Identification of 31 novel mutations in the F8 gene in Spanish hemophilia A patients: structural analysis of 20 missense mutations suggests new intermolecular binding sites

Deleterious changes in the human F8 gene reduce activity and/or circulating plasma levels of factor VIII (FVIII) protein causing hemophilia A (HA). Patients with HA are classified according to their plasma procoagulant levels of FVIII: severe (< 0.01 IU/mL), moderate (0.01-0.05 IU/mL), or mild (0.05-0.4 IU/mL).¹  The F8 gene is located at Xq28 and spans 186 kb genomic DNA, with 26 exons. Mature FVIII is composed of 2332 residues with an A1-a1-A2-a2-B-a3-A3-C1-C2 domain organization. The 3 A domains are homologous to the plasma ferroxidase, ceruloplasmin, while the 2 C-terminal domains belong to the widespread family of discoidin-like modules. The A and C domains of FVIII are also similar to those of coagulation factor V, and adopt similar quaternary structures in both activated cofactors.^2,3  By contrast, the heavily glycosylated B domain is related neither to the corresponding FV region nor to any other reported protein. Domains A1/A2, A2/B, and B/A3 are connected by relatively long linkers rich in acidic residues, which are termed a1 to a3. None of these peptides has homologs in ceruloplasmin, while a2 and a3, but not a1, have functional counterparts in FV.

FVIII circulates in plasma bound to the carrier protein, von Willebrand factor (VWF), as multiple heterodimeric forms containing various lengths of the B-subunit covalently attached to the A1-a1-A2-a2 region (the heavy chain) and an a3-A3-C1-C2 light chain. Cofactor activation requires thrombin-mediated cleavage of peptide bonds Arg372-Ser373 (in the a1-A2 linker), Arg740-Ser741 (a2-B), and Arg1689-Ser1690 (a3-A3); exosite-mediated thrombin interactions with the acidic a1-a3 peptides appear to be essential for these proteolytic cleavages. The resulting A1-a1/A2-a2/A3-C1-C2 heterotrimer no longer interacts with VWF but binds to the serine protease, FIXa, on negatively charged phospholipid membranes provided by activated platelets. This calcium-dependent complex of FVIIIa, FIXa, and acidic phospholipids, also termed intrinsic Xase, activates FX via cleavage of a single peptide bond (Figure 1A; recently reviewed by Fay and Jenkins⁴  and Graw et al⁵ ). In addition to its role in cofactor activation, the acidic region a1 is involved in important interactions with the Xase substrate, FX.^6,7 

The crystal structure of the FVIII C2 domain⁸  and those of human ceruloplasmin⁹  and bovine inactivated FVa (FVai; Adams et al² ) have provided valuable templates for developing FVIIIa models of increased accuracy (see, eg, Pemberton et al,¹⁰  Liu et al,¹¹  Stoilova-McPhie et al,¹²  and Autin et al¹³ ). The recently reported crystal structure of human FVIIIa at low resolution³  confirmed that the side-by-side arrangement of C1 and C2 domains previously observed in bovine FVai²  and incorporated in a recent modeling study¹³  is the most likely conformation for both cofactors, at least in their uncomplexed states. On the other hand, identification of novel mutations leading to HA and their thorough analysis has helped to define areas involved in important interdomain contacts, membrane binding, as well as in interactions with cognate FIXa and substrate, FX.^4,5  Therefore, characterization of F8 mutations is important for a better understanding of the structure-function relationship of the FVIII molecule. As of January 2007, more than 580 missense mutations have been deposited with the HAMSTeRS database,¹⁴  and recent investigations have added a large number of mutations in different populations (see, eg, David et al,¹⁵  Guillet et al,¹⁶  Repesse et al,¹⁷  and Rossetti et al¹⁸ ).

Here, we characterized 31 novel small mutations in the F8 gene, including 11 nonsense, frameshift, and splicing mutations as well as 20 missense mutations. We also provide a thorough evaluation of the probable structural and/or functional defects caused by these residue replacements.

We analyzed the F8 gene in 267 unrelated patients with clinical and laboratory confirmation of HA after obtaining their informed consent in accordance with the Declaration of Helsinki. Approval was obtained from the Ethical Committee from the Hospital of Sant Pau institutional review board for these studies. One hundred ninety-three patients were classified as severe HA and the remaining 72 were moderate to mild cases. Moreover, the patients were defined as familial (positive) or isolated (sporadic) according to their available pedigree information.

DNA was isolated from peripheral blood samples using the salting-out method. After excluding hemophiliacs with intron 1 and intron 22 inversions and with large deletions, all 26 exons and exon-intron boundaries of the F8 gene from the remaining patients were amplified using primers and polymerase chain reaction (PCR) conditions essentially as in David et al¹⁹  and as deposited in the HAMSTeRS database.¹⁴  Promoter analysis was performed as previously described.²⁰  The amplified fragments were purified with QIAquick columns (Qiagen, Isaza, Barcelona, Spain), and analyzed by direct forward and reverse sequencing using a DNA sequencing kit (PerkinElmer-Applied Biosystems, Madrid, Spain) on an ABI PRISM 3100 Avant DNA automatic sequencer (Applied Biosystems, Madrid, Spain).

Nucleotide numbering was based on the cDNA sequence (GenBank no. NM_000132.2.²¹ ), with nucleotide +1 corresponding to the adenine of the ATG translation initiation codon. The amino acid numbering system was based on the mature protein, while residues from the 19-residues–long signal peptide were numbered in reverse. Thus, the initial methionine was numbered as −19 and the first alanine of the mature protein was numbered as +1. The nomenclature used to describe mutations is that of den Dunnen and Antonarakis.²² 

For residues surrounding activation cleavage sites, we used the standard Schechter and Berger nomenclature²³ : substrate residues are denoted P_m, …, P₁, P_1′, … P_n′, with the scissile peptide bond located between P₁ and P_1′.

We ascertained that a given missense mutation is causally related to the patient's HA phenotype in accordance with the following criteria: (1) the mutation cosegregated in other affected individuals or carriers from the same family; (2) it was not detected in a panel of 100 X-chromosomes from the general Spanish population; (3) no other nucleotide change was found in the F8 coding region or in exon-intron boundaries; (4) the mutated residue was conserved or conservatively replaced in FVIII from different species, and in most cases in FV and/or ceruloplasmin; and (5) finally, most mutated residues are buried in the protein core, and their replacement would either disrupt networks of buried hydrogen bonds, create destabilizing internal cavities, and/or lead to clashes with neighboring residues. The predicted impact of missense mutations in FVIIIa as well as in equivalent positions of FVai and ceruloplasmin was assessed with PolyPhen (http://genetics.bwh.harvard.edu/pph/, Harvard University) and CUPSAT (http://cupsat.tu-bs.de/, Cologne University).²⁴ 

We separately submitted alignments of the A1-a1-A2-a2-A3 and C1 domains of FV, FVIII, and ceruloplasmin from different species to the Swiss-Model server.^25,26  For the A1-a1-A2-a2-A3 substructure, we used as a template the crystal structure of ceruloplasmin (PDB code 1KCW),^9,27  while the FVIII C1 domain was modeled based on the homologous C2 domain (Pratt et al⁸ ; PDB code 1D7P). (Proteolytic inactivation of FVa leads to the loss of domain A2 and a concomitant destabilization of nearby domain A1, and therefore the structure of ceruloplasmin represents a better template for modeling the A1-A2-A3 trimer.) The cysteine residue covalently linked to Cys2296 was removed from the 1D7P entry, and the partial models for FVIII A1/A2/A3, C1, and C2 domains were superimposed on the structure of inactivated bovine FVa (PDB 1SDD). Coordinates for bound calcium and copper ions were taken from the latter structure. After manual readjustment of side chains at the interdomain interfaces, especially those involved in cation-binding sites, the resulting trimer was finally minimized with CNS (http://cns.csb.yale.edu/v1.1/, Yale University). FX models were constructed using the partial crystal structures for EGF1-EGF2 catalytic domain (1XKA)²⁸  and Gla domain (1IOD),²⁹  as well as the solution structure of the Gla-EGF1 tandem (1WHF)³⁰ ; the structure of bovine chymotrypsinogen A was selected to generate FX zymogen conformation (2CGA).³¹  FIXa was modeled essentially following the structure of the porcine cofactor³² ; the Gla domain was derived from 1J34³³  and 1NL0.³⁴  Similar strategies have been followed by other authors to generate FIXa/FX models.^13,29,35  Model quality was assessed with MOLEMAN2 (http://xray.bmc.uu.se/cgi-bin/gerard/rama_server.pl, Uppsala Software Factory, Uppsala, Sweden); for FVIIIa, that is, 89% of all residues were found in core regions of the Ramachandran plot. Docking was performed with ZDOCK (http://zdock.bu.edu/, Boston University) and structure figures were generated with PyMOL (http://www.pymol.org, DeLano Scientific, Palo Alto, CA).

After submission of this work, a 3.7-Å crystal structure of human FVIIIa was presented.³  Although the coordinates for the latter are not yet available, the reported similarity to the previous FVai structure validates all relevant features of the current “compact” model. In particular, the side-by-side arrangement of the C1 and C2 domains is supported by a large hydrophobic/aromatic interface between domains A3 and C1, while C2 lacks important interactions with the remaining domains (Figure 1B). These features are in line with an overall conservation of residues that form the A3-C1 interface in the 2 cofactors.^2,3  Moreover, experimental evidence suggests absence of important contacts between the A2 domain and the cofactor light chain, while interactions with A1 account for 90% of the interchain binding energy.³⁶  Altogether, current evidence would seem to disfavor a FVIIIa model where domain C1 is located on top of C2 (and thus removed from the phospholipid membrane),¹²  and which has been commonly used to rationalize the effect of point mutations on FVIIIa structure.

By contrast, the compact arrangement of globular domains is likely to reproduce the conformation of membrane-bound cofactors Va and VIIIa. We note that topologically equivalent, exposed hydrophobic residues from both discoidin-like domains of FVa^37-39  and FVIIIa^11,40,41  appear to be involved in membrane binding (for a review on structures and interactions mediated by these discoidin-like domains, see Fuentes-Prior et al⁴² ). In particular, the exposed Tyr1956/Leu1957 pair in the first spike of the human FV C1 domain corresponds to human FVIII residues Lys2092 and Phe2093 (Figure 2B boxed), while arginine residues at positions 2023 and 2027 are conserved as Arg2159 and Arg2163 in human FVIII. As demonstrated for the FVa residues,³⁸  the side chains of these topologically equivalent, solvent-exposed FVIIIa residues (eg, Lys2092, Arg2159, Arg2163) could interact with negatively charged phosphatidylserine headgroups,¹¹  while Phe2093 could be inserted into the apolar membrane core, stabilizing the side-by-side arrangement on procoagulant phospholipid membranes (see Figure 1B, where side chains of some of these residues are shown in green).

Finally, we would like to stress that models for FVIIIa and FVIIIapi•FIXa complexes based on the bovine FVai structure have been developed by Autin et al following a similar strategy.¹³  These models, however, have not been used to date to predict the impact of missense mutations in the F8 gene.

We identified 137 small mutations in the 267 Spanish HA patients studied in the current work (the remaining 130 patients had intron 1 and 22 inversions and large deletions). Ninety-one of these mutations were missense, 10 were of the nonsense type, 29 were small deletions or insertions, and 7 were splice site mutations (data not shown; available upon request). In particular, we detected 31 novel changes (none deposited to date in the HAMSTeRS database nor reported in recently published articles), including 2 nonsense, 7 frameshift, and 2 splicing mutations (Table 1). Three of these hemophiliacs have developed inhibitors to FVIII.

The remaining 20 mutations were of the missense type for which the criteria described in “Analysis of missense mutations” apply; a detailed description of these mutations is shown in Table 2. According to available information, none of the patients from this group developed inhibitors. All globular domains of the activated cofactor were affected by these novel missense mutations (see Figure 1B, where mutated residues are shown with all their side chain atoms in the FVIIIa model). In line with the greater functional relevance of domains A1 and A2 for FIXa/FX binding, 60% (12/20) of the novel mutations affect residues within one of these domains (6 changes in each domain). Two further mutations were identified in domain A3, 4 in C1, and 2 in the second discoidin-like domain. To illustrate the location of affected positions, partial multiple alignments of FV, FVIII, and ceruloplasmin sequences from different species around all mutated residues in domain A1 are presented in Figure 2A. Similarly, partial alignments around mutations found in C1 and C2 domains from both coagulation factors are given in Figure 2B. These changes are representative of the 20 novel missense mutations identified in the current work, and range from strictly conserved, buried residues that belong to major secondary-structure elements (eg, p.Asp167Asn) to solvent-exposed loop residues that are unique to FVIII (eg, p.Ala375Ser).

Further, and based on structural analysis (Figures 3,4; Tables S1,S2, available on the Blood website; see the Supplemental Materials link at the top of the online article) and on consideration of previously reported mutants (Table 3), we interpreted missense mutations as those likely to result in type I or quantitative FVIII deficiency and as those likely to affect cofactor function—type II or qualitative cofactor deficiency. The majority of identified mutations within the first group replace well-conserved, fully or partially buried residues, which is similar to data recently reported by other authors.^14-17  Because these residues engage in multiple, mostly intradomain contacts, the ultimate result of these mutations would be an incorrectly folded and unstable cofactor molecule that is either poorly secreted and/or rapidly removed from circulation. There are, however, differences in the potential impact of a particular mutation as a function of its location (eg, whether the residue is part of a regular secondary structure element or found within a loop) and the nature of the replacing residue (“Discussion”).

Within the second group, 4 of the identified mutations affect residues that are exposed on the cofactor surface, which therefore do not engage in important intradomain or interdomain interactions (see Figure 1B, where labels for these residues are boxed). Further, these mutations map to regions that are only poorly conserved in FV/FVIII and ceruloplasmin (Figure 2), also suggesting a functional role. These considerations together with functional evidence presented for some related replacements allow us to hypothesize that these 4 mutants would lead to properly folded but functionally defective cofactor molecules.

We have detected 31 novel mutations in the F8 gene in Spanish HA patients, 9 of which were nonsense and small insertions and deletions. The deleterious mechanisms of these mutations are in general obvious given that they create premature termination codons leading to truncated F8 mRNA and FVIII protein.⁴³  A particularly interesting case is mutation c.4155_4195dup, which introduces a 41-bp tandem duplication starting at nucleotide 4056 in exon 14. This would in turn predict a stop codon 4 triplets after the beginning of the duplication (ie, following residue Thr1366). The truncated mutant protein would lack domains C1 and C2 along with the C-terminal part of the B domain and therefore, critical elements required for VWF association and membrane binding. As expected, all but one of these novel mutations were associated with a severe phenotype. However, one patient with mutation c.2766delC predicting a stop codon after residue Ser903 showed moderate HA. Transcription errors or ribosomal frameshifting may result in the production of full-length FVIII in this case, as previously reported for other mutations (Bogdanova et al⁴⁴  and references therein). The 2 novel splicing mutations were detected in patients with severe bleeding and are located at IVS13a-2g and IVS15g-1c affecting the specific consensus acceptor site. No RNA sample was available from these patients to confirm the effect of their mutations on splicing.

The predicted structural implications of the type I and type II novel missense changes are separately discussed below with emphasis on those positions where mutations of the same or topologically equivalent residues have previously been identified (Figure 3; Table 3; Table S1). Although the structural impact of these mutations was inferred from inspection of an FVIIIa homology model, quality and relatedness of substructures used as templates (crystal structures of human ceruloplasmin, human FVIII C2 domain, and bovine FVai) strongly support our analysis. In particular, most mutated residues are well conserved in FV and/or ceruloplasmin, and the effect of equivalent substitutions in the template structures can be directly assessed (Table S2).

The apparently conservative replacement of the strictly conserved Asp167 (Figure 2A) by an asparagine would disrupt a complex network of stabilizing hydrogen bonds centered on Asp167 carboxylate (Figure 3A). All these H-bonds could not be simultaneously donated by the carboxamide group of an asparagine side chain, thus compromising domain A1 stability. Several different mutations have been previously identified at this and at topologically equivalent positions (Table 3). All well-characterized mutations were associated with severe phenotypes, as in p.Asp167Asn, with exception of the conservative replacement Asp→Glu.²⁰ 

Residues Asn235 and Asn618, which occupy topologically equivalent positions in domains A1 and A2, were found mutated to serine and isoleucine, respectively. The carboxamide groups of these strictly conserved asparagine side chains engage in strong hydrogen bonds with main-chain atoms from distant polypeptide regions (compare Figure 3C and 3I). It is therefore not surprising that replacements by serine (instead of Asn235) or isoleucine (at position 618) destabilize the domain structures, and the patients presented severe and moderate HA, respectively. Again, disease severity is comparable with other mutations affecting these and the equivalent residue, Asn1922 (Table 3). Of note, the recombinantly expressed mutant p.Asn618Ser FVIII accumulates intracellularly,⁴⁵  and FVIII inhibitor development has been reported for patients carrying the Asn618Ser⁴⁶  and Asn1922Asp mutations.⁴⁷  Similar defects in protein secretion and inhibitor development might be expected for other mutations affecting Asn235, Asn618, or Asn1992, including the 2 novel mutations identified here.

Mutation p.His2082Asp represents the last example of a replacement that affects a polar residue. In the compact FVIIIa model, His2082 is essentially buried at the A3-C1 interface (Figure 1B); its imidazole ring engages in favorable interactions with residues from both domains (Figure 3K). Replacement by a negatively charged aspartate side chain would thus compromise the quaternary cofactor structure. On the other hand, formation of a salt bridge between Asp2082 and nearby Arg1869 might alleviate the consequences of this mutation, explaining the mild phenotype. We note that this residue has been alternatively hypothesized as part of a Mn²⁺-binding site in the quaternary arrangement of FVIIIa that locates domain C1 close to A1.⁴  Therefore, His2082 and neighboring residues represent a ground for a testable hypothesis about the prevalence of each conformation in circulating VWF-bound FVIII, its activated form, and FIXa-bound cofactor.

Six of the 20 novel missense mutations identified in the current work either introduce a proline instead of a conserved or conservatively replaced hydrophobic residue (p.Leu176Pro, p.Leu242Pro, and p.Gln2087Pro; Figure 2), or replace proline by arginine (at positions 477 and 550) or leucine (at 2153). The consequences of these mutations, however, depend on the precise location of the mutated residue. For instance, proline replacement for the strictly conserved Leu176 found in the middle of a major β-strand (Figures 2A,3B) would interfere with secondary structure formation. In addition, this change not only creates a cavity in the protein core, but also leads to clashes with residues from a neighboring strand (eg, Ser157). Thus, mutation p.Leu176Pro compromises cofactor folding at several levels; accordingly, it is associated with a severe bleeding phenotype, most likely related to absence of circulating FVIII. Severe HA results also from replacement of the strictly conserved Gln2087 by a proline (Figure 2B), as it disrupts an interaction network that clamps together sequentially distant polypeptide stretches of the C1 domain via strong H-bonds (Figure 3L). Interestingly, substitution of the equivalent Gln2246 by a polar arginine has been reported associated with moderate bleeding.⁴⁸ 

By contrast, the Leu242Pro replacement might be relatively well tolerated, as it does not affect a regular secondary structure element (Figure 2A). Moreover, a proline at this position retains favorable contacts with hydrophobic core residues while producing only minor clashes with the preceding Ser241. Finally, residues surrounding Leu242 are partially exposed and could be displaced to accommodate the mutant side chain (Figure 3D). These observations are in line with the moderate bleeding episodes described in this patient and with the mild disease form associated with the more conservative substitution, p.Leu625Val.⁴⁹ 

On the other hand, insertion of bulky, basic arginine side chains instead of strictly (Pro477) or well-conserved (Pro550) proline residues that are fully buried in the protein core (Figure 3G,H, respectively) would result in major disruption of the A2 domain fold, explaining the severe phenotypes of these patients. Similar phenotypes are associated with previously reported mutations of the Pro550-equivalent residue, Pro1854 (Table 3). We also note that these residues are found within repeated, well-conserved Gly-(Phe/Leu/Ile)-(Leu/Ile)-Gly-Pro motifs that appear to be critical for proper folding and trafficking of A domains.⁵⁰ 

Although apparently less deleterious, the strictly conserved Pro2153 (Figure 2B) is also buried in the protein core, and introduction of a bulkier, albeit aliphatic side chain (Leu) leads to major clashes with other core residues (Figure 3N). Gross fold destabilization would explain the severe bleeding associated with this mutation. Previous mutations of residue Pro2153 to Gln^48,51  and Arg^52,53  have been reported. It could have been expected that introduction of a polar side chain at this position has even more dramatic consequences than the Pro→Leu replacement identified here, and indeed p.Pro2153Arg is associated with severe HA.^52,53  However, replacement by a glutamine results in mild⁵¹  or moderate HA,⁴⁸  which is explained because the carboxamide group of Gln2153 could engage in H-bonding interactions with side chains of nearby polar residues (Figure 3N) compensating for steric clashes. The latter mutation has been previously associated with a reduced FVIII binding to von Willebrand factor.⁵¹  Because Pro2153 is essentially buried in the domain core, it is unlikely to engage in direct contacts with the carrier protein. On the other hand, we note that the neighboring residue, Arg2150 (Cα-Cα distance: 10 Å), is part of an important VWF-binding epitope.⁵⁴  The observed impairment in VWF binding caused by replacement Pro2153Gln most probably results from enforced displacements of Arg2150 and surrounding residues. Conceivably, mutation p.Pro2153Leu also leads to impaired VWF binding and rapid cofactor clearance from blood, contributing to the severe phenotype.

Two mutations that affect glycine residues conserved in FVIII, p.Gly261Asp (Figure 2A) and p.Gly455Val, result in a mild bleeding phenotype. We note that small side chains are well tolerated at these positions given that topologically equivalent residues in FV and in other FVIII domains are Ala, Ser, or Thr. However, insertion of a charged aspartate at position 261 is unfavorable, mainly because of electrostatic repulsion from the nearby Glu265 (Figure 3E). Similarly, the replacement of Gly455 by Val would predict several, albeit minor, clashes with side chains of both nearby and sequentially distant residues (Figure 3F). Nevertheless, in both cases neighboring residues possess enough freedom to rotate away and alleviate these unfavorable interactions. In addition, introduction of Asp261 could be partially compensated by donating hydrogen bonds to the main-chain nitrogen atoms of Val201 and/or Phe202. The more severe phenotypes associated with 2 further mutations at position 455 (p.Gly455Glu, Waseem et al⁵⁵  and p.Gly455Arg, Freson et al⁴⁹ ) are explained by unfavorable van der Waals contacts of these bulkier, polar side chains in a mostly aromatic/hydrophobic environment. Moreover, the negatively charged side chain of Glu455 would be repelled by the Asp459 carboxylate (Figure 3F).

Replacement of the partially buried, strictly conserved Met2124 by a valine (Figure 2B) would minimally affect domain structure (Figure 3M), and the proband presented mild bleeding. Because of the conservative nature of mutations p.Ala1701Val, p.Phe1743Leu, and p.Ile2317Phe, they would also seem to be less detrimental. However, in these cases, the precise location of the mutated residue determines disease severity. Although introduction of the bulkier Val side chain at position 1701 leads to clashes with several residues (Figure 3J), the well-exposed Gln1778 side chain might rotate away to accommodate the extra methyl groups from Val1701. The replacement Ala1701Val, associated with mild HA, would have less dramatic implications than the p.Ala1701Asp mutation described in a severe case.¹⁵ 

Similarly, mutation c.5286T>A that predicts replacement of the strictly conserved Phe1743 by the aliphatic leucine in a poorly conserved loop would seem to be well tolerated (Figure 3J). Indeed, mutation c.5284T>C leading to the same amino acid exchange has been previously reported in a patient suffering from mild bleeding.⁵⁶  However, the fact that our proband presented moderate HA suggests that lack of loop-stabilizing interactions mediated by the Phe1743 side chain might disrupt the proper structure of the convoluted Phe1738-Tyr1762 loop. We hypothesize that this could in turn affect interactions with domain C1, or intermolecular contacts with FIXa (Figure 1B).

By contrast, mutation p.Ile2317Phe introduces a bulkier phenyl moiety instead of the strictly conserved aliphatic side chain at position 2317 (Figure 2B), and would lead to severe clashes with other conserved residues (Figure 3O). Aside from destabilizing important structural elements, loop rearrangements enforced by the inserted Phe2317 benzene ring could compromise interactions with VWF and/or membrane binding mediated by the C2 domain explaining the moderate bleeding phenotype associated with this mutation.

Mutation p.Ala375Ser conservatively replaces the P_3′ position within the activation cleavage site that separates domains A1 and A2. Previously, several mutations of the critical P₁ residue Arg372 have been reported^57-61  with phenotypes ranging from mild to severe. Replacement of the P_1′ (p.Ser373Leu and p.Ser373Pro)^62,63  and P₂ residues (p.Ile371Thr)¹⁶  also compromises cleavage at this site and results in mild bleeding tendencies. Inspection of the thrombin structure,^64,65  together with work conducted using synthetic peptide libraries⁶⁶  and phage display,⁶⁷  suggests a minor influence of positions on the nonprimed side of the P₁-P_1′ scissile peptide bond. In fact, serine appears to be slightly preferred over Ala at P_3′ position, at least for thrombin inhibitors. However, these studies do not consider the key role played by interactions between thrombin exosites and acidic a1-a3 peptides during FVIII recognition and processing (eg, Glu331 to Asp363 for cleavage after Arg372). It is therefore conceivable that the mild bleeding phenotype associated with mutation p.Ala375Ser is due to a slightly decreased rate of cofactor activation.

Three additional mutations associated with mild or moderate bleeding phenotypes appear to either extend previously characterized or define novel protein-protein binding sites, based on their proximity to known interaction regions (Figure 1B). Mutation p.Pro64Arg replaces a proline conserved in FV and FVIII by a basic arginine side chain. By contrast, the topologically equivalent position is occupied by a glutamate or an aspartate in ceruloplasmin (Figure 2A), and the polar glutamate side chain is fully exposed in the crystal structure of the ferroxidase.⁹  Accordingly, inspection of the FVIIIa model revealed that an arginine side chain would be tolerated at this position. Because most residues within the long, irregular loop Thr55-Pro67 are well exposed, and considering their proximity to a1 (Figures 1B,5), it is conceivable that this area represents a binding site for substrate FX, which is partially masked by the mutant Arg64 side chain. Similar considerations apply to the p.Ala62Asp mutation, which has been reported in mild HA patients.^16,17  Finally, and also in line with the hypothetical role of this region as a FX-interactive site, more dramatic loop rearrangements due to in-frame deletion of Pro66 or the Arg65/Pro66 pair lead to moderately severe and severe HA, respectively.⁶⁸  Considering the predicted distance from the phospholipid membrane (Figure 1B), we speculate that this loop interacts with EGF2 and/or serine protease domains of substrate FX.

Mutation p.Gly494Val maps to a particularly mobile, surface-exposed loop,⁹  which represents the major inhibitor epitope in FVIII A2 domain.⁶⁹  Interestingly, both residues preceding and following Gly494 have been implicated in FX binding, in particular the basic cluster Arg489/Arg490/Lys493.^4,70  It is conceivable that a bulkier side chain at position 494 would impair these FVIIIa-FX interactions, thus compromising substrate recognition. Moderate bleeding has also been reported for mutant p.Gly494Ser,⁶⁸  also in line with the proposed involvement of this loop in FX activation. We note that a discrepancy between antigen and FVIII activity levels for the latter mutant (23% and 4%, respectively) suggests its involvement in intermolecular interactions. As alternative to FX binding, and because the 484 to 509 loop represents a major binding site for low-density lipoprotein receptor–related protein (LRP),^4,71  our mutation p.Gly494Val could reduce the half-life of circulating FVIII by enhancing binding to its hepatic clearance receptor.

Finally, mutation p.Asp2267Gly eliminates a partially exposed aspartate side chain within a loop that is poorly conserved between FV and FVIII (Figure 2B). One of the carboxylate oxygens from Asp2267 accepts hydrogen bonds from the main-chain nitrogen atoms of 2 consecutive loop residues (Figure 4). The topologically equivalent position, however, is occupied by Gln/Arg in FV (Figure 2B), and in spite of this fact the loop conformation is almost identical in all the experimentally solved structures of natural or recombinant C2 domains, both from FV^2,72  and FVIII^8,41  (Figure 4). Therefore, Asp2267 appears to be dispensable for C2 structural stability. Inspection of the model suggests that most residues within the convoluted, bowtie-shaped Gln2266-Lys2281 loop are well exposed in FVIIIa, and are thus available for intermolecular interactions (Figure 2B). In this regard, we also noticed that this region is the most variable between FV and FVIII discoidin-like domains (Figure 4), pointing to a cofactor-specific role. In the light of its neighborhood to other residues previously associated with VWF binding^11,73  (Figure 5), this area might represent an extension of the well-characterized VWF-binding site(s). Alternatively, and given the relatively close distance to the phospholipid membrane (Figures 1B and 5), residues from this loop and in particular Asp2267 might contact the Gla or EGF1 domains from substrate FX.

In conclusion, we characterized 31 novel mutations in the F8 gene including a further evaluation and interpretation of 20 novel missense mutations combining bioinformatics tools and current knowledge of FV/FVIII structure/function relationships. Our thorough analysis of the novel mutations provides further insights into the causes of hemophilia A, contributes to a better understanding of the functional consequences of F8 mutations, and suggests novel FX-interaction areas on the cofactor surface. Future investigations using recombinant proteins should validate these results, and refine current models for FVIIIa and FVIIIapi•FIXa (Xase) structures.

We thank all patients and their relatives for participating in this study. We are indebted to A. Alonso, C. Altisent, M. T. Calvo, L. Eciolaza, J. Fontcuberta, R. González Boullosa, F. Lucía, M. A. Lopez Aristegui, M. F. López Fernández, J. Martorell, J. Monteagudo, M. Moreno, R. Pérez Garrido, M. Quintana, J. Rosell, C. Sedano, J. Tusell, F. Velasco, and A. Villar for clinical information about these patients.

This work was supported by Fundació Catalana d′Hemofilia and grant SAF2004-00543 from Spanish Ministerio de Educación y Ciencia.

National Institutes of Health

Contribution: A.V., P.F.-P., and E.F.T. designed the research, analyzed data, and wrote the paper; P.F.-P. and M.A.C.-R. performed modeling and produced figures; M. Baena, M.C., and M.D. performed experiments to identify mutations; M. Baiget gave intellectual support and discussion on the paper; E.F.T. was primarily responsible for this work.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Eduardo F. Tizzano, Department of Genetics, Hospital de la Santa Creu i Sant Pau, Sant Antoni Ma. Claret 167, 08025 Barcelona, Spain; e-mail: etizzano@santpau.es.