Potent and selective antitumor activity of a Tcell-engaging bispecific antibody targeting amembrane-proximal epitope of ROR1Junpeng Qia, Xiuling Lia, Haiyong Penga, Erika M. Cookb, Eman L. Dadashianb, Adrian Wiestnerb, HaJeung Parkc,and Christoph Radera,1

aDepartment of Immunology and Microbiology, The Scripps Research Institute, Jupiter, FL 33458; bHematology Branch, National Heart, Lung, and BloodInstitute, National Institutes of Health, Bethesda, MD 20892; and cX-Ray Crystallography Core, The Scripps Research Institute, Jupiter, FL 33458

Edited by Ira Pastan, National Cancer Institute, NIH, Bethesda, MD, and approved May 7, 2018 (received for review November 16, 2017)

T cell-engaging bispecific antibodies (biAbs) present a promisingstrategy for cancer immunotherapy, and numerous bispecificformats have been developed for retargeting cytolytic T cellstoward tumor cells. To explore the therapeutic utility of T cell-engaging biAbs targeting the receptor tyrosine kinase ROR1,which is expressed by tumor cells of various hematologic andsolid malignancies, we used a bispecific ROR1 × CD3 scFv-Fc formatbased on a heterodimeric and aglycosylated Fc domain designedfor extended circulatory t1/2 and diminished systemic T cell activa-tion. A diverse panel of ROR1-targeting scFv derived from immuneand naïve rabbit antibody repertoires was compared in this bispe-cific format for target-dependent T cell recruitment and activation.An ROR1-targeting scFv with a membrane-proximal epitope, R11,revealed potent and selective antitumor activity in vitro, in vivo,and ex vivo and emerged as a prime candidate for further pre-clinical and clinical studies. To elucidate the precise location andengagement of this membrane-proximal epitope, which is con-served between human and mouse ROR1, the 3D structure of scFvR11 in complex with the kringle domain of ROR1 was determinedby X-ray crystallography at 1.6-Å resolution.

cancer immunotherapy | bispecific antibodies | antibody engineering |ROR1 | X-ray crystallography

ROR1 is a receptor tyrosine kinase uniformly expressed on thecell surface of malignant B cells in chronic lymphocytic leu-

kemia (CLL) (1–4) and mantle cell lymphoma (MCL) (5–7) butnot on healthy B cells or, with some exceptions (8), other healthycells and tissues of patients with cancer. In addition to leukemiaand lymphoma, ROR1 is also expressed in subsets of carcinoma,sarcoma, and melanoma, i.e., in all major cancer categories (9).Thus, ROR1 is a highly attractive candidate for targeted cancertherapy. Ongoing phase I clinical trials are assessing the safety andactivity of an mAb (ClinicalTrials.gov identifier NCT02222688)and chimeric antigen receptor (CAR)-engineered T cells (CAR-Ts; ClinicalTrials.gov identifier NCT02706392) targeting ROR1 inhematologic and solid malignancies.By using phage display, we previously generated a panel of

rabbit anti-human ROR1 mAbs from immune and naïve rabbitantibody repertoires (10, 11). These mAbs were shown to binddifferent epitopes on ROR1 with high specificity and affinity. AsCAR-Ts, they mediated selective and potent cytotoxicity towardROR1-expressing malignant cells (11, 12). One of these CAR-Tsbased on rabbit anti-human ROR1 mAb R12 demonstratedsafety and activity in nonhuman primates (13) and was translatedto the ongoing phase I clinical trial.With a mechanism of action (MOA) that is conceptually re-

lated to the MOA of CAR-Ts (14, 15), T cell-engaging bispecificantibodies (biAbs) are an alternative strategy for cancer immu-notherapy. Like CAR-Ts, T cell-engaging biAbs utilize the powerof T cells for tumor cell eradication, but, unlike CAR-Ts, aremanufactured and administered similarly to conventional mAbsand are cleared from the patient with cancer. The concept of

retargeting cytolytic T cells toward tumor cells by T cell-engagingbiAbs has had substantial clinical success with a CD19 × CD3biAb in bispecific T cell-engager (BiTE) format, blinatumomab,which received US Food and Drug Administration (FDA) ap-proval for the treatment of refractory or relapsed B cell-precursoradult lymphoblastic leukemia in 2014 (16), and with numerousother formats to various targets and indications under investiga-tion in clinical trials (17, 18). In view of their similar MOAs, CAR-Ts and T cell-engaging biAbs can be accompanied by potentiallydangerous adverse events, particularly cytokine release syndromeand central nervous system toxicity (19). Although both on-targetand off-target effects contribute to adverse events, targeting cellsurface antigens that are restricted to tumor cells is generallythought to afford superior safety profiles compared with targetingmore widely expressed cell surface antigens.In addition to the restricted expression of ROR1 on tumor

cells, its relatively large extracellular domain (ECD) consisting ofan Ig, a frizzled (Fz), and a kringle (Kr) domain, together com-prising ∼375 extracellular amino acids, provides an opportunity tocompare the therapeutic utility of T cell-engaging biAbs that bindto membrane-distal and membrane-proximal epitopes of ROR1.Efficient formation of cytolytic synapses (20) between T cells andtumor cells is thought to entail an optimal spacing between T cell

Significance

Harnessing and enhancing the innate and adaptive immunesystem to fight cancer represents one of the most promisingstrategies in contemporary cancer therapy. Although bispecificantibodies (biAbs) that combine a T cell-engaging arm with atumor cell-binding arm are particularly potent cancer immuno-therapeutic agents, they rely on the identification of tumor an-tigens with highly restricted expression. The receptor tyrosinekinase ROR1 is expressed by numerous cancers and is largelyabsent from postnatal healthy cells and tissues. Here we showthat T cell-engaging biAbs that target ROR1 are highly potent inin vitro, in vivo, and ex vivo models of cancer, in particular whentargeting a conserved site on ROR1 close to the tumor cellmembrane we precisely mapped by X-ray crystallography.

Author contributions: J.Q., E.M.C., A.W., H. Park, and C.R. designed research; J.Q., X.L.,H. Peng, E.M.C., E.L.D., and H. Park performed research; H. Peng and A.W. contributednew reagents/analytic tools; J.Q., E.M.C., E.L.D., A.W., H. Park, and C.R. analyzed data;and J.Q. and C.R. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.

Data deposition: The atomic coordinates and structure factors have been deposited in theProtein Data Bank, www.wwpdb.org (PDB ID code 6BA5).1To whom correspondence should be addressed. Email: crader@scripps.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1719905115/-/DCSupplemental.

Published online May 29, 2018.

www.pnas.org/cgi/doi/10.1073/pnas.1719905115 PNAS | vol. 115 | no. 24 | E5467–E5476

APP

LIED

BIOLO

GICAL

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

and tumor-cell membranes. This, in turn, depends on the spacingbetween the ROR1- and the CD3-engaging arm of the biAb aswell as on the location of its epitopes on ROR1 and CD3. With apanel of rabbit anti-human ROR1 mAbs and an established hu-manized mouse anti-human CD3 mAb at hand, we sought toidentify the best candidate for a heterodimeric and aglycosylatedscFv-Fc format designed for extended circulatory t1/2 and di-minished systemic T cell activation. Rabbit mAb R11, which bindsa conserved membrane proximal epitope on human and mouseROR1 with similar affinity (10), emerged as the most potentROR1-engaging arm. Its precise location and engagement wasdetermined by cocrystallization of scFv R11 with the Kr domain ofhuman ROR1. Collectively, our study encourages and enables thedevelopment of R11-based or R11 epitope-based T cell-engagingbiAbs and other T cell-engaging cancer therapeutic agents.

ResultsDesign, Generation, and Validation of ROR1 × CD3 biAbs. By usingthe established knobs-into-holes technology (21, 22), we designedIgG1-mimicking ROR1 × CD3 biAbs in scFv-Fc format. Thesecomprised (i) the variable light chain (VL) and variable heavychain (VH) domains of a rabbit anti-human ROR1 mAb linkedwith a (Gly4Ser)3 polypeptide linker; (ii) (Gly4Ser)3-linked VH andVL domains of humanized mouse anti-human CD3 mAb v9 (23),which was derived from mouse anti-human CD3 mAb UCHT1(24); and (iii) a human IgG1 Fc module with knobs-into-holesmutations for heterodimerization and with an Asn297Ala muta-tion to render the biAb aglycosylated (25) (Fig. 1A). To confirmpreferential heterodimerization, we combined the scFv-Fc knobscassette-encoding vector with a vector encoding the complemen-tary Fc holes cassette without scFv module. Analysis by SDS/PAGE revealed that >90% of Protein A-purified antibody as-sembled as heterodimer. Based on seven rabbit anti-human ROR1mAbs with diverse epitope specificity and affinity (10, 11) (SIAppendix, Table S1), a series of ROR1 × CD3 biAbs was con-structed that included R12 × v9, R11 × v9, XBR1-402 × v9,ERR1-403 × v9, ERR1-TOP43 × v9, ERR1-306 × v9, and ERR1-

324 × v9. The two polypeptide chains of the heterodimeric scFv-Fcwere encoded on two separate mammalian cell-expression vectorsbased on pCEP4 and combined through transient cotransfectioninto HEK 293 Phoenix cells. Nonreducing and reducing SDS/PAGE revealed the expected ∼100-kDa and ∼50-kDa bands, re-spectively, after purification by Protein A affinity chromatographywith a yield of ∼10 mg/L (Fig. 1B and SI Appendix, Fig. S1A).Further analysis by size-exclusion chromatography (SEC) revealedthat the seven ROR1 × CD3 biAbs eluted mainly as a single peakwith <10% aggregates (Fig. 1C and SI Appendix, Fig. S1B). Se-quential Protein A affinity chromatography and SEC was used topurify monomeric biAbs for the following functional studies.To study the dual binding specificity of ROR1 × CD3 biAbs, a

flow cytometry assay was performed by using CD3-expressinghuman T cell line Jurkat, a stable K562 cell line ectopicallyexpressing ROR1 (K562/ROR1), and MCL cell line JeKo-1,which expresses ROR1 endogenously. As shown in Fig. 2, andconsistent with the pairing of the same CD3-engaging arm withseven different ROR1-engaging arms, all ROR1 × CD3 biAbsrevealed similar binding to Jurkat cells but different binding toK562/ROR1 and JeKo-1 cells by flow cytometry. No binding toROR1-negative parental K562 cells was detected for any of thebiAbs (Fig. 2). To further confirm the dual binding specificity ofthe biAbs, we also cloned, expressed, and purified the corre-sponding eight homodimeric anti-ROR1 and anti-CD3 scFv-Fc.Flow cytometry analysis of R12, R11, XBR1-402, ERR1-403,ERR1-TOP43, ERR1-306, and ERR1-324 scFv-Fc showedbinding to JeKo-1 but not Jurkat cells (SI Appendix, Fig. S2). Bycontrast, the v9 scFv-Fc bound only to Jurkat cells (SI Appendix,Fig. S2). These data suggest that the ROR1 × CD3 biAbsretained the integrity and specificity of the parental mAbs.

In Vitro T Cell Recruitment and Activation Mediated by ROR1 × CD3biAbs. We next examined the functionality of the seven ROR1 ×CD3 biAbs in terms of target-dependent T cell recruitment andactivation in vitro. Primary human T cells were isolated andexpanded from healthy donor peripheral blood mononuclearcells (PBMCs) by using anti-CD3/anti-CD28 beads. The ex vivo-expanded human T cells served as effector cells and JeKo-1,K562/ROR1, and parental K562 cells as target cells. To studythe ability of ROR1 × CD3 biAbs to recruit effector cells totarget cells, we examined their cross-linking. T cells were stainedred and mixed with green-stained K562/ROR1 or K562 cells inthe presence of 1 μg/mL ROR1 ×CD3 biAbs for 1 h, gently washed,and fixed with paraformaldehyde. Double-stained events detectedand quantified by flow cytometry indicated cross-linked cell aggre-gates. As shown in Fig. 3 A and B, all seven ROR1 × CD3 biAbsexhibited highly efficient cross-linking (∼50%) between T cells andK562/ROR1 cells. A negative control scFv-Fc in which the ROR1-engaging arm was replaced with catalytic mAb h38C2 (26), whichdoes not have a natural antigen, revealed only background cross-linking (<10%). Likewise, only background cross-linking was ob-served when using ROR1-negative parental K562 cells (SI Appen-dix, Fig. S3).To analyze the capability of the ROR1 × CD3 biAbs in me-

diating cytotoxicity in vitro, specific lysis of target cells after 16-hincubation with concentrations ranging from 2 pg/mL to 15 μg/mLat an effector-to-target ratio of 10:1 was assessed. K562/ROR1and JeKo-1 cells but not K562 cells were killed in the presence ofall seven ROR1 × CD3 biAbs, indicating that target cell lysis wasstrictly dependent on ROR1 (Fig. 3 C–E). All biAbs had higheractivity against K562/ROR1 cells, which express substantiallyhigher levels of ROR1 compared with JeKo-1 cells (Fig. 2).Negative control biAb, h38C2 × v9, was inactive. One ROR1 ×CD3 biAb, R11 × v9, was significantly and consistently more po-tent than the other ROR1 × CD3 biAbs and killed K562/ROR1and JeKo-1 cells with EC50 values of 22 ng/mL (0.2 nM) and209 ng/mL (2 nM), respectively (Fig. 3 C and D).

VH αCD3 VL αCD3(Gly4Ser)3

CH2 CH3Hinge

VL αROR1 VH αROR1(Gly4Ser)3

CH2 CH3Hinge

aglycosylationholeknob

67.0

69.0

71.0mAU

0.0 5.0 10.0 15.0 mL

10.38

12.37VL VH

VH VL

CH 2

CH 3

CH 2

CH 3

anti-ROR1 anti-CD3

kDa190

11580

70

50

nr r

R12 x v9

A B

C

Fig. 1. Design and generation of bispecific ROR1 × CD3 scFv-Fc. (A) Schematicdiagram of ROR1 × CD3 biAb in scFv-Fc format combining a rabbit anti-humanROR1 scFv with the humanized mouse anti-human CD3 scFv v9 via a mutated Fcdomain of human IgG1. The Fc domain contains one mutation for aglycosylationin the CH2 domain (N297A) and two or four mutations in the CH3 knob (S354Cand T366W) and CH3 hole (Y349C, T366S, L368A, and Y407V) domains, re-spectively, for heterodimerization. (B) SDS/PAGE and Coomassie blue staininganalysis of purified representative R12 × v9 scFv-Fc showing the expected bandsat ∼100 kDa under nonreducing (nr) and ∼50 kDa under reducing (r) conditions.(C) SEC analysis of R12 × v9 scFv-Fc eluting as major peak at 12.37 mL. The10.38-mLminor highermolecular weight peak indicates the level of aggregation.

E5468 | www.pnas.org/cgi/doi/10.1073/pnas.1719905115 Qi et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

We next studied the ROR1 × CD3 biAbs for inducing T cellactivation. The biAbs induced T cell activation only in the pres-ence of K562/ROR1 cells but not K562 cells (Fig. 4). As analyzed

by flow cytometry, the early T cell activation marker CD69 was up-regulated after 16 h incubation with 1 μg/mL ROR1 × CD3 biAbscompared with negative control biAb, h38C2 × v9 (Fig. 4A). As

64

0 101 102 103 104

64

0 101 102 103 104

128

0 101 102 103 104

128

0 101 102 103 104

64

0 101 102 103 104

64

0 101 102 103 104

128

0 101 102 103 104

128

0 101 102 103 104

64

0 101 102 103 104

64

0 101 102 103 104

128

0 101 102 103 104

128

0 101 102 103 104

64

0 101 102 103 104

64

0 101 102 103 104

128

0 101 102 103 104

128

0 101 102 103 104

64

0 101 102 103 104

64

0 101 102 103 104

128

0 101 102 103 104

128

0 101 102 103 104

64

0 101 102 103 104

64

0 101 102 103 104

128

0 101 102 103 104

128

0 101 102 103 104

64

0 101 102 103 104

64

0 101 102 103 104

128

0 101 102 103 104

128

0 101 102 103 104

R12 x v9 R11 x v9 XBR1-402 x v9 ERR1-403 x v9 ERR1-TOP43 x v9 ERR1-306 x v9 ERR1-324 x v9

Fluorescence

Even

ts

Jurkat

K562/ROR1

JeKo-1

K562

Fig. 2. Cell surface binding of bispecific ROR1 × CD3 scFv-Fc. The indicated bispecific ROR1 × CD3 biAbs were analyzed for binding to human CD3-positiveT cell line Jurkat and human ROR1-positive cell lines K562/ROR1 and JeKo-1 by flow cytometry at a concentration of 5 μg/mL followed by APC-conjugated goatanti-human IgG pAbs. Parental K562 is a ROR1-negative control cell line. Open histograms show the background binding signal of the secondary pAbs.

K562/ROR1 cells

T ce

lls

101 102 103 104

101

102

103

104

R12 x v9

101 102 103 104

101

102

103

104

R11 x v9 XBR1-402 x v9

101 102 103 104

101

102

103

104

ERR1-403 x v9

101 102 103 104

101

102

103

104

101 102 103 104

101

102

103

104

101 102 103 104

101

102

103

104

101 102 103 104

101

102

103

104

101 102 103 104

101

102

103

104

ERR1-TOP43 x v9 ERR1-306 x v9 ERR1-324 x v9 h38C2 x v90

25

50

75

100 R12 x v9R11 x v9XBR1-402 x v9ERR1-403 x v9ERR1-TOP43 x v9ERR1-306 x v9ERR1-324 x v9h38C2 x v9

Dou

ble

posi

tive

even

ts (%

)

0.001 0.01 0.1 1 10 1000

20

40

60

Concentration (μμg/mL)

Spec

ific

Lysi

s (%

)

0.001 0.01 0.1 1 10 1000

20

40

60 R12 x v9R11 x v9XBR1-402 x v9ERR1-403 x v9ERR1-TOP43 x v9ERR1-306 x v9ERR1-324 x v9h38C2 x v9

Concentration (μμg/mL)

Spec

ific

Lysi

s (%

)

0.001 0.01 0.1 1 10 1000

10

20

30

40

Concentration (μμg/mL)

Spec

ific

Lysi

s (%

)

K562/ROR1 JeKo-1 K562

A B

C D E

Fig. 3. T cell engagement mediated by bispecific ROR1 × CD3 scFv-Fc. (A) Cross-linking of 5 × 104 primary human T cells (stained with CellTrace Far Red) and5 × 104 K562/ROR1 cells (stained with CellTrace CFSE) in the presence of 1 μg/mL ROR1 × CD3 and control scFv-Fc. Double-stained events were detected by flowcytometry. (B) Quantification of double-stained events from three independent triplicates (mean ± SD). (C) The cytotoxicity of ROR1 × CD3 scFv-Fc was testedwith ex vivo expanded primary human T cells and K562/ROR1 (C), JeKo-1 (D), or K562 (E) cells at an effector-to-target ratio of 10:1 and measured after 16 h.Shown are mean ± SD values from three independent triplicates.

Qi et al. PNAS | vol. 115 | no. 24 | E5469

APP

LIED

BIOLO

GICAL

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

analyzed by ELISA, the release of type 1 cytokines IFN-γ, TNF-α,and IL-2 was also strictly dependent on the presence of ROR1-expressing target cells and ROR1 × CD3 biAbs (Fig. 4 B–D).Whereas all seven ROR1 × CD3 biAbs induced high IFN-γ se-cretion, R11 × v9 induced significantly higher levels of TNF-αand IL-2 secretion.

In Vivo Activity of ROR1 × CD3 biAbs. To investigate whether the invitro T cell recruitment and activation mediated by ROR1 ×CD3 biAbs would translate into in vivo activity, we used a xe-nograft mouse model of systemic human MCL. For this, 0.5 × 106

JeKo-1 cells stably transfected with firefly luciferase (JeKo-1/ffluc)(12) were injected i.v. (tail vein) into five cohorts of eight fe-male NOD-scid-IL2Rγnull (NSG) mice per cohort, and, after1 wk, when the tumor was disseminated, mice were injected i.v.(tail vein) with 5 × 106 ex vivo expanded human T cells. Onehour later, 10 μg of the ROR1 × CD3 biAbs R11 × v9 (cohort2) and R12 × v9 (cohort 3), 10 μg of a positive control CD19 ×CD3 biAb (cohort 4), and 10 μg of a negative control HER2 ×CD3 biAb (cohort 5) were administered i.v. (tail vein). TheCD19 × CD3 biAb shared the same heterodimeric and agly-cosylated scFv-Fc format and the same CD3-binding arm withthe ROR1 × CD3 biAbs. Its CD19-binding arm was derivedfrom human anti-human CD19 mAb 21D4 (Medarex; US Patent8,097,703). Cohort 1 received only vehicle (PBS solution). Allfive cohorts were treated with a total of three doses of T cells(every 8 d) and six doses of biAbs (every 4 d). Mice were pre-conditioned with 250 μL human serum 24 h before every dose.To assess the response to the treatment, in vivo bioluminescenceimaging was performed before the first dose and then every 4 duntil day 39, when the signals were saturated in the control co-horts. Cohorts 1 (no biAb) and 5 (HER2 × CD3 biAb) revealedaggressive tumor growth (Fig. 5). In cohort 2, which receivedR11 × v9, we observed significant tumor growth retardation andtumor eradication starting on day 14 after one dose of T cells andtwo doses of biAb, comparable to the CD19 × CD3 biAb incohort 4 (Fig. 5 A and C). By contrast, R12 × v9 in cohort3 revealed only weak activity compared with the negative control

cohorts. As shown in Fig. 5B, no weight loss was observed duringthe treatment in the responding cohorts, including the R11-basedROR1 × CD3 biAb, which cross-reacts with mouse ROR1 (SIAppendix, Table S1). Weight loss in the nonresponding or weaklyresponding cohorts was noticeable with increasing tumor burden.The corresponding Kaplan–Meier survival curves showed that alltreatment groups except for cohort 5 (HER2 ×CD3 biAb) survivedsignificantly longer than cohort 1 (no biAb). This included bothROR1 × CD3 biAb cohorts (R11 × v9, P < 0.001; R12 × v9; P <0.05) and the CD19 × CD3 biAb cohort (P < 0.001; Fig. 5D). Al-though all mice with this aggressive xenograft eventually exhibitedrelapse, survival was longest in the CD19 × CD3 biAb cohort. Asanticipated from the in vivo bioluminescence imaging, mice treatedwith the R11-based ROR1 × CD3 biAb significantly outlived micetreated with the R12-based ROR1 × CD3 biAb (P < 0.05).We next carried out a pharmacokinetic (PK) study with R11 ×

v9 and R12 × v9 scFv-Fc in mice to examine their circulatory t1/2values. Five female CD-1 mice for each biAb group were injectedi.v. (tail vein) with 6 mg/kg of the biAb. Blood samples werewithdrawn at indicated time points (SI Appendix, Fig. S4) over aperiod of 2 wk, and plasma was prepared. The biAb concentra-tion in the plasma was measured with a sandwich ELISA usingimmobilized ROR1 ECD for capture and peroxidase-conjugatedgoat anti-human IgG pAbs for detection. Analysis of the PKparameters by two-compartment modeling revealed t1/2 values(mean ± SD) of R11 × v9 and R12 × v9, respectively, of 155 ±23 h (6.46 ± 0.96 d) and 152 ± 52 h (6.33 ± 2.17 d; SI Appendix,Table S2). For comparison, the t1/2 of glycosylated and aglyco-sylated [35S]Met-labeled chimeric mouse/human IgG1 in BALB/cmice was determined to be 6.5 ± 0.5 d (27).

Ex Vivo Activity of ROR1 × CD3 biAbs.We next tested the potency ofROR1 × CD3 biAbs against primary malignant B cells from13 patients with untreated CLL. CLL PBMCs with very lowautologous effector-to-target ratios typical for patients with CLL (SIAppendix, Table S3) were incubated with 6.6 nMR11 × v9 and R12 ×v9 scFv-Fc. For comparison, we included equimolar concentrationsof one-armed R11 and R12 scFv-Fc without the CD3-binding arm as

K562

K562/R

OR10

200

4005000

10000

15000

IFN

- γγ c

once

ntra

tion

(pg/

mL)

***

***

*********

******

***

K562

K562/R

OR10

50

100200

250

300

TNF-α α

con

cent

ratio

n (p

g/m

L) ******

***** ***

***

**

***

K562

K562/R

OR1

0

50

100200

250

300 R12 x v9R11 x v9XBR1-402 x v9ERR1-403 x v9ERR1-TOP43 x v9ERR1-306 x v9ERR1-324 x v9h38C2 x v9

IL-2

con

cent

ratio

n (p

g/m

L) ******

*****

K562

K562/R

OR10

10

20

30

40

Activ

ated

T c

ells

(%) ***

***

***

*

******

**

CD69+A B

C D

Fig. 4. T cell activation mediated by bispecific ROR1 × CD3 scFv-Fc. Ex vivo expanded primary human T cells were incubated with 1 μg/mL of the indicatedbiAbs in the presence of K562/ROR1 or K562 cells at an effector-to-target ratio of 10:1 for 16 h. (A) Percentage of activated T cells based on CD69 expression.Cytokine release measured by ELISA for IFN-γ (B), TNF-α (C), and IL-2 (D). Shown are mean ± SD values for independent triplicates. One-way ANOVA was usedto analyze significant differences between ROR1 × CD3 (colored) and control scFv-Fc (black; *P < 0.05, **P < 0.01, and ***P < 0.001).

E5470 | www.pnas.org/cgi/doi/10.1073/pnas.1719905115 Qi et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

well as the negative control HER2 × CD3 biAb. After 11 d, anaverage of 38%, 23%, and 13% CLL PBMCs were killed in thepresence of R11 × v9, R12 × v9 scFv-Fc, and HER2 × CD3 biAb,respectively (Fig. 6A). Averages of 4.9% and 9.3% CLL PBMCswere killed in the presence of the one-armed R11 and R12 scFv-Fc,suggesting that the cytotoxicity is mediated by T cells rather than byother effector cells in the CLL PBMCs. However, high variability inthe response to ROR1 × CD3 biAbs was noted among the 13 CLL

PBMCs, with only R11 × v9, but not R12 × v9, revealing a significantdifference vs. its one-armed counterpart (Fig. 6A). Nonetheless,there was a positive correlation between effector-to-target ratios andcytotoxicity for R11 × v9 and R12 × v9 scFv-Fc but not for thenegative control HER2 × CD3 biAb (Fig. 6B).

In Vitro Activity of ROR1 × CD3 biAbs Toward Carcinoma Cell Lines.To demonstrate the broader therapeutic utility of the ROR1 ×CD3

6 10 14 18 22 26 30 34 39

15

20

25

Day after tumor inoculation

Wei

ght (

g)

Day 6

Day 34

vehicle R11 x v9 R12 x v9 CD19 x CD3

Radiance (x105

p/sec/cm2/sr)

2.06.0 4.0

0.20.40.60.8

Radiance (x108

p/sec/cm2/sr)

HER2 x CD3

6 10 14 18 22 26 30 34 39102

103

104

105

106

107

108

Day after tumor inoculation

Avg

radi

ance

p/s

ec/c

m2 /s

r

****

*********

******

***

***

***

***

***

********* ***

***

0 50 100 1500

50

100

vehicleR11 x v9R12 x v9CD19 x CD3HER2 x CD3

Day after tumor inoculation

Perc

enta

ge s

urvi

val

* *** ***

A

B C D

Fig. 5. In vivo efficacy of bispecific ROR1 × CD3 scFv-Fc. Five cohorts of mice (n = 8) were inoculated with 0.5 × 106 JeKo-1/ffluc cells via i.v. (tail vein) injection.After 7 d, 5 × 106 ex vivo expanded primary human T cells and 10 μg of the indicated biAbs or vehicle alone were administered by the same route. The micereceived a total of three doses of T cells every 8 d and a total of six doses of biAbs or vehicle alone every 4 d. (A) Bioluminescence images of all 40 mice fromday 6 (before treatment) and day 34 (after treatment) are shown. (B) The weight of all 40 mice was recorded on the indicated days (mean ± SD). (C) Startingon day 6, all 40 mice were imaged every 4–5 d and their radiance was recorded (mean ± SD). Significant differences between cohorts treated with biAbs(colored graphs) and vehicle alone (black graph) were calculated by using a two-tailed and heteroscedastic t test (**P < 0.01 and ***P < 0.001). Red arrowsindicate the three T cell doses and black arrows the six biAb or vehicle-alone doses. (D) Corresponding Kaplan–Meier survival curves with P values [log-rank(Mantel–Cox) test] comparing survival between cohorts treated with biAbs (colored graphs) and vehicle alone (black graph; *P < 0.05 and ***P < 0.001).

0.01 0.1 1

-50

0

50

100R11 x v9R12 x v9HER2 x CD3

r=0.6978 p=0.0101

r=0.6209 p=0.0268

r=-0.4895 p=0.1098

E:T Ratio

Spec

ific

Lysi

s (%

)

R11x Fc

R11x v9

R12x Fc

R12 x v9

HER2 x CD3-50

0

50

100

Spec

ific

Lysi

s(%

)

**A B

Fig. 6. Primary CLL cell killing mediated by ROR1 × CD3 scFv-Fc. (A) Percent CLL-specific killing by treatment condition after 11 d in culture calculated by thefollowing formula: (% untreated CLL viability − % treated CLL viability)/(% untreated CLL viability) × 100. Each dot represents one sample and treatmentcondition. Color of data points represent patient-matched samples and correspond with colors in SI Appendix, Table S3. Mean ± SD values are shown. As-terisks indicate statistical significance by Dunn’s multiple comparisons test (**P < 0.01). (B) Spearman’s correlation between initial effector-to-target ratios inPBMC samples used and CLL cell-specific killing by R11 × v9 scFv-Fc (green), R12 × v9 scFv-Fc (blue), and HER2 × CD3 biAbs (red) after 11 d in culture.

Qi et al. PNAS | vol. 115 | no. 24 | E5471

APP

LIED

BIOLO

GICAL

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

biAbs beyond ROR1-expressing hematologic malignancies, weexamined breast adenocarcinoma cell line MDA-MB-231 and renalcell adenocarcinoma cell line 786-O as ROR1-expressing targetcells in the in vitro cytotoxicity assay. Prior analysis by flow cytom-etry confirmed that both carcinoma cell lines are recognized by allseven ROR1 × CD3 biAbs. Among these, the R12-, XBR1-402-,and ERR1-TOP43–based ROR1 × CD3 biAbs, which are thehighest-affinity binders and have overlapping membrane-distalepitopes (SI Appendix, Table S1) (11), revealed the strongestbinding to both carcinoma cell lines (Fig. 7A). By using the samecytotoxicity assay with ex vivo expanded human T cells as before(Fig. 3 C–E), all seven ROR1 × CD3 biAbs revealed dose-dependent killing of MDA-MB-231 and 786-O cells (Fig. 7 Band C). Negative control biAb, h38C2 × v9, was again inactive.Notably, despite its weaker binding, R11 × v9 significantly out-performed all other ROR1 × CD3 biAbs for both MDA-MB-231 and 786-O cells, revealing EC50 values of 100 ng/mL (1 nM)and 120 ng/mL (1.2 nM), respectively (Fig. 7 B and C). Collectively,these data suggest that the membrane-proximal epitope targeted byR11 is a preferred site for T cell-engaging biAbs in scFv-Fc format.

Crystallization of R11 scFv in Complex with the Kr Domain of HumanROR1. To elucidate the precise location of the R11 epitope, scFvR11 was crystallized in complex with the Kr domain of humanROR1 [Protein Data Bank (PBD) ID code 6BA5]. The complexcrystallized as P1 space group with four scFv:Kr complexes in theasymmetric unit (SI Appendix, Table S4). Overall, the rmsd ofatomic positions of each complex was <0.47 Å, indicating littlevariation among the independent structures. The Kr domainrevealed a typical kringle domain folding that appears in various

extracellular proteins (SI Appendix, Fig. S5). Notably, it showedhigher structural homology with the Kr domains of secreted pro-teins compared with the recently reported first 3D structure of aKr domain in the ECD of a transmembrane protein (28) (PDB IDcode 5FWW). For example, the rmsd with human plasminogen Krdomain 4 (PDB ID code 1KRN) is only 0.66 Å. However, theshallow surface pocket known to host a free lysine or lysine analogin many Kr domains (29–31) is constricted in the ROR1 Kr do-main as a result of an inward conformation of loop 5 and partialoccupation by the side chain of Lys369 (SI Appendix, Fig. S5).In the scFv:Kr complex, the N-terminal portion of β-strand A

of the VH domain undergoes domain swapping such that the first6 aa integrate into the BED β-sheet of the neighboring VH do-main (Fig. 8A). The domain swapping results in crystallographicas well as noncrystallographic twofold symmetry between twoneighboring complexes and is likely a crystallization artifact, asthe presence of CH1 and Cκ would block this dimerization. Also,it does not influence the binding of the ROR1 Kr domain, andpurification by SEC before crystallization revealed the mono-meric scFv:Kr complex as major peak.The total buried surface area between scFv R11 and ROR1 Kr

domain is 634 Å2, accounting for 6% and 13% of the total sur-face areas of scFv and Kr domain, respectively. Although theinteraction involves multiple salt bridges, hydrogen bonds, andhydrophobic interactions, it is mediated solely by the VH domain(Fig. 8B). No amino acid residue of the Vκ domain is involved inepitope recognition. Arg332 and Lys382 of the Kr domain formsalt bridges with Asp102 (HCDR3) and Asp30 (HCDR1), re-spectively, of the VH domain, enabling a strong interaction be-tween the antibody and the antigen that is further stabilized by

0.001 0.01 0.1 1 10 1000

20

40

60

Concentration (μμg/mL)

Spec

ific

Lysi

s (%

)

0.001 0.01 0.1 1 10 1000

20

40

60

80 R12 x v9R11 x v9XBR1-402 x v9ERR1-403 x v9ERR1-TOP43 x v9ERR1-306 x v9ERR1-324 x v9h38C2 x v9

Concentration (μμg/mL)

Spec

ific

Lysi

s (%

)

0 101 102 103 104

128

R12 x v9

128

0 101 102 103 104

0 101 102 103 104

ERR1-403 x v9

0 101 102 103 104

128

0 101 102 103 104

ERR1-TOP43 x v9

128

0 101 102 103 104

128

0 101 102 103 104

ERR1-306 x v9

128

0 101 102 103 104

128

0 101 102 103 104

ERR1-324 x v9

128

0 101 102 103 104

R11 x v9

128

0 101 102 103 104

128

0 101 102 103 104

XBR1-402 x v9

128

0 101 102 103 104

128

0 101 102 103 104

MDA-MB-231

786-O128

128

Fluorescence

Even

ts

A

B C

Fig. 7. Bispecific ROR1 × CD3 scFv-Fc targeting breast cancer and renal cell cancer cell lines. (A) The indicated bispecific ROR1 × CD3 biAbs were analyzed forbinding to human ROR1-positive triple-negative breast adenocarcinoma cell lines MDA-MB-231 and renal cell adenocarcinoma cell line 786-O by flowcytometry at a concentration of 5 μg/mL followed by Alexa Fluor 647-conjugated goat anti-human IgG pAbs. Open histograms show the background bindingsignal of the secondary pAbs. The cytotoxicity of ROR1 × CD3 scFv-Fc was tested with ex vivo-expanded primary human T cells and MDA-MB-231 (B) or 786-O(C) cells at an effector-to-target ratio of 10:1 and measured after 16 h. Shown are mean ± SD values from independent triplicates.

E5472 | www.pnas.org/cgi/doi/10.1073/pnas.1719905115 Qi et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

nine hydrogen bonds involving all three CDRs of the VH domain (SIAppendix, Table S5). Furthermore, multiple residues of the CDRsmake van der Waals interactions to establish a favorable shapecomplementarity with the Kr domain (Fig. 8C). These include fa-vorable contacts of Asp27 (HCDR1) and Tyr100 (HCDR3) withLys369 and Arg332, respectively, of the Kr domain, which delimitsthe paratope, and Tyr31 and Pro32 of HCDR1 with Phe381 of theKr domain.The amino acid sequence identities of the Kr domains of human

ROR1 vs. human ROR2 and human ROR1 vs. mouse ROR1 are62% and 99%, respectively (32, 33) (SI Appendix, Fig. S6). FabR11 binds human ROR1 and mouse ROR1 with essentially thesame association rate constant (kon), dissociation rate constant(koff), and Kd (2.7 vs. 3 nM) (10), but it does not cross-react withhuman ROR2. SI Appendix, Fig. S6, shows an amino acid se-quence alignment of the three Kr domains with highlighted resi-dues to indicate the epitope. These residues are highly diversebetween human ROR1 and human ROR2 and completely con-served between human ROR1 and mouse ROR1, explaining thereactivity of mAb R11.

DiscussionHarnessing and enhancing the innate and adaptive immune sys-tem to fight cancer represents one of the most promising strategiesin contemporary cancer therapy. The observation that T cell im-munity plays a critical role in the immune rejection of manycancers is a key premise for cancer immunotherapy (34). A criticalstep of T cell-mediated eradication of tumor cells is the formationof cytolytic synapses (20) between T cells and tumor cells. Thisstep can be mediated by biAbs combining a T cell-binding andactivating arm with a tumor cell-binding arm. Numerous formatsof T cell-engaging biAbs have been described and translated frombasic to preclinical to clinical investigations (18). Thus far, how-ever, only one T cell-engaging biAb, the CD19 × CD3 BiTE

blinatumomab (16), has been approved by the FDA and marketed.Here we describe the generation and characterization of apanel of ROR1 × CD3 biAbs in a heterodimeric and aglyco-sylated scFv-Fc format that confers a circulatory t1/2 of nearly1 wk and eludes systemic T cell activation. Built on these at-tributes, our panel of ROR1 × CD3 biAbs that was based onseven different rabbit anti-human ROR1 mAbs (10, 11) po-tently and selectively killed ROR1-expressing MCL, breastcancer, and kidney cancer cells in the presence of primaryT cells in vitro. A ROR1 × CD3 biAb that was based on a rabbitanti-human ROR1 mAb binding to a membrane-proximalepitope in the Kr domain was significantly more active invitro and dramatically more active in vivo compared withROR1 × CD3 biAbs with membrane-distal epitopes. Moreover,this ROR1 × CD3 biAb with membrane-proximal epitope sig-nificantly mediated the ex vivo killing of primary malignant B cellsby autologous T cells at the very low effector-to-target ratios foundin patients with CLL. Cocrystallization of the scFv in complex withthe Kr domain revealed a discontinuous epitope that is conservedbetween human and mouse ROR1 and located in proximity to thetransmembrane segment.A recent study reported ROR1 × CD3 biAbs in BiTE format

(35). The ROR1-binding arm was derived from two different ratanti-human ROR1 mAbs binding to an epitope in the Ig and Fzdomain, respectively. The CD3-binding arm was derived frommouse anti-human CD3 mAb OKT3 (36), which shares an over-lapping epitope on CD3δe with mouse anti-human CD3 mAbUCHT1 (37) and its humanized variant v9 (23, 38, 39), which weused for our ROR1 × CD3 biAbs in scFv-Fc format. Interestingly,the BiTE targeting the Fz domain was found to be significantlymore potent than the BiTE targeting the Ig domain (35). Al-though a 3D structure of the ROR1 ECD has not been reportedyet, it was assumed that the BiTE epitope in the Fz domain is incloser proximity to the cell membrane than the BiTE epitope in

Ig

Fz

Kr

A B

C

Fig. 8. Crystal structure of scFv R11 in complex with the Kr domain of ROR1. The 3D structure of scFv R11 in complex with the Kr domain of ROR1 wasdetermined by X-ray crystallography at 1.6-Å resolution (PBD ID code 6BA5). (A) Ribbon model of the scFv:Kr complex showing twofold noncrystallographicsymmetry. The scFv is shown in dark blue (VH) and cyan (VL), the Kr domain in orange. The domain swapping β-strand A of the VH domain is shown in green.(B) Cartoon showing the location of the R11 scFv epitope on the Kr domain of ROR1. The dashed area corresponds to the crystal structure of the scFv:Krcomplex shown on the right. Note that only the VH domain (dark blue) interacts with the Kr domain. (C) Interaction of the three CDRs of the VH domain (darkblue) with the Kr domain (orange surface model). The two enlarged areas show the salt bridges formed by HCDR1 (Asp30-Lys382) and HCDR3 (Asp102-Arg332), respectively. All interacting amino acid residues are listed in SI Appendix, Table S5.

Qi et al. PNAS | vol. 115 | no. 24 | E5473

APP

LIED

BIOLO

GICAL

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

the Ig domain, and that this closer proximity may augment theformation of cytolytic synapses between T cells and tumor cells.An increased potency of T cell-engaging biAbs that bind tomembrane-proximal epitopes on tumor cell antigens has also beenreported for BiTE-based MCSP × CD3 in melanoma (40) and forIgG-based FcRH5 × CD3 biAbs in multiple myeloma (41). Byusing confocal microscopy, the latter study provided evidence thatbiAbs with membrane-proximal epitopes augment the formationof cytolytic synapses by promoting target clustering and exclusionof CD45 phosphatase from the cell–cell interphase, which triggersefficient ZAP70 translocation. It was shown that this processreplicates the TCR-MHC/peptide-driven formation of cytolyticsynapses (41).Epitope location on ROR1 is also critical for the activity of

CAR-T cells. An R12-based CAR with membrane-distal epitopewas found to be more potent, with a shorter spacer between scFvand transmembrane segment, whereas a longer spacer was crit-ical for the activity of an R11-based CAR with membrane-proximal epitope (12). These findings were confirmed with anXBR1-402–based CAR, which shares an overlapping epitopewith R12, and an XBR2-401–based CAR, which binds to the Krdomain of human ROR2 (11). Collectively, the screening of apanel of mAbs with diverse epitopes is critical for identifyingsuitable candidates for T cell engagement by biAbs and CAR-T cells. Alternatively, if such panels are not available, custom-izing biAb and CAR formats may be required to incite activity.By determining the crystal structure of the R11 scFv in complexwith the Kr domain of ROR1 at 1.6-Å resolution, we preciselymapped the epitope to a discontinuous epitope located near theC-terminus of the Kr domain, 20 aa upstream of the predictedtransmembrane segment in the primary structure of ROR1 (32).Notably, the crystal structure revealed that only VH of scFvR11 makes contacts, covering ∼13% of the surface area of the Krdomain. In addition to single-domain antibodies (42), our studyencourages the development of R11 VH-mimicking non-Igscaffolds (43), such as affibodies (44) and designed ankyrin re-peat proteins (45), to explore alternative building blocks forROR1 × CD3 engagement.Despite being smaller, the scFv-Fc format (∼100 kDa) mimics

the natural IgG1 format (∼150 kDa) in using neonatal Fc re-ceptor (FcRn)-mediated recycling for extended circulatory t1/2. Arecent study measured circulatory t1/2 values in mice of ∼200 hfor a chimeric mouse/human IgG1, ∼100 h for a chimeric mouse/human scFv-Fc, and ∼0.5 h for a mouse scFv, with all threesharing identical mouse VL and VH amino acid sequences (46).In approximate agreement, our PK study revealed a circulatoryt1/2 of ∼150 h for the ROR1 × CD3 scFv-Fc. This prolongedpresence in the blood permitted a twice-weekly dosing in our invivo experiment that likely can be further reduced to onceweekly. By contrast, BiTEs, which are ∼50 kDa (scFv)2 withoutFc domain and have a circulatory t1/2 of ∼2 h in humans (47),require continuous administration via, e.g., a portable minipumpand i.v. port catheter. Although, in the case of blinatumomab, thissetting has the advantage that treatment can be interrupted toaddress adverse events, particularly cytokine release syndrome andneurologic events (48), an increasing number of biAb formats inpreclinical and clinical studies include an Fc domain to avoidcontinuous i.v. infusion (18). We chose an aglycosylated Fc do-main to retain FcRn-mediated recycling but elude binding to Fcγreceptors CD16, CD32, and CD64 (49). Although this designexcludes antibody-dependent cellular cytotoxicity and antibody-dependent cellular phagocytosis as contributing effector func-tions, it ensures that T cells are not activated systemically by Fcγreceptor-bound scFv-Fc that cross-link CD3 (50). The use of anaglycosylated Fc did not impact expression yields in mammaliancells, purification yields using standard Protein A affinity chro-matography, monodispersity, or solubility of the scFv-Fc, affirmingits suitability for manufacturing.

Despite limited insights into its role in cancer biology, the re-ceptor tyrosine kinase ROR1 is an attractive target for antibody-mediated cancer therapy because of its highly restricted expressionin a range of hematologic and solid malignancies (9). Postnatalhealthy cells and tissues are largely negative for cell-surface ROR1despite some exceptions (6–8, 51). Although solid malignancieshave remained challenging indications for antibody therapeuticagents in general, targeting ROR1 in hematologic malignanciessuch as CLL, MCL, and subsets of acute lymphoblastic leukemiacould circumvent concomitant depletion of healthy B cells by theFDA-approved CD19-targeting antibody therapeutic agents. Twoearly clinical trials are currently investigating ROR1 as a target formAb and CAR-T cell therapy, respectively. ROR1 × CD3 biAbs,which combine off-the-shelf availability of mAbs with T cell-engaging potency of CAR-T cells, could provide a best-of-both-worlds option for patients with ROR1-expressing cancers. In viewof its potency in vitro and in vivo, its recent humanization (52),and the precise mapping of its paratope and epitope by X-raycrystallography, mAb R11 has emerged as a prime candidate forfurther preclinical and clinical studies of ROR1 × CD3 biAbs.

Materials and MethodsCell Lines and Primary Cells. The K562, JeKo-1, and Jurkat cell lines wereobtained from the American Type Culture Collection (ATCC) and cultured inRPMI-1640 supplemented with L-glutamine, 100 U/mL penicillin/streptomy-cin, and 10% (vol/vol) FCS (all from Thermo Fisher Scientific). The K562/ROR1 and JeKo-1/ffluc cell lines (6, 12) were provided by Stanley R. Riddell(The Fred Hutchinson Cancer Research Center, Seattle, WA) and MichaelHudecek (University of Würzburg, Würzburg, Germany), respectively. HEK293 Phoenix (ATCC), MDA-MB-231 (ATCC), and 786-O cells (an NCI-60 panelcell line obtained from The Scripps Research Institute’s Cell-Based High-Throughput Screening Core) were grown in DMEM supplemented with L-glutamine, 100U/mL penicillin/streptomycin, and 10% (vol/vol) FCS. PBMCswere isolated from healthy donor buffy coats by using Lymphoprep (Axis-Shield) and cultured in X-VIVO 20 medium (Lonza) with 5% (vol/vol) off-the-clot human serum AB (Gemini Bio-Products) and 100 U/mL IL-2 (Cell Sci-ences). Primary T cells were expanded from PBMCs as previously described(31) by using Dynabeads ClinExVivo CD3/CD28 (Thermo Fisher Scientific).Cryopreserved treatment-naïve CLL PBMCs were derived from a naturalhistory study at the National Institutes of Health (NIH) Clinical Center reg-istered under ID code NCT00923507 at https://clinicaltrials.gov/. The studywas approved by the institutional review boards of the NIH Clinical Centerand The Scripps Research Institute. Informed consent was obtained in ac-cordance with the Declaration of Helsinki. Clinical and laboratory data werecollected from electronic medical records.

Cloning, Expression, and Purification of ROR1 × CD3 biAbs. All amino acidsequences have been previously published or patented. The variable domainencoding cDNA sequences of the rabbit anti-human ROR1 mAbs and thehumanized mouse anti-human CD3 mAb v9 were PCR-amplified fromphagemids (10, 11) and previously described DART-encoding plasmids (53),respectively. A (Gly4Ser)3 linker encoding sequence was used to fuse VL andVH by overlap-extension PCR. The Fc fragment with a N297A mutation andeither knob mutations, S354C and T366W, or hole mutations, Y349C, T366S,L368A, and Y407V, were synthesized as gBlocks (Integrated DNA Technol-ogies). Anti–ROR1-Fc knob and anti–CD3-Fc hole-encoding fragments wereassembled by overlap-extension PCR, respectively, and AscI/XhoI-cloned intomammalian cell expression vector pCEP4 and transiently cotransfected intoHEK 293 Phoenix cells (ATCC) with polyethylenimine (Sigma-Aldrich) as de-scribed (54). Supernatants were collected three times over a 9-d period,followed by 1-mL Protein A HiTrap HP columns in conjunction with an ÄKTAFPLC instrument (both from GE Healthcare Life Sciences). Subsequent pre-parative and analytic (10 μg biAb) SEC was performed with a Superdex 20010/300 GL column (GE Healthcare Life Sciences) in conjunction with an ÄKTAFPLC instrument using PBS solution at a flow rate of 0.5 mL/min. The purityof the biAbs was confirmed by SDS/PAGE followed by Coomassie bluestaining, and their concentration was determined by measuring the absor-bance at 280 nm. One-armed R11 and R12 scFv-Fc were cloned, expressed,and purified as described for the ROR1 × CD3 biAbs by replacing the anti–CD3-Fc hole with an empty Fc hole.

E5474 | www.pnas.org/cgi/doi/10.1073/pnas.1719905115 Qi et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

Flow Cytometry. Flow cytometry was performed on a FACSCanto system (BDBiosciences), and datawere analyzedwithWinMDI 2.9 software. APC or AlexaFluor 647-conjugated goat anti-human IgG pAbs were purchased fromJackson ImmunoResearch Laboratories. Alexa Fluor 647-conjugated mouseanti-human CD69 mAb was purchased from BioLegend. For the cross-linkingassay, target and effector cells were labeledwith CellTrace CFSE and CellTraceFar Red (both from Thermo Fisher Scientific), respectively, according to themanufacturer’s protocol. The labeled target and effector cells at a 1:1 ratiowere then incubated with 1 μg/mL biAbs in 100 μL final volume. Followingincubation for 1 h at room temperature, the cells were gently washed, fixedwith 1% (wt/vol) paraformaldehyde, and analyzed by flow cytometry asdescribed earlier.

In Vitro Cytotoxicity Assay. Cytotoxicity was measured by using CytoTox-Glo(Promega) following the manufacturer’s protocol with minor modifica-tions. Primary T cells expanded from healthy donor PBMCs as describedhere earlier were used as effector cells and K562/ROR1, JeKo-1, K562,MDA-MB-231, or 786-O cells were used as target cells. The cells were in-cubated at an effector-to-target ratio of 10:1 in X-VIVO 20 medium (Lonza)with 5% (vol/vol) off-the-clot human serum AB. The target cells (1 × 104)were first incubated with the biAbs before adding the effector cells (1 ×105) in a final volume of 100 μL per well in a 96-well tissue culture plate.The plates were incubated for 16 h at 37 °C with biAb concentrationsranging from 2 pg/mL to 15 μg/mL. After centrifugation, 50 μL of thesupernatant was transferred into a 96-well clear-bottom white-walledplate (Costar 3610; Corning) containing 25 μL per well CytoTox-Glo. Af-ter 15 min at room temperature, the plate was read by using a Spec-traMax M5 instrument with SoftMax Pro software. The samesupernatants (diluted 20-fold) used for the CytoTox-Glo assay were alsoused to determine IFN-γ, TNF-α, and IL-2 secretion with the human IFN-γ,human TNF-α, and human IL-2 ELISA Ready-SET-Go! reagent sets (eBioscience),respectively, following the manufacturer’s protocols.

Mouse Xenograft Studies. Forty 6-wk old NSG mice (The Jackson Labora-tory) were each given 0.5 × 106 JeKo-1/ffluc cells via i.v. (tail vein) injectionon day 0. On day 7, each mouse was i.v. injected (tail vein) with 5 × 106

primary T cells expanded from healthy donor PBMCs as described earlierand, 1 h later, with 10 μg biAbs or PBS solution alone. The mice received atotal of three doses of expanded primary T cells every 8 d and a total ofsix doses of biAbs or PBS solution alone every 4 d. Every 4–5 d, tumorgrowth was monitored by bioluminescent imaging 5 min after i.p. injec-tions with 150 mg/kg D-luciferin (Biosynth). For this, mice were anes-thetized with isoflurane and imaged by using an Xenogen IVIS imagingsystem (Caliper) 6, 8, 10, 12, and 14 min after luciferin injection in smallbinning mode at an acquisition time of 10 s to 1 min to obtain un-saturated images. Luciferase activity was analyzed by using Living ImageSoftware (Caliper), and the photon flux was analyzed within regions ofinterest that encompassed the entire body of each individual mouse. Theweight of the mice was measured every 4–5 d. All procedures were ap-proved by the institutional animal care and use committee of The ScrippsResearch Institute and were performed according to the NIH Guide forthe Care and Use of Laboratory Animals.

PK Study. Five female CD-1 mice (∼25 g; Charles River Laboratories) wereinjected i.v. (tail vein) with R11 × v9 or R12 × v9 at 6 mg/kg. Blood wascollected at 5 min and 12, 24, 48, 72, 108, 168, 240, and 336 h aftetr in-jection with heparinized capillary tubes. Plasma was obtained by centri-fuging the samples at 2,000 × g for 5 min in a microcentrifuge and storedat −80 °C until analysis. The concentrations of biAbs in the plasma sampleswere measured by ELISA. For this, each well of a 96-well Costar 3690 plate(Corning) was incubated with 200 ng recombinant human ROR1 ECD (10)in 25 μL carbonate/bicarbonate buffer (pH 9.6) at 37 °C for 1 h. Afterblocking with 150 μL 3% (wt/vol) BSA/PBS solution for 1 h at 37 °C, theplasma samples were added. Peroxidase AffiniPure F(ab′)2 Fragment GoatAnti-Human IgG pAbs (Jackson ImmunoResearch Laboratories) were usedfor detection. The concentration of the biAbs in the plasma samples wasextrapolated from a four-variable fit of the standard curve. PK parameterswere analyzed by using Phoenix WinNonlin PK/PD Modeling and Analysissoftware (Pharsight).

Ex Vivo Cytotoxicity Assay. Cryopreserved PBMCs were thawed and plated at3 × 106 cells per milliliter in 24-well plates in RPMI 1640 (Gibco), 10% FBS, 1%penicillin/streptomycin, 50 μM β-mercaptoethanol (Sigma-Aldrich). A total of60 IU/mL of IL-2 was added to cultures 6–12 h after initial plating. Cells werethen incubated with bispecific or one-armed scFv-Fc at 6.6 nM and harvested

on day 11. Untreated wells containing cells and medium alone were left asbackground controls. Cell viability was assessed with LSRFortessa (BD Bio-sciences) by using LIVE/DEAD fixable violet stain (Thermo Fisher Scientific).To analyze the cell viability, CLL cells were identified as CD8−/CD4−/CD5+/CD20+

and T cells as CD8+ or CD4+ by staining with commercial mAbs (BD Biosci-ences): CD4 (RPA-T4), CD8 (HIT8a), CD20 (L27), and CD5 (L17F12). Specifickilling was calculated by the following formula: (% untreated CLL viability −% treated CLL viability)/(% untreated CLL viability) × 100. Effector T cell totarget CLL cell ratios were determined based on frequencies of live cells.

Statistical Analyses. Statistical analyses were performed by using Prismsoftware (GraphPad). The in vitro data shown in Fig. 4 were subjected to one-way ANOVA, and the in vivo data shown in Fig. 5C were analyzed with atwo-tailed and heteroscedastic t test with a CI of 95%. Statistical analysis ofsurvival (Fig. 5D) was done by log-rank (Mantel–Cox) testing. The ex vivodata shown in Fig. 6A were analyzed with a Dunn’s multiple comparisonstest. Results with a P value <0.05 were considered significant.

Crystallization and Structure Determination.Cloning, expression, and purification. Leaderless R11 scFv and human ROR1 Krdomain (amino acids 319–391) encoding cDNA sequences were PCR-amplified from previously described phagemids and plasmids (10) andcloned as polycistronic assembly that also included a leaderless Escherichiacoli disulfide bond isomerase C (DsbC), which is a chaperone that aids properdisulfide bond formation (55), into E. coli expression vector pET-15b(Novagen). The resulting expression cassette included a ribosome bindingsite (RBS) with a start codon, an N-terminal (His)6 tag, a thrombin cleavagesite, the Kr domain (flanked by NdeI/XhoI), a second RBS with a start codonfollowed by the scFv (flanked by XhoI/XhoI), and a third RBS with a startcodon followed by DsbC (flanked by XhoI/BamHI). The prepared plasmid wastransformed into E. coli strain Rosetta-Gami 2(DE3) (Novagen). Followinginduction of protein expression by 0.3 mM isopropyl-D-thiogalactoside, thebacterial cultures were grown in Luria-Bertani (LB) medium containing am-picillin, tetracycline, and chloramphenicol at 20 °C and 220 rpm in anincubator shaker for 18 h.Protein purification. The proteins were purified by immobilized metal ionchromatography followed by SEC in conjunction with an ÄKTA FPLC in-strument. In brief, bacterial pellets were harvested by centrifugation andresuspended in sonication buffer [20 mM Hepes, pH 8.0, 300 mM NaCl, 15 mMimidazole, 10% (vol/vol) glycerol], sonicated in an ice-water bath, and centri-fuged for 25 min at 53,300 × g. The supernatant was loaded on a custom-packed10-mL HIS-Select column (Sigma-Aldrich) and washed with sonication buffer. ThescFv:Kr complex was eluted with a linear gradient of 15–500 mM imidazole andtreated overnight at 4 °C with thrombin (Sigma-Aldrich) to remove the N-terminal (His)6 tag on the scFv. The clipped scFv:Kr complex was then purifiedfurther by SEC on a Superdex 200 26/60 column (GE Healthcare Life Sciences)equilibrated with 50 mM NaCl, 10 mM Hepes (pH 7.4). Fractions containing thecomplex were combined and concentrated to ∼15 mg/mL with a 10-kDa mo-lecular weight cut-off Amicon ultrafiltration unit (Millipore).Crystallization and structure determination. Crystals of the purified scFv:Krcomplex were grown by vapor diffusion at room temperature by using 1.5 μLof 15 mg/mL protein and an equal volume of precipitant containing 0.2 Mammonium fluoride in 20% (wt/vol) PEG 3350, and were fully grown within2 d. The crystals were flash-frozen in liquid nitrogen by using nylon loopsafter removing excess mother liquor. A diffraction data set with Braggspacings to 1.6 Å was collected on an ADSC Quantum 315r detector at the24-ID-E beamline at the Advanced Photon Source, Argonne National Labo-ratory. Data were processed with iMOSFLM software (56). The structure ofthe scFv:Kr complex was solved by the molecular replacement method byusing BALBES with 2ZNW as the search model (57). Crystallographic re-finement was performed by using a combination of PHENIX 1.12 (58) andBUSTER 2.9 (59). Manual rebuilding and adjustment of the structure werecarried out by using the graphics program Coot (60). Data processing andrefinement statistics are shown in SI Appendix, Table S4. Molecular figureswere created by using PyMOL software (Schrödinger). Buried surface areawas calculated using PISA (61), and structure validation was carried outwith MolProbity (62). The crystal structure was deposited in PDB under IDcode 6BA5.

ACKNOWLEDGMENTS. We thank Drs. Stanley R. Riddell and MichaelHudecek for cell lines and Li Lin and Dr. Michael D. Cameron for theirhelp with analyzing the PK study. This study was funded by NIH GrantsR01 CA181258 and UL1 TR001114 and by donations from the PGA Na-tional Women’s Cancer Awareness Days, Peter and Janice Brock, and theAnbinder Family Foundation. This is manuscript 29607 from The ScrippsResearch Institute.

Qi et al. PNAS | vol. 115 | no. 24 | E5475

APP

LIED

BIOLO

GICAL

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020

1. Baskar S, et al. (2008) Unique cell surface expression of receptor tyrosine kinaseROR1 in human B-cell chronic lymphocytic leukemia. Clin Cancer Res 14:396–404.

2. Fukuda T, et al. (2008) Antisera induced by infusions of autologous Ad-CD154-leukemia B cells identify ROR1 as an oncofetal antigen and receptor for Wnt5a.Proc Natl Acad Sci USA 105:3047–3052.

3. Daneshmanesh AH, et al. (2008) Ror1, a cell surface receptor tyrosine kinase is ex-pressed in chronic lymphocytic leukemia and may serve as a putative target fortherapy. Int J Cancer 123:1190–1195.

4. Cui B, et al. (2016) High-level ROR1 associates with accelerated disease progression inchronic lymphocytic leukemia. Blood 128:2931–2940.

5. Barna G, et al. (2011) ROR1 expression is not a unique marker of CLL. Hematol Oncol29:17–21.

6. Hudecek M, et al. (2010) The B-cell tumor-associated antigen ROR1 can be targetedwith T cells modified to express a ROR1-specific chimeric antigen receptor. Blood 116:4532–4541.

7. Baskar S, Wiestner A, Wilson WH, Pastan I, Rader C (2012) Targeting malignant B cellswith an immunotoxin against ROR1. MAbs 4:349–361.

8. Balakrishnan A, et al. (2017) Analysis of ROR1 protein expression in human cancer andnormal tissues. Clin Cancer Res 23:3061–3071.

9. Hojjat-Farsangi M, et al. (2014) The receptor tyrosine kinase ROR1–An oncofetal an-tigen for targeted cancer therapy. Semin Cancer Biol 29:21–31.

10. Yang J, et al. (2011) Therapeutic potential and challenges of targeting receptor ty-rosine kinase ROR1 with monoclonal antibodies in B-cell malignancies. PLoS One 6:e21018.

11. Peng H, et al. (2017) Mining naive rabbit antibody repertoires by phage display formonoclonal antibodies of therapeutic utility. J Mol Biol 429:2954–2973.

12. Hudecek M, et al. (2013) Receptor affinity and extracellular domain modificationsaffect tumor recognition by ROR1-specific chimeric antigen receptor T cells. ClinCancer Res 19:3153–3164.

13. Berger C, et al. (2015) Safety of targeting ROR1 in primates with chimeric antigenreceptor-modified T cells. Cancer Immunol Res 3:206–216.

14. Fesnak AD, June CH, Levine BL (2016) Engineered T cells: The promise and challengesof cancer immunotherapy. Nat Rev Cancer 16:566–581.

15. Sadelain M, Rivière I, Riddell S (2017) Therapeutic T cell engineering. Nature 545:423–431.

16. Kantarjian H, et al. (2017) Blinatumomab versus chemotherapy for advanced acutelymphoblastic leukemia. N Engl J Med 376:836–847.

17. Frankel SR, Baeuerle PA (2013) Targeting T cells to tumor cells using bispecific anti-bodies. Curr Opin Chem Biol 17:385–392.

18. Brinkmann U, Kontermann RE (2017) The making of bispecific antibodies. MAbs 9:182–212.

19. Kroschinsky F, et al.; Intensive Care in Hematological and Oncological Patients(iCHOP) Collaborative Group (2017) New drugs, new toxicities: Severe side effects ofmodern targeted and immunotherapy of cancer and their management. Crit Care 21:89.

20. de la Roche M, Asano Y, Griffiths GM (2016) Origins of the cytolytic synapse. Nat RevImmunol 16:421–432.

21. Merchant AM, et al. (1998) An efficient route to human bispecific IgG. Nat Biotechnol16:677–681.

22. Atwell S, Ridgway JB, Wells JA, Carter P (1997) Stable heterodimers from remodelingthe domain interface of a homodimer using a phage display library. J Mol Biol 270:26–35.

23. Zhu Z, Lewis GD, Carter P (1995) Engineering high affinity humanized anti-p185HER2/anti-CD3 bispecific F(ab’)2 for efficient lysis of p185HER2 overexpressing tumor cells.Int J Cancer 62:319–324.

24. Shalaby MR, et al. (1992) Development of humanized bispecific antibodies reactivewith cytotoxic lymphocytes and tumor cells overexpressing the HER2 protooncogene.J Exp Med 175:217–225.

25. Hristodorov D, et al. (2013) Generation and comparative characterization of glyco-sylated and aglycosylated human IgG1 antibodies. Mol Biotechnol 53:326–335.

26. Rader C, et al. (2003) A humanized aldolase antibody for selective chemotherapy andadaptor immunotherapy. J Mol Biol 332:889–899.

27. Tao MH, Morrison SL (1989) Studies of aglycosylated chimeric mouse-human IgG. Roleof carbohydrate in the structure and effector functions mediated by the human IgGconstant region. J Immunol 143:2595–2601.

28. Zebisch M, Jackson VA, Zhao Y, Jones EY (2016) Structure of the dual-mode Wntregulator Kremen1 and insight into ternary complex formation with LRP6 and Dick-kopf. Structure 24:1599–1605.

29. Hochschwender SM, Laursen RA (1981) The lysine binding sites of human plasmino-gen. Evidence for a critical tryptophan in the binding site of kringle 4. J Biol Chem256:11172–11176.

30. Mulichak AM, Tulinsky A, Ravichandran KG (1991) Crystal and molecular structure ofhuman plasminogen kringle 4 refined at 1.9-A resolution. Biochemistry 30:10576–10588.

31. Ye Q, Rahman MN, Koschinsky ML, Jia Z (2001) High-resolution crystal structure ofapolipoprotein(a) kringle IV type 7: Insights into ligand binding. Protein Sci 10:1124–1129.

32. Masiakowski P, Carroll RD (1992) A novel family of cell surface receptors with tyrosinekinase-like domain. J Biol Chem 267:26181–26190.

33. Oishi I, et al. (1999) Spatio-temporally regulated expression of receptor tyrosine ki-nases, mRor1, mRor2, during mouse development: Implications in development andfunction of the nervous system. Genes Cells 4:41–56.

34. Blattman JN, Greenberg PD (2004) Cancer immunotherapy: A treatment for themasses. Science 305:200–205.

35. Gohil SH, et al. (2017) An ROR1 bi-specific T-cell engager provides effective targetingand cytotoxicity against a range of solid tumors. OncoImmunology 6:e1326437.

36. Kjer-Nielsen L, et al. (2004) Crystal structure of the human T cell receptor CD3 epsilongamma heterodimer complexed to the therapeutic mAb OKT3. Proc Natl Acad SciUSA 101:7675–7680.

37. Arnett KL, Harrison SC, Wiley DC (2004) Crystal structure of a human CD3-epsilon/delta dimer in complex with a UCHT1 single-chain antibody fragment. Proc Natl AcadSci USA 101:16268–16273.

38. Rodrigues ML, Shalaby MR, Werther W, Presta L, Carter P (1992) Engineering a hu-manized bispecific F(ab’)2 fragment for improved binding to T cells. Int J Cancer Suppl7:45–50.

39. Zhu Z, Carter P (1995) Identification of heavy chain residues in a humanized anti-CD3 antibody important for efficient antigen binding and T cell activation. J Immunol155:1903–1910.

40. Bluemel C, et al. (2010) Epitope distance to the target cell membrane and antigen sizedetermine the potency of T cell-mediated lysis by BiTE antibodies specific for a largemelanoma surface antigen. Cancer Immunol Immunother 59:1197–1209.

41. Li J, et al. (2017) Membrane-proximal epitope facilitates efficient T cell synapse for-mation by anti-FcRH5/CD3 and is a requirement for myeloma cell killing. Cancer Cell31:383–395.

42. Könning D, et al. (2017) Camelid and shark single domain antibodies: Structuralfeatures and therapeutic potential. Curr Opin Struct Biol 45:10–16.

43. �Skrlec K, �Strukelj B, Berlec A (2015) Non-immunoglobulin scaffolds: A focus on theirtargets. Trends Biotechnol 33:408–418.

44. Ståhl S, et al. (2017) Affibody molecules in biotechnological and medical applications.Trends Biotechnol 35:691–712.

45. Plückthun A (2015) Designed ankyrin repeat proteins (DARPins): Binding proteins forresearch, diagnostics, and therapy. Annu Rev Pharmacol Toxicol 55:489–511.

46. Unverdorben F, et al. (2016) Pharmacokinetic properties of IgG and various Fc fusionproteins in mice. MAbs 8:120–128.

47. Zhu M, et al. (2016) Blinatumomab, a bispecific T-cell engager (BiTE) for CD19 targetedcancer immunotherapy: Clinical pharmacology and its implications. Clin Pharmacokinet55:1271–1288.

48. Viardot A, et al. (2016) Phase 2 study of the bispecific T-cell engager (BiTE) antibodyblinatumomab in relapsed/refractory diffuse large B-cell lymphoma. Blood 127:1410–1416.

49. Saxena A, Wu D (2016) Advances in therapeutic Fc engineering–Modulation of IgG-associated effector functions and serum half-life. Front Immunol 7:580.

50. Weiner GJ, et al. (1994) The role of T cell activation in anti-CD3 x antitumor bispecificantibody therapy. J Immunol 152:2385–2392.

51. Dave H, et al. (2012) Restricted cell surface expression of receptor tyrosine kinaseROR1 in pediatric B-lineage acute lymphoblastic leukemia suggests targetability withtherapeutic monoclonal antibodies. PLoS One 7:e52655.

52. Waldmeier L, et al. (2016) Transpo-mAb display: Transposition-mediated B cell displayand functional screening of full-length IgG antibody libraries. MAbs 8:726–740.

53. Walseng E, et al. (2016) Chemically programmed bispecific antibodies in diabodyformat. J Biol Chem 291:19661–19673.

54. Sarkar M, et al. (2016) Targeting stereotyped B cell receptors from chronic lympho-cytic leukemia patients with synthetic antigen surrogates. J Biol Chem 291:7558–7570.

55. Park H, Adsit FG, Boyington JC (2005) The 1.4 angstrom crystal structure of the humanoxidized low density lipoprotein receptor lox-1. J Biol Chem 280:13593–13599.

56. Battye TG, Kontogiannis L, Johnson O, Powell HR, Leslie AG (2011) iMOSFLM: A newgraphical interface for diffraction-image processing with MOSFLM. Acta Crystallogr DBiol Crystallogr 67:271–281.

57. Long F, Vagin AA, Young P, Murshudov GN (2008) BALBES: A molecular-replacementpipeline. Acta Crystallogr D Biol Crystallogr 64:125–132.

58. Adams PD, et al. (2010) PHENIX: A comprehensive Python-based system for macro-molecular structure solution. Acta Crystallogr D Biol Crystallogr 66:213–221.

59. Bricogne G, et al. (2010) BUSTER(Global Phasing, Cambridge, UK), Version 2.9.60. Emsley P, Cowtan K (2004) Coot: Model-building tools for molecular graphics. Acta

Crystallogr D Biol Crystallogr 60:2126–2132.61. Krissinel E, Henrick K (2007) Inference of macromolecular assemblies from crystalline

state. J Mol Biol 372:774–797.62. Chen VB, et al. (2010) MolProbity: All-atom structure validation for macromolecular

crystallography. Acta Crystallogr D Biol Crystallogr 66:12–21.

E5476 | www.pnas.org/cgi/doi/10.1073/pnas.1719905115 Qi et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

1, 2

020