The creation of a 'fully humanized' or 'reshaped' variable region involves designing a variable region amino acid sequence that contains the rodent-derived CDRs and human-derived framework sequences. The rodent-derived CDRs from the monoclonal antibody of choice provides the specificity, so these residues are automatically included in the design of the reshaped variable region. On the other hand, framework residues in theory can be derived from any human variable region and the resultant designed variable region would be considered reshaped. But for a given antibody variable region, are all human framework sequences equally suitable for creating a reshaped variable region and retaining antibody affinity? If not, what is the best approach to get around this problem and create a reshaped antibody with acceptable affinity?
There have been a number of different strategies for framework selection that have been used over the past several years (summarised in Table 1). As described earlier, the first antibody reshaping experiments were performed on anti-NP [9] and anti-lysozyme antibodies [10]. In each case, the expression of rodent-derived CDRs along with the framework sequences of the human VH region NEW was successful. However as described above when the same human framework sequences were used in the reshaping a therapeutic CD52 antibody CAMPATH1G [11], the affinity of the antibody for antigen was unacceptable until the serine residue at position 27 had been replaced with phenylalanine, the corresponding residue in the rodent framework. This replacement was aided by computer modelling. Corrections in the NEW framework were also required in the creation of a reshaped antibody directed against the human respiratory syncytial virus [19]. In this case, three residues at the end of framework 3 (positions 91-94 were replaced with the corresponding rodent residues to restore antigen-binding ability. These data indicated that creating a reshaped antibody might require a more sophisticated approach than just splicing rodent CDRs into the most convenient human framework sequences, and suggests the importance of computer molecular modelling or other strategies in helping to resolve problems that may arise.
Subsequently, so-called "best-fit" strategies for framework selection and design were developed. The best-fit strategy holds that the human framework sequence used for antibody reshaping should be derived from the human variable region that is most homologous or similar to the rodent-derived variable region. This strategy is based upon the knowledge that the framework sequences serve to hold the CDRs in their correct spatial orientation for interaction with antigen, and that framework residues can sometimes even participate in antigen binding [3]. Logic dictates then, that if the selected human framework sequences are most similar to the sequences of the rodent frameworks, this will maximise the likelihood that affinity will be retained in the reshaped antibody.
The best-fit strategy was put to the test in the creation of a reshaped version of the CD4 antibody, CAMPATH-9 [16]. Two versions of this reshaped antibody were created, one deriving its frameworks sequences from the human antibody heavy chain variable region of greatest homology (KOL [26], 72% homology) and a second version deriving its framework sequences from a human antibody heavy chain variable region of lower homology (NEW, 47% homology). Immunofluorescence staining of a CD4+ cell line was used to compare the avidities of these two reshaped antibody versions to a chimeric form of the CD4 antibody containing the unaltered rat variable region sequences (Figure 2). Both the chimeric and KOL-based reshaped antibodies stained CD4+ cells well. The concentrations of chimeric and KOL-based reshaped antibodies needed to achieve 50% antigen saturation were determined to be 2.21 and 7.16 mg per ml, respectively. Thus the avidity of the KOL-based reshaped antibody is only slightly reduced when compared to the chimeric form. By contrast, the NEW-based reshaped antibody stained CD4+ cells only poorly even at the higher concentrations. The control CAMPATH-1H antibody did not stain cells at any concentration. Also, the chimeric and KOL-based reshaped antibodies were both effective in cell-mediated lysis, whereas the NEW-based reshaped and control CAMPATH-1H antibodies were ineffective. Thus in this case at least, the selection of a human variable region framework that is highly homologous to the rodent V region seems is the best strategy for framework selection. The best-fit strategy has also been used successfully by us in the reshaping of a CD3 antibody YTH12.5 [21] and similar strategies have been adopted for other antibodies as summarised in Table 1.
At its simplest level, the "best fit" strategy involves comparing the donor rodent V-region with all known human V-region amino acid sequences, and then selecting the most homologous to provide the acceptor framework regions for the humanization exercises. In reality there are several other factors which should be considered, and which may influence the final selection of acceptor framework regions.
Often, several human V region sequences may be identified by a search which have almost the same degree of homology to the donor V region. The availability of structural data for the V regions in question may then swing the balance in favour of one over the rest; this obviously makes sense as it improves the chances of identifying the causes of any reduction in affinity or loss of specificity resulting from the humanization procedure. The selection of a well characterized acceptor V region also lessens the probability of falling foul of errors in the sequence data base. For example, several entries exist for the sequence of the human type III VH gene VH26 [27], some of which refer to the first version deposited (EMBL code HSIGHVC, NBRF code HVHU26 Swisspot code HV3C-Human] and others which are the corrected version [28] (EMBL code HS1GV3BC). There are discrepancies in the amino acid sequence of the NEW VH region as reported by Saul et al [29] and deposited in the Brookhaven data base. Careful study of the chosen acceptor V region sequences is required to avoid such mistakes, and in fact, comparing the chosen acceptor V regions with the database will help in identifying them, by indicating if the genes in question have been independently sequenced by different laboratories.
The homology scores generated by computerized searches of databases are generally produced to cover an entire V region, and are therefore affected both by homologies in the frameworks and in the CDRs. As the process of humanization involves the joining of Fw sequences from the acceptor V region to the CDR sequences of the donor, the important factor to consider is the homology between the framework regions of the two V genes. If two possible acceptor V regions are identified having similar-homology scores to the donor, a choice can be made based on the number of framework region differences between them and the donor. The acceptor with the most homologous frameworks being the one of choice. This is illustrated by a comparison of the CAMPATH-1G VH region with the human germline genes VDH26 and VH26 [27],[28],[30]. They have very similar homology scores (68.3% and 69.4% identity respectively). However, when the number of framework region differences between these two and CAMPATH-1G VH are considered, VH26 is clearly the best choice (17 differences in framework 1 to 3, compared with 23 for VDH26; in addition, VDH26 is a pseudogene and should be ruled out on that basis alone).
As more human V region sequences become available, better choices of acceptor V regions which match the donor more closely, can be made. Queen et al 1989 [15] reported that the most homologous human VH region they found when searching the NBRF-PIR database for an acceptor for the mouse-anti Tac VH-region, was that of the myeloma protein Eu (identity score of 60% over the VH gene region, excluding CDR3 and framework 4). More recently, we performed a similar search of several databases, and found a match of 73.3% identity with the germline VHI gene DP-7 (see later). [31].
A variety of additional antibodies have been reshaped using a strategy similar to "best-fit", in which acceptor frameworks are chosen on the basis of homology to the rodent monoclonal antibody, and then altered based on molecular modelling predictions prior to any experimental work in an attempt to maximise the affinity of the resultant reshaped antibody. Essentially, the goal of the modelling is to predict which key residues (if any) of the most homologous human framework should be left as in the rodent to obtain the best affinity in the reshaped antibody. The reshaping of an antibody against the human epidermal growth factor 2 receptor [24] demonstrates the usefulness of this approach. Here, human framework sequences were derived using consensus sequences derived from the most abundant human subclasses (VK subgroup I and VH subgroup III). A reshaped antibody was created that derived its framework sequences from these consensus sequences, but the resultant antibody bound antigen 80-fold less tightly than the rodent antibody. Computer modelling was then used to try and predict which residues in the human framework should be replaced with the corresponding rodent residues to optimise affinity. These residues were chosen based on their possible influence on CDR conformation and/or binding to an antigen. The most potent modelled variant contained 5 framework residues from the mouse, and bound antigen 3-fold more tightly than the original rodent rodent antibody. Thus in this case, modelling provided a potentially useful addition to the best-fit strategy. Related versions of this approach have been reported for antibodies directed against CD33 [22], CD3 [23], placental tissue [32], the T cell receptor [17], and human epidermal growth factor receptor [20].
A theoretical approach to reshaping has been proposed by Padlan [33] whereby only the "surface" of the rodent antibody is reshaped. This is accomplished by replacement of the exposed residues in the framework regions with human sequences, thus retaining all the interior and contacting residues. The perceived advantage of this approach is that the resultant antibody would have reduced immunogenicity but retain affinity for antigen, but no experimental data are provided.
One VH framework residue, at position 71, has been proposed to be of strategic importance for maintainance of antigen binding [34]. In a series of solved immunoglobulin structures, it was observed that the conformation of CDR 2 was dependent upon the length of the sequence of this loop and its interaction with residue 71 (Kabat numbering system). Indeed, in the reshaping of an anti-human epidermal growth factor receptor antibody [20], the retention of residue 71 as found in the rodent proved vital to obtaining acceptable affinity in the reshaped antibody. However, it should be noted that CAMPATH-1H was successfully reshaped to the NEW-based framework despite having a substantially different CDR 2 length and sequence and also a different framework residue at position 71 [11]. Clearly, more experimental work is required to elucidate the importance of residue 71.
Thus a key feature of many reshaping strategies has been to leave many of the framework residues as found in the rodent rather than the human counterpart. The drawback of this approach is that the resultant antibodies are somewhat less "human", and may ultimately elicit an undesirable anti-globulin response when administered to patients. But these strategies consider this added risk worth the benefit of maximising antibody affinity. The ultimate value of all reshaping strategies will depend on whether the particular antibody evokes an antiglobulin response during therapy in a variety of patients using the appropriate treatment protocols.
The purpose of the above strategies is to optimise the intrachain interactions within the reshaped variable regions in order to design the best reshaped variable region and retain antigen affinity. However, these strategies ignore possible interchain interactions that might be important for affinity retention (see for example Figure 1). For example, the CDRs of the heavy and light chains participate in forming the antigen binding site of the antibody. There may be interactions between the heavy and light chain variable regions that indirectly help the CDRs maintain their correct spatial orientation. It stands to reason then that a given human framework sequence for a heavy chain may not interact equally well with all human light chain framework sequences. This is potentially a complicating factor for all attempts to 'reshape' or 'humanize' therapeutic antibodies.