Review pubs.acs.org/bc
The Dock-and-Lock Method Combines Recombinant Engineering with Site-Specific Covalent Conjugation To Generate Multifunctional Structures Edmund A. Rossi,*,† David M. Goldenberg,‡ and Chien-Hsing Chang†,§ †
IBC Pharmaceuticals, Inc.; ‡Garden State Cancer Center, Center for Molecular Medicine and Immunology; and §Immunomedics, Inc.; each in Morris Plains, New Jersey, United States ABSTRACT: Advances in recombinant protein technology have facilitated the production of increasingly complex fusion proteins with multivalent, multifunctional designs for use in various in vitro and in vivo applications. In addition, traditional chemical conjugation remains a primary choice for linking proteins with polyethylene glycol (PEG), biotin, fluorescent markers, drugs, and others. More recently, site-specific conjugation of two or more interactive modules has emerged as a valid approach to expand the existing repertoires produced by either recombinant engineering or chemical conjugation alone, thus advancing the range of potential applications. Five such methods, each involving a specific binding event, are highlighted in this review, with a particular focus on the Dock-and-Lock (DNL) method, which exploits the natural interaction between the dimerization and docking domain (DDD) of cAMP-dependent protein kinase (PKA) and the anchoring domain (AD) of A-kinase anchoring proteins (AKAP). The various enablements of DNL to date include trivalent, tetravalent, pentavalent, and hexavalent antibodies of monospecificity or bispecificity; immnocytokines comprising multiple copies of interferon-alpha (IFNα); and site-specific PEGylation. These achievements attest to the power of the DNL platform technology to develop novel therapeutic and diagnostic agents from both proteins and nonproteins for unmet medical needs.
T
activity.17,18 Indeed, the power of recombinant technology has facilitated the creation of diverse, multivalent, multifunctional fusion proteins. Nevertheless, we have recognized that innovative fusion proteins created by recombinant engineering may be built into more complex structures to gain additional attributes that are highly desirable, yet not technically attainable, in the individual engineered construct. To date, such goals are commonly achieved, with varied success, by judicious application of conjugation chemistries. Well-known examples include pegylated cytokines to increase serum halflives,19 biotinylated proteins to enable immobilization into microarrays,19,20 and intein-mediated assembly of protein− DNA chimeras to quantify specific molecules to which the protein binds.21 A number of methods exploiting the specific interaction between a receptor protein and its cognate small molecule ligand have been developed for the selective labeling of fusion proteins with small molecules, such as fluorescent dyes, and/or immobilization of fusion proteins. For this group of technologies, fusion proteins are recombinantly tagged with small molecule-binding receptor proteins, which include the Halo-Tag,22 SNAP-Tag,23 TC-Tag,24 dihydrofolate reductase,25 and FKBP-Tag.26 The purified tagged proteins can be labeled, either in vitro or in vivo,23−26 with a fluorescent dye coupled to
he continuing development of new technologies for both rational and combinatorial protein engineering has increased the variety and magnitude of potential molecules that may be designed and produced for biotechnological and biomedical applications. An essential factor for these accomplishments is that many natural proteins have their functional properties located in discrete domains, which can be manipulated as molecular modules and exploited as building blocks for devising artificial structures with multiple functions. Molecular modules derived from the binding domains of antibodies or nonantibodies, or those based on protein display scaffolds, currently have received the most attention, as described in several reviews.1−5 Molecular modules that confer effector functions of proteins, for example, the Fc of an IgG or the catalytic domain of an enzyme, also constitute an important class of components in the architecture of designer proteins.6,7 In addition, peptide motifs with the innate ability to selfassociate are often built into fusion proteins to facilitate the formation of dimeric, trimeric, or multimeric products composed of the same or different polypeptides.8−12 Examples of self-associating peptides that have been utilized successfully for this purpose include Fos/Jun, the C-terminal multimerization domain of the tumor suppressor protein p53, a streptavidin subdomain, the trimerization domain of collagen XVIII, and triplet-forming collagen-like peptides.13−16 Fusion of TNFα, which exists naturally as a trimer, to the C-terminus of a scFv antibody resulted in a homotrimeric structure with retained binding ability of the scFv as well as TNFα © 2011 American Chemical Society
Received: September 8, 2011 Revised: November 14, 2011 Published: December 14, 2011 309
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
Figure 1. Representative examples of linking interactive modules into covalent or quasi-covalent structures. (a) Self-assembled monolayer (SAM) projecting the capture ligand (phosphonate) is tethered on a gold surface. A capture protein (cutinase) fused to the protein of interest specifically binds to the capture ligand, which leads to covalent immobilization of the protein of interest without altering its activity or orientation (adapted from Proc Natl. Acad. Sci. U.S.A. 99, 5048−5052). (b) Multimeric complexes can be assembled by linking one module containing barstar to another module containing barnase. Bispecific binding constructs composed of two or more scFvs are generated using this approach (adapted from Nat. Biotechnol. 21, 1486−1492). (c) Multifunctional nanoparticles comprising a barnase-magnetic particle (MP), barstar-quantum dots (QD), and discFv-barnase. Adapted from Proc. Natl. Acad. Sci. U.S.A. 104, 14753−14758. (d) A docking tag derived from the first 15 amino acids of human RNase I is fused to a targeting protein to enable the attachment of a payload module that contains the S-protein of human RNase I as an adapter. The use of an optimized pair (C-tag and Ad-C) for this system further allows the formation of a disulfide-stabilized adduct (adapted from Bioconjug. Chem. 17, 912−919.). (e) The α helices of synaptobrevin (brevin), SNAP25, and syntaxin form a tetrahelical bundle known as the SNARE complex. A tag comprising the two α helices of SNAP25 plus the α helix of synaptobrevin is fused to the protein represented by the green diamond. The α helix of syntaxin is fused to the protein represented by a blue oval. (f) The α helices (DDD and AD) involved in the natural binding interaction of PKA and AKAPs are utilized to dock two types of modules (shown as fusion proteins of Fab) that are further locked by disulfide linkages (adapted from Proc. Natl. Acad. Sci. U.S.A. 103, 6841−6846).
original articles as well as reviews,31,32 it will not be included in this review. Relatively new to the bioconjugate field are promising strategies for tethering two or more molecular modules of distinct functions into covalent or quasi-covalent assemblies following a binding event. We present here an overview of five such strategies (Figure 1), each employing a different mode of attraction to assemble the interactive constituents, with an option for covalent stabilization of the resulting structures via directed ligation. We will provide a proof-of-concept example to elucidate each strategy, and then we will focus on the DockN-Lock (DNL) platform technology,33 which has the potential for making an unlimited number of bioactive molecules with multivalency, multifunctionality, and defined composition. We highlight the advantages of DNL, describe potential alternatives for DNL, and outline the challenges and prospects presented by DNL.
the synthetic small molecule ligand, which binds tightly and specifically to the tag. Alternatively, the tagged protein can be immobilized on a matrix, which is coupled with the ligand.27,28 While these methods are highly effective for site-specific coupling of small molecules to recombinant proteins, they are less appealing for the construction of complex, multifunctional proteins, because the requisite chemical conjugation of the small molecule ligand to a protein may be nonsite-specific and may interfere with biological activity. The most popular binary affinity system utilizes the uniquely strong biotin−streptavidin interaction. However, attachment of either biotin or streptavidin (normally tetrameric) to a target protein traditionally requires chemical conjugation, which is typically not site-specific. Recombinant methods have been developed for site-specific fusion of both streptavidin and biotin. Recombinant fusion of streptavidin to a precursor protein results in the formation of a tetrameric module, which may be advantageous or disadvantageous, depending on the intended application. Site-specific biotinylation can be accomplished by recombinant fusion of a biotin acceptor peptide to a precursor protein followed by enzymatic biotinylation with a biotin ligase.29 Biotinylation can even be accomplished in vivo when a biotin acceptor peptide-fused protein is expressed in a biotin ligase-tansfectant host production cell.30 Because the streptavidin−biotin system is well established with countless
■
ENZYME/INHIBITOR MODULES BASED ON CUTINASE AND PHOSPHONATE The introduction of protein microarrays has underscored a need for methods that can selectively, rapidly, efficiently, and irreversibly immobilize proteins in a defined orientation to preserve their folded conformation and function.34 This has been accomplished using some of the receptor−small molecule systems, such as the SNAP-tag, discussed above.27 Another 310
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
acids) and barstar (89 amino acids) as a heterodimerization pair adds significant protein mass (∼22 kDa), which, combined with their bacterial origin, will likely render the complexes immunogenic.
immobilization strategy that has been successfully applied to achieve these goals involves active site-directed capture of surface-attached suicide substrate by an enzyme-tagged protein, with subsequent reaction to secure a covalent link (Figure 1a).35 In essence, a protein of choice is fused to Fusarium solani pisi cutinase, a 22-kDa serine esterase that forms a site-specific adduct with phosphonates, to serve as the bait agent.36,37 Selfassembled monolayers of alkanethiolates with terminal 4nitrophenyl phosphates are tethered to gold-coated glass to present the capture ligand. The binding of cutinase to phosphonate leads to the displacement of p-nitrophenol by the catalytic serine, resulting in covalent immobilization of the protein onto the monolayer surface. Antibody chips thus prepared exhibit desirable characteristics that include a single orientation of the immobilized antibody with a controlled density, as well as high affinity and selectivity for the target antigens.38 This system has been applied to specifically and covalently label membrane proteins of live cells with quantum dots.39 Cutinase was recombinantly embedded in the extracellular domain of lymphocyte function-associated antigen-1 (LFA-1), and p-nitrophenyl phosphonate (pNPP)-conjugated quantum dots were covalently coupled to cutinase-LFA-1 on the surface of live transfectant cells. The dynamic redistribution of LFA-1 during locomotion of BAF cells, which were transfected with cutinase-LFA-1, was imaged using pNPP-Alexa488 on glass slides coated with ICAM-1 (for adhesion) and SDF-1 (for chemokinesis).
■
ADAPTER/DOCKING-TAG MODULES BASED ON MUTATED FRAGMENTS OF RNASE I The development of the adapter/docking-tag system based on the high affinity interaction between the two fragments of human pancreatic ribonuclease (RNase I) enables site-specific conjugation of a payload with a targeting protein.44 In this approach, a peptide derived from the first 15 amino acids of RNase I is fused to a targeting protein to serve as a docking tag for an adapter that comprises the S-protein of RNase I (residues 21−127) derivatized with a diagnostic or therapeutic agent (Figure 1d). The optimized method uses a docking tag, named C-tag, containing a cysteine in position 4 (R4C mutant), and an adapter, named Ad-C, containing a cysteine in position 118 (V118C mutant), to facilitate the formation of a C4−C118 disulfide bond upon association of C-tag and AdC.45 A further modification of Ad-C involves the introduction of an additional cysteine in position 88 (N88C) to allow sitespecific ligation of an effector. The source of Ad-C is a chimera (BH-RNase) consisting of the N-terminal 29 amino acids of bovine ribonuclease A fused to 30−127 amino acids of RNase I, which can be selectively cleaved with the bacteria protease subtilisin at the A20-S21 bond to release the S-protein.46 This technology has been adapted for covalent coupling of targeting fusion proteins containing the C-tag to liposomes decorated with Ad-C.47 Proof-of-principle was demonstrated using vascular endothelial growth factor fused with a C-terminal C-tag and a fragment of anthrax lethal factor fused with a Nterminal C-tag. Both proteins retained functional activity when coupled to drug-loaded Ad-C-liposomes.
■
ENZYME/INHIBITOR MODULES BASED ON BARNASE AND BARSTAR The barnase−barstar module provides an interesting approach for engineering multivalent, multifunctional protein complexes that takes advantage of a tight association (KD ∼ 10−14 M) of the ribonuclease barnase with the barstar inhibitor.40 Fusion proteins tagged with one barnase (or two in series) or barstar were generated separately, purified, and subsequently combined to form quasi-covalent heterodimers or heterotrimers (Figure 1b). Since the initial report of the production, refolding, and purification of barnase and barstar fusion proteins in E. coli, expressions of scFv-barnase in mammalian HEK 293T cells,41 and scFv-barstar in tobacco plants,42 also have been described. Recently, the barnase−barstar system has been utilized to generate trifunctional superstructures comprising magnetic iron oxide particles, fluorescent nanoparticles, and multiple scFv antibody fragments (Figure 1c).43 As core modules, magnetic particles of different sizes and chemical compositions were conjugated with multiple barnase groups. Two types of fluorescent quantum dots were chemically coupled with multiple barstar groups and used to decorate the magnetic particle−barnase core. A third layer of a dimeric anti-Her2/neu scFv-barnase module was added to the superstructure. The nanoparticles bound to and fluorescently labeled Her2/neu+ ovarian carcinoma cells (SCOV-3) and concentrated them in areas with a strong magnetic field gradient, demonstrating the designed trifunctionality. Although this system can be useful for the modular assembly of binary and more complexes, two significant drawbacks may hamper its practical application. First, the toxicity of barnase to the host cell requires the coexpression of barstar with barnase fusion proteins, thus entailing denaturation and chromatography under denaturing conditions for barstar removal and subsequent refolding. Second, utilization of barnase (110 amino
■
SNARE MODULES A modular method for making binary complexes via sitespecific combination of recombinant fusion protein modules, which is based on a tetrahelical bundle known as the SNARE (soluble N-ethylmaleimide-sensitive factor attachment protein receptor) complex, has been reported recently.48 The neuronal SNARE system is based on the natural interaction of three proteins, syntaxin, synaptobrevin, and SNAP25, which contribute two, two, and one α-helices, respectively, to form a very tight tetrahelical bundle commonly known as the SNARE complex.49 Although the interaction is not covalent, the SNARE complex is remarkably stable and resistant to harsh treatments, including urea and sodium dodecyl sulfate.50 In one embodiment of this method, recombinant modules are made as fusion proteins with a precursor protein fused with either SNAP25 (206 amino acids) or synaptobrevin (60 amino acids) tags. The modules are then stapled together using a synthetic syntaxin peptide (50 amino acids). Proof-of-principle was established using a modular assembly of botulinum neurotoxin type A from its subunits.51 Fusing the receptorbinding domain with the synaptobrevin SNARE motif allowed delivery of the active part of botulinum neurotoxin, which was tagged with SNAP25, into neurons, where it was able to cleave its intraneuronal molecular target and to inhibit release of neurotransmitters. To increase the utility of the system, incorporation of biotin into the synthetic syntaxin peptide 311
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
allowed the receptor binding portion of the toxin to be used to deliver streptavidin-coated quantum dots into neurons. As an alternative, the three-part system was adapted to a binary system where two of the three components were combined into a single tag (Figure 1e).48 In one variation, SNAP25 and synaptobrevin were fused with glutathione Stransferase (GST), which showed highly stable binding to syntaxin-coupled beads or to a syntaxin-coupled Biacore sensor chip. In another variation, a combination of synaptobrevin and syntaxin, is used as a single tag, referred to as Nano-Lock (NL), which binds to a SNAP25-tagged module. Proof-of-principle was demonstrated using gold nanoparticles, where GSTSNAP25 was immobilized on gold nanoparticles derivatized with NL. Similarly, GST-NL specifically coupled to nanoparticles was derivatized with SNAP25.
■
DNL TECHNOLOGY
The basis of the DNL method is the exploitation of the specific protein−protein interactions occurring in nature between the regulatory (R) subunits of cAMP-dependent protein kinase PKA and the anchor domain (AD) of A-kinase anchor proteins (AKAPs).52,53 Two types of R subunits (RI and RII) are found in PKA, and each has α and β isoforms. The R subunits have been isolated only as stable dimers, with the dimerization domain shown to consist of the first 44 amino-terminal residues.54 The AD of AKAPs for PKA is an amphipathic helix of 14−18 residues,55 which binds only to dimeric R subunits. For human RIIα, the AD binds to a hydrophobic surface formed by the 23 amino-terminal residues.56 Thus, the dimerization domain and AKAP binding domain of human RIIα are both located within the same N-terminal 44 amino acid sequence54,57 and are referred to as the dimerization and docking domain (DDD). For the DNL method, AD and DDD peptide sequences, which are modified with cysteine residues for covalent “locking” via disulfide bridges, are fused to a precursor protein (or other entity) to make DNL-modules, which are produced and stored independently (Figure 2a and c). The DDD-derivatized proteins (DDD-modules) spontaneously form stable homodimers; therefore, DDD-modules always comprise two copies of the precursor protein (Figure 2b). Upon design, any DDDmodule can be paired with any AD-module to generate a wide variety of stable conjugates comprising two and one copies of the DDD-module and AD-module precursors, respectively (Figure 2d). In our first study,33 we showed the DDD of human RIIα (designated DDD1) and the AD derived from AKAP-IS,58 a synthetic peptide optimized for RII-selective binding with a reported KD of 4 × 10−10 M (designated AD1), to be an excellent pair of linker peptides for docking two entities into a noncovalent complex, which could be further locked into a stably tethered structure through the introduction of cysteine residues into both the DDD and AD at strategic positions to facilitate the formation of disulfide bonds. The cysteinemodified versions of AD1 and DDD1, designated AD2 (CGQIEYLAKQIVDNAIQQAGC) and DDD2 (CGHIQIPPGLTELLQGYTVEVLRQQPPDLVEFAVEYFTRLREARA), respectively, are the preferred peptide linker pairing of the DNL method. Besides the unique feature that a DDD-derivatized precursor is always presented in two copies, additional advantages of the DNL method that together distinguish it from other site-
Figure 2. Basic features of the DNL method. (a) DDD-module of precursor A. (b) DDD-mediated dimer of precursor A. (c) AD-module of precursor B. (d) Stably tethered DNL conjugate comprising two copies of precursor A and one of precursor B. Green rings indicate SH groups of the engineered cysteine residues of DDD (blue helix) and AD (yellow helix). The “locking” disulfide bonds are depicted as interlocking green rings.
specific conjugation methods19−21,35,40,45 are summarized as follows. The completely modular nature of DNL allows the rapid and facile creation of a large number of potentially useful conjugates. Each DDD- or AD-containing entity serves as a module, and any DDD-module can be paired with any ADmodule. Such modules can be produced independently, stored separately, and combined on demand. There is essentially no limit on the types of precursors that can be converted into a DDD- or AD-module, so long as the resulting modules do not interfere with the dimerization of DDD or the binding of DDD to AD. Many DNL conjugates can be made combinatorially from a relatively small number of each type of module. For example, 400 different conjugates could be made from a pool of 20 AD- and 20 DDD-modules. DNL is versatile. Modules can be made recombinantly or synthetically. Recombinant modules, which may be produced in mammalian, microbial, or other expression systems, may include fusions of antibodies or antibody fragments, cytokines, enzymes, carrier proteins (e.g., human serum albumin and human transferrin), or a variety of natural or artificial nonantibody binding or scaffold proteins.1,2,4,5 Furthermore, DDD or AD can be coupled to the amino-terminal or carboxyl terminal end or even positioned internally within the fusion protein, preferably with a spacer containing an appropriate length and composition of amino acid residues, provided that the binding activity of the DDD or AD and the desired activity of the polypeptide fusion partners are not compromised. Two 312
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
Figure 3. Fab-based DNL-modules and TFs. (a) CH1-DDD1-Fab, a Fab-DDD-module without “locking” cysteine residues in the DDD, which is fused to the carboxyl-terminal end of the Fd chain. (b) N-DDD2-Fab, a Fab-DDD-module with added cysteine residues in the DDD, which is fused to the amino-terminal end of the Fd chain. (c) CH1-DDD2-Fab, a Fab-DDD-module with cysteine residues in the DDD, which is fused to the carboxyl-terminal end of the Fd chain. (d) CH1-AD1-Fab, a Fab-AD-module without “locking” cysteine residues in the AD, which is fused to the carboxyl-terminal end of the Fd chain. (e) CH1-AD2-Fab, a Fab-AD-module with cysteine residues in the AD, which is fused to the carboxyl-terminal end of the Fd chain. (f) Binary, noncovalent complex of CH1-DDD1-Fab-hMN-14 and CH1-AD1-Fab-h679. (g) TF1, a covalent complex of NDDD2-Fab-hMN-14 and CH1-AD2-Fab-h679. (h) TF2, a covalent complex of CH1-DDD2-Fab-hMN-14 and CH1-AD2-Fab-h679. Variable (V, blue or green) and constant (C, gray) domains of the heavy (H) and light (L) chains are represented as ovals. The DDD and AD peptides are shown as blue and yellow helices, respectively, with the locations indicated for the reactive sulfhydryl groups (SH) of the engineered cysteine residues in AD2 and DDD2, and the disulfide bridges (red line) that they form in TF1 and TF2. Adapted from Proc. Natl. Acad. Sci. U.S.A. 103, 6841−6846.
stable in serum and whole blood for more than 7 days at 37 °C.60−63 A potential disadvantage of DNL comes from a commercial standpoint. Although the modular nature of DNL is clearly an advantage for constructing many fusion proteins from a much smaller number of modules, and also allows the assembly of complex structures that would not be possible with conventional recombinant engineering, two (or possibly more) modules need to be synthesized in order to make a single product. Obviously, this results in increased production costs due to the establishment of two host cell lines and bioreactor production of each module. However, this concern is mitigated by current advances in manufacturing capabilities, which allow bioreactor production of some recombinant proteins at the scale of several grams per liter. Another potential technical complication, which we have yet to suffer from, is the potential for irreversible inactivation of precursor molecules during the mild redox conditions during DNL conjugation. The cysteine residues involved in DNL docking are exposed and readily reduced under mild conditions. Internal disulfides, which might be present in a precursor, are less affected under these conditions, and even if reduced, they are likely to re-form upon reoxidation. However, it is possible that some precursors might denature or aggregate during the redox reaction.
or more AD peptides can be incorporated into AD modules to create a scaffold for attachment of multiple DDD modules. Modules can also be made synthetically, to comprise peptides, polyethylene glycol (PEG), dendrimers, nucleic acids, chelators with or without radioactive or nonradioactive metals, drugs, dyes, oligosaccharides, natural or synthetic polymeric substances, nanoparticles, fluorescent molecules, or quantum dots, depending on the intended applications.59 DNL manufacturing is easy. The DNL method is basically a one-pot preparation and requires three simple steps to recover the product from the starting materials: (i) combine DDD- and AD-modules in stoichiometric amounts; (ii) add redox agents to facilitate the self-assembly of the DNL-conjugate; and (iii) purify by an appropriate affinity chromatography process. The DNL-modules can be purified and stored prior to their use in a DNL conjugation. However, purification of the modules is not necessary. DNL conjugation can be accomplished in mixtures of cell lysates and/or culture supernatant fluids containing the DNL-modules, with subsequent isolation of the DNL-conjugate by affinity chromatography. A single-step affinity purification process with commonly used affinity media, such as Protein A or immobilized metal, typically results in >95% purity of the DNL conjugate, which is sufficient for most preclinical applications. However, for manufacturing of clinical material, further processing steps, such as additional affinity chromatography, ion exchange chromatography, low pH treatment, and ultrafiltration, are required to ensure adequate removal of viral and other contaminants. The site-specific conjugation associated with DNL results in quantitative yields of a homogeneous product with a defined composition. The high-affinity binding between the DDD- and AD-modules results in nearly 100% conversion of each into the desired DNL product. The DNL method results in a preparation of defined and homogeneous molecular size and composition, for which the full activity of each module is usually preserved. Further, DNL conjugates are highly stable in vitro and in vivo. Various DNL conjugates were completely
■
DNL MODULES AND CONJUGATES
Fab-Based Modules and Tri-Fabs. In our first study,33 the DNL method was applied to generate bispecific trivalent complexes comprising three Fab fragments by combining DDD-modules constructed from the Fab of hMN-14, a humanized monoclonal antibody (mAb) with specificity for the A3B3 domains of human carcinoembryonic antigen (CEACAM5),64 with AD-modules constructed from the Fab of h679, a humanized mAb with specificity for the hapten histamine-succinyl-glycine (HSG).65 313
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
Three Fab-DDD-modules and two Fab-AD2-modules were used to generate three different Tri-Fabs (TFs), each comprising two hMN-14 Fabs and one h679 Fab. Fab-based modules are purified by affinity chromatography using Protein L or KappaSelect affinity media. CH1-DDD1-Fab-hMN-14 (Figure 3a) and CH1-AD1-Fab-h679 (Figure 3d) are FabDDD- and Fab-AD-modules of the respective parent mAbs, fused at the carboxyl-terminal end of the Fd chain (C-terminal end of CH1 domain) to DDD1 and AD1, respectively. Upon mixing of the two modules, the formation of a binary complex (Figure 3f) was readily demonstrated by SE-HPLC with the KD determined by equilibrium gel filtration analysis66 to be about 8 nM, which is presumably too weak of an affinity for in vivo applications. DDD1 was converted to DDD2 by incorporation of a cysteine residue at the amino-terminal end of the DDD peptide. Thus, the naturally dimeric DDD2-modules have two reactive cysteine residues. Two additional modules were generated for the hMN-14 Fab, N-DDD2-Fab-hMN-14 (Figure 3b) and CH1-DDD2-Fab-hMN-14 (Figure 3c), where the DDD2 peptide was fused to the amino- or carboxyl-terminal end of the Fd chain, respectively. A cysteine residue was added to each end of AD1 to create AD2, which was included in the CH1-AD2-Fab-h679 module (Figure 3e). Two stably tethered trivalent bispecific structures, referred to as TF1 (Figure 3g) for the conjugate of N-DDD2-Fab-hMN-14 and CH1-AD2-Fabh679 and TF2 (Figure 3h) for the conjugate of CH1-DDD2Fab-hMN-14 and CH1-AD2-Fab-h679, were produced in nearly quantitative yields and characterized extensively. TF1 and TF2 were isolated by affinity chromatography using the h679binding hapten HSG, and they were each resolved by SE-HPLC as a single peak of the expected molecular size (∼150 kDa). BIAcore demonstrated bispecific binding, which was confirmed by competition ELISA to be equivalent to hMN-14 IgG and h679 Fab, reflecting the retention of valency and binding affinity. Furthermore, TF1 and TF2 were stable for at least 7 days at 37 °C in human or mouse serum. The superiority of TF2 as a pretargeting agent for diagnostic imaging has been demonstrated in numerous animal studies.67−70 Recent clinical results have confirmed that TF2 is effective for pretargeted radioimmune detection in patients with CEA-CAM5+ tumors (Figure 4a). Since the generation of TF1 and TF2, the modular DNL method has allowed the rapid development of over 20 different trivalent bispecific Fab-based complexes, referred to as the TF series, by combinatorial pairing of monomeric Fab-AD2 and dimeric Fab-DDD2 modules.71−81 Several of the TFs have exhibited excellent performance in pretargeting applications (Figure 4b). The technology platform has also been expanded to generate a variety of conjugates, including multivalent, multifunctional antibodies and immunocytokines, which are highlighted below. IgG-AD2-Modules. After establishing proof of principle with the TF series using Fab-based DDD2- and AD2-modules, we next developed AD2-modules for IgG, which allowed the synthesis of a variety of complex, IgG-based DNL conjugates. CH3-AD2-IgG modules were produced recombinantly by appending, in frame, the coding sequence for AD2, preceded by a flexible peptide linker, to the 3′ end of the coding sequence of the CH3 domain of IgG (Figure 5a). This method allowed the simple conversion of any existing IgG-expression plasmid vector into one for IgG-AD2 expression. Because IgG naturally forms a heterotetramer comprising two heavy and two light chains, IgG-AD2-modules possess two AD2 peptides, having
Figure 4. (a) Transverse cross sections of an 18F-FDG-PET (top) and corresponding pretargeted immunoSPET image (bottom) of a patient with liver metastases of a colon carcinoma. The immunoSPECT image was taken 4 h after injection of an 111In-labeled hapten-peptide (100 ug,185 MBq), that was given 24 h after TF2 (75 mg). Arrows in the 18 F-FDG image show the regions of tumor uptake that are also seen in the immunoSPECT image. (b) Pretargeted imaging of pancreatic xenografts with TF10. Nude mice bearing CaPan-1 human pancreatic cancer xenograft in their upper right flank (top; arrows). Tumors were ∼0.5 cm in the longest diameter. The bottom panels show planar whole-body images of mice given 111In-labeled hapten-peptide alone (bottom right) or pretargeted 16 h prior with TF10, a bispecific TriFab directed against a pancreatic mucin and to the hapten HSG (bottom left). T, tumor; K, kidneys.
one on the carboxyl-terminal end of each heavy chain (Figure 5b). The IgG-AD2-modules are purified in a manner similar to that used for IgG, using Protein A affinity chromatography. Each AD2 binds a dimeric X-DDD2 module, resulting in defined DNL structures comprising an IgG fused at its carboxyl terminus to four X groups, where X denotes an entity that could be a protein, peptide, polymer, drug, or other molecule (Figure 5c). Hexavalent Antibodies. IgG-AD2-modules were first used to construct hexavalent antibodies (HexAbs), which were produced by combination of IgG-AD2 with the same FabDDD2 modules used in the TF series (Figure 3c), highlighting the advantage of the modular nature of DNL (Figure 5d). Initially, monospecific HexAbs were produced using IgG-AD2and Fab-DDD2-modules derived from the same parental mAb. As an example, a HexAb was made for the humanized antiCD20 mAb, veltuzumab (aka hA20),82 where CH3-AD2-IgGhA20 was combined with CH1-DDD2-Fab-hA20.61 The DNL conjugation resulted in a homogeneous preparation of a conjugate comprising an Fc and six functional anti-CD20 Fabs, which each retain the binding affinity of veltuzumab. The construct, originally named Hex-hA20, has been redesignated 20-(20)-(20), using a standardized naming system where the first code indicates the IgG-AD2-module and the codes in parentheses indicate dimeric Fab-DDD2-modules. Compared to veltuzumab, 20-(20)-(20) exhibited 3-fold higher binding avidity by ELISA, and a 3-fold slower off-rate from live NHL cells by flow cytometry, which demonstrated that all six binding arms can bind CD20 on cells (Figure 6). In vitro, 20-(20)-(20) inhibited proliferation of NHL cells at subnanomolar 314
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
Figure 5. IgG-AD2 modules. (a) Recombinant fusion of an AD2 peptide (yellow helix) to the carboxyl-terminal end of the heavy chain CH3 domain of an IgG. (b) IgG-AD2 module equipped with two AD2 peptides. (c) DNL conjugate of an IgG-AD2 with two dimeric X-DDD2 modules. (d) bsHexAb comprising an IgG and two stabilized Fab dimers. Variable (V, green) and constant (C, gray) domains of the heavy (H) and light (L) chains of the IgG-AD2 and Fab-DDD2 are represented as ovals. Yellow helix, AD2; Blue helices, DDD2; SH, sulfhydryl groups of the engineered cysteine residues in AD2; red line, “locking” disulfide bridges.
receptor (FcRn) is altered. Indeed, we have recently observed diminished FcRn binding of 20-(20)-(20) and other similarly designed HexAbs by surface plasmon resonance (BIAcore). Alternative formats of HexAbs, designed to maintain all Fc function including CDC and FcRn binding, are under development. Although 20-(20)-(20) has reduced circulating half-life and CDC, it still demonstrated antitumor efficacy, which was comparable to the case of veltuzumab at equivalent doses, in tumor-bearing mice. We have produced numerous bispecific hexavalent antibodies (bsHexAbs) by combining an IgG-AD2 module with a FabDDD2 module derived from a different parental mAb (Figure 5d). The first bsHexAbs produced were derived from the humanized anti-CD22 mAb, epratuzumab (aka hLL2),83 and veltuzumab.62 Combination of CH3-AD2-IgG-hA20 with CH1DDD2-Fab-hLL2 produced 20-(22)-(22), which comprises veltuzumab with four Fabs of epratuzumab. A bsHexAb of the opposite configuration, 22-(20)-(20), which has four Fabs of veltuzumab fused to epratuzumab, was generated from CH3AD2-IgG-hLL2 and CH1-DDD2-Fab-hA20. Characterization of the bsHexAbs demonstrated that the DNL conjugation resulted in highly purified covalent structures of the expected size and composition. Both 22-(20)-(20) and 20-(22)-(22) retain the binding properties of their parental Fab/IgGs, with all six Fabs apparently capable of binding simultaneously (Figure 6). The bsHexAbs exhibited biological activities that were not observed using a mixture of the parental mAbs. Treatment of cells with the bsHexAbs, but not veltuzumab plus epratuzumab, resulted in translocation of both CD22 and CD20 into lipid rafts, induction of apoptosis and growth inhibition without secondantibody cross-linking, and homotypic adhesion. The bsHexAbs induced significant increases in the levels of phosphorylated p38 and PTEN, and also notable differences in signaling events from those incurred by cross-linking veltuzumab or rituximab with a secondary antibody.84 Thus, the greatly enhanced direct toxicity of these bsHexAbs correlates with their ability to alter the basal expression of various intracellular proteins involved in regulating cell growth, survival, and apoptosis, with the net outcome leading to cell death. Indeed, the bsHexAbs killed lymphoma cells in vitro much more potently than the parental mAb mixture (Figure 7a).
Figure 6. Binding properties of hexavalent antibodies. (a) Competition ELISA showing relative binding avidity of veltuzumab, 22-(20)(20) (bivalent anti-CD22, tetravalent anti-CD20), 20-(22)-(22) (bivalent anti-CD20, tetravalent anti-CD22), 22-(22)-(22) (hexavalent anti-CD22), and 20-(20)-(20) (hexavalent anti-CD20) for binding to WR2, the anti-Id antibody to veltuzumab (left) and to WN, and the anti-Id antibody to epratuzumab (right). (b) Analysis of dissociation rates from live Raji. Cells were saturated with PE-mAbs, and the fluorescence intensity was measured over time by flow cytometry. % maximal binding (MFI T = 0) was calculated by dividing MFI T = x into MFI T = 0 and plotted vs time. Adapted from Blood 113, 6161− 6171.
concentrations without the need for an additional cross-linking antibody. For 20-(20)-(20), some of the Fc-associated effector functions, including antibody-dependent cellular cytotoxicity (ADCC), are retained, while others, such as complementdependent cytotoxicity (CDC), may be compromised. Even though 20-(20)-(20) is a nearly 2.5-fold larger molecule than veltuzumab, the circulating serum half-life is shorter for the former, suggesting that its interaction with the neonatal Fc 315
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
of events were observed at the molecular and organelle levels, which together culminated in cell death, following bsAbinduced juxtaposition of CD74 and CD20, but not treatment with the combined parental mAbs. Potentially important signaling events noted were a sustained phosphorylation of ERK and JNK MAP kinases, inhibition of Akt, and a reduction of p65 NF-κB nuclear translocation. Additionally, the antiapoptotic protein Bcl-xL was downregulated. The bsHexAbs induced a strong homotypic adhesion associated with reorganization of the actin cytoskeleton, an increase in lysosomal volumes coupled with membrane permeability and leakage of cathepsins and other lysosomal contents into the cytosol, a decrease in mitochondrial membrane potential, and increased levels of reactive oxygen species. The bsHexAbs induced a nonclassical apoptosis, which was caspase-independent and not associated with autophagy (Gupta et al., Blood, in press). More Antibody-Based-Modules and Multivalent Antibodies. We have produced many bispecific TF and HexAb DNL conjugates that are of 1 × 2 and 2 × 4 Fab formats, respectively. In addition, we have produced trispecific HexAbs (2 × 2 × 2) by combining two different Fab-DDD2-modules with an IgG-AD2-module of a third specificity (unpublished results). A wide variety of alternative formats of multispecific antibodies is attainable with the introduction of new types of DNL modules. For example, we have produced a Fab-based module with tandem AD2 peptides fused to the carboxylterminal end of the Fd (Figure 8a). Combination with a FabDDD2 module resulted in a 1 × 4 bispecific conjugate comprising five Fabs, without an Fc (Figure 8b), demonstrating that multiple AD2s can be incorporated into a single module (unpublished data). Application of this concept to an IgG module (IgG-AD2-AD2) would allow the creation of a multispecific IgG having a total of 10 Fabs, via combination with existing Fab-DDD2 modules. AD2- and DDD2-modules could be generated for other types of antibody-based proteins/ fragments, including scFv, diabody, minibody, dual variable domain IgG, etc., providing for a myriad of possible multispecific antibody formats. As a final example, we have generated a bispecific dual variable domain IgG-AD2 module (Figure 8c), which was combined with a Fab-DDD2 module to generate a 2 × 2 × 4 trispecific octavalent antibody (Figure 8d, unpublished data). Nonantibody-Based DNL Modules. DNL is not restricted to antibody-based constructs, as recombinant DDD2 or AD2 modules can be constructed for almost any protein. We have applied the DNL method to various cytokines, including interferon-alpha (IFNα), erythropoietin, and granulocyte colony-stimulating factor (G-CSF), to generate various immunocytokines (Figure 9).59,63,85 The cytokinemodules were produced in either E. coli or mammalian cell culture, with AD2 or DDD2 fused at their amino- or carboxylterminal ends. We have utilized a DDD2-module of human protamine, which, when paired with an IgG-AD2 module, can be used for targeted cellular delivery of nucleic acids, such as siRNA. As an example of a module of a precursor that is both an enzyme and a toxin, a DDD2-module of ranpirnase (Rap, a cytotoxic ribonuclease from Rana pipiens) has been combined with various IgG-AD2 modules for targeted delivery of the toxin to tumor cells. In addition to recombinant protein modules, nonprotein DNL-modules can be assembled using synthetic AD2 peptides coupled to a variety of synthetic precursor molecules.59
Figure 7. In-vitro antiproliferation with hexavalent antibodies. (a) Indicated NHL cells were seeded in T-flasks at 100k cells/mL and treated with 22-20-20, 20-22-22, or a combination of veltuzumab and epratuzumab at indicated concentrations. Viable cell densities (VCD) were determined daily by flow cytometry using Guava Viacount. On day 3, cultures were split 1:2 to maintain logarithmic growth over the course of the assay. The data are plotted as the VCD corresponding to the undiluted culture. Adapted from Blood 113, 6161−6171. (b) Invitro cytotoxicity of mantle cell lymphoma cell lines with hexavalent antibodies. Indicated MCL cells were treated for four days with increasing concentrations of the indicated hexavalent antibodies or a combination of veltuzumab (hA20) and milatuzumab (hLL1). The relative viable cell densities were measured with MTS. The % of the signal obtained from untreated cells was plotted vs the log of the nanomolar concentration. Error bars, SD.
As observed previously for the monospecific HexAbs, both bsHexAbs exhibited ADCC, but not CDC (in vitro), and had shorter circulating half-lives than the parent mAbs. Intriguingly, 22-(20)-(20) and 20-(22)-(22) killed human lymphoma cells in preference to normal B cells in whole human blood (ex vivo), whereas the parental veltuzumab depleted malignant and normal B cells about equally. In-vivo studies with NHL xenografts revealed that 20-(22)-(22), despite having a shorter serum half-life, had antitumor efficacy comparable to that of veltuzumab. The 22-(20)-(20) was less potent than 20-(22)(22) but more effective than epratuzumab and control bsAbs. We recently reported the potent activity of anti-CD20/CD74 bsHexAbs against various B-cell lymphomas including mantle cell lymphoma (MCL), which is an aggressive subtype, generally having a poor prognosis with no established standard of care (Gupta et al. Blood, in press). The same Fab-DDD and IgG-AD modules of veltuzumab were combined with IgG-AD2 and Fab-DDD2 modules, respectively, of milatuzumab (an antagonistic anti-CD74 humanized mAb) to generate the bsHexAbs 74-(20)-(20) and 20-(74)-(74). Similar to the anti-CD20/CD22 bsHexAbs, 74-(20)-(20) and 20-(74)-(74) inhibited growth of NHL cell lines without the need for additional cross-linking. They inhibited Burkitt lymphoma lines with a similar potency to the CD20/CD22 bispecifics. Of note, 74-(20)-(20) and 20-(74)-(74) effectively inhibited MCL lines, including JeKo-1, Granta-519, and Mino, which were relatively resistant to 20-(22)-(22) and 22-(20)(20) and unnaffected by combinations of the parental antibodies, veltuzumab and milatuzumab (Figure 7b). A variety 316
dx.doi.org/10.1021/bc2004999 | Bioconjugate Chem. 2012, 23, 309−323
Bioconjugate Chemistry
Review
more potently than a combination of veltuzumab and IFNα. Comparative pharmacokinetic (Pk) analysis in mice demonstrated a longer circulating serum half-life for 20-2b-2b (23.4 h) compared to the commercial PEGylated IFNα drugs peginterferonalfa-2a (14.9 h) or peginterferonalfa-2b (9.3 h). Due to high specific activity, extended Pk, and tumor targeting, 20-2b-2b demonstrated superior therapeutic efficacy compared to PEGASYS, veltuzumab, or nontargeting IgG-IFNα in human lymphoma xenograft models, even though mouse immune cells respond poorly to human IFNα2b (Figure 10a). Targeting IFNα with an anti-CD20 mAb makes the immunocytokine more potent than either agent alone or in combination. On the basis of encouraging preclinical results, 20-2b-2b, the prototype IgG-IFNα, is now under development for CD20targeted immunotherapy of NHL. CD20 is a preferred target for this disease because it is expressed at high levels on the cell surface of many B-cell lymphomas, and its expression on normal cells is essentially limited to B cells. The potential benefits of therapy with 20-2b-2b are likely limited to nonHodgkin lymphoma (NHL) and possibly chronic lymphocytic leukemia (CLL) patients. Using the combination of the IFNα2b-DDD2 module with CH3-AD2-IgG-hL243, we have recently developed an IgG-IFNα named C2-2b-2b, which has tetrameric IFNα2b fused to an anti-HLA-DR mAb (hL243).87 HLA-DR is an attractive target because it is expressed on the cell surface of many hematopoietic malignancies.88 The broad range and high-level expression of HLA-DR makes C2-2b-2b attractive for use in the therapy of diverse malignancies. In vitro, C2-2b-2b inhibited 20 cell lines, including B-cell lymphoma (Burkitt, mantle cell, and follicular), leukemia (hairy cell, AML, ALL, and CLL), and myeloma. In most cases, this immunocytokine was more effective than CD20-targeted mAb-IFNα or a mixture comprising the parental mAb and IFNα. The responsiveness of each hematopoietic tumor cell line correlated with HLA-DR expression/density and sensitivity to IFNα and hL243. C2-2b-2b induced more potent and longer-lasting IFNα signaling compared to nontargeted IFNα. In vivo, C2-2b-2b demonstrated superior efficacy compared to nontargeting mAb-IFNα, peginterferonalfa-2a, or a combination of hL243 and IFNα using human lymphoma and myeloma xenografts (Figure 10b and c).87 The modular nature of DNL enabled the production of the first bispecific immunocytokine, 20-C2-2b, which comprises two copies of IFN-α2b and a stabilized F(ab)2 of hL243 (humanized anti-HLA-DR) site-specifically linked to veltuzumab (Figure 9d).85 This was achieved by combining CH3-AD2IgG-hA20 (Figure 5b) with two DDD-modules, IFNα2bDDD2 (Figure 9a) and CH1-DDD2-Fab-hL243 (Figure 3c). Due to the random association of either DDD-module with the two AD2 groups, two side-products, 20-(C2)-(C2) and 20-2b2b, are formed in addition to 20-C2-2b, which was purified from the mixture by a multistep affinity chromatography process. Following reduction of 20-C2-2b, its five component polypeptides were identified by LC-MS (Table 2). We determined that ∼15% of the polypeptides comprising the IFNα2b-DDD2 module are O-glycosylated, with the observed mass indicating an O-linked glycan having the structure NeuGc-NeuGc-Gal-GalNAc, which was also predicted (
