Description and genetic analysis of the HFS patients
We performed the molecular diagnosis of four patients from four different families; the neonatal presentation of all index patients initially raised suspicion of the most severe form of the disease ISH (see Supplementary Information for more detailed description of each patient, ). In particular, painful contractures were the major clinical sign in each case at or shortly after birth, leading to progressive joint limitations and delay of motor development; intellectual development was normal. Typical skin eruptions and mucocutaneous lesions became apparent early in life. Because of feeding difficulties, gingivectomy was performed in at least three of the patients. Despite the similar initial clinical phenotype, the disease followed a clearly different course for each patient after birth (Supplementary Information). The most severe disease manifestation was observed in Patient 3 who showed the full-blown picture of ISH with recurrent, intractable diarrhoea and episodes of intermittent systemic inflammatory activity (SIRS), causing acute fatal circulatory failure in infancy. Clinical evidence of multiorgan involvement was also observed in Patient 4 with the diagnosis of protein-losing enteropathy at the age of 10 months. His general condition, however, remained stable under supportive therapy, and no instance of SIRS or generalized bacterial infection has occurred up to his current age of 3 years. In contrast, neither Patient 1 nor Patient 2, who showed the most favourable clinical course, developed further complications of ISH. Our patients thus collectively represent a continuum of phenotypes associated with HFS, ranging from a fatal multisystem disorder to a milder, yet disabling, oligosymptomatic connective tissue disease. We were interested in finding molecular explanations for these clinical differences with the aim of identifying genotype–phenotype correlations.
Clinical features of HFS patients
In the least affected families 1 and 2, different compound heterozygous mutations of cmg2
were detected (). Patient 1 carries a c.116G>T transversion predicted to cause a novel p.C39F amino acid substitution in the amino terminus of CMG2. In the second allele, a previously described (Hanks et al, 2003
) c.1074delT single nucleotide deletion in exon 13 was found, which modifies the open reading frame by a frame shift leading to a change in the cytosolic tail of the protein and a premature stop (). Patient 2 carries a missense mutation resulting in a c.928G>T transversion, leading to substitution of valine 310 in the ectodomain with a phenylalanine (). In the second allele, a single base insertion (c.1073_1074insC) was found again in exon 13, also leading to a frame shift and a premature stop. Both cases of severe HFS in families 3 and 4 proved to be associated with homozygous mutations. Patient 3 carried a biallelic novel c.945T>G transversion, leading to the change of cysteine 315 to tryptophan (). Patient 4 is homozygous for the same c.1073_1074insC insertion detected in Patient 2 (). The presence of insertions or deletions in exon 13 for three out of the four patients supports the previous observation that a GC-rich stretch in exon 13 is a mutational hot spot (Dowling et al, 2003
; El-Kamah et al, 2010
; Hanks et al, 2003
; Lee et al, 2005
HFS mutations analysed in the present work
Clinical presentation, pedigrees of HFS families and molecular characterization
The HFS mutations lead to drastic reduction of CMG2 protein levels
We made use of a newly generated monoclonal antibody (2F6) to analyse CMG2 in patient-derived fibroblasts. The 2F6 monoclonal antibody was obtained by genetic immunization of rats with a CMG2-expressing plasmid. Characterization of the antibody indicated that 2F6 specifically binds CMG2 on western blots of total cell extracts from different cell types, that it preferentially recognizes the non-reduced form and that it labels CMG2 by immunofluorescence staining but fails to recognized the improperly folded ER precursor forms (Supplementary ).
Based on the mutations identified in the patients, the following bands were expected to be revealed the patient fibroblasts: two bands of, respectively, 55 and 40 kDa for heterozygous Patient 1; two bands of 55 and 37 kDa for heterozygous Patient 2, a single
50 kDa full-length form for the homozygous Patient 3, and a single 37 kDa band for homozygous Patient 4. Surprisingly, CMG2 was undetectable using the 2F6 antibody in cell extracts of Patients 2–4, weakly detectable for Patient 1 and as expected readily detectable in control fibroblasts (Supplementary ). Enrichment of CMG2 by immunoprecipitation, however, allowed the detection in all patients but at very low levels (). Note that migration of full length CMG2 varied between Patients P1–P3 and the control. As will become apparent below, this is due to the fact that SDS–PAGE was performed under non-reducing conditions and that the patient mutations affect disulphide bonding of CMG2.
CMG2 protein and mRNA levels in HFS patient fibroblasts
Low abundance of 2F6 detectable CMG2 protein could be due to lower mRNA levels in patients. To test this possibility, we performed quantitative PCR on RNA extracts from patient-derived fibroblasts. Normalization was performed to three house-keeping genes (see Materials and Methods Section). In addition, we analysed an unrelated gene, RhoA
, and found that its mRNA level was identical in all patients (). Levels of cmg2
mRNA varied greatly amongst patients, with the lowest, 26 ± 17% (n
= 7), observed for Patient 4, homozygous for the c.1073-1074insC frame-shift mutation. The two heterozygous Patients 1 and 2, carrying a nonsense mutation on one allele and a frame-shift mutation on the other, had intermediate mRNA levels. These observations suggest that the nonsense mutations lead to mRNAs recognized by the NMD pathway (Rebbapragada & Lykke-Andersen, 2009
). Interestingly, the mRNAs containing a premature stop codon varied in susceptibility to the NMD pathway. Indeed, the cmg2
mRNA level of Patient 2, carrying the c.1073-1074insC mutation, was significantly lower (55 ± 11% of that of controls, n
= 7) than that in Patient 1 (76 ± 12%, n
= 4) carrying the c.1074delT—even more so considering that both patients also express a cmg2
mRNA from one allele that is not prematurely degraded.
While the amount of cmg2 mRNA can explain the very low level of truncated CMG2 protein in Patient 4, this is not the case for the other patients suggesting that CMG2 is affected at the protein level. To test this hypothesis, we investigated whether inhibition of the proteasome with MG132 would affect CMG2 protein levels in patient cells. MG132 led to an increase in the CMG2 levels in patient-derived fibroblasts indicating that the protein was partly degraded by the proteasome. Degradation by other pathways, such as lysosomes or autophagy, may also occur since the levels of CMG2 observed for example in Patient 3 upon MG132 treatment by no means match the levels predicted by the mRNA. Alternatively, CMG2 levels might be increased but undetectable using the 2F6 antibody. MG132 also led to a marked increase in CMG2 levels in cells from the control patient. This could be due to various reasons. Firstly, normal folding of wild-type (WT) CMG2 could be rather inefficient leading to significant degradation of newly synthesized protein by the ERAD pathway and MG132 would increase the amount of CMG2 that exits the ER. Secondly, MG132 could have an effect on the turnover rate of CMG2 at the plasma membrane and lead to an increase of its half-life.
Model of the CMG2 extracellular Ig-like domain
Our subsequent aim was to understand the mechanisms leading to the premature degradation of CMG2 mutants in patients. Mutation p.C39F found in Patient 1 maps to the stem of the vWA domain (). Upon expression of the vWA domain in Escherichia coli
, this cysteine forms a disulphide bridge with Cys-218 (Lacy et al, 2004
). The two other mutations, p.V310F (Patient 2) and p.C315W (Patient 3) map to the Ig-like domain, the structure of which is unknown. We therefore modelled this domain using a combination of fold recognition (Biegert et al, 2006
; Soding, 2005
) and manual modelling (Guex & Peitsch, 1997
). The model shows an Ig-like fold () where, quite remarkably, four cysteine residues are positioned in a manner compatible with the formation of two disulphide bridges: C230-C315 and C255-C279. In further support of the model, the two potential N
-glycosylation sites, at least one of which is known to be modified (Deuquet et al, 2009
), localize to the solvent exposed surface of the domain.
We next sought experimental evidence for the existence of the predicted disulphide bonds. Single cysteine to alanine mutations were generated for each of the seven ectodomain cysteine residues () as well as double cysteine mutants corresponding to the predicted pairs. Expression was analysed in two cell lines, human Hela cells and Chinese Hamster Ovary (CHO) cells, since these two cell types vary in their protein glycosylation and/or folding capacity, leading to differences in the CMG2 migration pattern on SDS gels (Deuquet et al, 2009
). In CHO cells, CMG2 shows a very distinct pattern with a single sharp band, corresponding to the glycosylated ER precursor form, and a broad higher molecular weight smear corresponding to mature CMG2, in which the N
-linked sugars have been modified by Golgi enzymes into more complex oligosaccharides (Deuquet et al, 2009
). Hela cells have the advantage of being easily transfected at high rates; however, the migration pattern of CMG2 in these cells does not show a clear boundary between the precursor and mature forms, although both are present (Deuquet et al, 2009
Single mutants of the vWA domain cysteines did not drastically affect the migration pattern, i.e. both the precursor and the mature form were observed (), indicating that a significant proportion of the protein was able to exit the ER. Immunofluorescence analysis and surface biotinylation experiments further showed that mutation of Cys-39 or Cys-218 to alanine does not significantly affect targeting of CMG2 to the cell surface ().
Mutations of Ig-like domain cysteines lead to ER retention of CMG2
In marked contrast, mutation of any cysteines in the Ig-like domain led to an almost complete disappearance of mature CMG2 () and rendered the protein fully sensitive to Endoglycosidase H (EndoH; ), an enzyme that can only remove non-complex oligosaccharides, suggesting that transport of CMG2 to the Golgi was impaired. Immunofluorescence analysis of the mutants revealed a typical reticulate ER staining (illustrated in for C315A).
To establish the exact disulphide pairing, extracts from cells expressing the single cysteine mutants were treated with N
-ethylmaleimide (NEM) to block any free cysteines and analysed by SDS–PAGE under non-reducing conditions. Migration of WT CMG2 was very similar to that observed under reducing conditions (). A higher
114 kDa molecular weight form, corresponding to a dimer (see below and ), was, however, observed for the two single, C39A and C218A, cysteine mutants (). The presence of this disulphide-linked dimer implies that, upon the single cysteine to alanine mutation, another cysteine in the protein became available for inter-molecular disulphide bond formation and thus, that in the WT protein, these two cysteines are bonded. The absence of the dimeric band for the double C39A–C218A mutant confirms the pairing of these residues.
CMG2 cysteine mutants form higher order homo-complexes
The pattern for the Ig-like domain single cysteine mutants was more complex, with higher order complexes consistent with CMG2 trimers and tetramers (). The presence of higher order complexes for all single cysteine mutants not only indicates that all four Ig-like domain cysteines are involved in disulphide bond formation but that more than one cysteine must have become available for inter-molecular disulphide bond formation. This in turn implies that disruption of one bridge in the Ig-like domain impairs formation of the other.
The migration pattern of the C230A–C315A double mutant was simpler than that of the two corresponding single mutants confirming disulphide bonding between these two residues in the WT protein (). A disulphide-linked dimer was, however, still observed revealing the presence of one free cysteine and suggesting that the remaining Cys-255–Cys-279 bridge failed to form. The free cysteine in the C230A–C315A mutant is likely to be Cys-255, since Cys-279, even in a misfolded protein, probably resides in the core of the domain and is thus inaccessible for inter-molecular disulphide bond formation. The migration pattern of the last C255A–C279A double mutant remained as complex as that of the corresponding single mutants, indicating that the two remaining cysteines, Cys-230 and Cys-279, both of which are exposed, became available for inter-molecular disulphide bridge formation.
Altogether, these observations determine the identity of the three-disulphide bridges in the CMG2 ectodomain and show that, while the bridge that caps the vWA domain is not essential for passing the quality control leading to ER exit, the two bridges in the Ig-like domain are. The ER retention phenotype could not be rescued by mutating all Ig-like domain cysteines () indicating that the disulphide bridges are essential for proper folding/stability of CMG2.
The above-mentioned conclusion that the
114 kDa complexes observed for CMG2 cysteine mutants correspond to covalent dimers is based on the following experiments. Cells were co-transfected with C218A harbouring an HA tag and C218A harbouring a V5 tag. Immunoprecipitation was performed using an antibody against the HA-tag followed by Western blotting, under non-reducing conditions, with an antibody against the V5 tag. A V5-positive band was observed at
114 kDa, while no band was observed as expected at
55 kDa corresponding to the monomer. The monomer was, however, detected with the HA antibody. Similar co-transfection experiments were performed for the WT protein and for the C230A mutant. A far more pronounced co-immunoprecipitation was observed with the C230A mutant, consistent with the higher abundance of complexes of different sizes, while no co-immunoprecipitation was observed for the WT proteins.
C39F and C218R HFS mutations affect folding of the ectodomain
We next investigated whether the HFS mutations involving cysteine residues had similar effects as those observed above for cysteine to alanine mutations. We included in this analysis the previously described p.C218R mutation found in a homozygous HFS patient (; Hanks et al, 2003
Upon expression of C39F or C218R CMG2, both the mature and the ER precursor forms were observed under reducing conditions (), but the relative abundance of the precursor was higher than for the WT protein, as even more apparent after EndoH treatment (). Partial ER retention was confirmed by immunofluorescence analysis, where we could detect both plasma membrane and ER staining, as particularly indicated by the staining of the nuclear membrane (illustrated for C39F in ).
C39F and C218R HFS mutations lead to inter-disulphide bond formation and ER retention
Moreover, analysis of C39F and C218R CMG2 by non-reducing SDS–PAGE showed that, in contrast to the corresponding alanine mutants, which only formed dimers, patient mutations led to the formation of higher order complexes () which were not seen under reducing conditions (Supplementary ). This reveals the availability of more than one cysteine to inter-molecular disulphide bonding. In patients, the presence of a bulky phenylalanine residue at position 39 or a long charged arginine side chain at position 218 is therefore significantly more disruptive than merely preventing formation of the 39-218 bridge. Folding of the vWA domain was, however, not significantly impaired as revealed by the ability of the mutant CMG2 proteins to bind the anthrax protective antigent (PA) in cell lysates (). WT and the previously analysed ER retained HFS L329R mutant () were used as positive controls (Deuquet et al, 2009
), while D50A, mutated in the vWA ligand binding site (Scobie & Young, 2006
) and L45P affected in vWA domain folding (Deuquet et al, 2009
), were used as a negative control.
Finally, surface biotinylation experiments showed that plasma membrane targeting of C39F and C218R was impaired () as we previously reported for the C218R mutation (Deuquet et al, 2009
). When comparing and , it is again apparent that the mutations found in the patients have a more severe effect on transport than cysteine to alanine mutations.
Ig-like domain HFS mutations lead to aberrant disulphide bond formation and severe ER retention
We next analysed the effects of the mutations observed in Patients 2 and 3 as well as of a previously reported mutation in the Ig-like domain that leads to the insertion of a glutamine residue at position 293 (p.InsQ293) (Hanks et al, 2003
). Upon expression in CHO cells, V310F, C315W, and InsQ293 migrated as essentially a single (), EndoH sensitive () band. Immunofluorescence analysis showed that the mutant proteins were retained in the ER (). Non-reducing SDS–PAGE revealed that these mutants efficiently formed high order inter-molecular disulphide bonded complexes () indicative of the availability of two or more cysteine residues. Finally, surface biotinylation analysis showed that only a small fraction reached the plasma membrane (). ER retention of CMG2 could not be alleviated by the mere removal of the mutation-sensitive Ig-like domain as indicated by the analysis of an isoform of CMG2 (isoform 2, Uniprot Ref P58335-2), which lacks exons 8–11 encoding the six last residues of the vWA domain—including Cys-218—as well as the entire Ig-like domain. CMG2 isoform 2 indeed failed to reach the plasma membrane () and localizes to the ER as demonstrated by immunofluorescence (Bell et al, 2001
Various HFS mutations in the CMG2 Ig-like domain lead to inter-disulphide bond formation and ER retention
That HFS mutations in the Ig-like domain affect folding is supported by analysis of the Ig-like domain model: Val-293 is predicted to be located in the middle of a β-strand. Inserting a residue at this position would lead to an out-of-register β-strand, leading to the exposure of initially buried hydrophobic residues, a change expected to have major impact on folding. Val-310 makes a direct hydrophobic contact with Val-293, in the core of the Ig-like domain. Introduction of a bulkier side chain—phenylalanine versus valine—should not be easily accommodated by the structure.
The CMG2 vWA and Ig-like domains are two independent folding units
The above described observations show that HSF mutations in the Ig-like domain affect folding of this domain. As readout for proper vWA domain folding, we investigated whether the insQ293, V310F, and C315W CMG2 mutants are able to bind the anthrax PA. PA binding was again performed in cell lysates as above (). Ig-like domain mutants were competent of PA binding (), indicating that even if the Ig-like domain did not fold properly, the vWA domain did.
To further evaluate the ability of the vWA and Ig-like domains to fold independently, we investigated whether previously analysed vWA domain HFS mutations (Deuquet et al, 2009
) lead to the formation of inter-molecular disulphide bond formation, used as an indication of defect Ig-like domain folding. Dimeric forms were observed for the previously described L45P, G105D, and I189T mutations, but bands corresponding to trimers and tetramers were absent, suggesting that these mutations affected the formation of the Cys39–Cys218 bond (Supplementary ), but not those in the Ig-like domain.
Altogether, these observations indicate that the vWA and Ig-like domains constitute independent folding units. This is important since it indicates that if Ig-like domain HFS CMG2 mutants were rescued from ER retention, they would be competent for ligand binding and likely functional, and thus that the ER folding/quality control and ERAD components are potential therapeutic targets for HFS.
Proteasome inhibitors can rescue CMG2 in HFS patient cells
In view of a possible therapeutic intervention, we tested the effect of BZ, also known as Velcade, the first therapeutic proteasome inhibitor tested in humans (Adams & Kauffman, 2004
), on CMG2 levels in HFS patient fibroblasts. The effect of BZ () was very similar to that of MG132 () with an increase in the levels of CMG2 for all patients including control. The highest levels of CMG2 were reached for Patient 1. We wondered whether this was because Patient 1 harbours a mutation in the vWA domain, while the missense mutations in Patients 2 and 3 map to the Ig-like domain. To address this possibility, we analysed CMG2 in fibroblasts from a previously described patient (Patient 5) heterozygous for the p.I189T mutation in the vWA domain and a frame-shift mutation in exon 13 (Deuquet et al, 2009
; Dowling et al, 2003
). We have previously shown that the I189T mutation does not affect the structure of the vWA domain per se
, but slows down folding of the domain as to render CMG2 detectable by the ER quality control in the cellular context (Deuquet et al, 2009
). As shown in , the mRNA levels of CMG2 in Patient 5 were very similar to that found in Patient 2. In agreement with our hypothesis, full-length CMG2 was detectable in fibroblasts from Patient 5, albeit at lower levels than for Patient 1 () and this level was increased upon treatment with either MG132 (Supplementary ) or BZ () but to a lesser extent than for Patient 1.
Rescue of CMG2 cell surface expression and function in HFS patient cells upon exposure to proteasome inhibitors
We next analysed plasma membrane targeting of rescued CMG2 by surface biotinylation of control fibroblasts and fibroblasts from Patient 1, 2, and 5. Patients 3 and 4 were not included in this analysis because the sensitivity of surface biotinilyation would not allow the detection of the low levels of CMG2 observed in these patients even after treatment with proteasome inhibitors. Equal amounts of proteins were submitted to precipitation with streptavidin beads. Treatment with MG132 (Supplementary ) or BZ (, Supplementary ) led to an increase in the amount of CMG2 detectable at the plasma membrane both for WT and mutant proteins, albeit to different extents. This indicates that inhibition of the proteasome not only rescued mutant CMG2 from degradation but allowed it to exit the ER and reach the plasma membrane. The increase in WT protein at the cell surface in BZ treated fibroblasts is likely due to both an increase in ER exit and a reduction in surface removal as discussed above for MG132 treatment.
We have previously shown, using anthrax PA as a ligand, that CMG2 undergoes ligand-triggered endocytosis (van der Goot & Young, 2009
) and that this requires ligand-induced signalling via CMG2 (Abrami et al, 2010b
). More specifically, we have shown that the toxin bound to the extracellular domain of CMG2 triggers ubiquitination on the cytosolic tail and that ubiquinitation is required for efficient endocytosis of the ligand-receptor complex via clathrin-coated pits (Abrami et al, 2003
). As evidence for the ability of CMG2 to signal, we have monitored CMG2 endocytosis in BZ-treated Patient 1 fibroblasts upon exposure to anthrax PA as a proof of concept. PA led to rapid degradation of CMG2 showing that the toxin led to receptor endocytosis and targeting to lysosomes (). Moreover, PA triggered ubiquitination of CMG2 not only in BZ-treated control but also patient fibroblasts (), revealing the ability of CMG2 to signal. This ubiquination signal was weakly detectable for the WT protein in the absence of BZ treatment, as expected due to the lower levels of CMG2, and undetectable for the mutant protein (Supplementary ) consistent with the absence of surface CMG2 in Patient 1 cells.