Crosstalk Between RPE Cells and Choroidal Endothelial Cells via the ANXA1/FPR2/SHP2/NLRP3 Inflammasome/Pyroptosis Axis Promotes Choroidal Neovascularization

One type of age-related macular degeneration (AMD), neovascular (nAMD), characterized by choroidal neovascularization (CNV), accounts for the majority of the severe central vision impairment associated with AMD. Endothelial cells (ECs) in direct contact with retinal pigment epithelial (RPE) cells are more prone to the pathological angiogenesis involved in CNV. Herein, we investigated the effect of crosstalk between RPE cells and choroidal endothelial cells (CECs) via the ANXA1/FPR2/NLRP3 inflammasome/pyroptosis axis on the development of choroidal neovascularization (CNV) in vitro and in vivo. ANXA1 expression and secretion from ARPE-19 cells were upregulated by hypoxia. FPR2 expression, especially on the plasma membrane, in HCECs was upregulated under hypoxic conditions. ANXA1 secreted from ARPE-19 cells inhibited NLRP3 inflammasome activation and NLRP3 inflammasome-mediated pyroptosis in HCECs by activating the FPR2/SHP2 axis. Moreover, ANXA1 secreted by ARPE-19 cells promoted behaviors of HCECs, including proliferation, migration, and tube formation, by activating the FPR2/SHP2 axis and inhibiting NLRP3 inflammasome-mediated pyroptosis. Inhibiting the upregulated ANXA1/FPR2/SHP2/NLRP3 inflammasome/pyroptosis axis decreased the volume of CNV. Our data suggest that the crosstalk between RPE cells and CECs via the ANXA1/FPR2/NLRP3 inflammasome/pyroptosis axis promotes CNV. This finding could identify a potential target for the prevention and treatment of CNV.


INTRODUCTION
Age-related macular degeneration (AMD), a neurodegenerative disease, acts as is one of the leading causes of vision impairment in elderly individuals worldwide [1]. Advanced AMD is segmented into two types: neovascular AMD (nAMD) and nonneovascular AMD (aAMD). Among them, nAMD is characterized by choroidal Manhui Zhu and Ying Wang contributed equally to this study. 1 neovascularization (CNV) that penetrates Bruch's membrane into the subretinal pigment epithelial space and the subretinal space, resulting in leakage, hemorrhage, edema, pigment epithelial detachment, and subretinal fibrosis [2], accounting for approximately 80% of the severe central vision impairment associated with AMD [3]. Recently, the first-line therapeutic strategy for nAMD is repeat intravitreal injection of anti-vascular endothelial growth factor (VEGF) agents, which can effectively delay the progression of nAMD. However, resistance to anti-VEGF drugs can occur from the beginning of treatment or be induced gradually. For example, the comparison of age-related macular degeneration treatments trials (CATT) reveals that despite monthly treatments with anti-VEGF agents for 2 years, 51.5% of patients receiving intravitreal ranibizumab (RAN) and 67.4% of patients that are treated with bevacizumab (BEV) exhibit evidence of persistent fluid as determined by time-domain optical coherence tomography (OCT) [4]. Anti-VEGF treatment resistance indicates that in addition to VEGF, other factors also contribute to the development of CNV.
Clinicopathological studies show that 50% of sightthreatening nAMD occurs when choroidal endothelial cells (CECs) are induced to migrate toward and make contact with the retinal pigment epithelium (RPE) and its extracellular matrix. After making contact with RPE, CECs can migrate across the RPE into the neurosensory retina, where CNV occurs [5]. VEGF 189 secreted by RPE cells facilitates the migration of CECs across the RPE [6]. Additionally, co-culture of endothelial cells (ECs) and RPE cells under conditions enabling direct EC-RPE cell contact enhances the proangiogenic potential of ECs under normoxic conditions to an extent similar to that induced by hypoxia, suggesting that ECs in direct contact with RPE cells might be more prone to the pathological angiogenesis involved in CNV [7].
Annexin A1 (ANXA1) is a major driver of inflammatory resolution. In vitro studies confirm that ANXA1 levels in RPE cells are upregulated after 60 and 120 min of infection with Toxoplasma gondii [8]. Using a false discovery rate of 20% as the threshold, the researchers identified 1,133 genes, including ANXA1 [log (AMD/ cnt) = − 0.41], in GSE29801 [transcriptome data which is from macular and extramacular RPE-choroid tissue isolated from AMD patients with Rotterdam grades of 2a, 2b, or 3 and from individuals with no features of AMD (controls)], and these genes are dysregulated in macular RPE-choroid tissue from AMD patients compared with that of the controls [9]. Additionally, hypoxia induces ANXA1 expression in multiple cancers, such as colon cancer [10] and prostate cancer [11]. Therefore, we hypothesize that RPE cells release increased ANXA1 under hypoxic conditions during CNV.
Secreted ANXA1 exerts the inflammation resolution function by binding to its receptor formyl peptide receptor 2 (FPR2) [12], which is a G protein-coupled receptor (GPCR). Neither formyl peptide receptor 1 (FPR1) nor formyl peptide receptor 3 (FPR3) is affected by ANXA1, except FPR2 protein levels, which are increased in mouse laser-induced 15 d CNV lesions [13]. After ANXA1 binds to FPR2, ANXA1 interacts with SH2 domain-containing protein tyrosine phosphatase (SHP2), a ubiquitously existing non-receptor protein tyrosine phosphatase [14]. Inflammasomes are multimolecular signaling complexes that have crucial roles in host defense against various autoimmune and inflammatory disorders. Among the numerous types of inflammasomes that have been determined, the Nod-like receptor family pyrin domain containing 3 (NLRP3) inflammasome is the most extensively studied type owing to its mighty activation in response to multiple stimuli, including hypoxia [15]. Following activation, the NLRP3 inflammasome brings about caspase 1-dependent secretion of the pro-inflammatory cytokines interleukin-1β (IL-1β) and interleukin-18 (IL-18), accompanied with gasdermin D (GSDMD)-assisted cell death named as pyroptosis. GSDMD contains an amino-terminal cell death domain (GSDMD Nterm ), a central short linker region, and a carboxy-terminal autoinhibition domain. Caspase 1 cleaves GSDMD, dislodging its carboxyl terminus and discharging it from intramolecular inhibition [16]. GSDMD Nterm then binds to phosphatidylinositol phosphates and phosphatidylserine locating in the cell membrane inner leaflet, initiating pyroptosis [17]. SHP2 negatively modulates NLRP3 inflammasome activation in macrophages, consequently alleviating NLRP3 inflammasome-mediated pyroptosis [18]. Actually, recent studies find that the activation of NLRP3 inflammasome happens in human AMD and facilitates AMD-like pathologies in animal models as well [19,20]. The researchers test the role of IL-18 in a novel genetic mouse model of nAMD, Vegfa hyper mice, in which increased VEGF-A levels in the RPE lead to increased inflammasome activation products in RPE/choroids of these mice as well as in spontaneous CNV formation in an age-dependent manner without experimental injury [21]. In addition, inactivation of IL-18 does not obviously affect CNV lesion formation in Vegfa hyper mice, whereas inactivation of NLRP3, caspase-1, or interleukin-1 receptor (IL-1R) potently inhibits the formation of CNV lesion in these mice [21,22]. Based on the previous data, we sought to resolve the problems whether ANXA1 released by RPE cells could interact with FPR2 on the surface of HCECs to inhibit the activation of NLRP3 inflammasome inside HCECs, subsequently effecting on the pathogenesis of CNV formation.
Herein, we investigated the effect of crosstalk between RPE cells and CECs via the ANXA1/FPR2/ NLRP3 inflammasome/pyroptosis axis on CNV. Our data further clarify the crosstalk between RPE cells and CECs during CNV and hint potential targets for the more effective treatment of CNV.

Western Blot
The cells were cultured to 70% confluence and subjected to the designated treatments. The cells were lysed at 4 °C using radioimmunoprecipitation assay (RIPA) lysis buffer (R0278, Sigma Aldrich, USA). Protein concentrations were determined using a BCA protein assay kit (A53225, Thermo Fisher Scientific, USA). Samples (80 µg protein) were resolved by 4-20% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE The membranes were washed three times with phosphatebuffered saline with Tween detergent (PBST) and incubated with horseradish peroxidase (HRP)-conjugated goat-anti-rabbit (SA00001-2, Proteintech; 1:5000) or HRP-conjugated goat-anti-mouse antibodies (SA00001-1, Proteintech; 1:5000) at room temperature for 2 h. GAPDH (600,004-1-IG, Proteintech; 1:50,000) and caveolin 1 (Cav1; 16,447-1-AP, Proteintech) were used as loading controls. Unless otherwise indicated, the dilution for each antibody was 1:1000. Following an additional 10 min of washing with PBST, the protein bands were visualized using an electrochemiluminescence (ECL) assay kit (JP001B250, CLINX, China). The density of each band was measured using ImageJ software. The expression level of each target protein was normalized to the GAPDH or Cav1 protein expression level and compared with that of the normal group, which was assigned a value of 1.

Enzyme-Linked Immunosorbent Assay (ELISA)
The levels of ANXA1 were measured using a human ANXA1 ELISA kit (NBP2-60,538, Novus Biologicals, USA) according to the manufacturer's instructions. The detection range of the kit was 0.313-20 ng/ml. The absorbance was measured using a microplate reader (Bio-Rad Laboratories, USA) at a wavelength of 450 nm and a correction at 655 nm.

Plasma Membrane and Cytosol Isolation
Plasma membrane and cytosolic fractions were isolated from HCECs using a plasma membrane protein extraction kit (ab65400, Abcam, USA).

Terminal Deoxynucleotidyl Transferase dUTP Nick-End Labeling (TUNEL) Assay
HCECs were seeded on glass coverslips in 24-well plates. After different treatments for 24 h, the TUNEL assay was done using an in situ cell death detection kit (11,684,795,910,Roche), and the cells were observed under a TCS-SP2 confocal microscope (Leica Microsystems, Germany).

5-Ethynyl-2′-Deoxyuridine (EdU) Incorporation Assay
Cell proliferation was analyzed using the EdU Apollo 488 assay kit (C10310-1, RiboBio, China). The images were viewed using a confocal microscope at 100 × magnification. The number of stained nuclei was counted and used to determine the percentage of the total number of nuclei in each image.

Wound Healing Assay
Confluent monolayers of HCECs in 6-well plates were scratched with pipet tips, leading to one acellular 0.5-mm-wide lane per well. After the cells were washed twice, the HCECs were exposed to different treatments. Photographs were taken after 24 h of incubation. The amount of migration was determined by ImageJ software.

Tube Formation Assay
A 24-well culture plate was coated with 100 μl of 10 mg/ml precooled growth factor-reduced Matrigel (354,230, Corning, USA). HCECs were seeded on Matrigel-coated plates at concentrations of 5 × 10 4 cells per well and exposed to different treatments. After 24 h of culture, images were acquired using an optical microscope (× 100 magnification). The average length of the tube branches in four areas of each sample was assessed under an inverted light microscope.

Mouse Laser-Induced CNV Model
After anesthesia and pupil dilation, male C57BL/6 J mice (purchased from the Laboratory Animal Center of Soochow University) (aged 10 weeks) were subjected to laser photocoagulation using a PASCAL diode ophthalmic laser system (neodymiumdoped yttrium aluminum garnet [Nd:YAG], 532 nm; Topcon Medical Laser Systems, Santa Clara, USA) with the following parameters including 200 mm spot size, 0.02 s duration, and 100 mW power. Four laser spots were made around the optic nerve heads of both eyes in each mouse to induce CNV. The rupture of Bruch's membrane was verified by observing a bubble at each laser spot. Approval was obtained from the Animal Research Ethics Committee of Soochow University in agreement with the Chinese National Standard.

Statistical Analysis
All results are represented as the means ± SEM. Statistical significance between groups was evaluated with Student's unpaired t-tests (two-tailed). A value of P < 0.05 was considered to be statistically significant.

ANXA1 Expression and Secretion from ARPE-19 Cells Are Upregulated by Hypoxia
Ocular Toxoplasma gondii infection contributes to a significant increase in ANXA1 expression in the mouse RPE at 48 h and 72 h [8]. First, we measured ANXA1 protein levels in hypoxia-exposed ARPE-19 cells at different time points and found that ANXA1 protein levels increased in ARPE-19 cells in response to hypoxic conditions at 16 h, peaked at 24 h, and then declined (Fig. 1A and B). Moreover, ANXA1 secreted by ARPE-19 cells under hypoxic conditions was measured by ELISA and showed a similar tendency (Fig. 1C). These data suggested that ANXA1 expression and secretion from ARPE-19 cells were upregulated by hypoxia.

FPR2 Expression, Especially on the Plasma Membrane, in HCECs is Upregulated Under Hypoxic Conditions
ANXA1 binds to FPR2 to exert an anti-inflammatory effect [25], and FPR2 is expressed in multiple types of ECs, including human retinal microvascular ECs [26]. Based on previous studies, we hypothesize that HCECs express FPR2, which binds to ANXA1 produced by RPE cells under hypoxic conditions. As expected, FPR2 protein levels in normal HCECs were low. After hypoxia exposure, FPR2 protein levels increased at 16 h, reaching a peak at 24 h ( Fig. 2A and B). Additionally, following culture under hypoxic conditions for 24 h, HCEC FPR2 expression on the plasma membrane but not in the cytosol increased (Fig. 2C). These results suggested that FPR2 expression, especially on the plasma membrane of HCECs, was upregulated by hypoxia.

ANXA1 Secreted by ARPE-19 Cells Inhibits NLRP3 Inflammasome Activation in HCECs by Activating the FPR2/SHP2 Axis
The ANXA1 tripeptide suppresses NLRP3 inflammasome activation, thus mitigating microglial activation The relative protein level of ANXA1 was analyzed. C The protein level of ANXA1 in ARPE-19 cell culture supernatant was measured by ELISA. In Fig. 1B and C, * P < 0.05 and ** P < 0.01 versus the 0 h hypoxia group. n = 4/each group. and hippocampal-dependent memory deficits [27]. In addition, FPR2 promotes SHP2 activation [28]. Therefore, whether ANXA1 interacted with FPR2 to activate SHP2 and subsequently inhibited NLRP3 inflammasome activation was examined. Hypoxia induced SHP2 phosphorylation compared to that of normal (normoxic) conditions. Human ANXA1 recombinant protein also induced the phosphorylation of SHP2, while the FPR2 antagonist WRW4 and SHP2 inhibitor SHP099 impaired the positive effect of ANXA1. CCM from ARPE-19 cells cultured under hypoxic conditions for 24 h also enhanced the phosphorylation of SHP2, while the ANXA1 neutralizing antibody WRW4 or SHP099 inhibited the effect of CCM. Except in the normoxia and hypoxia groups, the NLRP3 inflammasome-associated molecules ASC, NLRP3, and IL-1β showed the opposite tendencies of p-SHP (Fig. 3A-E). Pro-IL-1β showed few changes in each group. These data suggested that ANXA1 secreted by ARPE-19 cells inhibited NLRP3 inflammasome activation in HCECs by activating the FPR2/SHP2 axis.

ANXA1 Secreted by ARPE-19 Cells Inhibits NLRP3 Inflammasome-Mediated Pyroptosis in HCECs by Activating the FPR2/SHP2 Axis
Next, the effect of the ANXA1/FPR2/SHP2 axis on NLRP3 inflammasome-mediated pyroptosis in HCECs was examined. N-GSDMD and cleaved caspase-1 were upregulated in the hypoxia group compared to the normoxia group. The administration of ANXA1 protein decreased N-GSDMD and cleaved caspase-1 protein levels, while WRW4, SHP099, the NLRP3 inflammasome agonist ATP, or caspase-1 activation plasmid reversed the effect of ANXA1. Similarly, CCM downregulated N-GSDMD and cleaved caspase-1 protein levels, and WRW4, ANXA1 neutralizing The relative protein level of FPR2 was analyzed. * P < 0.05 and ** P < 0.01 versus the 0 h hypoxia group. C Following normoxic or hypoxic culture for 24 h, the plasma membrane and cytosolic fractions of HCECs were isolated for analysis. Lysates were prepared, and Western blot was used to measure FPR2 protein levels. D The relative protein level of FPR2 in the plasma membrane and cytosol was analyzed. ** P < 0.01 versus the normal group. The ns indicated no significance. n = 4/each group. antibody, SHP099, ATP, or caspase-1 activation plasmid inhibited the effects of CCM (Fig. 4A-C). Moreover, the ratio of TUNEL-positive HCECs showed a similar tendency as the N-GSDMD and cleaved caspase-1 protein levels (Fig. 4D-E). Therefore, these results suggested that ANXA1 secreted by ARPE-19 cells inhibits NLRP3 inflammasome-mediated pyroptosis in HCECs by activating the FPR2/SHP2 axis.

DISCUSSION
First, we found that ANXA1 expression and secretion from hypoxia-exposed ARPE-19 cells were higher than those in normal (hypoxia for 0 h) cells. A previous study reveals that ANXA1 is upregulated under low-oxygen conditions by hypoxia-inducible factor (HIF-1) overexpression and not by binding to a hypoxia-responsive element (HRE; -430 5′-CACCT-3′ -426) in the ANXA1 promoter [29]. However, another study shows that the loss of HIF-1α can retain tumor metabolism and proliferation by boosting ANXA1, and these proteins are expressed in a mutually exclusive way in 37 gastric cancer (GC) cell lines. ANXA1 is a crucial protein in the downstream pathway regulated by HIF-1α under hypoxic conditions [30]. Whether ANXA1 is upregulated by HIF-1α at the transcriptional level requires further examination. ANXA1 promotes cardiac macrophages to release high level of VEGF-A (VEGF is commonly referred to VEGF-A), hence triggering cardiac neovascularization and repair [31]. Moreover, ANXA1 promotes EC migration and angiogenesis [32]. The proangiogenic effects of ANXA1 can be explained by the signal amplifier role of ANXA1 to boost the release of second messengers that impact cellular proliferation and migration [33]. In our study, we found that ANXA1 promoted proliferation, migration, and tube formation of HCECs by downregulating downstream NLRP3 inflammasome activation and pyroptosis, indicating that the proangiogenic effects of ANXA1 are associated with inhibiting inflammation.
NLRP3 inflammasome activation in non-RPE cells, but not in RPE cells, promotes CNV in an nAMD mouse model [38]. However, NNLRP3 inflammasome activation occurs in macrophages and results in IL-18 activation, which inhibits laser-induced CNV [39]. The difference in these conclusions may be derived from the different animal models used in the two studies: the former uses a genetic model of nAMD Vegfa hyper mice, and the latter uses a laser-induced CNV mouse model. In our study, we find that SHP2 inhibits NLRP3 inflammasome activation in HCECs to exacerbate the formation of laser-induced CNV in mice, while the SHP2 inhibitor SHP099 alleviates CNV.
In summary, we elucidated a mechanism by which the crosstalk between RPE cells and CECs via the ANXA1/FPR2/SHP2/NLRP3 inflammasome/pyroptosis axis promotes CNV. There were several limitations in our study worth mentioning, such as the lack of primary cell experiments and the absence of FPR1 and FPR3 detection in HCECs. However, our identification of crosstalk may facilitate the manipulation of FPR2 expression to develop therapeutics for CNV.

DATA AVAILABILITY
Not applicable.

DECLARATIONS
Ethics Approval All mice studies were conducted according to protocols approved by the Animal Ethics Committee of Soochow University.

Consent for Publication
All authors have reviewed the manuscript and have given consent for publication.