset of HIF2␣ target genes. In the present manuscript, we provide a mechanism that is essential for the regulation of the iron response genes by HIF2␣. We show that SMAD3 and SMAD4 are iron-regulated transcription factors that limit HIF2␣ transcriptional activity under normal cellular iron. Under low-iron conditions, SMAD3 and SMAD4 are degraded by the proteasome, leading to a derepression of HIF2␣ activity and an increase in iron-absorptive genes (Fig. 7D).
Our data demonstrate that SMAD3 and SMAD4 do not directly bind to HIF2␣, and the consensus SBE is not required for SMAD3 or SMAD4 repression. Although it is possible that noncanonical SBEs are present in the proximal promoters of iron transporters, an in-depth search for cryptic SBEs still did not identify any additional interactions. Interestingly, a 0.2-kb DMT1 promoter luciferase construct was repressed by SMAD4, indicating a noncanonical and novel SMAD4 regulatory site. Data mining identified forkhead box protein A1 (FOXA1), which is known as a pioneer transcription factor and is essential in providing chromatin access to SMAD4 and its subsequent DNA binding (40). The deletion of FOXA1 on DMT1 promoter luciferase completely abolished SMAD4 induced DMT1 activity inhibition, suggesting a SMAD4 -FOXA1-DMT1 interaction. Furthermore, the DNAbinding mutants of SMAD4 completely abolished the repressive activity of SMAD4. Taken together, these data suggest that SMAD signaling via indirect mechanisms and possibly through FOXA1 transcriptional activation of a repressor may lead to HIF2␣ inhibition. Further work is being done to identify the precise mechanism.
Our data indicated that SMAD3 and SMAD4 are constitutively expressed and highly active in tissues under basal iron conditions. Currently the mechanism for driving SMAD3 and SMAD4 activity in normal intestine is not known; however, BMP signaling is constitutively high in the villus. Moreover, other signals that are necessary for cell differentiation and intestinal stem cell homeostasis may cross-talk with SMAD3 and SMAD4 (41). Under low-iron diet, SMAD3 and SMAD4 are degraded via a proteasome-mediated mechanism. The bestcharacterized regulation of SMAD3 and SMAD4 signaling is via ligands such as TGF␤ and BMPs. However, Smurf2 is a specific ubiquitin-protein isopeptide ligase specific for SMADs, and future work will focus on understanding whether the ligases are activated in low-iron conditions.
In conclusion our study reports evidence of a novel cross-talk between HIF2␣ and SMAD3 and SMAD4. The data demonstrate that this repressive mechanism is an important molecular switch to specifically shift HIF2␣ regulation to iron gene targets.

SMAD4 Suppresses HIF2␣ activity
mice had access to food and water ad libitum. All the animal studies were carried out in accordance with the Institute of Laboratory Animal Resources guidelines and approved by the University Committee on the Use and Care of Animals at the University of Michigan.

Cell lines and cell culture
Human colon cancer cell lines HCT116 and SW480 and human embryonic kidney cell line HEK293T were used. These cell lines express SMAD proteins basally and HIF2␣-dependent iron-regulatory gene promoters are responsive in these cell lines. SMAD4shRNA are commercially available from Dharmacon (Lafayette, CO). HCT116 SMAD3 and SMAD4 knock-out cell lines were generated by using CRISPR-Cas9 technology according to the Zhang lab protocol (43); the sgRNA sequences used for SMAD3 and SMAD4 are in Table 1. All cells were maintained in complete Dulbecco's modified Eagle's medium (supplemented with 10% fetal bovine serum and 1% antibiotic/ antimycotic agent) at 37°C in 5% CO 2 and 21% O 2 .

Luciferase assay
The cells were seeded into a 24-well plate at a cell density of 5 ϫ 10 4 cells/well. The promoter luciferase constructs for divalent metal transporter 1 (Dmt1) (1 kb unless otherwise indicated as 0.4 or 0.2kb), duodenal cytochrome B (Dcytb), ferroportin (Fpn), enolase (eno1), or tumor necrosis factor (Tnf) ␣ The left panel shows that constitutively expressed and highly active SMAD proteins restrict hyperinduction of iron-related genes by HIF2 under iron deficiency, whereas the right panel shows that decreased iron level results in SMAD protein degradation and a release of the molecular brake to increase HIF2 iron target gene. **, p Ͻ0.01; ***, p Ͻ0.001; ****, p Ͻ0.0001 compared as indicated on the graph.

SMAD3 and SMAD4 protein half-life
HCT116 cells were pretreated with cycloheximide (50 mg/ml) for 1 h and then treated with or without DFO (150 M) for indicated time points. SMAD3 and SMAD4 half-lives were determined by relative densitometric analysis from SMAD3, SMAD4, and GAPDH Western blots.

Chromatin-associated protein isolation
Cytosol, nucleus, chromatin-bound, and non-chromatinbound proteins were isolated from cell lines or mouse duodenal tissue using previously described methods (47). In brief, 1 ϫ 10 7 to 2 ϫ 10 7 cells were collected and washed with 2 ml of PBS. For tissues, mucosal duodenal scrapes were collected. The samples pellets were resuspended in 100 -200 l of buffer A (10 mM HEPES, pH 7.9, 10 mM KCl, 0.34 M sucrose, 10% glycerol, 1 mM DTT, 0.1% Triton X-100, 1.5 mM MgCl 2 , and protease inhibitor mixture). The pellet was homogenized with a glass Dounce homogenizer (20 -30 strokes), then transferred to 1.5-ml centrifuge tubes, and incubated on ice for 8 min. The supernatant and nuclei pellet were separated by centrifugation (5 min at 1,300 ϫ g in 4°C). For the cytosolic fraction the supernatants were transferred to a new tube and spun at high speed (5 min at 20,000 ϫ g in 4°C). The nuclei were washed once in buffer A and processed in a Dounce homogenizer for an additional 20 -30 times. Following centrifugation (5 min at 1,300 ϫ g in 4°C) the supernatants were discarded. The nuclei were lysed for 30 min in 50 -80 l of buffer B (3 mM EDTA, 0.2 mM EGTA, 1 mM DTT, and protease inhibitor mixture), and insoluble chromatin (chromatin bound protein) and soluble (non-chromatin-bound protein) fractions were separated by centrifugation (5 min, 1,700 ϫ g, 4°C). The pellet was washed with 0.5-1 ml of buffer B and resuspended in nucleus digestion buffer (radioimmune precipitation assay ϩ phosphatase inhibitor ϩ protease inhibitor mixture).

Quantitative real-time RT-PCR
RNA extraction, reverse transcription, and quantitative PCR were described previously (48,49). The primers used in the study were listed in Table 1. The data were normalized to ␤-actin and expressed as fold difference from controls.

Statistical analysis
The results were expressed as means Ϯ S.D. Significance between two groups were calculated by independent t test, and the significance among different groups was tested using one-way analysis of variance followed by Dunnett's post hoc comparisons.