To date, the major factor which has governed the choice of human acceptor V-regions used in humanization exercises has been the desire to retain antigen binding specificity and affinity in the final humanized antibody. However, of equal importance is the need to reduce the immunogenic potential of the antibody (in humans) to the minimum possible, after all, it is this which is the driving force behind the humanization process. By allowing our knowledge of immunoglobulin V region gene genetics, gene expression and affinity maturation to influence the final choice of human acceptor V-region, it should be possible to improve on past performance and reduce the immunogenic potential of humanized monoclonal antibodies still further.
A similar analysis we performed suggested that the VH genes of the myelomas KOL, Eu and NEW which feature prominently in reshaping exercises performed to date, may have up to 15, 11 and 26 amino acid changes respectively due to somatic mutation. Of these, 8, 4 and 11 respectively are in the framework regions, and are therefore of particular concern during humanization. This data should be treated with a degree of caution because it is based on the assumption that the relevant germline counterparts (or closely related genes) are represented in the Tomlinson germline directory [31], and if they are, that they will be the ones most homologous to the rearranged genes. For example, NEW, is classified by Kabat as a human type II VH gene, but was found to be most homologous to the human type IV germline genes VH4.11 (= DP-71) and VH4.34 [31],[37],[38] (72.2% and 73.7% identical respectively). We based our estimate of the possible number of somatic mutations on a comparison of NEW VH to VH4.34. Had we assumed that NEW VH was derived from one of the 5 VHII germline genes listed by Tomlinson et al 1992 [31], the estimated number of amino acid changes would rise to approximately 46. The analyses also makes no allowance for VH gene allotypic variation between individuals, i.e., for the fact that the directory of germline genes wasn't derived from the same individual as the myeloma protein concerned. However, it does serve to illustrate that myeloma proteins, solely on the grounds of potential immunogenicity, may not be the best choice to supply acceptor framework regions, especially when the number of somatic mutations in the myeloma Fw regions may be approaching the number of Fw differences between a rodent donor V region and a human germline V gene (the number of Fw1 to Fw3 differences between the rat monoclonal antibody YTH12.5 VH and the human VH26 gene was only 14; [21].
The number of human germline VH genes now known which possess different nucleotide sequences currently stands at approximately 122 [31]; additional sequences are continually being added to the sequence databases. On the basis of nucleotide homology, they can be divided into 6 families (VHI -> VI) of various sizes (41 VHI, 5 VHII, 46 VHIII, 22 VHIV, 7 VHV, 1 VHVI). Of these genes, 83 have open reading frames, the rest being classified as pseudo-genes due to the presence of frameshift mutations or stop codons. Many of these genes have been isolated from different individuals, and so it is often difficult to say whether or not two similar genes represent allelic variants of the same gene, or are truly different genes. The total repertoire of germline VH genes in the population is at present unknown, as is the exact number possessed by any one individual. Berman et al 1988 [40] detected between 60 and 80 hybridising bands in Southern blots of restriction enzyme digested DNA. Walter et al 1990 [41], using pulse-field gel electrophoresis of homozygous DNA, estimated the number as 76, while Tomlinson et al 1992 [31] cloned and sequenced a total of 74 by PCR. All of these are acknowledged as likely underestimates due to DNA fragment comigration or PCR primer induced bias. However, the sequences now known probably represent the major part of the available repertoire, as Tomlinson et al [31] demonstrated that 215/292 (= approx. 74%) of rearranged VH genes from different individuals had germline counterparts in the donor from which their set of 74VH sequences were derived.
Using germline V genes as a source of acceptor Fw sequences requires that a modified method be used for searching for homologous sequences. The sequence of the donor V region has to be broken into sections corresponding to the germline V region gene elements (VH DH JH, VL, JL) from which it is constructed. The V and J region sequences are compared separately to databases of the corresponding human germline elements. Performing similar analyses using the DH segment of the donor VH region is unlikely to yield any useful information, because of the great degree of sequence variation seen at the VH DH and DH/JH junctions (CDR3) caused by integration/deletion of random nucleotides and consequent unpredictable DH reading frame usage which result from the mechanism of VHDHJH gene recombination during immunoglobulin gene rearrangement.
Separate comparisons of V and J elements can be advantageous. The possible dilemma, which may arise when searching rearranged V regions for a homologous acceptor, of finding a closely matching V gene which is by chance linked to a poorly matching J segment, is avoided. This situation occurred during the humanization of the mouse anti-Tac monoclonal antibody VH region described by Queen et al 1989 [15]. Comparison of the anti-Tac VH region to the NBRF-PIR database indicated that the VH region of the human myeloma protein Eu had the highest available degree of homology (57% identical over VDJH). However, framework 4 of the Eu VH region has several amino acids, presumably encoded by the Eu JH segment, which are atypical of human JH segments and which additionally resulted in a poor match between the Eu framework 4 and that of anti-Tac. (Figure 3). Separate comparison of the anti-Tac JH region (framework 4 and the framework 4 - proximal end of CDR3) to the amino acid sequences of the known functional human JH segments (of which there are 6; Figure 4) indicates that human JH4 is a much better-match than the Eu JH. In the event, Queen et al didn't conserve the EuVH framework 4 during the humanization of anti-Tac; it was effectively converted into human JH4 by the incorporation of 4 of the mouse anti-Tac residues into the final reshaped H chain.
We have performed homology searches to identify possible human germline VH and JH acceptors for the VH regions of several therapeutic rodent monoclonal antibodies. The antibodies included in the study were the CD52 antibody CAMPATH-1G [11], the anti-IL-2 receptor (CD25) antibody Anti-Tac [15], and the CD3 antibody OKT3 [23]. Accounts of the successful humanization of these antibodies utilising rearranged human myeloma acceptor V regions have already been published (see above), and although, until tested, there is no guarantee that using homologous germline genes would work equally as well, a paper comparison of the two strategies yields some interesting points (Figure 5a, b and c, and Table 2).
A general observation made when searching the sequence databases (these included EMBL, Genbank, NBRF-PIR) was that the overall best homologies for the above 3 rodent VH genes were found to be against human germline VH genes or genes expressed in the foetal repertoire, despite the presence of other rearranged human V region sequences in the databases. Perhaps in retrospect this isn't surprising, assuming that rodent and human VH genes initially originated from a common ancestral gene. Somatic mutation of rearranged rodent or human genes will presumably amplify the degree of evolutionary divergence seen in their germline counterparts, hence a rearranged rodent V gene is likely to have a higher degree of homology to a germline human V gene than to its rearranged, somatically mutated derivative. This does imply that, if the overall humanization strategy being followed is one of "best fit", nothing is lost by choosing germline V genes as acceptors over rearranged V genes.
As described above CAMPATH-1G, the first rodent therapeutic monoclonal antibody to be successfully humanized, utilized the VH framework regions of the human myeloma protein NEW in its reshaped form, CAMPATH-1H [11]. NEW is classified by Kabat as a type II human VH gene [3] and it was chosen because its crystallographic structure is known. Its amino acid sequence is only 39% identical to that of CAMPATH-1G (excluding DH and JH segments), there being 40 amino acid differences in framework regions 1 to 3. This is itself did not prevent the success of the reshaping project, however, it was found that residue Ser 27 in the NEW VH framework region 1 had a serious negative effect on the binding affinity of the CDR grafted antibody (see Figure 1). The occurrence of serine at position 27 is atypical of VHII genes; in fact serine isn't found at position 27 in any of the human VH germline sequences listed in the directory compiled by Tomlinson et al. 1992 [31]. All 5 of the known VHII germline genes possess phenylalanine at this position and the type IV VH genes to which NEW is most similar, have a glycine residue. Mutation of Ser27 to Phe in the humanized version of CAMPATH-1G was required to restore antigen binding activity. Comparison of the CAMPATH-1G VH region to human germline VH sequences indicated a close homology with type III VH genes, the best match (69.4% identity) being with VH26 (see Figure 5a and Table 2). This gene has only 17 differences to the CAMPATH-1G VH region in frameworks 1 to 3, compared with the 40 differences of NEW. VH26 also has a phenylalanine residue at position 27 in framework 1.
In addition to the unusual nature of the NEW VH gene segment amino acid sequence, the NEW VH region also has an atypical residue (serine) at position 107 in the NEW JH encoded framework 4. This may represent a somatic mutation, and it has been incorporated into the humanized version CAMPATH-1H. Separate "best fit" analysis for CAMPATH-1G VH and JH segments eliminates this by matching CAMPATH-1G JH to human JH4.
The VH region of the mouse anti-Tac (CD25) antibody [15] was reshaped using the myeloma protein Eu VH (type I) region to supply the frameworks, chosen on the basis of its close homology with anti-Tac (57% identical over the entire V region, 61.2% over the V gene segment). As discussed earlier, the Eu VH gene segment may contain several somatic mutations, as does the J segment. Queen et al identified 5 atypical framework residues in Eu which they replaced with the corresponding donor residues (which were the same as human consensus residues for these positions; at 89, 91, 104, 105 and 107; Kabat numbering). They also, using molecular modelling, identified a further 7 framework residues (27, 30, 48, 66, 67, 94 and 103) in the mouse anti-Tac VH region framework, which may be important in maintaining the conformation of the antigen binding site; these residues were also included in the final reshaped VH region in place of the corresponding Eu residues.
Comparison of the anti-Tac V region to human germline VH segments gave the closest match with the VHI gene DP-7 (= VH21-2 and VH3-1) [31],[40]. with an identity score of 73.5% (Figure 5b and Table 2). The anti-Tac JH segment, as discussed earlier, has the greatest homology with human JH4. The Eu VH region has a total of 29 amino acid differences in frameworks 1-4 when compared to anti-Tac VH, whereas the combination of DP-7 + JH4 has only 20 differences. In addition, 9 of the 12 "donor" framework residues which were included in the reshaped version of anti-Tac are already in place. This of course is no guarantee that anti-Tac reshaped using DP-7 and JH4 to supply the acceptor frameworks would retain binding activity either with or without the inclusion of the 3 remaining changes incorporated by Queen et al into the Eu VH framework. Looking at things from a slightly different perspective, the reshaped anti-Tac monoclonal antibody has sometimes been referred to as hyper-chimaeric [15], because of the large number of donor framework residues which have been included in the reshaped V regions. The closest matching germline gene we found to the Eu VH gene segment was DP-10 [31]. These two were 89.7% identical, there being 4 amino acid differences in framework 1-3 and a total of 11 differences overall (excluding CDR3 and Fw4). The altered Eu VH frameworks 1-3 incorporated into the humanized anti-Tac have only 5 residues different from DP-10, because whereas the inclusion of some donor residues increased the number of differences, the inclusion of others decreased the number. This, along with the effective conversion of the unusual Eu VH J region into the human JH4 sequence, means that in reality the reshaped anti-Tac VH is no more hyper-chimaeric than monoclonal antibodies reshaped to other myeloma protein VH regions which have been subjected to somatic mutation.
The VH region of the mouse anti-human CD3 monoclonal antibody, OKT3, has been humanized [23] using the myeloma protein KOL as the source of acceptor VH region frameworks. The KOL VH region has a human type III VH gene segment which isn't particularly homologous to that of OKT3 (42.9% identical excluding CDR3 and Fw4), and presumably it was chosen, like NEW for the reshaping of the CAMPATH-1G VH region, on the grounds that it is structurally well characterized. Straight forward transplantation of the OKT3 VH region CDRs into the KOL VH region wasn't sufficient to transfer antigen binding activity. It was also necessary to replace 11 of the KOL VH framework amino acids with the corresponding residues from the OKT3 VH region (in addition, several donor framework VL residues were included in the humanized OKT3 VL region). A comparison of the OKT3 VH gene segment with human germline VH genes identified the VH type I gene DP-7, the same gene picked for anti-Tac, as having the highest degree of homology (72.4% identity) (see Figure 5c and Table 2). This is not surprising, as the anti-Tac and OKT3 VH regions are themselves highly homologous, being 92.9% identical (excluding CDR3 and Fw4). Similar to the situation seen with the humanized anti-Tac VH, 7 of the 11 donor framework residues included in the reshaped OKT3 VH region are already present in the germline DP-7 gene.
By way of example, several studies have shown that the largest human VH gene family, VHIII, is highly polymorphic [42],[43],[44]. Despite this, certain individual genes are known to be remarkably conserved. The type III gene VH26, or closely related genes with 100% or 99% identical amino acid sequence, have been isolated independently from over 10 different individuals [27],[28],[31],[45],[46],[47],[48],[49],[50],[51]. This suggests that VH26 may be generally good choice as an acceptor VH gene, possibly even for donor V regions which have a greater degree of homology with other human VHIII genes that are not well conserved. To assess the potential degree of "foreignness" of the VH26 V region in individuals who may not possess the VH26 gene, we compared its polypeptide sequence to those of 16 other known homologous human germline VHIII gene sequences (Table 3). Homology scores ranged from 88.8% identity (DP-58) [31] down to 84.7% identity (DP-50) [31] with the number of different residues ranged from 11 to 15. However, as the differences in the CDRs can be ignored (because they will all be replaced by donor region CDR residues), the comparison suggests that VH regions humanized using VH26 frameworks will probably have only 4 potentially immunogenic framework residues in individuals who don't carry the VH26 gene (assuming that they do possess at least one of the other relevant VHIII genes listed).
Another VHIII germline gene which may be a good acceptor candidate is 1.9111 (= DP-49) [31],[40]. A study by Sasso et al 1990 [55], which utilized combinations of probes to detect single germline VH gene elements, indicated that the 1.9111 VH gene was present in 92% of 52 unrelated individuals. In addition, two other closely related VHIII genes, T5M10 (likely to be the previously described hv3005; [54]) and 56p1 [50], which have a maximum of 2 residues different from 1.9111, (one in CDR1 and one in Fw3) were similarly found in 35% and 62% of the 52 individuals respectively. The distribution of the three genes within the test group suggested that they weren't alleles of each other and, of greater interest from a "humanization" point of view, all of the individuals tested had at least one of the genes. This means that if VH 1.9111 frameworks were used in a humanized VH region, only 8% of the above individuals could potentially see them as foreign, and only by recognising a single residue difference in Fw3.
Although not as much information is available about the second largest VH family, VHI, it is probably as polymorphic as VHIII. The exact number of gene segments isn't known, but it has been mostly recently estimated at between 20 and 25 [31],[40],[41]. These may well be under-estimates, because of the difficulty in determining whether a hybridising band contains one or several VHI gene segments. As with the VHIII family, a considerable proportion are pseudogenes [56],[57]. How well the germline gene DP-7 (selected as a suitable acceptor for anti-Tac and OKT3 VH regions) is represented in the population isn't known, but an identical gene has been described by Berman et al 1988 [40], (gene 21-2) and at least two other very similar sequences, one in the germline configuration with only 1 amino acid difference in framework1 (gene HG3; [53], and the other a rearranged V region with single differences in each of CDRs 1 and 2 (Clone H20C3H, EMBL Code HSIGHRL, N.P. Weng and D.M. Marcus, unpublished] are also known. Several other germline VHI genes isolated from the same individual as DP-7 are also fairly similar, (identity scores of 89.8% to 84.7%) having no more than 4 differences in the framework regions. (DP-75, DP-27, DP-15 and DP-14). This suggests that even if DP-7 is poorly conserved in the population, using it as an acceptor will be no worse than using a myeloma protein VH region, and, if it or HG3 or the germline equivalent of H20C3H are well represented, it could be considerably better.
The VHII family was originally thought to contain between 10 and 25 gene segments based on Southern filter hybridization [39],[56] but more recently the number has fallen to 5 [41]. This discrepancy may in part be due to the similarity between the VHII and VHIV gene families; several VH genes initially described as type II have since been reclassified as type IV [56], [58]. Polymorphism in the VHII family have been described, including two loci characterized by hybridizing bands which are either present or absent in the individuals tested [59].
More is known about the polymorphisms of the remaining VH families, IV, V and VI. The VH IV family contains in the region of 10 to 15 gene segments [37],[38],[40],[41]. They can be placed in groups according to the length of CDR1, and these groups are probably the result of diversification following the duplication of a common ancestral gene. The variation in the nucleotide sequences of allelic counterparts in the VH IV family has been estimated as being only 1%, and the minimum degree of homology between different genes is approximately 90% [37],[38],[58]. Although these observations suggest a high degree of conservation between members of the VH IV gene family, significant RFLPs have been detected in different individuals [39], and this family may be larger and more diverse than originally thought [38].
At present, only 2 functional genes have been described for the VH V family (VH 32 and VH 251), and one of these genes (VH32) is absent from 50% of the population [60]. Allelic differences at the nucleotide level have been described for both of these genes, but at the amino acid level they account for only one difference (in framework 3) for VH 251, and no differences for VH32 [37]. Overall there are a maximum of 8 residues different between VH251 and VH32, but only 2 of these are in the framework regions (1 in Fw1, 1 in Fw3).
The VH VI gene family contains only one gene segment and it is probably the most highly conserved VH gene. It has been isolated and sequenced from over 10 individuals without a single nucleotide difference [31],[37],[40],[48]. It is expressed in the foetal repertoire [50] and it is highly represented in the population; all individuals screened to date using hybridization techniques possess this gene, 14/14 unrelated individuals [44] and 10/10 unrelated individuals [61].
Overall, data currently available on variation within the smaller VH gene families VH IV to VI suggests that in general they would make good choices as acceptor V regions from the point of view of minimizing antigenicity, especially VH251 and VHVI. Obviously their selection must be compatible with the maintenance of antigen binding in the final humanized monoclonal antibody.