The Intrinsic Factor-Vitamin B 12 Receptor and Target of Teratogenic Antibodies Is a Megalin-binding Peripheral Membrane Protein with Homology to Developmental Proteins

From the ‡Department of Medical Biochemistry, University of Aarhus, 8000 Aarhus C, Denmark, iInstitut National de la Santé et de la Recherche Médicale, U489, Hôpital Tenon, 75020 Paris, France, **Tulane University Medical and Astrobiology Centers and Veterans Affairs Medical Center, New Orleans, Louisiana 70112, ‡‡Département de Biochimie, Hôpital Tenon, 75020 Paris, France, and the §§Department of Cell Biology, Institute of Anatomy, University of Aarhus, 8000 Aarhus C, Denmark

Intestinal uptake of vitamin B 12 (B 12 ) 1,2 is facilitated by a receptor recognizing B 12 in complex with gastric intrinsic factor (IF) (for a review, see Ref. 1). Failure of either IF production or receptor expression leads inevitably to a B 12 deficiency state and disease (2,3). Although the existence of a receptor for IF-B 12 was recognized nearly 3 decades ago (4), its biochemical properties and structure have remained elusive, possibly because of difficulties in purifying it in high amounts from the terminal ileum (5,6). The observation by Seetharam et al. (7) that the receptor was expressed in considerably greater amounts in the kidney and yolk sac facilitated the production of antibodies inhibiting IF-B 12 binding (8,9) and confirmed the receptor as a glycoprotein of Ͼ200 kDa. We have recently estimated a size of 460 kDa of the IF-B 12 affinity-purified receptor (9).
Since IF is only detected at very low levels in the circulation and in nongastrointestinal tissues (1), the high expression of the receptor in kidney and yolk sac may suggest that it has other functions/ligands. This is further indicated by several lines of evidence. First, patients with hereditary intestinal malabsorption of IF-B 12 , known as Imerslund-Grä sbeck syndrome (10,11), have proteinuria, indicating that the receptor facilitating IF-B 12 uptake in the intestine is also important for normal kidney function. The significance of this observation is reinforced by the presence of proteinuria in a family of dogs that synthesize a nonfunctional receptor (12). Second, we have recently shown that the receptor binds RAP to a site distinct from the binding site for IF-B 12 (9). RAP is a chaperone-like protein that protects multiple ligand binding sites of processed low density lipoprotein receptor family proteins, in particular the two giant receptors megalin and low density lipoprotein receptor-related protein (13,14). Finally, immunopathological studies indicate a key role of the IF-B 12 -binding protein in embryonic development. We have recently demonstrated that the target antigen (initially designated "gp280") of teratogenic antibodies (15) was associated with the endocytic pathway of yolk sac epithelial cells (16,17) and identical to the IF-B 12 receptor (8).
The present study provides novel molecular insight into the multifaceted IF-B 12 receptor structure and membrane interactions. Determination of its primary structure revealed a modular 3603-amino acid sequence characterized by the presence of a hitherto unreported cluster of 27 CUB 3 domains, which constitute multiple potential sites for interaction with proteins, carbohydrates, and phospholipids. Furthermore, the receptor lacks a transmembrane domain and cytoplasmic tail, although it is internalized and recycles (9,(17)(18)(19). Direct binding of ligand affinity-purified IF-B 12 receptor to megalin and colocalization at the ultrastructural level of the two proteins indicate that the internalization of the receptor may, at least in part, occur by interaction with megalin. Due to the predominant content of CUB domains, the receptor is here designated cubilin.
Amino Acid Sequencing and Estimation of N-Linked Carbohydrate of Cubilin-CNBr fragments and tryptic digests of a 100-kDa CNBr fragment of rat cubilin were purified by reverse phase high pressure liquid chromatography. Seven isolated peptides were subjected to Edmann degradation using an Applied Biosystems model 477 A sequencer equipped with a model 120 A online chromatograph. A cross-flow reaction and the Doublot reaction and conversion cycles were used. Deglycosylation with PGNase F of 5 g of purified rabbit cubilin was carried out as described previously (23).
cDNA Cloning, Sequencing, and Northern Blotting-Total RNA was extracted from renal cortex and BN cells using Trizol (Life Technologies, Inc.) as described by the manufacturer. mRNA required for library construction was isolated using the Qiagen Oligotex kit. Northern blots were made with 1 g of mRNA and revealed with 32 P-labeled riboprobes (base pairs 1205-1645 and 1702-2175).
In total, four libraries were used. Two conventional libraries were constructed in the laboratory using cDNAs synthesized by oligo(dT) and random priming of poly(A)-selected RNA from BN cells using the superscript Kit (Life Technologies, Inc.). After ligation to EcoRI adapters and size fractionation, they were introduced in the Zap or gt11 EcoRI site. Subsequently, screening was performed on a commercial Zap cDNA library (Stratagene) prepared from yolk sac-derived L2 epithelial cells. Finally, to identify the 5Ј-end, a library was constructed in gt11 using the 5Ј-Cap Finder library from CLONTECH. Immunoscreening was carried out on the Zap-BN library using previously reported polyclonal antibodies to gp280. cDNA probes were constructed from known sequences by polymerase chain reaction using a 1:19 mixture of digoxygenin-labeled nucleotide (Boehringer Mannheim) and used to identify overlapping clones. RACE was carried out using Marathon ready cDNA prepared from rat renal cortex (CLONTECH). Specific primers were from base pairs 838 -859 for 5Ј-RACE and base pairs 6872-6891 and 7152-5172 for 3Ј-RACE. Inserts were prepared by the ex vivo excision system for Zap clones (Stratagene). cDNAs from gt11 clones were isolated by EcoRI digestion and inserted in Bluescript. Sequencing was carried out by cycle sequencing in both directions with IRD-41-labeled primers, and the sequence reactions were analyzed on a LICOR 4000 automatic sequencer.
Release of Cubilin from Renal Cortex Membranes-Rabbit renal cortex (0.6 g) was suspended in 3 ml of PBS, pH 7.4, containing 0.1 mM phenylmethylsulfonyl fluoride and Pefablock (Boehringer Mannheim) and homogenized on ice using an ultrathorax homogenizer (23,000 rpm/min) for 20 s. The fluid and solid phases were separated by cen-trifugation (20,800 ϫ g) for 20 min. The saline-soluble and salineinsoluble samples were analyzed by immunoblotting with anti-cubilin and anti-megalin monoclonal antibodies (9). The amounts loaded on the gels were adjusted so that both fractions were derived from 20 g of original cortex. IF-B 12 affinity chromatography of the fluid phase was performed as described (9) except that the buffer contained no detergent.
Rabbit renal membranes were prepared as described previously (20). For release of cubilin, 2 mg of membranes were incubated in 525 l of PBS, 250 units/ml heparin (LEO, Denmark), 20 mM EDTA, or 5 mM phosphatidylethanolamine (Sigma) for 1 h at 22°C followed by centrifugation at 20,800 ϫ g for 20 min.
Immunocytochemistry-Rat kidneys were fixed by retrograde perfusion through the abdominal aorta with 8% paraformaldehyde in 0.1 M sodium cacodylate buffer, pH 7.2. The tissue was trimmed into small blocks, further fixed by immersion for 1 h in the same fixative, infiltrated with 2.3 M sucrose containing 2% paraformaldehyde for 30 min, and frozen in liquid nitrogen. Rat embryos at day 12 of gestation were dissected free of the decidua and parietal layer to expose yolk sac epithelial cells. The tissue was then fixed by immersion and further processed as described above. For electron microscopy, 70 -90-nm cryosections were obtained at Ϫ100°C with an FCS Reichert Ultracut S cryoultramicrotome as described previously (24). For double immunolabeling, the sections were incubated with the two primary antibodies overnight at 4°C after preincubation in PBS containing 0.05 M glycine and 1% bovine serum albumin. Sheep anti-rat megalin serum (20) was diluted 1:200,000, and mouse monoclonal MAB75 against cubilin (16) was diluted to ϳ2 g/ml. The sections were then incubated for 30 min with rabbit anti-sheep serum (1:20,000) (Dako A/S, Glostrup, Denmark) and finally incubated with 10-nm goat anti-rabbit gold particles and 5-nm goat anti-mouse gold particles (BioCell, Cardiff, UK). The sections were embedded in methylcellulose and studied in a Philips CM100 electron microscope. As controls, sections were incubated with secondary antibodies alone or with nonspecific monoclonal antibodies or normal sheep serum. None of the controls showed any labeling.
Analyses of Megalin-Cubilin Interaction by Surface Plasmon Resonance-Surface plasmon resonance measurements were performed on a BIAcore 2000 instrument (Pharmacia). BIAcore sensor chips (type CM5, Pharmacia) were activated with a 1:1 mixture of 0.2 M N-ethyl-NЈ-(3-dimethylaminopropyl) carbodiimide and 0.05 M N-hydroxysuccimide in water. Rabbit megalin was immobilized as described (25) at a concentration of 40 g/ml in 10 mM sodium acetate, pH 4.5, and the remaining binding sites were blocked with 1 M ethanolamine, pH 8.5. The flow buffer was 10 mM Hepes, 150 mM NaCl and 1.5 mM CaCl 2 , 1 mM EGTA, pH 7.4. The binding data were analyzed using the BIAevaluation program.

RESULTS
cDNA Cloning of Cubilin-By immunoscreening of a Zap cDNA library from rat yolk sac BN cells (17) we identified an initial 0.7-kb clone encoding a portion of cubilin. The 5Ј-sequence of this clone was used to design two nested primers to perform 5Ј-RACE on kidney cDNA, allowing identification of the 5Ј-end of cubilin. A number of clones were identified using polymerase chain reaction-generated probes for further screening of yolk sac libraries. Fig. 1 schematizes three overlapping clones used to construct the final cDNA. The last clone contained a polyadenylation signal and a poly(A) tail. The 3Ј-and 5Ј-ends of the 11.6-kb sequence were further confirmed, respectively, by sequencing a 3Ј-end RACE product and a gtll clone selected from a Cap Finder library. Northern blot analysis of yolk sac mRNA (Fig. 1) identified a mRNA of the same size as the cDNA. The presence of the same mRNA in kidney and intestinal mucosa, but not in liver, was confirmed by Northern blotting and reverse transcriptase polymerase chain reaction (data not shown).
Primary Structure of Cubilin-The assembled cDNA revealed an uninterrupted open reading frame of 10.8 kb encoding a 20-amino acid signal peptide and a 3603-amino acid protein with 42 potential N-glycosylation sites (Fig. 2). The molecular size of the protein backbone was calculated as 397 kDa. The seven amino acid sequences determined by N-terminal microsequencing of tryptic and CNBr peptides confirmed the identity of the sequence (boldface letters in Fig. 2). The size of the protein was confirmed by SDS-PAGE. As shown in Fig. 3, deglycosylation of the IF-B 12 affinity-purified rabbit receptor by peptide N-glycosidase F increased its electrophoretic mobility corresponding to a size of 400 kDa. Compared with the 460-kDa size of the untreated glycoprotein, this indicates a carbohydrate content of ϳ13%. Fig. 4A shows the predicted domain organization of the receptor. A stretch of approximately 110 amino acids with no apparent homology to known proteins is followed by a cluster of eight EGF type B repeats preceding 27 contiguous CUB domains accounting for 88% of the protein mass. The high degree of internal homology (overall similarity of 45%) between the CUB domains is evident from the dot plot display in Fig. 4B. A total of 76 disulfide bridges is predicted if all of the extracellular modules fold normally. The only cysteine outside of the CUB domains and EGF repeats is located in the 110-amino acid N-terminal sequence. This residue might account for the partial disulfide bond-dependent dimerization of a minor part of purified receptors (9,17). Fig. 5 shows alignment of the EGF repeats and CUB domains of some of the most homologous regions of other proteins. Two of the EGF repeats (repeats 2 and 4) contain the discontinuous consensus sequence (Asp/Asn)-(Asp/Asn)-(Gln/Glu)-(Asp/Asn)-(Tyr/Phe) for Ca 2ϩ binding and ␤-hydroxylation of Asp/Asn (26). The 110-amino acid CUB domains contain four cysteines except for CUB domain 13, which is missing the first two cysteines suggested to form the upstream disulfide bond (27). The high homology of the CUB domains of bone morphogenic factor, the Drosophila dorsal-ventral patterning gene product tolloid, the embryonic protein Uvs2 in Xenopus laevis, tumor necrosis factor-stimulating gene 6, C1r/C1s, and spermadhesin is seen in the two lower panels of Fig. 5.
Except for the leader peptide, no sequence compatible with a transmembrane domain could be identified. This excludes the possibility that the protein is a type 1 membrane protein or a glycosylphosphatidylinositol-anchored protein that is synthesized with a cleavable hydrophobic C terminus. Furthermore, since almost the entire protein sequence consists of extracellular modules, it is very unlikely that the protein is a type II or III protein with a noncleaved hydrophobic signal peptide inserted in the membrane (28).
Cubilin Is a Peripheral Membrane Protein-To verify that cubilin is a peripheral membrane protein, as predicted by the lack of a transmembrane segment and cytoplasmic tail, we investigated its release from renal cortex membranes by pro- cedures that do not involve solubilization of the membranes or enzymatic treatment. Fig. 6 shows the identical size of the receptor from kidney, yolk sac, and intestinal mucosa (lanes 1-3). As seen in lane 4 versus lane 3, approximately 50% of cubilin was released into the fluid phase by mechanical grinding of renal cortex in PBS, whereas megalin, the 600-kDa transmembrane protein (29) expressed in the same tissues, was released in minimal amounts. IF-B 12 affinity chromatography performed as recently described for solubilized membranes (9), but in the present experiment, without the use of detergent in the running or elution buffer, yielded a comparable amount of purified receptor (data not shown). Cubilin, which remained membrane-associated, was tightly bound but could be released partly by EDTA, heparin, and, to a low extent, phosphorylethanolamine (Fig. 6, lanes 5-11). Heparin and phosphorylethanolamine have previously been reported to bind to the spermadhesin CUB domain (30,31). The same treatments released virtually no megalin (Fig. 6). The size of the released cubilin, as estimated by SDS-PAGE, was not different from the membrane-associated cubilin.
Cubilin Traffics with Megalin-Previous studies have demonstrated the presence of megalin and cubilin in the same endocytic vesicles of the same absorptive epithelia in the intestine, kidney, and yolk sac (9,16,21). Fig. 7 shows electron microscopic examination of rat yolk sac and kidney sections subjected to double immunogold labeling using a sheep antimegalin polyclonal antibody and a mouse anti-cubilin monoclonal antibody. The large gold particles label the megalin antibody and the small particles label the cubilin antibody. An almost identical localization of the two sizes of gold particles is seen. This close localization of the particles led us to test for the formation of cubilin-megalin complexes.
Cubilin Binds Megalin-As shown in Fig. 8, 125 I-cubilin binds to megalin covalently linked to Sepharose 4B. Bound radiolabel was released from the column by EDTA. Surface plasmon analysis (Fig. 9A) confirmed this binding. No difference in the rate of dissociation of cubilin from megalin was seen in the pH interval 4 -8 (not shown). Binding of cubilin to megalin was reduced partially (ϳ75%) when RAP was prebound to megalin, indicating that cubilin binds to the extracellular domain of megalin. Megalin-bound cubilin was still capable of binding IF-B 12 as shown by subjecting the megalin chip to flow with IF-B 12 after the binding of cubilin (Fig. 9B). Thus, the response after adding IF-B 12 represents the formation of a megalin-cubilin-IF-B 12 complex. Control experiments showed no binding of IF-B 12 to megalin (Ref. 9 and data not shown).

Cubilin: A Novel Type of Multifunctional Giant Receptor-
The present study provides novel molecular information on cubilin, previously known as the yolk sac target antigen of teratogenic antibodies and the intestinal receptor for IF-B 12 . The primary structure predicts 35 extracellular modules uniquely organized in a cluster of eight EGF repeats followed by, from a molecular point of view, a huge cluster of 27 CUB domains accounting for 88% of the protein. Northern and Western blotting of kidney, yolk sac, and intestine did not indicate differences in size of the receptor in these organs.
The EGF type B repeats are similar to the carboxyl-terminal extracellular modules of megalin and low density lipoprotein receptor-related protein. Cubilin has otherwise very little homology to these two giant receptors, which also bind RAP and  4. The extracellular modules of cubilin. A, schematic representation of the 460-kDa receptor (designated cubilin) and related developmental control proteins, human tumor necrosis factor-stimulating gene 6 (TSG6), pig spermadhesin aqn3, and the Drosophila protein tolloid human bone morphogenic protein-1. The EGF repeats and CUB domains encode the whole protein except for the 110 residues after the signal peptide. B, a dot plot display of the high internal homology of the CUB domains in cubilin. mediate endocytosis of a variety of ligands. Also, cubilin does not display homology to sortilin, the 95-kDa putative vesicular sorting receptor, which also binds RAP (32).
The CUB domains conform to the description of Bork and Beckmann (27) based on the analysis of 31 copies of a module initially identified in the C1r and C1s components of complement and subsequently in a variety of proteins associated with fetal development. They consist of 110 amino acids defining a characteristic hydrophobicity pattern predicted to form antiparallel ␤-barrels (Refs. 33-35 and see "Note Added in Proof "). The four conserved cysteines, generally thought to form two S-S bridges (Cys 1 -Cys 2 , Cys 3 -Cys 4 ), are found in all but domain 13 of cubilin that lacks the first two cysteines as already described in the first CUB domains of C1r/s and the homologues MASP1/2 (see Ref. 36 and references therein). When analyzed individually, the CUB domains of cubilin are more closely related to those seen in developmental control proteins.
On the functional level, there is compelling evidence that the CUB domains are involved in the binding of proteins, as de-scribed for the Ca 2ϩ -dependent formation of the C1 complex (37), as well as for binding of phospholipids and carbohydrates, as demonstrated for spermadhesins (30,31,38). In addition to the CUB domains, the EGF repeats might also account for some of the binding properties of cubilin. EGF repeats are widely expressed and participate in a number of receptorligand interactions (for a review, see Ref. 39). Two of the EGF repeats in cubilin have the consensus sequence for calcium binding (26,40) and may be involved in the calcium-dependent binding of e.g. RAP or IF-B 12 (9).
Membrane Binding and Internalization of Cubilin-The lack of a transmembrane segment was surprising because cubilin is internalized via clathrin-coated organelles (18) and recycles to the membrane (19). However, early studies have indicated that both an intrinsic factor-B 12 -binding protein (6) and the target protein of teratogenic antibodies (41) could be released, at least in part, from intestinal or renal tissue using mechanical dissociation in the absence of detergents. We further showed that some cubilin was released by heparin, phosphorylethanol- amine, and EDTA, whereas the membrane association was resistant to acid conditions. These observations are in line with the membrane binding properties of spermadhesins. These pro-teins consist of a single CUB domain and lack a transmembrane segment yet are tightly bound to the surface of sperm cells via nonionic interactions with phospholipids (31), whereas another region of the CUB domain is free to bind to carbohydrates of the zona pellucida surrounding the mammalian egg. The lectin binding characteristics of the spermadhesins are not fully characterized but include heparin and a variety of carbohydrates (30,38). In view of its 27 CUB domains, cubilin may link to the membrane in a heterogeneous manner via multiple sites, which might account for our inability to release it entirely from the membrane.
The identification of the membrane components interacting with cubilin is also essential for explaining its internalization and recycling. The present data show that the cubilin is capable of binding to the endocytic receptor megalin, which colocalizes at the subcellular compartments of yolk sac, kidney, and probably also the intestine (9). Co-internalization of a receptor that lacks internalization signal(s) by means of another receptor has previously been shown. Thus, the glycosylphosphatidylinositolanchored urokinase receptor is endocytosed by coupling of urokinase receptor-bound urokinase-inhibitor complex to low density lipoprotein receptor-related protein (42,43). Megalin, which also binds the urokinase-inhibitor complex (20), may perform a similar function in regard to both the urokinase receptor-and cubilin-ligand complexes. Once internalized, IF-B 12 is segregated from the receptor and directed to lysosomes for degradation of IF, whereas the receptor is recycled to the membrane (9,19,44). Since the in vitro cubilin-megalin complex is stable at pH 5, the two receptors might remain in complex during the entire recycling pathway. In contrast, the urokinase receptor recycles to the plasma membrane without being linked to the low density lipoprotein receptor-related protein (42). When we analyzed the effect of polyclonal megalin antibodies and RAP on the endocytosis of 125 I-IF-B 12 in uptake in cultured yolk sac cells, we only found a reduction of 10 -15%. 4 This modest effect might be accounted for by a short cell surface expression of megalin and cubilin due to a rapid recycling of the two proteins and thereby a too short time for the cubilin-megalin to dissociate, a prerequisite for RAP to block binding. Furthermore, a continuous incubation with RAP will probably have no effect on intracellular receptors, since exter-4 S. K. Moestrup and P. J. Verroust, unpublished data.  nally receptor-bound RAP is transported to lysosomes for degradation (45). To further characterize the partnership of these two giant receptors, studies will be initiated to investigate cubilin trafficking in megalin-deficient or megalin-mutated cells expressing cubilin. However, this study might be complex, since recent data on megalin-deficient mice indicate a key role for normal development of the endocytic apparatus in the proximal kidney tubules and for the survival of the mice in general (13).
Cubilin and Fetal Development-The high expression of cubilin at the functional materno-fetal interface in the yolk sac and the teratogenic effect of anti-cubilin antibodies in the rat suggest an important role of cubilin for fetal development. The mode of action of the teratogenic antibodies is not known. They have previously been shown to inhibit endocytosis, thus reducing the amount of maternal proteins internalized and consequently the amount of protein-derived amino acid that can be incorporated into embryonic tissue (46 -48). However, there is yet no direct evidence that a decreased amino acid supply is responsible for fetal malformations. Alternatively, the teratogenic effect might relate to a more specific disturbance of the materno-fetal barrier such as an impaired transfer of B 12 or of other nutrients. Interestingly, the pattern of antibody-induced fetal malformations, which includes abnormal cranio-facial development, in particular of the eyes and hypophysis (16), resembles to some extent the holoprosencephalic syndrome induced by anti-cholesterol agents (49) and knockout of the cholesteroldependent Sonic hedgehog gene (50) or of the megalin gene (51). It has been proposed (52) that the defective development of the central nervous system in megalin-deficient mice was related to a decreased megalin-mediated uptake of cholesterol-containing lipoproteins, which in turn altered the addition of cholesterol to the Sonic hedgehog protein. It is therefore intriguing to speculate that the anti-cubilin antibodies could interfere with cholesterol uptake either directly or indirectly via binding of cubilin to megalin in the yolk sac.
In conclusion, the present data establish cubilin as a novel type of peripheral membrane receptor with multiple potential sites for interaction with other proteins and membrane components. It is at present known that cubilin can bind IF-B 12 , RAP, megalin, and most likely also calcium, phospholipids, and carbohydrates. However, a number of ligands may remain to be identified to explain the role of the receptor in kidney function and its importance in fetal development. Our future studies on this protein will concentrate on its multiple interactions to define its role as a receptor for ligands in the fluids lining the yolk sac, ileum, and kidney epithelia. FIG. 9. Characterization of the cubilin-megalin interaction by surface plasmon resonance analysis. Rabbit megalin was immobilized to a sensor chip, and the on rates and off rates for the binding of cubilin were recorded by the flow of ϳ20 nM purified cubilin along the chip surface. For control, cubilin was exposed to a blank chip. The displayed values are the recordings from the megalin chip subtracted from the recordings from the blank chip. A, sensorgram of the binding of cubilin to megalin. The binding curves in the presence of 10 mM EDTA or after prebinding of RAP to megalin are also shown. B, demonstration of the formation of an IF-B 12 -cubilin megalin complex by subsequent flow with cubilin, running buffer, and IF-B 12 . Evaluation of the binding data suggests a complex binding. By fitting the binding data to a one-binding site model, a K d of ϳ7 nM was measured.