The Development of Bispecific Hexavalent Antibodies as a Novel Class of DOCK-AND-LOCK TM (DNL TM) Complexes

The DOCK-AND-LOCKTM (DNLTM) method provides a modular approach to develop multivalent, multifunctional complexes of defined structures, of which bispecific hexavalent antibodies (bsHexAbs) are prominent examples with potential applications in targeted therapy for malignant, autoimmune, and infectious diseases. Currently, bsHexAbs are constructed by derivatizing a divalent IgG, at the carboxyl termini of either the heavy chain (the CH3-format) or the light chain (the Ck-format), to contain two stabilized dimers of Fab having a different specificity from the IgG. In this review, we briefly outline the features of the DNLTM method and describe key aspects of bsHexAbs examined with diverse preclinical studies, which include binding affinity to target cells, induction of signaling pathways, effector functions, serum stability, pharmacokinetics, and antitumor activity in human tumor xenograft models. Our findings favor the selection of the CK- over the CH3-format for further exploration of bsHexAbs in clinical trials.


Introduction
With more than 30 antibody-based products now commercialized, and an additional 28 in advanced clinical trials for various indications [1], the prospect of monoclonal antibodies (mAbs) as next-generation therapeutics has been realized.Since the efficacy of mAbs often can be enhanced when combined with other antibodies targeting different antigens [2][3][4][5][6][7][8], or distinct epitopes of the same antigen [9,10], parallel efforts to explore the potential of combination therapy with dissimilar antibodies are being explored.In principle, combination therapy involving two unlike antibodies could be accomplished more cost-effectively with a cognate dual-targeting bispecific antibody (bsAb).Accordingly, there has been an emergence of new bsAbs that differ in design, structure, valency, and specificity [11,12].
We have advanced an alternative approach of constructing bsAbs using the DOCK-AND-LOCK TM (DNL TM ) method [32][33][34][35], which enables the site-specific self-assembly of two modular components only with each other, resulting, after combining under mild redox conditions, in a covalent structure of defined composition with retained bioactivity.The initial proof-of concept was provided by linking a stabilized dimer of Fab specific for one antigen to a monomeric Fab with specificity for a different antigen to generate a bispecific trivalent antibody composed of three stably-tethered Fab-arms [32].Since then, we have applied the DNL method to develop bispecific hexavalent antibodies (bsHexAbs) by derivatizing a divalent IgG, at the carboxyl termini of either the heavy chain (the C H 3-format), or the light chain (the C k -format), to contain two stabilized dimers of Fab with a different specificity from the IgG [36][37][38][39][40].In this review, we present archetype examples of bsHexAbs and discuss the notable advantages of the C K -format over the C H 3-format as revealed by preclinical results.

The DNL TM Method
The DNL TM (henceforth, DNL) method is based on the specific protein/protein interactions occurring in nature between the regulatory (R) subunits of cAMP-dependent protein kinase A (PKA) and the anchoring domain (AD) of an interactive A-kinase anchoring protein (AKAP) [41,42].There are two types of R subunits (RI and RII) found in PKA and each has α and β isoforms.The R subunits contain a dimerization domain in the first 44 amino-terminal residues and have been isolated only as stable dimers [43].The AD of AKAPs for PKA is an amphipathic helix of 14-18 residues [44], which binds only to dimeric R subunits.For human RIIα, the AD binds to a hydrophobic surface formed by the 23 amino-terminal residues [45].Thus, the dimerization domain and AKAP binding domain of human RIIα are both located within the same N-terminal 44 amino acid sequence [43,46], and are referred to as the dimerization and docking domain (DDD).
While pursuing a trivalent bispecific antibody-based agent best suitable for pretargeting applications [47][48][49], we recognized the prospect of exploring a DDD and its cognate AD as an attractive pair of linkers, and envisioned the feasibility of specifically docking a module containing the DDD of human RII, referred to as DDD1, with a module containing AKAP-IS [50], a synthetic peptide optimized for RII-selective binding with a reported K D of 4 × 10 −10 M, referred to as AD1, to form a noncovalent complex.This would be locked into a covalently-tethered structure to improve in vivo stability by introducing cysteine residues into DDD1 and AD1, resulting in DDD2 and AD2, respectively, to facilitate the formation of disulfide bonds.The amino acid sequences of DDD1, DDD2, AD1 and AD2, as well as a schematic of a basic DNL complex, are shown in Figure 1.Besides the unique feature that a module derivatized with the DDD is always presented in two copies, there are additional merits of the DNL method, as summarized below.
DNL is modular.Each DDD-or AD-containing module can be produced independently, stored separately "on shelf," and combined "on demand."In principle, any DDD module can be paired with any AD module, and 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.In addition to the DDD sequence of human RIIα, other DDD sequences may be selected from human RIα, human RIβ, or human RIIβ, and the selected DDD sequence will be matched with a highly interactive AD sequence, which can be deduced from the literature [51] or determined experimentally.

DNL is versatile.
The modular nature of the DNL method also makes it versatile, since these modules can be made recombinantly or chemically.A recombinant module may be produced in mammalian or microbial systems, and may derive from antibodies or antibody fragments, cytokines (as shown for interferon-2b [52]), enzymes, carrier proteins (such as human serum albumin and human transferrin), or a variety of natural or artificial non-antibody binding or scaffold proteins [53][54][55][56].Although each recombinant module would usually be produced in a separate expression system, we have engineered certain pairs of DDD-and AD-modules for expression in the same host cell without affecting the formation of the DNL complexes.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.
Modules may also be made synthetically, as demonstrated with linking AD2 to either polyethylene glycol or peptides [57], and depending on the intended applications, it should be feasible to develop chemistries for preparing modules that contain peptide mimetics, oligo-or poly-nucleotides, small interfering RNA, chelators with or without radioactive or non-radioactive metals, drugs, dyes, oligosaccharides, natural or synthetic polymeric substances, nanoparticles, dendrimers, fluorescent molecules, or quantum dots.
DNL manufacture is relatively trouble-free and results in quantitative yields of a homogeneous product with a defined composition and retained bioactivities.We have refined the production of the DNL complex into a one-pot reaction followed with 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 complex; and (iii) purify by an appropriate affinity chromatography.The spontaneous binding between the DDD and AD modules as well as their site-specific conjugation effects nearly 100% conversion of each into the desired DNL product and assures that the full activity of each module is preserved, the molecular size is homogeneous, the composition is defined, and in vivo integrity is largely sustained.

The C H 3-format of bsHexAbs
One established application of the DNL method is the generation of the C H 3-format of hexavalent antibodies (HexAbs), all of which comprise a pair of Fab-DDD2 dimers linked to a full IgG at the carboxyl termini of the two heavy chains, thus having six Fab-arms and a common Fc domain, as illustrated Figure 2A.To identify these HexAbs, we assign each of them a code of X-(Y)-(Y), where X and Y are specific designations given to differentiate the antibodies, with the Fab distinguished from the IgG by enclosing its designation in a parenthesis.The present notation is applicable to denote HexAbs that are either bispecific [36][37][38] or monospecific [58].As an example, 20-( 22)- (22), defines the bsHexAb comprising a divalent anti-CD20 humanized IgG (veltuzumab or hA20) and a pair of dimeric anti-CD22 humanized Fab's (epratuzumab or hLL2).Likewise, 22-( 20)- (20) specifies the bsHexAb comprising a divalent hLL2 IgG and a pair of dimeric hA20 Fab's, whereas 20-( 20)- (20) describes the monospecific HexAb comprising a divalent hA20 IgG and a pair of dimeric hA20 Fab's.The designation of each antibody used in the construction of bsHexAbs is provided in Table 1.

Generation and Biochemical Analysis
To date, we have made a variety of bsHexAbs in the C H 3-format (Table 2) by combining, under mild redox conditions, a C H 3-AD2-IgG module with a C H 1-DDD2-Fab module, followed by purification with Protein A affinity chromatography.The individual modules used to assemble these HexAbs were produced in mammalian cell cultures and purified by either Protein A (to obtain C H 3-AD2-IgG) or kappa-select (to obtain C H 1-DDD2-Fab).The ensuing DNL reaction typically proceeded uneventfully, resulting in each final conjugate shown by SE-HPLC to consist of a major peak of the expected molecular size (~365 kDa), by SDS-PAGE to be of high purity, and by dynamic light scattering to have an averaged particle diameter of 15.83 nm, about 5 nm larger than an intact IgG [59].
We also investigated the extent of internalization of 20-( 22)-( 22) and 22-( 20)-( 20) into Raji cells by flow cytometry and made an intriguing observation [36].Live cells were incubated with PE-conjugated antibodies at 37 C for 1 h before trypsin digestion to remove non-internalized antibodies.The MFI of cells stained with PE-veltuzumab and PE-22-( 20)-( 20) was reduced by 90% and 85%, respectively, indicating that 22-( 20)-( 20) behaves like veltuzumab with a slow internalization rate.On the other hand, we found approximately 50% of the 20-( 22)-( 22) internalized, similar to the results obtained for epratuzumab.Thus, the internalization property of a bsAb composed of a rapid internalizing antibody, such as epratuzumab, and a slowly or non-internalizing antibody, such as veltuzumab, appears to depend on the relative valency of the two antibodies with different internalization characteristics.
Using a cell counting assay, we further found that 22-( 20)-( 20) and 20-( 22)-( 22) effectively inhibited the growth of three Burkitt lymphoma cell lines, Ramos, Raji and Daudi, at 15 nM or lower, whereas under the same conditions, epratuzumab alone was ineffective and veltuzumab alone or in combination with epratuzumab was less effective [36].Based on the EC 50 values determined from dose-response curves, 22-( 20)- (20) was more potent than 20-( 22)- (22) in the three lymphoma cell lines examined, and the observed direct toxicity was not appreciably affected for either 22-( 20)-( 20) or 20-( 22)-( 22) upon the addition of a crosslinking anti-human Fc antibody which, however, markedly increased the inhibitory activity of veltuzumab, but not epratuzumab, alone.
Finally, we have explored the signaling pathways involved in evoking direct toxicity of 20-( 20)-( 20), 22-( 20)-( 20), and 20-( 22)-( 22) in Daudi cells [37], and compared the results with cells treated with anti-IgM antibodies, or with veltuzumab or rituximab in the presence of a crosslinking antibody.Collectively, our findings indicate that the potent direct toxicity of the three CD20-targeting HexAbs is due to their multivalent binding ability, which lowers the threshold for modifying multiple signaling pathways, resulting in a new distribution of pro-and anti-apoptotic proteins that promotes growth arrest, apoptosis, and eventually cell death.Specifically, we showed the following.(i) The signaling events triggered by 20-( 20)-( 20), 22-( 20)- (20), or 20-( 22)-( 22) were quantitatively and qualitatively similar in Daudi cells, but distinct from those induced by anti-IgM; (ii) Although ligation of CD22 by epratuzumab failed to induce appreciable changes in the basal expression of a variety of signaling molecules examined, ligation of CD20 by veltuzumab or rituximab incurred the signaling events associated with the ERK and NF-B pathways, similar to the three CD20-targeting HexAbs.However, both veltuzumab and rituximab required a higher concentration to be effective and were less efficient in modulating the cell cycle regulators known to promote growth arrest (e.g., up-regulation of p21, p27 and Kip2 and down-regulation of cyclin D1 and phosphorylated Rb).Other notable differences included the inability of veltuzumab or rituximab to alter the levels of phosphorylated p38 and PTEN from untreated controls, whereas all three HexAbs increase phosphorylated p38 and PTEN levels significantly.The decrease in phosphorylated ERKs and the increase in phosphorylated p38 were also observed for the three CD20-targeting HexAbs in Raji cells; (iii) The apoptosis and inhibition of cell proliferation resulting from crosslinking veltuzumab or rituximab with a secondary antibody involves signaling events that were distinguishable from those associated with the CD20-targeting HexAbs, as manifested in phosphorylated ERK (increase vs. decrease), intracellular calcium (increase vs. no change), and mitochondrial membrane potential (loss vs. no change).Intriguingly, these effects translated to notable differences with regard to their relative potency for killing normal human B cells vs. human Burkitt lymphoma cells ex-vivo, because the bispecific 22-( 20)-( 20) and 20-( 22)-( 22) showed a higher toxicity to malignant than normal B cells, compared to veltuzumab and rituximab [36].Thus, the potential advantages of lacking CDC and the moderate but significant enhancement of ADCC observed for 22-( 20)- (20), as compared to epratuzumab, may lead to a more potent anti-lymphoma agent for clinical use.

bsHexAbs that Target CD20 and CD74
During the course of evaluating 22-( 20)-( 20) and 20-( 22)-( 22), we noted that neither bsHexAb was capable of inducing direct cytotoxicity in JeKo-1, a mantle cell lymphoma (MCL) line expressing comparable levels of CD22 and CD20 as Daudi cells.On the other hand, 20-( 74)-( 74) and 74-( 20)-( 20), the bsHexAbs derived from veltuzumab and milatuzumab, were highly cytotoxic when tested in JeKo-1 and two other MCL lines (Granta-519 and Mino), as well as in primary tumor cells from patients with MCL or CLL, all of which, in the absence of a crosslinking Ab, responded poorly to veltuzumab or milatuzumab alone or combined [38].Follow-up experiments to interrogate the intracellular events triggered by simultaneously ligating both CD20 and CD74, which could conceivably result only from the binding engagement via either bsHexAb or from crosslinking the 2 parental Abs with a secondary Ab, revealed the prominent roles of actin reorganization and lysosomal membrane permeabilization in the mechanisms of cell death.In addition, the juxtaposition of CD20 and CD74 on MCL cells by the bsHexAbs also induced homotypic adhesion and set off intracellular changes that included loss of mitochondrial transmembrane potential, production of reactive oxygen species, rapid and sustained phosphorylation of ERKs and JNK, and down-regulation of pAkt and Bcl-xL.In an ex vivo setting, both 20-( 74)-( 74) and 74-( 20)-( 20) displayed a higher potency in depleting lymphoma cells than normal B cells from whole blood [38].

Pharmacokinetics (PK) and in Vivo Anti-Tumor Activity
We used a bispecific ELISA to quantify the amount of 20-( 22)-( 22) and 22-( 20)- (20) in serum samples collected from PK studies in BALB/c mice and found [36] the two bsHexAbs displayed a shorter circulating half-life than their parental antibodies (24 to 37 h vs. 46 to 52 h).Because both bsHexAbs were stable in serum when assessed in vitro, their faster blood clearance was likely due to intracellular breakdown of the modular components, which presumably occurred after their uptake via the FcRn in the vascular endothelium of mice.Evidence for the slow dissociation of the bsHexAbs in vivo was provided by SE-HPLC analysis of the PK samples obtained 72 h after injecting radiolabeled 20-( 22)-( 22), which identified the presence of a new peak with a size of an IgG shown to be derived from veltuzumab, not epratuzumab, since it failed to bind the anti-idiotype antibody to epratuzumab.A parallel study using radiolabeled 22-( 20)-( 20) also revealed the presence in the 72-h PK samples of a new peak with a size of an IgG shown to be derived from epratuzumab, not veltuzumab, using an anti-idiotype antibody to veltuzumab.
In the third study, we examined the role of effector cells in the ability of 22-( 20)-( 20) or 20-( 22)-( 22) to inhibit tumor growth.Groups of 5 mice depleted of NK cells and neutrophils were administered i.v. with 230 µg of 22-( 20)-( 20) or 20-( 22)-( 22) on days 1, 3, 5, and 9.As controls, four groups of mice without depletion of NK cells and neutrophils were each treated with saline, epratuzumab (100 µg), or the two bsHexAbs at the same dose and schedule as the depleted groups.Treatments with 22-( 20)-( 20) or 20-( 22)-( 22) resulted in no survival benefit compared to animals in the depleted groups, since there was insignificant difference in the MST from the saline control (18 days vs. 21 days).In contrast, treatment of animals in the non-depleted groups with either 22-( 20)- (20) or 20-( 22)-( 22) significantly (P < 0.002) increased their survival, with MSTs of 63 and 91 days, respectively, compared to 21 days of the saline control and 28 days of the epratuzumab control.However, the observed difference in the MSTs between the 22-( 20)-( 20) and 20-( 22)- (22) was not statistically significant, perhaps because the relatively small number of animals included in these groups.These initial results underscore the importance of ADCC as the major mechanism of action in retarding tumor growth in animal models by antibodies that target CD20, CD22, or both.

The C K -format of bsHexAbs
The search for increased effector functions as well as a more favorable PK of the C H 3-based bsHexAbs led us to explore the potential of their C K -based counterparts (Figure 2B), in which the AD2 sequence was fused at the C-terminus of the kappa light chain of the intact IgG, resulting in a new series of bsHexAbs denoted as X*-(Y)-(Y), where the C K -AD2-IgG-X and the dimeric C H 1-Fab-DDD2-Y are identified as X* and (Y), respectively.Such notations allow a prompt discern of the C Kbased from the C H 3-based HexAbs.For example, 22*-( 20)-( 20) and 22-( 20)- (20) represent two different bsHexAbs, comprising a pair of dimeric C H 1-Fab-DDD2-hA20 linked to C K -AD2-IgG-hLL2 and C H 3-IgG-AD2-hLL2, respectively.The results obtained from comparing 22*-( 20)-( 20) with 22-( 20)- (20), as reported recently [40] and highlighted below, indicate the bsHexAbs with the C K -format not only bind to target cells as efficiently as the C H 3-format, but also exhibit superior Fc effector functions in vitro, as well as improved PK, stability, and anti-lymphoma activity in vivo, thus being favored for future clinical development.
Unlike 22-( 20)- (20), which was previously noted to display no measureable CDC and only weak ADCC, we found that 22*-( 20)-( 20) induced moderate CDC (Figure 4B) and exhibited a potent ADCC similar to veltuzumab (Figure 4C).These results implicate a potential strategy for imparting CDC and ADCC to an IgG with little or weak effector functions, such as epratuzumab, by constructing a bsHexAb of the C K -format in which the IgG of interest is linked to multiple antigen-binding fragments derived from a different antibody with potent effector functions.

Conclusions and Future Challenges
Since its inception in 2005, the DNL method, by combining DDD2-and AD2 modules derived from assorted classes of molecules that include antibodies [36,38,58,59], antibody fragments [32,61,62], cytokines [40,52,63,64], polyethylene glycols [57], and enfuvirtide [65], has provided over 100 different complexes with potential applications for targeted therapies of malignant, autoimmune and infectious diseases.As more modules develop, we envision a continuous expansion of the repertoire such that the ultimate promise of multivalent and multispecific agents for treating certain unmet medical needs may be fulfilled with their advantages of more selective and sustained binding to the target cells, potent effector functions, designed capability to inhibit multiple survival pathways, acquired ability to impart novel apoptotic signals, and constitutive cytotoxic activity.For HexAbs in general and bsHexAbs in particular, the future challenges lie in the identification of optimal pairs of parental antibodies for assembly into the C K -format to deliver effective therapy of solid cancers.We anticipate 22*-( 20)- (20) to be the first DNL-based bispecific hexavalent antibody for clinical evaluation in patients with diseases involving abnormal B cells or B-cell malignancies.
In closing, we should note that there are other multimerization approaches to generate bispecific antibodies, among which the "trimerbodies," as described only recently by Blanco-Toribior and her coworkers [66], were also designed as hexavalent antibodies, whereby its bispecific format is capable of binding to each target antigen trivalently.Such "trimerbodies" lack the Fc domain, can be generated by fusing single-chain variable fragments (scFv) with the same or different specificity to both N-and C-terminus of the trimerizing scaffold domain derived from human collagen XVIII noncollagenous-1 domain, and are produced as soluble proteins in mammalian cells.Beside HexAbs and "trimerbodies," alternative designs for hexavalent antibodies via recombinant methods, chemical conjugation, or a combination of both, are certainly conceivable.However, a discussion assessing the DNL and other potential approaches is beyond the scope of the current review, whose main purpose is to provide an up-to-date account on the development of bispecific, hexavalent antibodies using the DNL method.

Figure 1 .
Figure 1.Amino acid sequences (AD1, AD2, DDD1, and DDD2) and a schematic of a basic DNL complex comprising a dimer of A-DDD2 linked to a monomer of B-AD2.

Figure 2 .
Figure 2. Schematics of a C H 3-based HexAb (A) and a C K -based HexAb (B).

Table 1 .
Designations of antibodies used in the construction of bsHexAbs.

Table 2 .
Codes, modules, and targets of selected C H 3-based HexAbs.