KAI1(CD82) is a key molecule to control angiogenesis and switch angiogenic milieu to quiescent state

Adult C57BL/6 mice (6–12 weeks, male or female) were purchased from The Jackson Laboratory (Bar Harbor, ME). All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at Seoul National University Hospital and complied with the National Research Council (NRC) ‘Guidelines for the Care and Use of Laboratory Animals’.

Eyeballs including the optic nerve were collected from the mice at postnatal day 2 and 5. The retinae were stained with Lectin from Bandeiraea simplicifolia fluorescein isothiocyanate (FITC) conjugate (catalog no. L9381; Sigma-Aldrich), anti-CD31 phycoerythrin (PE) conjugate (catalog no. 340297; BD Biosciences), anti-NG2 Alexa 488 conjugate (catalog no. AB5320A4; Merck Millipore), anti-NG2 cy3 conjugate (catalog no. AB5320C3; Merck Millipore), anti-collagen IV antibody (catalog no. AB748; Merck Millipore), and anti-GFP (catalog no. A6455; Thermo Fisher Scientific). MS1 and 10T1/2 cells were fixed in 2% cold paraformaldehyde (Wako) for 10 min on ice. After washing with phosphate-buffered saline (PBS), the samples were blocked with 1% bovine serum albumin and 0.1% Triton X-100, followed by staining with anti-KAI1 (catalog no. SC-1087; Santa Cruz), anti-CD9 (catalog no. SC-13118; Santa Cruz), anti-alpha-smooth muscle actin Cy3 conjugate (catalog no. C6198; Sigma-Aldrich), anti-CD31 PE conjugate (BD Biosciences), and Bandeiraea simplicifolia fluorescein isothiocyanate conjugate (Sigma-Aldrich). After staining, the tissue and cells were examined by confocal microscopy (Zeiss).

Total RNAs were separated and purified and cell harvesting at representative time points. RNeasy® mini kit (catalog no. 74104; QIAGEN) and QIAshredder (catalog no. 79654; QIAGEN) were used to separate and purify RNA from cells. To synthesis the cDNA from RNA, we used qPCR RT master mix from Toyobo (catalog no. FSQ-201; TOYOBO). Real-time PCR was performed with SyBR Green I Mastermix (Applied Biosystems) using an ABI PRISM TM 7500 Sequence Detection System (Applied Biosystems). RT-PCR was performed using the S1000 Thermal Cycler (BIO-RAD) and detected the agarose gel with Gel Doc XR + Gel Documentation System (BIO-RAD). Information on the primers used is provided in the Supplemental Materials.

Cells were lysed with lysis buffer (catalog no. #9803s; Cell Signaling Technology) containing protease inhibitor cocktail (catalog no. #3100-001; Gendepot). Total protein was immunoblotted with primary antibodies against KAI1 (catalog no. ab135779; Abcam), LIF (catalog no. AB-449-NA; R&D Systems), tSrc, pSrc (Tyr⁴¹⁶), and pSrc (Tyr⁵²⁷) (Cell Signaling Technology), Myc (catalog no. 05-724; Sigma-Aldrich), p53 (catalog no. MABE327; Merck Millipore), Pbx1 (catalog no. sc-889; Santa Cruz), Cav-1 (Caveolin-1, catalog no. sc-894; Santa Cruz), Flot1 (Flotillin1, catalog no. #3253s; Cell signaling Technology), p-VEGFR2 (catalog no. ab38473; Abcam), total VEGFR2 (catalog no. #2479s; Cell Signaling Technology), p-PDGFRβ (catalog no. #3166s; Cell signaling Technology), total PDGFRβ (catalog no. #4564s; Cell signaling Technology), PDGF-BB (catolg no. 07-1437; Sigma-Aldrich), and VEGF-A (catalog no. #ABS82; Merck Millipore) followed by incubation in horseradish peroxidase-conjugated secondary antibodies (Jackson Laboratory). GAPDH (catalog no. ab9485; Abcam) and beta-actin (catalog no. ab8227; Abcam) were used as an internal control.

Cells were collected in FACS buffer [dPBS (Thermo Fisher Scientific) containing 1% FBS and 0.1% bovine serum albumin (Amresco)]. Cells were stained with antibodies, anti-CD82 (Miltenyi Biotec.), CD31 (Santa Cruz), and PDGFRβ (R&D Systems) for 30 min on ice washed out with FACS buffer after staining and analyzed with a BD FACSCanto II™ (BD Biosciences), followed by sorting using a BD FACSAria™ (BD Biosciences) instrument.

Adenovirus carrying mouse KAI1 was created with the pAdEasy vector system and titrated using the plaque-forming unit assay. 10T1/2 and MS-1 cells were transduced at a multiplicity of infection of 500 and 1000 pfu (plaque-forming unit), respectively, with 1 μg/mL polybrene (Sigma-Aldrich). For lentivirus transduction, MS1 cells were transduced by GFP encoded lentivirus using polybrene.

For the Matrigel two-dimensional tube formation assay, confocal dishes (Ibidi) were coated with 80 μL matrigel (Corning) and incubated for 30 min in a 37 °C incubator for polymerization. For EC-pericyte co-culture experiments, 3 × 10⁴ MS1 (EC) and 1 × 10⁴ WT/Kai1 K/O mouse primary PVCs were seeded onto the polymerized GFR-matrigel with EBM 5% media. For the EC-conditioned medium set, 3 × 10⁴ MS1 cells were seeded onto a matrigel coated µ-Dish (Ibidi) with conditioned media of primary PVCs from WT or Kai1 K/O mouse. For the three-dimensional spheroid sprouting assay, spheroids of PVCs, ECs, and B16 cells combinations were generated using the hanging drop method. Spheroids seeded onto polymerized matrigel. For the tube formation assay using an overconfluent PVC monolayer, a high density of 10T1/2 cells was incubated for 5 days in growth media and GFP expressing-ECs (MS1) were co-cultured with the PVC monolayer for 3 days.

Tube formation images were acquired by fluorescence microscopy and using a Zeiss LSM-710 META confocal microscope. Random fields were measured in terms of tube length and number of branching points by ImageJ software (National Institutes of Health).

Lipid raft isolation in mouse PCs (10T1/2) and ECs (MS-1) was performed. For lipid-protein crosslinking, the cells were treated with 1.25 mM DTSSP (catalog no. 21578; Thermo Fisher Scientific) solution and incubated for 1 h on ice at 4 °C. The cells were washed with cold PBS and 5 mM EDTA and then 0.75 mL cold PBS and 5 mM EDTA were added. The samples were frozen overnight at − 80 °C. The cells were lysed with 0.8 mL 0.1% Triton X-100 membrane raft isolation buffer [1 M Tris–HCL, pH7.4, 1 M NaCl, 100 mM EDTA, Triton X-100, DW, protease inhibitor cocktail (500×)] by passing through a 23-gauge needle on a 5-mL syringe 20 times. Next, 500 μL supernatant 1 mL OptiPrep (catalog no. 1114542; Proteogenix) separation medium (60% iodixanol) were added, resulting in a 40% iodixanol solution of lysed cells. Using a Pasteur pipet, the 40% iodixanol solution was carefully overlaid with equal amounts of 30% iodixanol solution, followed by 5% iodixanol solution. The gradient was visible to the naked eye. Samples were ultra-centrifuged for 5 h at 132,000 g and 4 °C. The membrane rafts were present in the second fraction and were visible. Western blotting was performed as described above.

Leukemia inhibitory factor (LIF) concentrations in the conditioned media of adenovirus-transduced MS1 and 10T1/2 cells were determined using a mouse LIF Quantikine ELISA Kit (R&D Systems). 1:2 diluted standard, control, and samples were incubated for 2 h at room temperature followed by a total of five washes. Next, 100 μL of mouse LIF Conjugate was added and incubated for 2 h at room temperature, followed by five washes, and then 100 μL of substrate solution was added and incubated for 30 min at room temperature in the dark. Finally, 100 μL of stop solution was added and the optical density of each sample was determined using a microplate reader (Promega). The concentration was calculated by four-parameter logistic (4-PL) regression.

RNA was extracted from pericytes of WT and Kai1 K/O mice. RNA sequencing reads were aligned to the mouse genome build mm9 (NCBI37) using TopHat 2.0.9 and Bowtie 0.12.9 with a segment-length of 21, which allowed 2 mismatches per read. Expression levels for 23,170 RefSeq genes were measured by reads per kilobase per million mapped reads (RPKM). Differentially expressed genes between conditions were tested by Cuffdiff. Differentially expressed genes were defined as having more than a twofold change and an adjusted p value less than 0.05. Hierarchical clustering of samples was performed by R (www.​R-project.​org). Gene Ontology (GO) analysis was performed (http://​geneontology.​org/​). Gene Set Enrichment Analysis (GSEA) was performed against custom-made lists from the GO database. The RNA-seq data have been deposited in the Gene Expression Omnibus (GEO) database (GSE114465).

Putative p53 and pbx1 binding sites were predicted using the EpiTect ChIP qPCR Primers website (Qiagen, http://​www.​sabiosciences.​com/​chipqpcrsearch.​php?​app=​TFBS&​qs=​1490097319). 10T1/2 cells were transfected with FLAG-tagged mouse KAI1 plasmid (KAI1 plasmid from R&D Systems, catalog no. RDC0331; pCMV-Tag 2B vector carrying FLAG obtained from Agilent Technologies) that had been incubated with polyethylenimine (Sigma-Aldrich) to improve transfection efficiency. For Src inhibition, 5 µM of PP2 (catalog no. P0042; Sigma-Aldrich) was added to the normal growth medium. The cells were fixed in 1% formaldehyde (catalog no. F8775; Sigma-Aldrich) for 10 min at room temperature for crosslinking and collected in a tube. The supernatant was discarded, and ChIP RIPA buffer (50 mM Tris–Cl, 150 mM NaCl, 10 mM Triton X-100, 1 mM EDTA, 0.1% sodium deoxycholate, 0.1% SDS, 1× protease inhibitor cocktail, 1 mM phenylmethane sulfonyl fluoride) was added to the pellet. To shear the DNA, both fractions were sonicated 10 times with a BIORUPTOR (30 s on and 30 s off per cycle; Diagenode). The immunoprecipitated sample was supplemented with a p53 (catalog no. MABE327; Merck Millipore) and pbx1 (catalog no. SC-889; Santa Cruz) antibody (2 µg) and rotated overnight. Protein A/G sepharose (catalog no. ab193262; Abcam) were added to pulldown the antibodies, and the samples were washed sequentially three times with homemade washing buffers 1, 2, and 3 (washing buffer 1: 20 mM Tris–Cl, 140 mM NaCl, 0.5 mM Triton X-100, 0.1 mM EDTA; washing buffer 2: 20 mM Tris–Cl, 500 mM NaCl, 0.5 mM Triton X-100, 0.1 mM EDTA; washing buffer 3: 20 mM Tris–Cl, 250 mM LiCl, 0.5 mM Triton X-100, 0.1 mM EDTA). Samples were incubated overnight at 65 °C for decrosslinking and recovered using a PCR purification kit (Qiagen). Recovered DNA was analyzed by semi-quantitative PCR.

10T1/2 cells were transiently co-transfected with firefly pGL3-control luciferase reporter vector (Promega) and renilla luciferase reporter plasmid pRL-TK (catalog no. E2241; Promega) by using Neon transfection system (Thermo Fisher Scientific). Lif promoter binding affinity-luciferase activity was detected using Dual-Glo Luciferase Assay system (Promega), following the protocol provided by the manufacturer.

The surface plasmon resonance (SPR) binding studies were performed by Woojung BSC Inc. (Seoul, Korea) using a Biacore T200 (GE Healthcare, Sweden) instrument optical biosensor and CM5 chip (GE healthcare, Cat#. BR-1005-30, Sweden). The data were analyzed by using BIAevaluation software version 3.0 (GE Healthcare, Sweden). Real-time data showing the binding of recombinant KAI proteins to angiogenic cytokines was obtained using a Blitz instrument (ForteBio).

Kai1-Gfp fusion vector and Dhhc-Ha vectors were co-transfected into 293T cells using ViaFect (Promega) according to the manufacturer’s instructions. pEF-Bos-zDHHC-HA constructs were a generous gift from Professor Masaki Fukata (National Institute for Physiological Science (NIPS). 293T cells were harvested in IP buffer (20 mM Tris–Cl, 140 mM NaCl, 10 mM NP-40, 10 mM Triton X-100, 0.5 mM EDTA, and 1× protease inhibitor cocktail) to confirm the interaction between Kai1-VEGF-A and Kai1-PDGF-BB, respectively, and incubated for 10 min at 4 °C. Following incubation, centrifuge for 10 min at 4 °C and lysates transfer to a new tube. Lysates were incubated for overnight at 4 °C with anti-VEGF-A (Merck Millipore) (or anti-PDGF-BB; Sigma-Aldrich) antibody or normal rabbit IgG (catalog no. sc-2027; Santa Cruz Biotechnology). After incubation with the antibody, the lysates were incubated with protein A/G sepharose (Abcam) for 2 h at 4 °C. After washing the beads were resuspended in 2× reducing sample buffer and heated for 5 min at 95 °C to dissociate captured antigen from beads. Samples obtained through the immunoprecipitation were performed through the western blotting described above.

10T1/2 cells were used for Kai1 and PDGF-BB, Kai1 and VEGF-A binding assay (PLA; Duolink Proximity Ligation). Cells were seeded on confocal dishes (Ibidi) and incubated for 24 h. After cells were attached, we treated 2-bp (5 μM) for 24 h. Then, cells were fixed 2% cold paraformaldehyde (Wako) for 10 min on ice and incubated with primary antibodies, Kai1 (catalog no. ab140238; Abcam), PDGF-BB (catolg no. 07-1437; Sigma-Aldrich), and VEGF-A (catalog no. #ABS82; Merck Millipore), as for IF. Duolink™ (catalog no. DUO92101; Sigma-Aldrich) experiments were performed using the protocol provided by Sigma-Aldrich. After PLA, the cells were examined by confocal microscopy (Zeiss).

For efficacy evaluation of rhKAI1 and KAI1 overexpressing cells, 2 × 10⁶ PC-3 (human prostate cancer cell line), B16 (mouse melanoma cancer cell line) cells and MDA-MB231 cells were mixed with rhKAI1 protein (catalog no. 12275-H08H; SinoBiological) 4 μg or conditioned media of vehicle or Kai1 O/E 10T1/2 cell line or KAI1 peptide (wild-type sequence; NRPEVTYPCSCEVKGEEDNS, mutant sequence; NRPEVTAPASCEVKGAAANS), and subsequently embedded in 100 μL of matrigel (catalog no. 356231; Corning). NSG and C57BL/6 mice were subcutaneously injected with 100 μL of matrigel-cell mixture in the dorsal area. After 14 days, mice were killed in order to harvest tumor tissues. The tissues were fixed with 4% paraformaldehyde (Wako), paraffin-embedded and mounted on slide glasses. The samples were examined by immunofluorescence analysis.

Newborn mice were exposed to hyperoxia (75 ± 0.5% O₂) from postnatal day 7 to 12 and returned to normal room air. At postnatal day 14, we intravitreally injected 1 μL of PBS, KAI WT, rhKAI1 and KAI Mut into the right eyes of mice (n = 4). At P17, the enucleated eyes were prepared for immunofluorescent staining of whole-mount retinas with Alexa Fluor® 594 isolectin GS-IB4 conjugate (5 μg/mL; Invitrogen). The whole-mount retinas were viewed with a fluorescence microscope (Eclipse 90i; Nikon). Then, the neovascular tufts were marked and calculated using Image J (NIH). The area of neovascular tufts was normalized to the area of whole retina.

All tests were repeated several times, independently, to verify reproducibility and the replicate number for each test is shown in each figure. We performed statistical analysis and used two-tailed Mann–Whitney rank sum tests using Prism software (version 5.00; GraphPad Software). The resulting p values are indicated as follows: NS, *p < 0.05, **p < 0.01, and ***p < 0.001. Data show the mean ± SEM.

Kai1 knockout (Kai1−/−) mice exhibited the accelerated retinal vascularization, as observed in the immunofluorescence (IF) images of the retinal vascular bed compared to in wild-type (WT) mice (Fig. 1a). Additionally, Kai1−/− retinae at postnatal 4 day displayed more filopodia (Fig. 1b) and fewer empty sleeves (i.e. less vessel regression) compared to in WT mice (Additional file 1: Fig. EV1A).

To determine whether the ECs or PCs were responsible for the enhanced angiogenesis in Kai1−/− mice, we analyzed Kai1-expressing cells in the vascular niche. To trace the Kai1 protein expression among various types of cells in vivo, we generated transgenic mice expressing the Kai1-GFP fusion protein using the CRISPR system. PCs and ECs of Kai1-GFP mice were identified by NG2 and IB4, respectively, located at the cell surface. Interestingly, KAI1 was predominantly expressed in PCs (NG2⁺IB4⁻) rather than in ECs (NG2⁻IB4⁺) of the Kai1-GFP mouse retinae (Fig. 1c). Similar results were obtained by immunofluorescent staining for KAI1, NG2, and CD31 in WT mice retinae (Additional file 1: Fig. EV1B). In the brain and heart, KAI1 was expressed in the PCs (Additional file 1: Fig. EV1C). Fluorescence-activated cell sorting (FACS) analysis revealed that 43% of PDGFRβ⁺ PCs were Kai1⁺ cells in the bone marrow of the Kai1-GFP fusion transgenic mice, in contrast, only 9% of ECs were Kai1 positive (Additional file 1: Fig. EV1D). Additionally, we analyzed the expression of this molecule in the mouse cell lines (ECs; MS1, PCs; 10T1/2) and mouse primary-cultured aortic ECs and PCs. PCs exhibited considerably higher Kai1 expression levels compared to ECs (Fig. 1d, e). Similar results were obtained by FACS analysis of HUVECs versus human brain vessel PCs (HBVPs) (Additional file 1: Fig. EV1E). Next, we evaluated the mRNA expression of Kai1 and other tetraspanin superfamily members in PCs and ECs. Cd9 and Cd151 as well as Cd82 (Kai1) were expressed predominantly in PCs (Additional file 1: Fig. EV1F and G).

To evaluate the effects of Kai1 in PCs on angiogenesis, we co-cultured Kai1−/− mice primary PCs with MS1 cells (mouse EC line) and found that the co-culture matrigel tube formation assay was significantly enhanced, compared to co-culture of WT mice primary PCs and MS1 cells (Fig. 1f). In contrast, Kai1-overexpressing PC suppressed ECs sprouting in the co-culture spheroid assay in mouse primary cells and cell lines (Fig. 1g, Additional file 1: Fig. EV1H and Additional file 1: movie EV1A, and B). To confirm the cell of origin of Kai1 for inhibiting angiogenesis, we prepared mock- or Kai1-O/E MS1 or 10T1/2 cells for a matrigel tube formation assay under various combinations of co-culture. We found that overexpression of Kai1 in the PCs led to strong suppression of tube formation, whereas overexpression in ECs showed a relatively weak effect (Fig. 1h).

To understand the different effects of KAI1 in ECs and PCs, we analyzed the localization of KAI1. Interestingly, surface expression of Kai1 was much stronger on mouse PCs (10T1/2) than on mouse ECs (MS1) (Fig. 2a). Next, we performed subcellular fractionation to isolate the lipid raft membrane, non-lipid raft membrane, and cytosolic proteins in ECs and PCs. Endogenous Kai1 was detected in the membranous and cytosolic fractions of 10T1/2, but not detected in MS1 (Additional file 1: Fig. EV2A). When Kai1-GFP was introduced, both PCs and ECs expressed Kai1 in the cytosolic fraction. Interestingly, however, Kai1 expression in the membranous or lipid-raft fraction differed; expression was strong in 10T1/2 and very weak in MS1 (Fig. 2b). Expression of Kai1 in the lipid raft on 10T1/2 rather than on MS1 was confirmed by analyzing the expression of a lipid raft marker, cholera toxin B, and its co-localization with Kai1 on 10T1/2 (Fig. 2c). Two different types of the lipid rafts have been reported: flotillin-enriched planar lipid rafts or caveolin-enriched caveolae. We found that PCs had flotillin-enriched planar lipid rafts whereas ECs had the caveolae type (Fig. 2b).

Next, we examined the mechanism of Kai1 localization at the lipid raft membranous fraction of PCs, but not ECs. Palmitoylation is an important mechanism to regulate functions of the tetraspanin family members including CD9, CD81, KAI1/CD82, and CD151. KAI1 is palmitoylated at cytoplasmic cysteine residues, and then localizes to the cell membrane [12]. It is known that the functions of KAI1 depend on its palmitoylation status [12, 13]. Therefore, we hypothesized that the different distribution or effect of KAI1 between ECs and PCs were related to the different status of KAI1 palmitoylation between these cells. To detect the palmitoylated KAI1 molecules in ECs and PCs, we performed acyl-biotin exchange (ABE) assay. The palmitoylation level of Kai1 was significantly higher in PCs than in ECs (Fig. 2d). Protein palmitoylation is catalyzed by zDHHC palmitoyl transferases, which contain cysteine-rich domains (CRDs) [14]. Members of the zDHHC family, except for zDHHC12, showed higher expression in PCs than in ECs (Additional file 1: Fig. EV2B). We demonstrated that Kai1 directly interacts with zDHHC3 and zDHHC4 among the 23 zDHHC enzymes screened (Additional file 1: Fig. EV2C). To evaluate Kai1 palmitoylation by these two enzymes, we performed an ABE assay following the knockdown of these genes. Interestingly, Kai1 palmitoylation was reduced by zDHHC4 knockdown, but not by zDHHC3 knockdown (Fig. 2e). Inhibition of palmitoylation using 2-bromopalmitate (2-bp) significantly inhibited the Kai1 localization in the lipid raft or membrane of 10T1/2 (Fig. 2f, g).

To investigate the mechanisms by which KAI1 negatively regulates angiogenesis, we analyzed the transcriptome of WT and Kai1^−/− primary PCs and performed gene set enrichment analysis (GSEA). We found that Kai1^−/− PCs expressed significantly lower level of “negative angiogenic regulators” (GO: 0016525) and “blood vessel remodelling genes” (GO: 0001974) compared to control PCs did (Fig. 3a).

Among top 17 of “negative angiogenic regulators” (GO: 0016525), Sulf1 and Lif levels were considerably decreased in Kai1−/− primary PCs to less than 20% of those in control PCs (Additional file 1: Fig. EV3A). Interestingly, Leukemia inhibitory factor (Lif) which is both a negative regulators of angiogenesis and the blood vessel remodelling genes (Fig. 3a, b) emerged as a member of the leading edge subset that contributing to the enrichment score (ES) [15]. A previous study reported that Lif−/− mice have increased endothelial filopodia, branching points, and capillary density in the retina [16], which is very similar to our findings in Kai1−/− mice. Thus, we examined Lif mRNA expression in PCs and ECs after Kai1 O/E (Fig. 3c), and found that Kai1 significantly increased the mRNA levels of Lif in PCs, but not in ECs (Fig. 3d). The Lif protein concentration in the conditioned medium obtained from Kai1-O/E PCs was shown to be higher than that in the medium obtained from control PCs (Fig. 3e). And G0 arrest was increased in the ECs treated with conditioned medium obtained from Kai1-O/E PCs (Additional file 1: Fig. EV3B).

The baseline expression level of Lif was also high in PCs compared to in ECs in mouse and human cells. However, the expression level of its receptor LIFR was comparable between PCs and ECs in both species, mouse and human (Additional file 1: Fig. EV3C). To examine the functional relevance of Lif during angiogenesis, we performed a tube formation assay after neutralization of Lif in the supernatant of Kai1-O/E PCs (10T1/2). Interestingly, the supernatant of Kai1-O/E PCs inhibited the tube formation of ECs (MS1), which was reversed by neutralization of Lif (Fig. 3f).

To further investigate the relevance of LIF in angiogenesis, we examined the effect of the recombinant mouse Lif on the expression of angiogenic genes in the mouse ECs and PCs. Lif significantly suppressed the expression of angiogenic genes, such as, Ang2, Esm1, Dll4, Vegfr2, Sox7, Sox17, and Sox18 in MS1 (Fig. 3g). In a previous study, these genes were upregulated by members of the Sox family [17], which play important roles in the vascular system development [18]. We found that Lif significantly suppressed Sox17, but not Sox7 or Sox18, in ECs (Fig. 3g). In PCs, Lif suppressed Vegfa, Fgf2, Pdgfbb, Dll4, Vegfr2 and Ang2 expression (Fig. 3h). Interestingly, inhibition of palmitoylation by 2-bromopalmitate (2-bp) significantly inhibited the expression of Lif which was induced by Kai1 (Fig. 3i).

Next, we investigated the signalling pathways of the KAI1-LIF axis in PCs. Among the inhibitors of several signalling pathways examined, Src inhibitor nearly completely blocked the induction of Lif mRNA by Kai1 O/E in PCs (Additional file 1: Fig. EV4A). In Kai1^−/− primary PCs, the active form of Src was decreased (phosphorylation at tyrosine 416 of Src), whereas inactive form of Src was increased (phosphorylation at tyrosine 527 of Src) compared to in WT PCs. In Kai1 O/E PCs, the opposite results were obtained (Fig. 4a, b). To reveal the relevance of the Src at the downstream of KAI1, PP2, the specific inhibitor of Src was used [19, 20]. The supernatant of Kai1-O/E PCs (10T1/2) inhibited endothelial cell tube formation and ECs migration, which was reversed by the Src inhibitor (PP2) (Additional file 1: Fig. EV4 B and C).

And we evaluated the downstream of the KAI1/Src pathway by examining transcription factor-binding sites in the Lif promoter region. A consensus p53-binding element in mouse Lif gene was previously reported at position + 4547 (p53 binding site 2) [21]. We identified an additional p53-binding element in this gene at position + 605 (p53 binding site 1) from the transcription start site using a prediction program. Additionally, a consensus Pbx1 binding element in mouse Lif gene was confirmed using prediction program. Interestingly, we found that Kai1 induced binding of p53 at + 605 position of the Lif intron, which was reversed by the Src inhibitor (PP2) (Fig. 4c). In contrast, Pbx1 binding at the Lif promoter by Kai1 O/E was increased by the Src inhibitor (Fig. 4c). To investigate whether the binding of Pbx1 and p53 in the Lif gene-regulating element is important for KAI1/Src mediated Lif expression, we performed luciferase assay. Since Kai1 induced the protein level of p53 and Pbx1, it might effect on the enhancement of Lif gene-regulating element’s activation (Additional file 1: Fig. EV4D and E). We also assessed the specific effects of p53 and pbx1 on the Lif promoter or gene-regulating element, by constructing a luciferase vector with a mutation in the p53 or pbx1 binding sites, respectively. Interestingly, we found that KAI1 induced luciferase activity in the Lif promoter of wild type, which was reversed by the Src inhibitor. And the mutant p53 binding site in the Lif gene-regulating element was not activated by Kai1. However, the mutation of Pbx1 binding site did not affect activation of the Lif promoter (Fig. 4d). From these results, p53 is the specific transcription factor to activate Lif gene-regulating element.

Previously, anti-KAI1 antibody was shown to inhibit VEGF signalling by regulating the KAI1 distribution in the lipid rafts [22]. We predicted that KAI1 binds VEGF or VEGFR directly to inhibit VEGF signalling. We found that recombinant human KAI1 (rhKAI1) bound to recombinant human VEGF (rhVEGF) and rhPDGF-BB, but not rhFGF2 (Fig. 5a–c). KAI1 did not bind to cytokine receptors such as VEGFR2 (Additional file 1: Fig. EV5A). Binding of KAI1 to VEGF or PDGF resulted in sequestration of angiogenic cytokines, leading to inhibition of VEGFR2 and PDGFRβ phosphorylation in HUVECs and HBVPs even in the presence of rhVEGF-A or rhPDGF-BB (Fig. 5d, e). Next, to visualize the interaction of KAI1 and VEGF-A or PDGF-BB in living cells, we performed a bimolecular fluorescent complementation assay based on the complementation between two non-fluorescent fragments of fluorescent protein that are brought together by an interaction between proteins fused to each fragment. These data clearly visualized the direct binding between KAI1 and VEGF-A or PDGF-BB in living cells (Fig. 5f). In addition, we carried out an in vitro co-immunoprecipitation assay with KAI1-Myc and VEGF-A or PDGF-BB. And we confirmed that KAI1 interacted VEGF-A and PDGF-BB. When we treated 2-bp, there was no binding between KAI1 and VEGF-A or PDGF-BB (Fig. 5g). These findings provide the evidence of the requirement of membrane-localized KAI1 for interaction between KAI1 and VEGF-A or PDGF-BB. Additionally, rhKAI1 inhibited endothelial cell tube formation and ECs migration (Additional file 1: Fig. EV5B and C).

Taken together, KAI1 exerts anti-angiogenic effects by two mechanisms. First, KAI1 induces LIF through the Src/p53 axis, which downregulates angiogenic genes. Second, KAI1 directly binds to VEGF-A and PDGF-BB, leading to inhibition of VEGF/PDGF signalling.

To assess the relevance of KAI1 and angiogenesis in tumor, we examined the expression level of KAI1 in human melanoma, prostate cancer and pancreatic cancer tissues by immunohistochemistry. We found that KAI1 was highly expressed in the normal tissues, in contrast, KAI1 level was significantly decreased in cancer tissues (Additional file 1: Fig. EV6A, D, and G). Similar results were obtained by immunofluorescent staining for KAI1 in these tissues (Additional file 1: Fig. EV6B, C, E and F). Interestingly, KAI1 was predominantly expressed in PCs (PDGFRβ⁺) rather than in ECs (VE-cadherin⁺) of the normal tissues (Additional file 1: Fig. EV6B and E).

To investigate the therapeutic potential of KAI1 against tumor growth, we first examined the anti-angiogenic role of rhKAI1 using three-dimensional hybrid spheres consisting of cancer and blood vessel cells (melanoma cells; B16 (mouse melanoma cancer cells), ECs; MS1, and PCs; 10T1/2). Spheres treated with rhKAI1 showed a reduced sprouting activity compared to cells treated with vehicle (Fig. 6a). Interestingly, spheres including KAI1 knockdown PC showed the most angiogenic phenotype and spheres including wild-type PC treated with rhKAI1 showed the most anti-angiogenic.

To investigate the therapeutic potential of KAI1 for inhibiting cancer growth in an in vivo mouse model, we subcutaneously transplanted B16 and PC3 (human prostate cancer cells) with and without KAI1 supplementation (combination of two strategies, firstly, rhKAI1 to bind and sequester VEGF or PDGF and, secondly, supernatant from Kai1-O/E PCs containing Lif to inhibit angiogenesis). Before transplant cancer cells into the mouse, we determined the Lif mediates Kai1 function in this system. First, we measured the secreted Lif level in the media of three-dimensional hybrid spheres. We detected more secreted Lif in Kai1 overexpressed group. In contrast, the Vegfa level was decreased in Kai1 overexpressed group (Fig. 6b, c). Interestingly, the phosphorylation of Vegfr2 was also decreased in Kai1 overexpressed group (Fig. 6d). These findings provide the evidence of LIF mediated KAI1 function in anti-angiogenesis. In vivo, melanoma and prostate tumor growth was remarkably retarded by supplementation with KAI1 (Fig. 7a, b). According to histologic examination, vessel formation in tumors was significantly decreased by supplementation with KAI1 (Fig. 7c). However, KAI1 did not effect on the proliferation of melanoma cells (Additional file 1: Fig. EV7). This means that KAI1 reduced the tumor growth through inhibition of angiogenesis, not through inhibition of proliferation (Fig. 7d). Interestingly, when we transduced the expression of Kai1 gene in mouse Tibialis Anterior muscle, we could detect the Kai1 and Lif simultaneously in the same location (Additional file 1: Fig. EV8A and B). Therefore, KAI1 shows potential for preventing cancer angiogenesis and tumor growth by gene therapy.

Large extracellular loop (LEL) of KAI1 has been demonstrated to be essential for at least some of the biological activities of tetraspanins [23]. In this study, we demonstrated that KAI1 interact with VEGF and PDGF but not with FGF. To design peptide which interact with VEGF and PDGF, we screened the consensus amino acid sequences of VEGF- and PDGF-receptor but exclusive with FGF-receptor sequence. Since domain 2, 3 and 4 of these two receptors are known as essential for ligand binding [24], we narrowed down to these sequences. Lastly, we confirmed these sequences are matched in amino acids sequences of LEL of KAI1 (Additional file 1: Fig. EV9A and B). And we found that 166–185 amino acids of KAI1 protein matched this condition and confirmed that this peptide interact with VEGF and PDGF as much as recombinant KAI1 (Additional file 1: Fig. EV10). We also constructed mutant peptide which are substituted predicted binding sites with alanine and the mutant peptide did not interact with VEGF and PDGF (Additional file 1: Fig. EV10). Surprisingly, a 20-mer peptide derived from the large extracellular loop (LEL) of KAI1 has been shown to have anti-angiogenic effects to block retinal neovascularization (Fig. 7e) and the progression of breast cancer in vivo (Fig. 7f, g). However, mutant peptide did not show any effect in this in vivo models. According to histologic examination, vessel formation in tumor was significantly decreased by KAI peptide (Fig. 7f, g).