Pigment epithelium-derived factor binds to cell-surface F1-ATP synthase


  • Luigi Notari,

    1.  Section of Protein Structure and Function, Laboratory of Retinal Cell and Molecular Biology, National Eye Institute, NIH, Bethesda, MD, USA
    Search for more papers by this author
  • Naokatu Arakaki,

    1.  Section of Protein Structure and Function, Laboratory of Retinal Cell and Molecular Biology, National Eye Institute, NIH, Bethesda, MD, USA
    2.  The University of Tokushima Graduate School, Japan
    Search for more papers by this author
  • David Mueller,

    1.  Department of Biochemistry and Molecular Biology, Rosalind Franklin University of Medicine and Science, The Chicago Medical School, IL, USA
    Search for more papers by this author
  • Scott Meier,

    1.  Department of Biochemistry and Molecular Biology, Rosalind Franklin University of Medicine and Science, The Chicago Medical School, IL, USA
    Search for more papers by this author
  • Juan Amaral,

    1.  Section of Protein Structure and Function, Laboratory of Retinal Cell and Molecular Biology, National Eye Institute, NIH, Bethesda, MD, USA
    Search for more papers by this author
  • S. P. Becerra

    1.  Section of Protein Structure and Function, Laboratory of Retinal Cell and Molecular Biology, National Eye Institute, NIH, Bethesda, MD, USA
    Search for more papers by this author

S. P. Becerra, NIH-NEI, Building 6, Room 134, 6 Center Drive, Bethesda, MD 20892-0608, USA
Fax: +1 301 451 5420
Tel: +1 301 496 6514
E-mail: becerrap@nei.nih.gov
Website: http://www.nei.nih.gov/intramural/protein_struct_func


Pigment epithelium-derived factor (PEDF), a potent blocker of angiogenesis in vivo, and of endothelial cell migration and tubule formation, binds with high affinity to an as yet unknown protein on the surfaces of endothelial cells. Given that protein fingerprinting suggested a match of a ∼ 60 kDa PEDF-binding protein in bovine retina with Bos taurus F1-ATP synthase β-subunit, and that F1Fo-ATP synthase components have been identified recently as cell-surface receptors, we examined the direct binding of PEDF to F1. Size-exclusion ultrafiltration assays showed that recombinant human PEDF formed a complex with recombinant yeast F1. Real-time binding as determined by surface plasmon resonance demonstrated that yeast F1 interacted specifically and reversibly with human PEDF. Kinetic evaluations revealed high binding affinity for PEDF, in agreement with PEDF affinities for endothelial cell surfaces. PEDF blocked interactions between F1 and angiostatin, another antiangiogenic factor, suggesting overlapping PEDF-binding and angiostatin-binding sites on F1. Surfaces of endothelial cells exhibited affinity for PEDF-binding proteins of ∼ 60 kDa. Antibodies to F1β-subunit specifically captured PEDF-binding components in endothelial plasma membranes. The extracellular ATP synthesis activity of endothelial cells was examined in the presence of PEDF. PEDF significantly reduced the amount of extracellular ATP produced by endothelial cells, in agreement with direct interactions between cell-surface ATP synthase and PEDF. In addition to demonstrating that PEDF binds to cell-surface F1, these results show that PEDF is a ligand for endothelial cell-surface F1Fo-ATP synthase. They suggest that PEDF-mediated inhibition of ATP synthase may form part of the biochemical mechanisms by which PEDF exerts its antiangiogenic activity.

Structured digital abstract

  • • MINT-7711286: angiostatin (uniprotkb:P00747) physically interacts (MI:0915) with F-ATPase alpha subunit (uniprotkb:P07251), F-ATPase beta subunit (uniprotkb:P00830), F-ATPase gamma subunit (uniprotkb:P38077), F-ATPase delta subunit (uniprotkb:Q12165) and F-ATPase epsilon subunit (uniprotkb:P21306) by competition binding (MI:0405)
  • • MINT-7711113: angiostatin (uniprotkb:P00747) physically interacts (MI:0915) with F-ATPase epsilon subunit (uniprotkb:P21306), F-ATPase delta subunit (uniprotkb:Q12165), F-ATPase gamma subunit (uniprotkb:P38077), F-ATPase beta subunit(uniprotkb:P00830) and F-ATPase alpha subunit (uniprotkb:P07251) by surface plasmon resonance (MI:0107)
  • • MINT-7711060: F-ATPase gamma subunit (uniprotkb:P38077), F-ATPase beta subunit (uniprotkb:P00830), F-ATPase alpha subunit (uniprotkb:P07251) and PEDF (uniprotkb:P36955) physically interact (MI:0915) by molecular sieving (MI:0071)
  • • MINT-7711313: F-ATPase epsilon subunit (uniprotkb:P21306), F-ATPase delta subunit (uniprotkb:Q12165), PEDF (uniprotkb:P36955), F-ATPase alpha subunit (uniprotkb:P07251), F-ATPase beta subunit (uniprotkb:P00830) and F-ATPase gamma subunit(uniprotkb:P38077) physically interact (MI:0915) by molecular sieving (MI:0071)
  • • MINT-7711083: PEDF (uniprotkb:P36955) physically interacts (MI:0915) with F-ATPase epsilon subunit (uniprotkb:P21306), F-ATPase delta subunit (uniprotkb:Q12165), F-ATPase gamma subunit (uniprotkb:P38077), F-ATPase beta subunit (uniprotkb:P00830) and F-ATPase alpha subunit (uniprotkb:P07251) by surface plasmon resonance (MI:0107)

bovine retina endothelial cell


cytochrome c oxidase


Endothelial Cell Basal Medium-2


the F1 portion of ATPase

F1Fo-ATP synthase

ATP synthase


human F1-ATPase


human microvascular endothelial cell immortalized with telomerase


human umbilical vascular endothelial cell


kringles 1–3


kringles 1–5


pigment epithelium-derived factor


pigment epithelium-derived factor receptor


surface plasmon resonance


Pathological vessel growth in the posterior segment of the eye can perturb the structure and morphology of the retina, and lead to visual loss. If this angiogenesis is prevented, retinal degeneration is dramatically restricted. Therefore, endogenous angiogenic inhibitors are likely to play an important role in ocular neovascularization development. Pigment epithelium-derived factor (PEDF) is a potent antiangiogenic, neurotrophic and antitumorigenic factor [1–5]. It is an extracellular protein present in the interphotoreceptor matrix and vitreous [6,7], believed to be responsible for the avascularity of these compartments under physiological conditions. Moreover, the concentrations of PEDF in the eye are inversely correlated with ocular angiogenic development, and overexpression of PEDF or local PEDF protein delivery prevents ocular neovascularization and tumorigenesis, and delays retinal cell death in vivo [4,8–18]. PEDF induces endothelial cell apoptosis, inhibits the proliferation and migration of endothelial cells, and blocks the formation of endothelial capillary-like networks and vessel sprouting ex vivo from chick aortic rings [1,18,19]. However, little is known about the molecular mechanisms by which PEDF functions to regulate endothelial cell behavior.

PEDF is a member of the serpin superfamily by structural homology, but does not have inhibitory activity against serine proteases [20]. Its biological activities are associated with receptor interactions at cell-surface interfaces and changes in protein expression. There is evidence for high-affinity PEDF-binding sites and proteins in retinoblastoma cells, normal retina cells, cerebellar granule cell neurons and motor neurons, as well as in endothelial human umbilical vein endothelial cells (HUVECs) [21–24]. We have recently identified an ∼ 85 kDa PEDF-binding protein in the retina that is a phospholipase-linked membrane protein, termed PEDF receptor (PEDF-R) [25]. PEDF has high affinity for this protein, and stimulates its phospholipase A enzymatic activity.

It is unclear whether the only receptor for PEDF is PEDF-R. Studies on PEDF binding partners have also revealed a PEDF-binding protein of ∼ 60 kDa in membrane extracts from bovine retinal tissues and retinoblastoma Y-79 tumor cells [21,22] of as yet unknown identity. Preliminary investigations by peptide fingerprinting suggested a match of the bovine retinal protein to Bos taurus F1-ATP synthase β-subunit (Table S1). Until recently, F1Fo-ATP synthase expression was assumed to be strictly confined to mitochondria, where it generates most of the cellular ATP. Current evidence for extramitochondrial expression of its components is derived from immunofluorescence, biochemistry and proteomics studies [26,27]. F1Fo-ATP synthase components have been identified as cell-surface receptors for apparently unrelated ligands during studies performed on angiogenesis, lipoprotein metabolism, innate immunity, hypertension, or regulation of food intake. One of these ligands is angiostatin, which also inhibits ocular angiogenesis [28] and is antitumorigenic [29], like PEDF. It has been reported that angiostatin binds and inhibits the F1 catalytic domain of F1Fo-ATP synthase on HUVEC surfaces, leading to inhibition of migration and proliferation of endothelial cells [30–32]. HUVECs possess high ATP synthesis activity on the cell surface [33]. Extracellular ATP generation by HUVECs can be detected within 5 s after addition of ADP and inorganic phosphate, and is inhibited by mitochondrial F1Fo-ATP synthase inhibitors (e.g. efrapeptins, resveratrol, and piceatannol) targeting F1 [33]. Furthermore, these F1-targeting ATP synthase inhibitors can block tube formation and proliferation of HUVECs without affecting intracellular ATP levels [33,34]. These observations agree with the idea that the mechanisms of blocking angiogenesis might involve binding and inhibition of the endothelial cell-surface F1Fo-ATP synthase.

In the current study, we examined the potential interactions between PEDF and ATP synthase. We used highly purified recombinant yeast F1-ATPase and recombinant human PEDF in size exclusion ultrafiltration assays and surface plasmon resonance (SPR) spectroscopy. We also assessed the binding of PEDF to endothelial cell-surface ATP synthase, and examined the effect of PEDF on the extracellular ATP synthesis activity of human microvascular endothelial cells (HMVECs) and bovine retinal endothelial cells (BRECs). Our results provide evidence for high-affinity interactions between PEDF and F1, as well as for PEDF-mediated inhibition of extracellular ATP synthesis activity in endothelial cells. We discuss how these interactions provide insights into the mechanisms of action for angiogenesis inhibition.


Direct binding of PEDF and F1-ATPase

To investigate the potential interactions between PEDF and ATP synthase, mixtures of highly purified recombinant yeast F1-ATPase (∼ 360 kDa) and human PEDF (50 kDa) were first assayed by complex formation. Solutions containing F1-ATPase (14.4 μg) and PEDF (2 or 20 μg) were mixed and incubated at room temperature for 1 h before the mixtures were subjected to size-exclusion ultrafiltration through membranes with 100 kDa exclusion limits (C-100). Figure 1 shows PEDF immunostaining and Ponceau Red staining of bands for F1 subunits and PEDF after SDS/PAGE. One-tenth of the reaction mixture was removed before size-exclusion ultrafiltration, and analyzed in separate lanes as control of starting material (Fig. 1, lanes 10 and 11). PEDF immunostaining was proportional to the PEDF amount added to the reactions. Ponceau Red bands for α-subunits and β-subunits were detected in both reaction mixtures. In lane 10 of Fig. 1, the relative intensities of Ponceau Red bands suggested a lower ratio of PEDF to α/β-subunit (about or less than 1 : 10) than in lane 11, in which the band intensities for each α/β-subunit and PEDF appeared at an approximately 1 : 1 molar ratio for these components. After ultrafiltration, only the reactions with equimolar amounts of PEDF and α/β-subunits showed detectable levels of PEDF (Fig. 1, lane 2), indicating the formation of PEDF complexes with F1-ATPase. Omitting F1-ATPase (Fig. 1, lanes 3 and 4) or replacing it with BSA (66 kDa) (lane 9) did not result in PEDF complexes. A 1 mm Mg2+/ATP combination is known to increase the stability of the F1-ATPase multimeric protein, as detected by an increase in Ponceau Red staining of the F1 subunits (see bottom of Fig. 1 and compare lanes 1, 2 and 7 with lanes 5, 6 and 8). Binding reactions in the presence of 1 mm Mg2+/ATP resulted in a proportional increase in the amount of PEDF–F1 complexes (compare lanes 2–6 in Fig. 1). Formation of fluorescein-conjugated PEDF complexes with F1-ATPase was also observed (Fig. S3). These observations revealed that PEDF bound specifically to F1 complexes.

Figure 1.

 Assays for complex formation between soluble recombinant human PEDF and recombinant yeast F1-ATPase. Proteins were incubated for 1 h at room temperature, and the mixtures were then subjected to size exclusion ultrafiltration, using membranes with size exclusion limits of 100 kDa (lanes 1–9, indicated by C-100). The amounts of each component in each reaction mixture are indicated at the top. The total protein complexes retained by the membrane for each reaction were applied to lanes 1–9 of a 10–20% polyacrylamide gel, and resolved by SDS/PAGE. One-tenth of the reactions corresponding to lanes 1 and 2 before being subjected to ultrafiltration were applied to lanes 10 and 11 (indicated by 1/10 rxn, No C-100) of the same gel. Proteins were transferred from the gel to a western blot, stained with Ponceau Red (bottom blot), and then immunostained with antibodies against PEDF (top blot). The migration positions of PEDF, F1α-subunit, F1β-subunit and F1γ-subunit are indicated to the right, and those of protein standards to the left (BSA, ∼ 66 kDa; Ova, ovalbumin, ∼ 48 kDa; CA, carbonic anhydrase, ∼ 31 kDa).

To determine the biophysical binding parameters for the PEDF–F1-ATPase interactions, real-time SPR spectroscopy was performed. Sensorgrams with PEDF immobilized on the surface of a CM5 sensor chip revealed binding response units for the yeast F1-ATPase that were above those of reference cells (without PEDF) (Fig. 2A). They indicated specific, reversible and concentration–response binding of F1 to PEDF (Fig. 2B). The kinetic parameters for the SPR interactions between F1-ATPase and PEDF were consistent with 1 : 1 Langmuir binding, implying one-site binding between F1 and PEDF. They revealed high binding affinities (KD = 1.51 nm) with high association rates and low dissociation rates between PEDF and F1-ATPase in vitro (Fig. 2B). Similarly, the SPR interactions between F1 and angiostatin kringles 1–5 (K1–5) were assessed (Fig. 2C). Table 1 summarizes the results obtained with several batches of F1-ATPase proteins. The yeast F1-ATPase had higher affinity for PEDF surface sensor chips than for angiostatin K1–5 surface sensor chips (> 10-fold). Altogether, these results implied that soluble and immobilized PEDF can interact with F1.

Figure 2.

 Real-time SPR binding analyses of F1-ATPase and PEDF interactions. (A) SPR spectroscopy of recombinant yeast F1-ATPase with recombinant human PEDF immobilized on a CM5 sensor chip. Sensorgrams of SPR responses (relative units, RU) of 200 nm F1-ATPase solutions injected onto surfaces with PEDF or without PEDF (reference surface) are shown. (B, C) Sensorgrams were recorded with PEDF (B) or human angiostatin K1–5 (C) immobilized on CM5 sensor chips, and injections of F1-ATPase solutions [100, 50, 20, 10, 5, 1 and 0 nm F1-ATPase in (B); 500, 300, 200, 100, 50, 20 and 0 nm F1-ATPase in (C)], using a BIAcore 3000 biosensor and biaevaluation software. The SPR responses for the blank surface and for the 0 nm F1-ATPase were subtracted from the ones obtained at the various concentrations during the evaluation with biaevaluation software (y-axis), and are shown as a function of time (s, x-axis). The kinetic and thermodynamic values were ka (1/M × s) = 6.89 × 103; kd (s−1) = 1.04 × 10−5 and KD = 1.51 nm for PEDF in (B), and ka (1/M × s) = 962; kd (s−1) = 1.88 × 10−4 and KD = 195 nm for angiostatin in (C).

Table 1.   Summary of SPR kinetic parameters for the interactions between yeast F1-ATPase and human PEDF or human angiostatin K1–5. ND, not determined.
SPR SurfaceF1-ATPase (batch No.)Fit methodka
(1/M × s ± SEa)
(1/s ± SEa)
(1/m ± AVEDEVb)
(nm ± AVEDEVb)
  1. a SE values were obtained from the files of the SPR kinetic analyses using the biaevaluation software program. b AVEDEV values were calculated from ka ± SE and kd ± SE values using excel‘s Statistical functions. c An additional SPR bioevaluation estimated the kd value to be 2.10E-05 1/s and the KD value to be 230 nm for the interactions between F1-ATPase (batch No. 3) and angiostatin surfaces (P. Schuck, personal communication).

PEDF11:1 (Langmuir) binding8.8 × 103 ± 1275.5 × 10−5 ± 9.4 × 10−71.6 × 108 ± 2.7 × 1066.30 ± 0.11
PEDF11:1 (Langmuir) binding with drifting baseline6.9 × 103 ± 761.0 × 10−5 ± 3.0 × 10−76.6 × 108 ± 1.9 × 1071.51 ± 0.04
PEDF11:1 (Langmuir) binding3.6 × 104 ± 3231.7 × 10−4 ± 7.8 × 10−72.1 × 108 ± 1.9 × 1064.82 ± 0.04
PEDF21:1 (Langmuir) binding8.4 × 104 ± 10301.5 × 10−4 ± 8.4 × 10−75.6 × 108 ± 6.8 × 1061.79 ± 0.02
PEDF21:1 (Langmuir) binding with drifting baseline8.6 × 104 ± 8837.2 × 10−4 ± 4.8 × 10−61.2 × 108 ± 1.2 × 1068.39 ± 0.09
PEDF31:1 (Langmuir) binding with mass transfer4.7 × 105 ± 92001.4 × 10−3 ± 2.5 × 10−53.2 × 108 ± 6.4 × 1063.08 ± 0.06
Angiostatin21:1 (Langmuir) binding1.3 × 103 ± 292.0 × 10−4 ± 2.2 × 10−66.5 × 106 ± 1.5 × 105154 ± 3.5
Angiostatin21:1 (Langmuir) binding0.5 × 103 ± ND6.9 × 10−5 ± ND7.3 × 106 ± ND137 ± ND
Angiostatinc31:1 (Langmuir) binding0.96 × 103 ± 4.51.9 × 10−4 ± 2.9 × 10−65.1 × 106 ± 7.9 × 104195 ± 3.0

Competition between PEDF and angiostatin for F1-ATPase binding

Angiostatin binds the α/β-subunits of F1-ATPase [31]. To determine whether PEDF and angiostatin share a binding site(s) on F1-ATPase, the SPR interactions between angiostatin and F1-ATPase were subjected to competition by PEDF. Injections of yeast F1-ATPase mixed with increasing concentrations of PEDF decreased the SPR response to angiostatin surface sensor chips in a dose–response fashion (Fig. 3A) and with an estimated half-maximum inhibition, IC50, of ∼ 12 nm PEDF. Control injections of yeast F1-ATPase mixed with PEDF onto PEDF surfaces also decreased the SPR response of F1-ATPase (Fig. 3B; estimated IC50 of ∼ 17 nm PEDF), and PEDF by itself was deficient in binding to either surface (data not shown). Competition between fluorescein-conjugated PEDF and angiostatin or unmodified PEDF for F1-ATPase binding was also observed by size-exclusion ultrafiltration (Fig. S4). These results indicated that PEDF efficiently blocked the F1-ATPase interactions with angiostatin by competing for the angiostatin-binding site(s).

Figure 3.

 Ligand competition for F1-ATPase binding to angiostatin (A) or PEDF (B) surfaces was performed. F1-ATPase (100 nm) was premixed with increasing concentrations of PEDF (as indicated), and injected onto each surface for 300 and 250 s, respectively, at a flow rate of 20 mL·min−1. Dissociation was performed with running buffer for 600 and 300 s, respectively. SPR response differences with respect to blank surfaces were aligned to 0 in the region preceding the injections (Δ Resp. Diff.), and are shown as a function of time. Half-maximal inhibition values determined by nonlinear regression of SPR response differences at saturation and dissociation time points as a function of PEDF concentration were as follows: IC50 = 11.8 ± 0.3 nm PEDF for angiostatin surface, and IC50 = 17.3 ± 2.1 nm PEDF for PEDF surface.

Binding of PEDF to endothelial cell-surface ATP synthase

As illustrated in Fig. 4A,B, PEDF bound to BRECs with high affinity (KD = 3.04–4.97 nm) and with 39 000–78 000 sites per cell (two different batches of cells). Competition of radioligand PEDF binding with unlabeled PEDF showed an EC50 (4.1–4.6 nm) similar to the KD. The physicochemical parameters of these interactions are in agreement with previously reported ones for the binding of PEDF to HUVECs (KD = 5.2 ± 2.3 nm; Bmax = 42 000–54 000 sites per cell; EC50 = 5.1 nm [24]), and the affinity for purified PEDF and yeast F1-ATPase subunits (see above). These results demonstrated that the binding of PEDF to the surface of endothelial cells was specific, was concentration-dependent, was saturable, and had high affinity, and suggested that PEDF interacts with a protein(s) at the surface of endothelial cells.

Figure 4.

 PEDF binding to endothelial cell surfaces. (A, B) Radioligand-binding assays were performed with 2 nm [125I]PEDF and 0–200 nm unlabeled ligand on BRECs attached to collagen-coated plates at 4 °C for 90 min. Cells were washed with binding medium, and bound radioactivity was determined in cell extracts detached with 0.1 m NaOH. Binding competition with unlabeled PEDF (A), and saturation isotherm, nonlinear regression of transformed binding in function of PEDF concentration (B) with a Scatchard plot in the inset are shown. The saturation isotherm was calculated by nonlinear regression of transformed binding data in function of PEDF concentration. The Scatchard plot was calculated by linear regression of the transformed binding data. Both were determined using GRAPHPAD software. (C) Western blots of HMVEC total lysate (Lys) and plasma membrane (Mb) extracts with antibodies against Na+/K+-ATPase, a plasma membrane marker, and to COX-1, a mitochondrial membrane marker, are shown. Samples were loaded onto the gel as follows: lane 1, total homogenate from HMVECs (52 μg of protein); and lane 2, HMVEC membrane fraction (4 μg of protein). (D) Western blots of BREC, Y-79, HMVEC and bovine retinal (Bov. ret.) membrane extracts with antibodies to F1β-subunit. Western blots of the same samples of HMVECs and bovine retina with antibodies to PEDF-R are also shown (bottom). Detergent-soluble plasma membrane protein fractions were prepared and loaded onto gels as follows (protein amounts): lane 1, BRECS (8 μg); lane 2, Y-79 cells (8 μg); lane 3, HMVECs (5 μg), and lane 4, bovine retina (5 μg). Lane 5 contained human heart mitochondria (HH Mit.) (1 μg), a positive control for F1-ATP synthase.

To determine whether the endothelial PEDF-binding component was related to cell-surface F1Fo-ATP synthase, we prepared subcellular fractions of plasma membrane proteins from endothelial cells. We confirmed that they were depleted of mitochondrial membrane markers and contained plasma membrane markers (Fig. 4C). In western blots of detergent-soluble membrane protein fractions from HMVECs and BRECs, we detected proteins that were immunoreactive to antibody to the β-subunit of human heart mitochondrial F1Fo-ATP synthase (anti-hF1), which comigrated with ∼ 60 kDa proteins of yeast and human heart mitochondrial F1-ATPase controls (Fig. 4D). The β-subunit-immunoreactive band was also detected in plasma membrane extracts from normal bovine retina and human retinoblastoma Y-79 tumor cells. However, PEDF-R was undetectable in endothelial cell membrane extracts.

SPR interactions of PEDF and endothelial cell membrane proteins

To investigate whether the endothelial cell-surface F1Fo-ATP synthase binds to PEDF, real-time SPR spectroscopy was performed with detergent-soluble plasma membrane extracts from HMVECs on a PEDF surface sensor chip. Sensorgrams revealed binding response units with injections of membrane extracts that were above those of reference cells (without PEDF) (Fig. 5A), indicating specific binding of a component(s) in HMVEC membranes to PEDF. Upon stopping the injection of extracts, the bound components remaining on the PEDF sensor chip become available to be selectively captured with injections of specific antibodies. This was clearly demonstrated by capturing purified yeast F1-ATPase on PEDF sensor chips with polyclonal antiserum against yeast F1-ATPase (Fig. 5D). To determine whether the PEDF-binding component(s) in endothelial membranes included F1Fo-ATP synthase, solutions of antibodies to F1-ATPase were subsequently injected onto the surface. As shown in Fig. 5A, injections of anti-hF1 increased the SPR response units above those of HMVEC plasma membrane extracts. In contrast, an F1-unrelated antibody that immunorecognized Na+/K+-ATPase in HMVEC plasma membrane extracts (Fig. 4A) did not increase the SPR response (Fig. 5B), and anti-hF1 alone (control injections) did not bind to the PEDF surface (Fig. 5C). Figure 5E,F shows that the F1β-subunit and the previously identified PEDF-R [14] from bovine retina plasma membranes bound to PEDF. PEDF-R was undetectable in endothelial cell membranes extracts by SPR capture (L. N., personal observations), in agreement with western blotting results (see above). Altogether, these results clearly demonstrated that the ATP synthase F1β-subunit in plasma membrane extracts of endothelial cells was a PEDF-binding component. They suggest that interactions of extracellular PEDF ligands with the β-subunit of F1 on endothelial cell surfaces may regulate ATP metabolism.

Figure 5.

 PEDF-binding proteins in cell membranes from HMVECs and bovine retina. SPR spectroscopy on PEDF surfaces of detergent-soluble membrane proteins from HMVECs (A, B) and bovine retina (BR) (E, F), and no extracts (C) and control yeast F1-ATPase (yF1) (D). Antibody capture was performed with antibodies against human F1-ATPase β-subunit (Ab-hF1), yeast F1-ATPase β-subunit (Ab-yF1) and PEDF-R (Ab-RA). Protein extracts (34 μg·mL−1) were injected for 300 s at a flow rate of 20 μL·min−1, and after 600 s of dissociation, the flow rate was decreased to 5 μL·min−1 and specific antibodies (5 μg·mL−1) were injected for 600 s. Sensorgrams relative to the reference surface (without PEDF) are shown. Dashed lines in the sensorgrams point to time of injection of proteins and antibodies, as well as cessation of injection.

Effects of PEDF on the extracellular ATP synthesis activity of endothelial cells

First, we determined the ATP synthesis activity of HMVECs. The cell-surface ATP synthase activity was measured by extracellular ATP production after addition of ADP and inorganic phosphate to intact HMVECs. Extracellular ATP production increased linearly during the first 60 s of incubation, whereas the intracellular ATP levels did not change significantly with incubation time or when inorganic phosphate was not included in the reactions (Fig. 6A). These results demonstrate extracellular ATP synthase activity in these cells, as observed before for HUVECs [33].

Figure 6.

 ATP production by HMVECs. (A) Extracellular ATP production by and intracellular ATP levels of HMVECs. ATP synthesis was initiated by the addition of a solution containing ADP and inorganic phosphate to a culture of HMVECs. At the indicated times, extracellular medium and intracellular pools were prepared, and the ATP content in those pools was determined. Each point corresponds to an average of triplicate samples for: extracellular medium (•); intracellular pools (○); and reactions without inorganic phosphate (bsl00001, extracellular; ×, intracellular). (B) HMVECs were incubated in EBM2/BSA in the presence of PEDF (1 nm) or piceatannol (20 μm) for an increasing period of time (top). Extracellular ATP synthesis activity was determined after incubation for 60 s with ADP and inorganic phosphate in the presence of the indicated inhibitors (x-axis). (C) HMVECs were incubated in EBM2/BSA containing increasing PEDF concentrations, angiostatin K1–5 or piceatannol for 30 min. Extracellular ATP synthesis activity was determined as in (B). Box-and-whisker plot representations of replicates for extracellular ATP synthesis determination are shown. Each point corresponds to a measurement from one well, measurements in each condition were performed in triplicate wells, and measurements in all conditions were repeated with three batches of cells. Values inside the boxes correspond to the central 50% of measurements, the internal horizontal bars correspond to median values, and the vertical lines outside the boxes correspond to variances of measurements. Inhibitor concentrations are indicated on the x-axis. PEDF and the positive controls angiostatin (10 nm) and piceatannol (2 μm) inhibited ATP synthesis.

Then, we examined the extracellular ATP synthesis activity of endothelial cells in the presence of PEDF. The cell-surface ATP synthase activity was measured in HMVECs treated with PEDF for the indicated periods of time. Extracellular ATP generation was assayed within 60 s after addition of ADP and inorganic phosphate, in the presence or absence of PEDF. Treatment for 30 min with 1 nm PEDF decreased extracellular ATP synthesis (Fig. 6B). The positive control, piceatannol, was also a potent inhibitor, requiring ≤ 5 min of preincubation time for effective blocking. Other investigators have demonstrated inhibition of extracellular ATP synthesis by pretreatment of HUVECs for 30 min with much higher doses of angiostatin kringles 1–3 (K1–3) (50 μm [35]) and piceatannol (1–20 μm [33]; 500 μm [35]) than those used here. As shown in Fig. 6C, pretreatment with PEDF for 30 min inhibited extracellular ATP synthesis activity in a dose-dependent fashion. The range of distribution of the measurements reflected the variability of the assay. The median value of the inhibitory activity of PEDF on extracellular ATP synthesis varied between 27%, 43% and 53% with 0.1, 1 and 10 nm PEDF, respectively. No significant statistical difference was observed between PEDF and angiostatin at 10 nm (P ≤ 0.096). Moreover, treatment with PEDF or angiostatin for up to 48 h did not decrease the intracellular levels of ATP; if anything, it slightly increased them (Fig. S1). These results demonstrated that extracellular PEDF additions inhibited the extracellular ATP synthesis activity of endothelial cells.


PEDF, a potent inhibitor of neovascularization, targets endothelial cells [3]. We have shown that PEDF directly binds and inhibits endothelial cell-surface F1Fo-ATP synthase. These two proteins interact when they are in solution and when either one is immobilized. PEDF can bind recombinant yeast F1 in a purified version or native mammalian F1 in membrane cell extracts or in intact cells. We observed that PEDF chemically modified at primary amines (e.g. fluorescein-conjugated PEDF) also binds to F1-ATPase (Figs S3 and S4). The interactions are specific, reversible, and of high affinity, and take place between PEDF and the F1β-subunit. Furthermore, inhibition of extracellular ATP synthesis in intact endothelial cells demonstrates that the PEDF interaction blocks the structural determinant required for the activity of the cell-surface ATP synthase. PEDF shares these properties with angiostatin, and the observed competition for binding to F1-ATPase between these two factors implies that the β-subunit of F1-ATPase has an overlapping site(s) for binding both proteins. These conclusions suggest that interactions between extracellular PEDF ligands and the F1β-subunit on endothelial cell-surfaces may regulate ATP metabolism. They imply that inhibition of ATP synthase may form part of the biochemical mechanisms by which PEDF exerts its antiangiogenic activity.

Previous reports have described a PEDF-binding protein of 60 kDa in plasma membranes from HUVECs [36], normal bovine retina [22], and human retinoblastoma Y-79 tumor cells [21], but have not shown its identity. The present results reveal that the ∼ 60 kDa PEDF-binding protein is the β-subunit of F1Fo-ATP synthase in endothelial cells, as well as in retina and Y-79 cells (Figs 4D and 5E,F; Table S1 [21,22]). Other subunits of the F1Fo-ATP synthase holoenzyme, such as the α-subunits and β-subunits of F1, and the b-subunits and d-subunits of Fo, have also been identified in plasma membranes of HUVECs, several tumor cells, adipocytes, and myocytes [27]. Interestingly, the entire F1Fo-ATP synthase has demonstrable activity in the endothelial cell-surface, with the ability to synthesize ATP and transport protons [27,30]. Our data provide further lines of evidence for the extramitochondrial expression of ATP synthase in the surfaces of endothelial cells. The presence of the F1β-subunit in retinoblastoma Y-79 cell surfaces is consistent with previously reported expression of F1Fo-ATP synthase in tumor cell surfaces [27], and suggests a role for interactions between cell-surface F1Fo-ATP synthase and PEDF in mediating differentiating activity in retinoblastoma cells. PEDF affinity column chromatography of plasma membrane extracts revealed different migration patterns of PEDF-binding proteins among bovine retinal cells, Y-79 cells, and BRECs (Fig. S2 [21,22]). All gave bands corresponding to F1-ATPase α/β-subunits of ∼ 60 kDa, but only bovine retinal cells and Y-79 cells gave detectable bands for PEDF-R of ∼ 85 kDa. Peptide fingerprinting of the PEDF-binding protein of 60 kDa matched it to the F1-ATPase β-subunit (Table S1). The inability to detect PEDF-R in endothelial cell membranes supports the idea that endothelial cell surfaces express a different set of PEDF-binding protein(s) than neural retinal cell surfaces, which may distinctly and specifically trigger angiostatic activities upon interacting with PEDF ligand.

We compared the interactions of the purified F1-ATPase and PEDF proteins, and those that occur with cells. The KD values of the SPR binding of yeast F1-ATPase to immobilized human PEDF match those for the interactions between PEDF and the surface of endothelial cells (KD = 3–7.5 nm) (Fig. 4A,B [24]), as well as the concentration of PEDF capable of inhibiting about 50% of the maximum extracellular ATP synthase activity in HMVECs (Fig. 6C). The estimated IC50 values of PEDF for blocking binding of yeast F1-ATPase to immobilized angiostatin (∼ 12 nm) or PEDF (∼ 17 nm) suggest similar affinities for PEDF when in solution and when immobilized on sensor chips. This observation implies that only minimal changes in affinity occurred upon PEDF immobilization. In contrast, angiostatin K1–5 at concentrations ≤ 270 nm (five-fold the F1-ATPase concentration) could not compete with immobilized PEDF on sensor chips (L. N. unpublished observations), in agreement with a lower affinity for the yeast F1-ATPase–angiostatin interactions. In spite of the higher affinity of yeast F1-ATPase for immobilized human PEDF than for human angiostatin K1–5 as determined by SPR (Table 1), no significant statistical difference was observed between PEDF and angiostatin in inhibiting endothelial extracellular ATP synthase activity (Fig. 5C). A previously reported value of an apparent dissociation constant [Kd(app) = 14.1 nm] for binding of human angiostatin K1–3 to purified bovine heart F1-ATPase immobilized on plastic [30] suggests higher affinity for these interactions than for binding of yeast F1-ATPase to angiostatin K1–5 sensor chips (KD = 130–237 nm; Table 1). The affinity of F1-ATPase–angiostatin interactions is likely to be species-specific, and the observed affinity of the F1-ATPase-angiostatin interaction as determined by SPR is lower than that in mammalian cells. In addition, alterations of structural determinants in angiostatin that are critical for binding F1-ATPase might also affect the affinity of these interactions. For example, immobilization of molecules on the SPR sensor chips by conjugation of primary amines (lysines and N-terminal ends) to the CM5 surfaces may decrease the affinity of the angiostatin molecule for F1-ATPase. As mentioned above, PEDF is not affected by this. Moreover, piceatannol, which is known to target the catalytic F1-ATPase/ATP synthase domain at the β-subunit [37], does not affect the SPR interactions of F1-ATPase with PEDF, either when it is coinjected or when it is included in the SPR running buffer (L. N. and S. P. B., personal observations). This implies that the structural determinants required for binding PEDF and piceatannol do not overlap.

Our results have biological implications. The interactions of extracellular PEDF ligands with cell surface F1Fo-ATP synthase molecules may regulate the levels of ATP and ADP, which in turn may affect the behavior of endothelial cells; for example, PEDF may interact with the ATP–P2X and ADP–P2Y receptor-mediated signaling pathways by regulating the availability of the ATP and ADP ligands, similarly to angiostatin [27]. It has been shown that blocking the ATP synthase by targeting the F1 catalytic domain with angiostatin or piceatannol can trigger caspase-mediated endothelial cell apoptosis, and inhibit the tube formation and proliferation that are necessary for antiangiogenesis [32–34]. Similarly, blocking the ATP synthase with PEDF may trigger signal transduction to mediate apoptosis in endothelial and/or tumor cells.

In summary, this is the first report demonstrating that PEDF binds the endothelial cell-surface F1-ATPase/ATP synthase β-subunit, and inhibits endothelial extracellular ATP synthesis activity. The findings imply that F1Fo-ATP synthase may act as a receptor for PEDF on the surfaces of endothelial cells, and that PEDF can inhibit this extramitochondrial ATP synthase, which catalyzes ATP synthesis. The interactions between PEDF and ATP synthase might be a critical biochemical step for the angiostatic effects exerted by PEDF on the neovasculature.

Experimental procedures


PEDF was human recombinant PEDF, as described previously [38]. Recombinant yeast F1-ATPase was obtained and highly purified as described previously [39]. Human angiostatin K1–5 was purchased from Calbiochem (La Jolla, CA, USA). Human angiostatin K1–3 was from Sigma (St Louis, MO, USA). Polyclonal antibodies directed against the β-subunit of the yeast F1-ATPase were made in rabbits using β-subunit purified from recombinant yeast F1ATPase by SDS/PAGE. Mouse monoclonal antibody against human F1Fo-ATP synthase β-subunit (anti-F1Fo-β; Ab-hF1) (cat. no. MS503), and human heart mitochondrial extracts (cat. no. MS801-50) were from MitoSciences (Eugene, OR, USA).


HMVECs immortalized with telomerase were a generous gift from R. Shao, and were cultured as described previously [40]. BRECs were from Vec Technologies, Inc. (Rensselaer, NY, USA). These cells were sensitive to the angiostatic effects of PEDF.

Size-exclusion ultrafiltration

Complex formation was analyzed by size exclusion ultrafiltration, using Centricon-100 devices with membranes with 100 kDa exclusion limits, as described previously [41]. This assay is based on the fact that PEDF of 50 kDa passes through the membranes, but PEDF in complexes of ≥ 100 kDa does not. The components retained by the membrane after centrifugation and washes of the devices were analyzed by western blotting.

SPR spectroscopy

The interactions between PEDF and yeast F1-ATPase were analyzed by SPR using a BIAcore 3000 instrument (BIAcore, Uppsala, Sweden) with immobilized PEDF ligands, as described previously [42]. PEDF ligand (4 ng) was immobilized on a CM5 sensor chip by N-hydroxysuccinimide/1-ethyl-3-[3-dimethylaminopropyl] carbodiimide hydrochloride] (NHS/EDC) activation, and this was followed by covalent amine coupling of the protein to the surface. A reference surface without protein was prepared by the same procedure. Both surfaces were preconditioned with two injections of 50 mm NaOH, washed with 0.5 m NaCl, and re-equilibrated with binding buffer (10 mm Tris/HCl, pH 7.5, 0.15 m NaCl, 0.25 m sucrose). Ten different dilutions of F1-ATPase solutions with concentrations ranging between 0.3 and 200 nm were injected on both surfaces. Each injection was followed by a 50 mm NaOH regeneration step. The results were analyzed using biaevaluation software. The data were then fitted to several binding models for kinetic analysis. The best fittings were obtained with a simple 1 : 1 Langmuir model for the PEDF surface-binding assay. Background baseline noise was slightly adjusted, using drifting baseline or accounting mass transfer limitation, to produce the best fit.

Radioligand-binding assays

Assays were performed using a given concentration of [125I]PEDF as radioligand and unlabeled PEDF as competitor on endothelial cells attached to the wells, as described previously [21]. The radioligand-binding data were analyzed by nonlinear regression with a one-site competition equation for ligand competition with unlabeled PEDF. The radioligand-binding data were transformed to calculate bound PEDF per assay, and then analyzed by nonlinear regression. The best fittings for saturation binding isotherm were obtained with a classic equation for one binding site (hyperbola). graphpad prism version 4.00 for Windows (GraphPad Software, San Diego, CA, USA) was used for data analysis and generation of plots.

Ligand competition assays

F1-ATPase (100 nm) was premixed with increasing concentrations of PEDF (1–300 nm) and injected with a kinject procedure on a CM5 BIAcore chip with immobilized PEDF (5000 RU) or angiostatin K1–5 (1400 RU). Injections lasted for 300 s, and were followed by 600 s of dissociation, and regeneration of the chip surface with 50 mm NaOH. SPR differential responses were plotted as a function of PEDF concentration.

Cell-surface ATP synthesis activity

The extent of extracellular ATP synthesis by HMVECs was determined as described previously [32], with the following modifications. HMVECs were serum-starved overnight at 37 °C in Endothelial Cell Basal Medium-2 (EBM2)/epidermal growth factor/hydrocortisone (Cambrex Bio Science, Walkersville, MD, USA) plus 0.2% BSA, and then for 1 h at 37 °C in EBM2/vascular endothelial growth factor/basic fibroblast growth factor (Cambrex Bio Science) plus 0.2% BSA, following the manufacturer’s instructions. Cells were preincubated with effectors for 30 min at 37 °C. After 10 min at room temperature, cells were rinsed once with Hepes buffer (10 mm Hepes, pH 7.4, 150 mm NaCl). Then, Hepes buffer containing 1 mm MgCl2 with effectors were added to the wells. After 1 min, a solution of the same buffer with final concentrations of 100 μm ADP, 10 mm potassium phosphate and 1 mm MgCl2 was added. The cells were then incubated at room temperature for the indicated periods of time for up to 120 s, and the extracellular ATP content was determined in the media using the ATP bioluminescence CellTiter-Glo assay kit (Promega, Madison, WI, USA), according to the manufacturer’s instructions. The intracellular ATP content was also determined in lysed cells using a Cell Titer Glo assay kit. All measurements were performed using the Wallac Victor2 1420 multilabel counter (Wallac Oy, Turku, Finland), and the results were analyzed using an Excel spreadsheet (Microsoft, Redmond, WA, USA).

Plasma membrane extracts

Endothelial cell membrane extracts were prepared as described previously [25]. Briefly, cells were grown to confluence, starved in serum-free medium for 16 h, harvested, homogenized, and subjected to differential centrifugation. Separation of the final supernatant (cytosolic fraction) and particulate material (membrane fraction) from transfected cells was performed by centrifugation at 150 000 g. Detergent-soluble plasma membrane fractions were subjected to SPR spectroscopy or to western blotting.

Western blotting

Western transfers were performed as described previously [7]. Immunoreactions were performed with mouse monoclonal antibody against PEDF (Chemicon, Temecula, CA, USA) diluted 1 : 1000, rabbit polyclonal antibody against PEDF-R [25] diluted 1 : 5000, mouse antibody against hF1 (MitoScience, Eugene, OR, USA) diluted 1 : 5000, mouse antibody against cytochrome c oxidase (COX-I) complex IV (Santa Cruz Biotechnology, Santa Cruz, CA, USA) diluted 1 : 2000, or rabbit anti-(human Na+/K+-ATPase) serum (Santa Cruz Biotechnology) diluted 1 : 5000 in 5% BSA in NaCl/Tris-Tween (50 mm Tris/HCl, pH 7.5, 150 mm NaCl, 0.1% Tween-20). Secondary antibodies were rabbit anti-(mouse biotinylated IgG-POD) (Vector Laboratories, Burlingame, CA, USA) diluted 1 : 1000 for western blotting with antibodies against PEDF, peroxidase-labeled goat [anti-mouse IgG (H + L)] (KPL, Gaithersburg, MD, USA) diluted 1 : 100 000 for western blotting with antibody against COX-1, or peroxidase-labeled goat [anti-rabbit IgG (H+L] (KPL) diluted 1 : 100 000 for western blotting with antibody against Na+/K+-ATPase. Immunoreactive bands were detected by chemiluminescence (Super Signal West Dura Extended Duration Substrate; Pierce Biotechnology, Rockford, IL, USA), and signal was acquired using a Typhoon 9410 laser-based scanner (Amersham, Piscataway, NJ, USA). For PEDF analyses, western blotting was performed as described previously [7].


This research was supported in part by the Intramural Research Program of the NIH, NEI, and NIH Grant R01-GM066223. We thank T. Higuti for starting the collaboration between N. Arakaki and our group, P. Schuck for interesting discussions on SPR kinetic evaluations and confirming SPR evaluations with our data, and I. Rodriguez and V. Notario for critically reading the manuscript.