Single-domain brain-targeting antibody fragments derived from llama antibodies

This application is a division of U.S. patent application Ser. No. 10/450,036, filed Oct. 24, 2003, currently pending, which is a national stage filing under 35 U.S.C. § 371 of international application PCT/CA01/00783, filed May 25, 2001, which was published under PCT Article 21(2) in English; this application also claims the benefit under 35 U.S.C. 119(e) of U.S. provisional application Ser. No. 60/207,234, filed May 26, 2000, and U.S. provisional application Ser. No. 60/263,108, filed Jan. 22, 2001, all of which are incorporated herein by reference.

The invention relates to single-domain brain-targeting antibody fragments derived from llama antibodies, in particular to antibody fragments comprising at least a part of the variable heavy domain (VH or V_HH) of llama antibodies, which fragments selectively bind to and are internalized by human cerebromicrovascular endothelial cells (HCEC), and transmigrate HCEC monolayer.

The immune system in vertebrates provides a defense mechanism against foreign intruders, such as foreign macromolecules or infecting microorganisms. The foreign invaders (antigens), both macromolecules (proteins, polysaccharides, or nucleic acids) and microbes (viruses or bacteria), are recognized through specific binding of the proteins of the host immune system to specific sites on the antigen surface, known as antigenic determinants.

As part of the immune system, B-cells of vertebrate organisms synthesize antigen-recognizing proteins known as antibodies or immunoglobulins (Ig). According to the clonal selection theory, an antigen activates those B-cells of the host organism that have on their surface immunoglobulins that can recognize and bind the antigen. The binding immunoglobulins into the bloodstream. Antibodies secreted by B-cells bind to foreign material (antigen) to serve as tags or identifiers for such material. Antibody-tagged antigens are then recognized and disposed of by macrophages and other effector cells of the immune system or are directly lysed by a set of nonspecific serum proteins collectively called complement. In this way a small amount of antigen can elicit an amplified and specific immune response that helps to clear the host organism of the source of antigen. Through a complex process of gene splicing combined with additional mutation mechanisms, human B-cells have been estimated to produce a “library” (repertoire) of more than a billion (10⁹) different antibodies that differ in the composition of their binding sites.

For most vertebrate organisms, including humans and murine species, their antibodies show a common structural pattern which consists of two identical light polypeptide chains and two identical heavy polypeptide chains linked together by disulfide bonds and numerous non-covalent interactions, resulting in a Y-shaped molecule. In humans, there are two different classes (isotypes), λ and κ, of the light chains, with no known functional distinction between them. The heavy chains have five different isotypes that divide immunoglobulins into five different functional classes (IgG, IgM, IgA, IgD, IgE), each with different effector properties in the elimination of antigen.

Of the above five classes, immunoglobulins of the IgG class are the major type in normal serum of humans and many other species and have the four-chain structure shown schematically in FIG. 1. Each chain of an IgG molecule is divided into domains of about 110 amino acid residues, with the light chains having two such domains and the heavy chains having four. Comparison of amino acid sequences between different IgGs shows that the amino-terminal domain of each chain (both light and heavy) is highly variable, whereas the remaining domains have substantially constant sequences. In other words, the light (L) chains of an IgG molecule are built up from one amino-terminal variable domain (VL) and one carboxy-terminal constant domain (CL), and the heavy (H) chains from one amino-terminal variable domain (VH) followed by three constant domains (CH1, CH2, and CH3).

The variable domains are not uniformly variable throughout their length. Three small regions of a variable domain, known as hypervariable regions (loops) or complementarity determining regions (CDR1, CDR2, and CDR3) show much more variability than the rest of the domain. These regions, which vary in size and sequence among various immunoglobulins, determine the specificity of the antigen-antibody interaction. The specificity of an antibody of the type shown in FIG. 1 is determined by the sequence and size of six hypervariable loops (regions), three in the VL domain and three in the VH domain.

By partial digestion with papain, which cleaves the heavy chains in the hinge region, the IgG molecule can be broken down into two identical Fab fragments (Fragment, antigen binding) and one Fc fragment (Fragment, crystallizes easily). Each Fab fragment comprises one complete light chain (consisting of VL and CL domains) linked by a disulfide bridge and noncovalent interactions to a fragment of the heavy chain consisting of VH and CH1 domains. The Fc fragment comprises CH2 and CH3 domains from both heavy chains, also linked by disulfide bridges and noncovalent interactions. The part of the Fab fragment consisting of variable domains of the light and the heavy chain (VL and VH) is known as Fv fragment (Fragment, variable). In an Fv fragment, the variable domains VL and VH are not covalently bound. In an scFv (single chain Fv) fragment, the VL and VH domains are covalently linked by a short peptide linker (spacer), usually 15 to 20 amino acids long, introduced at the genetic level (see FIG. 2).

scFv fragments are recombinant fusion proteins and are produced by techniques of genetic engineering, by expressing in a suitable host, usually in bacteria, a chimeric gene coding for the fragment. Various other recombinant antibody fragments have been designed to substitute for large intact immunoglobulin molecules (see FIG. 2). Other than scFv fragments, these options include Fab or Fv fragments that are stabilized or covalently linked using various strategies (see, for example, Bird et al., Science, 242, 423-426 (1988); Huston et al., Proc. Natl. Acad. Sci. USA, 85, 5879-5883 (1988); Glockshuber et al., Biochemistry, 29, 1362-1376 (1990); Jung et al., Proteins, 35-47 (1994); Reiter et al., Biochemistry, 5451-5459, 18327-18331 (1994); Young et al., FEBS Lett., 135-139 (1995)). Small antigen-binding fragments of natural antibodies are advantageous for medical applications, for example cancer targeting and imaging, when small antigen-biding molecules are required to penetrate into solid tumors.

Recent advances in gene technology have greatly facilitated the genetic manipulation, production, identification and conjugation of recombinant antibody fragments and broadened the potential utility of antibodies as diagnostic and therapeutic agents. Of particular importance to such applications is the possibility to alter the fine specificity of the antibody binding site, to create small stable antigen-binding fragments, to prepare fusion proteins combining antigen-binding domains with proteins having desired therapeutic properties, for the purpose of immunotargeting, or to “humanize” antibodies of other species, for example murine antibodies (see FIG. 2).

The genetic engineering has also made possible to screen in vitro for antibodies having a predetermined binding specificity. This may be achieved by constructing first a gene library of antibodies or antibody fragments, for example by polymerase chain reaction (PCR)-amplification of cDNA derived from B-lymphocytes using suitable primers, or by in vitro gene synthesis. The gene library may contain sequences corresponding to certain fragments of natural antibodies, or randomized antigen-binding regions, or new combinations of heavy/light chains, thus creating the potential for generating antibodies which could never be obtained from natural sources, for example, antibodies to highly toxic substances or antigens tolerated by the human immune system. By random or designed mutations, the affinity or specificity of the antigen binding can be manipulated, for example, to reach affinities never observed with natural antibodies.

To screen a gene library, which may contain many millions or even billions of different clones, for genes of antibodies having the desired binding specificity, a selection system comparable to that of the immune system is required. Such a selection system can be achieved by inserting the library genes into the genome of microorganisms capable of displaying on their surface the antibody corresponding to the inserted gene, in analogy to the expression of an immunoglobulin antigen receptor on the surface of a B-cell. Microorganisms most frequently used for providing such a display are filamentous bacteriophages, such as fd or M13 phages (phage display). The collection of phage particles having inserted genes of a library of proteins, such as antibodies, and displaying these proteins on the particles\' surface is known as a phage display library. The display of the library of antibodies on the surface of phage particles provides a physical link between the antigen-binding function of an antibody and the antibody gene. Using the affinity to a preselected antigen, the whole organism (phage) displaying this affinity can be identified and separated out of billions of non-specific clones, usually through binding to the antigen immobilized on a support, technique usually referred to as panning (see, for example, Scott et al., Science, 249, 386-390 (1990); Winter et al., Annual Rev. Immunology, 12, 433-455 (1994)). Phage clones binding to the antigen can be then amplified and used to produce the specific antibody or antibody fragment in E. coli or in other suitable organism.

For naturally occurring antibodies, there are examples that whole heavy chains alone retain a significant binding ability in the absence of light chains. It is also well established, from structural studies, that the CDR3 of the heavy variable domain generally contributes the most to antigen binding, because CDR3 amino acid residues are responsible for most of the surface contact area and molecular interaction with the antigen (Padlan, E. A., Mol. Immunology, 31, 169-217 (1984); Chothia et al., J. Mol. Biol., 196, 904-917 (1987); Chothia et al., J. Mol. Biol., 186, 651-663 (1985)). Less binding activity was observed for light chain. In view of these findings, attempts were made to isolate single VH domains. For example, VH domains were isolated from expression libraries derived from immunized mice (Ward et al., Nature, 341, 544-546 (1989)). In another report, antigen-binding VH domains were rescued from an antibody phage library that was made from a vaccinated patient (Cai et al., Proc Natl. Acad. Sci. USA, 93, 6280-6285 (1996)). Antigen-binding antibody fragments consisting of a single VH domain, known as dAbs or sdAbs (single-domain antibodies), are becoming an attractive alternative to single chain Fv (scFv) fragments. Despite smaller binding surface, their demonstrated affinity is comparable to that demonstrated by scFv fragments (Davies et al., Biotech., 13, 475-479 (1995)). Because of their smaller size, being half of the size of scFvs, sdAbs are amenable to detailed NMR structural studies (Davies et al., FEBS Letters, 339, 285-290 (1994)). Additionally, due to their simpler structure, sdAbs are more stable and have simpler folding properties.

Recently, a new class of antibodies known as heavy chain antibodies (HCA, also referred to as two-chain or two-chain heavy chain antibodies) have been reported in camelids (Hamers-Casterman et al., Nature, 363, 446-448 (1993); see also U.S. Pat. No. 5,759,808; U.S. Pat. No. 5,800,988; U.S. Pat. No. 5,840,526; and U.S. Pat. No. 5,874,541). Compared with conventional four-chain immunoglobulins of IgG-type, which are also produced by camelids, these antibodies lack the light chains and CH1 domains of conventional immunoglobulins. One of the salient features of these naturally occurring heavy chain antibodies is the predominant presence of Glu, Arg and Gly at VL interface positions 44, 45 and 47 (Kabat numbering), respectively, of their variable domain (designated V_HH). The same positions in the variable domain of the heavy chain of conventional four-chain antibodies (designated VH) are almost exclusively occupied by Gly, Leu and Trp. These differences are thought to be responsible for the high solubility and stability of camelid HCA variable domain (V_HH), as compared with the relative insolubility of VH domain of the conventional four-chain antibodies. Two more salient features of camelid V_HH domains are their comparatively longer CDR3 and high incidence of cysteine pairs in CDRs. It appears that cysteine pairs mediate the formation of a disulfide bridge and are therefore involved in modulating the surface topology of the antibody combining site. In the crystal structure of a camel sdAb-lysozyme complex, a rigid loop protruding from the sdAb and partly stabilized by a CDR disulfide linkage extends out of the combining site and penetrates deeply into the lysozyme active site (Desmyter et al., Nature Struct. Biol., 3, 803-811 (1996)).

More recently, a number of camelid sdAbs phage display libraries have been generated from the V_HH repertoire of camelids immunized with various antigens (Arbabi et al., FEBS Letters, 414, 521-526 (1997); Lauwereys et al., EMBO J., 17, 3512-3520 (1998); Decanniere et al., Structure, 7, 361-370 (1999)). By creating polyclonal libraries, many highly soluble sdAbs with high affinity and specificity have been isolated. However, it has been questioned whether sdAbs with desired affinity and defined conformations can be generated in the absence of prior immunization, i.e., with a naïve library (Lauwereys et al., supra). Immunization of domesticated valuable animals, such as camelids, raises serious ethical implications related to experiments with animals. Moreover, this approach has serious drawbacks because most of the pathogenic antigens cannot be injected into camelids, as this could endanger their lives. Considering the above drawbacks and limitations of the prior art, there exists a strong need for the generation of phage display libraries of sdAb antibody fragments derived from naïve libraries of camelid antibodies, in particular sdAb fragments of camelid heavy chain antibodies, which libraries may become a universal source of sdAbs for in vitro selection against any antigen of interest as a target. By choosing antigen targets located in tissues of therapeutic or diagnostic interest or importance, such libraries may provide new vectors for targeted delivery of therapeutic and diagnostic agents. Of particular interest to the present invention are antibody fragments targeting antigens of the endothelial tissue of the blood-brain barrier (BBB), which fragments may be used for the delivery of therapeutic and diagnostic agents into neuronal tissues.

The effective delivery of molecules into neuronal tissues remains one of the most perplexing challenges facing the pharmaceutical and biotechnology industries. The brain is isolated from the rest of the body by a specialized endothelial tissue known as the blood-brain barrier (BBB). The endothelial cells of the BBB are connected by tight junctions and efficiently prevent many therapeutic compounds from entering the brain. In addition to low rates of vesicular transport, one specific feature of the BBB is the existence of enzymatic barrier(s) and high level(s) of expression of ATP-dependent transporters, including P-glycoprotein (Gottesman et al., Ann. Rev. Biochem., 62, 385-427 (1993); Watanabe, T., Acta Oncol., 34, 235-241 (1995)), which actively degrade/extrude various pharmaceuticals from the brain (Samuels B. L., J. Clin. Pharmacol. Ther., 54, 421-429 (1993). As a result, a plethora of compounds with demonstrated efficacy in vitro cannot be used as brain-targeting pharmaceutical agents in vivo unless appropriate delivery vehicles capable of overcoming the impermeability of the BBB are employed.

Only small (<600 Daltons) and hydrophobic (Pardridge, W. M., Adv. Drug Delivery Reviews, 15, 5-36 (1995)) molecules can easily pass the BBB, a constraint that places enormous restrictions on drug development strategies. Current brain drug delivery practices either employ invasive neurosurgical procedures or non-invasive strategies such as pharmacological methods to facilitate transport of drugs via intercellular or transcellular routes. In addition to invasive and highly limited neurosurgical strategies (e.g., intraventricular drug infusion, cerebral implants) and osmotic BBB opening applied clinically, strategies based on 1) physiological- and 2) pharmacological modulation of BBB permeability are being developed.

Strategies based on physiological approaches to drug delivery through the BBB use pseudonutrients that are substrates for BBB nutrient carrier systems (Pardridge, W. M., supra). At least eight different nutrient transport systems have been identified in cerebromicrovascular endothelial cells (i.e., glucose transporter, the neutral amino-acid carrier, the basic amino acid carrier, the monocarboxylic amino acid carrier, the purine nucleoside transport carrier, the purine base carrier, choline carrier, and glutamate transporter) many of which are being exploited to carry drugs that ‘mimic’ the respective natural ligands for these transporters into the brain. While this strategy constitutes a clear advance over current alternatives, it is limited by the fact that such drugs will have to compete with endogenous substrates normally transported by these systems.