Expression analysis of endoglin missense and truncation mutations: insights into protein structure and disease mechanisms

Abstract

Hereditary hemorrhagic telangiectasia (HHT) is an inherited autosomal dominant vascular dysplasia caused by mutations in either endoglin (HHT1) or activin-like kinase receptor-1 (ALK-1) (HHT2). The majority of the mutations in endoglin cause frameshifts and premature stop codons. Although initial reports suggested a dominant-negative model for HHT1, more recent reports have suggested that mutations in endoglin lead to haploinsufficiency. In this study, we investigated six different missense mutations and two truncation mutations in the endoglin gene to examine whether mechanisms other than haploinsufficiency might be involved in HHT1. Expression of the missense mutants alone revealed that they are misfolded and that most show no cell surface expression. When co-expressed with wild-type endoglin, the missense mutants are able to dimerize with the normal endoglin protein and are trafficked to the cell surface. We also show that although one truncation mutation acts through haploinsufficiency, the other acts in a dominant-negative way. This implies that either dominant-negative protein interactions or haploinsufficiency can cause HHT1. The biochemical analyses for the different mutations suggest that the endoglin N-terminus is important for correct protein folding and that cysteine residues in the first 350 amino acids are involved in intramolecular disulfide bonds, whereas cysteines located closer to the C-terminus of the extracellular domain are responsible for inter­molecular disulfide bond dimerization.

Received 22 November 1999; Revised and Accepted 18 January 2000.

INTRODUCTION

Hereditary hemorrhagic telangiectasia (HHT) is an inherited autosomal dominant vascular dysplasia found in the range of 1 in 2351 to 1 in 39 216 in different geographical regions (1–5). The disorder is characterized by mucocutaneous telangiectases affecting the tongue, lips, fingers, ears and conjunctivae, in association with frequent epistaxis and gastrointestinal hemorrhages. Further clinical manifestations are pulmonary arteriovenous malformations, cerebral arteriovenous malformations, hepatic arteriovenous malformations and, in rare cases, spinal cord arteriovenous malformations (4,6,7). Phenotypic penetrance is age dependent and nearly complete by age 40. The expression of the disease shows great variability and is seen with a wide disparity of clinical features even among members of the same family. This indicates that the inherited mutation alone does not determine the individual phenotype but that other factors are involved.

Mutations in either of two genes, endoglin or activin-like kinase receptor-1 (ALK-1), can cause HHT (8–12). Endoglin maps to chromosome 9q33–q34, the HHT1 locus, and ALK-1 maps to chromosome 12q13, the HHT2 locus. ALK-1 is a type I receptor for transforming growth factor (TGF)-β1, TGF-β3 and a third so far unidentified ligand (13). Endoglin and ALK-1 can be found in a ligand-independent complex, suggesting that these two proteins share a common signaling pathway (13).

Endoglin is a homodimeric integral membrane glycoprotein composed of disulfide-linked 90–95 kDa subunits (14–16) and is expressed primarily in the vascular endothelial cells of capillaries, arterioles and venules, as well in activated monocytes, syncytiotrophoblasts and some leukemic cells (16,17). Endoglin is a receptor for both TGF-β1 and -β3 and is found in a signaling complex with transforming growth factor-β receptor-(TβR)-I and -II after ligand binding (18–20). Binding of TGF-β to endoglin requires the presence of the type II receptor (21,22). Endoglin is also an accessory protein for activin A and bone morphogenic protein 7 utilizing the activin type II receptors (21).

To date, at least 30 different mutations have been identified in the endoglin gene, all in the extracellular domain (8,9, 12,21,23,24). The majority are nonsense mutations, larger genomic deletions, splice site changes, small nucleotide insertions and deletions leading to frameshifts and premature stop codons. One of the first hypotheses proposed for endoglin dysfunction was a dominant-negative model based on the idea that the truncated endoglin proteins might interrupt proper signaling either by aberrant binding to normal endoglin or by being secreted and then binding to and sequestering TGF-β. (9)

Since this initial proposal, the model of haploinsufficiency has been proposed for HHT1, and certain evidence for this model is compelling. For example, endoglin mutations that eliminate large portions of the protein, including the polyadenylation site, have been described (12), as well as three endoglin mutations that were found to be undetectable at the mRNA level (12,24). The additional finding of missense mutations that disrupt the start codon suggested instead that the majority of the mutations in endoglin represent null alleles, and that the disease phenotype is the result of haploinsufficiency (12,24). Other reports have claimed to show evidence of haploinsufficiency through studies using monoclonal antibodies to detect intracellular and cell surface endoglin levels (21,25). We show here, however, that monoclonal antibodies do not detect all mutant proteins.

However, one subset of mutations, the missense mutations which lead to single amino acid changes in the endoglin protein are likely to be expressed and therefore might act in a dominant-negative fashion. To address this question, we studied six different missense mutations and two truncation mutations. Expression of the mutants alone reveals that they are misfolded and that most of them show no cell surface expression. However, when co-expressed with wild-type endoglin, the missense mutants are able to dimerize with the normal endoglin protein and are trafficked to the cell surface. Although some truncation mutants lead to haploinsufficiency, we show that at least one acts in a dominant-negative way. Thus, it seems that either dominant-negative protein interactions or haplo­insufficiency can cause HHT1. The individual mutations determine the mechanistic pathway of pathogenesis in HHT1.

RESULTS

In this study, we investigated the role of six different endoglin missense mutations, one nonsense mutation and one frameshift mutation. Figure 1 shows a schematic diagram of the endoglin protein with the different features that are relevant for this report. We recently reported the G52V, C53R, W149C and L306P changes (24). A160N is a mutation previously detected in a Japanese HHT family (23). G413V was found in patients from the Netherlands Antilles as well in a Dutch HHT family (C.J. Gallione et al., manuscript in preparation).

The stop codon mutation C350X in exon 8 and the GC nucleotide deletion mutation (ΔGC) in exon 11 were reported previously (8,9). The GC deletion at nucleotide positions 1553–1554 causes a frameshift with a premature stop eight amino acids later, resulting in a 525 amino acid peptide. The positions of the four putative N-linked glycosylation sites are noted (15). G52V and C53R are near the glycosylation site at amino acid position 59. W149C and A160N are in relative proximity to the third glycosylation site at position 135, and L306P is very close to glycosylation site number four at position 308.

Endoglin missense mutations are expressed intracellularly

Missense mutations are likely to cause conformational changes in affected proteins. Therefore, we added either the hemag­glutinin (HA) epitope or the myc epitope to the 3′ ends of the different endoglin expression constructs. This would allow specific detection of wild-type and mutant endoglin proteins independently of their conformational structures.

HA- and myc-tagged missense mutations G52V, C53R, W149C, A160N and L306P were expressed in COS-1 cells and immunoprecipitated with either the anti-HA antibody 12CA5, the anti-myc antibody 9E10 or the anti-endoglin monoclonal antibody (mAb) P3D1. The precipitated proteins subsequently were separated by SDS–PAGE, blotted and immunostained with a polyclonal anti-endoglin antibody (13). Figure 2A shows that mAb P3D1 was able to precipitate the L306P mutant protein and the normal endoglin protein, but not the G52V, C53R, W149C and A160N mutant proteins. Precipitations using the anti-HA (Fig. 2A) as well as the anti-myc (Fig. 2B) antibodies demonstrated that the mutant proteins were expressed, although not fully processed, as shown by western blot staining with our polyclonal endoglin antibody. This confirmed that mAb P3D1 was unable to detect the mutant proteins and was therefore not suitable for this study. In contrast, our polyclonal anti-endoglin antibody can detect all the mutant proteins tested as well as normal endoglin protein.

To test for cell surface expression of the mutant endoglin proteins, COS-1 cells were transiently transfected with the different myc-tagged endoglin constructs and, prior to cell lysis, cell surface proteins were biotin labeled. Subsequent anti-myc immunoprecipitations and streptavidin immunoblot staining showed that none of the missense mutants were expressed on the cell surface (Fig. 2B). Reprobing with polyclonal anti-endoglin antibody demonstrated that all mutants were expressed.

Missense mutations inhibit full glycosylation and cell surface expression of endoglin

Under reducing conditions, the normal endoglin fraction shows two molecular weight bands, whereas only the lower molecular weight band is seen for the missense mutations. To test whether the lower molecular weight band represents the partly glycos‐ ylated endoglin fraction, COS-1 cells were transiently transfected with normal endoglin and then treated with two different glycosylation inhibitors, tunicamycin and castanospermine. Tunicamycin blocks the initial steps in the glycosylation process, whereas castanospermine acts further downstream, allowing partial glycosylation of proteins. Figure 3A shows that, in tunicamycin-treated cells, endoglin has a lower molecular weight than in cells treated with castanospermine, where only the partly processed endoglin protein is seen. As an example of the different missense mutations, G52V mutant endoglin was gel separated in parallel to endoglin from the glycosylation inhibitor-treated cells. The G52V mutant protein corresponds in size to the castanospermine-treated endoglin, suggesting that the missense mutant proteins are only partially glycosylated.

To determine whether the prevention of endoglin glycosylation disrupts trafficking, we mutated the different putative glycosylation sites and looked for cell surface expression of endoglin (Fig. 3B). All four putative glycosylation sites are actual glycosylation sites. The mutation of each glycosylation site independently resulted in lower molecular weight proteins than normal endoglin, and the mutation of any one or two sites concurrently had no effect on endoglin cell surface expression. However, the mutation of three glycosylation sites together (Endo^{glycomut1+3+4}) eliminated cell surface expression of endoglin. The Endo^{glycomut1+3+4} protein has a much lower molecular weight than either the missense mutations or the partly processed protein fraction of normal endoglin. The double glycosylation site mutant endoglin proteins are approximately the same size as the partly processed normal endoglin protein. Thus, we concluded that the block in trafficking for the different missense mutants is not due to lack of or aberrant glycosylation.

Missense mutant proteins form heteromers and show normal surface expression when co-expressed with the endoglin wild-type protein

Endoglin missense mutations expressed in COS-1 cells exhibit processing and trafficking defects of the protein, but does this reflect the normal situation in a patient in whom both alleles, wild-type and mutant, are expressed at the same time?

First, we tested the mutants to see whether they could dimerize with wild-type endoglin. Substituting a truncated form of endoglin, EndoΔ^cyto, for the normal protein allowed us to distinguish between mutant proteins and the otherwise ‘wild-type’ endoglin by size. EndoΔ^cyto has an intact transmembrane domain and a cytoplasmic domain of only seven amino acids (Fig. 1). EndoΔ^cyto does not have an epitope tag. COS-1 cells were co-transfected with EndoΔ^cyto and different HA-tagged missense mutation constructs, and the cell lysates were used for anti-HA immunoprecipitations. The fully processed EndoΔ^cyto protein shows approximately the same molecular weight by SDS–PAGE as the partly processed missense mutants, whereas the partly processed EndoΔ^cyto fraction is of lower molecular weight. If the missense mutant proteins are able to complex with the normal endoglin protein, then an immunoprecipitation for the mutants should also precipitate EndoΔ^cyto. This would appear as two bands on an anti-endoglin western blot; one lower molecular weight band representing the partly processed EndoΔ^cyto and one of higher molecular weight representing the partly processed missense mutant protein. As shown in Figure 4, this is indeed the case.

Next, HA-tagged missense mutants were co-expressed in COS-1 cells with myc-tagged wild-type endoglin. Cell surface proteins were biotin labeled and, subsequently, an anti-HA immunoprecipitation was carried out for the mutant proteins. Surface proteins were detected on a western blot by streptavidin. Co-expression of mutant proteins with the full-length normal endoglin protein enabled some of the mutants to be processed and expressed on the cell surface (Fig. 5A). The G413V mutant protein showed the highest surface expression, followed by L306P and then C53R: G52V, W149C and A160N were still trapped intracellularly. Western blots with whole-cell lysate showed that for wild-type endoglin, >50% of the total endoglin protein present is fully processed. For most of the mutants, the ratio between fully processed and partly processed protein was shifted towards the partly processed fraction (Fig. 5A). With these mutants, less than half of the total endoglin protein visualized was present in a fully processed form, whereas more than half of the total protein was only partly processed. It appears that only fully processed endoglin protein is found on the cell surface.

We repeated the co-expression study using HA-tagged wild-type endoglin and mutants. It appeared that the missense mutant proteins were now able to reach the cell surface to at least some degree (Fig. 5B). This was not surprising as the HA epitope on both the mutant protein and the normal endoglin protein were targeted in our immunoprecipitations. Western blots with whole-cell lysates showed that the ratio between processed and partly processed protein for the mutants G52V, C53R, W149C and A160N had shifted to approximately even amounts (Fig. 5B). Again, G413V and L306P showed a wild-type or almost wild-type endoglin expression pattern. Expression of HA-tagged missense mutants alone showed surface expression for the L306P and G413V mutants, albeit at 50% or less than normal endoglin (data not shown).

To reflect the in vivo situation more accurately, we repeated our co-expression experiments with non-tagged normal and mutant endoglin constructs. To identify specifically the mutant protein and the normal endoglin protein, we looked again in whole-cell lysate western blots at the ratio of processed to partly processed proteins. As shown in Figure 6, there was no difference between endoglin wild-type expression and mutant–wild-type co-expressions. Again, expression of missense mutants alone showed surface expression for the L306P and G413V mutants but not for the other four mutants (data not shown). Thus, we concluded that the myc epitope tag blocks trafficking of the endoglin wild-type–mutant heterodimer and that the otherwise normal endoglin helps the mutant proteins to reach the cell surface.

Exon 11 frameshift mutation ΔGC acts in a dominant-negative manner

We tested the C350X and the ΔGC mutations to determine their roles in endoglin expression and dimerization and their potential for secretion. Both mutations were transiently expressed in COS-1 cells either alone or with normal endoglin. Western blots with cell lysate and cell supernatant under reducing and non-reducing conditions showed that the C350X mutant is expressed as a protein of ∼52 kDa and the ΔGC mutant protein is ∼75 kDa (Fig. 7A). The ΔGC protein is secreted as demonstrated by the appearance of a specific band in the supernatant under reducing conditions which disappears under non-reducing conditions. Thus, we concluded that ΔGC is able to dimerize or form oligomers/heteromers of higher order.

The picture for C350X is less clear. Under reducing conditions, we see two bands in the C350X supernatant that do not appear in the ΔGC supernatant. These two bands have a size of ∼60 kDa each, and therefore do not correspond in terms of molecular weight to the intracellular C350X peptide of 52 kDa. We do not know whether the C350X protein becomes modified during the secretion process. Under non-reducing conditions, these two bands persist. Thus, if they indeed represent the C350X truncation mutation, then they are not able to dimerize/oligomerize like the ΔGC mutant protein does.

To test for possible protein interactions, endoglin wild-type (myc-tagged) and ΔGC or C350X, respectively, were co-expressed in COS-1 cells. Prior to cell lysis and anti-myc immunoprecipitation, cell surface proteins were labeled with biotin. Using streptavidin immunoblots, cells expressing either wild-type endoglin alone or co-expressing wild-type endoglin plus the C350X mutant showed equal amounts of normal size endoglin protein but no protein of the size of the C350X mutant (Fig. 7B). This implies that the C350X mutant protein does not bind to endoglin and does not interfere with the processing and trafficking of normal endoglin. Co-expression of the ΔGC mutant and wild-type endoglin reduced cell surface expression of endoglin by at least 50% (Fig. 7B). We could not detect any biotin-labeled ΔGC protein. When we reprobed the blot with our polyclonal endoglin antibody, we did detect a small amount of ΔGC protein that was co-precipitated with the myc-tagged wild-type endoglin. We conclude then that the ΔGC protein interferes with endoglin processing and cell surface expression by forming a ΔGC–wild-type endoglin heterodimer. Thus, the ΔGC muta‐ tion must be considered a dominant-negative mutation.

Endoglin cytoplasmic domain is involved in dimerization and/or trafficking

We have shown that the endoglin missense mutations require the wild-type endoglin protein for expression on the cell surface. Missense mutants are able to dimerize with the EndoΔ^cyto peptide; however, none of the mutants become fully processed when co-expressed with EndoΔ^cyto (Fig. 4). When the missense mutations were expressed with the full-length wild-type endoglin, they were processed correctly. One explanation for the lack of full processing when co-expressed with EndoΔ^cyto could be that the mutants were epitope tagged. To test whether the epitope tags on the mutants interfere with the maturation and trafficking process, we co-expressed EndoΔ^cyto with the various missense mutations using untagged mutants. Except for the G413V mutation, the missense mutant proteins were still not fully processed (data not shown). This result suggests that the C-terminus of the endoglin protein plays a role in dimerization and/or trafficking.

DISCUSSION

Germline mutations in one of two distinct genes, endoglin or ALK-1, cause HHT, an autosomal dominant disorder of localized angiodysplasia (5,8,10). Several studies suggest that the majority of the mutations represent null alleles (8,10–12,24–26). The current model therefore, is that the disease phenotype is the result of inherited haploinsufficiency for either endoglin or ALK-1. Missense mutations that are stably expressed can result in constitutively active proteins, in a gain of function, or can display a dominant-negative mechanism (27). We investigated six different endoglin missense mutations and two truncation mutations to investigate their expression and localization in cells and to examine whether mechanisms other than haploinsufficiency are involved in HHT1. Further knowledge about the underlying molecular defects that contribute to the development of HHT might help in finding new therapeutic methods.

We demonstrate here that all missense mutations are expressed and show normal processing when co-expressed with wild-type endoglin. However, when expressed individually, the missense mutations exhibit severe defects in protein processing and trafficking and are retained intracellularly. Our results show that endoglin normally is processed as a dimer and we demonstrate unambiguously that the missense mutations dimerize with the normal endoglin protein. We propose that the normal endoglin protein acts as a chaperone for the mutant proteins by dimerizing with them and escorting them to the cell surface where they exist as mutant–wild-type heterodimers.

Of critical importance in our studies are the issues of antibody recognition and the correct use and interpretation of data when using monoclonal versus polyclonal antibodies for mutant protein detection. A recent study also investigated the expression of missense mutations G52V, C53R, W149C and one other, L221P (21). Using mAb P3D1, mutant proteins were not detected on the cell surface and only minimal levels of partially processed mutant proteins were detected intracellularly. The location of the binding epitope for P3D1 is encoded by exons 2–5 (28). The first three of these missense mutations are located in exons 2 and 4 and the fourth is in exon 5, so it is most probable that each of these mutations disrupt mAb P3D1 recognition and binding. However, our polyclonal antibody, which was also raised against the same region encoded by exons 2–5, was able to detect all of the missense mutants that we tested. This is explained by the heterogeneous character of polyclonal antibodies. The polyclonal antibody has multiple recognition epitopes, so if a mutation alters the structure of one epitope, other epitopes are still available for recognition. When an mAb epitope is destroyed, it totally abrogates recognition of the epitope by the mAb.

Mutations L306P and G413V are located in exons 8 and 9a, respectively, and P3D1 is able to recognize both of these missense mutants. Although detection of L306P is reduced compared with wild-type endoglin, P3D1 binds to G413V at levels similar to wild-type (data not shown). This suggests that mAb binding in endoglin is dependent on the spacing between the specific epitope and the missense mutation. This agrees with data presented in the aforementioned study (21). In their tests of the missense mutations G52V, C53R, W149C and L221P, they used two additional mAbs, P4A4 and RMAC8, which are directed against epitopes located in exon 7 and exons 8–12, respectively (28). These two mAbs were able to detect these missense mutants at increasing levels as the distance between mAb epitope and missense mutation increased. These data and our own data clearly imply that the missense mutations alter protein folding not only at the specific site of the mutation but along the entire protein, and thus abrogate epitope-specific detection by the different mAbs. This is also stated by the authors of this recent study (21).

These two studies present similar and mutually supportive data. Why are the conclusions contradictory? The authors of the previous study conclude that because they cannot detect the mutant proteins at wild-type levels using mAbs, then the missense mutants must not be stably expressed in the cell and thus haploinsufficiency is the disease model. We, however, are able to detect stable mutant protein in the cells and on the cell surface. We show this in numerous ways using a polyclonal antibody and by using C-terminal epitope tags to detect the missense mutant proteins. Furthermore, we show that the missense mutant proteins are able to dimerize with wild-type endoglin. Therefore, we propose that missense mutations are likely to act in a dominant-negative way.

The addition of epitope tags to the missense mutations helped us to determine that mAb P3D1 is not able to detect mutant endoglin proteins. Our immunoprecipitation data for the missense mutant proteins, using anti-HA or anti-myc epitope tag-specific antibodies and subsequent immunodetection with our polyclonal anti-endoglin antibody, demonstrate that the missense mutants are in fact expressed and are not degraded rapidly. This suggests that mAb P3D1 and, perhaps, other commonly used endoglin mAbs are not suitable for mutation analyses, or that the results must be interpreted with caution.

The epitope specificity of mAbs could be turned into a diagnostic technique for HHT or other diseases. Using mAbs directed against different epitopes of the wild-type endoglin, detection of either truncated or reduced levels of endoglin in a patient would indicate mutations in endoglin. This technique could also distinguish between patients having HHT1 or HHT2.

The missense mutations provide some very useful insights into endoglin protein folding, processing and trafficking. The primary reason for the disrupted expression pattern seen in the mutants is protein misfolding. Protein misfolding has been implicated in numerous diseases from Alzheimer’s disease and cystic fibrosis to the various spongiform encephalopathies (29). Misfolded proteins are known to be retained intracellularly in either the endoplasmic reticulum (ER) or the Golgi, or they are recycled from the Golgi back to the ER where they are accumulated or degraded (29–33). Our experiment with castanospermine, a glucosidase I inhibitor, implies that misfolded endoglin proteins are retained in the ER. Glucosidase I takes part in the initial step in the calnexin/calreticulin binding cycle which is involved in folding and processing of nascent protein strands in the ER (32,33). This cycle exhibits a cellular quality control over the proteins that enter the cycle, as only properly folded proteins may leave. When cells expressing wild-type endoglin were treated with castanospermine, the only endoglin seen was the partially processed fraction which is normally seen as only a minor fraction in cells. This product was identical in size to the endoglin seen with all the mutants. Thus, it appears likely that mutant endoglin proteins fail the quality control in the calnexin/calreticulin binding cycle and remain in the ER.

We confirmed four previously predicted N-linked glycosylation sites (15). Mutation analyses of these sites demonstrate that glycosylation is not a major factor in endoglin protein trafficking. One function of N-linked glycosylation might be to stabilize the secondary and/or tertiary structure of a protein during the folding and trafficking process (34). Two forms of endoglin are seen in cell lysates: a fully processed form found at the cell surface and a partially processed form found intracellularly. Intermediate sized fractions of endoglin were not found, suggesting that correctly folded endoglin is fully glycosylated and transported to the cell surface rapidly.

The loss of cell surface expression for the individually expressed missense mutations G52V, C53R, W149C and A160N demonstrate that this part of the N-terminus is an important region of endoglin for correct protein folding, whereas L306P and G413V, which are furthest away from the N-terminus, show some cell surface expression. Previously, we proposed that C53 might be important in disulfide bridging (24). The data now suggest that this cysteine is involved in intramolecular bonding. Similarly, the W149C mutation may introduce an additional intramolecular disulfide bond. The change of cysteine residue 53 to arginine and the introduction of a new cysteine at position 149, as well as the change of nearby amino acids, G52 and A160, would therefore alter or disrupt the secondary and/or tertiary endoglin structure. A recent report showed that cysteine residues between C330 and C412 are responsible for endoglin dimerization (35). Our expression and co-immunoprecipitation data for C350X, ΔGC and EndoΔ^cyto support this. The ΔGC and the EndoΔ^cyto truncated proteins each form homodimers as well as heterodimers with the normal endoglin protein. The C350X truncated protein does not dimerize with the wild-type endoglin. This suggests that cysteine residues in the first 350 amino acids are involved in intramolecular disulfide bonds, whereas cysteines located closer to the C-terminus of the extracellular domain are responsible for intermolecular dimerization via disulfide bonds. However, this does not exclude other parts of the protein in dimerization, as our results with the cytoplasmic domain demonstrate.

The addition of HA and myc epitope tags to the C-terminus revealed the importance of the cytoplasmic domain of endoglin in dimerization and trafficking. The HA sequence might actually facilitate dimerization and therefore increase the trafficking of wild-type–mutant endoglin HA-tagged dimers. In contrast, myc tags have been shown to interfere with protein sorting (36). Wild-type endoglin proteins with HA and myc tags, respectively, were processed normally when co-expressed (Fig. 5A). Our experiments suggest that for the co-expression of wild-type and mutants, the presence of different epitope tags, in particular the myc epitope tag, causes additional blocks for processing and trafficking of the mutants. It is likely that the presence of two different tags blocks dimerization at the C-terminus. These results suggest that functional studies employing epitope-tagged proteins should be interpreted with some caution.

Many secreted, transmembrane or membrane-clustered proteins contain a specific three amino acid sequence that usually is located in the C-terminus. Frequently, these three amino acids immediately precede the stop codon (37). This short sequence is composed of S/T(X)Z, where X can be any amino acid and Z tends to be a hydrophobic residue (36,38,39). Proteins that interact with this C-terminus motif contain PDZ domains. It has been shown that the PDZ domain binds to the S/T(X)Z sequence (37). Additionally, PDZ domains are protein-binding modules evolutionarily conserved from bacteria to higher plants and animals. One of their binding functions is to organize both signaling and cytoskeletal networks on the cytoplasmic surface of the plasma membrane. Endoglin contains this S/T(X)Z amino acid motif at the extreme C-terminus as SMA. Betaglycan, which shares regions of homology with endoglin and is also involved in TGF-β signaling, contains the S/T(X)Z motif STA at its C-terminus. We postulate that a PDZ domain protein binds endoglin and directs it towards a particular cellular location and function.

One of the central questions in HHT research has been the cause of the disease phenotype. Is it due to haploinsufficiency or to dominant-negative interactions between the mutant and normal proteins? Our data suggest that both mechanisms are present and that the individual nature of the mutation determines the pathogenesis. Specifically, our results from experiments with truncation mutations support this proposal. Assuming that missense mutations act in a dominant-negative way, only ∼25% of the endoglin dimers at the cell surface would be fully functional endoglin wild-type homodimers. In contrast, mutations which are null alleles reduce the functional levels of endoglin at the cell surface by 50% (haploinsufficiency). This implies that HHT1 patients with mutations leading to haploinsufficiency may exhibit slightly less severe phenotypes than those with a dominant-negative mutation. There are no indications for this in the literature; however, a thorough genotype–phenotype correlation study has not yet been undertaken.

Further investigations are necessary to elucidate whether HHT1 is due to reduced levels of TGF-β signaling or signal modulation, or whether it is the result of a function of endoglin that is not involved directly in TGF-β signaling.

MATERIALS AND METHODS

Cell lines

Cell medium was obtained from Gibco BRL (Grand Island, NY). Fetal bovine serum (FBS) was obtained from Sigma (St Louis, MO). COS-1 cells were cultured in Iscove’s modified Dulbecco’s medium + 5% FBS.

Antibodies and cytokines

Rabbit antiserum against endoglin was prepared against part of the endoglin extracellular domain, encoded by exons 2–5, as described previously (13). The endoglin-specific mAb P3D1 (hybridoma cell supernatant) was obtained from Karen Jensen (University of Iowa, Iowa City, IA). Proteins tagged with the influenza virus HA epitope were immunoprecipitated with the mAb 12CA5. 12CA5 was obtained from the supernatant of a mouse hybridoma cell line expressing and secreting the mAb. Proteins tagged with the myc epitope were immunoprecipitated with the mAb 9E10 obtained from Santa Cruz Biotechnology (Santa Cruz, CA).

Endoglin expression constructs

Endoglin was cloned into the EcoRI and BamHI sites of the mammalian expression vector pCMV5 (pCMV5-Endo^wt). To create the truncated endoglin version EndoΔ^cyto, pCMV5-Endo^wt was digested with EcoRI and MluI. The fragment containing endoglin was gel purified and ligated into EcoRI–MluI-digested pCMV5, resulting in pCMV5-EndoΔ^cyto, a 617 amino acid peptide.

All primer sequences for this study are given from 5′ to 3′. To add an HA or myc epitope tag to the C-terminus of endoglin, the sequence encoding the cytoplasmic domain was amplified with the following primer combinations. For the HA tag, GCACACGCGTTCCCCCAGCAAG and GCTCTAGAGCGGATCCCTAAGCGTAGTCTGGGACGTCGTATGGGTATGCCATGCTGCTGGTGGAGCAGG, and for the myc tag, GCACACGCGTTCCCCCAGCAAG and GCTCTAGAGCGGATCCCTACAAGTCTTCTTCAGAAATAAGCTTTTGTTCTGCCATGCTGCTGGTGGAGCAGG were used. The resulting PCR products were cut with MluI and BamHI, gel purified and ligated into MluI–BamHI-pre-cut pCMV5-Endo^wt. The sequences for the two PCR products were confirmed by sequencing.

Extracellular endoglin sequence changes were introduced into the normal endoglin sequence by site-directed mutagenesis using the QuickChange Site-Directed Mutagenesis kit (Stratagene, La Jolla, CA). The constructs were made with the following primer pair combinations:

EndoG52V, CTCGAAGGTCTGCGTGGCTC

and GAGCCACGCAGACCTTCGAG;

EndoC53R, CTCGAAGGGCCGCGTGGCTCAG

and CTGAGCCACGCGGCCCTTCGAG;

EndoW149C, TCCTTGAGTGCGCAGCTGAGAG

and CTCTCAGCTGCGCACTCAAGGA;

EndoA160N, CATCACCTCTGCTGATGAGCTGAATGACC

and GGTCATTCAGCTCATCAGCAGAGGTGATG;

EndoL306P, GCCCGGATGCCCAATGCCAGC

and GCTGGCATTGGGCATCCGGGC;

EndoC350X, GGACACTTGAAGCCCGGAGC

and GCTCCGGGCTTCAAGTGTCC;

EndoG413V, CTCCAGCTGTGTCATGCAGGTGTCAG and CTGACACCTGCATGACACAGCTGGAG;

EndoΔGC, GGGCAACTGTGTGACTGCTGTCCCCAAG

and CTTGGGGACAGCAGTCACACAGTTGCCC;

EndoGlyco1, ATCCAAGCAAACTGGCACCTGGC

and GCCAGGTGCCAGTTTGCTTGGAT;

EndoGlyco2, CTTGGCCTACACTTCCAGCCTGG

and CCAGGCTGGAAGTGTAGGCCAAG;

EndoGlyco3, CCCGGGGGTCACCACCACAGAGC

and GCTCTGTGGTGGTGACCCCCGGG;

EndoGlyco4, CCGGATGCTCACTGCCAGCATTG

and CAATGCTGGCAGTGAGCATCCGG.

The sequences of all mutagenized endoglin constructs were confirmed by sequencing.

Immunoprecipitation and western blots

COS-1 cells were cultured in six-well plates to 70–80% confluence. The LipofectAMINE method was used for transfections according to the manufacturer’s instructions (Gibco BRL). Two days post-transfection, cells were lysed at 4°C for 30 min in 500 µl of lysis buffer [50 mM Tris–HCl pH 7.5, 150 mM NaCl, 0.5% NP-40, 50 mM NaF, 1 mM Na₃VO₄, 1 mM dithiothreitol (DTT), 1 mM phenylmethylsulfonyl fluoride (PMSF), 2.5 µg/ml leupeptin, 2.5 µg/ml aprotinin, 1 mM benzamidine, 10 µg/ml trypsin inhibitor]. The cell lysate was pre-cleared for 1 h with 50 µl of protein A–Sepharose (Amersham Pharmacia Biotech, Piscataway, NJ) [resolved in lysis buffer (w/v)]. A 10 µl aliquot of the pre-cleared lysate was used for estimating the amount of protein with the BCA Protein Assay according to the manu­facturer’s instructions (Pierce, Rockford, IL). For immuno­precipitations, equal amounts of proteins were incubated at 4°C for at least 3 h with 20 µl of mAb P3D1, 20 µl of mAb 12CA5 or 10 µl of mAb 9E10 together with 30 µl of protein A–Sepharose beads. The protein A–Sepharose/immunocomplexes were then washed three times with 500 µl of lysis buffer and resolved in 15 µl of Laemmli buffer. The precipitated proteins were separated by SDS–PAGE and blotted onto a PVDF transfer membrane, Hybond-P (Amersham Pharmacia Biotech), for immunodetection with the indicated antibody. Immunodetection was performed with the ECL western blotting analysis system (Amersham Pharmacia Biotech) according to the manufacturer’s instructions.

Glycosylation inhibitor treatment

COS-1 cells were treated with tunicamycin and castano­spermine, respectively (Roche, Indianapolis, IN). At 24 h post-transfection, 10 µg/ml tunicamycin or 100 µg/ml castano­spermine were added to the cells, which were then incubated overnight in Iscove’s modified Dulbecco’s medium + 5% FBS. Subsequently, cells were lysed and subjected to SDS–PAGE and western blot analysis as described above.

Cell surface biotinylation

Two days after transfection, COS-1 cells were surface-labeled with biotin. Cells were washed three times with ice-cold phosphate-buffered saline (PBS) and then incubated for 30 min at room temperature in HEPES buffer (150 mM NaCl, 5 mM KCl, 1.3 mM CaCl₂, 1.2 mM MgCl₂, 10 mM HEPES pH 7.4) containing 0.5 mg/ml sulfosuccinimidyl-6-(biotinamido)hexanoate (EZ-Link Sulfo-NHS-LC-Biotin; Pierce, Rockford, IL). After incubation, cells were washed with ice-cold PBS. After three washes, cells were lysed in lysis buffer as described above and lysates were subjected to immunoprecipitations and western blots as described above.

ACKNOWLEDGEMENT

This work was supported by NIH grant HL49171.

+

To whom correspondence should be addressed. Tel: [phone]; Fax: [phone]; Email: [e-mail]

Figure 1. Schematic diagram of the endoglin protein. The numbers within the boxes denote the exons in endoglin. The signal peptide is indicated by shading, and the transmembrane domain (tm) is indicated by the hatched area. The four putative glycosylation sites are also noted above the diagram. The missense and truncation mutants used in this report are denoted by the arrowheads beneath the diagram. Cyto, cytoplasmic domain.

Figure 2. Endoglin missense mutations cause conformational changes resulting in intracellular expression only. COS-1 cells were transiently transfected with wild-type endoglin or missense mutation expression constructs. Constructs were either HA or myc epitope tagged. Cell lysates were used for immunoprecipitations. Immunoprecipitated endoglin was detected with anti-endoglin rabbit serum after SDS–PAGE under reducing conditions and western blot. (A) Cells were transfected with HA-tagged wild-type endoglin (wt) and five different HA-tagged missense mutations (G52V, C53R, W149C, A160N and L306P). Endoglin was immunoprecipitated with either anti-endoglin mAb P3D1 or anti-HA mAb 12CA5. P3D1 precipitated only wild-type endoglin and the L306P mutant, whereas 12CA5 precipitated wild-type and all mutant endoglins. (B) Cell surface proteins were biotin labeled prior to cell lysis. Myc-tagged endoglin proteins were immunoprecipitated with mAb 9E10. Streptavidin immunoblot staining shows cell surface expression only for the wild-type endoglin protein but not for the mutant proteins. Reprobing with a polyclonal endoglin antibody reveals mutant protein expression, but only intracellularly.

Figure 3. Endoglin missense mutant proteins are distinct from glycosylation site mutant proteins. (A) COS-1 cells were transiently transfected with wild-type endoglin (wt). Prior to cell lysis, cells were treated for 24 h with the glucosidase I inhibitor castanospermine or with tunicamycin, an inhibitor of the lipid-linked oligosaccharide precursor. Cell lysates from glycosylation inhibitor-treated cells were separated by SDS–PAGE in parallel with lysates from wild-type endoglin- and G52V mutant endoglin-transfected cells under reducing conditions. Immunoblot staining with anti-endoglin rabbit serum shows that castanospermine-treated endoglin corresponds in size to both the processed normal endoglin and the G52V mutant endoglin. Tunicamycin-treated endoglin represents the non-glycosylated core protein. (B) COS-1 cells were transiently transfected with wild-type endoglin and different endoglin constructs containing mutations in the putative glycosylation sites (glycomut). The numbers 1, 2, 3 and 4 refer to the putative glycosylation sites at amino acid positions 59, 122, 135 and 308, respectively. All constructs are HA tagged and were immunoprecipitated with mAb 12CA5. Cell surface proteins were labeled with biotin. Streptavidin immunoblot staining shows that eliminating one or two glycosylation sites does not inhibit cell surface expression. Only the destruction of three sites concurrently (glycomut 1+3+4) prevents cell surface expression. Expression of endoglin was confirmed by reprobing with a polyclonal anti-endoglin antibody.

Figure 4. Endoglin missense mutant proteins dimerize with a surrogate ‘wild-type’ endoglin protein. COS-1 cells were transiently transfected with endoglin expression constructs either alone or in different combinations, as indicated. All constructs are HA tagged except for EndoΔ^cyto. EndoΔ^cyto has intact extracellular and transmembrane domains but a truncated cytoplasmic domain. This distinguishes by size between missense mutant proteins and the otherwise ‘wild-type’ endoglin protein. Immunoprecipitations were done with anti-HA mAb 12CA5 or, where indicated, with anti-endoglin mAb P3D1. The last lane is a control lane for EndoΔ^cyto of whole-cell lysate that was not subjected to anti-HA immunoprecipitation, which shows both the fully and partly processed forms of EndoΔ^cyto. SDS–PAGE was done under reducing conditions and immunoblots were stained with anti-endoglin rabbit serum. The anti-HA immunoprecipitations demonstrate that the missense mutants co-precipitate EndoΔ^cyto (right). EndoΔ^cyto is the lower molecular weight fragment in the lanes in the right panel.

Figure 5. Endoglin missense mutants show cell surface expression when co-expressed with wild-type endoglin. COS-1 cells were transiently transfected with the different endoglin constructs as indicated. Cell surface proteins were biotin labeled prior to cell lysis. Immunoprecipitations were done with anti-HA mAb 12CA5. Proteins were separated by SDS–PAGE under reducing conditions. Endoglin cell surface expression is shown by streptavidin staining. Endoglin is visualized in immunoblot stainings with anti-endoglin rabbit serum. (A) COS-1 cells were transfected with either 1 or 2 µg of HA-tagged wild-type endoglin, or with 1 µg of myc-tagged wild-type endoglin plus 1 µg of the different HA-tagged endoglin missense mutants. Co-expression of myc-tagged wild-type endoglin and HA-tagged endoglin missense mutants allows partial cell surface expression of the C53R, L306P and G413V mutants. (B) COS-1 cells were transfected with either 1 or 2 µg of HA-tagged wild-type endoglin or with 1 µg of HA-tagged wild-type endoglin plus 1 µg of the different HA-tagged endoglin missense mutants. Co-expression of HA-tagged wild-type and HA-tagged missense mutant endoglin shows increased protein processing and cell surface expression for all mutants. This is demonstrated by the increased amount of fully processed protein in the whole-cell lysate blot (bottom) as compared with the whole-cell lysate blot in (A, bottom).

Figure 6. Co-expression of untagged wild-type endoglin and endoglin missense mutants shows the normal endoglin expression pattern. COS-1 cells were transiently transfected with the indicated endoglin expression constructs and DNA amounts. Equal amounts of cell lysate were separated by SDS–PAGE under reducing conditions. Anti-endoglin rabbit serum was used for immunoblot staining.

Figure 7. Expression analysis of the endoglin truncation mutations C350X and ΔGC. (A) COS-1 cells were transiently transfected with the endoglin wild-type (wt), the endoglin C350ochre (C350X) and the endoglin ΔGC expression constructs alone or with wild-type endoglin in combination with the C350X or ΔGC constructs. Cell lysates and cell supernatants were separated by SDS–PAGE under reducing and non-reducing conditions as indicated. Supernatant was obtained from cells incubated for 8 h in medium containing 0.5% FBS. Immunoblot stainings with anti-endoglin rabbit serum show that the ΔGC mutant protein is secreted (supernatant, reducing conditions) and that, in contrast to the C350X mutation, it is able to oligomerize under non-reducing conditions as demonstrated by the disappearance of the ΔGC truncated protein in the supernatant. (B) Transient transfections in COS-1 cells were done as in (A) but with myc-tagged wild-type endoglin. Myc-tagged endoglin was immunoprecipitated with mAb 9E10 and precipitates were separated by SDS–PAGE under reducing conditions. Streptavidin immunoblot staining shows no cell surface expression for the ΔGC and C350X mutations (left). Wild-type endoglin shows a much lower level of surface expression when co-expressed with the ΔGC mutation compared with co-expression with the C350X mutation, indicating that the ΔGC mutant blocks the trafficking of wild-type endoglin to the cell surface. Small amounts of the ΔGC protein were co-immunoprecipitated with wild-type endoglin, as seen on the anti-endoglin blot (middle).