Pro-inflammatory cytokines activate hypoxia-inducible factor 3α via epigenetic changes in mesenchymal stromal/stem cells

Human mesenchymal stromal/stem cells (hMSCs) emerged as a promising therapeutic tool for ischemic disorders, due to their ability to regenerate damaged tissues, promote angiogenesis and reduce inflammation, leading to encouraging, but still limited results. The outcomes in clinical trials exploring hMSC therapy are influenced by low cell retention and survival in affected tissues, partially influenced by lesion’s microenvironment, where low oxygen conditions (i.e. hypoxia) and inflammation coexist. Hypoxia and inflammation are pathophysiological stresses, sharing common activators, such as hypoxia-inducible factors (HIFs) and NF-κB. HIF1α and HIF2α respond essentially to hypoxia, activating pathways involved in tissue repair. Little is known about the regulation of HIF3α. Here we investigated the role of HIF3α in vitro and in vivo. Human MSCs expressed HIF3α, differentially regulated by pro-inflammatory cytokines in an oxygen-independent manner, a novel and still uncharacterized mechanism, where NF-κB is critical for its expression. We investigated if epigenetic modifications are involved in HIF3α expression by methylation-specific PCR and histone modifications. Robust hypermethylation of histone H3 was observed across HIF3A locus driven by pro-inflammatory cytokines. Experiments in a murine model of arteriotomy highlighted the activation of Hif3α expression in infiltrated inflammatory cells, suggesting a new role for Hif3α in inflammation in vivo.


Results
HIF3α is regulated by cytokines in hMSCs in an oxygen-independent manner. Human mesenchymal stem cells (hMSCs) were cultured under standard oxygen conditions or hypoxic conditions for 24 hours and HIF3α protein was assessed by immunofluorescence.
Hypoxia was obtained either by Gas-Pak method, used to achieve 1% oxygen tension (H) or by the addition of cobalt chloride (CoCl 2 250 μM), a widely used hypoxia mimicker 26 . HIF1α expression, known to be expressed under hypoxia 7 , was used as positive control. HIF-1α was absent in standard oxygen conditions (N), but was nicely accumulated either in 1% O 2 (H) or in CoCl 2 -induced hypoxia (CoCl 2 ) (Suppl. Fig. 1a). To compare the effects of 1% O 2 and CoCl 2 -induced hypoxia, we investigated also whether known hypoxia target genes were indeed activated. Expression analysis showed a significant induction of FLT1, GLUT1 and KDR mRNAs in hMSCs in 1% O 2 (H) and in CoCl 2 -induced hypoxia (CoCl 2 ) (Suppl. Fig. 1b), indicating that both methods (Gas-Pak vs. CoCl 2 ) gave similar results in term of induction of hypoxia signalling, and therefore we used CoCl 2 -induced hypoxia for our experiments.
In contrast to HIF1α, immunofluorescence (IF) analysis indicated that HIF3α was expressed in hMSCs cultured in normoxic conditions (21% O 2 ) and it is predominantly localized to the cytoplasm of MSC cells (Fig. 1a). HIF3α was induced in CoCl 2 -induced hypoxic conditions and the protein remained into the cytoplasm, with a much weaker nuclear staining (Fig. 1a).
Since HIF proteins are subjected to oxygen-dependent and -independent mechanisms of regulation, we then explored whether HIF3α expression was influenced by a set of inflammatory cytokines. hMSCs were grown in presence of pro-inflammatory (IL6, IFNγ, TNFα, MCP1) and pro-angiogenic (EGF, VEGF) cytokines, and exposed to normoxia and CoCl 2 -induced hypoxia for 24 h. Given the dependence of hMSC proliferation on different cytokines, we first examined cell cycle under these conditions. As shown by flow cytometry analysis, none of the cytokines altered the G 1 phase both in normoxia and in CoCl 2 -induced hypoxia (Suppl. Fig. 2a), although the presence of cytokines had a protective effect on apoptosis compared to control cells (Suppl. Fig. 2b).
IF staining further revealed that HIF3α expression was higher when hMSCs were cultured in presence of cytokines, both in normoxic and in CoCl 2 -induced hypoxic conditions (Fig. 1a). Remarkably, HIF3α was stimulated in presence of IL6, TNFα, MCP1, EGF and VEGF in normoxic conditions, whereas IFNγ had a negative effect. Furthermore, HIF3α showed similar staining pattern among cytokines in hypoxic conditions, in terms of quantity and localization.
To verify that the observed staining was HIF3α-specific, we used an RNA interference approach to silence HIF3α expression. hMSCs were transfected with a pool of HIF3α-specific siRNAs (ID 64344; NM_152795.3, exon 3-5) or with a scramble siRNA as negative control, and HIF3α expression was assessed by IF. A dramatic reduction of HIF3α expression was obtained both in normoxia and hypoxia, confirming the staining specificity (Fig. 1b).
We then looked at HIF3α mRNA expression. Six splicing isoforms with three alternative start sites were identified with different expression patterns in foetal and adult tissues 14,15 . We evaluated the expression levels of the three alternative exons of HIF3α mRNAs (exon 1a for isoforms 2, 4 and 9; exon 1b for isoforms 7 and 8; exon 1c for isoforms 1, Suppl. Fig. 3a) by qRT-PCR 15 . The expression of HIF3α mRNAs, starting from exon 1a and exon 1b, was greatly induced when hMSCs were exposed to cytokines in standard oxygen conditions (N), except IFNγ, paralleling the IF data, whereas expression isoform starting from exon 1c was barely induced (Fig. 1c) CoCl 2 -induced hypoxia, mRNAs starting from exon 1a, 1b and 1c were further induced only in presence of IFNγ, MCP1 and EGF (Fig. 1c).
Taken together, these data suggested that hMSCs expressed HIF3α in standard oxygen conditions, but, more important, the endogenous level of HIF3α is induced by pro-inflammatory cytokines in a oxygen-independent manner and it is localized to the cytoplasm of hMSCs.
HIF3α is regulated by NF-kB in a oxygen-independent manner. Given that HIF3α showed an increased expression in presence of inflammatory cytokines independent from oxygen, we investigated whether NF-κB could be implicated in its activation 27 .
To test this, we treated hMSCs with TCPA-1 28,29 , a compound with the potential to inhibit NF-κB by preventing the degradation of IκBα. We first assessed the IκBα expression in hMSCs grown under normoxic conditions in presence of TCPA-1. As shown in Fig. 2a, hMSCs accumulated IκBα when cultured in presence of cytokines in standard oxygen conditions in presence of TCPA-1, compared to untreated cells, with the exception of TNFα (Fig. 2a). In these conditions, IF analysis revealed that the expression of HIF3α was reduced by TCPA-1, indicating that the cytokines became ineffective in inducing HIF3α (Fig. 2b). In contrast, in CoCl 2 -induced hypoxia, IκBα was not accumulated (Suppl. Fig. 4a), and no effects were seen on HIF3α expression by IF (Suppl. Fig. 4b), probably because hypoxia signalling was prevalent. Thus, NF-κB is seemingly involved in HIF3α activation under standard oxygen conditions. NF-κB binds a large number of genes activated by inflammation in the human genome 30 . Then, we searched for κB sites in promoter sequence of HIF3A using JASPAR software 31 and we found in intron 1a a putative κB binding site (Fig. 2c). Chromatin immunoprecipitation with a specific NF-κB (RelA) antibodies was performed in hMSCs cultured in standard oxygen conditions and stimulated by indicated cytokines. After cytokines stimulation, 2-to 3-fold enrichment of NF-κB binding in this region was observed and the pre-treatment of cells with TCPA-1 reduced this binding, suggesting that HIF3α expression in hMSCs is induced by cytokines via NF-κB (Fig. 2c).

HIF3α expression correlates with promoter methylation and histone modifications.
To gain further insights on HIF3A gene regulation, we examined the methylation status of the three different promoters. We selected regions by the presence of restriction enzyme sites sensitive to methylation (HpaII and MspI). Two CpG islands, defined as DNA sequences with at least 50% GC content over a minimum of 300 bp, were identified on human HIF3A gene, overlapping exon 1a and exon 1c. Our analysis encompassed also the regions identified by high-density DNA methylation array experiments and covering the exon 1b and the intron 1, as depicted in Fig. 3a, and found hypermethylated 25 . Genomic DNA was extracted from hMSCs exposed to pro-inflammatory cytokines in normoxic and CoCl 2 -induced hypoxic conditions and subjected to methylation-sensitive-restriction enzyme PCR (MS-PCR) analysis 32 . Amplicons #1 and #2 were efficiently digested both by HpaII and MspI enzymes in presence of cytokines under normoxic and CoCl 2 -induced hypoxic conditions, indicating the absence Figure 2. HIF3α activation is dependent on NF-κB activation: (a) Immunoblotting analysis of IκBα: hMSCs cultured in normoxia and treated with IL6, IFNγ, TNFα, MCP1, EGF and VEGF for 24 h. Cells were pre-treated with TCPA-1 for 1 h before cytokines supplementation. (b) Immunofluorescence analysis of HIF3α protein in hMSCs cultured in standard oxygen conditions (Normoxia) in absence and in presence of indicated cytokines and probed with antibodies against HIF3α. Cells were pre-treated with TCPA-1 for 1 h before cytokines supplementation. Scale bars: 10 μm). (c) Schematic diagram of the HIF3A promoter region, where a putative NF-KB binding site is depicted (black triangle). NFκB (RelA) binding on HIF3A promoter: ChIP analysis was performed in hMSCs cultured in standard oxygen conditions and treated with IL6, IFNγ, TNFα, MCP1, EGF and VEGF for 24 h (black bars), or pre-treated with TCPA-1 for 1 h (grey bars) before adding cytokines. As control, species matched IgG were used. Data obtained by qRT-PCR are expressed as enrichment of chromatinassociated DNA fragments immunoprecipitated by NF-κB antibody compared with input (% Input) and expressed as means ± SEM of two independent experiments performed in triplicate.
SCiEntifiC RepoRts | (2018) 8:5842 | DOI:10.1038/s41598-018-24221-5 of any DNA methylation in these regions (Fig. 3b). In contrast, a peculiar restriction pattern was seen over the intron 1 (amplicon #3), which was methylated in hMSCs grown either in normoxia and in CoCl 2 -induced hypoxia and resulted unmethylated in presence of IL6, IFNγ and EGF in hypoxic conditions, indicating that this modification is reversible. Loss of methylation in this site suggested that pro-inflammatory cytokines are able to activate the expression of HIF3α isoforms starting from exon 1b. In contrast, the region covering exon 1c (amplicon #4) showed cleavage resistance to HpaII in DNA extracted from hMSCs grown in all conditions, indicating a persistent DNA methylation and suggesting that the HIF3α isoform starting from exon 1c is barely transcribed in hMSCs.
Next, we verified whether pro-inflammatory cytokines and/or CoCl 2 -induced hypoxia might induce changes in the chromatin structure around the HIF3α transcription start sites. Chromatin immunoprecipitation assay (ChIP) was performed using specific antibodies against trimethylation of lysine 4 in histone H3 (H3K4me3), mark of active transcription, and trimethylation of lysine 27 in histone H3 (H3K27me3), associated with gene silencing. Six different regions spanning HIF3α gene were analysed on the basis of currently available methylome data (Fig. 4a). When hMSCs were exposed to cytokines and cultured at standard and low oxygen tension, we found a global enrichment of H3K4me3 marks on all regions analysed and low H3K27me3 levels, compared to untreated hMSCs (Fig. 4b).
Notably, an enrichment of H3K27me3 was observed when hMSCs were exposed to VEGF both in normoxia and hypoxia across the region chIP1, 2 4 and 5, possibly indicating that these regions are paused and might be activated by additional signals. (Fig. 4b).
In conclusion, our data support the hypothesis that the HIF3α locus is sensitive to inflammatory cytokines that are able to modify the epigenetic profile independent from oxygen conditions. Immunohistochemical analysis of Hif3α in a murine model of arteriotomy. We then investigated the relevance of the in vitro data to an acute pathophysiological condition in vivo. In more detail, Hif3α activation was assessed in a well-established model of rat carotid arteriotomy, in which vascular damage triggers a cascade of events, where inflammation and hypoxia coexist 33 . Rat common carotid arteries were subjected to surgical arteriotomy and animals were administered via tail vein with allogenic MSCs or with DMEM soon after the vascular injury. Our previous studies demonstrated the efficacy of MSC administration in limiting the inflammation triggered by arteriotomy 5 , and consequently this model of acute injury has been considered advantageous and potentially informative in the setting of Hif3α research. Arteriotomy-injured rat carotids were harvested 7 days after injury. Injured carotids exhibited a decreased lumen area (Fig. 5b) compared to controls (Fig. 5a), resulting mainly from negative remodelling and thickening of neoadventitia close to the polypropylene stitch, causing a lumen reduction due to ab-estrinseco compression of the artery 33 . The induction of hypoxia at the arteriotomy site was verified by HIF1α expression (Suppl. Fig. 5c-e). Immunohistochemical analysis (IHC) revealed that HIF1α was expressed only in sporadic endothelial cells at basal level in uninjured carotids, possibly due to hypoxia occurring during the surgical procedure for carotid removal (Suppl. Fig. 5c). Conversely, HIF1α expression was markedly increased 7 days after injury in the carotid region proximal to the arteriotomy site, with particular reference to infiltrating cells in the adventitia, and to intimal and adventitial endothelial cells in vasa vasorum (Suppl. Fig. 5e). In the region distal to the arteriotomy site within the same carotid cross-section, adventitial vasa vasorum and infiltrating cells were negative and HIF1α expression was limited to sporadic endothelial and smooth muscle cells (Suppl. Fig. 5d).
We then carried out IHC analysis to assess whether neoadventia relied also on infiltration of inflammatory cells. Cd45, a general marker of leukocytes, is expressed in infiltrating round-shaped cells in neoadventitia (infiltrated leukocytes), and in fibrocyte-like cells in outer media of injured carotids (Fig. 5h). On adjacent carotid cross-sections, Hif3α was markedly expressed in the same round-shaped cells in neoadventitia tissue (Fig. 5d), suggesting a consistent activation of this factor in inflammatory cells. The absence of staining in negative controls of immunohistochemical reactions confirmed the specificity of the assay ( Fig. 5f and l), suggesting that Hif3α expression is inducible upon inflammation.
The same IHC analysis performed in MSC-treated rats submitted to carotid arteriotomy confirmed Hif3α activation in infiltrating cells (Fig. 5e), suggesting that the MSC immunomodulatory role did not inhibit Hif3α expression. Noticeable, Hif3α was expressed also in perivascular cells and in adventitial endothelial cells in vasa vasorum of arteriotomy-injured carotids in MSC-treated rats (Fig. 5e, small inset), suggesting its possible involvement in tissue damage recovery, whereas it was absent in adventitial endothelial cells in vasa vasorum of arteriotomy-injured carotids in DMEM-treated rats (Fig. 5d). All together, our data showed that Hif3α is expressed in inflamed tissue and might contribute to the repair process.

Discussion
Low oxygen concentration within a tissue activates a compensatory mechanism to counteract the detrimental effects of lack of oxygen and nutrients. Concomitantly, it induces an inflammatory status, and several studies have pointed to inflammation as an essential process to activate the repair process during which inflammatory cells infiltrate the hypoxic tissue, produce different cytokines to sustain inflammation, and activate neoangiogenesis and cell metabolism 27 . HIFs and NF-κB play a crucial role in this mechanism, activating the expression of genes involved in neoangiogenesis [9][10][11]34 , and pro-inflammatory factors 35 , indicating a tight biochemical and functional relationship between hypoxia and inflammation. Our data indicate a role for HIF3α in the early response to inflammation, and activate neoangiogenesis to repair the damaged tissue.
Up to now, little is known about HIF3α. This is in part due to the existence of multiple isoforms; according to Ensembl database, the human Hif3A has 19 transcripts in total, of which only 6 experimentally confirmed with different functions. Some HIF3α isoforms suppress HIF1α and HIF2α expression, while others inhibit HIF1α activity in a dominant negative fashion 14,15,18,19 . Recent transcriptomic data in zebrafish suggested that Hif3α activates the inflammatory response 21 .
In this study, we report for the first time that human HIF3α expression is modulated by pro-inflammatory cytokines in vitro, and is associated with inflammation and neoangiogenesis in vivo, providing a direct involvement of HIF3α in repair process.
Multiple lines of in vitro evidence support our conclusions: (a) IF data showed that HIF3α increases when hMSCs are cultured in standard oxygen conditions and exposed to pro-inflammatory cytokines; (b) Hypoxia further induced HIF3α accumulation, as for other HIF family members. Moreover, IF staining showed a peculiar cytoplasmic localization for HIF3α, a protein supposed to be a transcription factor; (c) HIF3α accumulation is regulated by NF-κB in oxygen-independent manner: hMSC treatment with TCPA-1, a IκB inhibitor able to sequester NF-κB in the cytoplasm, reduces HIF3α activation; (d) chromatin immunoprecipitation reveals that NF-κB binds the HIF3A intronic region and pre-treatment with TCPA-1 reduces its binding; (e) exposure of hMSCs to cytokines modifies the methylation status of HIF3A promoter regions with a subsequent enrichment of the H3K4me3 marks, indicating that cytokines are able to open the HIF3A locus; (f) in a murine model of carotid arteriotomy, Hif3α is activated in vivo and is expressed in infiltrating inflammatory cells of damaged tissue and in vasa vasorum endothelial cells in repairing process. Furthermore, Hif3α protein localisation was slightly different from Hif1α in arteriotomy-injured carotids, with a prevalence of Hif3α expression in adventitia, whereas Hif1α is mainly accumulated in the intima, reflecting a gradient of expression between the two factors.
The new findings suggest that HIF3α gene is sensitive to inflammatory cytokines and is involved not only in inflammation, but also in neoangiogenesis associated with injury resolution and tissue repair. Infiltration of inflammatory cells producing proangiogenic factors is one of first event that then stimulate neoangiogenesis in the context of inflamed and hypoxic tissue. Our data showed that HIF3α is expressed in the infiltrating inflammatory cells, suggesting the possibility that it functions as sensor of inflammation/hypoxia in injured tissue.

Conclusions
Inflammation and neoangiogenesis are both involve in tissue repair. Our data revealed that human HIF3α expression is regulated by inflammatory cytokines in hMSCs in a oxygen-independent manner, showing a distinct mechanism of regulation, compared to HIF1α. Furthermore, the expression is in part under the control of NF-κB. In addition, to the best of our knowledge, the expression of Hif3α in both inflammatory and endothelial cells has been revealed for the first time in vivo in a murine model of vascular restenosis, suggesting that Hif3α might play a role in the activation of tissue repair and its acceleration in presence of allogenic MSCs, endowed with immunomodulatory and secretory properties.
Despite the progress, a lot of questions remain open. The human Hif3A has 19 transcripts in total, of which only 6 experimentally validated in different mammalian cells, including cancer cells. Have these isoforms a cell-context function? What is the role of HIF3α in immune cells? Is HIF3α important for the interaction of different immunomodulatory cells? In humans with high BMI, often associated with metabolic syndrome characterized by chronic inflammation, is HIF3α silenced pointing out a role in metabolic surveillance? In cancer cells, is HIF3α involved in neoangiogenesis to sustain tumour growth? or induces metastasis by altering the cytoskeleton of cancer cells 36 ?
Additional studies will be necessary to clarify the exact role of HIF3α in different normal and cancer cells and in different growth conditions; the newly CRISPR/CAS technology will help us to better define its role.
We believe that any progress toward the comprehension of this progress will have important clinical implications for treatment of ischemic disorders and cancer.

Methods
hMSC culture and treatment. hMSCs were obtained and characterized, as previously described 37 . Cells were grown in RPMI-1640 medium (Euroclone SPA, Italy), containing 10% heat-inactivated foetal bovine serum (FBS), 1% Penicillin-streptomycin, and 1% L-Glutamine, and maintained as monolayers in a humidified atmosphere containing 5% CO 2 at 37 °C. Hypoxic culture conditions were achieved in a BD GasPak EZ Anaerobe Gas Generating Pouch System (BD Biosciences, San Diego). As certified by the manufacturer, the Anaerobe Gas Generating Pouch System produces an atmosphere containing 1% oxygen after 1 h. Hypoxic culture conditions were also obtained adding 250 μM CoCl 2 (Sigma-Aldrich, Saint Louis, MO, USA), a hypoxia mimetic agent, in culturing medium.
Oligonucleotides used in this study are reported in Table 1.
Immunofluorescence analysis. Cells were seeded the day before and grown on coverslips in 12-well

MS-PCR.
The DNA methylation analysis has been performed through the methylation sensitive restriction enzyme-polymerase chain reaction (MS-PCR) assay, a well-established method already applied in different settings 32,39 . Genomic DNA was extracted by lysing cultured hMSCs with 1% SDS followed by proteinase K digestion, ethanol precipitation, and phenol-chloroform purification. MS-PCR was performed on purified genomic DNA (1 μg) that was previously restriction endonuclease-digested for 48 h with the isoschizomers MspI and HpaII (New England Biolabs, USA).
ChIP-qRT-PCR. Chromatin immunoprecipitation was performed as reported 30  Rat carotid arteriotomy and MSC treatment. Arteriotomy of rat common carotid artery was performed as reported 33 . Male Wistar rats (Charles River, France) were anesthetized with i.p. injection of 100 mg/kg ketamine and 0.25 mg/kg medetomidine, and carefully placed onto a warm surface and positioned for surgery. All the surgical procedures were conducted under sterile conditions and vital signs were continuously monitored by pulsioxymeter. A plastic Scanlom clamp for coronary artery grafting was placed for 10 s on the carotid causing a crushing lesion to the vessel. At the same point where the clamp was applied, a 0.5 mm longitudinal incision was made on the full thickness of the carotid. The incision did not cross to the other side of the vessel. Haemostasis was obtained with a single adventitial 8.0-gauge polypropylene stitch. Once bleeding stopped, the carotid artery was carefully examined and blood pulsation was checked distally to the incision. A reabsorbable suture approximated the skin. Animals were allowed to wake up through an intramuscular injection of 1 mg/kg atipamezol. Postoperative systemic analgesia was administered through subcutaneous injection of 0.1 mg/kg buprenorphine every 8 h. Antibiotic therapy was administered through subcutaneous injection of 5 mg/kg enrofloxacin once a day for 3 days following the arteriotomy procedure. Rats submitted to arteriotomy were administered with 5 × 10 6 MSCs suspended in 200 μl DMEM via tail vein injection (n = 5) soon after arteriotomy, while control rats were administered with 200 μl DMEM (n = 5). instructions. Primary antibodies were omitted in negative controls. Nuclei were counterstained with Mayer's haematoxylin (Sigma-Aldrich). Image screening and photography were performed using a Leica microscope IM1000 System.
Animal studies approval. The experiments with animals were performed in compliance with the institutional guidelines and approved by the Local Committee for 'Good animal experimental activities' . The experimental protocol was approved by the Animal Care and Use Committee of Università della Campania "L. Vanvitelli" (1966/7.17.2012). Animal care complied with Italian regulations on protection of animals used for experimental and other scientific purposes (116/1992) as well as with the EU guidelines for the use of experimental animals (2010/63/EU). Mice were housed in the Animal Facility of Università della Campania "L. Vanvitelli".
Statistical analysis. To analyse differences in HIF3α expression/methylation, statistical significance was calculated through a paired two-tailed t-test using the SPSS 17 software (SPSS, Inc. Chicago, IL, USA). Data are expressed as mean ± SEM of three biological replicates. p-values < 0.05 were considered significant.
Data availability. All data generated or analysed during this study are included in this published article (and Suppl. Information files). For layout reasons, some data generated were modified, but raw data are included in Suppl. Information files. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.