Factor XIII (FXIII) is the last enzyme in the clotting cascade. Its main function is to convert the loose fibrin polymer into a firm, highly organized, cross-linked structure with increased tensile strength, firmly anchored to the site of the wound and possessing an in-built resistance to fibrinolysis. In FXIII deficiency, standard clotting tests are normal, as the clotting end point is not affected by the absence of FXIII. It is the quality of the clot which is abnormal. Unless this is assessed, the diagnosis may be missed. Soon after its discovery by 94), FXIII was aptly named: the ‘fibrin stabilizing factor’ or ‘FSF’. Fibrin formed in the absence of FSF was ‘unstable’: it dissolved in weak acids, weak bases and 5 m urea. Addition of a small amount of plasma to the system ‘stabilized’ the fibrin, it was no longer soluble in these reagents and it was also more resistant to fibrinolysis. Solubility of clots in 1% monochloroacetic acid or in 5 m urea still forms the basis of the standard laboratory screening test for inherited FXIII deficiency. Initially it was believed that FSF combined with fibrin stoichiometrically, acting as a kind of glue, sticking molecules of fibrin together ( 64; 67). Later it became obvious that FSF was an enzyme ( 18; 66; reviewed in detail by 15) and the cross-link introduced into fibrin by FSF was found to be the ɛ(γ-glutamyl)lysine link (reviewed by 81), identifying FXIII as a member of the transglutaminase family of enzymes (see reviews by 36; 1). FXIII is the only transglutaminase found both intra- and extracellularly and the only one requiring thrombin as well as calcium for activation. Thus, in common with other clotting factors, it exists as a pro-enzyme ( 68). In plasma, FXIII circulates in the form of a tetramer composed of two catalytic A subunits bound to two carrier protein B subunits (A2B2)( 97). FXIII is found in plasma, platelets and monocytes ( 85, 84). Intracellular FXIII is a dimer of two catalytic A subunits (A2). In inherited FXIII deficiency the A subunit is absent from plasma, platelets and monocytes. The plasma level of the B subunit is usually reduced, and very rarely both A and B subunits are absent ( 35). The clinical relevance of FXIII became apparent 16 years after its discovery, when 27) described the case of a boy with a severe bleeding diathesis in whom the only abnormality in the clotting tests was the solubility of his clots in 5 m urea. His clots became insoluble when some normal plasma was mixed with his plasma in vitro. He formed insoluble clots, after blood or plasma transfusion which also controlled his bleeding temporarily. Since that original description, over 200 cases have been reported from all parts of the world. Most patients with inherited FXIII deficiency suffer from a severe, lifelong, crippling bleeding diathesis with a very high risk of death in early life from intracranial haemorrhage. In a minority of patients defective wound healing has been reported. Women with the deficiency are unable to carry a pregnancy to term and habitual abortion is usual ( 31). Inherited FXIII deficiency affects all races and both sexes equally. Inheritance is autosomal recessive. The genes for both the A and B subunits were cloned about 10 years ago (detailed in 15). Five years ago only five mutations were tentatively identified as probably being responsible for the deficiency ( 15). In recent years developments in DNA technology have facilitated detailed studies of the FXIII gene in families with inherited deficiency as well as in normal individuals. Thirty-six different mutations resulting in FXIII deficiency have now been identified. Studies on the structural and functional consequences of the various mutations and the normal polymorphisms in the gene have resulted in a much better understanding of FXIII function. Discussion of these exciting developments in the field will form the major part of this review. The three-dimensional structure of FXIII was first studied by electron microscopy of rotary-shadowed molecules ( 19). The A dimer was found to consist of two globular particles of about 6 × 9 nm in size. The B subunit appeared to be a filamentous flexible strand with kinks, being around 30 nm long and 2–3 nm in diameter. The electron microscope images of the A2B2 tetramer revealed particles which are about 10 × 12 nm across in size, compact and slightly oblong ( 19). However, the individual A and B subunits could not be distinguished. More recently, the three-dimensional X-ray crystallographic structure of recombinant FXIIIA subunit, produced in yeast, was resolved ( 115). These X-ray crystallography studies showed that the A subunit folds into four distinct, sequential domains. From the N-terminus activation peptide at residues 1–37, the domains are the beta sandwich (residues 38–183), the catalytic core region (residues 184–515), beta barrel 1 (residues 516–627) and beta barrel 2 (residues 628–730). These structural analyses have been very clearly documented and the reader is referred to the primary literature for further information ( 115, 116). It is thought that the core region is composed of two subdomains, with the boundary at residue 332. The two beta barrel domains are thermally stable whereas the beta sandwich and the two subdomains of the catalytic core are thermolabile ( 63; 26). The thermolabile domains appear to be important in dimerization of the A subunits through intra- and/or intermolecular interactions ( 63). Cys314 has been known to be essential for FXIII activity for many years ( 34). It is now clear that the catalytic mechanism of cross-linking by transglutaminases requires the ‘catalytic triad’ formed through hydrogen bond interactions between Cys314, His373 and Asp396 ( 88). These three residues are absolutely conserved among all members of the transglutaminase family. In fact when the sequences of all 16 members of the transglutaminase family are aligned, 179 residues are found to be highly conserved (i.e. identical in at least 13 of the sequences), and the majority of these (77%) lie within the core domain ( 116). Very recently, two non-proline cis peptide bonds between Arg310 and Tyr311 (close to the active site residue Cys314), and Gln425 and Phe426 (at the dimerization interface) have been proposed to be important for FXIII function ( 113). Further evidence for the structure/function relationship for FXIIIA, and of residues important for function, comes from the missense mutations identified in FXIII-deficient families. These are discussed in detail below. The allelic variance of FXIIIA has been known for many years. Initial studies, using agarose gel electrophoresis, had indicated that FXIIIA was encoded by two common alleles ( 13). Subsequently, isoelectric focusing in denaturing polyacrylamide gels suggested there were four common alleles ( 106; 117). More recently, molecular analysis of the FXIIIA gene has enabled the identification of a number of normal variants ( 105, 104; 6, 2); these are summarized in Table I. There are five recognized sequence changes which give rise to ‘normal’ polymorphic amino acid residues in the FXIIIA polypeptide: Val34Leu, Tyr204Phe, Leu564Pro, Val650Ile and Glu651Gln ( Table I; 2). There are also two silent polymorphisms, in codons 331 and 567 ( 6 ). The allele frequencies for all the polymorphisms have been determined in the Caucasian and Japanese populations ( Table I). Interestingly, the Leu34 allele seems to be rare in the Japanese but its prevalence in Caucasians is estimated from three independent studies to be 0.16–0.28 ( 104; 61; 2). The allele frequencies for codons 204, 564, 650 and 651 appear to be similar in the two populations. The silent polymorphisms in codons 331 and 567 are present only in Caucasians, being absent from the Japanese population. The polymorphism at codon 567 appears to be in linkage disequilibrium with Pro564. The sequence GAG at codon 567 has been found only in individuals who are homozygous Pro564 or heterozygous Pro564/Leu564, and never in individuals who are homozygous for Leu564 ( 104; 2). A large number of haplotypes is possible when all the ‘normal’ variants present in the FXIIIA gene are considered. Indeed, over 25 unique haplotypes have been identified in the normal population, the most prevalent being Val34-Tyr204-Pro564-Val650-Glu651 ( 104; 2 ). In addition to the polymorphic sequences described above, a short tandem repeat (STR) polymorphism is also present in the 5′ promoter region of the FXIIIA gene, giving rise to 14 different alleles ( 90). Thirteen of these alleles contain 4–16 repeats of the AAAG sequence while the fourteenth contains four repeats and a two base deletion. This STR polymorphism is useful in investigating disease segregation in FXIII-deficient families and hence genetic diagnosis. The application of this STR has been reported in prenatal diagnosis in a FXIII-deficient family ( 55). Very recent studies have demonstrated a relationship between the different FXIIIA gene polymorphisms and FXIII activity. The Leu34 variant gives rise to higher plasma FXIII activity, and specific activity (activity per unit level; 2), compared to the Val34 variant ( 52; 59; 2). Since the amino acid 34 is only three residues away from the thrombin cleavage site it is possible that the Val34 and Leu34 variants differ in their ability to be activated by thrombin and how they release the activation peptide. The Leu564 variant also induces higher specific activities compared to molecules containing Pro564 ( 2). The Phe204 variant, which is rare in the normal population, results in FXIII molecules with a lower specific activity compared to Tyr204-containing molecules ( 2). In addition, some genotypes, when considering all the amino acid polymorphisms, induce higher FXIII-specific activities whereas others result in lower specific activities. These have been presented in detail very recently by 2) and will therefore not be discussed at length here. The molecular basis of FXIII deficiency has now been investigated in almost 30 unrelated families world-wide. The majority of FXIII-deficient patients lack plasma and platelet FXIII. The first mutations in the FXIIIA gene were reported by 14) and 51). Since then molecular genetic analysis has demonstrated that FXIII deficiency is a highly heterogenous disorder. The gene defects causing FXIII deficiency were last summarized by 75). The number of mutations responsible for FXIII deficiency has more than doubled since then. To date, 36 sequence changes have been described in the FXIIIA gene and three in the FXIIIB gene ( Table II). These are discussed in relation to their effects on the FXIIIA and FXIIIB mRNAs and polypeptides. Nucleotide substitutions resulting in missense or nonsense mutations and splice defects do not appear to be clustered in specific regions, but are scattered throughout the FXIIIA gene and its mRNA (Fig 1). There appear to be more deletions and/or insertions around the exon 2 and 3 regions. To date, no sequence changes have been located in exons 13 and 15. Furthermore, there have been no mutations reported at the residues involved in formation of the catalytic triad (Cys314, His373, Asp396). Fig 1. Sequence changes identified in factor XIII deficiency. All the mutations identified to date are located on the FXIIIA cDNA. The size of each of exons I–XV is given in base pairs. UTR = untranslated region. A total of 15 missense mutations have been presented ( Table II). Ten of these lie within the core domain, two in each of the barrel domains 1 and 2 and one in the beta-sandwich domain ( Figs 2 and 3 ). All mutations have been identified in single unrelated FXIII-deficient families with the exceptions of: (a) the Val414Phe mutation which has been described in two patients, both of Indian origin suggesting a founder effect for this sequence change; and (b) the Leu660Pro mutation identified in 10 patients from three unrelated families of Palestinian Arab origin, again suggesting a founder effect for this sequence change in the Arab population. Fig 2. Sequence changes identified in factor XIII deficiency. Locations of the missense and nonsense mutations on the linear FXIIIA polypeptide. AP = activation peptide. Fig 3. Sequence changes identified in factor XIII deficiency. Locations of the missense mutations on the FXIIIA three-dimensional structure. Each mutation is presented by a filled red circle. The N- and the C-termini are indicated. The Asn60Lys mutation is a change from a polar uncharged amino acid to a charged basic residue with a longer side chain ( 6). This mutant has been expressed in the yeast Saccharomyces cerevisiae ( Table III; 23 ). However, no recombinant protein could be detected although there was mRNA present in the transformed yeast cells ( Table III). In addition, the patient carrying this mutation also had FXIIIA mRNA in PBMCs but no immunodetectable plasma FXIII ( 6 ). This suggests that the Asn60Lys mutation results in an unstable FXIIIA molecule. Seven of the 10 missense mutations within the core domain have been expressed in recombinant yeast cells or mammalian COS cells ( Table III). All of these mutants show steady-state FXIIIA mRNA levels comparable to the wild type, suggesting they are transcribed normally and that their mRNAs are probably stable. The recombinant proteins have extremely low activities and are highly unstable. The only exception to this was the Gly501Arg mutation which had normal transglutaminase activity ( Table III). However, its extremely unstable nature made it impossible to purify this recombinant protein for further analysis ( 23 ). The effects of most of these mutations on the FXIIIA structure have been studied through molecular modelling: (i) Met242 is a buried residue whose side-chain is surrounded by a predominantly hydrophobic environment. It also packs against the side-chain of Arg252 ( 76). Replacement of Met242 with a smaller threonine residue, as seen in one patient, would create a void in the structure and probably yield a destabilized FXIIIA molecule ( 77). (ii) The Arg252 residue is involved in stabilizing the conformation of a loop, between two core domain helices that are near the beta-sandwich, through hydrogen bond interactions with Met247 and Asp243 ( 77). Its substitution with a small hydrophobic residue Ileu would result in a space in this region of the molecule and elimination of the hydrogen bond interactions. This is likely to interfere with protein folding and result in a destabilized structure. (iii) Arg260 is located on a core domain helix at the dimer interface ( 115). It forms salt links with Asp404 and Asp427 of the same subunit and Asp404 from the adjacent subunit ( 115). These interactions may maintain the integrity of the active site. In the mutants Cys260 and His260 these salt links would be lost, resulting in a destabilized molecule ( 45; 53). (iv) Arg326 is completely buried in the core domain forming hydrogen bonds with Ala332 and Asn207. The Arg326Gln mutation is also expected to prevent correct folding of the molecule through creating a destabilizing hole in the molecule and eliminating the hydrogen bonds observed with Arg326 ( 77). (v) Molecular modelling of the Ala394Val mutation predicts that this change has no effect on its structure, conformation being similar to the wild type ( 50). The patient who presented with this sequence change also carried a 20 bp deletion at the exon 1/intron 1 boundary ( Table II) on the same allele ( 50). These investigators were unable to detect FXIIIA mRNA transcribed from this allele in this patient, suggesting that either this allele is not transcribed or the RNA is unstable. Since there is no expression data for this mutation it is difficult to predict a priori the effect it may have on the activity and stability of the molecule. It is therefore not fully established that the Ala394Val mutation is actually a deficiency causing sequence change. (vi) Val414Phe is located in the core domain in a hydrophobic pocket near the surface of the molecule. It would not be possible to incorporate a phenylalanine residue at position 414 without creating short contacts which would need to be relieved by some conformational change of the surrounding protein backbone. It is likely that the bulky and inflexible aromatic side-chain of this Phe residue dramatically interferes with correct protein folding and destabilizes the molecule ( 11; 53). (vii) Residue Leu498 is found near the carboxy terminus of a helix that is on the surface of the core domain and is exposed to solvent. Substitution with a Pro at this position would introduce a kink in the helix. This would again lead to destabilization of the molecule and probable interference with dimer formation ( 77; 116). The two mutations reported in the barrel domain 1 are Asn541Lys and Gly562Arg: (i) Asn541 is located towards the surface of the barrel 1 domain and appears to be important in heavily stabilizing a turn in the molecule through hydrogen bond interactions with Ser543, Asn545 and Leu577. The much longer side-chain of Lys cannot be accommodated within the space available and the hydrogen bond interactions will be eliminated ( 3). The resulting molecule is again likely to be unstable. (ii) In the Gly562Arg mutant, molecular modelling shows that the elongated side-chain of Arg562 cannot be accommodated within the space available. Furthermore, expression data indicate that the Arg562 mutant has a much increased total potential energy compared to the Gly562 containing molecule, suggesting that Arg562 is less stable than its wild- type form ( 108). Both of the disease-associated mutations identified in the barrel domain 2 are Leu to Pro changes at residues 660 and 667 respectively. (i) The side-chain of Leu660 is buried in the hydrophobic core of the barrel 2 domain, making two hydrogen bonds with main chain atoms of Ileu683 ( 48). The Leu660Pro mutation would eliminate hydrogen bond formation probably resulting in a misfolded structure that is unstable or more susceptible to degradation. (ii) In the FXIIIA crystal structure, Leu667 is located in the middle of the sheet that flanks the catalytic core domain ( 115). The side-chain of this residue is directed toward the centre of the barrel domain and is surrounded by a number of other hydrophobic residues. Proline in this position will disrupt at least three hydrogen bonds integral to the beta sheet structure creating a hole in the hydrophobic centre of the domain ( 9). This would significantly affect protein folding and the anticipated stability of the mutant molecule. It is now apparent that all the missense mutations identified to date can be predicted to have a detrimental effect on protein stability, probably through misfolding. This conclusion arises from their structures as predicted via computer molecular modelling as well as from recombinant protein expression data (see Table III). Four mutations have been found which result in truncations of the FXIIIA polypeptide ( Figs 1 and 2; Table II). The Arg661* mutation was found in six Finnish families and accounts for around two-thirds of the mutant alleles in this population. This is likely to be due to the genetic isolation of the Finnish population and a consequent founder effect. The remaining three termination mutations have been identified in single unrelated families. With regard to mRNA levels, the Arg661* mutation results in a reduction in the steady-state transcript level of FXIIIA mRNA to < 10% of normal ( 76). FXIIIA mRNA analysis by Northern blotting has not been performed in the remaining families, although FXIIIA mRNA was detectable by RT-PCR in the patient carrying the Tyr441* mutation ( 6 ). Seven nucleotide substitutions are known at FXIIIA splice sites. Interestingly, six of these have occurred in patients of English origin ( 6, 3, 4). The G→A substitution at position −1 in intron 5 has been identified in three unrelated families ( 110; 3, 4). All of these splice site mutations give rise to abnormally spliced mRNA transcripts which result in either frameshifts and early translation termination or in-frame deletions ( Table IV). The T→C mutation at position +6 in intron 3 allows a low level of correct mRNA splicing, giving rise to a small quantity of active protein. In one family, splice site mutations in introns 7 and 8 both occur on one allele, whereas the intron 7 mutation is present only on the second allele ( 4). The mRNA transcript produced by the skipping of the last 41 bases of exon 7 probably originated from the intron 7 mutation. However, the origins of the other two mRNA transcripts observed in this family cannot be determined with certainty. It is probable that these two transcripts (lacking 41 bases of exon 7 and all of exon 8, or lacking all of exons 7 and 8) result from the allele containing both the intron 7 and intron 8 mutations ( Table IV; 4 ). A two-base AG deletion at the intron 2 and exon 3 splice junction also gives rise to an abnormally spliced transcript where the first two nucleotides of exon 3 are skipped ( Table IV; 51). A 20-base deletion at the exon 1 and intron 1 splice junction has also been described ( 50 ). However, the consequence of this splice site defect on mRNA splicing is not known. An insertion of a T nucleotide at position +2 in intron 4 causes the skipping of exon 4 ( Table IV; 50 ). Thus, splice defects reported to date consist of nucleotide substitutions, deletions and an insertion. All these defects appear to affect the processing of FXIIIA pre-mRNA. There are two single-base insertions described: a C nucleotide inserted at position 1286 in codon 400 resulting in a frameshift such that translation terminates at codon 403 ( 10); a T base insertion at the exon 4/intron 4 splice junction ( 50). This affects the FXIIIA RNA processing and is detailed above in the splice site defects section. Four deletions have been reported within the coding region of FXIIIA, not including those at splice junctions, ranging from one to 13 nucleotides ( 77; 8; 56; 54). Three of these result in frameshifts and one produces an in-frame deletion of amino acid 344. This variant, lacking the Asn344 residue, has been expressed in Saccharomyces cerevisiae. The recombinant mutant protein lacked transglutaminase activity and was also unstable ( 54). It has been postulated that residues His342 and Asp343 guide the lysyl substrate residue into the active site ( 115). Because His342 and Asp343 residues are adjacent to Asn344 they are likely to be reorientated in the FXIIIA molecule lacking Asn344. It is possible that it is this reorientation which contributes to loss of transglutaminase activity. There has been one small insertion/deletion identified in a family of English origin where a GG dinucleotide in exon 3, at positions 375 and 376 in the FXIIIA cDNA, was replaced by the sequence TCGTCC ( Table II; 6). Interestingly, the sequence TCGTCC also occurs upstream, at positions 354–359, in the FXIIIA cDNA ( 38 ). It is possible that the conversion GG→TCGTCC may have arisen due to a replication error in this family, particularly as it is transmitted through the paternal line. This sequence change results in a frameshift and early translation termination. One gross deletion, probably of > 100 kb, was reported in a patient originating from West Yorkshire in England ( 5). This deletion has been characterized at the mRNA level where exons 4–11 are absent in the FXIIIA transcripts isolated from the FXIII- deficient patient and his father who was heterozygous for this deletion. Overall, it is worth noting that regions around exons 2, 3 and 11 are involved in deletion events in six of the eight patients whose FXIII deficiency is due to deletions. It would be interesting to analyse the intronic DNA sequence of these areas to locate any particular DNA elements, or other characteristic features, that may promote an increased level of deletion events or DNA replication errors, in these regions. The molecular genetics of FXIIIB subunit deficiency has been investigated in one Japanese and two unrelated Italian families ( Table II). The deficient patient from the Japanese family was a compound heterozygote for a deletion of nucleotide A at position 4161 at the intron 1/exon 2 splice junction and a Cys430Phe mutation ( 40). The effects of the splice site mutation have not been determined but are presumed to lead to abnormal splicing of the FXIIIB pre-mRNA. The Cys430Phe defect destroys a disulphide bond in the seventh Sushi domain, altering the conformation of the mutant protein sufficiently to impair its intracellular transport ( 39). The three FXIII-deficient patients from the two Italian families were all homozyotes for an AAC insertion within the codon for Tyr80, TAC→TAAACC, in exon 3 ( 49; 102 ). This insertion results in a stop codon, Tyr80* in the second Sushi domain. The range of plasma FXIII activity within the normal population is very wide, ranging between 53.2% and 221.3% (mean 105% ± 28.56% standard deviation) of the standard normal plasma value ( 2). It has been recognized for many years that low plasma FXIII levels, < 5% of normal, are sufficient to control bleeding ( 111). FXIII-deficient patients completely lack plasma and platelet FXIII and yet some suffer from only a mild bleeding tendency ( 69). Recently, two families with the mild phenotype have been studied in detail and the results may provide possible explanations for this paradox. In the first family, of Finnish origin, two sisters, aged 16 and 19, with FXIII subunit A deficiency (absent subunit A by immunoelectrophoresis; clots soluble in 5 m urea) suffered from very mild bleeding ( 74). Molecular analysis of their FXIIIA gene revealed that a C→T splice site mutation at position +6 in intron C, inherited through the paternal line, was contributing to the mild phenotype in this family. RT-PCR of the FXIIIA mRNA from these patients produced PCR products representing four differentially spliced FXIIIA transcripts ( Tables II and IV). Sequence analysis of these showed that three RT-PCR products were from abnormally spliced transcripts, which would result in a truncated FXIIIA protein molecule, or in-frame deletions of 63aa and 147aa respectively. The fourth RT-PCR product, which ironically was least abundant, represented the correctly spliced FXIIIA mRNA and would produce FXIIIA protein of the correct normal sequence. This small amount of FXIIIA protein was sufficient to cause some γ-γ dimerization of fibrin in these patients and thus prevent severe bleeding. This family also carried a nonsense Arg661→stop FXIII deficiency-causing mutation segregating through the maternal line. The second family was described by 53). In this case the patient was of Indian origin and was diagnosed with FXIII deficiency at the age of 20 when she presented with a severe bleeding episode but had no past history of a significant bleeding syndrome. The mutation responsible was a homozygous Val414Phe missense change ( Table I). However, this mutation has previously been described, in the homozygous state, by 11) in an unrelated patient, also of Indian origin, who suffered from and had a family history of the severe coagulation FXIII deficiency phenotype. The patient studied by 53) also carried the Leu34 variant whereas the patient reported by 11) did not, and was homozygous for Val34. It has now been established by three independent studies that alleles containing the Leu34 variant give rise to FXIIIA molecules with higher activity, and specific activity, compared to alleles containing the Val34 variant ( 52; 59; 2). It is therefore possible that the presence of a Leu34 residue produces some residual FXIII activity in the patient suffering the mild phenotype, despite the presence of the Val414Phe mutation. Indeed, it has recently been shown that there is a correlation between genotype (comprising the known ‘normal’ variations at codons 34, 204, 564, 650 and 651) and phenotype ( 2 ). Clearly then, the mechanisms linking ‘mutant’ genotype, combined with ‘normal’ variants, and clinical phenotype are far more complex than our previous understanding suggested. This is still controversial. In vitro and in vivo studies point to the hepatocyte as the most likely site of synthesis of the B subunit. The A subunit is most likely to be synthesized by precursor cells of platelets and monocytes in the bone marrow but a contribution from hepatocytes cannot be discounted (see review by 84). In recipients of liver transplants the B subunit phenotype changes to that of the donor whereas the A phenotype remains unchanged. The converse is true in recipients of bone marrow transplants ( 15). It is very likely that the major source of plasma subunit A in vivo is a population of cells of marrow origin, i.e. megakaryocytes, monocytes and monocyte-derived macrophages. A mechanism for transport of the intracellular A subunit into the plasma has not been identified. In plasma, FXIII circulates tightly bound to f