Hypoxia Regulated Inhibin Promotes Tumor Growth and Vascular Permeability By ACVRL1 and CD105 Dependent VE-Cadherin Internalization

Hypoxia, a driver of tumor growth and metastasis, regulates angiogenic pathways that are targets for vessel normalization and ovarian cancer management. However, toxicities and resistance to anti-angiogenics limits their use making identication of new targets vital. Inhibin, a heteromeric TGFb ligand, is a contextual regulator of tumor progression acting as an early tumor suppressor, yet also an established biomarker for ovarian cancers. Here, we demonstrate a previously unknown role for inhibins and nd that hypoxia increases inhibin levels in ovarian cancer cell lines, xenograft tumors, and patients. Inhibin is regulated specically through HIF-1, shifting the balance from activins to inhibins. Hypoxia regulated inhibin promotes tumor growth, endothelial cell invasion and permeability. Targeting inhibin in vivo through knockdown and anti-inhibin strategies robustly reduces permeability in vivo and alters the balance of pro and anti-angiogenic mechanisms resulting in vascular normalization. Mechanistically, inhibin regulates permeability by increasing VE-cadherin internalization via ACVRL1 and CD105, a receptor complex that we nd stabilized directly by inhibin. Our ndings are the rst to demonstrate direct roles for inhibins in vascular normalization via TGF-b receptors providing new insights into the therapeutic signicance of inhibins as a strategy to normalize the tumor vasculature in ovarian cancer.


Introduction
Changes in angiogenesis are associated with metastasis in most cancers, including ovarian cancers, with signi cant impact on tumor progression and ascites development in advanced disease 1,2 . As such, antiangiogenic therapies have had signi cant impact in the management of ovarian cancers 3 . However, their effectiveness can be frequently limited due in part to toxicities and acquired resistance, leading to challenges with long term use and marginal improvements in overall survival 3 . Discovery of new and safer angiogenic targets is thus critical.
TGFβ family members, particularly BMP9 and TGFβ, are the most examined regulators of angiogenesis but have not been effective as targets for angiogenic therapy due to their pleiotropic functions in cancer and normal physiology 4,5 . Similar to TGFβ and BMP9, activins' have controversial and context dependent roles for regulating angiogenesis. Speci cally activin A has been shown to increase VEGF induced angiogenesis in some instances 6 and in others has been demonstrated to inhibit angiogenesis 7 . Inhibins' are a distinct and unique member of the TGFβ family as the only endocrine hormone and a functional heterodimer of an alpha (α) subunit (INHA) and a beta (β) activin subunit (INHBA or INHBB) forming either inhibin A or inhibin B respectively 8 . Inhibins' are distinct from activins which are comprised of homodimers of either beta subunit 8 . Inhibinα is synthesized as a pro-peptide with a pro-domain, αN region, and αC region. The pro-domain and αN region can be cleaved to produce the mature Inhibinα subunit comprising the αC region. Physiological Inhibinα production by the sertoli cells of the testes, granulosa cells of the ovary, and the adrenal and pituitary glands 9 is regulated primarily by FSH and LH 10,11 via a cAMP-PKA pathway resulting in cAMP response element binding (CREB) to the cAMP response element (CRE) on the INHA promoter 12 .
While inhibin levels (inhibin A and B) cycle across the lifespan of healthy females and dramatically decrease at the onset of menopause 13 , elevated inhibinα levels are found in ovarian, gastric, hepatocellular, and prostate cancers 14−17 . Total inhibin levels comprising free inhibinα, inhibin A and inhibin B are also an established diagnostic marker alone and/or in combination with CA125, for ovarian cancers 18 and have been proposed as a potential tumor speci c target for therapy 8,15,17,19,20, . Inhibinα levels are also predictive of survival in multiple cancer types with gene signatures that correlate with INHA expression, providing a highly accurate prognostic model for predicting patient outcomes 20 . However, the mechanism of inhibin expression in cancers have not been delineated.
Hypoxia is a key mediator of angiogenic responses, regulating pro-and anti-angiogenic genes impacting tumor growth, metastasis, and immune evasion 21 and is driven by the hypoxia inducible factor (HIF) family of transcription factors. Hypoxia induced changes, speci cally in tumors, are characterized by ine cient oxygen delivery, leading to leaky vessels and altered permeability, build-up of uid and ascites in ovarian cancer, and metastasis by facilitating intra/extravasation of tumor cells 21,22 . We previously reported decreased ascites accumulation in mice bearing tumor cells with INHA knockdown 19 , indicating a potential role for inhibin in regulating metastasis and vascular functions, a key contributing factor to ascites accumulation. Moreover, inhibin secreted by tumor cells induces angiogenesis via SMAD1/5 signaling in endothelial cells in a paracrine manner dependent on the type III TGFβ receptor endoglin/CD105 and the type I TGFβ receptor ALK1 19 .
To precisely delineate inhibin's signi cance in cancer and mechanism of action, we now determine the impact of hypoxia, a key mediator of the angiogenic and metastatic response in cancer 21 , and the contribution of inhibin to the hypoxia adaptive response. We discover that hypoxia in ovarian xenograft tumors, cancer cells, and patient samples leads to an increase in inhibin synthesis in a hypoxia inducible factor (HIF) dependent manner. We nd that hypoxia induced tumor growth and vascular permeability in vivo is driven by inhibin. Moreover, intervention using an antibody based therapeutic strategy to inhibin can suppress hypoxia driven tumor biology. Mechanistically, inhibin promotes vascular permeability via endoglin and ALK1. Notably, we also describe for the rst time using sensitive biophysical methods the nature and stability of the endoglin and ALK1 interaction at the cell surface in response to inhibin. Our ndings not just strongly implicate inhibins as part of the hypoxia adaptive response, but also suggest anti-inhibins' as an alternative or companion to current anti-angiogenic therapies that may not be well tolerated.

Materials And Methods
Cell Lines and Reagents: Ovarian epithelial carcinoma cell lines were obtained as described in resource Table 1 and were from ATCC, the NCI cell line repository through an MTA, or were as indicated. Cell line authentication was performed at the He in Center for Genomic Science Core Laboratories at UAB. HMEC-1s were grown per ATCC instructions. COS7 cells were grown in Dublecco's modi ed Eagle's medium with 10% FBS, 100 U penicillin/streptomycin and L-glutamine. Mouse embryonic endothelial cells (MEEC) WT and ENG -/-were grown as previously described 23 . Epithelial carcinoma cell lines HEY, OVCA420, SKOV3 and PA1 were cultured in RPMI-1640 containing L-glutamine, 10% FBS and 100 U of penicillin-streptomycin 24 . OVCAR-5 and HEK293 were cultured in DMEM containing 10% FBS and 100U of penicillin streptomycin. ID8ip2Luc was a kind gift from Jill Slack-Davis 25 and cultured in DMEM containing 4% FBS, 100U of penicillin streptomycin, 5µg/mL of insulin, 5µg/mL of transferrin, and 5ng/mL of sodium selenite. All cell lines were maintained at 37°C in a humidi ed incubator at 5% CO 2 , routinely checked for myco-plasma and experiments were conducted within 3-6 passages depending on the cell line. For hypoxia experiments, a ProOx Model C21 was used and set to 0.2% O 2 and 5% CO 2 . Anti-inhibin PO/23 and R1 antibodies were obtained from Oxford-Brookes university through an MTA and from Biocare Medical. INHA luciferase construct was generated through restriction cloning into pGL4.10 luciferase plasmid. Primers were designed to 547 base pairs of the INHA promoter containing the rst HRE site with Nhe1 and Xho1 restriction sites on the ends. Insert was ampli ed from PA1 genomic DNA. Insert was ligated into pGL4.10 plasmid with T4 DNA ligase and INHA promoter region was veri ed through Sanger sequencing. Additional details on resource is provided in Table 1.  Table 2.  Table 3. Cells were grown to 80% con uency in 24 well plates before media was replaced with fresh full serum media. Cells were placed in hypoxia chamber for 24hrs and media was collected and concentrated using Amicon Ultra centrifugal lter.

IN VITRO ASSAYS
In vitro Permeability Assay was adapted from Martins-Greene 27 . 1x10 5 HMEC-1 cells were plated onto a Matrigel coated 3µM trans-well lter in full serum media. After 24h hours, a second layer of 1x10 5 HMEC-1 was plated on top to obtain a con uent monolayer of cells. After an additional 24hrs, media was replaced with serum free media in the top of the trans-well and either conditioned media (with 2µg of either R1, PO/23, or IgG) or serum free media containing growth factor in the bottom chamber as indicated in legends. FITC-dextran was added to the lower chamber (10µg/ml). At indicated time points 10µL aliquots were taken from the top chamber in triplicate and measured using microplate reader for FITC-dextran passage. At end point, lters were stained with crystal violet to con rm equal monolayers were achieved.
Trans-well Migration Assay: 75 000 HMEC-1 were plated on a bronectin coated (10µg/mL) 8µM transwell lter in serum free media. Conditioned media (with 2µg of either R1, PO/23, or IgG) or serum free media containing 1nM inhibin A or VEGF A was used as a chemoattractant in the bottom chamber. After 24hrs, unmigrated cells were scraped off the apical side, migrated cells were xed in methanol:acetic acid, and nuclei were stained with Hoechst. Three random images were taken per lter using 10X objective on EVOS M7000 microscope. Nuclei were counted using ImageJ.

Trans-endothelial Migration Assay
HMEC-1 were grown on 8µm trans-well lters as per permeability assay. HMEC-1 monolayer was treated with 1nM inhibin A or untreated for four hours. After four hours of treatment, 150 000 HEY-LucGFP expressing cells were plated on top of the HMEC-1 monolayer and allowed to invade for 18hrs. Filters were xed in 4% paraformaldehyde, cells on the apical side of the lter were scraped off, and lters were mounted on glass slides for imaging. Migration of GFP + cells was visualized using 10x objective on EVOS M7000 microscope. Three random elds were captured per lter and GFP + cells were counted using ImageJ software. Thresholding, circularity and size gating were used to exclude unmigrated cells and artifacts.
Chromatin Immunoprecipitation protocol was adapted from ABCAM. Brie y, OV-90 or OVCAR-5 cells were grown in 150cm 2 dishes until 80% con uency was reached. Cells were kept under normoxia or placed in the hypoxia chamber set at 0.2% O 2 for either 12hrs (OVCAR-5) or 24hrs (OV-90). DNA was crosslinked using 0.75% formaldehyde and sheared by sonication to a fragment sizes between 100-400bp. DNA was immunoprecipitated with Dyna-beads and either HIF-1α antibody or Normal Rabbit IgG as a control. DNA was puri ed using Purelink PCR Puri cation kit and ampli ed using RT-qPCR with ChIP primers.
Luciferase Assay HEK293 cells were seeded into 24 well plate and co-transfected with a luciferase reporter containing 547 base pairs of the INHA promoter (pGL4.10 INHA) and a SV40 (Renilla internal control vector). For HIF-1 overexpression, cells were also co-transfected with pcDNA3-HA-HIF1aP402A/P564A or PCDNA3.1. One day after transfection, cells were left in a normoxia incubator or moved to hypoxia chamber (0.2% O 2 ) for 24hrs. Luciferase activity was measured using the Dual Luciferase Reporter Assay System by calculating the ratio between luciferase and Renilla and normalized to normoxia or PCDNA3.1 as indicated in legends.
Immuno uorescence HMEC-1 cells were grown to con uence on bronectin (10µg/mL) and treated with either 1nM inhibin A or VEGF A for 30 minutes in serum free media. Cells were xed in 4% paraformaldehyde and permeabilized with 0.2% TritonX-100, followed by blocking with 5% BSA in PBS for 1hr. VE-cadherin was labeled with anti-VE-cadherin antibody overnight at 4°C followed by AlexaFluor 488 secondary antibody. F-actin was stained with rhodamine-phalloidin and nuclei were labeled with DAPI. Immuno uorescence imaging was performed on EVOS M7000 microscope or Nikon A1 confocal microscope. Actin bers were quanti ed by measuring anisotropy using the FibrilTool Plugin in ImageJ 28 .  FRAP and patch/FRAP COS7 cells co-expressing epitope-tagged receptors labeled uorescently by anti-tag Fab′ fragments as described above were subjected to FRAP or patch/FRAP experiments as described 34 . FRAP studies were conducted at 15°C, replacing samples after 20 min to minimize internalization. An argon-ion laser beam (Innova 70C, Coherent, Santa Clara, CA) was focused through a uorescence microscope (Axioimager.D1; Carl Zeiss MicroImaging, Jena, Germany) to a Gaussian spot of 0.77 ± 0.03 µm (Planapochromat 63x/1.4 NA oil-immersion objective). After a brief measurement at monitoring intensity (528.7 nm, 1 µW), a 5 mW pulse (20 ms) bleached 60-75% of the uorescence in the illuminated region, and uorescence recovery was followed by the monitoring beam. Values of D and R f were extracted from the FRAP curves by nonlinear regression analysis, tting to a lateral diffusion process 34 . Patch/FRAP studies were conducted analogously, except that IgG-mediated cross-linking of epitope-tagged endoglin preceded the measurement 34 .

Patient Ascites
Specimens from patients diagnosed with primary ovarian cancer was collected and banked after informed consent at Duke University Medical Center, with approval for the study from Duke University's institutional research ethics board. ELISA's were conducted using ELISA for Total inhibin from Ansh labs (#AL-134).

Public Data Mining
Clinical data and normalized RNA-seq were obtained from cBioportal 35 . The ovarian serous cystadenocarcinoma (TCGA, PanCancer Atlas) and breast invasive carcinoma (TCGA, PanCancer Atlas) were assessed for INHA expression and hypoxia (Buffa or Winter) scores. INHA expression was plotted against hypoxia score for each patient for correlation analysis.
IN VIVOASSAYS. All animal studies and mouse procedures were conducted in accordance with ethical procedures after approval by UAB's IACUC prior to study commencement.

Matri-gel Plug Assay
Matrigel plugs were formed using 200µL of Matrigel mixed with 50µL of HEY conditioned media and injected subcutaneously into the underside of BALB/c female mice aged 5-6 weeks. For conditioned media, HEY cells were grown until 80% con uence in 24 well plate before media was replaced with fresh full serum media. Cells were placed in hypoxia chamber for 24hrs and media was collected and concentrated to 50µL Savant SpeedVac SPD1030. Conditioned media was incubated with 2µg of either R1 or IgG overnight before injection. Plugs were harvested 12 days after injection and hemoglobin content was determined according to Drabkin's method 19 .
In vivo subcutaneous tumor growth and permeability analysis 3x10 6 HEY cells either exposed to normoxia or hypoxia (0.2% O 2 ) for 24hrs were subcutaneously injected into right ank of 6-week old Ncr Nude mice (Taconic). Tumor volume ((LxW 2 )/2) was calculated by caliper measurements every other day starting at day 10 until harvest at day 30. In animals receiving antiinhibin treatment, R1 (BioCare) was administered IP at 2mg/kg three times weekly. Da Vinci Green diluent (BioCare) was administered as vehicle.
For measurement of permeability, tumors were harvested between 700-800mm 3 . At end point, Rhodamine Dextran 70 000 MW was intravenously injected at 2mg/kg two hours before euthanasia. Tumors were xed in 10% NBF and sections were analyzed for rhodamine-dextran by immuno uorescence on EVOS M7000. Three sections per tumor were quanti ed and four images per section were taken. Thresholding was performed in ImageJ and kept constant for all images. ROUT analysis (Q= 10%) was performed to test for outliers.
For tumor hypoxia analysis, tumors were harvested at varying sizes between 200-1400mm 3 .
Immuno uorescence on Tissues: Brie y, formalin xed, para n-embedded tissues from subcutaneous tumors were depara nized by sequential washing with xylene, 100% ethanol, 90% ethanol, 70% ethanol and distilled water for 10 min each. Antigen retrieval was performed by boiling tissues in sodium citrate buffer (pH 6.0). Blocking was performed with Background Punisher. Primary antibodies, antipimonidazole (1:50) and anti-CD-31(1:100), were diluted in Da Vinci Green Diluent and incubated overnight at 4°C in a humidi ed chamber followed by AlexaFluor 594 secondary antibody. Nuclei were stained with DAPI. 10x images were acquired on EVOS M7000 microscope.
Quantitation of CD-31 labeled vessel size and number as well as pimonidazole was performed in ImageJ.
Images were converted to binary and thresholding mask was applied equally to all images. For CD-31, objects smaller than 25 pixels were removed as were deemed too small to be vessels. For each image, average vessel size (area) and average vessel number was measured. Four images per section and two sections per tumor were used for quantitation. For pimonidazole, a 10x stitched image comprising the whole tumor section was used. The total area covered by signal was acquired and divided by total tumor area to calculate the % hypoxic area for each tumor.
Angiogenesis Proteome Array was performed according to manufacturer's instruction (R&D Systems, Table 1). Brie y, tissues were homogenized in PBS with 1% TritonX-100 and PI cocktail. 200µg of protein was used per sample (two samples for shControl and shINHA tumors each). Pixel intensity was quanti ed for each dot using ImageStudio software after background subtraction.

Statistical Analysis
All data are representative of three independent experiments, unless otherwise described in legends. Statistical analyses were performed using GraphPad Prism 9, with statistical test chosen based on experimental set up and speci cally described in the gure legends. Data are expressed as mean ± SEM.
Difference between two groups was assessed using a two-tailed t-test. Multiple group comparisons were carried by the analysis of variance (ANOVA) using One or Two-way ANOVA followed by appropriate posthoc tests as indicated in Figure legends.

Expression and secretion of inhibin is regulated by hypoxia in ovarian xenograft tumors cell lines and in patients
We and others have previously demonstrated increased expression of inhibinα mRNA and protein in a broad spectrum of cancers leading to increased angiogenesis in vitro and in vivo impacting metastasis 17,19,36 . Based on the potential role of inhibins' in cancer angiogenesis, we tested the impact of hypoxia, a key regulator of angiogenesis, on INHA expression. A panel of ovarian cancer cell lines representing a broad spectrum of ovarian cancer subtypes, including HEY, OV90, OVCAR5 of high grade serous origin, PA1 a teratocarcinoma cell line of the ovary, and ID8ip2 a mouse ovarian cancer cell line, were grown for twelve or twenty-four hours under either hypoxic conditions (0.2% O 2 ) or normoxic control tissue culture conditions (17-21%). Changes to INHA were evaluated by semi quantitative RT-PCR with VEGFA as a positive control (Fig. 1Ai-ii). We nd a 3-6 times increase in INHA expression across all four cell lines (HEY: 4.87-times, OVCAR5: 4.4-times, PA1: 5.28, OV90: 3.1-times, ID8ip2: 4-times, Fig. 1Ai). All cell lines showed maximum INHA increases after 24hrs of hypoxia growth except for OVCAR5 which increased INHA expression within 12hrs under hypoxia. VEGFA was evaluated side by side as a positive control and representative of the hypoxia response 37 in all four cell lines and was elevated 2-6-times (HEY: 3.5-times, OVCAR5: 3.1-times, PA1: 5.18-times, OV90: 2.5-times, Fig. 1Aii). HIF-1α stabilization in all cell lines con rmed by westerns indicated an active response to hypoxia in indicated cell lines (Fig. 1Aiii).
INHA translates into the protein inhibinα which can be secreted as a free monomer or can dimerize with INHBA or INHBB to produce dimeric functional inhibin A or inhibin B 12 . Thus, total inhibin ELISA, which detects all three, was used to test if the changes in INHA mRNA resulted in alterations to secreted protein.
We nd that conditioned media collected from HEY and OV90 exposed to hypoxia increased total inhibin protein secretion as well, (4.2-times in HEY and 3.8-times OV90, Fig. 1B). These data suggest that INHA mRNA and functional secreted inhibin protein, is increased by hypoxia.
Since total inhibin protein, re ecting either inhibin A/B and free inhibinα, increased in response to hypoxia (Fig. 1B), we evaluated mRNA changes in INHBA and INHBB subunits in HEY and OV90 cells. While INHA was increased three to ve-times in response to hypoxia (Fig. 1A), INHBA and INHBB levels were unchanged in the two cell lines evaluated (Fig. 1C), indicating that changes in inhibin protein levels ( Fig. 1B) were largely related to increases in inhibinα. The INHA response to hypoxia was also more robust in tumor cells as compared to endothelial cells (HMEC-1) grown under hypoxia (0.2% O 2 ) for 24hrs exhibited (Supp Fig. 1) indicating that inhibinα increases in response to hypoxia occur more signi cantly in tumor cells.
To evaluate other pathologically relevant hypoxic conditions pertinent to ovarian cancer growth and metastasis, we evaluated hypoxia and INHA expression in cells grown in spheroids under anchorage independence, an environment that is often hypoxic 38 . PA1 and OVCA420 cells were chosen due to their ability to form spheroids 39,40 . Cells were grown on poly-hema coated plates for either 72hrs (PA1) or 48hrs (OVCA420). Under such 3D conditions, where HIF-1α was stabilized (Fig. 1Di), INHA was increased 7.8-times in PA1 and 4.6-times in OVCA420 when compared to 2D growth conditions in a dish (Fig. 1Dii).
Previous studies have established that in healthy pre-menopausal women, inhibin levels cycle across the menstrual cycle reaching a peak of 65.6 pg/mL, while in post-menopausal women, total serum inhibin levels are below 5 pg/mL 41 . Ovarian cancer patients are commonly postmenopausal 42 and tumor tissues can display higher inhibin levels 19 . We thus wanted to assess if the peritoneal ascites uid of advanced ovarian cancer patients, which has been shown to be a hypoxic environment 43 and contains disseminated ovarian cancer spheroids 22 , also displays detectable or elevated inhibin levels. To test if inhibin protein is secreted and detectable in clinical ascites, total inhibin ELISA was performed on a cohort of 25 patient ascites (Methods). We nd total inhibin levels in the range of 6.7 to 120.53pg/mL in the ascites uid with increasing concentrations found in higher stages of disease (Fig. 1E).
We next evaluated if INHA expression was elevated in vivo with increasing xenograft tumor size. 5 million HEY cells were subcutaneously implanted and harvested at varying tumor sizes. Tumors greater than 500mm 3 were found to be hypoxic based on pimonidazole staining (4.8-times, Fig. 1Fi and Supp Fig. 2A).
INHA expression was increased 9.8 times in tumors greater than 500mm 3 than in tumors less than 500mm 3 (Fig. 1Fii). INHA expression was also signi cantly correlated with tumor size (Supp Fig. 2B). To further examine the potential clinical relevance of inhibinα expression in response to hypoxia, we analyzed the TCGA/PanCancer Atlas patient data set from cBioportal 35,44 and obtained hypoxia scores from two different hypoxia gene signatures (Buffa and Winter) 45,46 . The signatures consisted of 51 (Buffa) and 99 (Winter) hypoxia related genes from a large meta-analysis of breast and head and neck squamous cell cancer that were independently veri ed for prognostic value 45,46 . Using these signatures, inhibinα (INHA) expression was signi cantly correlated with both hypoxia Buffa (r=0.1961, p=0.0221) and Winter hypoxia (r=0.223, p=0.009) scores in the ovarian cancer data set (Fig. 1Gi-ii). Analysis of breast cancer data revealed a similar trend as INHA expression was signi cantly correlated (r=0.2026, p=0.0165) with the Buffa hypoxia score (Fig. 1Giii). Taken together, these data strongly indicate that inhibinα mRNA and protein expression are increased under hypoxia conditions in ovarian cancer cell lines, xenograft tumors and in patients.

INHA is a direct HIF-1 target under hypoxia
Hypoxia inducible factors (HIFs) are key transcriptional regulators of the hypoxia adaptive response and increase expression of critical pro-angiogenic genes 21 . To test whether HIF proteins are regulators of INHA expression, we rst utilized cobalt chloride (CoCl 2 ), a well characterized chemical stabilizer of HIF's 47 . HIF-1α was stabilized in PA1 and OVCAR5 cells treated with 100µM of CoCl 2 for either 6, 12, or 24 hrs (Fig. 2Ai). We nd that INHA expression was signi cantly increased; 10-times in OVCAR5 after 12hrs and 11.5-times in PA1 cells after 24hrs of CoCl 2 treatment (Fig. 2Aii). Maximum increases in INHA expression with CoCl 2 occurred at the same time points as exposure to hypoxia (12hrs for OVCAR5 and 24hrs for PA1, Fig. 1Ai). VEGFA, used as a positive control increased 4.8 and 4.3-times at 12hrs and 2.7 and 3.7-times at 24hrs in both OVCAR5 and PA1, respectively (Fig. 2Aii). To test if INHA could be a direct hypoxia target leading to increased inhibinα expression, we evaluated the effect of reducing the levels of HIF-1β/ARNT which is the binding partner for all HIF's 48 . Stable ARNT knockdown cells were generated in HEY cells (Methods). We nd that control HEY cells increase INHA levels 2.8-times under 0.2% hypoxia (Fig. 2B). However, shRNA ARNT lead to a 2.7-times reduction in hypoxia induced increase in INHA mRNA levels (Fig. 2B) indicating direct contributions of HIFs' to the regulation of inhibin.
To determine the roles of the HIF-1 and HIF-2 heterodimeric transcriptions factors, that both require ARNT 48 , in the transcriptional regulation of INHA we used siRNA to knockdown the levels of HIF-1α and HIF-2α (Methods). HEK293 cells were used as they express relatively equal levels of both HIF isoforms (Fig. 2Ci). HEK293 cells with either control or HIF1/2α siRNAs were exposed to hypoxia for 24hrs and e cacy of HIF1/2α knockdown was con rmed by immunoblotting (Fig. 2Ci). Notably, siRNA to HIF-1α (siHIF-1α; Fig. 2Cii) decreased hypoxia induced INHA expression 1.8-times as compared to scramble controls (siScr; Fig. 2Cii). However, siRNA to HIF-2α resulted in a smaller (1.25-times) and non-signi cant reduction in INHA expression compared to siScr when exposed to hypoxia (Fig. 2Cii). These data suggest that increases in INHA under hypoxia were more signi cantly impacted by HIF-1 as compared to HIF-2 .
In silico, analysis of the INHA gene, which is located at Chr:2q35 revealed two hypoxia response element (HRE) consensus sites within 2Kb of the promoter, GGCGTGG and CGCGTGG, at -144 and -1789 bp from the transcription start site (TSS) (Supp Fig. 3A) respectively. These HRE sites conform precisely to the (G/C/T)(A/G)CGTG(G/C) consensus sequence 48 . Two hypoxia ancillary sequences (HAS) (CAGGG and CACGG) were also found directly anking the proximal HRE sequence at -169 and -173 bp from the TSS, respectively. One HAS sequence (CACGT) was found anking the distal HRE sequence at -1761 bp from TSS (Supp Fig. 3A). A previously well characterized CREB binding site (CRE) is designated for reference (Supp Fig. 3A).
To test direct interactions between HIF-1 and the INHA promoter, chromatin immunoprecipitation (ChIP) was performed using OVCAR5 and OV90 cells. Primers were designed to amplify the region including the HRE site closest to the transcription start site (HRE1) and chromatin shear size optimized accordingly (Methods). We nd that exposure to hypoxia led to a 4-times increase in enrichment of HIF-1 binding to INHA's HRE site in OVCAR5 and 3-times in OV90 (Fig. 2D). The second HRE site is GC rich which lead to modest ampli cation. Despite this, a 2-times increase in HIF-1 enrichment at this site in OV90 cells was observed (Supp Fig. 3B) which was however not statistically signi cant.
Given the poor enrichment of HIF-1 at the distal promoter site (Fig. S3B), we next evaluated if the proximal promoter was su cient to increase INHA levels under hypoxia and if this was dependent on HIF-1. To achieve this, we utilized 547 base pairs of the INHA promoter, containing the rst HRE site (Fig.  S3A), in a luciferase reporter assay (Fig. 2E). The effect of HIF-1 on INHA promoter activity, was evaluated in HEK293 cells exposed to hypoxia (0.2% O 2 ) for 24hrs and compared to cells under normoxia (Fig. 2Ei), or in the presence or absence of HIF-1 ODD (pcDNA3-HA-HIF1aP402A/P564A) (Fig. 2Eii) that prevents degradation of the HIF1α subunit 26 . We nd that in un-transfected or control vector expressing cells (pcDNA3.1), INHA promoter luciferase activity is increased two times in response to hypoxia (Fig. 2Ei) that was mimicked by stabilization of HIF-1α (HIF-1 ODD) under normoxia conditions (Fig. 2Eii). These  Fig. 3C). This relationship appeared to be additive and not synergistic as addition of the PKA inhibitor, H89, was not able to reduce hypoxia induced INHA expression (Supp Fig. 3C). Taken together, these data implicate HIF-1 as being the key transcriptional factor responsible for increase of INHA in hypoxia.

Inhibin promotes hypoxia induced angiogenesis and stimulates endothelial cell migration and vascular permeability
Hypoxia is a key driver of endothelial cell migration and blood vessel permeability within the tumor leading to alterations in angiogenesis 37 . To determine the overall contribution of inhibin to hypoxia induced angiogenesis in vivo, we utilized an in vivo Matrigel plug assay. Conditioned media (CM) from HEY tumor cells exposed to normoxia or hypoxia was used to stimulate angiogenesis into the plugs and a well-established anti-inhibinα antibody, R1 (recognizing the junction between the αN region, and αC region) 51 was used to block inhibin in the CM with IgG as a control. We nd that CM from hypoxia grown cells increased hemoglobin in the plugs 2.9-times compared to CM from normoxia grown cells (Fig. 3Ai-ii). Anti-inhibinα in the hypoxic CM fully reduced the hemoglobin content in the plug (2.1-times suppression, Fig. 3Ai-ii) indicating that inhibin is required for hypoxia induced blood vessel formation in vivo.
Since blood vessel ow is an indication of endothelial cell functionality 52 , we sought to de ne the speci c effects of increased inhibinα on hypoxia induced endothelial cell biology, speci cally endothelial cell chemotaxis and vascular permeability. To determine the impact on endothelial chemotaxis to hypoxic CM, CM from either hypoxia (24 hrs, 0.2% O 2 ) or normoxia grown OV90 or HEY cells were used as a chemoattractant to measure migration of human microvascular endothelial cells (HMEC-1; Fig. 3B). Two anti-inhibinα antibodies, R1 and a second well established antibody PO23 (recognizing the C-terminus of the αC region) 51 , were used with IgG controls to test the effect of blocking/sequestering hypoxia produced inhibinα. We nd that CM from hypoxia grown tumor cells signi cantly increased migration of endothelial cells (IgG, Fig. 3B) and incubation of hypoxic CM with anti-inhibinα R1 signi cantly suppressed hypoxia induced endothelial migration (2.1 and 1.6-times for OV90 and HEY conditioned media respectively, Fig. 3Bi-ii). Anti-inhibinα PO23 was also able to signi cantly suppress CM stimulated endothelial migration (1.5 and 1.75-times for OV90 and HEY CM, respectively, Fig. 3Bi-ii). Similar to the effects of hypoxic CM, recombinant inhibin A was also able to stimulate HMEC-1 migration to similar extents as VEGF A at equimolar amounts (Fig. 3Biii).
We next evaluated the effect of CM from hypoxic tumor cells on changes to permeability across an endothelial monolayer using a trans-well permeability assay that measures solute (FITC-dextran) ux across endothelial monolayers. Permeability was monitored across a four-hour time course and CM from hypoxic tumor cells was used to induce permeability across the HMEC-1 monolayer. Effect of inhibin in the CM was evaluated either in the presence of anti-inhibinα (PO23 and R1) or IgG control (Fig. 3Ci-ii). We nd that both inhibinα antibodies (R1 and PO23) signi cantly decreased solute ux induced by hypoxic CM from two tumor cell lines, albeit with moderate differences in the kinetics and time to inhibition ( Fig. 3Ci-ii). Speci cally, signi cant inhibition of permeability was seen beginning at two hours for CM treated with PO23 and three hours for R1. PO23 was moderately more effective than R1 as it effectively reduced permeability within 1 hour (Fig. 3Ci-ii). Recombinant inhibin was also able to induce endothelial cell permeability to similar extents as LPS (Fig. 3D), an established permeability inducing factor 43 . Since perturbations to the endothelial barrier are critical to invasion and extravasation of cancer cells during metastasis 53 , we tested whether inhibin induced vascular permeability facilitates tumor cell extravasation. To test this, we used a trans-endothelial cell migration assay to mimic the process. HEY tumor cells infected with GFP adenovirus to distinguish them from migrated non-GFP endothelial cells were plated on top of a non-GFP endothelial cell monolayer that was then either pre-treated with 1nM inhibin A for 4 hours or left untreated. We nd that HEY GFP tumor cells, were 2.9-times more invasive across the inhibin treated monolayer than untreated conditions (Fig. 3Eii-iii). All together, these data implicate inhibin as a robust contributor to hypoxia mediated angiogenesis, vascular permeability and thereby tumor cell extravasation across the vascular endothelium.
Inhibin promotes vascular permeability through increased VE-cadherin tra cking.
Endothelial permeability is regulated through changes in junctional proteins which are maintained through contacts with the actin cytoskeleton 54 . VE-cadherin is a critical junctional protein involved in regulating endothelial cell permeability 54 . To delineate the mechanism of inhibin's effects on vascular permeability, we rst evaluated the effect of inhibin on endothelial cell junctions and the actin cytoskeleton through immuno uorescent staining of VE-cadherin and actin (Fig. 4A). Examination of the actin cytoskeleton revealed signi cant contractile actin staining, with a signi cant increase in stress ber formation after 30 minutes of inhibin A treatment (two times increase, Fig. 4Ai-ii). VEGF A treatment was used as a comparison that also led to similar changes in actin stress ber formation (Fig. 4A). VEcadherin localization also appeared to be reduced qualitatively at the cell-cell junctions after 30 minutes of inhibin treatment as compared to untreated cells, suggestive of perturbation of the endothelial cell barrier at the level of the cytoskeleton (Fig. 4A). Loss of VE-cadherin at the cell junctions was also observed in VEGF A treated cells (Fig. 4A). However, total VE-cadherin levels were unchanged in response to inhibin as evaluated over a time course of 60 minutes (Supp Fig. 4) indicating no change in the total pool of VE-cadherin in response to inhibin A. Actin contractility and stress ber assembly is regulated through phosphorylation of myosin light chain (MLC) 54 . In accordance, we nd that phosphorylation of MLC-2 (Ser19) increased within 5 minutes of inhibin A treatment and was sustained across a 60-minute time course (Fig. 4Bi-ii). Based on the qualitative changes in VE-cadherin in response to inhibin A treatment (Fig. 4A), we tested whether alterations in VE-cadherin at the cell-cell junctions were due to inhibin induced VE-cadherin internalization (Fig. 4C). To determine this, HMEC-1 membrane localized VE-cadherin was labeled at 4°C with an anti-VE-cadherin antibody recognizing the extra-cellular domain. HMEC-1 cells were washed with acid to remove membrane bound anti-VE-cadherin leaving only any internalized VE-cadherin that may have been labeled at 4°C prior to treatment with inhibin A or VEGF A (Fig. 4Ci). Stripping of cell surface VE-cadherin was veri ed by cell surface immunostaining of VEcadherin with little to no internalized VE-cadherin detected (Fig. 4Cii,iv). Cells were then either left untreated or treated for 30 minutes with inhibin A at 37 o C and VE-cadherin evaluated by immuno uorescence (Fig. 4Ciii). We nd that inhibin A increased the internalized VE-cadherin pool compared to untreated cells 1.4-times (Fig. 4Cv) and to similar extents as VEGF A (1.6-times, Fig. 4Cv). These results indicate that inhibin induces rapid changes in the actin cytoskeleton and tra cking of VEcadherin from the cell junctions of endothelial cells. We next evaluated if internalization of VE-cadherin by inhibin was dependent on endoglin using mouse embryonic endothelial cells (MEEC) that are either wild type (WT) or null for endoglin expression 51 (Supp Fig. 6). Cell surface biotinylation of VE-cadherin was used to quantitatively assess VE-cadherin internalization. Towards this, cell surface proteins were labeled with Sulfo-NH-SS biotin and allowed to internalize for 30 minutes at 37°C in the presence or absence of inhibin followed by stripping of cell surface biotin, immunoprecipitation with neutravidin resin and immunoblotting to detect internalized biotin labeled VE-cadherin (Fig. 5Bi). Treatment with inhibin A increased internalized VE-cadherin 1.9times in MEEC WT compared to control (Fig. 5Bi), similar to extents seen by immuno uorescence in HMEC-1 cells (Fig. 4C). However, in the absence of endoglin in MEEC ENG-/-cells inhibin A did not change the internalized VE-cadherin pool (Fig. 5Bii). This data indicates that endoglin is essential for inhibins effects on VE-cadherin.
Based on the signi cant dependency of inhibin's effects on endothelial cell permeability and VE-cadherin internalization on endoglin and ALK1 respectively (Fig. 5A,B), we evaluated biophysically, in a sensitive and quantitative manner, the extent of the endoglin-ALK1 interaction in response to inhibin. We utilized a patch/FRAP ( uorescence recovery after photobleaching) methodology to measure interactions between endoglin and ALK1 at the surface of live cells. This method differentiates between stable and transient interactions as described in detail previously 57 . Herein, one receptor carrying an extracellular epitope tag is patched and immobilized through cross-linking with a double layer of IgGs. The effects of this immobilization on the lateral diffusion of a co-expressed, differently-tagged receptor labeled exclusively with Fab' fragments are then measured by FRAP (Methods). Stable complex formation between the two co-expressed receptors (complex lifetimes longer than the characteristic FRAP uorescence recovery time) reduces the mobile fraction (R f ) of the Fab'-labeled receptor, since bleached Fab'-labeled receptors associated with immobilized receptors do not appreciably dissociate from the immobile patches during the FRAP measurement. On the other hand, transient complexes (short complex lifetimes) would reduce the apparent lateral diffusion coe cient (D), since each Fab'-labeled receptor molecule can undergo multiple association-dissociation cycles during the FRAP measurement 57 . For these studies, COS7 cells were transfected with myc-ALK1, HA-endoglin or co-transfected with both, and subjected to patch/FRAP experiments in the absence or presence of 4 nM of inhibin A (Fig. 5C). Fig. 5Ci-iii depict representative FRAP curves showing the lateral diffusion of myc-ALK1 (Fig. 5Ci), IgG-crosslinked and immobilized HAendoglin (Fig. 5Cii), and myc-ALK1 co-transfected with HA-endoglin followed by IgG cross-linking of HAendoglin in the presence of inhibin (Fig. 5Ciii). Average values derived from multiple independent experiments are shown in (R f in Fig. 5Cv, D values in Fig. 5Cv). Singly expressed myc-ALK1 had lateral mobility resembling other TGF-β superfamily receptors 34 , which was insensitive to inhibin treatment ( Fig. 5Ci and iv). Immobilization of HA-endoglin ( Fig. 5Cii and iv) reduced R f of myc-ALK1 by about 45%, and the presence of inhibin increased this reduction signi cantly (from 45-70% reduction) (Fig. 5Ciii and  iv). Under all these conditions, the lateral diffusion coe cient (D) of myc-ALK1 was not signi cantly affected (Fig. 5Cv), indicating that endoglin and ALK1 form stable complexes at the plasma membrane which are enhanced and stabilized by inhibin.
Previous studies indicate that inhibinα may bind to ALK4 58 , an established Type I receptor for the Activin family of proteins 8 . We thus employed patch/FRAP to determine the interactions between endoglin and ALK4 and to examine whether inhibin A enhanced these interactions. To this end, we expressed HA-ALK4, myc-endoglin or both in COS7 cells, and subjected them to patch/FRAP studies on the lateral diffusion of HA-ALK4 without and with IgG cross-linking of myc-endoglin, and with or without inhibin A. In the absence of inhibin A, endoglin and ALK4 exhibited signi cant stable interactions, as demonstrated by the reduction in R f of HA-ALK4 upon immobilization of myc-endoglin (40% reduction in R f , with no effect on the D value) (Fig. 5Di-ii). However, in contrast to the observations with endoglin-ALK1 complexes, the interactions between endoglin and ALK4 were weakened in the presence of inhibin A (the reduction in R f decreased to 20%) (Fig. 5Di). Taken together, these results indicate that inhibin shifts the balance of endoglin complexes from interactions with ALK4 to interactions with ALK1, both of which (endoglin and ALK1) are required for inhibin mediated vascular permeability.

Inhibin promotes hypoxia induced tumor growthin vivothrough alterations in permeability and angiogenesis
The signi cance of hypoxia in ovarian cancer is well documented and we previously demonstrated increased ascites accumulation in tumor bearing mice in the presence of inhibin 19 . To precisely de ne the contribution of inhibin to hypoxia induced tumor growth and angiogenesis, we rst evaluated the effects of pre-exposure to hypoxia on tumor growth in a subcutaneous model in vivo, a model that allows for quantitative analysis of the vasculature in tumors 59 . HEY pLKO.1 control vector (shControl) cells were preexposed to hypoxia (0.2% O 2 ) for 24hrs or kept under normoxia followed by injection into the right ank of Ncr nude mice. Tumors were measured throughout and harvested after 30 days (n=10 mice). HEY cells pre-exposed to hypoxia produced rapid growing tumors compared to those that originated from normoxia grown cells (Fig. 6A, purple versus black line). In parallel, we utilized two methods to perturb inhibin: 1) shRNA knockdown of INHA in HEY cells (Supp Fig. 7A,B) and 2) intraperitoneal administration of antiinhibinα antibody (R1). R1 is a human antibody 51 and consistent with this no overall toxicity was noted in pilot toxicity studies that utilized daily injections of R1 (Supp Fig. 7C). shINHA cells exposed to hypoxia maintained their knockdown to INHA at the end of the study (Supp Fig. 7B) and produced tumors with signi cantly slower growth rates than shControl hypoxia tumors (Fig. 6A, blue versus black lines). In complementary ndings, hypoxia exposed tumor cells had signi cantly reduced tumor growth upon receiving treatment with the R1 antibody when compared to tumors in mice that received vehicle only (Fig. 6A, red versus blue line, n=6 for R1 treated mice). The group receiving anti-inhibinα (R1) grew at a similar rate as the shINHA hypoxia tumors (Fig. 6A, red versus blue line). In mice with shINHA tumors, treatment with R1 further reduced tumor growth albeit moderately compared to vehicle shINHA (Fig. 6A, blue versus green line). These data indicate that perturbation of inhibin through shRNA targeting and antiinhibin antibody treatment reduces tumor growth. We hence sought to determine the effect of shINHA on the angiogenic cytokine pro le of the tumors using a proteome array of 55 different human angiogenesis targets. We nd that the most up-regulated proteins in control tumors compared to shINHA tumors were a subset of pro-angiogenic cytokines IL8 (2.5-times) and EGF (2.1-times) (Fig. 6Bi) indicating a proangiogenic pro le of the tumor cells in the presence of inhibin. In contrast, the shINHA hypoxia tumors showed increases in proteins including ADAMTS-1 (1.6-times) and Pentraxin-3 (1.3-times), indicating an anti-angiogenic pro le in shINHA tumor cells as both have been demonstrated to be anti-angiogenic 60,61 . Activin A and endoglin were also found to be elevated in shINHA tumors (Fig. 6Bi). To complement the human tumor array, we analyzed changes in the mouse angiogenic proteome as well to delineate any host differences in response to shControl and shINHA tumor cells. We nd that host cells also upregulated signi cantly more pro-angiogenic proteins, including CXCL16 62 , PIGF-2 63 , and NOV 64 in shControl tumors compared to shINHA tumors ( Fig. 6Bii). Taken together, these data suggest that altering inhibin in the tumors results in a change in the balance of angiogenic factors leading to a signi cant reduction in pro-angiogenic factors and slower overall tumor growth.
To rule out whether the reduction in tumor growth in shINHA cells was due to slower proliferation of tumor cells, growth rate of HEY shINHA and HEY shControl was evaluated in culture under hypoxia for 3 days. No signi cant change was observed (Supp Fig. 7D) suggesting that the major effect of inhibin on tumor growth are likely through effects on the tumor vasculature due to the effects of hypoxia regulated inhibin on angiogenesis and vascular permeability in vitro (Fig. 4,5). We thus determined the effect of inhibin on the tumor vasculature and associated permeability changes as a contributing factor to the altered tumor growth in shINHA and antibody treated hypoxia tumors (Fig. 6A). To this end, HEY shControl or shINHA cells pre-exposed to hypoxia for 24hrs were injected subcutaneously into the right ank of Ncr nude mice (n=4 mice, Fig. 6B) and tumors in all groups were harvested upon reaching 700-850mm 3 (Fig. 6Ci) to eliminate any tumor size effects on angiogenesis. These tumors (Fig. 6Ci) were evaluated for changes in vascular permeability by visualization of a rhodamine-dextran dye that leaks from the blood vessels into the tumors when administered into mice prior to sacri ce. We nd that rhodamine-dextran was present at 5.5-times higher levels in shControl tumors compared to shINHA tumors indicating higher vascular permeability within the tumors in the presence of inhibin (Fig. 6Cii-iii).
To further characterize the differences in the vasculature between shControl and shINHA tumors, blood vessels were stained with CD-31 to evaluate vessel number and size (Fig. 6D). We nd an increase in the total number of blood vessels in shControl tumors compared to shINHA tumors (Fig. 6Di,iii). Quantitation of the size of the vessels revealed signi cantly smaller vessels in shControl tumors as compared to the shINHA tumors (Fig. 6Dii,iii). These data together demonstrate that reducing inhibin in the tumor decreases vascular leakiness, alters vessel size and numbers and promotes more normalized vasculature in the tumors.

Discussion
Hypoxia signi cantly impacts several aspects of tumor progression by regulating pathways that can be targeted for cancer management, particularly angiogenic mechanisms. We have for the rst time identi ed inhibins', that are well established biomarkers for ovarian and other cancers and a member of the TGFβ superfamily, to be targets of the hypoxic response. We signi cantly extended our previous ndings 19 to demonstrate that hypoxia induced tumor growth, angiogenesis and vascular leakiness is accompanied with, and dependent on inhibin levels in cells and tumors, and relevant to the ovarian cancer patient population. In keeping with this, hypoxia induced tumor growth can be suppressed by treatment with a selective inhibin antibody that leads to a shift in the angiogenic balance in tumors. We also provide mechanistic evidence for the involvement of ALK1 and endoglin in inhibin's effects on permeability via increased VE-cadherin internalization. Due to the lack of systemic inhibin expression in post-menopausal women, establishing the therapeutic signi cance of targeting inhibin in this patient population may be particularly bene cial to evade systemic side effects seen with targeting other hypoxia associated angiogenic pathways.
Signi cant information exists on the cycling levels of inhibins' in premenopausal women, the decline of inhibin during peri-menopause, and as a marker whose decline de nes the onset of menopause leading to complete absence of inhibin in normal post-menopausal women 13 . Contrastingly, several studies have reported elevated levels of Inhibin in a subset of cancers 14,15,17,36 . Our studies shed light on the potential mechanisms leading to elevated inhibin. We also nd that total inhibin is elevated in the ascites uid of patients with ovarian cancer, a hypoxic environment that aids in dissemination of shed ovarian cancer spheroids 22,43 (Fig. 1). Serum inhibin and CA125 levels are both markers for ovarian cancer 17,18 and were also positively correlated with each other in these patient ascites uid (Supp Fig. 8A). Menopause status was unknown in these patients however the median age of the cohort was 62 and only two patients were below 50 years of age (Supp Fig. 8B). INHA expression and hypoxia are also correlated through a hypoxia gene score (Fig. 1G). Supporting our hypothesis that inhibin is regulated by hypoxia, we also found that exposure of ovarian cancer cells to hypoxia increased INHA expression and inhibin secretion (Fig. 1). Surprisingly, we did not note consistent and statistically signi cant increases in the activin subunits, INHBA or INHBB which only appeared to be moderately elevated (Fig. 1) indicating that increased secretion levels were driven by inhibinα. Previous reports indicate activin, speci cally INHBA, increases in response to hypoxia in endothelial cells 65 . However, here the increase in INHA levels in endothelial cells in response to hypoxia was only moderate as compared to in tumor cells (Fig. S1)  Mechanistically, we nd through knockdown studies and ChIP studies that INHA expression is regulated through the HIF-1 transcription factor binding directly to the INHA promoter (Fig. 2). Our ndings on a hypoxia response in a pathological condition as seen here, is consistent with a previous report demonstrating that FSH can drive INHA expression in granulosa cells dependent on HIF-1 in what appeared to be in an indirect manner 66 . Intriguingly, evidence for INHA regulation by hypoxia, speci cally dependent on HIF isoforms, has been demonstrated in cytotrophoblasts 67 . Here we present detailed and direct evidence of regulation by HIF-1, with HIF1 interacting at INHA's promoter to drive expression under hypoxia (Fig. 2). cAMP and PKA can be activated in response to hypoxia as well. However, the PKA inhibitor, H89, was not able to reduce hypoxia induced INHA expression indicating that cAMP may not be involved in the hypoxia transcriptional regulation of INHA (Fig. S3). This does not preclude a role for cAMP-PKA in the regulation of INHA as it is well established that the cAMP-PKA signaling axis enhances tumorigenesis in ovarian cancer 68 . As the effect of forskolin was additive on INHA expression, cAMP and PKA could represent an alternative or additive mechanism of regulation of INHA in ovarian cancer.
In prostate and adrenocortical cancers reports of both increased and decreased inhibinα levels have been reported 15,69,70 . In adrenocortical tumors with lower INHA levels, methylation of the INHA promoter was reported to occur at the CpG island within the proximal HRE site that we identi ed, suggesting potential roles for epigenetic regulation of INHA as well 70 . HIF transcription factors have reduced binding to methylated hypoxia response elements 71 . To this end, it is possible that not all cell lines will increase inhibin expression in response to hypoxia. If this is the case, methylation of INHA' s promoter may play a role making further understanding of the regulation of INHA expression, particularly in patients necessary in the future.
Previously we also demonstrated inhibin's effects broadly on angiogenesis 19 . Here, we sought to de ne more precisely the outcomes of inhibin's effects on angiogenesis, speci cally in the context of hypoxia.
Using recombinant inhibin and antibodies to the alpha subunit of inhibin, we nd novel roles for inhibin as a permeability inducing factor with implications for tumor cell extravasation (Fig. 3). Inhibin induced permeability was dependent on ALK1 and endoglin (Fig. 5A). The VE-cadherin dependent mechanism of permeability observed by us (Fig. 4) is consistent with prior ndings on the effects of other TGFβ family members' roles in promoting vascular permeability, speci cally BMP6 30 . BMP6 induced vascular permeability was mediated through the Type 1 receptor ALK2 30 , whereas we expect the Type 1 receptor ALK1 to be more critical for inhibin induced vascular permeability. Interestingly, inhibin strongly increased the stable interaction between ALK1 and endoglin (Fig. 5C), in line with our observation that both endoglin and ALK1 are required for permeability, and endoglin being critical for VE-cadherin internalization (Fig. 5). These ndings have broad implications for other TGFβ family members that may regulate permeability dependent on Type 1 receptors. The patch/FRAP studies (Fig. 5) support our current and previous ndings 19 . Although there are some reports suggesting that inhibin can bind to ALK4 58 , our ndings show that inhibin does not enhance endoglin-ALK4 complex formation but rather weakens it (Fig. 5D). We have previously demonstrated that endothelial cells such as HMEC-1 express very little ALK4 compared to ALK1 19 , supporting the idea that inhibin acts in endothelial cells preferentially via ALK1 in line with a potential physiological relevance of inhibin-mediated increase in endoglin-ALK1 interactions. However, these ndings in endothelial cells do not contradict the current understanding of inhibin's function in non-endothelial cells, which may express more ALK4 than ALK1. These ndings also do not allow us to conclude whether the ALK1-endoglin complex, which is enhanced by the binding of inhibin, is signaling or kinase competent, as non-signaling receptor complexes may exist and impact signaling in an indirect manner. Such complexes were previously reported in the context of activin and ALK2 72,73 and need further examination for inhibins.
Targeting inhibin through shRNA knockdown and antibody treatment was found to be an effective antiangiogenic strategy leading to reduced vascular permeability increased blood vessel size but fewer number of vessels and a likely more normalized vasculature (Fig. 6D). Interestingly, in our analysis of the angiogenic proteome of HEY tumors, permeability promoting cytokines, EGF, IL-8, and DPP4 were signi cantly lower in shINHA tumors which were less permeable as compared to shControl tumors ( Fig. 6B). Interestingly, the shINHA tumor cells produced more activin and endoglin compared to shControl tumors (Fig. 6Bi). Increased activin ts the pro le of the shINHA tumors expressing more antiangiogenic proteins as activin has been shown to inhibit angiogenesis 7 which could also be a result of decreased inhibinα thus shifting the balance to increased dimerization of INHBA/B and thereby activin production. Similarly, increases in tumor cell endoglin levels in shINHA tumors in vivo may re ect compensatory responses to changes in inhibin expression consistent with recent reports on endoglin expression changes in ovarian cancers 74 . Whether these changes impact metastasis and angiogenesis and are directly related to changes in inhibin levels in patients remains to be examined. Which of these altered proteins contributes the most to the either pro or anti-angiogeneic tumor microenvironment remains to be determined as we unravel new roles for inhibins herein.
In the mouse host cells where inhibin is likely to interact with endoglin from the endothelia to affect angiogenesis, endoglin levels were slightly higher in shControl receiving hosts that had more vessels compared to shINHA (Fig. 6Dii). These ndings also suggest that blocking inhibin could shift the balance between pro and anti-angiogenic genes.
We also demonstrate for the rst time that anti-inhibin in a therapeutic regimen can reduce tumor growth in vivo (Fig. 6A). The subcutaneous model utilized does not induce ascites formation, unlike the intraperitoneal model used previously, where mice with shINHA tumors produced less ascites than those with shControl tumors 19 . However, this model was chosen as it better allows for evaluation of the vasculature in vivo. Our ndings that inhibin is elevated in patient ascites (Fig. 1E) supports the idea that inhibin may promote ascites formation, likely through increased vascular permeability. The effectiveness of anti-angiogenic therapies is attributed to increased vascular normalization resulting in reduced intratumoral hypoxia, perfused and functional vessels that improve delivery of other chemotherapeutics and enhanced immune response 75 . Resistance to current anti-angiogenic therapies is also common and inhibin A levels have been reported to be increased in patients non-responsive to anti-angiogenic therapy 76 (combination of TRC105 and Bevacizumab) indicating inhibin as a potential alternative mechanism of angiogenesis in tumors resistant to other anti-angiogenic therapies. Further studies exploring the impact of anti-inhibin therapy on the effectiveness of chemotherapeutics and anti-tumor immune response as well is most certainly warranted.

Conclusion
In conclusion, our study shows that targeting inhibin is an effective anti-angiogenic strategy. We demonstrate for the rst time a contextual mechanism for the regulation of inhibin directly driven by hypoxia and HIF-1 and fully de ne inhibin's contributions to hypoxia induced angiogenesis. Based on our ndings and the previously known physiological functions of inhibin, we speculate that targeting inhibin may have potential improved therapeutic value in post-menopausal cancers including a signi cant percentage of ovarian cancers.    corresponding shControl normoxia. Mean SEM, (n=4). n.s.,not signi cant; *, p<.05; **, p<.01, unpaired ttest. C) (i) Representative western blot from HEK293 lysates con rming HIF-1 and HIF-2 knockdown. (ii) qRT-PCR analysis of INHA and VEGF mRNA expression in HEK293 cells transfected with either siScr, siHIF-1 , or siHIF-2 and exposed to hypoxia (0.2% O2) for 24hrs. Mean SEM, (n=4) *, p<.05; **, p<.01, Twoway ANOVA followed by Tukeys multiple comparison test. D) qRT-PCR analysis of primers that amplify a proximal HRE region in the INHA promoter (Methods and Supp Fig. 3A) after ChIP of HIF-1 with HIF-1 antibody (Methods) in OVCAR5 and OV90 cells. ChIP qRT-PCR results were quanti ed as normalized enrichment over IgG and normalized to normoxia. Mean SEM, OVCAR5 (n=3), OV90 (n=2). n.s.,not signi cant; *, p<.05; **, p<.01, Two-way ANOVA followed by Fishers LSD test E) Luciferase activity of HEK293 cells transfected with pGL4.10 luciferase reporter containing a 547bp piece of the INHA promoter and SV-40 Renilla control vector. Cells were either (i) exposed to hypoxia (0.2% O2) or (ii) co-transfected with HIF-1 overexpression plasmid (HIF-1 ODD) and luciferase activity was measured as described in Methods and normalized to either normoxia or PCDNA3.1. Mean SEM, n=3 (Hypoxia), n=2 (HIF-1ODD) *, p<.05; **, p<.01, unpaired t-test. Inhibin increases hypoxia induced angiogenesis and endothelial cell migration and permeability in vivo and in vitro respectively. A) (i) Hemoglobin content in Matrigel plugs collected 12 days after subcutaneous injection of HEY conditioned media collected from cells exposed to normoxia or hypoxia for 24hrs and mixed with either 2 g of IgG or of anti-inhibin R1 antibody. Mean SEM, n=6 plugs per condition. n.s., not signi cant; ***, p<.001, One-way ANOVA followed by Tukey's multiple comparison test.  Inhibin promotes endothelial cell permeability via ALK1 and endoglin and speci cally increases ALK1endoglin cell surface complexes while reducing ALK4-endoglin complexes. A) Quantitation of endothelial cell permeability by measuring FITC-dextran changes across a HMEC-1 monolayer treated with 1nM inhibin A in the presence or absence of (i) 100 g/mL TRC-105 or (ii) 10ng/mL ALK1-Fc. FITC-dextran diffusion across the HMEC-1 monolayer at 4hrs is presented. Mean SD, n=4 for i and n=3 for ii. n.s., not Hypoxia regulated inhibin promotes tumor growth and regulates vascular normalization in vivo. A) Growth curves of subcutaneously implanted HEY shControl or shINHA tumors exposed to either normoxia or hypoxia (0.2% O2) 24hrs prior to injection. 10mg/kg R1 antibody or vehicle control was intraperitoneally injected three times a week. Mean SEM, n=10 for vehicle and n=6 for R1 receiving groups. **, p<.01; ****, p<.0001, Two-way ANOVA followed by Tukey's multiple comparison test. B) Fold change of proteins most altered in shControl and shINHA tumors (Fig 6A) using the (i) human or (ii)