Aledort et al.  recently reported the results of a meta-analysis of prospective clinical trials of the two classes of recombinant factor VIII (rFVIII) products, B-domain-deleted (BDD) and full-length (FL). The authors claimed that exposure to BDD-rFVIII was associated with an increased risk of de novo inhibitors. A detailed critique of the meta-analysis conducted in this report was provided by Iorio et al.  and the conclusions of Aledort and coworkers remain highly controversial.
We and others have previously developed computational approaches to predict potential T-cell epitopes in therapeutic proteins. We have assessed this approach using FVIII in the treatment of hemophilia A as a model. These studies show an association between a mismatch in the primary sequences of the infused FVIII protein drug and the endogenous FVIII (albeit non-functional) synthesized by the patient and the development of inhibitory antibodies. A similar strategy can be used to evaluate engineered protein drugs that result in novel sequences. In this computational study, we investigate whether the novel junction that results from protein engineering to generate the BDD-rFVIII product could result in T-cell epitopes. We would like to emphasize that this approach permits identifying potential T-cell epitopes, which are a necessary condition to elicit an immune response that may not be sufficient to induce inhibitory antibodies (see below). Moreover, it is also possible that the immune response generated may not result in inhibitory antibodies.
We used a computational approach to evaluate whether potential T-cell epitopes may exist within the set of novel overlapping peptides that span the synthetic junction incorporated in the BDD-rFVIII products. Several lines of evidence show that CD4+ T cells play a central role in the development of inhibitory antibodies to FVIII [3,4]. Our current understanding, which admittedly is evolving, suggests that an immunogenic CD4+ T-cell response to an exogenous protein requires that: (i) at least one of the peptides derived by proteolytic processing of the infused protein must be foreign (‘non-self’) to the patient; (ii) at least one of the distinct isomers of class-II human-leukocyte antigens (HLA-II) comprising the patient’s individual MHC-class-II (MHC-II) repertoire must be able to bind a foreign peptide with sufficient affinity and stability so that it can be presented by the antigen-presenting cells (APCs); (iii) at least one of the patient’s subpopulations of CD4+ T cells has a T-cell antigen receptor (TCR) capable of high affinity binding to an HLA-II/foreign-FVIII-peptide complex; and (iv) the above requirements occur in the presence of danger signals as the co-stimulatory molecules whose expression they induce help to ensure that the interactions between APCs and specific T cells are of sufficient avidity . The sequence of an infused FL-rFVIII drug-product is always mismatched with respect to the endogenously synthesized FVIII due to the patient’s HA-causing FVIII gene (F8) defect; it can also be mismatched at up to three non-synonymous F8 polymorphisms . The engineered F8 cDNA used to express BDD-rFVIII creates additional potential T-cell epitopes; viz. a novel protein segment not present in endogenous FVIII proteins. Any peptides liberated from this segment would thus have the potential to elicit an immune response.
We first analyzed the cDNA in the vector that expresses the BDD-rFVIII products [7–9]. This construct involves the deletion of 894 internal codons and splicing between the third and first nucleotides, respectively, of codons 762 and 1657. Codons 762 and 1657 encode serine (S) and glutamine (Q), respectively, at positions 743 and 1638 in the circulating form of wild-type FVIII. Removal of 894 B-domain residues places these amino acids adjacent (S743-Q1638) in the BDD-rFVIII, which thus contains 1438 residues compared with the 2332 amino acids in the FL-rFVIII products. Thus, the BDD-rFVIII protein contains a synthetic junctional peptide sequence (SFS- QNPPVLKRHQR), formed by the covalent attachment of the three N-terminal-most residues of the B-domain, S741F742S743, to the 11 C-terminal-most residues, Q1638N1639P1640P1641V1642L1643K1644R1645H1646Q1647R1648 (Fig. 1A). The 14 residue engineered BDD-junction, which is referred to as the ‘SQ-peptide’, is thus numbered as follows: S741F742S743- Q744N745P746P747V748L749K750R751H752Q753R754.
We analyzed 18 overlapping 15-mer peptides in a 32 amino acid segment encompassing the engineered junction (15 residues upstream and downstream of the dipeptide S743-Q744). We determined that 11 of these peptides (subset-2, Fig. 1B) cannot be encoded by the wild-type F8. These 11 peptides were screened using the Consensus Coding Sequence (CCDS) Project, a database of all known human polypeptide sequences based on the wild-type human reference genome (NCBI build 37.3) . This analysis demonstrated that foreign FVIII peptides generated from the BDD-junction are not encoded by genes elsewhere in the genome and thus confirmed that these 11 peptides are not ‘self’ to HA patients.
While our understanding of the details of proteolytic processing of exogenous proteins within APCs has improved markedly over the last decade , we are still unable to predict accurately which peptides will be released from any given protein without performing complex and labor-intensive assays. It is known that >80% of the BDD-rFVIII protein exists in a heterodimeric form comprised of a 90 kDa heavy-chain linked non-covalently to an 80 kDa light-chain (Fig. 1A, inset) . As shown, the SQ-peptide is located at the C-terminus of the heavy-chain. Therefore, a single proteolytic event N-terminal to this linker (e.g. within the a2-loop) can liberate a set of overlapping foreign peptides. Such peptides, which are located at the N- or C-termini of proteins and thus require only one proteolytic event for release, are likely to be present at higher concentrations in the endosomal/lysosomal compartments of APCs, where loading of peptides onto MHC-II molecules occurs.
Although necessary, the generation of foreign peptides is not always sufficient to induce an immune response. We therefore assessed the binding of the 11 foreign 15-mer BDD-junction-spanning peptides to a subset of 30 DR-isomers of HLA-II molecules, each of which contains a distinct DRB1 polypeptide chain that is common in one or more human populations. We adopted this approach because (i) several lines of evidence indicate that FVIII immunogenicity is mediated by CD4+ T cells (see above), (ii) the binding of a foreign peptide by an HLA molecule is a necessary first step in T-cell-mediated immune responses, and (iii) prior studies have shown that MHC-II restriction is mediated by DR molecules . Recent advances have improved our knowledge of the parameters that mediate interactions between peptides and HLA-II molecules [13,14] and have led to the development of publically available computational tools , which yield reasonably accurate estimations of binding affinities for peptides and specific MHC-II molecules (http://tools.immuneepitope.org/analyze/html/mhc_II_binding.html). Using the consensus approach developed by Wang et al. , we estimated the binding affinities of the 30 common DR isomers described above for the set of 18 overlapping 15-mer peptides spanning the synthetic BDD-linker (Fig. 1B). As described previously , a peptide’s affinity for a given MHC-II molecule is represented by a consensus score calculated as the median of the percentile ranks obtained from (up to) three top-performing prediction algorithms. The percentile score for a peptide-MHC-II complex is based on the affinities of 5 million peptides generated from proteins in the Swiss Prot database to that MHC-II allele. In this scale, a low percentile rank indicates high binding-affinity. According to our analysis, the foreign 15-mer BDD-junction-spanning peptides (subset-2) are predicted to bind with high-affinity to only certain DR molecules and, consequently, patients having these alleles could be at increased risk of generating an immune response if treated with BDD-rFVIII. For example, all but four of the subset-2 peptides bind with very high to high affinity to the DR14 isomer containing the DRB1*1401 gene product. Furthermore, three to seven of these peptides are capable of high to intermediate affinity binding to the DR10, DR11, DR13 and DR14 molecules containing the polypeptide chains encoded by the DRB1*1001, DRB1*1101, DRB1*1303 and DRB1*1402 alleles, respectively.
It is noteworthy that recent research characterizing the naturally occurring peptides eluted from HLA-II molecules reveals that the most abundant peptides are frequently in the range of 17- to 21-mers . If peptides in this size-range are generated from BDD-rFVIII, more patients could be at risk of an immune response because 12 of the 30 HLA-II proteins examined appear to bind the foreign SQ-junction peptides of this length with high to very high affinity. The collection of 30 distinct DR molecules defined by the DRB1 alleles depicted in Fig. 1(B) together occur in >90% of the European and North American populations and 87% of the African population. Because the foreign BDD-junction-derived peptides may bind to a subset of HLA-II molecules that occur in almost 25% of each population, a T-cell-mediated immune response could theoretically be initiated by the bioengineered portion of BDD-rFVIII in almost one in four patients. This does not mean that infused BDD-rFVIII will elicit a deleterious immune response in all of these patients as, among other reasons, only a subset will (i) have CD4+ T cells with TCRs capable of high affinity binding to the HLA-II/foreign peptide complex and (ii) develop anti-FVIII antibodies that are inhibitory.
The present letter provides a computational tool to assess the potential immunogenicity of non-naturally occurring sequences introduced into engineered proteins using BDD-rFVIII as an example. Importantly, these data suggest that some patients may be at increased risk of having an immune response elicited by the foreign peptides spanning the synthetic SQ-junction. It is important to mention that the notion that BDD-rFVIII products may be more immunogenic (introduced recently by Aledort et al. ) is controversial  and we provide a preliminary computational assessment of the immunogenic potential of this non-naturally occurring portion of the engineered protein. However, it is increasingly becoming clear that there is a pharmacogenetic basis for the immunogenicity to FVIII [6,17]. For example, when an identical rFVIII product is infused into different patients, some develop an immune response whilst others do not. In the last decade we have come to understand that the nature of individual F8 defects and non-synonymous F8 polymorphisms may be pharmacogenetic risk factors for immunogenicity in conjunction with individual HLA-II repertoires . In the context of these findings, the report by Aledort et al.  and the results reported here suggest that the immunogenic consequences of bioengineering rFVIII products should be assessed carefully on an individual patient-by-patient basis. Even if such products do not pose a significantly higher risk than FL-FVIII products at the population level, they may present a risk to specific individuals. Finally, the authors would like to emphasize that the sole purpose of this letter is to advance the debate on the pharmacogenetic basis of immunogenicity of FVIII and that no recommendations for clinical practice or product efficacy and safety should be construed. Clearly, more clinical and experimental studies are needed to resolve this debate.