Both anti-CD30 and anti-CD137 antibodies were isolated by in vitro biopanning from a human Fab phage display library that has a diversity of 30 billion clones (16). The experimental procedures for selection, phage preparation, Fab expression and purification closely followed the protocols described by De Haard et al. (17). Target proteins (R&D Systems, Minneapolis, MN, USA) were biotinylated using the EZ-Link NHS-PEG4-Biotin labeling kit (Pierce, Thermo Fisher Scientific, Waltham, USA). Biotinylated targets were first immobilized on M280 streptavidin-coated magnetic beads (Life Technologies Carlsbad, CA, USA). In the first round of biopanning, 10¹³ cfu phage were mixed with the bead-immobilized target in 1 ml casein-PBS blocking buffer; in the second and third rounds of biopanning, 10¹¹ cfu phage were used in 0.5 ml blocking buffer. The phage library/target were incubated for 1 h at room temperature and the beads were washed with the PBS buffer containing 0.1% Tween-20. The concentrations of target proteins used in the biopanning step were 100, 20, and 5 nM, and the number of washes were 5, 10, and 25 times in rounds 1, 2, and 3, respectively. Bound phage were recovered by incubating the washed beads in 0.1 M triethylamine (pH 11) for 10 min followed by neutralization with 1 M Tris-HCl, pH 8. The recovered phages upon each round of biopanning were amplified. Following three rounds of biopanning, the selected Fab clones were expressed in Escherichia coli TG1 cells (Stratagene, Agilent Technologies, Santa Clara, California, USA) to screen for CD30 or CD137 binders by ELISA. Target binding unique clones were identified by DNA fingerprinting technique (17) and confirmed by DNA sequencing. To express full length IgG, the sequences of variable regions were cloned into human IgG1 backbone in the pTT5 vector. The anti-CD30/anti-CD137 bispecific antibodies were generated in CrossMab format (18), in which the variable region sequences of the heavy chain of anti-CD30 and anti-CD137 antibodies were cloned in the knob arm and hole arm, respectively. All full length antibodies were produced in HEK293-6E cells (19). Both the cells and the vector were obtained from National Research Council of Canada. Antibodies were purified from the culture supernatant using Protein G resin (Merck Millipore) following standard protocols.

The HRS cell lines L-428, L-1236, KM-H2, and HDLM-2 were purchased from the German Collection of Microorganisms and Cell Cultures (DSMZ, Braunschweig, Germany) (20), and were cultured in RPMI 1640 (Sigma, St. Louis, MO, USA) supplemented with 10 or 20% fetal bovine serum (Biowest, Kansas City, MO, USA) at 37°C with 5% CO₂. Stable cell lines expressing CD137 were generated by lentiviral transduction (11).

KM-H2 mutant cells were generated by deletion of CD30 or CD137 or both by transduction with lentiviruses expressing the sgRNA sequences 5′GTCGGTGACAGAACCCGTCG and 5′GCGCTGGAGAAACTATTTGG for CD30 and CD137, respectively.

Buffy coats from healthy donors were obtained from the National University Hospital, Singapore. Human PBMCs were isolated by density gradient centrifugation using Ficoll-Paque Plus (GE Healthcare, Chalfont St. Giles, UK). The protocol was approved by the National University of Singapore (NUS) IRB number B15-320E.

NK cells were isolated from PBMC by negative selection using EasySep™ Human NK Cell Enrichment Kit (Stemcell Technologies, Vancouver, Canada). 10⁸ PBMC were resuspended with Stem cell separation buffer (2% fetal bovine serum and 1 mM EDTA) in PBS. PBMC were incubated with NK enrichment antibody cocktail for 10 min, followed by incubation with Magnetic D beads in a polystyrene tube for 5 min. The tube was placed in EasySep magnet for 2.5 min and unlabeled cells were poured out from the tube. Enriched NK cells were resuspended in RPMI with 10% FBS and Pen-Strep.

The ADCC assay was carried out using the Delfia EuTDA cytotoxicity kit (PerkinElmer, Waltham, MA, USA) according to manufacturer's instruction. In brief, target HRS cells (L-428-control or L-428-CD137 or L-1236-control or L-1236-CD137) were washed with RPMI with 10% FBS (R10) once and loaded with DELFIA BATDA reagent at 1 μl per 10⁶ cells in R10 supplemented with 2 mM probenecid and 50 μM β-mercaptoethanol (R10 PS PM media) for 20 min at 37°C. Labeled target cells were washed 4 times in R10 PS PM media, and were seeded into “V”-bottom 96-well plates at 10⁴ cells per well together with 5 μg/ml of bispecific or control antibodies for 40 min at 4°C. Effector cells (isolated NK cells) were added to the target cells at an E:T ratio of 20:1 and incubated for 3 h at 37°C. The Europium substrate solution was added to the supernatants for 15 min before the fluorescence was measured using a time-resolved fluorometer (VICTOR³ multi-label reader, PerkinElmer).

KM-H2 cells were treated with 10 μg/ml of control HA4 or bispecific antibodies on ice. Cells were washed twice with ice-cold PBS and treated with 5 μg/ml of anti-human Fc-Biotin for 10 min on ice to crosslink the antibodies on cell surface. Cells were washed and incubated at 37°C with media for several time points. Cells were stained with streptavidin-APC and analyzed by Fortessa Analyzer (BD Biosciences, New Jersey, USA).

Internalization was further analyzed by treating KM-H2 cells with 50 μM of monodansylcadaverine (MDC, Sigma-Aldrich). The presence of the bispecific antibodies was analyzed 24 h later by flow cytometry.

Five micrometer tissue microarray slides containing 48 cases of HL (HL481) were purchased from US Biomax, Inc. (Derwood, MD, USA). Opal multiplex immunofluorescent system (opal staining) was adopted for multi-color staining (Perkin Elmer, Waltham, USA). Sections were incubated with the primary antibodies mouse anti-human CD137 (clone BBK-2, Thermo Fisher Scientific), rabbit anti-human CD3 (Dako), and rabbit anti-human CD30 (clone EPR4102, Abcam), followed by the secondary antibodies anti-mouse and anti-rabbit horseradish peroxidase (HRP) polymers (GBI Labs, Bothell, USA), and then developed with the 4-color Opal IHC kit (Perkin Elmer). Nuclei were stained with DAPI. Unstained sections of HL served as controls to collect autofluorescence signature for image unmixing. Images were acquired with the Vectra imaging system (Perkin Elmer) and analyzed with the inForm® advanced image analysis software (Perkin Elmer) at the Genome Institute of Singapore, A^*STAR.

Cells were stained with PE-conjugated anti-CD137 antibody (clone 4B4-1, BD Biosciences) or FITC-conjugated anti-CD30 (clone BY88, Biolegend, San Diego, CA, USA) or mouse-IgG1k-PE (eBioscience, San Diego, USA) at 4°C in the dark for 15 min. Flow cytometry was performed with Fortessa flow cytometry (BD Biosciences) and analyzed with Flow Jo software (FlowJo, LLC, Ashland, USA). The rate of apoptosis was determined by Annexin V and 7-AAD staining (Biolegend).

Statistical significance was determined by a two-tailed unpaired Student's t-test.

CD137 is expressed by HRS cells of 86% of classical HL tumor samples (10, 11). To test if HRS cells co-express CD30 and CD137, we performed multiplex immunoflourescence staining (Figures 1A,B). There is a clear co-localization of CD30 (white) and CD137 (green) on giant cells, confirming co-expression of the two molecules on HRS cells. The reason why the CD30 staining looks different from the CD137 stainings is that CD30 is mainly on the cell surface while most of CD137 is cytoplasmic. This is due to the fact that CD137 yields a complex with CD137 ligand which then gets internalized (11). T cells (red) present in the tumor stroma do not stain positive for CD137, an activation marker, indicating that the cells are not activated (21).

In addition, for the HRS-derived cell lines KM-H2 and HDLM-2 cells around 84 and 87%, respectively, of cells were CD30⁺CD137⁺ (Figure 1C). These data clearly demonstrate a co-expression of CD30 and CD137 on HRS cells.

Both, CD30 and CD137, can be expressed on activated lymphocytes. In order to evaluate whether a simultaneous targeting of CD30 and CD137 would also affect activated lymphocytes, we determined whether CD30 and CD137 are co-expressed on these cells.

A small minority of resting PBMC stained positive for CD30 while no CD137 could be detected in unstimulated PBMC. Activation of PBMC by immunocult, PMA + ionophore, PMA + ionomycin, PHA, or anti-CD3 induced expression of CD30 or CD137 to various degrees but the cells expressed either CD30 or CD137 but not both together. Only in PHA-activated PBMC there was a small percentage of CD30⁺CD137⁺ cells (Figure 2). This finding confirms an earlier report that noticed a mutual exclusion of CD30 and CD137 expression on T cells (22), implying that co-expression of CD30 and CD137 is rather specific for HRS cells, and that targeting the two molecules would not affect activated lymphocytes.

We generated 4 antigen binding Fab for anti-CD30 and anti-CD137 through phage display. Binding efficiency of the anti-CD30 and anti-CD137 Fabs were tested by dose response ELISA. The HL cell line L-428 homogenously expresses CD30 on its surface (Supplementary Figure 1). Hence, L-428 cells were treated with 10 μg/ml of anti-CD30 Fabs (clones 30A, 30B, and 30C) washed and probed with mouse anti-myc and anti-mouse-AF488. All Fabs efficiently bound to CD30 on the surface of L-428 cells. Fabs 30A and 30C showed a strong binding to CD30 with geometric mean fluorescence (GMF) of around 2,000 or more, while clone 30B had a moderate binding with GMF of around 1,000 (Supplementary Figure 2). We selected 30A and 30B for the anti-CD30 arm of the bispecific antibody as they had the least binding affinities thereby reducing the likelihood of the bispecific antibody binding to CD30-single positive cells.

The anti-CD137 Fabs 137A, 137B, 137C, and 137D were tested for their efficiency to bind to CD137-expressing HRS cell lines. CD137-transfected L-428 cells (L-428-CD137) were treated with 10 μg/ml of Fabs, followed by anti-Fab-AF647. All Fabs bound efficiently to CD137⁺ cells (Supplementary Figure 3). No binding was observed to CD137⁻ L-428-control cells (not shown). 137A and 137C showed weaker binding affinity than the other Fabs, making them suitable for the generation of the bispecific antibody.

As CD137 signaling in HRS cells induces IL-13 release and thereby favors immune evasion, we aimed to select an antagonistic anti-CD137 arm for the generation of the bispecific antibodies. Hence, the Fabs were converted to full anti-CD137 antibodies and tested in a functional assay for co-stimulating T cells. T cells were seeded on wells coated with 5 μg/ml of anti-CD3 (clone OKT3) and 10 μg/ml of anti-CD137 antibodies. Anti-CD137 clone 137D increased secretion of IFN-γ suggesting it having an agonistic activity. The commercially available agonistic anti-CD137 antibody (clone JG1.6a) and antagonistic anti-CD137 antibody (clone BBK-2) have been used as controls. Treatment with 137A and 137B did not increase IFN-γ levels (Figure 3). The anti-CD137 Fab (137A) was selected for anti-CD137 arm of bispecific antibody, because it lacks agonistic activity and because of its moderate binding affinity to CD137-single positive cells.

A bispecific antibody in CrossMab format (18) was generated with 137A as the anti-CD137 arm and 30B as the anti-CD30 arm, and its binding specificity was tested on HL cell lines, being single positive or double positive or double negative for CD30 and CD137 (Supplementary Figure 4). Although, this antibody preferentially bound to double-positive (CD30⁺CD137⁺) L-428-CD137 cells, it still retained binding to single positive cells (not shown). Therefore, this bispecific antibody was re-engineered by site directed mutagenesis to reduce the affinity of each arm. To this end, each tyrosine or tryptophan residue in the heavy and light chain CDR3 loops of these two antibodies was individually replaced with an alanine to create 6 single mutation bispecific antibody variants.

The binding specificity of 6 re-engineered clones was tested by flow cytometry. None of the clones bound double negative cells (white bars), and all bound to varying degrees single CD30- or CD137-positive cells (patterned bars). Clone 2 and 5 displayed highest binding to double-positive cells (black bars) (Figure 4). Affinities were measured by surface plasmon resonance. Clone 2 has a 9.3 nM and 16.3 nM affinity to CD30 and CD137, respectively; clone 5 has a 10.0 nM and 6.2 nM affinity to CD30 and CD137, respectively. These affinities are a few fold lower that that original 137A for CD137 (1.2 nM) and 30B for CD30 (2.7 nM). These two affinity-reduced bispecific antibodies were selected for functional analysis.

The FDA approved anti-CD30-drug conjugate exerts its therapeutic activity via being internalized and delivering the toxin auristatin E (23). We therefore tested if the bispecific antibodies can be internalized into HL cells, and whether that is dependent on CD30 and CD137. Binding of mutant 2 and 5 to the KM-H2 cells was almost abolished in the absence of either CD30 or CD137 as seen by the reduced MFI (Figure 5). In vivo, Fc receptors will crosslink the cell-bound antibodies and accelerate internalization of antibody. To simulate this in vitro, anti-Fc-Biotin was used to crosslink the antibodies. Internalization was measured by the reduction in the surface staining as a function of time. Both antibodies were internalized in WT cells. There was some residual internalization when only CD137 was present, i.e., in CD30^−/− cells, but internalization was absent in CD30, CD137 double negative cells (Figures 5A,B), demonstrating that these antibodies may be suitable as toxin carriers and specific to CD30, CD137 double positive HRS cells.

To further demonstrate internalization, we made use of Monodansylcadaverine (MDC), an inhibitor of internalization. Treatment of KM-H2 cells with 50 μM MDC increased the level of both bispecific antibodies on the surface of KM-H2 cells, demonstrating that internalization indeed takes place (Figure 5C). Internalization is likely mediated by the Fcγ receptor CD32a, since KM-H2 cells express CD32a while CD16 and CD64 could not be detected (Supplementary Figure 4).

An alternative to killing HRS cells by an antibody drug conjugate is via immune effector functions such as ADCC. Therefore, clone 2 and 5 were tested for their efficiency to kill HL cells by ADCC using primary NK cells as effector cells. Spontaneous lysis of all 4 HL cell lines was normalized to 1. Treatment with clone 2 or 5 led to a 4-fold increase in the killing of double positive (CD30⁺CD137⁺) cells (black vs. white bars). Killing of single positive cells by clone 2 and 5 was close to spontaneous lysis levels (patterned bars compared to white bars) (Figure 6). This demonstrates the selectivity of the bispecific antibody to double positive (CD30⁺CD137⁺) cells.

While drug-conjugated antibodies or antibodies that exert ADCC activity are useful anti-cancer agents, they require the use of toxins or the presence of NK cells, respectively. The ideal situation for a therapeutic antibody is to induce cell death directly upon binding to its target. While clone 5 showed no activity, clone 2 induced cell death in KM-H2 cells, increasing the rates of early and late apoptosis from 5.6 to 21.9% and from 3.9 to 8.7%, respectively, compared to the isotype control antibody (Figure 7). While clone 2 induced cell death in KM-H2 cells, we could not detect any change in cytokine levels (not shown).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.