Affinity maturation, humanization, and co-crystallization of a rabbit anti-human ROR2 monoclonal antibody for therapeutic applications

Antibodies are widely used as cancer therapeutics, but their current use is limited by the low number of antigens restricted to cancer cells. A receptor tyrosine kinase, receptor tyrosine kinase-like orphan receptor 2 (ROR2), is normally expressed only during embryogenesis and is tightly down-regulated in postnatal healthy tissues. However, it is up-regulated in a diverse set of hematologic and solid malignancies, thus ROR2 represents a candidate antigen for antibody-based cancer therapy. Here we describe the affinity maturation and humanization of a rabbit mAb that binds human and mouse ROR2 but not human ROR1 or other human cell-surface antigens. Co-crystallization of the parental rabbit mAb in complex with the human ROR2 kringle domain (hROR2-Kr) guided affinity maturation by heavy-chain complementarity-determining region 3 (HCDR3)-focused mutagenesis and selection. The affinity-matured rabbit mAb was then humanized by complementarity-determining region (CDR) grafting and framework fine tuning and again co-crystallized with hROR2-Kr. We show that the affinity-matured and humanized mAb retains strong affinity and specificity to ROR2 and, following conversion to a T cell–engaging bispecific antibody, has potent cytotoxicity toward ROR2-expressing cells. We anticipate that this humanized affinity-matured mAb will find application for antibody-based cancer therapy of ROR2-expressing neoplasms.

Infrequent identification of new suitable cancer antigens restricts the indications and patients suited for mAb therapies. Due to their overexpression on cancer cells, receptor tyrosine kinases (RTKs) have proven their general suitability as cancer antigens. However, only a few out of more than 30 RTKs are currently targeted by FDA-approved mAbs (2). Two RTKs not yet targeted by FDA-approved mAbs are ROR1 and ROR2, which are expressed in embryogenesis and tightly down-regulated in postnatal tissues (3)(4)(5)(6). A number of solid and hematologic malignancies have been shown to express ROR1 or ROR2, suggesting utility as targets for antibody-based cancer therapies (7). Multiple clinical trials for cancer therapies with ROR1-targeting antibodies are under way. Although ROR2-targeting campaigns were only recently translated from preclinical to clinical investigations (NCT03504488, NCT03393936, NCT03960060), they underscore the suitability and attraction of ROR2 as a candidate antigen for antibody-based cancer therapy.
ROR2 shares 58% amino acid sequence identity with ROR1 along with the same extracellular domain composed of an N-terminal Ig domain, a frizzled (Fz), and a kringle (Kr) domain (7,8). ROR2 is involved in the WNT signaling pathway when associated with its ligand WNT5A and facilitates polarization of cells during embryonic development along with regulating migration and differentiation (9 -12). While largely down-regulated after birth in mice (6) and humans (8,13), ROR2 is overexpressed in several cancers (7, 9), including solid malignancies, such as renal cell adenocarcinoma and subsets of breast cancer, and hematologic malignancies, such as multiple myeloma (14 -16). Among solid malignancies without FDA-approved and marketed antibody-based cancer therapies, a notable indication is sarcoma, where ROR2 overexpression was found in osteosarcoma, leiomyosarcoma, and gastrointestinal stromal tumor (13,17). Numerous studies show that ROR2 increases invasiveness and takes part in tumorigenesis, making ROR2 a promising cancer target and biomarker.
We previously reported the generation of a diverse panel of ROR2-targeting mAbs derived from a large naive rabbit antibody library by phage display (18). Among these, mAb XBR2-401 ("401") was shown to bind ROR2 with high affinity and exclusive specificity. Once converted into a chimeric antigen receptor T cell (CAR-T) format, 401 triggered selective killing in vitro (18). The epitope of 401 was mapped to the Kr domain. This plasma membrane-proximal location makes 401 a preferred candidate for T cell-engaging ROR2 ϫ CD3 bispecific antibodies (biAbs). This is suggested by our ROR1 ϫ CD3 biAb study, where ROR1-targeting mAb R11, with an epitope in the Kr domain, had superior in vitro and in vivo activity compared with ROR1 ϫ CD3 biAbs with plasma membrane-distal epitopes (19). To map this epitope, we co-crystallized R11 in single-chain variable fragment (scFv) format with the Kr domain of ROR1 (Protein Data Bank (PDB): 6BA5 (55)). Notably, R11 does not cross-react with the Kr domain of ROR2, and 401 does not cross-react with the Kr domain of ROR1. Determination of the precise interaction of 401 and the Kr domain of ROR2 would allow for a full understanding of the mAb and aid its preclinical and clinical development.
mAb 401 was selected as a chimeric rabbit/human Fab with rabbit variable domains and human constant domains (18). An issue with utilizing chimeric mAbs for therapy is potential immunogenicity. Studies have shown that the risk of immunogenicity is lower with humanized mAbs versus chimeric mAbs (20). The chimeric mAbs in this comparison contained murine variable domains, and it is not known whether chimeric mAbs with rabbit variable domains have comparable immunogenicity. A well-established strategy of humanization is to graft murine complementarity-determining regions (CDRs) into human frameworks. This was later optimized by identifying human frameworks with the highest identity to the original murine antibody and using them as the backbone for the murine CDRs (21,22). We and others showed that rabbit mAbs can also be humanized by CDR grafting (23). A number of humanized mAbs originating from the rabbit antibody repertoire are currently in clinical trials (24), with brolucizumab, an scFv-targeting VEGF in wet age-related macular degeneration, becoming the first to have received FDA approval on October 7, 2019. The United States Adopted Names Council and World Health Organization International Nonproprietary Names sector work together to update and determine classifications for the evolving world of nonhuman, chimeric, humanized, and human antibodies for diagnostic and therapeutic applications (25,26). The current parameters that define a humanized antibody include the variable region identifying closer with human than any other species sequences. It also requires the identity of the final variable region sequences, including CDR regions, to be Ͼ85% human using IMGT (26).
Here we report the generation of a humanized mAb that binds to a membrane-proximal epitope of ROR2 with high affinity and specificity and as such can be utilized as an antibody-based cancer therapy. First, we co-crystallized 401 in scFv format with the hROR2-Kr domain and used this information for affinity maturation by phage display. The in vitro evolved mAb was humanized by CDR grafting and rational back-mutations, confirmed to have retained its exclusive specificity to ROR2, and was again co-crystallized with the hROR2-Kr domain, providing a detailed picture of the paratope and epitope. Finally, conversion to a T cell-engaging ROR2 ϫ CD3 biAb demonstrated potent in vitro killing of ROR1Ϫ/ROR2ϩ but not ROR1ϩ/ROR2Ϫ cancer cells. We anticipate broad therapeutic utility of this reagent as T cell-engaging biAb, CAR-T, and other antibody-based cancer therapies.

Crystallization of mAb XBR2-401 in complex with the human ROR2 kringle domain
We previously reported a panel of 12 chimeric rabbit/human Fabs that were selected from a naive rabbit antibody library for binding to human ROR2 (18). Among these, mAb XBR2-401 ("401") in Fab and IgG1 format (Fig. 1A) was shown to be specific for the kringle domain of ROR2 (hROR2-Kr) and to recognize both human and mouse orthologs but not its closest relative, ROR1 (18). To define the 401 paratope and epitope, we used X-ray crystallography to solve the structure of 401 in scFv ( Fig. 1A) format in complex with hROR2-Kr at 1.2-Å resolution ( Fig. 1B and Table S1) (Protein Data Bank ID (PDB): 6OSH). The crystal contained one complex in the asymmetric unit. All residues in the crystal were well-resolved except for the 15amino acid scFv linker. The buried surface area between 401 and hROR2-Kr was 720 Å 2 , which comprised 7.0 and 15.9% of the total surface area of 401 and hROR2-Kr, respectively. The van der Waals contacts were dominated by HCDR2 and LCDR3 of 401. Notably, Ala-95 (Kabat numbering) from LCDR3 was nestled in a shallow hydrophobic pocket created by hROR2-Kr loop 3, 5, and 6 residues (27), Leu-350, Pro-368, Gln-371, Trp-376, Phe-378, and Met-386. On the other hand, His-349 of loop 3 from hROR2-Kr projected into the main pocket formed by the CDRs and made a salt bridge to Asp-32 of LCDR1 (Table 1). The interface also contained numerous direct and water-mediated hydrogen bond interactions dominated by residues from HCDR2 and LCDR1 (Table 1). Compared with the residues from HCDR2 and LCDR3 that are heavily involved in epitope recognition, Trp-96 (Kabat numbering) from HCDR3 provided limited interactions with hROR2-Kr through a suboptimal hydrogen bond with His-348 and van der Waals interactions with His-349 with the potential to form abond (Fig. 1B). This observation posed an opportunity for optimizing HCDR3 binding to the kringle domain.
hROR2-Kr and hROR1-Kr share 58% amino acid sequence identity (3). When the epitope residues of hROR2-Kr that are recognized by 401 were compared with those mediating R11: hROR1-Kr recognition (19), no residue overlap was found (Fig.  S1). This observation explains why there is no cross-reactivity between 401 and R11 despite hROR2-Kr and hROR1-Kr's homologous amino acid sequences. When comparing the antibody-bound kringle domains from the 401:hROR2-Kr and the R11:hROR1-Kr complex, the root mean square deviation (RMSD) of C␣ positions was found to be 0.695 Å, revealing a highly conserved tertiary structure of the two Kr domains.
We also crystallized and solved the structure of antibodyunbound hROR2-Kr at 1.1 Å resolution (PDB: 6OSN) (Fig. S1B (right) and Table S1). The RMSD of the unbound hROR2-Kr Therapeutic monoclonal antibody to ROR2 structure to 401-bound hROR2-Kr was 0.383 Å, revealing only minor differences between the coordinates. Notably, in the unbound hROR2-Kr structure, Arg-385 formed a mixed salt bridge/hydrogen bond interaction with an acetate from the crystallization solution (Fig. S1B, right). The binding site, which overlaps with the canonical lysine-binding sites (LBSs) in other kringle domains (28), was partially covered by 401 in the crystal structure of the 401:hROR2-Kr complex. Superposition of the unbound hROR2-Kr structure to the 401-bound hROR2-Kr showed a minor shift in loop 5 due to the bound acetate ion (Fig.  S1B, right).
Overall, our findings of no overlap between ROR2 and ROR1 epitopes along with suboptimal binding of 401's HCDR3 to hROR2-Kr revealed opportunities for in vitro affinity maturation.

Affinity maturation via phage display
A phage display library was constructed to conduct focused mutagenesis on 401's HCDR3 residues 96 and 97 (Kabat numbering; Fig. 2) with 0, 1, or 2 additional randomized residues ( Table 2). Additional randomized positions were investigated because the co-crystal structure of 401:hROR2-Kr depicted an open cavity between hROR2 and the HCDR3, which could be filled by a longer matured HCDR3 and improve mAb affinity. Selection for hROR2-Fc binding was performed three ways: surface, surface competition, and solution competition panning (see "Experimental procedures"). Both competition panning protocols applied selection pressure toward Fabs with lower dissociation rate constants (k off ) and thus higher affinity. From the three combined libraries, 144 clones were selected and analyzed via ELISA for ROR2 binding and Fab expression  Therapeutic monoclonal antibody to ROR2 from supernatants. The top 12 clones with the highest absorbance ratio (hROR2 binding to expression) were purified. Using surface plasmon resonance (SPR), thermodynamic (K D ) and kinetic parameters (k on and k off ) of the interaction with hROR2 were determined. Clone XBR2-401-X3.12 (X3.12), which was obtained from the X3 library, revealed the highest affinity (K D ϭ 0.7 nM) ( Table 3 and Fig. S2A), which is at least a 5-fold improvement from 401. The X3.12 HCDR3 sequence differs from 401 at two residue positions and is one residue longer, changing the HCDR3 sequence from DWTSLNI to DDRWSLNI (Table 3). To make the affinity-matured X3.12 more therapeutically relevant, the next step was to employ humanization.

Humanization by CDR grafting
Humanization of the affinity-matured chimeric rabbit/human X3.12 Fab was performed in three main steps. First, the human germlines with the closest identity to X3.12's variable light-chain (V L ) and variable heavy-chain (V H ) amino acid sequences were identified using IgBlast (see "Experimental Procedures" for online tools). The IMGT repertoire was then referenced to eliminate germlines with more than three polymorphisms. The human germlines that had the highest amino acid sequence identity to X3.12 with the least number of polymorphisms were heavy-chain germlines IGHV3-66*03 and IGHV3-48*03 and light chain germline IGKV1-NL1*01, which are 54.3, 54.8, and 65.6% identical to the X3.12 heavy and light chain, respectively. Second, CDRs from X3.12 determined using Kabat numbering were grafted into these three framework sequences (Fig. 2, V L 1, V H 1, and V H 2). Third, residues determined to preserve affinity (29) were back-mutated from the human germline residues to the original rabbit residues. From these three steps comprised of CDR grafting and rational back-mutations, four heavy chain (V H 1-V H 4) and two light chain variants (V L 1 and V L 2) were formed (Fig. 2). These humanized chains were compared with the IMGT database of human germline antibody sequences using the IMGT/Domain-GapAlign tool to determine the human identity percentage. The human identity of V L 1, V L 2, and V H 1-V H 4 was 88.6%*, 87.1%, 86.6%, 87.8%, 73.5%*, and 80.4%*, respectively, where the percentages with an asterisk indicate that the first "hit" on IMGT DomainGapAlign was not human. World Health Organization standards state that the first "hit" on IMGT Domain-GapAlign tool must be human along with human identity being above 85% for the antibody to be considered humanized (26). All heavy-and light-chain Fab combinations were cloned into a pET11a variant (30) and expressed in the Escherichia coli Rosetta strain followed by quantification of Fab expression and hROR2 binding via ELISA to eliminate nonbinding clones. The remaining clones hX3.12.5, hX3.12.6, hX3.12.7, and hX3.12.8 had higher binding/expression ratios compared with X3.12. The overall percentage identity of these variants to human germlines was 87, 87, 80, and 84%, respectively, and all utilized V L 2 (Table 4). Therefore, hX3.12.5 and hX3.12.6 are considered humanized, whereas hX3.12.7 and hX3.12.8 are considered chimeric mAbs by World Health Organization standards.

Characterization of affinity-matured and humanized Fabs
Following expression and purification (Fig. S3A), the affinities of hX3.12.5, hX3.12.6, hX3.12.7, and hX3.12.8 were deter- The location of the four framework regions (FWRs) and the three CDRs is indicated. The numbers refer to Kabat numbering of variable domain residues shown in single-letter code. Residues shown in red were back-mutated to the original rabbit residue. Dots, identical residues in the alignment to V L 1 or V H 1. CDR residues are shown in boldface type. Top, the V L 1 FWRs are derived from human germline IGKV1-NL1*01; the V L 1 CDRs are grafted from X3.12. The V L 2 amino acid sequence is the same as V L 1 but with seven back-mutations from the human germline to the original rabbit residues of X3.12. Bottom, the V H 1 and V H 2 FWRs are derived from human germlines IGHV3-66*03 and IGHV3-48*03, respectively, and their CDRs are grafted from X3.12. V H 3 and V H 4 vary from V H 1 and V H 2 by back-mutating FWR residues from the human germline to the original rabbit residues of X3.12.
Similar to 401, in hX3.12.6, Ala-95 was buried in hROR2-Kr, and the salt bridge between light-chain Asp-32 and hROR2-Kr was also retained. All hydrogen bond interactions and van der Waals contacts present in 401:hROR2-Kr remained intact in the hX3.12.6:hROR2-Kr complex except for the changes in the HCDR3 due to affinity maturation, which include Asp-96 and Arg-97. Whereas these two residues do not directly interact with hROR2-Kr, they help to properly position Trp-98, which does contact hROR2-Kr. Trp-98 in hX3.12.6 further improves interactions with hROR2-Kr, having been optimized from 401's Trp-96, located at the tip of the HCDR3 loop (Fig. 1, B and C). Unlike in 401, the side chain of Trp-98 made an optimal hydrogen bond interaction with the backbone oxygen of hROR2-Kr's His-348. The side chain of Trp-98 also displayed geometric characteristics of -/-cation interactions with hROR2-Kr His-349 (Fig. 1C (bottom) and Table 1). The RMSD between 401 and hX3.12.6 in their respective co-crystal structures was found to be 0.446 Å, suggesting subtle differences between the structures (Fig. S1A). The crystallized kringle domains in complex with either 401 or hX3.12.6 had an RMSD of 0.279 Å, indicating no relevant change between the two kringle domains. Collectively, these findings confirmed that our rational design of affinity maturation was critical in the improvement of ROR2 binding.
To confirm that these biAbs are specifically killing via binding ROR2 on 786-O cells and CD3, we tested a ROR2Ϫ, ROR1ϩ cell line, MDA-MB-231, and found that all ROR2 ϫ CD3 biAbs were inactive up to 1 g/ml (Fig. 5B). The positive control ROR1 ϫ CD3 biAb did show specific cell lysis as expected due to its binding to ROR1 on MDA-MB-231 cells. T-cell activation

Therapeutic monoclonal antibody to ROR2
was quantified by flow cytometry using an anti-CD69 mAb, a known marker of early T-cell activation. Humanized biAbs hX3.12.6 ϫ v9, hX3.12.5 ϫ v9, and parental 401 ϫ v9 incubated with T cells at 0.2 g/ml up-regulated CD69 on over 50% of T cells in the presence of 786-O but not MDA-MB-231 cells (Fig.  5C). The negative control hX3.12.6 scFv-Fc did not reveal upregulation of CD69. The release of type 1 cytokine IFN-␥ was assessed by ELISA, where all ROR2 ϫ CD3 biAbs caused cytokine release in the presence of ROR2ϩ but not ROR2Ϫ target cells (Fig. 5C). As shown previously (19), R11 ϫ v9 caused comparable cytokine release in the presence of ROR1ϩ target cells.

Discussion
Obtaining a humanized antibody with high affinity for ROR2, a cancer target with relatively low cell surface density, facilitates investigations for therapeutic utility. A previous study (18) took the first step in selecting a well-defined anti-ROR2 mAb, 401, from a naive rabbit antibody library by phage display and confirming its high affinity and specificity. In the current study, we further characterized 401 by co-crystallization with its antigen, hROR2-Kr. The crystal structure revealed that the hROR2-Kr epitope residues bound by 401 did not overlap with the hROR1-Kr epitope residues bound by R11. This further defined 401 as specific to hROR2-Kr and not cross-reactive with ROR1. The dominant interactions between 401:hROR2-Kr were within the LCDR3 and HCDR2. Notably, LCDR3 domi-nance in rabbit antibody paratopes is not uncommon (39). Between HCDR3 and the hROR2-Kr there was weak hydrogen bonding and potential for ainteraction between Trp-96 and hROR2-Kr His-349. From these findings in the crystal structure, we hypothesized that elongating and mutating the HCDR3 would improve 401's affinity.
Affinity maturation of 401 was done by generating and selecting an HCDR3-targeted Fab-phage display library. This library allowed selection from variants that incorporated mutations in positions 96 and 97 of V H with the addition of up to two additional residues in the HCDR3. Selecting from this library produced 10 Fabs sustaining affinities below 10 nM as determined by SPR. The top 10 Fabs displayed varying lengths of HCDR3, with four containing one additional residue and three containing two additional residues. This suggests that seven of the 10 top clones would not have been selected from a strictly mutagenized HCDR3 library. Also, within the top four clones, three contained a Trp at position 98 or 99 compared with 401's Trp at 96, suggesting that improvedinteractions were critical in the affinity maturation. The top Fab, X3.12, had a K D of 0.72 nM, a 5-10-fold improvement from parental mAb 401. The parental HCDR3 sequence starting at position 95 is DWTS, where the affinity-matured sequence is DDRWS. The affinitymatured HCDR3 sequence of X3.12 originated from the X3 library, as there is an extra residue in addition to the two ran- Based on independent triplicates shown as mean Ϯ S.D. (error bars), one-way analysis of variance was used to analyze significant differences between ROR2 ϫ CD3 (or ROR1 ϫ CD3) biAbs and the monospecific scFv-Fc negative control (****, p Ͻ 0.0001).

Therapeutic monoclonal antibody to ROR2
domized residues at positions 96 and 97. This selection emphasizes the value of structural guidance in affinity maturation as additional residues were incorporated in order to fill the cavity seen in the crystal structure between HCDR3 and hROR2-Kr.
With the matured affinity of X3.12, the next step was to make the mAb therapeutically relevant by humanization. Humanization of the rabbit variable domains is expected to lower the risk of possible immunogenicity after repeated administration (39). CDR grafting and rational back-mutations to X3.12 produced eight humanized variants. The optimal humanized anti-ROR2 mAb was hX3.12.6, which had a K D of 3.8 nM, a Ͼ5-fold improvement from h401.6, which contains the parental HCDR3 and the humanized frameworks from hX3.12.6, improving similarly to the affinity maturation. Thus, reversing the order of affinity maturation and humanization would have likely yielded similar clones. As discussed for parental 401, the light chain contained a large portion of the hydrogen bonds along with burial of the LCDR3 in the hROR2-Kr. These interactions are critical for retention of affinity and paratope structure, which supported the idea of conserving a higher percentage of the original rabbit residues in the light chain than the heavy chain when humanizing. All anti-ROR2 humanized variants presented (hX3.12.5, hX3.12.6, hX3.12.7, and hX3.12.8) maintained single-digit nanomolar affinity for hROR2-Kr. Humanization did not impair the mouse/human ROR2 crossreactivity, which is an important characteristic for preclinical studies.
To distinguish the structural differences due to affinity maturation and humanization, the hX3.12.6:hROR2-Kr complex crystal structure was obtained. The percentage of buried surface area of each constituent in the 401:hROR2-Kr and hX3.12.6:hROR2-Kr complexes was 7.0%:15.9% and 6.8%:16%, respectively. The RMSD between the two co-crystal structures was 0.474 Å, with respect to the C␣ positions, showing that they are essentially identical. The only noticeable difference between 401 and hX3.12.6 crystal structures is the HCDR3 loops due to affinity maturation changing the sequence, as discussed above, and a slight shift in HCDR1 due to changing the framework residue from Leu-29 to Phe-29 during humanization, located in the human framework region directly upstream of HCDR1. However, we mutated this residue back to Leu-29 and did not observe a notable difference in affinity (data not shown).
The kringle domain takes part in protein-protein interactions and is thought to be a binding mediator (27). In plasminogen, kringle domains bind to effector molecules such as fibrin via LBSs (28). Our crystal structures of ROR2-Kr in complex with 401 and hX3.12.6 revealed that it contains an LBS that may be involved in receptor-ligand or receptor-receptor (cis and trans) interactions. Whereas a functional role of the LBS of ROR2-Kr remains unknown, it is interesting to note that 401 and hX3.12.6 partially cover it and as such may sterically hinder it from binding a ligand or co-receptor.
Recent immunohistochemistry studies utilizing a mAb binding to an intracellular epitope found that ROR1 is more widely expressed than originally thought (40). There is currently limited knowledge of ROR2 compared with ROR1, mostly due to poor reagents and inconsistency. For example, one study found two of three commercially available anti-human ROR2 anti-bodies have off-target cross-reactivity (41). This issue is widespread, but unreliable studies can be reduced by having mAbs with defined and disclosed V H and V L sequences that are wellcharacterized (42). This is the case for the anti-ROR2 mAbs that we report here, laying the groundwork for consistency in the field. However, as in the case of ROR1, a mAb binding to an intracellular epitope of ROR2 may be a preferred reagent for immunohistochemistry and, as such, a suitable companion diagnostic. A recent study by Hellmann et al. (43) described the selection of fully human anti-human ROR2 mAbs and their conversion to antibody-drug conjugates. Although these mAbs are human, most are not cross-reactive with mouse ROR2. Further studies are needed to confirm the exclusive specificity of these mAbs for ROR2 and their epitopes.
Previous studies have shown that membrane-proximal epitopes of ROR1 can mediate more potent responses against cancer cells when targeted by CAR-Ts and T cell-engaging biAbs (18,19,44). These studies prompted us to convert hX3.12.6, which binds a membrane-proximal epitope on hROR2-Kr, into a T cell-engaging biAb. In the current study, we show that hX3.12.6 ϫ v9 in an scFv-Fc aglycosylated format can cause T-cell activation specifically as seen in the in vitro cytotoxicity toward 786-O (ROR2ϩ) but not MDA-MB-231 (ROR2Ϫ) cells. This suggests therapeutic utility of hX3.12.6 as the ROR2-targeting arm of T cell-engaging biAbs. In addition, as we reported previously, parental mAb 401 can be converted to potent and specific CAR-Ts (18). Its epitope mapping by co-crystallization, affinity maturation, and humanization in the current study affords a highly attractive molecule for investigating ROR2 targeting by T cell-engaging biAbs and CAR-Ts in preclinical models of ROR2-expressing malignancies.
Protein purification-Bacterial pellets were resuspended in sonication buffer (20 mM HEPES, pH 8.0, 500 mM NaCl, 15 mM imidazole, 10% (v/v) glycerol), sonicated in an ice-water bath, and centrifuged for 25 min at 53,300 ϫ g. The supernatants were loaded on a custom-packed 10-ml HIS-Select column (Sigma-Aldrich) and washed with sonication buffer. Bound proteins were eluted with a linear gradient of imidazole from 15 to 500 mM. The eluted proteins were treated overnight at 4°C with thrombin (Sigma-Aldrich) to remove the N-terminal hexahistidine tags at hROR2-Kr. The cleaved proteins were purified further on a Superdex 200 26/60 column (GE Healthcare) equilibrated with 50 mM NaCl, 10 mM HEPES, pH 7.4.
Crystallization and structure determination-Crystals of the scFv401:hROR2-Kr complex were grown by vapor diffusion at room temperature (RT) using 1.5 l of 14 mg/ml protein and an equal volume of precipitant containing 0.1 M sodium citrate tribasic dihydrate, 15% (w/v) PEG 3350 and were fully grown within 2 days. hROR2-Kr produced clustered crystals in vapor diffusion at RT using 2 l of 14 mg/ml protein with 1 l of precipitant containing 0.2 M lithium acetate, 20% (w/v) PEG 3350. The crystal clusters were crushed and seeded to drops equilibrated with a protein/precipitant ratio of 3:1 to obtain single crystals. Crystals of the scFv hX3.12.6:hROR2-Kr complex were grown by vapor diffusion at RT using 1.5 l of 3 mg/ml protein and an equal volume of precipitant containing 10 mM MgCl 2 hexahydrate, 5 mM nickel (II) chloride hexahydrate, 0.1 M Na-HEPES, pH 7.0, 13% (w/v) PEG 4000. The crystals were flash-frozen in liquid nitrogen using nylon loops after removing excess mother liquor. Diffraction data sets with Bragg spacings set to 1.1 Å for both scFv401:hROR2-Kr and hROR2-Kr were collected on a Rayonix MX300 detector at the Advanced Photon Source (APS) beamline LS-CAT 21-ID-F synchrotron facility (Argonne National Laboratory). A diffraction data set with Bragg spacings set to 1.3 Å for scFv hX3.12.6: hROR2-Kr was collected on a PILATUS3 S 6M detector at the Advanced Light Source (ALS) beamline 5.0.2 synchrotron facility (Lawrence Berkeley National Laboratory). Data sets were processed with autoPROC using XDS as the processing engine (46). The structures were solved by the molecular replacement method using PHASER (47) with PDB entry 6BA5 (55) (scFvR11:hROR1-Kr) (19) as the search model. Crystallographic refinements were performed using PHENIX version 1.14 (48). Manual rebuilding and adjustment of the structures were done in Coot (49). Data processing and refinement statistics are shown in Table S1. Molecular images (Fig. 1 (B and C) and Fig. S1 (A and B)) were created using PyMOL (50). Interaction interfaces were analyzed using PDBePISA (51,52). Structure validations were carried out with MolProbity (53).
Library selection-Following published protocols for the selection of chimeric rabbit/human Fab by phage display (54), three different panning approaches were investigated, where the first was conventional surface panning (3 rounds) using 1 g of the hROR2-Fc (18) in 25 l of PBS for immobilization on a 96-well ELISA plate (Costar 3690; Corning), 3% (w/v) BSA in PBS for blocking, and 10 wash steps using 0.05% (v/v) Tween 20 in PBS (TPBS). The second approach was surface competition panning (using the second round of the conventional surface panning, one round was conducted) using 100 ng of hROR2-Fc, immobilized and blocked as above, followed by a 2-h pre-incubation with a 10-fold molar excess of the parental Fab 401 and 15 wash steps with TPBS. The third approach was solution competition panning (5 rounds) using 2-fold decreasing amounts of biotinylated (54) hROR2-Fc (100 to 6.25 ng in PBS) and pre-incubation with a 10-fold molar excess of the parental Fab 401 to the hROR2-Fc at the various concentrations. At Therapeutic monoclonal antibody to ROR2 each step, capturing with streptavidin-coated magnetic beads (Dynabeads MyOne Streptavidin C1; Thermo Fisher Scientific), acid elution using 100 mM glycine-HCl (pH 2.2), coupling of panning rounds 2 and 3 and rounds 4 and 5 (i.e. without intermittent reamplification), and increasingly stringent wash steps (0.05% (v/v) to 0.5% (v/v) Tween 20 in PBS) were conducted. Using published protocols (54), final output colonies were screened by a Fab ELISA using immobilized hROR2-Fc, and the HCDR3 of positive clones was determined by DNA sequencing.

Humanization
Humanization of X3.12 was done by finding the closest human germline(s) using IgBlast (RRID:SCR_002873) with the least amount of polymorphisms, which was determined using IMGT's IGHV and IGKV mammalia human (Homo sapiens) links (RRID:SCR_018220). Rational mutations (29) were performed in varying severity to determine mutations that were necessary to retain ROR2 affinity.

Fab cloning, expression, and purification
All selected Fab variants were cloned as described with modifications (18). Briefly, Fab variants were cloned into bacterial expression vector pET11a (30), transformed into E. coli strain Rosetta (DE3) (EMD Millipore), and tandemly purified from culture supernatants using a 1-ml HiTrap Kappa Select HP column followed by a 1-ml HisTrap HP column in conjunction with an Ä KTA FPLC instrument (all from GE Healthcare). (The initial top 12 chimeric rabbit/human anti-human Fabs were purified using only the 1-ml HisTrap HP column.) Purity of protein was analyzed by SDS-PAGE and Coomassie Blue staining, and A 280 absorbance was used to determine the concentration of purified Fab variants.

Surface plasmon resonance
Kinetic and thermodynamic parameters for the ROR2 binding of purified anti-ROR2 Fab variants were measured by the use of SPR as described previously (18), performed on a Biacore X100 instrument using Biacore reagents and software (GE Healthcare). Briefly, a CM5 sensor chip was immobilized with a mouse anti-human IgG C H 2 mAb to capture the hROR2-Fc antigen. Each Fab variant was diluted to 100 nM 1ϫ HBS-EPϩ running buffer and further diluted 2-fold using 1ϫ HBS-EPϩ running buffer to make five dilutions in total with a replicate of the lowest concentration after measuring the highest concentration to confirm regeneration of the sensor chip.

Thermostability assay
Steps were followed as described in the LightCycler 480 Instrument Quick Guide (Roche Applied Science) for protein melting. Optimal conditions were determined using parental Fab 401 at 1 mg/ml and the suggested Optimization Table 1 in the Quick Guide. Roche Protein Melting software was used for analysis. The optimal conditions required 0.5 l of 1 mg/ml Fab, 1.0 l of SYPRO Orange Dye 100ϫ stock, 8.50 l of Dulbecco's PBS. All Fabs were tested under these conditions and in triplicates.

Functionality studies
Production of ROR2 ϫ CD3 bispecific antibody-Cloning, expression, and purification of ROR2 ϫ CD3 biAbs in heterodimeric aglycosylated scFv-Fc format followed a previously described protocol with modifications (34). In short, the scFvencoding sequences were synthesized as gBlocks containing a signal peptide-encoding sequence at the N terminus (Integrated DNA Technologies). Overlap extension PCR was used to include sequences encoding hinge and heavy-chain constant domains C H 2 and C H 3 of human IgG1. Previously described hole and knob mutations (37) were included in the CD3 scFvhinge-C H 2-C H 3-and ROR2 scFv-hinge-C H 2-C H 3-encoding sequences, respectively. The aglycosylation mutation N297A in C H 2 was included in both. These scFv-Fc-encoding sequences were then inserted into mammalian expression vector pCEP4 using KpnI and XhoI restriction sites. Following DNA sequencing (Eton Bioscience) for verification, the plasmids were transfected into HEK 293F cells (Thermo Fisher Scientific) using polyethyleneimine (Polysciences) at 3 ϫ 10 6 cells/ml cultured in 150 ml of FreeStyle medium (Thermo Fisher Scientific) shaking at 37°C in an atmosphere of 8% CO 2 and 100% humidity. After 6 -12 h, an additional 150 ml of FreeStyle medium was added. Supernatants were collected after 3 days followed by filtration and purification using a 1-ml HiTrap Protein A HP column (GE Healthcare) in conjunction with an Ä KTA FPLC instrument (GE Healthcare) followed by size-exclusion chromatography using a Superdex 200 10/300 GL column (GE Healthcare) equilibrated with water followed by Dulbecco's PBS. Yields were typically ϳ5-10 mg/liter. The purity of the biAbs was confirmed by SDS-PAGE followed by Coomassie Blue staining and quantified by A 280 absorbance.
Flow cytometry-Similar to standard methods and as described previously (18,19), staining of 1 ϫ 10 5 target cells was done with 5 g/ml Fab or biAb in 100 l of cytometry buffer (PBS supplemented with 1% (w/v) BSA and 0.1% (w/v) sodium azide). After washing, the cells were incubated with a 1:1,000 dilution of phycoerythrin-conjugated goat anti-human IgG F(abЈ) 2 fragment-specific pAbs or Alexa Fluor 647-conjugated donkey anti-goat IgG (HϩL) pAbs (both in F(abЈ) 2 format from Jackson ImmunoResearch) in 100 l of flow cytometry buffer on ice for 1 h. Alexa Fluor 647-conjugated mouse anti-human CD69 mAb was purchased from BioLegend. Cells were analyzed using a FACSCanto instrument (BD Biosciences) and FlowJo analytical software (Tree Star).
In vitro cytotoxicity and T-cell activation assays-Cytotoxicity was measured by using CytoTox-Glo (Promega) following the manufacturer's protocol and a previous publication (34) with minor modifications. Primary T cells expanded from healthy donor PBMCs were used as effector cells, and 786-O or MDA-MB-231 cells were used as target cells at an effector/ target cell ratio of 10:1. Cells were incubated in X-VIVO 20 medium (Lonza) with 5% (v/v) off-the-clot human AB serum. Target cells (2 ϫ 10 4 ) were first incubated with the biAbs before Therapeutic monoclonal antibody to ROR2 adding the effector cells (2 ϫ 10 5 ) in a final volume of 100 l/well in a 96-well tissue culture plate followed by incubation at 37°C for 16 h. A biAb concentration range from 2 ng/ml to 1 g/ml was used. Plates were centrifuged, and 50 l of the supernatants were transferred into a 96-well clear-bottom whitewalled plate (Costar 3610, Corning) containing 25 l/well CytoTox-Glo reagent. After a 15-min incubation at RT, a Spec-traMax M5 instrument was used to read the plates with Soft-Max Pro software set to luminescence. Following the ELISA Ready-SET-Go! Reagent protocols (eBioscience), additional supernatant from the previous study was diluted 20-fold and used in a human IFN-␥ ELISA.

Data availability
Crystal structure files have been deposited to the Protein Data Bank and can be found by searching accession numbers 6OSH, 6OSV, and 6OSN, which correlate to 401:hROR2-Kr, hX3.12.6:hROR2-Kr, and hROR2-Kr crystal structures, respectively. All other data are contained within the article and supporting material.