Antibody Humanisation for Therapeutic Applications

Therapeutic antibodies are currently in vogue. A large number of new biological drugs in clinical development from the biotechnology industry are based on recombinant antibodies and several have recently been approved by the FDA.

Currently a number of different strategies exist for making whole IgG therapeutic antibodies. As a result of their method of production there are some real and some perceived differences. Briefly the different antibodies can be classified as follows.

This diagram illustrates schematically the concept of mouse, chimaeric, humanised and human antibodies.

1. Murine Monoclonal antibodies from mice and rats: The original Kohler and Milstein technology provided rodent monoclonal antibodies to any antigen to which the animals could be immunised. These have been used therapeutically and the first monoclonal antibody to be licensed for therapy by the FDA was OrthoClone OKT3 a mouse IgG2a (FDA approved 1986). Rodent antibodies tend to provoke strong Human anti-Murine Antibody (HAMA) immune responses which restricts their usefulness for repeated application in the same patient.
2. Chimaeric Antibodies: Are antibodies in which the whole of the variable regions of a mouse or rat antibody are expressed along with human constant regions. This provides the antibody with human effector functions and also reduces immunogenicity (HAMA) caused by the murine Fc region.
3. Humanised / CDR grafted / Reshaped antibodies: This is an alternative to chimaeric antibodies in which only the complimentarity determining regions from the rodent antibody V-regions are combined with framework regions from human V-regions. The idea is that these antibodies should be more human like than chimaeric and thus perhaps less immunogenic than chimaeric antibodies.
4. Human antibodies from immune donors: Some antibodies have been rescued from immune human donors using either EBV transformation of B-cells or by PCR cloning and phage display. By definition these antibodies are completely human in origin.
5. Fully human antibodies from phage libraries: Synthetic phage libraries have been created which use randomized combinations of synthetic human antibody V-regions. By selection on antigen so called 'fully human antibodies' can be made in which it is assumed the V-regions are very human like in nature.
6. Fully human antibodies from transgenic mice: Transgenic mice have been created which have a repertoire of human immunoglobulin germline gene segments. These mice when immunised thus make human like antibodies.

The common misconceptions

Unfortunately the diagrams above and similar ones in the academic and biotechnology related literature have led to the misconception that there are big sequence differences between chimaeric, humanised and fully human antibodies. It is frequently stated that 'chimaeric' antibodies are about 25% murine / 75% human in sequence, 'humanised' about 5% murine / 95% human in sequence and 'human' are of course 0% murine / 100% human in sequence.

As a follow on from the above it is then concluded that the more murine a sequence is the more foreign it will look to the human immune system. So it is thus assumed that 'chimaeric' antibodies will be much more immunogenic in patients than will 'humanised', and, in turn, 'humanised' will be more immogenic than human or 'fully human' antibodies. Thus calling an antibody fully human or humanised gives some perception of added value to a therapeutic product. This perception is common in company advertising, corporate funding and in some academic and clinical circles.

There is however a big flaw in this whole logic which should be apparent to anyone who takes a look at the immunoglobulin sequences in the Kabat or IMGT databases. Many features of immunoglobulin sequences are conserved between species and thus there is no concept of an immunoglobulin sequence looking 100% mouse and at the same time 0% human and vice versa! It may surprise some to learn that there is considerable homology between many mouse V-region sequences and human V-regions sequences. In 1993 Ed Routledge, Scott Gorman and I wrote a review on antibody humanisation in which we compared different strategies and the effects this had on sequence homology. You can read this review elswhere on this website. However some interesting results are summarised below.

Sequence homology details of the heavy chain V-regions of three humanised therapeutic monoclonal antibodies are summarised in this table (from a review by Routledge et al 1993). They are the CD52 specific antibody Campath-1H, the CD25 specific antibody anti-Tac, and the CD3 specific antibody OKT-3. These antibodies were all 'humanised' by making use of human myeloma heavy chain sequences as the acceptor sequences. Three human myeloma sequences were used called NEW, EU and KOL. The results show the number (and also the percentage) of amino acid differences between the humanised antibodies and the human acceptor antibodies. They also show a comparison of the original murine sequences with the closest matched human germline sequences found in the databases available in 1992. The human myeloma sequences were also compared to the known human germline sequences. The data is presented for Framework Regions (FR), Complimetarity Determining Regions (CDR)and also the whole V-region. The FR and CDR regions were determined according to alignments with the Kabat database of protin sequences.

You might be surprised to see from this data that there is in fact very little overall difference in the sequence homology of the humanised antibodies and the equivalent chimaeric antibodies (had they been made and used for therapy). So you will I hope see that this makes nonsense of the statement that humanised are always much more homologous to human than are chimaeric antibodies. What humanisation has done for these three antibodies is to maximise homology for the V-region framework regions at the expense of homology for the CDR regions. But an important question is does this impact on the immunogenicity of the antibodies when used for therapy? Is it more important to have an overall homology for FR and CDR regions or just for FR regions?

An additional feature which becomes apparent in the above table is the fact that human myeloma sequences which are of course by definition 100% human do not seem to be completely homologous to the human germline sequences. There are two possible reasons, either they are derived from human germline sequences not yet included in the databases, or, they have diverged from the germline through then accumulation of somatic mutations (which of course are expected in immunoglobulins).

The above analysis of sequence data from therapeutic antibodies, which was derived from data in the public domain, raises I think very important issues. Just how homologous to human sequences are many of antibodies currently in development for human therapy? What impact does this have on analysis of results for immunogenicity? Clearly the classification of an antibody as 'chimaeric' or as 'humanised' is not helpful for these questions. Some so called 'chimaeric' antibodies may be more homologous to human germline than some 'humanised' antibodies. Also some 'fully human' antibodies may have drifted from germline sequences through accumulation of somatic mutations.

A solution to the problem?

There is of course an entirely objective way to reclassify therapeutic antibodies. To take their sequences and to run them against the databases and determine their homology. I have started such a project, you can find further details on the Therapeutic Antibody Human Homology Project on this site.

Links

• An introduction to the Therapeutic Antibody Human Homology Project
• TAHHP Antibody summaries
• Selected sequence alignments
• Complete search results for individual antibodies.

Requests for this page since the counter was last reset.

This page is from Mike Clark
"An antibody engineer who also enjoys the mountains."
mrc7@cam.ac.uk
Mike's home-page

---

© Mike Clark, PhD., Cambridge University, UK
Last updated on 15th March 2000