The HIF-2 Transcription Factor Mediates Resistance to Ferroptosis in Pancreatic Cancer

Summary

Ferroptosis is an iron-dependent form of cell death converging on lipid peroxidation first identified by examining compounds with enhanced lethality to KRAS mutant cells. Despite over 90% of pancreatic adenocarcinoma (PDAC) tumors harboring KRAS mutations, PDAC exhibits relative resistance to ferroptosis compared to other tumor types, and the mechanisms behind this resistance remain unclear. Here we report exposure to pancreatic tumor interstitial fluid in synergy with hypoxia induced robust protection against ferroptosis in a manner dependent on the hypoxia-inducible transcription factor 2 (HIF-2). HIF-2 upregulates expression of both components of the system Xc- cystine transporter and trans-sulfuration pathway enzymes CBS and CTH to increase intracellular cysteine level, enabling anti-ferroptotic glutathione production. HIF-2 also induces the Parkin mitophagy factor and suppressed mitochondrial function and ROS generation. Altogether, our findings uncover an unforeseen role of the HIF-2 transcription factor as a coordinator of anti-ferroptotic mechanisms in pancreatic cancer

eTOC Blurb:

Hubbi et al. reveal that hypoxia and pancreatic tumor interstitial fluid cooperate to suppress ferroptosis in pancreatic cancer through HIF-2 activity. By transcriptionally regulating glutathione metabolism and mitochondrial function, HIF-2 enables tumor survival in metabolically hostile environments, defining a tissue-specific role in pancreatic ductal adenocarcinoma.

Graphical Abstract

Introduction

Pancreatic adenocarcinoma is a leading cause of cancer mortality with poor survival rates¹. Tumors are typically diagnosed at an unresectable stage, with limited response to both chemotherapy and immunotherapy. There is thus a major unmet clinical need for new treatment concepts.

More than 95% of pancreatic ductal adenocarcinoma (PDAC) cases harbor mutations in KRAS¹. Mutant KRAS-specific drugs hold significant hope², though resistance remains a challenge. Before the emergence of KRAS-specific drugs, two key compounds that selectively target KRAS mutant cells were discovered: RSL-3 (Ras-selectively lethal)³ and erastin (eradicator of Ras and ST-expressing cells)⁴. Erastin blocks cystine import through the system Xc- transporter (a heterodimer of SLC7A11 and SLC3A2), preventing production of the antioxidant glutathione. In contrast, RSL-3 inhibits glutathione peroxidase 4 (GPX4) which uses glutathione to diminish lipid peroxidation⁵. Both drugs were ultimately found to induce an iron-dependent cell death pathway, which converges on lipid peroxidation, termed ferroptosis⁶. A key pathway that regulates ferroptosis and sensitivity to both GPX4 inhibitors and system Xc- inhibitors is the glutathione/GPX4 axis⁷. There are other pathways that selectively regulate sensitivity to ferroptosis induced by either erastin or RLS3. In particular, mitochondrial function and downstream mitochondrial reactive oxygen species (ROS) were found to regulate sensitivity to erastin or ferroptosis induced by cysteine deprivation^8,9.

A key feature of the pancreatic environment is severe hypoxia, driven by the combination of a thick fibrotic stroma and reduced vascularity. In fact, in one ranking of tumor types by vascularity, PDAC was found to be the most hypovascular tumor type¹⁰. Much of the adaptation to hypoxia is mediated by the hypoxia inducible transcription factors HIF-1 and HIF-2. These are composed of heterodimers of oxygen-sensitive alpha subunits and a shared and constitutively expressed beta subunit (HIF-1β or ARNT). Under normoxic conditions, the alpha subunits are subject to proline hydroxylation reactions which allow binding to the von Hippel Lindau (VHL) tumor suppressor, protein ubiquitination, and proteasomal degradation¹¹. Under hypoxic (<10% oxygen) conditions, the oxygen-sensitive hydroxylation reactions do not occur, VHL cannot bind, and the alpha subunits dimerize with ARNT to form HIF-1 or HIF-2. Upon DNA binding and recruitment of co-factors, the HIF transcription factors mediate the expression of hundreds of genes involved in the adaptation to hypoxia, including angiogenesis, metabolic changes, and erythropoiesis¹².

We report here that hypoxia protects a panel of PDAC lines from ferroptosis induced with either erastin or RSL3, and this effect was synergistic with exposure to pancreatic tumor interstitial fluid (TIFM). TIFM was capable of inducing a hypoxia-like response, even under normal oxygen levels, and had a synergistic effect with hypoxia at a number of target genes. We find that the protective effect of hypoxia is mediated by the HIF-2 transcription factor, which upregulates several genes involved in ferroptosis resistance. These include: (1) both components of the cystine transporter, SLC7A11 and SLC3A2, (2) key enzymes in the transsulfuration pathway CBS and CTH, (3) Parkin, a regulator of mitochondrial function that we identify as a regulator of ferroptosis, and (4) coordinated downregulation of genes involved in oxidative phosphorylation. We conclude HIF-2 mediates a coordinated response to protect against ferroptosis in PDAC cells.

Results

Hypoxia induces a protective effect against ferroptosis in PDAC lines

Despite their dependency on activated KRAS, PDAC lines under normoxia with regular media do not have increased susceptibility to ferroptosis inducers compared to other cell types in DepMap¹³. Since a previous report indicated that HIF activity enhanced ferroptosis susceptibility in renal cell carcinoma¹⁴, we tested the hypothesis that PDAC lines would be sensitive to ferroptosis inducers when HIF is induced under hypoxia. Surprisingly, hypoxia led to substantial resistance to ferroptosis induced by the GPX4 inhibitor RSL-3 in all PDAC cell lines tested (Fig. 1A).

Figure 1: Hypoxia Induces a Protective Response in PDAC Cell Lines.

(A) A panel of human PDAC cell lines were exposed to the indicated doses of RSL3 under normoxic (21% O₂) or hypoxic (1% O₂) conditions after 24 hours pre-treatment exposure to either normoxia or hypoxia. After 72 hrs of treatment, cell viability was measured with the Cell-Titer Glo assay.

(B) A panel of human PDAC cell lines were exposed to normoxic (21% O₂) or hypoxic (1% O₂) conditions, with regular media (DMEM + 10% FBS) or TIFM for 24 hours. Cells were then exposed to the indicated doses of erastin. After 72 hrs of drug treatment, cell viability was measured with the Cell-Titer Glo assay.

(C) PANC-1 or MiaPaCa cells (200,000) were seeded into 6 well plates and next day exposed to regular media/normoxia or hypoxia/TIFM for 24 hours, then treated with 5 μM erastin for 24 hours. Samples were processed using ELISA-based 4-HNE assay kit.

Data are presented as mean ± SEM of 3–4 biological replicates.

We next tested the effect of hypoxia against ferroptosis induced by the system Xc- inhibitor erastin. Whereas in CFPAC-1 cells hypoxia induced substantial protection against erastin-induced ferroptosis, in three other human PDAC cell lines the effect of hypoxia alone was mild or absent (Fig. 1B). To better recapitulate the metabolic milieu PDAC is exposed to in the tumor microenvironment, we tested the effect of previously described pancreatic tumor interstitial fluid medium (TIFM) alone or in combination with hypoxia. Unlike regular medium (DMEM + 10% FBS), TIFM recapitulates the approximate concentrations of amino acids, sugars, and metabolites as measured directly from pancreatic cancer samples^15,16. Like hypoxia, TIFM alone could induce a protective response against erastin with some cell line variability. However, the combination of hypoxia and TIFM led to nearly complete protection against high doses of erastin (up to 15 μM) in all PDAC cell lines tested (Fig. 1B). We used an ELISA-based assay to detect 4-hydroxynonenal (4-HNE)-protein adducts as a measure of lipid peroxidation, a more specific hallmark of ferroptosis. Whereas erastin induced an increase in lipid peroxidation measured by this assay in both PANC-1 and MiaPaCa cells, pre-exposure to hypoxia/TIFM completely blunted this effect (Fig. 1C). This striking effect of TIFM with hypoxia led us to ask how this protection against ferroptosis is mediated.

Tumor interstitial fluid medium induces a hypoxia-like gene response

To understand the mechanism by which TIFM enhances the protective effect of hypoxia, we used RNA-sequencing analysis of PANC-1 and MiaPaCa cells exposed to regular media or TIFM. Surprisingly, the gene set with strongest induction by TIFM was the hypoxic gene set itself (even though these cells were cultured under normoxic conditions) (Fig. 2A–C and Fig. S1A). Western blot analysis showed a mild increase in expression of HIF-1α and HIF-2α levels by TIFM when combined with hypoxia (Fig. 2D). qRT-PCR analysis looking at RNA expression of two classical HIF target genes using previously described primers¹⁷, the glucose transporter Glut1¹⁸ and the pro-angiogenic factor VEGF-A¹⁹, confirmed TIFM alone could induce certain hypoxia-responsive genes and had a synergistic effect when combined with hypoxia (Fig. 2E, Fig. S1B). Further, analysis of VEGF levels using a mouse VEGF ELISA assay showed that TIFM had a similar effect as hypoxia on VEGF levels in a mouse PDAC line 6694 (Fig. 2F). We conclude that TIFM can induce a hypoxia-like gene response. This observation led us to hypothesize that the protective effect of TIFM with hypoxia against ferroptosis is mediated by induction of HIF activity.

Figure 2: Pancreatic Tumor Interstitial Fluid Media Induces a Hypoxic-Like Gene Response.

PANC-1 cells were cultured in regular media or exposed to TIFM for 48 hours. RNA was isolated and analyzed by RNA-sequencing.

(A) GSEA analysis is shown;

(B) the enrichment plot for the hypoxic gene set is displayed;

(C) a heat map of the indicated genes within the hypoxic gene set is presented. PANC-1 cells were exposed to regular media or TIFM for 48 hours.

(D) PANC-1 cells were exposed to combinations of hypoxia (.5% O₂) and TIFM as indicated. Protein lysates were collected and analyzed by western blot using the indicated antibodies;

(E) RNA was isolated, and the expression levels of VEGF and Glut1 mRNA were determined by qRT-PCR;

(F) 6694 cells were exposed to regular media or TIFM under normoxia, or regular media under hypoxia, for 48 hours. Culture supernatants were collected, and VEGF levels were measured by ELISA.

Data are presented as mean ± SEM of 3–5 biological replicates.

The protective effect of hypoxia on ferroptosis is mediated by the HIF-2 transcription factor

We considered the possibilities that the effect of hypoxia on ferroptosis susceptibility are mediated by HIF-1, HIF-2, both, or neither. We used CRISPR guides to target HIF-1α, HIF-2α, or the shared β subunit HIF-1β/ARNT (which would eliminate activity of both transcription factors) in PANC-1 cells stably expressing Cas9. We typically achieve 80–90% lentiviral infection efficiency of guide RNAs in this cell type. The results reveal that knockout of either HIF-2α or its binding partner ARNT largely eliminated the protective response of hypoxia/TIFM on erastin-induced ferroptosis, but the protective effect was completely preserved in HIF-1α knockout cells (Fig. 3A). To further confirm these findings, we generated several PANC-1 clones with knockout of either HIF-1α or HIF-2α (Fig. 3B). In all HIF-1α knockout clones the protective effect of hypoxia/TIFM was preserved, whereas the knockout of HIF-2α completely abolished this effect (Fig. 3C). We further confirmed rescue of erastin-mediated cell death in HIF-2α knockout cells by the ferroptosis inhibitor ferrostatin (Fig. 3D).

Figure 3: The HIF-2 Transcription Factor Mediates the Protective Effect of Hypoxia on Ferroptosis.

(A) PANC-1-Cas9 expressing cells were infected with a virus encoding guide RNAs targeting ARNT/HIF-1β, HIF-1α, or HIF-2α. Cells were exposed to regular media under normoxia for 24 hours or TIFM under hypoxia for 24 hours, followed by treatment with the indicated concentrations of erastin. Cell viability was measured 72 hours after treatment.

(B) PANC-1 HIF-1α or HIF-2α knockout (KO) clones were exposed to hypoxia for 24 hours, and lysates were subjected to Western blot analysis.

(C) PANC-1 clones, either wild-type, HIF-1α KO, or HIF-2α KO were exposed to normoxia with regular media or hypoxia with TIFM for 24 hours, then treated with 10 μM erastin for 72 hours, after which cell viability was measured. The percentage of viability in erastin-treated versus untreated cells is shown.

(D) PANC-1 cells, either wild-type or HIF-2α KO, were exposed to normoxia with regular media or hypoxia with TIFM for 24 hours, then treated with indicated combinations of 10 μM erastin or ferrostatin for 48 hours. The percentage of viability relative to untreated cells is shown.

(E) MiaPaCa cells, either wild-type, HIF-1α KO, or HIF-2α KO, were exposed to hypoxia and then harvested for Western blot analysis.

(F) The indicated MiaPaCa cell lines were exposed to normoxia or hypoxia (1% O₂) and treated with the indicated concentrations of RSL-3 for 48 hours, after which cell viability was measured.

(G) PANC-1 cells were infected with a virus encoding a constitutively active HIF-2α mutant or GFP control, then seeded and exposed to the indicated drugs for 72 hours, after which cell viability was measured.

Data are presented as mean ± SEM of 3–4 biological replicates.

We confirmed these findings in MiaPaCa cells and generated stable clones with knockout of either HIF-1α or HIF-2α (Fig. 3D). The results reveal enhanced susceptibility to ferroptosis induced by RSL-3 even at baseline in the HIF-2α knockout clone, and complete abolishment of the protective effect of hypoxia, whereas there was no difference in baseline susceptibility or the protective effect of hypoxia in the HIF-1α knockout cells (Fig. 3E).

We tested whether these findings are relevant in vivo. Given the importance of the pancreatic tumor metabolic microenvironment implied by TIFM, these studies ideally would have been done in the autochthonous setting. However, HIF-2α knockout in autochthonous models of pancreatic cancer completely block tumor development⁴³, so this approach was not possible. Therefore, we implanted PANC-1 wild type or HIF-2α knockout cells into NSG (NOD-SCID gamma) mice subcutaneously and used the drug imidazole ketone erastin (IKE) intraperitoneally²⁰. At a low dose of 25 mg/kg injected three times weekly (MWF), we observed no effect on control tumors, but a trend towards impaired growth in HIF-2α knockout cells (p-value .08) (Fig. S2).

To test whether HIF-2 activity is sufficient to protect against ferroptosis, we infected PANC-1 cells with lentivirus encoding a HIF-2α mutant with both hydroxylation sites mutated, so that the protein is stable under normoxic conditions. Lentiviral infection of PANC-1 cells with the HIF-2 overexpression construct led to modest resistance to erastin or a second system Xc- inhibitor sulfasalazine (Fig. 3G), an effect that was substantially less than the effect of hypoxia/TIFM. In contrast, we generated a PANC-1 line with doxycycline inducible expression of HIF-1α hydroxylation-defective mutant (Fig. S3A), and found induction of HIF-1 did not confer protection to either erastin or sulfasalazine (Fig. S3B). We further used pharmacological inducers of HIF-2—Roxadustat, a prolyl hydroxylase inhibitor, and dimethyloxalylglycine (DMOG), an α-ketoglutarate antagonist that also inhibits prolyl hydroxylases— and found that neither compound conferred protection against ferroptosis (Fig. S3C). Thus, we conclude that HIF-2 is necessary but largely insufficient to protect against ferroptosis.

Pancreatic TIFM can Protect Renal Cell Carcinoma Cells From Ferroptosis.

HIF-2 induced by VHL loss in renal cell carcinoma was previously shown to enhance susceptibility to ferroptosis, rather than protect against ferroptosis³¹. Since chemical induction of HIF-2 was insufficient to protect against ferroptosis in PDAC cells (Fig. S3), we tested whether VHL loss in PDAC cells would be similarly ineffective. We used two different guide RNAs to knockout VHL by CRISPR-Cas9 technology (Fig. S4A) and showed that knockout of VHL in either PANC-1 or AsPC-1 cells offered no protection against ferroptosis (Fig. S4B). (MiaPaCa cells with VHL deleted were not viable.)

Given that the combination of hypoxia/pancreatic TIFM protected against ferroptosis while suppressing pro-ferroptotic genes such as HILPDA (Table S1), we tested whether pancreatic TIFM could protect renal cell carcinoma cells against ferroptosis. RCC4 cells, a renal cell carcinoma line with VHL loss, were not viable in pancreatic TIFM, but re-introduction of VHL (RCC4-VHL cells) allowed them to survive (Fig. S4C). RCC4-VHL cells were sensitive to erastin with exposure to hypoxia and pancreatic TIFM offering a protective response similar to PDAC lines (Fig. S4D). Overall, our results reveal that the metabolic microenvironment can reshape ferroptosis sensitivity, with pancreatic TIFM protecting against ferroptosis even in highly sensitive renal cell carcinoma cells.

HIF-2 induces expression of transsulfuration enzymes CBS and CTH

To determine possible effector genes by which HIF-2 protects against ferroptosis, we performed RNA-sequencing of PANC-1 cells, a HIF-1α KO clone, or two HIF-2α KO clones with different guides in response to normoxia and regular media, hypoxia and regular media, and hypoxia with TIFM. In our RNA-sequencing data, we sought genes induced significantly by both hypoxia and hypoxia/TIFM, with hypoxic induction preserved in the HIF-1α KO line and completely lost in both HIF-2α clones. We looked for overlap with the ferroptosis gene set, a list of a few dozen genes encoding key proteins implicated in ferroptosis. With these rigid criteria, three anti-ferroptotic genes were identified: CBS, GCLC, and Steap3. In contrast, there were no genes induced by hypoxia in a HIF-1 dependent manner in the ferroptosis gene set. We confirmed that 24 hours of hypoxic exposure resulted in the induction of the three anti-ferroptotic RNAs, however, only CBS had a consistent induction at the protein level in PANC-1 or MiaPaCa cells. Hence, GCLC or Steap3 were not analyzed further. We further generated two ARNT knockout PANC-1 clones (Fig. S5A) and confirmed that ARNT knockout impaired hypoxia-mediated protection against ferroptosis induced by RSL3 (Fig. S5B) like erastin (Fig. 3A). RNA-sequencing of PANC-1 cells and both ARNT KO clones was performed under the same three conditions (normoxia and regular media, hypoxia and regular media, and hypoxia with TIFM). We found that CBS induction was also impaired by ARNT KO in our RNA-sequencing results (Table S3).

CBS and CTH are rate-limiting enzymes involved in the transsulfuration pathway, which allows intracellular synthesis of cysteine from methionine stores and ultimately synthesis of glutathione. The two differ in tissue expression with CBS the predominant protein in pancreas²¹. Induction of several other genes involved in glutathione synthesis (such as GCLC, GCLM, SLC7A11, SLC3A2) were also impaired or partially impaired in HIF-2α or ARNT knockout cells, so we tested whether glutathione supplementation could rescue the effect of HIF-2 loss. Indeed, glutathione supplementation rescued cell survival when HIF-2α knockout cells were treated with either erastin or RSL-3 (Fig. 4A). Thus, we hypothesized that HIF-2 maintains ferroptosis resistance by sustaining the genetic program required for glutathione synthesis

Figure 4: HIF-2 Promotes Expression of the Transulfuration Pathway Enzymes CBS and CTH.

(A) PANC-1 cells, either wild-type or HIF-2α KO, were exposed to indicated combinations of regular media, TIFM, or hypoxia for 24 hours and/or GSH supplementation for 8 hours, then treated with indicated combinations of erastin or RSL-3 for 48 hours, after which cell viability was measured.

(B) PANC-1 cells were exposed to the indicated treatment conditions for 48–72 hours. RNA was then isolated and subjected to qRT-PCR analysis of CBS or CTH as indicated, relative to 18S rRNA control. Results are normalized to the normoxic condition with regular media.

(C) PANC-1 or MiaPaCa cells were exposed to combinations of hypoxia and/or TIFM for 48 hours, after which lysates were subjected to Western blot analysis with the specified antibodies.

(D) MiaPaCa cells were infected with a virus encoding GFP, CBS, or CTH. After infection, cells were seeded and left untreated or treated with erastin, sulfasalazine, or RSL-3. Seventy-two hours post-treatment, cell viability was measured using the CellTiter-Glo assay and normalized to the untreated control.

(E) PANC-1 control cells or HIF-2α knockout (KO) clones were exposed to either regular media/normoxia or TIFM/hypoxia for 48 hours. RNA was then isolated and subjected to qRT-PCR analysis. Fold induction relative to the normoxic control for each cell line is plotted.

(F) PANC-1 cells that were wild-type or HIF-2α knockout (left) or ARNT knockout (right) were exposed to regular media/normoxia or TIFM/hypoxia for 48 hours. Cells were then lysed and lysates were subjected to Western blot analysis with the specified antibodies.

(G) PANC-1 cells with stable expression of doxycycline inducible HIF-2α mutant (P405A/P531A) were left untreated or treated with doxycycline for 24 hours, then exposed to hypoxia for an additional 24 hours. Lysates were then subjected to Western blot analysis with the specified antibodies. Separately, PANC-1 cells were transfected with a plasmid encoding a CBS promoter upstream of firefly luciferase or a control reporter with Renilla luciferase. 12 hours post-transfection cells were left untreated or treated with doxycycline in either regular media or TIFM for additional 48 hours. Luciferase activity was measured, and the ratio of firefly to Renilla luciferase was calculated.

(H) PANC-1 cells stably expressing inducible shRNA against HIF-2α were either left untreated or treated with doxycycline for 24 hours, then exposed to hypoxia for an additional 24 hours. Lysates were then subjected to Western blot analysis with the specified antibodies. Separately, PANC-1 cells were transfected with a plasmid encoding a CBS promoter upstream of firefly luciferase or a control reporter with Renilla luciferase. After 24 hours, cells were left untreated or treated with doxycycline and exposed to the indicated combinations of TIFM and hypoxia for an additional 48 hours. Luciferase activity was measured, and the ratio of firefly to Renilla luciferase was calculated.

(I) MiaPaCa control cells or HIF-2α KO clones were exposed to hypoxia for 24 hours, after which lysates were subjected to Western blot analysis with the specified antibodies.

(J) MiaPaCa wild-type or HIF-2α knockout cells were transfected with a plasmid encoding a CBS promoter upstream of firefly luciferase or a control reporter with Renilla luciferase. After 24 hours, cells were left untreated or treated with doxycycline and exposed to the indicated combinations of TIFM and hypoxia for an additional 48 hours. Luciferase activity was measured, and the ratio of firefly to Renilla luciferase was calculated.

(K) MiaPaCa wild-type or HIF-2α knockout cells were exposed to hypoxia for 24 hours. RNA was then isolated and subjected to qRT-PCR analysis.

Data are presented as mean ± SEM of 3 biological replicates.

Using qRT-PCR in PANC-1 cells, we confirmed induction of CBS and CTH mRNAs by hypoxia and/or TIFM (with separate samples than those used for RNA-sequencing) (Fig. 4B). Further, we found induction of both CBS and CTH at the protein level in both PANC-1 and MiaPaCa lines after treatment with hypoxia/TIFM (Fig. 4C). Importantly, ectopic lentiviral mediated expression of either CBS or CTH protected MiaPaCa cells from ferroptotic death induced by erastin, sulfasalazine, or RSL-3 (Fig. 4D).

To verify these enzymes are HIF-2 inducible, we examined induction of CBS or CTH mRNA in control PANC-1 cells or two HIF-2α KO clones in hypoxia/TIFM vs normoxia/regular media. HIF-2 knockout substantially blunted the induction of either gene (Fig. 4E) at the RNA level. Similarly, knockout of HIF-2α or ARNT impairs induction of either CBS or CTH at the protein level (Fig. 4F).

Notably, the CBS promoter has three potential hypoxia response elements (HREs) in close proximity roughly 800 bp upstream of the transcriptional start site. We used a previously generated luciferase reporter²²with the CBS promoter upstream of firefly luciferase and looked at ratios with a Renilla luciferase reporter as a transfection control. We generated a PANC-1 line with doxycycline-inducible expression of a mutant HIF-2α construct with both hydroxylation sites mutated, rendering the protein oxygen-stable (Fig. 4G). In a CBS promoter assay, induction of HIF-2α and exposure to TIFM media synergistically induced promoter activity (Fig. 4G). We also generated a PANC-1 line with inducible shRNA against HIF-2α (Fig. 4H) and found that HIF-2α knockdown abolished reporter activity induced by either TIFM, hypoxia, or the combination of TIFM and hypoxia (Fig. 4H). These observations indicate that hypoxia and TIFM synergistically induce HIF-2 activity to upregulate CBS.

To extend these findings to a second PDAC line, we generated two HIF-2α knockout clones in MiaPaCa cells (Fig. 4I). We found that HIF-2α knockout abolished CBS promoter reporter activity in MiaPaCa cells (Fig. 4J) and induction of either CBS or CTH mRNA levels measured by qRT-PCR (Fig. 4K). We conclude that CBS and CTH, whose overexpression protects against ferroptosis, are induced in a HIF-2 dependent fashion in PDAC lines.

HIF-2 induces expression of System Xc-

Generation of glutathione also involves the import of extracellular cystine. System Xc- is a cystine transporter composed of two subunits, SLC7A11 and SLC3A2. Our RNA-sequencing indicated expression of both genes in response to hypoxia and/or TIFM in either PANC-1 or MiaPaCa cell lines. We confirmed induction of SLC7A11 and SLC3A2 by qRT-PCR analysis in separate samples from PANC-1 cells (Fig. 5A). Induction of both subunits at the protein level in response to hypoxia + TIFM was confirmed through Western blot analysis (Fig. 5B). We used a PANC-1 clone with doxycycline inducible expression of oxygen-stable mutant HIF-2α (Fig. 4G) and showed that induction of HIF-2 activity, in combination with TIFM, induced expression of both SLC7A11 and SLC3A2 by western blot analysis (Fig. 5C).

Figure 5: HIF-2 Promotes the Expression of system Xc Components SLC7A11 and SLC3A2.

(A) PANC-1 cells were exposed to indicated treatment conditions for 48–72 hours, after which RNA was isolated and subject to qRT-PCR analysis of indicated target genes.

(B) The indicated cells were exposed to combinations of hypoxia and/or TIFM for 48 hours, after which lysates were subject to Western blot with indicated antibodies.

(C) PANC-1 cells with doxycycline-inducible oxygen-stable mutant HIF-2α (from Fig. 4F) were exposed to TIFM for 24 hours, then exposed to doxycycline in combination with regular media or TIFM as indicated for additional 48 hours, after which cells were lysed and lysates subject to western blot analysis with indicated antibodies.

(D-E) MiaPaCa cells (D) or PANC-1 cells (E) or that were control, HIF-2α KO, or HIF-1α KO were exposed to normoxia, hypoxia with regular media, or hypoxia with TIFM as indicated for 48 hours. RNA was then isolated and subject to qRT-PCR analysis of indicated target genes.

(F) PANC-1 cells that were control or knockouts for ARNT, HIF-2α, or HIF-1α were exposed to normoxia or hypoxia/TIFM for 48 hours, after which lysates were collected and subject to Western blot with indicated antibodies.

(G) Single cell sequencing datasets from pancreatic cancer, colon cancer, and lung cancer were analyzed. Correlation between expression levels of the indicated target genes and a previously published hypoxic gene score are plotted.

Data are presented as mean ± SEM of 3 biological replicates.

We showed induction of SLC3A2 or SLC7A11 in response to hypoxia + TIFM was blunted in two HIF-2α KO clones in the MiaPaCa line (Fig. 5D). SLC7A11 was previously identified as a HIF-1 target gene in breast cancer stem cells²³. We found in PANC-1 cells that induction of either SLC7A11 or SLC3A2 was abolished in two HIF-2α KO clones (with different guide RNAs) but preserved in HIF-1α knockout cells (Fig. 5E). Similarly, induction by hypoxia/TIFM of SLC7A11 or SLC3A2 at the protein level was impaired in HIF-2α or ARNT knockout cells but preserved in HIF-1α knockout cells (Fig. 5F).

To test whether the effects we observed in cell culture are also relevant in human PDAC samples, we asked whether the expression of anti-ferroptotic genes was correlated with hypoxia gene expression. We analyzed previously published single-cell datasets from human PDAC samples²⁴ with a previously described hypoxia gene expression score²⁵. Results show that induction of CBS, CTH, SLC7A11, and SLC3A2 were all correlated with hypoxic gene expression at the single cell level (Fig. 5G). To determine whether these findings might extend to other solid tumors, we performed similar analysis with single cell sequencing datasets from colon cancer and lung cancer and found that hypoxic gene expression correlates with expression of these anti-ferroptotic genes in colon²⁶ and lung²⁷ cancer (Fig. 5G). We conclude that HIF-2 induces expression of system Xc- in PDAC cells in vitro and provide evidence that these anti-ferroptotic genes are correlated with hypoxia in human PDAC samples and potentially other solid cancers.

HIF-2 Suppresses Mitochondrial Function and Oxidative Phosphorylation

Whereas production of glutathione protects against ferroptosis, mitochondrial production of reactive oxygen species contributes to lipid peroxidation and promotes ferroptosis. In this respect, we determined whether genes involved in mitochondrial function are altered by hypoxia/TIFM and dependent on HIF-2 activity. GSEA analysis showed that processes related to mitochondrial function were downregulated by hypoxia/TIFM in a HIF-2 dependent fashion (Fig. 6A). Heat maps of genes with functions in oxidative phosphorylation show coordinated downregulation of these genes at the RNA level, which was largely abolished by HIF-2α knockout (Fig. 6B). Western blot analysis in three different PDAC lines showed that exposure to hypoxia/TIFM led to a decrease in levels of mitochondrial proteins SDHA (complex II of the electron transport chain), COXIV (complex IV subunit), PHB1 (a mitochondrial inner-membrane scaffolding protein involved in cristae integrity and respiratory chain assembly), and cytochrome c (an intermembrane-space electron carrier between complexes III and IV) (Fig. 6C). As another rough estimate of mitochondrial content, we isolated DNA from cells in regular medium vs 72 hours of hypoxia/TIFM and looked at copy number ratios of mitochondrial-encoded cytochrome c oxidase III (MT-CO3) versus a nuclear encoded gene thymidine kinase 2 (TK2).

Figure 6: HIF-2 Mediates Downregulation of Genes Involved in Mitochondrial Function.

(A) PANC-1 cells - control, HIF-1α KO, or HIF-2α KO were cultured in regular media and normoxia or exposed to TIFM and hypoxia for 48 hours. RNA was then isolated and sent for bulk RNA-sequencing.

(B) Heat map showing average change in expression levels upon treatment with hypoxia and TIFM compared to normoxia and regular media for the control, HIF-1α KO, or HIF-2α KO line is shown. Indicated target genes are grouped by their role in the electron transport chain. Hypoxia/TIFM condition is shown on left, normoxia/regular media (i.e. control) is shown on right

(C) PANC-1, MiaPaCa, or AsPC-1 cells were exposed to regular media/normoxia or hypoxia/TIFM for 48–72 hours, after which cells were lysed in RIPA buffer and lysates subject to Western blot with indicated antibodies.

(D) PANC-1 cells that were control, or either of two HIF-2α clones were exposed to TIFM/hypoxia for 72 hours or to regular media/normoxia. DNA was then isolated and subject to qPCR against a mitochondrial gene, CytoC3, and nuclear gene, TK2. The fold-change in mitochondrial/nuclear DNA ratios under hypoxia/TIFM vs normoxia/regular condition are plotted. * p < .05 by student’s T-test. Data are presented as mean ± SEM of 3 biological replicates.

(E) PANC-1 cells that were control or either of two HIF-2α clones were exposed to TIFM/hypoxia for 72 hours, after which cells were pelleted and stained with MitoSox-Red then analyzed by flow cytometry.

(F) PANC-1 cells were treated with regular media under normoxia or TIFM and hypoxia for 72 hours. Cells were then seeded and 2 hours after seeding analyzed by Seahorse analysis. Phase I – basal; II – with oligomycin; III – with FCCP; IV – with rotenone and antimycin.

(G) PANC-1 cells with doxycycline-inducible oxygen-stable mutant HIF-2α (from Fig. 4F) were exposed to TIFM for 24 hours, then either untreated or treated with doxycycline for additional 48 hours. Cells were then seeded and analyzed by Seahorse analysis. Data are presented as mean ± SD of 6–7 replicates.

Results reveal that hypoxia/TIFM led to a decrease in mitochondrial DNA content in control PANC-1 cells which was largely abolished in either HIF-2α knockout clone (Fig. 6D). We also used MitoSox to look at mitochondrial ROS in control PANC-1 cells compared to either HIF-2α clone and found that HIF-2 knockout was associated with higher levels of mitochondrial ROS under hypoxia/TIFM (Fig. 6E). Further, we examined mitochondrial function with Seahorse assay. PANC-1 cells were kept in regular media with normoxia or TIFM with hypoxia for 72 hours, then analyzed by Seahorse assay. The results indicate a decrease in baseline oxygen consumption (phase I) and maximal mitochondrial respiration (phase III) in TIFM/hypoxia vs regular media/normoxia (Fig. 6F). These observations are consistent with decreased mitochondrial mass and a decrease in the ratio of oxygen consumption rate over extracellular acidification rate (OCR/ECAR) indicating a switch away from mitochondrial respiration with TIFM/hypoxia (Fig. 6F).

Having observed that HIF-2 is necessary for the reduction of mitochondrial mass and function, we then tested whether ectopic expression of inducible HIF-2 would conversely be sufficient to reduce mitochondrial function. Consistent with the HIF-2 loss-of-function findings, the induction of HIF-2α in a PANC-1 line cultured under TIFM media was associated with decreased mitochondrial respiration, compensatory increased glycolysis, and decreased OCR/ECAR ratio (Fig. 6G). We conclude that HIF-2 inhibits mitochondrial abundance and function in PDAC cells.

HIF-2 Induces the mitophagy regulator Parkin

Because HIF-2 alters mitochondrial function in PDAC cells, we sought to determine whether HIF-2 affects mitophagy. We note from RNA sequencing (Table S1) that the mitophagy regulator Parkin mRNA was induced by TIFM in PANC-1 and MiaPaCa lines. Further, we found across a panel of human and mouse PDAC lines that the combination of hypoxia and TIFM had a synergistic effect of inducing Parkin at the protein level (Fig. 7A). Notably, Parkin has not been documented as a HIF target gene. To confirm that HIF activity was responsible, we used chemical induction of HIF with the iron chelator desferrioxamine (DFX) or the hydroxylase inhibitor DMOG and showed chemical induction of HIF had similar effect on Parkin levels as hypoxia in both a human PDAC line (PANC-1) and mouse PDAC line (6694) (Fig. 7B).

Figure 7: HIF-2 Promotes the Expression of Parkin.

(A) A panel of PDAC cell lines was exposed to hypoxia and/or TIFM for 48 hours, as indicated, and then subjected to Western blot analysis with the specified proteins.

(B) PANC-1 or 6694 cells were exposed to hypoxia, DFX (100 μM), or DMOG (0.5 mM) for 48 hours. Lysates were then subjected to Western blot analysis with the specified proteins.

(C-D) PANC-1 control cells or ARNT KO cells were exposed to normoxia or hypoxia/TIFM for 48 hours, after which either RNA was isolated (C) and subject to qRT-PCR, or lysates were obtained and subject to Western blot (D) with indicated antibodies.

(E) Wild-type MiaPaCa cells or two HIF-2α KO clones were exposed to normoxia, hypoxia with regular media, or hypoxia with TIFM for 24 hours. RNA was isolated and subjected to qRT-PCR analysis with the specified genes. Fold-change relative to wild-type cells under normoxic conditions are shown.

(F) PANC-1 cells—either wild-type, HIF-1α knockout (KO), or HIF-2α KO—were exposed to normoxia or hypoxia for 24 hours. RNA was isolated and subjected to qRT-PCR analysis, and fold-change relative to normoxia is shown.

(G) PANC-1 control cells or HIF-2α KO cells were exposed to normoxia or hypoxia/TIFM for 48 hours, after which lysates were collected and subject to Western blot with indicated antibodies.

(H) PANC-1 cells were exposed to hypoxia in either regular media or TIFM for 48 hours. Cells were collected and subject to chromatin immunoprecipitation with anti-HIF-2α or ARNT antibody and qPCR of primers spanning putative HRE in Parkin promoter was performed.

(I) PANC-1-Cas9 cells were infected with guide RNA targeting VHL and single cell clone isolated. Lysates were collected and subjected to Western blot analysis with the specified proteins.

(J) PANC-1 VHL knockout cells were collected and subject to chromatin immunoprecipiration with indicated antibodies, after which qPCR of Parkin promoter was performed.

(K) PANC-1 or MiaPaCa cells were infected with lentivirus expressing Parkin or an empty vector control. 48 hours post-infection, cells were seeded and next day treated with the indicated drugs. 72 hours after treatment, cell viability was measured using the CellTiter-Glo assay.

(L) HIF-2α knockout PANC-1 cells were infected with lentivirus expressing Parkin or empty vector control. Cells were then seeded and exposed to hypoxia/TIFM as indicated for 48 hours, then treated with erastin for additional 48 hours after which cell viability was measured.

(M) PANC-1 or MiaPaCa cells were infected with lentivirus expressing Parkin that was wild-type or mitophagy-defective S65A mutant. Infected cells were seeded and 48 hours later treated with erastin for additional 48 hours after which cell viability was measured.

(N) PANC-1 cells were infected with vector encoding FLAG-tagged HIF-2α mutant plasmid, and 24 hours post-infection media was replaced with TIFM for 48 hours or fresh regular media (DMEM + 10% FBS). Cells were collected and a fraction taken for input, with remainder of cells subject to chromatin immunoprecipitation with anti-HIF-2α or anti-FLAG antibody, followed by qPCR using primers spanning putative HREs in indicated gene promoters.

Data are presented as mean ± SEM of 3–4 replicates.

Parkin induction at both the protein (Fig. 7C) and RNA (Fig. 7D) level were abolished by ARNT knockout. Similarly, induction of Parkin mRNA by exposure to hypoxia or hypoxia/TIFM was abolished in two HIF-2α KO clones in MIaPaCa cells (Fig. 7E). Similar results were obtained in PANC-1 cells at both the RNA (Fig. 7F) and protein levels (Fig. 7G), but hypoxic induction of Parkin was preserved in HIF-1α knockout cells. Analysis of the Parkin promoter with the bioinformatics tool MotifMap revealed a high-probability HRE at the −9 position. We performed ChIP of PANC-1 cells exposed to hypoxia in regular media or TIFM and confirmed binding of HIF-2α and its partner ARNT to this site which was enhanced by TIFM (Fig. 7H).

We also induced HIF activity with CRISPR knockout of its main negative regulator VHL and showed VHL knockout was sufficient to induce HIF-1α and HIF-2α protein and associated with higher Parkin levels (Fig. 7I). Further, we found that chIP in PANC-1-VHL knockout cells had detectable binding of both HIF-2α and ARNT to this site but no binding of HIF-1α was observed (Fig. 7J). We conclude that HIF-2 is a specific regulator of Parkin expression.

To test whether Parkin protects against ferroptosis in these cells, we infected PANC-1 or MiaPaCa cells with lentivirus encoding Parkin and showed Parkin overexpression in either cell line protected against erastin- or sulfasalazine- induced cell death but had no protection against RSL-3 (Fig. 7K). This is consistent with previous reports that mitochondrial function regulates sensitivity to ferroptosis induced by glutathione depletion but has no effect on GPX4 inhibitors^8,9. We further found that lentiviral Parkin overexpression protected HIF-2α knockout cells from erastin-induced ferroptosis with a synergistic effect when combined with hypoxia/TIFM (Fig. 7L). Finally, we used a previously-described²⁸ Parkin S65A mutant which is unable to induce mitophagy and found that the mitophagy-defective Parkin mutant was significantly impaired in its ability to protect against ferroptosis in both PANC-1 and MiaPaCa cells (Fig. 7M). Thus, our data reveal a previously unappreciated HIF-2–Parkin axis that links hypoxic signaling to mitophagy-dependent protection from ferroptosis

Belzutifan is Ineffective in Inhibiting HIF-2 in PANC-1 cells

Belzutifan is a HIF-2 inhibitor FDA-approved for treatment of renal cell carcinoma, which blocks HIF-2α/ARNT heterodimerization²⁹. We tested whether Belzutifan would block induction of anti-ferroptotic target genes. Previous studies have shown that Belzutifan inhibits HIF-2 in renal cell carcinoma cells at a dose of less than 1 uM. Surprisingly, there was no effect on induction of anti-ferroptotic genes induced by hypoxia/TIFM at a high dose of 10 uM Belzutifan at either the RNA (Fig. S6A) or protein (Fig. S6B) level. Predictably, Belzutifan also had no effect on hypoxia-mediated protection against RSL3 (Fig. S6C). However, we could not detect inhibition of HIF-2 activity by Belzutifan at a bona fide HIF-2 target gene VEGF (Fig. S6B). This suggests that pancreatic cancer cells can export Belzutifan and may explain previous reports that Belzutifan inhibits pancreatic cancer progression largely through effects on the pancreatic stroma³⁰.

Pancreatic TIFM Enhances HIF-2 Binding to Anti-Ferroptotic Target Gene Promoters

To determine whether pancreatic TIFM has differential effects on HIF-1 or HIF-2 specific target genes, we generated a list of potential HIF-1 and HIF-2 target genes in PANC-1 cells by looking at genes with significant induction with hypoxia alone in control cells, but induction lost in either HIF-1α knockout cells or one of the HIF-2α knockout clones (Table S2). We then looked at PANC-1 cells treated with TIFM alone (Table S1) and found that TIFM preferentially induced HIF-2 target genes while inhibiting HIF-1 target genes. Bubble plots showing the degree of induction or repression for relevant target genes by TIFM or hypoxia (Fig. S7–S8). Thus, TIFM induces a hypoxic gene response generally and promotes a HIF-2 dominant response specifically.

We hypothesized that TIFM enhances HIF-2 binding at specific target gene promoters. SLC3A2, SLC7A11, and Parkin have not previously been described as HIF-2 target genes. An analysis of the SLC3A2 promoter using MotifMap showed a high-probability HRE present 42 bp upstream of the transcriptional start site and a high-probability HRE 9 bp upstream of the Parkin transcriptional start site. A manual search showed five potential HRE sites within a 200 bp stretch downstream of the SLC7A11 promoter. We generated primers spanning these putative HREs. We infected PANC-1 cells with lentivirus encoding a mutant HIF-2α with a FLAG tag and hydroxylation sites mutated so that it would be constitutively expressed. Cells were then exposed to regular media or pancreatic TIFM and subject to chromatin immunoprecipitation (ChIP) with both a FLAG tag antibody or HIF-2α antibody and qPCR was performed. The results confirm HIF-2 binds the HREs of CBS, SLC3A2, SLC7A11, and Parkin with enhanced binding in pancreatic TIFM (Fig. 7N). Another group previously reported binding of HIF-1 to inter-intronic HREs of the SLC7A11 gene²³. However, we could not detect binding of HIF-2 using those previously published primers, indicating that HIF-1 and HIF-2 use distinct binding sites to induce SLC7A11 in a context-dependent fashion.

Discussion

Pancreatic adenocarcinoma, which is largely driven by mutant KRAS, is poorly responsive to both chemotherapy and immunotherapy due to the hypoxic desmoplastic tumor microenvironment. The development of mutant KRAS specific drugs holds tremendous hope, but the emergence of drug resistance remains a challenge. As a result, there remains great interest in finding targetable vulnerabilities. Ferroptosis has attracted interest as a potential therapeutic approach, as there have since been agents discovered that can induce ferroptosis in vivo and cancer cells that persist after treatment with chemotherapeutic agents retain sensitivity to ferroptosis. In this study we show that hypoxic pancreatic tumor interstitial fluid provides robust protection against ferroptosis in a HIF-2 dependent manner, providing insights into potential novel therapeutic strategies.

HIF-2 protects against ferroptosis through multiple effectors (Graphical Abstract). A key mechanism is through enhancing glutathione synthesis. HIF-2 induced both components of the system Xc- cystine transporter – SLC7A11 and SLC3A2. Intracellular cystine is converted to cysteine and ultimately glutathione. Further, HIF-2 induced CBS and CTH, both key enzymes involved in the transsulfuration pathway which allows intracellular synthesis of cysteine from methionine. In addition to their role upstream of glutathione synthesis, CBS and CTH are important as producers of hydrogen sulfide gas and persulfides^32,33 which can scavenge free radicals and thereby inhibit ferroptosis. Hydrogen sulfide has also been identified as signaling molecule through regulation of key cysteine residues.³⁴ Most studies have indicated H2S protects against ferroptosis although this may be context-dependent. For example, H2S has been noted to protect against ferroptosis through upregulation of Keap1/Nrf2 signaling³⁵ and through regulation of SLC7A11³⁶, but in non-small cell lung cancer a recent study identified H2S as a pro-ferroptotic molecule³⁷.

A second mechanism of HIF-2 in protecting against ferroptosis is through regulation of mitochondrial function. We found that HIF-2 reduces expression of oxidative phosphorylation genes and reduce mitochondrial DNA which tracks with diminished mitochondrial ROS. We further identify the key mitophagy regulator Parkin as a HIF-2 target gene and found in PDAC lines that Parkin overexpression is protective against ferroptosis induced by system Xc- inhibition (i.e., erastin and sulfasalazine). Parkin had no effect on RSL3-mediated ferroptosis which is consistent with previous findings that mitochondrial function is selectively related to cystine-deprivation or erastin induced ferroptosis, but not ferroptosis related to GPX4 inhibition³⁸. Parkin is mutated in familial forms of Parkinson’s disease³⁹ (from which the gene name is derived) and given that ferroptotic cell death has been observed in the brains of Parkinson’s disease patients and in animal models of Parkinson’s disease⁴⁰, it is plausible that Parkin might confer protection against ferroptosis in the substantia nigra. The HIF-1 transcription factor has previously been identified as a regulator of the mitophagy regulator BNIP3^41,42. Our RNA-sequencing data confirm that HIF-1 is a specific regulator of BNIP3, whereas HIF-2 is a specific regulator of Parkin. Thus, both transcription factors can induce mitophagy through distinct effectors, with the relevant importance of each cell-type or tissue-specific.

It is likely that additional HIF-2 target genes are involved in mediating protection against ferroptosis. RNA seq analysis suggested other potential HIF-2 target genes with known roles in ferroptosis. These include glutamate-cysteine ligase catalytic subunit (GCLC) which catalyzes the rate-limiting step in glutathione synthesis and Steap3, a metalloreductase involved in iron regulation which has been shown to protect against ferroptosis in renal cell carcinoma⁴³. Both GCLC and Steap3 had high probability hypoxia response elements in their promoters. However, in our experiments, cells were generally exposed to hypoxia for 24 hours before drug treatment and on this short time scale we could not confirm induction of these genes at the protein level. Nonetheless, it is possible that HIF-2 regulation of these genes during exposure to hypoxia on a longer time scale or to more extreme levels of hypoxia would be significant. Further, while this paper was under revision a report was published indicating O-GlcNAcylation is protective against ferroptosis in hepatocellular carcinoma⁴⁴, and the gene GFPT2 - which is directly upstream of O-GlcNAcylation as the rate-limiting step in the hexosamine biosynthetic pathway - is induced by hypoxia/TIFM in a HIF-2 dependent manner. Thus we believe HIF-2 acts as a master regulator of multiple anti-ferroptotic mechanisms in pancreatic cancer.

We note this is an active field of research and there are likely many anti-ferroptotic proteins yet to be identified. Finally, we note that hypoxia and TIFM led to broad changes in many genes related to lipid metabolism that are relevant for ferroptosis, and these changes were mainly HIF-2 independent. This may explain why HIF-2 was necessary for the protective effect of hypoxia/TIFM but HIF-2 overexpression alone, although capable of offering mild-moderate protection against ferroptosis, could not recapitulate the full effect. Prior research has shown that the combination of hypoxia and TIFM induced changes in proliferation and susceptibility to chemotherapeutic agents through upregulation of Bcl-xL, an anti-apoptotic protein, indicating that the combination of hypoxia and nutrient deprivation in the tumor microenvironment can protect against multiple cell death pathways⁴⁵.

The protective role of HIF-2 against ferroptosis in PDAC is in contrast to other tumor types. It is documented that across cell lines, renal cell carcinoma (RCC) is especially sensitive to ferroptosis inducers in cell culture. As renal cell carcinoma almost always has inactivating mutations in VHL⁴⁶, these are marked by high HIF activity at baseline. Zou et al showed that in RCC, HIF-2 mediates ferroptosis sensitivity through induction of the pro-ferroptotic protein hypoxia-inducible lipid droplet-associated protein (HILPDA)¹⁴. Consistent with that study, Yang et al reported in the Caki1 RCC line HIF-2 is pro-ferroptotic whereas HIF-1 is protective against ferroptosis by inducing expression of SLC1A1 and indirectly driving cystine import⁴⁷. In our RNA-sequencing data from PANC-1 and MiaPaCa lines, as well as other datasets such as FANTOM5 or GTEx, SLC1A1 is present at exceedingly low levels in both normal pancreas and PDAC cell lines, so the effect of the HIF-1/SLC1A1 axis is negligible in this tissue. A recent study showed in bladder cancer cell lines, hypoxia mediated protection against ferroptosis through repression of KDM6A activity, an oxygen-sensitive histone demethylase⁴⁸. Clearly that mechanism is not at play in the cell lines studied here, as MiaPaCa cells have inactivating mutations in KDM6A and we showed that DMOG, which also inhibits KDM6A, could not protect PANC-1 cells from ferroptosis. It seems that hypoxia can offer protection against ferroptosis through multiple non-overlapping and tissue-specific mechanisms.

Consistent with prior reports, we do find HILPDA is a hypoxia-inducible gene in regular media, but HILPDA expression is suppressed by TIFM (see Table S1 and S2). TIFM appears to modulate HIF activity such that the anti-ferroptotic proteins we identified as HIF-2 targets have gene expression enhanced by TIFM, but at certain other genes (including pro-ferroptotic HILPDA) gene expression is suppressed. Surprisingly, there was also a differential effect where TIFM tended to promote the expression of HIF-2 target genes while inhibiting expression of specific HIF-1 target genes. The overall effect is that pancreatic TIFM modulates the hypoxic landscape to promote resistance to ferroptosis.

The precise mechanism by which TIFM modulates the hypoxic response by increasing HIF-2 DNA binding remains under investigation in our laboratory. The effect of TIFM on gene expression may explain previous observations that PDAC robustly expresses a hypoxic gene expression signature even at the earliest cancer stages, in contrast to other solid tumor types where hypoxic gene expression is more pronounced at advanced stages²⁵. It is likely that the metabolic milieu of other cancers can also modulate HIF expression, such that activity of HIF-2 might be pro- or anti-ferroptotic depending on tumor type.

Our study adds to the body of evidence that HIF-2 is a key player in PDAC development and progression. Data from clinical samples show that HIF-2α expression in PDAC was associated with pathological grade, metastases, and poor patient outcomes⁴⁹. Data from mouse models lends support for this as well. In Ptf1aCre;Kras^G12D mice, HIF-2α deletion blocked the development of high-grade mPanINs through modulation of Wnt-signaling⁵⁰. Further, HIF-2 activity in the stromal population of PDAC promotes infiltration of immunosuppressive regulatory T cells and M2 macrophages⁵¹. Thus HIF-2 activity in different cell types in the tumor environment can promote PDAC growth through multiple synergistic mechanisms. Collectively, these studies support targeting HIF-2 as a therapeutic strategy in PDAC.

Limitations of the Study: Although our findings identify HIF-2 as a central mediator of ferroptosis resistance in PDAC, several limitations should be noted. Most experiments were performed in human PDAC cell lines, and we observed pronounced differences in baseline ferroptosis susceptibility between human and mouse PDAC lines, suggesting that species-specific factors contribute to ferroptotic responses. In addition, because HIF-2α deletion prevents PDAC tumor formation in genetically engineered mouse models, we were unable to assess ferroptosis sensitivity in autochthonous tumors, and in vivo analyses were therefore limited to subcutaneous xenografts, which do not fully recapitulate the metabolic or stromal features of pancreatic cancer. Finally, while we identify several HIF-2–regulated anti-ferroptotic pathways, this list is by no means exhaustive and it is likely that additional mechanisms beyond those detailed in this study contribute to the observed phenotype.

Resource Availability

Lead Contact:

For information about reagents or materials used for this study, please contact the first author Dr. Maimon Hubbi (mhubbi1@jh.edu).

Materials Availability:

Plasmids and unique cell lines generated for this study are available upon reasonable request and completion of a Materials transfer agreement.

Data and Code Availability:

RNA-sequencing data are deposited at Gene Expression Omnibus under accession numbers GSE320200, GSE319951, and GSE319526.

This paper does not report original code.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request

STAR★Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell Lines/Tissue Culture:

All cell lines used in this paper were cultured in DMEM with 10% FBS and 1% penicillin/streptomycin (regular media) or tumor interstitial fluid medium (TIFM) prepared as previously described¹⁵. Hypoxia was induced by exposing cells to 1% O₂ in an InvivO₂ hypoxia workstation (SciPro). Knockout clones were obtained by infecting cells stably overexpressing Cas9 with lentivirus encoding guide RNA against target proteins, then growing single cell clones by limiting dilution. Cell line with inducible shRNA against HIF-2α was generated by infecting PANC-1 cells with lentivirus encoding inducible shRNA then growing single cell clones by limiting dilution. CRISPR sequences and shRNA sequences are listed in Table S5.

Mouse Xenograft Studies:

Female NSG mice were injected subcutaneously in the flank with 1 million PANC-1-Cas9 control cells or PANC-1 HIF-2α knockout cells, and when tumors reached a size of 60–110 mm³ were randomized to vehicle control injection or injection with imidazole ketone erastin (IKE) 25 mg/kg three-times weekly (MWF). Drug was dissolved in 5% DMSO in HBSS at pH 4 as previously described²⁰. 5–7 mice were used per group. Tumor dimensions were measured with calipers prior to injection. Tumor volume was calculated using the formula (length × width²) / 2, where length is the longest tumor diameter and width is the perpendicular diameter measured by calipers. Tumor growth curves were analyzed using a two-way ANOVA with treatment and time as factors. All animal studies were conducted in accordance with institutional guidelines and approved by the appropriate IACUC.

METHOD DETAILS;

Cell Viability Assay:

Cells were seeded at 1000 cells/well in 96-well plate. Next day cells were subject to treatment (hypoxia or TIFM for 24–48 hrs as indicated). Cells were then treated with indicated concentrations of RSL-3, erastin, or sulfasalazine. 48–72 hrs later viability was measured with Cell-Titer Glo (Promega) according to manufacturer’s protocol.

RNA-sequencing and Quantitative Real-Time Reverse Transcriptase-PCR Assay

Cells were lysed by Trizol (ThermoFisher 15596026), extracted by 0.2 volume of chloroform, followed by adding equal volume of 100% ethanol. RNeasy kit (Qiagen) was then used to purify the RNA. 10 μg of purified RNA was then treated with DNAse according to manufacturer protocol (Thermo-Fisher Scientific). This was submitted for RNA-sequencing (Novogene). For RT-PCR analysis, DNAse-treated RNA (1 μg) was used for first-strand synthesis with the iScript cDNA Synthesis system (Biorad). Real-time PCR was performed using IQ SYBR Green Supermix and the iCycler Real-Time PCR Detection System (BioRad). Expression of target mRNA relative to 18S rRNA was calculated based on the threshold cycle (C_T) for amplification as 2-^(ΔCT), where ΔC_T = C_T,target − C_T,18S. RT-PCR primer sequences are provided in Table S5.

Plasmids:

Parkin and HIF-2α (P405A/P531A) were subcloned into PCDH-EF1-FHC viral vector. HIF-2α (P405A/P531A) or HIF-1α (P402A/564A) was also subcloned into pCW57-MCS1-P2A-MCS2 to generate a doxycycline inducible construct. Viral plasmids expressing CBS and CTH were generated by gene synthesis. CRISPR guides against HIF-1α. HIF-2α, HIF-1β/ARNT, and VHL were cloned into LRG lentiviral plasmid. CRISPR guide sequences are provided in Table S5.

Chromatin Immunoprecipitation:

PANC-1 cells were infected with lentivirus encoding Flag-tagged HIF-2α (P405A/P531A). Two days post-infection, cells were exposed to regular media or TIFM for an additional 2 days. Chromatin immunoprecipitation was performed with SimpleChip chromatin IP kit (Cell Signaling Tech) according to manufacturer’s protocol with indicated antibodies.

Mitochondrial DNA Quantitation:

DNA was isolated from cells with PureLink genomic DNA mini kit (Invitrogen) and treated with RNAse according to manufacturer’s protocol. For qPCR analysis, previously described primers⁵³ encoding mitochondrial encoded cytochrome c oxidase (MT-CO3) and nuclear-encoded TK2 were used in combination with IQ SYBR Green Supermix. Quantitation of mitochondrial DNA relative to nuclear DNA was calculated based on the threshold cycle (C_T) for amplification as 2-^(ΔCT), where ΔC_T = C_T,mito − C_T,nuclear.

Luciferase Reporter Assay:

20,000 cells were seeded into 24-well tissue culture plates. The next day cells were transfected with CBS promoter luciferase plasmid and Renilla luciferase plasmid in 10:1 ratio. 24 hours later, cells were treated with indicated conditions including doxycycline treatment when applicable for 48 hours. Cells were then lysed and Firefly and Renilla luciferase activity measured according to manufacturer’s instructions (Promega).

Bioinformatics:

RNAseq reads were aligned to the GRCh39 genome using HISAT2. The reads were counted against standard gene models using featureCounts. Gene wise p-values were calculated using DESeq2. Pathway analysis is performed using the Kolmogorov-Smirnov test for Gene Set Enrichment Analysis method. All heatmaps were plotted using z-scores (gene-wise normalization) of the transcripts per million expressions. Single cell lung, colon, and pancreatic cancer single cell datasets were processed to calculate the cell-wise HIF1A activation as previously published⁵⁴. For each gene, correlation coefficients were calculated for correlation of gene expression with HIF activation, and statistically significant genes were selected.

QUANTIFICATION AND STATISTICAL ANALYSIS;

Unless otherwise stated, the results are expressed as mean ± Standard Error of Mean (SEM). Statistical analyses were performed using GraphPad Prism (version 10, GraphPad Software). The exact value of n within the figures and replicates is indicated in the figure legends.

Supplementary Material

1

Document S1 – containing Figures S1–S8

2

Table S1: Effects of pancreatic tumor interstitial fluid on gene expression changes by RNA-sequencing, related to Figure 2

3

Table S2: HIF-1 and HIF-2 mediated gene expression changes in PANC-1 cells, related to Figure 3

4

Table S3: Analysis of HIF-1 and HIF-2 mediated changes by pathway analysis, related to Figure 3

5

Table S4: Analysis of ARNT-mediated gene expression changes in PANC-1 cells, related to Figure 3

6

Table S5: Primer sequences used in study, related to STAR Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
ARNT	Cell Signaling Tech	Cat# 5537
CBS	Cell Signaling Tech.	Cat# 14782
COX IV	Cell Signaling Tech	Cat# 4850
CTH	ProteinTech	Cat# 12217-1-AP
Cytochrome c	Cell Signaling Tech	Cat# 4280
FLAG	Cell Signaling Tech	Cat# D6W5B
HIF-1α (for WB)	BD Biosciences	Cat# 610959
HIF-1α (for chIP)	Novus	Cat# NB100-479
HIF-2α	Invitrogen	Cat# PA1-16510
Parkin	Cell Signaling Tech	Cat# 2132
PHB1	Cell Signaling Tech	Cat# 2426
SDHA	Cell Signaling Tech	Cat# 11998
SLC7A11	Cell Signaling Tech	Cat# 12691
SLC3A2	Cell Signaling Tech	Cat# 47213
β-tubulin	Cell Signaling Tech	Cat# 86298
β-actin	Cell Signaling Tech	Cat# 4967
Bacterial and virus strains
Biological samples
Chemicals, peptides, and recombinant proteins
DFX	Millipore Sigma	Cat# 138-14-7
DMOG	Cayman Chemicals	Cat# 71210
Erastin	Selleck Chemical	Cat# S7242
Ferrostatin	Med Chem Express	Cat# SML0583
Imidazole Ketone Erastin	Selleck Chem	Cat# S8877
Roxadustat	Cayman Chemicals	Cat# 15294
RSL-3	Med Chem Express	Cat# 15294
Sulfasalazine	Med Chem Express	Cat# HY-14655
Mitosox Red	ThermoFisher Scientific	Cat# M36008
Sybr Green PCR Master Mix	ThermoFisher Scientific	Cat# 4301955
Critical commercial assays
Cell-Titer Glo	Promega	Cat# 7570
Lipid Peroxidation (4-HNE) Assay Kit	Abcam	Cat# ab238538
Dual Luciferase Assay Kit	Promega	Cat# E1910
Chromatin IP Kit	Cell Signaling Tech	Cat# 9003
RNeasy Mini Kit	Qiagen	Cat# 74104
iScript cDNA Synthesis system	BioRad	Cat# 1708890
PureLink Genomic DNA Mini Kit	Invitrogen	Cat# K182001
Deposited data
RNA-seq data	This paper	GEO: GSE320200, GSE319951, GSE319526
Experimental models: Cell lines
Human: PANC-1	ATCC	Cat# CRL-1469
Human: PANC-1 VHL KO clone (CRISPR-Cas9); mCherry+	This paper	N/A
Human: PANC-1 HIF-1α KO Clone 1 (CRISPR-Cas9); mCherry+	This paper	N/A
Human: PANC-1 HIF-1α KO Clone 2 (CRISPR-Cas9); mCherry+	This paper	N/A
Human: PANC-1 HIF-1α KO Clone 3 (CRISPR-Cas9); mCherry+	This paper	N/A
Human: PANC-1 HIF-2α KO Clone 1 (CRISPR-Cas9); GFP+	This paper	N/A
Human: PANC-1 HIF-2α KO Clone 2 (CRISPR-Cas9); GFP+	This paper	N/A
Human: PANC-1 HIF-2α KO Clone 3 (CRISPR-Cas9); GFP+	This paper	N/A
Human: PANC-1 ARNT KO Clone 1 (CRISPR-Cas9); GFP+	This paper	N/A
Human: PANC-1 ARNT KO Clone 2 (CRISPR-Cas9); GFP+	This paper	N/A
Human: PANC-1 Dox-Inducible shRNA-HIF-2α	This paper	N/A
Human: PANC-1 Dox-Inducible HIF-1α (P402/564A)	This paper	N/A
Human: PANC-1 Dox-Inducible HIF-2α (P402/564A)	This paper	N/A
Human: PANC-1 Cas9+; GFP+	This paper	N/A
Human: PANC-1 Cas9+; mCherry+	This paper	N/A
Human: MiaPaCa HIF-1α KO Clone 1 (CRISPR-Cas9); GFP+	This paper	N/A
Human: MiaPaCa HIF-2α KO Clone 1 (CRISPR-Cas9); GFP+	This paper	N/A
Human: MiaPaCa HIF-2α KO Clone 2 (CRISPR-Cas9); GFP+	This paper	N/A
Human: AsPC-1	ATCC	Cat# CRL-1682
Human: SW1990	ATCC	Cat# CRL-2172
Human: CFPAC-1	ATCC	Cat# CRL-1918
Human: BxPC-3	ATCC	Cat# CRL-1687
Human: RCC4	Laboratory of Celeste Simon	N/A
Mouse: 6694	Laboratory of Ben Stanger 52	N/A
Experimental models: Organisms/strains
Nod-Scid Gamma (NSG) Mice	Johns Hopkins Animal Facility	N/A
Oligonucleotides
All primers and sgRNAs used in this study (see Table S5)	This paper	Table S5
Recombinant DNA
pCDH-EF1-FHC	Addgene	Cat# 64874
pCW57-MCS1-P2A-MCS2	Addgene	Cat# 80921
psPAX2	Addgene	Cat# 12260
pMD2.g	Addgene	Cat# 12259
LRG (lenti-sgRNA-EFS-GFP)	Addgene	Cat# 65656
LRG-sgRNA-HIF-2α-guide 1 - GFP	This paper	N/A
LRG-sgRNA-HIF-2α-guide 2 - GFP	This paper	N/A
LRG-sgRNA-HIF-2α-guide 3 - GFP	This paper	N/A
LRG-sgRNA-ARNT-guide A - GFP	This paper	N/A
LRG-sgRNA-ARNT-guide B - GFP	This paper	N/A
LRCherry2.1	Addgene	Cat# 108099
LRG-sgRNA-HIF-1α-guide A- mCherry	This paper	N/A
LRG-sgRNA-HIF-1α-guide B- mCherry	This paper	N/A
LRG-sgRNA-VHL-guide A- mCherry	This paper	N/A
LRG-sgRNA-VHL-guide B- mCherry	This paper	N/A
LentiCRISPRv2 Puro	Addgene	Cat# 98290
Myc-Parkin	Addgene	Cat# 17612
pCDH-EF1-FHC; myc-Parkin	This paper	N/A
pCDH-EF1-FHC; CBS	This paper	N/A
pCDH-EF1-FHC; CTH	This paper	N/A
HIF-2α (P405A/P531A)	Addgene	Cat# 18956
pCW57-Dox-Inducible HIF-1α(P402A/P564A)	This paper	N/A
pCW57-Dox-Inducible-HIF-2α(P405A/P531A)	This paper	N/A
pCDH-EF1-FHC; HIF-2α(P405A/P531A)	This paper	N/A
pSV-Rl	Laboratory of Gregg Semenza 22	N/A
CBS promoter-luciferase	Laboratory of Gregg Semenza 22	N/A
Parkin	Addgene	Cat# 213559
Parkin(S65A)	Addgene	Cat# 213557
Software and algorithms
Graphpad Prism	Graphpad	https://www.graphpad.com/
Adobe Photoshop	Adobe	https://www.adobe.com/
FlowJo	FlowJo	FlowJo 11 - Download | FlowJo, LLC
RNASeq aligner	HISAT2	https://daehwankimlab.github.io/hisat2/
RNASeq read rummmarizer	featureCounts	https://subread.sourceforge.net/featureCounts.html
RNASeq differential analyzer	DESeq2	https://bioconductor.org/packages/release/bioc/html/DESeq2.html

Highlights.

Hypoxia and pancreatic tumor interstitial fluid protect against ferroptosis

HIF-2 mediates the protective effect of hypoxia against ferroptosis

HIF-2 transcriptionally regulates genes involved in glutathione metabolism

HIF-2 upregulates Parkin and suppresses mitochondrial ROS generation