Biparatopic binding of ISB 1442 to CD38 in trans enables increased cell antibody density and increased avidity

ABSTRACT

ISB 1442 is a bispecific biparatopic antibody in clinical development to treat hematological malignancies. It consists of two adjacent anti-CD38 arms targeting non-overlapping epitopes that preferentially drive binding to tumor cells and a low-affinity anti-CD47 arm to enable avidity-induced blocking of proximal CD47 receptors. We previously reported the pharmacology of ISB 1442, designed to reestablish synthetic immunity in CD38+ hematological malignancies. Here, we describe the discovery, optimization and characterization of the ISB 1442 antigen binding fragment (Fab) arms, their assembly to 2 + 1 format, and present the high-resolution co-crystal structures of the two anti-CD38 Fabs, in complex with CD38. This, with biophysical and functional assays, elucidated the underlying mechanism of action of ISB 1442. In solution phase, ISB 1442 forms a 2:2 complex with CD38 as determined by size-exclusion chromatography with multi-angle light scattering and electron microscopy. The predicted antibody-antigen stoichiometries at different CD38 surface densities were experimentally validated by surface plasmon resonance and cell binding assays. The specific design and structural features of ISB 1442 enable: 1) enhanced trans binding to adjacent CD38 molecules to increase Fc density at the cancer cell surface; 2) prevention of avid cis binding to monomeric CD38 to minimize blockade by soluble shed CD38; and 3) greater binding avidity, with a slower off-rate at high CD38 density, for increased specificity. The superior CD38 targeting of ISB 1442, at both high and low receptor densities, by its biparatopic design, will enhance proximal CD47 blockade and thus counteract a major tumor escape mechanism in multiple myeloma patients.

Introduction

The landscape of cancer therapy has undergone a transformative shift with the advent of new antibody formats such as bi- and multispecific antibodies.^1–5 This class of engineered molecules is designed to simultaneously engage several targets, offering novel functionalities for treating cancers that cannot be mediated by conventional monoclonal antibodies (mAbs). Beyond immune cell retargeting, bispecific (bsAbs) also have the potential to outperform mAbs’ efficacy and overcome drug resistance by taking advantage of the avidity and synergy by targeting different antigens or antigenic epitopes.^6–8 Biparatopic antibodies, co-engaging two distinct epitopes on the same target, and modulating antigen-binding capabilities have been described to enhance antibody efficacy.⁹ The mechanisms of action of biparatopic antibodies include enzyme inhibition, receptor clustering and down-regulation by internalization or increased drug conjugate uptake, increased affinity, inverse agonism, clearance of free target, prevention of mutational escape or enhanced effector Fc functions such as complement-dependent cytotoxicity (CDC), antibody-dependent cell-mediated cytotoxicity (ADCC), and antibody-dependent cellular phagocytosis (ADCP).^9–11 There are now at least 27 antibodies, alternative scaffold proteins, antibody-drug conjugates (ADCs) or CAR-T, all of which utilize biparatopic targeting of antigens, including HER2, Met, CXCR4, amyloid-beta, BCMA, CD37, CD38, GPRC5D, FR-alpha, Factor XI, HIV or SARS-Cov2,⁹ in clinical development.

CD38 is an ADP-ribosyl cyclase and cyclic ADP-ribose hydrolase, expressed on the cell surface, that catalyzes the first step in the conversion of NAD+, released from damaged cells, to immunosuppressive adenosine in the tumor microenvironment (TME).¹² CD38 is a promising target for an antibody-based therapeutic as it is expressed in several hematologic malignancies, including multiple myeloma (MM), acute myeloid leukemia (AML) and diffuse large B cell lymphoma (DLBCL).^13,14 This has led to the development of daratumumab (DARZALEX®) and isatuximab (SARCLISA®), two anti-CD38 mAbs approved for the treatment of MM.^15,16 Although targeting CD38 has demonstrated clinical efficacy, several resistance mechanisms, including CD38 down-regulation, upregulation of complement inhibitory receptors, downregulation of FcγRIIIa and CD47 upregulation, have been described^17,18 and are proposed to result in reduced clinical activity and relapses from anti-CD38 therapy.

CD47 is an immunoglobulin superfamily transmembrane protein whose ligands include signal-regulatory protein (SIRP), thrombospondin and integrins. Blockade of CD47 by antagonistic antibodies or recombinant SIRPα proteins, resulting in inhibition of CD47-SIRPα “don’t eat me” signaling and subsequently enhanced phagocytosis, has demonstrated anti-tumor activity.^19,20 However, because of the ubiquitous expression of CD47 on various tissues, and particularly, high expression on red blood cells (RBCs) and platelets, it is difficult to selectively target tumor cells with anti-CD47 mAbs. Thus, such antibodies may exhibit poor pharmacokinetic properties due to target-mediated drug disposition and serious side effects (i.e., anemia).²¹ To address this limitation, bsAbs with reduced affinity for CD47 and high affinity to a tumor-associated antigen (TAA), to avoid direct targeting of CD47, have been successfully constructed and proven to be active preclinically.^22–25

We previously described the engineering and in vitro and in vivo functional activity of ISB 1442, a CD47 × CD38 bispecific biparatopic antibody (BsBpAb), currently in a Phase 1 clinical trial in relapsed refractory multiple myeloma.²⁶ ISB 1442 was designed in a 2 + 1 BsAb format with one low affinity anti-CD47 antigen binding fragment (Fab) arm, two high affinity, biparatopic anti-CD38 Fab arms and with enhanced Fc effector functions. This enables highly efficient targeting of CD38-expressing cancer cells, avidity-enabled CD47 blockade and enhanced CDC, ADCC and ADCP.²⁶

Here, we describe the antibody discovery and optimization activities that enabled ISB 1442 and present atomic-level detail of the engagement of CD38 by the anti-CD38 dual Fab arm and the consequences of biparatopic binding for ISB 1442’s mechanism of action. As previously described, ISB 1442 was assembled by exploiting the BEAT® technology.^26–28 The co-crystal structures of the two anti-CD38 Fabs in complex with CD38 show that both common light chain (cLC) Fabs bind unique epitopes. Using physicochemical characterization, cryo-electron microscopy (EM), and structural modeling of biparatopic interactions with CD38, we demonstrate that the molecular architecture of ISB 1442 and the mode of binding provide a distinct advantage compared to monoparatopic 1:1 antibody/antigen binding. Finally, the trans biparatopic binding by ISB 1442 enabled increased avidity to cells with higher CD38 surface densities, and thus improved specific targeting of tumor relative to normal CD38+ cells, with a potential of broadening of the therapeutic window. Multiple myeloma patients eventually relapse when treated with monospecific anti-CD38 antibody therapeutics such as daratumumab¹⁷ due to either CD38 downregulation or upregulation of CD47. The unique design of ISB 1442, with biparatopic targeting of CD38, can counteract these escape mechanisms by several modes of action. Firstly, enabling increased Fc density on cancer cells that express low levels of CD38, leading to enhanced killing where the target is down-regulated. Secondly the trans-binding of ISB 1442 to CD38 on cancer cells leads to both an increased load of therapeutic on the cancer cell together with reduced off rate, which leads to more efficient CD47 co-targeting. Therefore ISB 1442 could eventually offer an alternative treatment option to MM patients.

Results

Design and construction of a human synthetic common light chain antibody phage display library

ISB 1442 utilizes cLC Fab arms^29,30 to enable correct light chain pairing and to generate a multispecific IgG format for superior developability and manufacturability.³¹ For this purpose, a synthetic cLC antibody phage display library using a fixed Vκ3–15/Jκ1 light chain was generated. The selection of this cLC framework was initially motivated by its compatibility with OmniFlic^Ⓡ transgenic rat technology (OmniAb^Ⓡ), using the same fixed light chain, that could be used as a complementary approach for generating cLC antibodies.³² Four germline sequences (V_H1–69, V_H3–15, V_H3–23 and V_H3–53³³ were selected as variable heavy chain framework based on their structural diversity, favorable developability profile, and representation in the human antibody repertoire when paired with Vκ3–15 LC.³⁴ Briefly, optimized natural position-specific diversity was introduced in heavy chain complementarity-determining regions (HCDRs). HCDR1 and HCDR2 randomization is based on germline specific diversity and HCDR3 mimics natural length distribution encompassing length 6 to 20 amino acids (according to Kabat nomenclature) (Figure 1a). Diversified single-chain variable fragments (scFv) were transformed into E. coli to achieve a total library size of 5.6 × 10¹⁰. Next-generation sequencing (NGS) was used to confirm good library quality with 77% of in-frame sequences and diversity matching the original design.

Figure 1.

Identification of anti-CD47 and anti-CD38 common light chain fabs for the generation of ISB 1442. a Design of common light chain scFv phage display library. Library framework is based on 4 VH germlines, VH1–69, VH3–15, VH3–23 and VH3–53, and a fixed Vκ3–15/Jκ1 light chain. HCDRs were randomized using trinucleotide primers mimicking natural position-specific diversity and HCDR3 length distribution. b overview of the antibody discovery workflow for the generation of ISB 1442. c epitope binning results showing plot of saturating antibody in rows against competing antibodies in columns. Values indicate percentage of binding of competing antibody to CD38 relative to maximum binding in the absence of saturating antibody. Values less than 30 were assigned as competing antibodies and shaded in black, values between 30 and 50 were assigned as partially competing antibodies and shaded in light gray, while values greater than 50 were assigned as non-competing antibodies. Self-blocks are outlined by a dark-gray box. NT = not tested. Cross-competition results between: E2 and B6; E2 and daratumumab; B6 and daratumumab are outlined in blue, yellow and red, respectively. d-e. Fitted sensorgrams of a representative measurement show the binding of E2 and E2-RecA fabs (d), and B6 and B6-D9 fabs (e) at increasing concentrations onto immobilized human CD38. Colored curves represent experimental data and black curves represent 1:1 kinetic fits.

Identification of V_H domains targeting CD47 and multiple epitopes on CD38

To generate diverse anti-CD47 and anti-CD38 Fab arms and find the optimal combination of antigen-binding moieties to achieve potent ISB 1442, the phage display library was panned against both targets and antigen-specific scFv were reformatted to Fabs for characterization (Figure 1b). As described by Grandclément et al.,²⁶ a low-affinity anti-CD47 Fab blocking the CD47-SIRPα interaction was selected for avidity-induced blocking of CD47 upon CD38 engagement. To enable the identification of Fab arms, which bind to non-overlapping binding sites on CD38 and do not compete with the standard of care daratumumab, a low stringency phage display panning was applied to maximize epitope coverage. Fab epitope binning revealed that E2, B6 and daratumumab Fabs recognize distinct epitopes (Figure 1c and reference 26). E2 and B6 Fab have affinities (K_D) to human CD38 of 110 nM and 225 nM, respectively, as measured by surface plasmon resonance (SPR) (Figure 1d–e, Table 1). To enable strong anchorage of ISB 1442 to cancer cells through CD38, an affinity maturation of both E2 and B6 Fab was carried out. Phage display libraries were generated by randomizing HCDR1, HCDR2 or HCDR3 of both lead candidates and panning was performed in high stringency conditions to select high affinity antibodies. ScFv, expressed from phage display outputs, were screened for binding to CD38 by SPR and sequence-unique clones with a slower dissociation rate (off-rate) than their respective parental clones were reformatted and expressed as Fabs. Affinity measurement of E2 variants by SPR revealed that E2-A10 and E2-D6, originated from HCDR1 and HCDR2 libraries, with 12- and 22-fold affinity increase to human CD38, compared to the parental sequence, respectively. In addition, the combination of E2-A10 HCDR1 and E2-D6 HCDR2 led to the generation of E2-RecA with affinities of 1.6 nM and 2.7 nM to human and cynomolgus monkey CD38, corresponding to 69-fold and 33-fold affinity increases compared to E2, respectively (Figure 1d, Tables 1 and 2). Affinity maturation of B6 was successfully completed by selecting B6-D9 with affinities of 0.6 nM and 1.4 nM to human and cynomolgus monkey CD38, respectively (Figure 1e, Tables 1 and 2). Here, a unique set of HCDR1 substitutions led to 375-fold and 592-fold affinity increases to the human and cynomolgus monkey CD38, respectively (Table 2). To generate ISB 1442, optimized anti-CD38 Fab arms B6-D9 and E2-RecA were linked together through a flexible glycine-serine peptide linker of 15 amino acids (G₄S)₃ and fused to the BEAT^Ⓡ B chain²⁸ and the anti-CD47 Fab arm was linked to the BEAT^Ⓡ A chain (Figure 1b). The Fc also included the Fc receptor-enhancing mutations (S239D, I332E), on the BEAT^Ⓡ B chain, to increase tumor cell killing via ADCC and ADCP^35–37 and the S324N mutation on both the A and B chains to enhance CDC.³⁸

Table 1.

Binding kinetic parameters of anti-CD38 E2 and B6 parental and affinity matured variants fabs to human and cynomolgus monkey CD38 measured by surface plasmon resonance. Affinities were assessed by capturing CD38 and using fab fragments as analyte. k_a = association constant, k_d = dissociation constant, K_D = equilibrium constant. Representative single replicate is shown.

Lineage	Library	Clone name	Human CD38 ka (1e-05, 1/Ms)	Human CD38 kd (1e + 02, 1/s)	Human CD38 KD (nM)	Cynomolgus monkey CD38 ka (1e-05, 1/Ms)	Cynomolgus monkey CD38 kd (1e + 02, 1/s)	Cynomolgus monkey CD38 KD (nM)
E2	Parental	E2	5.5	5.8	110	14	13	89
	HCDR1	E2-A10	37	3.5	9.4	68	5.8	8.6
	HCDR2	E2-D6	2.2	0.10	4.9	8.5	0.10	1.1
	Combination	E2-RecA	5.2	0.09	1.6	5.1	0.14	2.7
B6	Parental	B6	2.4	5.9	225	1.6	13	829
	HCDR1	B6-D9	15	0.10	0.6	7.7	0.11	1.4
Benchmark		Daratumumab	3.5	0.21	6.1	No binding

Table 2.

Sequences of anti-CD38 E2 and B6 parental and affinity matured variants HCDRs. Residues that differ from the parental sequences are highlighted in underlined bold.

Lineage	Library	Clone name	HCDR1 (kabat 26–35)	HCDR2 (kabat 50–58)	HCDR3 (kabat 93–102)
E2	Parental	E2	GGSFSNYAIG	RIIPVFGSAH	ARGLGYYLYSSYYFDI
	HCDR1	E2-A10	GLPDAT YAIG	RIIPVFGSAH	ARGLGYYLYSSYYFDI
	HCDR2	E2-D6	GGSFSNYAIG	RIIPRLDAEH	ARGLGYYLYSSYYFDI
	Combination	E2-RecA	GLPDATYAIG	RIIPRLDAEH	ARGLGYYLYSSYYFDI
B6	Parental	B6	GGHISNSAIN	RVIPVIDDAY	ARSRGYYGSFYFDI
	HCDR1	B6-D9	GGFFSHYIIN	RVIPVIDDAY	ARSRGYYGSFYFDI

Co-crystal structure of the B6-D9/CD38 and E2-RecA/CD38 complexes

To elucidate the mode of binding of the anti-CD38 biparatopic dual Fab arms and to obtain atomic-level resolution of the epitope/paratope interactions, we determined the crystal structures of the B6-D9 and E2-RecA Fabs in complex with the human CD38 extracellular domain (ECD) at 1.8 Å and 2.7 Å resolution, respectively (Figure 2).

Figure 2.

Crystal structure of human CD38 in complex with the anti-CD38 B6-D9 and anti-CD38 E2-RecA fabs. a Overall structure of CD38 in complex with the B6-D9 fab from ISB 1442 BEAT® as determined from the crystal structures of CD38/B6-D9 (PDB 9GOX). hsCD38 (uniprot ID P28907) as a space filling representation is shaded in white with B6-D9 epitope colored in light blue, ribbon diagrams of the fab heavy chain in blue and fab common light chain (cLC) in gray. b footprint of B6-D9 fab on human CD38 highlighting CD38 epitope residues interacting with B6-D9. c surface representation of the B6-D9 epitopes on CD38 with side chains of interacting paratope residues from the fab shown and colored by CDR loop. d structure representation of CD38 in complex with the E2-RecA fab as determined from the crystal structures of CD38/E2-RecA (PDB 9GOY). hsCD38 is shaded in white with E2-RecA epitope colored in light cyan, fab heavy chain in cyan and fab cLC in gray. e footprint of E2-RecA fab on human CD38 epitope with E2-RecA interacting residues numbered. f surface representation of the E2-RecA epitope on CD38 with side chains of interacting paratope residues from E2-RecA shown and colored by CDR loop. HCDR1, blue; HCDR2, orange; HCDR3, pink; LCDR3, grey.

The two Fabs bind CD38 on diametrically opposed regions. B6-D9 recognizes an epitope in the membrane distal region of CD38, opposite the catalytic pocket,³⁹ whereas E2-RecA binds an epitope close to the membrane, at the bottom lateral side of CD38 (Figure 2a–d). The buried surface area (BSA) between the Fab/Antigen interfaces is 1,304 Ų for B6-D9 and 1,297 Ų for E2-RecA, which is very low compared with other antibody/antigen interactions,⁴⁰ including daratumumab (PDB 7DHA; 2,274 Ų and isatuximab (PDB 4CMH, 2,492 Ų. ^16,41 For both Fabs, the recognition of CD38 is dominated by the heavy chain, with 86% and 75% of the interface that arises from the heavy chain for B6-D9 and E2-RecA, respectively. Unsurprisingly, the contribution of the heavy chain for these two cLC Fabs is higher than reported for natural antibodies. For example, in a study of 1425 antibody-antigen complexes, an average of 67% of the paratope residues belonged to the heavy chain.⁴²

The crystal structure reveals that B6-D9 recognizes an epitope of 14 residues using the three HCDR loops and LCDR3 (Figure 2b,c). There are two salt bridges (Arg50_HCDR2 to Glu233 and Asp56_HCDR2 to Arg269), five hydrogen bonds (His31_HCDR1 to Asn270, Arg50_HCDR2 to Gln231, Tyr58_HCDR2 to Lys234, Gly97_HCDR3 to Asn270 and Y99_HCDR3 to Asp202), two water-mediated hydrogen bonds (Asn35_HCDR2 and Trp96_LCDR3 to Glu233), and several hydrophobic interactions. HCDRs via Ile33_HCDR1, Ile52_HCDR2, Val53_HCDR2, Ile54_HCDR2, Tyr99_HCDR3, Gly100_HCDR3 and Phe100B_HCDR3 make hydrophobic contacts with the epitope, including, Pro232, Val235, Ile265, as well as the aliphatic portions of residues Glu198 and Gln231 and Lys268. To study the energetic stabilization provided by the putative hydrogen bonds, we ran molecular dynamic (MD) simulations and monitored the key epitope-paratope intermolecular bonds remaining at the interface between the B6-D9 Fab and the antigen (Supplementary Fig. S1 and S2). This analysis confirmed the energetic stabilization provided by the hydrogen bonding pattern observed in the crystal structure, and this suggests that the experimentally observed hydrogen bonding network reflects the situation dynamically, in addition to its presence in the static picture presented by crystal structures. Arg50_HCDR2 forms very stable H-bonding interactions simultaneously with the Gln231 and Glu233 residues of CD38. Interestingly, the water-mediated hydrogen bonding involving HCDR1 Asn35 and LCDR3 Trp96 was observed in about half of the simulation frames, suggesting a dynamic equilibrium with the water shell. Lastly, to identify the binding hotspot,⁴³ we conducted molecular mechanics generalized Born surface area (MM-GBSA) free energy decomposition analysis using the snapshots collected during MD simulations.^44–46 The hotspot residues that strongly interact with CD38 were labeled and shown in Supplementary Fig. S1 and Supplementary Table S1. In agreement with the analysis of the hydrogen network, the residues His31_HCDR1, Arg50_HCDR2, Tyr58_HCDR2 and Tyr99_HCDR3 residues, with a free energy per residue exceeding −4.6 kcal/mol, were identified as the main contributors to the interaction with CD38. Another set of neighboring residues from all three HCDRs, namely Ser30_HCDR1, Ile52 _HCDR2, Ile54_HCDR2, Tyr98_HCDR3, as well as LCDR3 Trp94 and Trp96, demonstrate significant values of stabilizing energies in the range between −4.6 and −1.0 kcal/mol (Supplementary Fig. S1a, c, and e, Supplementary Table S1).

E2-RecA binds the membrane-proximal part of CD38, recognizing an epitope of 15 residues (Figure 2e,f and Supplementary Fig. S1b, d and f). The relative orientation of the chains suggests that E2-RecA Fab approaches CD38 in parallel to the cellular membrane (Figure 2d). In contrast to B6-D9, HCDR1 does not contribute directly to the interaction with CD38. The residues within the HCDR2 and HCDR3 bind CD38 via hydrogen bonds (Arg50_HCDR2 and His58_HCDR2 to Glu72, Tyr99_HCDR3 to His79 and Tyr100A_HCDR3 to the backbone carbonyl oxygen of Val80), a salt bridge (Arg50_HCDR2 to Glu72) and hydrophobic interactions. HCDR2, via Ile52, Leu54 and Ala56 make hydrophobic contacts with the CD38 residues Leu64, Ala65 and Val68. Three aromatic residues at the tip of HCDR3, Tyr99, Tyr100A and Tyr100D interact with a hydrophobic patch on CD38 as well as with surrounding residues, including Val68, Arg78, His79, Val80, Asp81 and Cys82 (Figure 2e,f). The cLC also contributes to the hydrophobic interactions where LCDR3 Trp94 and Trp96 interact with Ile73 and the aliphatic portion of Glu72. In addition, Tyr91_LCDR3 and Asn92_LCDR3 make hydrogen bonds with the side chain of Glu72 (Figure 2). Using MD simulation, we confirmed the absence of interaction from HCDR1 and the formation of the salt bridge between HCDR2 Arg50 and CD38 Glu72 (Supplementary Fig. S1). The analysis reveals the loss of direct hydrogen bonding from Tyr99_HCDR3 to His79 and Tyr100A_HCDR3 to Val80. The side chains of these two residues move to cover a larger hydrophobic patch in CD38 (Supplementary Fig. S3). This results in Arg50 and His58 being the principal determinants of HCDR2 antibody-antigen specificity. These two residues, flanking HCDR2, interact with the carboxylic moiety of Glu72 on CD38, with the Arg50 forming a strong interaction directing its guanidine moiety toward the Glu72 carboxylate moiety with hydrogen bonding geometry, and the His58_CDRH2 N$\mathit{\delta }$- and Glu72 h-bond also constantly present. Notably, despite no water molecule present in the X-ray structure at the complex interface, during the MD simulation a distinct water molecule occupies the cavity between the backbone of the heavy chain of E2-RecA and Glu72 of CD38 and mediates a H-bond bridge between the hydroxyl group of Ser100C_HCDR3, the backbone oxygen of Y100A _HCDR3 and the carboxylic group of Glu72 (Supplementary Fig. S4). Another water molecule also enters and mediates a hydrogen bond between the carboxylic group of Glu72 (CD38) and the backbone amine of Ser100C _HCDR3 (Supplementary Fig. S5). This interaction, while mediated by different water molecules, is always present during the entire equilibrium phase of the MD simulation. The molecular mechanics with generalized Boltzman and surface area continuum solvation⁴⁷ (MM-GBSA) free energy decomposition analysis confirms the importance of Arg50_HCDR2, Leu54_HCDR2, His58_HCDR2 and Tyr100D_HCDR3 for the interaction, with the corresponding per residue binding free energy (ΔG) values exceeding −5.0 kcal/mol. It also identifies Ile52_HCDR2, Tyr99_HCDR3, Tyr100A_HCDR3, Ser100B_HCDR3, Ser100C_HCDR3, as well as the light chain residues Asn92_LCDR3, Asn93_LCDR3, Trp94_LCDR3 and Trp96_LCDR3 as hotspot residues with energies in the range between −4.6 and −1.0 kcal/mol (Supplementary Table S1).

The X-ray structures offer a deeper understanding of how the cLC engages with the epitopes on CD38. As intended by design, the cLC interacts with CD38 to a much lesser extent than the heavy chain; however, for both B6-D9/CD38 and E2-RecA/CD38 complexes, the residues of the LCDR3 loop are directly involved in the interaction with the epitope. For both complexes the MM-GBSA free energy decomposition analysis reveals significant stabilizing values for the Trp94_LCDR3 and Trp96_LCDR3 residues, with a free energy per residue between −1.27 and −4.08 kcal/mol (Supplementary Fig. S1 and Table 1). The inspection of the B6-D9/CD38 structure suggests hydrophobic contacts of these two tryptophan sidechains with the aliphatic portions of CD38 Glu233 and Lys234, i.e., there are significantly stabilizing stacking interactions in the sidechain packing. It also suggests the formation of H-bonding between the (dynamical) water molecules and CD38 in the case of B6-D9, and direct H-bonds between E2-RecA and CD38. The MD analysis confirms the possibility of the water-mediated H-bonding with Trp96_LCDR3 of B6-D9, with water molecules occupying a particular position at the interface (Supplementary Fig. S4 and S5). The MD simulation also shows the formation of additional hydrogen bonding between E2-RecA LCDR3 and CD38, namely between the side chain of the CD38 Arg78 and the Tyr91_LCDR3 and Asn92_LCDR3. Interestingly, in both complexes, Trp94_LCDR3 and Trp96_LCDR3 side chains demonstrate very stable π-π stacking intramolecular contact, but with the different rotation angles between the indole rings (Supplementary Fig. S6), which can be a demonstration of adaptivity of the LCDR3 and rendering its importance for the stabilization and relative orientation of the paratope loops. Comparison of the MM-GBSA ΔG values for the contribution of the light and heavy chains in the B6-D9/CD38 complex shows only a marginal contribution of the cLC. In the complex, the sum of the ΔG values for the heavy chain residues involved in the binding is −29.73 kcal/mol, while that of the cLC is only −3.31 kcal/mol (Supplementary Table S1). For the E2-RecA/CD38 complex the picture is quite different. While the heavy chain is still the major force driving the interaction (−36.56 kcal/mol) the cLC shows a larger effect, with a ΔG of −12.14 kcal/mol (Supplementary Table S1). This might be due to both the previously mentioned presence of interactions between CD38 and the cLC LCDR3 residues, as well as better complementarity and larger surface of interaction between the epitope in CD38 and the cLC of E2-RecA.

Comparison of the binding modes of B6-D9, E2-RecA, daratumumab and isatuximab

The comparison of the co-crystal structures of B6-D9/CD38 and E2-RecA/CD38 with the previously reported crystal structures of CD38 in complex with daratumumab and isatuximab enabled us to compare their binding modes.^16,41 From Figure 3 it is evident that B6-D9, E2-RecA, daratumumab and isatuximab bind to distinct epitopes on CD38.

Figure 3.

Anti-CD38 B6-D9, anti-CD38 E2-RecA, daratumumab and isatuximab bind to distinct epitopes on human CD38. a Footprint of B6-D9 and E2-RecA on the hsCD38 highlighting CD38 residues interacting with B6-D9 in light blue and E2-RecA in light cyan. b footprint of daratumumab on the hsCD38 highlighting CD38 residues interacting with daratumumab in limon. c footprint of isatuximab on the hsCD38 highlighting CD38 residues interacting with isatuximab in yellow. d surface representation of the epitopes of B6-D9, E2-RecA, daratumumab and ixatuximab. Residues labeled correspond to overlapping residues between the epitope of B6-D9 and daratumumab (light green), the epitope of B6-D9 and isatuximab (brown) and the epitope of E2-RecA and isatuximab (green). e the binding modes of anti-CD38 B6-D9, E2-RecA and daratumumab on CD38 show that all fabs can bind simultaneously to the surface of CD38. CD38 is represented as a space filling representation with B6-D9 and E2-RecA epitopes in light blue and light cyan, respectively. The B6-D9 and E2-RecA fabs are depicted and colored as described in Figure 1. f while E2-RecA and isatuximab can bind simultaneously to CD38, the binding orientations of B6-D9 and isatuximab clearly cause steric collision between both fabs, preventing their simultaneous binding to CD38. The light chain of the anti-CD38 B6-D9 fab sterically clashes with the heavy chain of isatuximab (heavy chain in orange and light chain in yellow space filling). The steric collision of the two antibodies is indicated with a dashed ellipse. B6-D9 epitope is highlighted in light blue.

B6-D9 recognizes a distinct but overlapping epitope with daratumumab and isatuximab (Figure 3a–d). Two residues are shared between the epitopes of B6-D9 and daratumumab (Asp202 and Gln236), and three residues are shared between the epitopes of B6-D9 and isatuximab (Glu198, Gln231 and Glu233) (Figure 3d). While the binding modes of B6-D9 and daratumumab show that both Fabs can bind simultaneously to the surface of CD38 (Figure 3e), the binding orientations of B6-D9 and isatuximab clearly cause steric collision between both Fabs, preventing their simultaneous binding to CD38 (Figure 3f). These data confirm the results obtained from the epitope binning (Figure 1c), and from the competition assay previously reported.²⁶

E2-RecA recognizes a distinct epitope relative to B6-D9, daratumumab, and isatuximab and does not compete with these antibodies (Figure 3e–f). The epitopes of E2-RecA and isatuximab share two common residues (Arg78 and His79), but, due to the binding orientation of both Fabs, they can still bind to CD38 simultaneously (Figure 3d–f). Thanks to the absence of competition between E2-RecA and isatuximab, ISB 1442 binds to CD38 after saturation of the CD38 surface with isatuximab, as measured by bio-layer interferometry (BLI), while a complete competition was observed when CD38 was saturated with ISB 1442 (Supplementary Fig. S7a). Binding response of ISB 1442 on CD38 surface saturated with isatuximab reaches 50% of the binding response of ISB 1442 on non-saturated CD38 (Supplementary Fig. S7b).

Biparatopic ISB 1442 binds CD38 in trans, not cis

To understand how ISB 1442 interacts with CD38, we assessed whether the interaction of the two biparatopic anti-CD38 Fabs of ISB 1442 with a single CD38 monomer is sterically possible (cis binding). As observed previously, the two Fabs bind CD38 on diametrically opposed regions. The distance between the C-terminal extremity of B6-D9 and the N-terminal extremity of E2-RecA is more than 96 Å (Figure 4, left panel). A structural model built by superimposing the crystal structures of the two Fabs, simultaneously engaging the same CD38 molecule and linked by an artificial linker, was generated (Figure 4, right panel). This revealed that, to allow simultaneous binding of a dual Fab arm to the same antigen, at least a (G4S)8 linker to cover the distance of 96 Å would be required (yellow line, Figure 4, left panel). Therefore, the (G4S)3 linker used in ISB 1442, which has a maximum theoretical length of about 42 Å as deduced from modeling the unbound (G4S)3 peptide with AlphaFold,^48,49 will not allow for both Fabs to bind simultaneously to the same antigen. ISB 1442 would have two key advantages if both arms do not bind the same CD38 antigen simultaneously: 1) two ISB 1442 would bind to a CD38 monomer on the cancer cell instead of one, resulting in twice the cell surface Fc density and thus augmented cytotoxicity; and 2) avid binding to shed soluble monomers of CD38 would be minimized, preventing reduction of ISB 1442 cytotoxicity in the presence of soluble CD38.²⁶ This design of ISB 1442, where the biparatopic Fab arms are constrained by a short linker, has a potential advantage compared to where the biparatopic Fabs are spaced wider, with a flexible IgG1 hinge.

Figure 4.

ISB 1442 dual fab arm cannot bind CD38 in cis. B6-D9 and E2-RecA Fabs bind distinct epitopes located on opposite side of hsCD38. a The distance between the C-terminal extremity of B6-D9 and the N-terminal extremity of E2-RecA is more than 95 Å. b to allow both Fabs to bind simultaneously to the same antigen it would require at least a (G4S)₈ linker to cover the distance of 96 Å (yellow line, left panel)). Artificial (G₄S)₈ linker is colored in red (right panel). V_H and C_H1 domains of the anti-CD38 B6-D9 and E2-RecA are represented blue and magenta, respectively.

ISB 1442 forms 2:2 heterodimer complexes in solution

Size-exclusion chromatography with multi-angle light scattering (SEC-MALS) was conducted to analyze the binding stoichiometry of the ISB 1442/CD38 complex. After incubation, using CD38 in excess, relatively homogeneous 2:2 complexes of ISB 1442/CD38 of ~ 488 kDa are formed (Figure 5a). Smaller and larger heterogenous complexes were also observed on the chromatogram, suggesting that interaction events can also occur with 1:1 and 2:1 ISB 1442:CD38 ratios (one or two antibodies binding to one antigen), or that ISB 1442 can form large bridging immunocomplexes with CD38.

Figure 5.

ISB 1442 forms 2:2 heterodimer complexes in solution binding to two CD38 in an antiparallel fashion. a ISB 1442:CD38 binding stoichiometry. SEC-MALS profiles for CD38, ISB 1442 and ISB 1442/CD38 complex in the presence of excess of CD38 after overnight incubation. The relative absorbance at 280 nm (right y-axis) is plotted against elution volume from a Shodex column (protein LW-803) and overlaid with the molar mass determined for each peak (left y-axis). Predicted and calculated molecular masses are shown in the table. b Representative cryo-em 2D class average (left) and a schematic diagram of the ISB 1442/CD38 complex (right). c low resolution negative-stain (NS) EM maps showing two ISB 1442 dual fab arm bound to two CD38 in an antiparallel fashion. d overlay of the model NS EM map with the crystal structure.

To identify how ISB 1442 binds CD38 in solution, we generated a 3D reconstruction of their complexes using negative stain EM (Figure 5b–c). Most particles observed were those of two ISB 1442 molecules bound to two CD38 molecules in an antiparallel fashion. This observation supports the data obtained by SEC-MALS that a 2:2 ISB 1442:CD38 complex is formed in solution. ISB 1442 shows the typical “Y” shape antibody profile with three distinct domains, two blurry domains corresponding to the anti-CD47 Fab and to the Fc, and the third appearing as a two “bean” shaped domain corresponding to two anti-CD38 Fabs on the same arm.

ISB 1442 shows superior binding properties compared to monoparatopic antibodies at low and at high CD38 surface densities by SPR

To study the possibilities of cooperative interaction of several ISB 1442 molecules with surface-bound CD38, the binding of ISB 1442 to human CD38, immobilized at various surface densities, was measured by SPR and compared to a set of control molecules (Figure 6a). Control molecules included the 2 + 1 bivalent and monovalent (for CD38) monoparatopic controls.

Figure 6.

Binding comparison of ISB 1442 and monoparatopic control antibodies to different human CD38 surface densities by SPR. a schematic view of ISB 1442, CD38-monoparatopic bivalent and CD38-monoparatopic monovalent control antibodies used in the assay. Immunoglobulin domains are shown as rectangles. VH of anti-CD38 fab domains are shown in two different shades of blue. VH of anti-CD47 is depicted in pink. All fab domains make use of an identical common light chain (cLC) depicted in gray. The BEAT® interface in the CH3 domains is depicted by the yellow and black dots. b plots of binding response at end of association (in response units, RU, y-axis) vs analyte concentration (in nM, x-axis) for ISB 1442 and control antibodies to human CD38 immobilized at a 10 RU surface density (left panel) or at a 300 RU surface density (right panel) on a sensor CHIP as determined by SPR. All blank-subtracted binding levels were normalized to capture levels. Curves are colored by antibody used as analyte. c table shows percent of theoretical rmax (% Rmax_theo) calculated for a 1:1 antibody:CD38 interaction using a 1:1 kinetic fit. d table shows percent Rmax_theo of control antibodies as compared to ISB 1442 at different CD38 surface density. * in (c) and (d): data less precise (estimates) as CD47_CD38-2_DU did not reach surface saturation.

At low CD38 surface density, ISB 1442 demonstrated approximately 2-fold higher maximal binding at surface saturation compared to both monovalent and bivalent monoparatopic antibodies (Figure 6b, left panel). If the anti-CD38 dual Fab arm of ISB 1442 could bind CD38 in cis, the maximal binding at saturation of ISB 1442 may expect to be comparable to monovalent or bivalent monoparatopic antibodies. These SPR results showed that, at low density, two ISB 1442 can bind to both epitopes on the same CD38-ECD, as suggested by the crystal structure analysis and heterodimer complexes formation in solution. This would result on two Fc per CD38 antigen when ISB 1442 binds to low density CD38. At high CD38 surface density, both ISB 1442 and anti-CD38 monovalent antibodies resulted in approximately a 2-fold higher maximal binding compared to the anti-CD38 bivalent monoparatopic antibodies (Figure 6b, right panel). Higher maximal binding of ISB 1442 as compared to anti-CD38 bivalent monoparatopic antibodies at high CD38 surface density could be explained by binding to both epitopes on CD38 of two ISB 1442 molecules in a 2:2 mode (see Figure 5), resulting in one Fc per CD38 antigen. The anti-CD38 bivalent monoparatopic antibodies bind to CD38 in trans. This enables the binding of two CD38 molecules by one antibody and prevents the binding of a second antibody. This can explain the difference in binding stoichiometry and lower maximal binding at saturation compared to ISB 1442. ISB 1442 also showed a 1.1-fold higher antibody density as compared to CD47_CD38-1_DU at high CD38 surface density, likely because not all ISB 1442 molecules bind in a 2:2 mode, but also by engaging only one anti-CD38 arm in the other binding stoichiometry (e.g., 3:2), resulting in > 1 Fc per CD38 (Figure 6b, right panel). CD47_CD38-2_DU did not reach saturation at the highest concentration used, but the curve shape suggests that it would also reach a higher maximal binding response than the anti-CD38 bivalent monoparatopic molecules if the analyte injection concentration was increased further. CD47_CD38-2_DU may not have achieved saturation at an analyte concentration of 100 nM because the anti-CD38 E2-RecA Fab arm is in the “inner” position (Figure 6a) and this may restrict the Fab arm conformational freedom and subsequently limit binding to its target. Comparison of the surface analyte binding capacity (Rmax) derived from 1:1 kinetic fits, applied to different surface densities of human CD38, showed around 1.2–1.4-fold increased analyte binding capacity at low surface density (10 RU) compared to high surface density (300 RU) for ISB 1442 and anti-CD38 bivalent monoparatopic antibodies; no increased analyte binding capacity was observed for anti-CD38 monovalent antibodies (Figure 6c). Comparison of Rmax also showed approximately 1.6–2.0-fold lower analyte binding capacity for anti-CD38 bivalent monoparatopic antibodies compared to ISB 1442 at each CD38 surface density tested; anti-CD38 monovalent antibodies showed comparable analyte binding capacity to ISB 1442 at high CD38 surface density (300 RU), but reduced analyte binding capacity at lower CD38 surface densities (Figure 6d). Altogether, these data show that 2-fold more ISB 1442 than anti-CD38 monoparatopic molecules can bind at low CD38 surface density, indicating that more than one ISB 1442 can engage a single CD38 molecule. The higher number of Fc domains displayed on CD38 for the biparatopic antibody may enable more potent Fc effector function than for monoparatopic antibodies, especially through increased CDC, ADCC and ADCP.^7,10,26

We further compared binding of ISB 1442 and a combination of anti-CD38 monovalent antibodies to different surface densities of CD38. At low CD38 surface density, ISB 1442 and the combination of CD47_CD38-1_DU and CD47_CD38-2_DU showed comparable binding, confirming that two molecules of ISB 1442 or one molecule of each anti-CD38 monovalent antibodies bind simultaneously to one CD38. At high CD38 surface density, the combination of both anti-CD38 monovalent antibodies showed approximately 1.6-fold higher binding as compared to ISB 1442, indicating that one molecule of each anti-CD38 monovalent antibody can bind simultaneously to one CD38 molecule, while ISB 1442 mainly interacts with CD38 in a 2:2 mode of binding, resulting in one Fc per CD38 molecule (Supplementary Fig. S8). While the higher number of Fc present at the surface of cells expressing high levels of CD38 when using a combo of anti-CD38 monovalent antibodies as compared to ISB 1442 is expected to lead to increased Fc functions, Grandclément et al.²⁶ showed no benefit of using a combination of monoparatopic monovalent antibodies as compared to ISB 1442 in a CDC assay on cells expressing high levels of CD38.

ISB 1442 shows superior binding kinetics at high CD38 surface densities

ISB 1442 and the bivalent monoparatopic molecules exhibited reduced dissociation at high versus low CD38 surface density. In comparison, anti-CD38 monovalent antibodies showed no reduced dissociation at high versus low CD38 surface densities (Figure 7a–b). Reduced dissociation of ISB 1442 from CD38, measured as the binding at the end of dissociation minus the binding at the start of dissociation, showed 5.8-fold slower dissociation at 300 RU compared with 10 RU surface density. This decrease in dissociation correlated with an increase in avidity-mediated apparent affinity (K_D^app) of ISB 1442 at high compared to low CD38 density. The apparent affinity of ISB 1442 at high CD38 density was increased by at least 14-fold compared to low CD38 density (K_D^app <0.01 nM vs 0.139 nM, Supplementary Tables S2 and S3); the K_D^app could not be precisely determined due to the kinetic parameters reaching the limitations of the instrument. The affinity of the anti-CD38 monovalent antibodies at any CD38 density was comparable or lower than the affinity of ISB 1442 at low CD38 density (Supplementary Tables S2 and S3).

Figure 7.

ISB 1442 binding affinity to CD38 is increased at high CD38 surface density. a. SPR binding sensorgrams of ISB 1442 and control antibodies injected in single-cycle at concentrations from 0.41 nM to 100 nM in 1/3 dilution series over immobilized human CD38. Binding response curves to human CD38 immobilized at 10 RU, 30 RU, 100 RU and 300 RU are displayed in single plot per antibody. Binding response curves were aligned on the y-axis on the CD38 capture level and normalized to the end of the last concentration of antibody injection to allow direct comparison of dissociation curves. Plots embedded in larger plots show amplified dissociation phases, respectively. Data show normalized response (RU, y-axis) over time (s, x-axis). Curves are colored by CD38 capture level. b table shows fold increase stability of each antibody at 30 RU, 100 RU and 300 RU CD38 capture level as compared to 10 RU CD38 capture level.

In summary, ISB 1442 binds with a larger number of molecules than anti-CD38 monovalent and bivalent monoparatopic antibodies at low CD38 surface density, where no avidity is present. ISB 1442 also binds with a larger number of molecules than anti-CD38 bivalent monoparatopic antibodies and with higher affinity than anti-CD38 monovalent antibodies at high CD38 surface density due to avidity.

Binding of ISB 1442 and monovalent antibodies on CD38-expressing tumor cells

To translate the SPR findings to cell binding properties, we assessed ISB 1442 on tumor cells expressing different level of CD38 and compared it to control antibodies selectively replacing each binding Fab by an irrelevant dummy Fab (DU) (Figure 8a). Antibody binding was assessed on three different cell lines, KMS-12-BM, NCI-H929, and Daudi, with CD38 copy numbers of 11 $\times$ 10³, 31 $\times$ 10³ and 180 $\times$ 10³, respectively. There was no statistically significant difference in maximal binding for ISB 1442 compared with the same molecule lacking the CD47 binding arm (CD38–1 + 2_DU), for each cell line, indicating that the CD38 binding arms were the main drivers for cancer cell binding. As predicted by structural modeling, and observed by biochemical binding experiments with recombinant protein, the biparatopic anti-CD38 antibodies showed higher maximum specific binding (Bmax) than the monovalent antibodies on low and intermediate CD38-expressing cells, indicating a higher antibody density at the cell surface (Figure 8b–c). The Bmax of ISB 1442 was not significantly different from the monovalent controls in terms of maximum binding to the Daudi cell line with a higher copy number (Figure 8d). Despite not being significantly different, ISB 1442 lacking the CD47 arm (CD38–1 + 2_DU) showed a 1.1-fold higher Bmax as compared to the monovalent CD38-1_DU on Daudi cells, which is in line with SPR data and indicates that ISB 1442 likely engages CD38 high expressing cells with slightly more than 1 Fc per CD38, on average. The difference between the Bmax of the biparatopic and the monovalent B6-D9 antibody decreases with increasing CD38 expression levels of the cell lines. A lower Bmax observed for the E2-RecA monovalent antibody correlates with the lower binding observed by SPR (Figure 6), as well as with its slower on-rate (Table 1 and Figure 7). ISB 1442 shows a slightly higher Bmax compared with the biparatopic control antibody (CD38–1 + 2_DU), likely due to the additional avidity of binding contribution of the CD47 binding arm, which would result in a slower off-rate during the cell washing steps. These findings are in alignment with the biological activity of ISB 1442 previously reported, whereby employing a 2 + 1 biparatopic antibody resulted in higher binding to CD38-expressing tumor cells and enhanced CDC relative to bivalent monoparatopic antibodies.²⁶

Figure 8.

Biparatopic versus monovalent CD38 targeting. a schematic view of ISB 1442 and dummy control antibodies. b-d cancer cell lines were incubated with varying antibody concentrations and analyzed by flow cytometry. Upper row: representative experiment shows relative fluorescent intensity (RFI) plotted against antibody concentration for CD38^low cells (KMS-12-BM) (b), CD38^{low-intermediate} cells (NCI-H929) (c) and CD38^high cells (Daudi) (d). Lower row: respective plots of mean RFI ± standard deviation (SD) in at least 3 independent experiments. Statistics: ns = not significant, *p < 0.05 **p < 0.01 (RM one-way ANOVA followed by šidák’s multiple comparisons tests).

To assess whether the relative positions of the anti-CD38 binding domains affect function, an alternative trispecific construct, in which the positions of the two CD38 binding domains were interchanged relative to ISB 1442 (B6-D9 in the “inner position” and E2-RecA in the “outer position”), was generated and compared to ISB 1442. Interestingly, we found that cytotoxicity of ISB 1442 was comparable to the alternative construct with B6-D9 placed at the “inner position”, suggesting that, using the ISB1442 bispecific biparatopic format, the position of the anti-CD38 Fabs could be interchanged (Supplementary Fig. S9).

Grandclément et al. tested ISB 1442 potency in the absence or presence of soluble CD38 (sCD38).²⁶ They demonstrated that the molecule potency is not reduced in the presence of sCD38 at physiological concentration. Here, we compared the binding of ISB 1442 and its monoparatopic-monovalent and monoparatopic-bivalent controls in the presence of excess of sCD38 and showed that ISB 1442 binding to NCI-H929 is less affected by the presence of sCD38 than the monovalent anti-CD38 controls (Supplementary Fig. S10). While the IC50 of sCD38 is reached with a 2.32 molar excess of CD38 to ISB 1442, lower concentrations of sCD38 are sufficient to reach IC50 for the monovalent controls, with 1.03, 1.43, 0.52 and 0.92 molar ratio of CD38/antibody for CD47_CD38-1_DU, CD47_CD38–1, CD47_CD38-2_DU and CD47-CD38-2, respectively (Supplementary Fig. S10c). CD47_CD38–1 + 1 monoparatopic bivalent control showed equivalent resistance to sCD38 as compared to ISB 1442 due to avid binding to membrane-bound CD38, while CD47_CD38–2 + 2 showed to be significantly more impacted by the presence of sCD38, with an IC50 close to equimolar ratio (0.93). This is in line with other findings showing that the E2-RecA Fab binds less than the B6-D9 Fab to immobilized CD38 (either at the cell surface or on SPR sensors), likely due to lower accessibility of its epitope, which is closer to the membrane. In such case, E2-RecA is likely to bind preferentially to sCD38 where its epitope is expected to be more accessible. Overall, these data are in accordance with our findings where ISB 1442 trans binding to CD38 molecules promotes avid interaction with adjacent receptors at the cell surface over monomeric shed soluble CD38.

Structural modeling of biparatopic interactions between ISB 1442 and CD38

Based on the data generated, including the X-ray structures of both anti-CD38 Fab complexes, the SPR experiment, and the evaluation of the cell surface binding to CD38 expressing cancer cells, we generated several structural models of a complex of ISB 1442 with membrane bound CD38. We used as templates the previously reported BEAT^Ⓡ Fc structures,²⁷ with the crystal structures of the B6-D9 and E2-RecA Fabs in complex with CD38, which was embedded into the modeled lipid bilayer (Figure 9a). All models were tested for a lack of molecular steric clashes with Schrödinger software. The obtained structural models clearly reveal that several ISB 1442/CD38 complexes are possible, depending on the surface density of CD38. For example, at low CD38 density one CD38 monomer can bind two different ISB 1442 molecules, as illustrated by Figure 9a, left panel resulting in an ISB 1442:CD38 ratio of 2:1. At intermediate CD38 cell density, the antiparallel 2:2 heterodimer complex previously observed in solution by negative staining was also modeled with the membrane integrated CD38 (Figure 9a, second panel). The bridging of two CD38 monomers by a single ISB 1442 does not prevent the remaining available epitopes from binding to two additional ISB 1442 molecules, resulting in a complex with a 3:2 stoichiometry (Figure 9a, third panel). At high CD38 density, three CD38 monomers can theoretically be linked by four different ISB 1442 molecules in a complex with an ISB 1442: CD38 stoichiometry of 4:3 (Figure 9a, right panel). Here two ISB 1442 can bind additional targets and, thus, potentially bridge other ensembles of CD38 monomers resulting in extensive clustering.

Figure 9.

Structural modeling of biparatopic interactions of ISB 1442 BEAT® with human CD38. a from left to right, computational model of two ISB 1442 molecules interacting with the same membrane integrated CD38 monomer (2:1); two ISB 1442 molecules interacting with two different membrane integrated CD38 monomers (2:2); three ISB 1442 molecules interacting with two different membrane integrated CD38 monomers (3:2); and four ISB 1442 molecules interacting with three different membrane integrated CD38 monomers (~4:3). VH domains of the anti-CD38 B6-D9 and E2-RecA are represented in blue and cyan, respectively. VH domain of the anti-CD47 binder is represented in purple. All binders make use of a common light chain (cLC) depicted in gray (VL) and light gray (CL). CH1 domain is represented in black and the Fc domain in white. Membrane lipid bilayer is depicted in brown. b-d. Schematic representation of the different binding modes of ISB 1442 (b), a CD38 monoparatopic bivalent anti-CD38×CD47 bispecific antibody (c) and a CD38 monovalent anti-CD38×CD47 bispecific antibody (d) depending on the expression of CD38 at the cell surface. Dual biparatopic arm enables to increase target occupancy and antibody densities on the target cell surface (particularly for low CD38 expressing cells) favoring overall functional activity. Antibodies colored as in Figure 8a, with chain a IgG1 CH2 and chain B engineered human IgG1 CH2-CH3 represented as solid line white square and chain a engineered IgG3 CH3 represented in dash line white square. CD38 monomer is depicted in green. The BEAT® interface shown in the CH3 domains is depicted by the green and black dots.

The models of ISB 1442 in complex with CD38 suggest that several structures are possible depending on the cell surface density of CD38. To illustrate the advantage of the ISB 1442 design in terms of increased Fc cell density, schematic illustrations of ISB 1442 bound to low, intermediate or high density CD38 are shown in Figure 9b (left, middle and right panels, respectively). This was compared with schematics showing bivalent, monoparatopic (Figure 9c) or monovalent monoparatopic CD38 binding (Figure 9d). Various permutations of complex formation are possible, but from low to high CD38 surface density, the biparatopic design of ISB 1442 enables a higher Fc density compared with both bivalent monoparatopic and monovalent monoparatopic designs. These data and the X-ray-derived models demonstrated the mechanism of ISB 1442 engagement and show that ISB 1442 dual biparatopic arm binds CD38 in trans. The trans binding of ISB 1442 enables increased target occupancy and local Fc-domain density compared to bivalent monoparatopic antibody at both high and low CD38 cell expression levels and compared to monovalent antibody at low CD38 cell expression levels (Figure 9b–d).

Discussion

In this study, we provided details on the generation and structural characterization of ISB 1442, a BsBpAb, currently being evaluated in a Phase 1/2 clinical trial (NCT05427812) for the treatment of relapsed refractory multiple myeloma.²⁶ The discovery and engineering of ISB 1442 were enabled by two key technologies: the implementation of a cLC antibody library, that was used to identify the binding domains recognizing CD47 and CD38, and the BEAT^Ⓡ Fc heterodimerization interface^27,28 to assemble the Fabs into an asymmetric bispecific antibody. Structural analysis of both anti-CD38 cLC Fabs in complex with CD38 confirmed that ISB 1442 binds to two non-overlapping epitopes on CD38, without competing with daratumumab and only partially competing with isatuximab. In addition, the combination of X-ray crystallography, SEC-MALS and negative staining EM demonstrated that co-engagement of CD38 by ISB 1442 can only happen in trans, leading to efficient target cross-linking at the cell surface. The inability of ISB 1442 to co-engage soluble shed CD38 in cis is an advantage because soluble shed CD38 in patient sera will not effectively inhibit the avid binding to cancer cells.²⁶

SPR experiments confirmed that ISB 1442 mediates a higher target saturation compared to bivalent monoparatopic antibodies at both low and high CD38 density. ISB 1442 also achieves increased target saturation compared to monovalent antibodies at low CD38 density. SPR data correlate with the superior maximal binding of ISB 1442 to CD38^low KMS-12-BM and CD38^int NCI-H929 cells compared to monovalent antibodies, as well as with the superior maximal binding to CD38^high Daudi cells compared to bivalent monoparatopic antibodies reported previously.²⁶ At high CD38 density, ISB 1442 also showed a 1.1-fold superior antibody density as compared to monovalent antibodies both by SPR and on Daudi cells. Although the antibody densities were not significantly different, this suggests that not all ISB 1442 molecules bind to CD38 in a 2:2 ratio, but some ISB 1442 molecule may bind with one arm only, resulting in a slightly higher number of Fc per CD38 as compared to monovalent antibodies.

ISB 1442 also shows a higher avidity-mediated apparent affinity, and slower off rate, than monovalent antibodies at high CD38 density by SPR. This could explain superior phagocytosis and killing of Daudi cells in CDC compared to monovalent bispecific control antibodies.²⁶ Although a combination of two anti-CD38 monovalent bispecific antibodies results in a higher number of Fc per CD38 as compared to ISB 1442 at high CD38 surface density, no superior cell killing was observed in CDC.²⁶ This indicates that Fc density does not necessarily follow a linear correlation with potency in CDC. Other mechanisms, such as the avidity of ISB 1442 or formation of optimal immune complex architecture may compensate for the lower Fc density at the surface as compared to a combination of two anti-CD38 monovalent bispecific antibodies. Alternatively, there might be a threshold of Fc density after which Fc functions such as CDC are not further increased. Monovalent control antibody comprising the E2-RecA Fab consistently showed lower maximal binding to CD38 positive tumor cells and was reported to induce inferior killing of tumor cells compared to the monovalent control antibodies comprising the B6-D9 Fab.²⁶ This difference may be due to the E2-RecA Fab being in the sterically constrained “inner position”, thus reducing its capacity to freely bind to CD38, although other characteristics of both Fabs, such as the different targeted epitopes, may also account for observed differences in cytotoxicity.⁵⁰ Taken together, these results provide insight into the superior activity of ISB 1442 compared to daratumumab described previously²⁶ and support the observed preclinical superiority of a biparatopic bivalent bispecific antibody compared to a monoparatopic bivalent bispecific antibody or a monovalent bispecific antibody to kill multiple myeloma cells.

To promote anti-tumor activity, we are advancing the biparatopic antibody concept by combining bispecific targeting of two different antigens (CD38 and CD47) and biparatopic targeting of one of these antigens (CD38). Biparatopic molecules possessing variable formats and different mechanisms of action (e.g., receptor clustering and down-regulation by internalization, increased drug conjugate uptake, increased affinity, or enhanced effector Fc functions) have been developed.⁹ Two main approaches have been utilized in the field to isolate cLC Fabs either starting from existing antibodies while engineering a tailor-made LC compatible with the different heavy chains or directly isolating antibodies with a generic cLC from discovery platforms.^9,51 A generic cLC greatly facilitates the generation of antibodies composed of more than two binding moieties, such as ISB 1442, contrary to the tailor-made approach that would require extensive LC engineering. In our study, we demonstrated that biparatopic antibodies can also be isolated from a fully human synthetic phage display cLC library. In addition, the concept was demonstrated by targeting CD38, a rather small protein (ECD composed of 1 domain and 257 amino acids) with a reduced druggable surface to exclude the daratumumab footprint, further demonstrating the potential of the cLC technology to target the desired epitopes. We have also showed that the cLC Fab/antigen interface is differentiated from other antibody/antigen interactions with a reduced antigen interaction surface area (1,304 Ų and 1,297 Ų for B6-D9 and for E2-RecA, respectively, compared to 2,274 Ų and 2,492 Ų for daratumumab and isatuximab, respectively^16,41 and superior contribution to the heavy chain (86% and 75% for B6-D9 Fab/human CD38 and E2-RecA/human CD38, respectively, 67% for other antibodies.⁴² Interestingly, the cLC also contributes to the interacting interface through LCDR3, in which W94 and W96 were key binding contributors in both complexes.

The optimization of ISB 1442 involved the affinity maturation of anti-CD38 Fabs. Here, the affinity of E2 to human CD38 was increased by 69-fold by selecting 10 residue substitutions in HCDR1 and HCDR2. This affinity improvement can be partially explained when analyzing the structure of E2-RecA Fab in complex with human CD38 because the selected F54L and S56A HCDR2 residues were shown to directly interact with CD38. In addition, MD simulations demonstrated that residue F54L was a main contributor to the interaction with human CD38, with a free energy of −5.5 kcal/mol. Surprisingly, our study also revealed that some affinity-improving substitutions are not located at the interacting interface. Indeed, the affinity of E2 can be increased by 12-fold by selecting 5 substitutions within HCDR1 although this CDR is not in contact with CD38. A possible explanation could be that a residue of the parental antibody clashed with CD38. These mutations may also play a role in the overall rigidification of the paratope, a mechanism described to contribute to antibody affinity increase.⁵² Similarly, the affinity of B6 to human CD38 was increased by 375-fold by selecting 5 amino-acid substitutions in HCDR1. This result can also be partially explained by the structure of the complex of B6-D9 Fab with human CD38 demonstrating that the selected N31H and A33I residues interacted with the antigen. In addition, MD simulations confirmed the preponderant role of N31H in the interaction with human CD38, with a free energy of −5.4 kcal/mol. Interestingly, E2-RecA F54L and B6-D9 N31H selected during affinity maturation and likely contributing to the antibody affinity increase, are located at the periphery of the Fab-antigen interface. This observation is in accordance with previously reported observations⁵² and is likely due to the periphery of the interface being more permissive to affinity-enhancing mutations.

CDC is an important effector mechanism of anti-CD38 immunotherapies. However, low CD38 expression level in some hematological malignancies was associated with reduced CDC and lack of efficacy of unmodified anti-CD38 IgG1 mAbs such as daratumumab. To overcome this limitation, alternative approaches (HexaBody-CD38⁵³ or biparatopic molecules^10,11,54 aiming at augmenting CDC via CD38 crosslinking were described.^10,11,55,56 The HexaBody® technology is based on the introduction of mutations in the Fc part of CD38 antibodies enhancing the formation of hexamers upon target engagement and leading to high CDC activity. Although the HexaBody® format can kill cancer cells with a lower antigen expression than regular antibodies, this requires careful assessment of safety and the therapeutic window for targets that are also expressed on healthy tissues.⁵⁷ Therefore, this technology may not be applicable in the context of ISB 1442 due to the high CD47 expression in RBCs that could result in poor pharmacokinetics and on-target off-tumor depletion of RBCs. Alternatively, Schütze et al.¹⁰ demonstrated that biparatopic engagement of CD38 using a bispecific llama/human heavy chain antibody (hcAb) mediated stronger CDC of myeloma cells compared to daratumumab. This molecule may maximize CD38 crosslinking on the cell surface and facilitate the complement activation. In addition, hcAbs are symmetric bispecific molecules that can bivalently engage both CD38 epitopes. The authors hypothesized that the tetravalency of their molecule could more efficiently induce C1q-activating oligomers compared to bivalent biparatopic asymmetric antibodies. However, the tetravalent biparatopic VHH format described by Schütze et al. may possess several disadvantages. Firstly, this format would result in hexamers with an Fc to CD38 ratio of 6:12, whereas ISB 1442 has been proposed to form hexamers with a higher Fc to CD38 ratio of 6:6, potentially resulting both in a greater number of clusters per cell and enabling cluster formation in cells with a lower CD38 copy number. Secondly, the relatively long spatial distances between the Fab arms and separation by the flexible IgG1 hinge may enable binding to CD38 in cis, strong avid binding to soluble shed CD38 and a reduction in efficacy for patients with a high tumor burden and increased soluble CD38 levels in their serum. Cis binding to CD38 would also not enable the antibody to CD38 binding ratio of 2:1 of which ISB 1442 has been shown to be capable at low CD38 surface density.

We have demonstrated that the biparatopic engagement of two different CD38 epitopes induces more CDC compared to its bivalent monoparatopic engagement,²⁶ likely due to an increased number of antibodies at the cell surface enabling Fc hexamer assemblies to enhance C1q binding and induce complement activation.⁵⁵ This is supported by the higher antibody density of the biparatopic over bivalent monoparatopic controls observed on Daudi cells, thanks to the likely predominant 2:2 mode of binding of the biparatopic versus the likely predominant 1:2 mode of binding of the monoparatopic at high CD38 density. Moreover, slightly higher antibody density of the biparatopic over monovalent CD38 controls observed on Daudi cells also translated into superiority in CDC. Avidity-mediated binding to CD38 of the biparatopic antibody might also account for the superior cell killing in CDC. We hypothesize that the distinct clustering properties of ISB 1442 promote a superior ability to form hexamers and engage C1q to activate the complement cascade.^7,10,56 As illustrated in Supplementary Fig. S11, ISB 1442 could form an optimal immune complex assembly by creating hexameric antibody/antigen complexes on the cell surface, facilitating efficient complement binding and activation. Nevertheless, other Fc functions are also dependent on thresholds of IgG clustering at the cell surface⁷ and the higher antibody density of the biparatopic over monovalent CD38 controls on KMS-12-BM (CD38 low) cells translated into superior ADCP.²⁶ A biparatopic antibody targeting HER2, zanidatamab (Ziihera®), was also shown to enhance CDC and tumor cell killing compared to the combination of trastuzumab (Herceptin®) and pertuzumab (PERJETA®), two mAbs approved for the treatment of HER2-positive breast cancer.¹¹ Zanidatamab is an asymmetric bispecific antibody composed of one scFv targeting HER2-ECD4 and one Fab targeting HER2-ECD2. ISB 1442 and zanidatamab present some similarities regarding the mode of recognition of their respective targets: 1) monovalent engagement of each target epitope; 2) one target molecule can be bound by two antibody molecules; and 3) target binding is possible only in trans, i.e., one antibody molecule cannot co-engage both arms on the same target molecule but can cross-link two target molecules. This comparison suggests that these CDC-enhancing antibody features may be applicable to other TAA.

To conclude, we demonstrated that our multispecific antibody platform based on BEAT^Ⓡ technology and cLC Fabs can be used to engineer multispecific antibodies with optimized biological activity using a combination of TAA-engaging arms targeting the relevant epitopes, with adequate affinity. Moreover, our study provides structural insights into the mechanism of action of ISB 1442, demonstrating that the optimal antibody activity lies in the biparatopic engagement of CD38 in a trans configuration where each epitope is engaged monovalently and one antibody cannot reach both epitopes on the same CD38 molecule, thus maximizing antibody presence at the cell surface. Beyond this study, this mechanism of action could be further exploited to improve cancer immunotherapies and more efficiently target low-expressing or downregulated TAA.

Materials and methods

Synthetic common light chain library generation

Human antibody sequences were uploaded from the abYsis database.⁵⁸ For each variable heavy chain (VH) germline, all different HCDR1 or HCDR2 amino-acid sequence regions (Kabat 25–36 and Kabat 47–65, respectively) were aligned and the natural residues frequencies determined. Based on structural analysis and natural diversity, buried residues with low mutation rate were excluded from the library design (Kabat 25–26, 36, 47–49, 59–65). Other animo acids were randomized according to natural diversity, excluding residues occurring < 1% and C, and M unless localized in a buried canonical position. Most exposed residues were “hard randomized” using all amino acids, independently from the natural diversity, with exceptions: C, M and W were removed, and P were not considered unless naturally occurring. HCDR3 (Kabat 95–102) design is based on natural diversity, excluding C, N and M (except that M was kept at 101–1 position). The final engineered designs’ diversity was encoded as trimer oligonucleotides using an E. coli optimized codon set (Ella Biotech). Library generation was performed by PCR assembly of the different fragments encompassing the HCDRs diversity. An initial step consisted of the generation of HCDR1/2 cassette using E. coli codon optimized germline FR1 or FR3 as DNA template and trimer oligonucleotides. HCDR3 synthetic diversity was introduced using a pool of trimer oligonucleotides encoding 15 hCDR3 lengths (6–20). DNA fragments encompassing E. coli codon optimized HCDR3, IGHJ1/4/5, linker and germline Vκ3–15/Jκ1 variable light chain (VL) were obtained by PCR and pooled to mimic natural HCDR3 length distribution. Finally, HCDR1/2 cassette and synthetic HCDR3-containing fragments were assembled by PCR, scFv were cloned into the pNGLEN (in-house modified pUC119 phagemid vector) using NcoI/Not1 restriction sites and the resulting ligation electroporated into E. coli TG1 cells. Bacteria were superinfected with M13K07 helper phage for assembly and production of recombinant phages. Phages were purified by two precipitations steps with 1/3 v/v of 20% PEG-6000, 2.5 M NaCl and resuspended in phosphate-buffered saline (PBS). Library quality was assessed by NGS using Illumina MiSeq 2 × 200bp analysis and custom analysis pipeline (Fasteris).

Cloning and transient expression of full-length human and cynomolgus monkey CD47 and cynomolgus monkey CD38

The human codon-optimized sequences of full-length cynomolgus monkey CD38 (accession number: Q5VAN0; residues 1–301), full-length human CD47 (accession # Q08722; residues 1–323) and full-length cynomolgus monkey CD47 (accession # A0A2K5X4I2; residues 1–323) were cloned in a modified pcDNA3.1 plasmid (ThermoFisher Scientific, Catalog NO: V79020). The vector also contained the enhanced green fluorescent protein (eGFP). The aforementioned plasmids were transfected into suspension-adapted CHO-S cells (Invitrogen, Catalog NO A1136401) using polyethyleneimine (PEI; Polysciences, Catalog NO: 23966). Briefly, cells were resuspended in CD CHO (Gibco, Catalog NO: 10743011) and transfected with a DNA-PEI mixture at 37°C. Four hours post-transfection, the cell culture was diluted 1:1 PowerCHO 2 (Lonza, Catalog NO: LZ-BELN12-771Q) supplemented with 4 mm L-Glutamine (GE, Catalog NO: GE SH300.34.02) and incubated with orbital shaking at 37°C, 5% CO₂ and 80% humidity. The expression of the target antigens was assessed by monitoring the expression of the eGFP reporter protein with a fluorescence microscope.

Cloning, expression and purification of human CD47(C33S), cynomolgus monkey CD47(C33S), human CD38, and cynomolgus monkey CD38 ECD

The human codon-optimized sequences of human CD47(C33S) ECD (accession #Q087221, residues 19–141) fused to a C-terminal Avi-10His tag (human CD47-Avi-His), of cynomolgus monkey CD47(C33S) ECD (accession# A0A2K5X4I2, residues 19–141) fused to a C-terminal Avi-10His tag (cyno CD47-Avi-His), of human CD38 ECD (accession#P28907, residues 45–300) fused to a C-terminal 6-His tag (human CD38-His) and of cynomolgus monkey CD38 ECD (accession # Q5VAN0, residues 45–301) fused to a C-terminal 6-His tag (cyno CD38-His) were cloned in a modified pcDNA3.1 plasmid. The plasmid contained the OriP sequence to ensure episomal replication during production. HEK-EBNA cells (ATCC, Catalog NO: CRL-10852) were resuspended in RPMI1640 (Biowest, Catalog NO: L0501–500) and transfected with the previously mentioned plasmids using PEI and incubated at 37°C, 150 rpm, 80% humidity, 5% CO₂ for 4 to 5 hours. The culture was then fed with the same volume of Ex-cell 293 media (Sigma Aldrich, Catalog NO: 24571C-50 L) and incubated for 5 days at 37°C, 150 rpm, 80% humidity and 5% CO₂. Cell cultures were then centrifuged to collect the supernatants that were then 0.2-μm filtered and pH-adjusted at pH 7.4. Ni Sepharose Excel beads (GE Healthcare, Catalog NO: GE173711201) were added to the supernatants and incubated overnight (ON) at 4°C, under agitation. Purification was done by gravity flow with the following steps: wash with 1X PBS, pH 7.4; second wash with 1X PBS, 20 mm imidazole (Sigma Aldrich, Catalog NO: I5513), pH 7.4; elution was performed by either a single step elution with 500 mm imidazole or a step elution approach (40 mm imidazole, 80 mm imidazole, 250 mm imidazole and 500 mm imidazole). Selected elution fractions were pooled and dialyzed against 1X PBS, pH 7.4 at 4°C. Quality and functionality of the purified proteins were assessed by SE-HLPC, SPR and endotoxin levels were measured. When required, human CD47-Avi-His was biotinylated using the BirA biotin-protein ligase reaction kit (Avidity, catalog NO: Bulk BirA).

Synthetic common light chain library selection on CD47 and CD38

The panning strategy against CD47 consisted of two rounds (round 1 and 2) of selection using biotinylated human CD47-Avi-His, followed by 2 rounds (round 3 and 4) using CHO cyno CD47. For the first round, purified phage particles from VH3–53 sub-library (10¹² plaque-forming units) and magnetic BEAT® BEAT® Streptavidin C1 beads (Invitrogen, catalog NO: 65002) were blocked with PBS containing 3% (w/v) skimmed milk (3% MPBS) for 1 h at room temperature (RT). Phages were deselected against pre-blocked beads for 1 h at RT. Deselected phages were incubated with 100 nM of biotinylated human CD47-Avi-His for 2 h at RT. Antigen bound phages were captured on streptavidin beads for 30 min at RT and beads were washed five times with PBS containing 0.1% (v/v) Tween 20 (PBS-Tween 0.1%) and twice with PBS. Phages were eluted with 100 mm triethylamine for 10 min at RT and neutralized using Tris-HCl 1 M pH 8. Eluted phages were used to infect 10 ml of exponentially growing E. coli TG1 cells. Infected cells were grown in 2YT medium for 1 h at 37°C and 100 rotation per minute (RPM), then spread on 2YTAG (2TY medium supplemented with 100 μg mL⁻¹ ampicillin and 2% glucose) agar plates and incubated ON at 30°C. Colonies were scrapped off the plates into 10 ml of 2YT and 15% glycerol (v/v) was added for storage at −80°C. TG1 cells from glycerol stocks were grown at 37°C and 240 RPM in 2YTAG medium until the optical density (OD) at 600 nm reached 0.5. Cells were then superinfected with the M13K07 helper phage using a multiplicity of infection of 10 for 1 h at 37°C and 100 RPM. Culture medium was then changed for 2YT medium supplemented with 100 μg mL⁻¹ ampicillin and 50 μg mL⁻¹ kanamycin and cells were further cultured ON at 30°C and 280 RPM. The next day, 10 μL of phage containing cell-free supernatants were used for the second round of selection in the same experimental setup as the first round. For the third and fourth round, 10 μL of phage supernatant were blocked with PBS containing 3% (w/v) bovine serum albumin (PBS/BSA 3%) for 1 h at RT. 2 × 10⁷ non-transfected CHO cells (Invitrogen, A1136401) and 2 × 10⁷ CHO cyno CD47 were blocked with PBS/BSA 3% supplemented with 0.1% azide to avoid receptor internalization for 1 h at RT. Phage were deselected against non-transfected cells for 1 h at RT. The deselected phages were then incubated with the transfected cells for 2 h at RT. To remove nonspecific phages, cells were washed four times with PBS-Tween 0.1% and twice with PBS. Phages were eluted with citric acid 76 mm, pH 2.0 for 10 min at RT and neutralized using Tris-HCl 1 M pH 8. Eluted phages were used to infect 10 mL of exponentially growing E. coli TG1 cells. For the third round, phage amplification was performed as described above. For the last round, the process ended with bacteria storage at −80°C.

The selections against CD38 were performed as described above. A first panning strategy consisted of two rounds of selection using biotinylated human CD38-Avi-His (Acrobiosystems, catalog NO: CD8-H82E7) followed by two rounds using human CD38 (accession number: P28907, residues 1–300) stably transfected CHO cells. A second panning strategy consisted of two rounds of selection using the biotinylated human CD38-His-Avi followed by two rounds using CHO cyno CD38. A pool of VH1–69, VH3–23, and VH3–15 cLC sub-libraries was used for the first panning strategy. VH1–69 sub-library was used for the second panning strategy.

Anti-CD47 and anti-CD38 scFv screening

The binding of scFv clones to CHO human CD47 and to CHO human CD38 was assessed by flow cytometry. Briefly, scFv containing periplasmic extracts from individual E. coli colonies were incubated with CD47- or CD38 expressing and non-transfected CHO cells and bound scFv were revealed with a biotinylated anti-c-Myc antibody (Gallus Immunotech, catalog NO: ACMYC-B), followed by incubation with streptavidin APC (eBioscience, catalog NO: 17–4317). Cell fluorescence was then measured using a FACSCalibur flow cytometer (BD biosciences).

SPR analysis was used to confirm specific binding activity of the scFv clones. Measurements were performed on a Biacore 2000 instrument (Cytiva) using the Biacore 2000 Control Software (v3.2) at RT and analyzed with the Biacore T200 Evaluation Software (v3.1). Human CD47-Avi-His, cyno CD47-Avi-His, human CD38-His or cyno CD38-His) proteins were individually immobilized on CM5 Sensor Chips (Cytiva, catalog NO: BR100012) using an amine coupling kit (Cytiva, catalog NO: BR100050). Filtered periplasmic extracts were injected directly on the covalently coupled CM5 Sensor Chip. Samples were injected on the flow-path 1, 2, 3 and 4 (flow-path 1 being used as reference) at a 30 μL min⁻¹ flow rate for 3 min, followed by a dissociation time of 3 min in running buffer. After each binding event, the surface was regenerated with 10 mm Glycine pH 1.5 solution (Cytiva, catalog NO: BR100354) injected for 1 min at 30 μL min⁻¹. Each measurement included zero-concentration samples as well as irrelevant scFv periplasmic extracts for referencing and specificity, respectively.

Competition assays by BLI

Competition of antibodies and epitope binning of Fabs was assessed using BLI. Measurements were done on an OctetRED96e instrument (Sartorius) and analyzed using the Data Analyis HT version 11.1 software (Sartorius). Biotinylated human CD38-Avi-His protein (Acrobiosystems, catalog NO: CD8-H82E7) was loaded at 1 μg mL⁻¹ in kinetic buffer (Sartorius, catalog NO: 18–1105) on a streptavidin SA Biosensor (Sartorius, catalog NO: 15–5019) for 5 min. Streptavidin biosensor coated with biotinylated human CD38-Avi-His was dipped into a solution of 200 nM of antibody 1 (saturating antibody) for 10 min, followed by a successive dip into a mixed solution of 200 nM of antibody 1 and 200 nM of antibody 2 (competing antibody) for 5 min. Saturation of the CD38 sensor surface was verified by dipping antibody-saturated CD38 sensor surface into a solution of the same antibody at 400 nM. Fresh streptavidin biosensors were coated with biotinylated human CD38 before each cycle. Data was analyzed using Octet HT 11.1 software and user-defined thresholds on percentage of binding of competing antibody to CD38 relative to maximum binding in the absence of saturating antibody were applied to classify antibodies as either competing (<30%), partially competing (≥30% ≤ 50%), or non-competing (>50%) pairs.

ISB 420-E2 and ISB 420-B6 affinity maturation library generation and selection

Affinity maturation phage display libraries were generated by introducing diversity in the CDRs of the heavy chain. HCDR1, HCDR2 and HCDR3 were randomized using degenerate NNK codon oligonucleotides (wherein N is any of the four deoxyribonucleotides and K is G or T) at Kabat residues 27–35, 50–58, 95–101 minus 2, respectively. Each library was generated using a pool of overlapping oligonucleotides containing five consecutive degenerate codons. CDR-H1 and CDR-H2 were also diversified using the trimer oligonucleotides described above. The resulting five library PCR products were cloned into the pNGLEN and the individual ligation reactions were electroporated into E. coli TG1 cells. Phages were prepared as described above. Phage display panning consisted of three rounds using the biotinylated human CD38-Avi-His as described above with the following modifications. Phages were incubated with 50 nM, 5 nM and 0.5 nM of antigen for round 1, round 2 and round 3, respectively. After 1 h incubation, 1 µM of non-biotinylated recombinant homo sapiens (hs) CD38-His was added for 3 h during rounds 2 and 3.

Affinity measurements of anti-CD38 fabs by SPR

SPR analysis was used to measure the association and dissociation rate constants for the binding kinetics of anti-CD38 Fabs to human CD38 and cynomolgus monkey CD38 proteins. The binding kinetics were measured at 25°C on a Biacore T200 instrument (Cytiva). Biotinylated human CD38-Avi-His (Acrobiosystems, catalog NO: CD8-H82E7) was immobilized on Series S BiotinCAPture Chips (Cytiva, catalog NO: 28920234) and cyno CD38-His was immobilized on Series S CM5 sensor chips (Biacore, Cytiva, catalog NO: BR100530) previously coated with anti-histidine antibody (Cytiva, catalog NO: 28995056). Fabs were then injected in single cycle kinetics at different concentrations ranging from 3.9 to 2000 nM (parental Fabs) or from 0.4 to 100 nM (optimized Fabs) in 1:4 dilution series. Experimental data were processed using the 1:1 Langmuir kinetic fitting model on double reference subtracted sensorgrams (reference surface and buffer injection subtractions).

Cloning, expression and purification of human CD38 antigens for crystallization

cDNAs encoding the wild-type hsCD38 ECD, amino acid residues Arg45–Ile300 of Uniprot entry P28907), and the deglycosylated quadruple mutant (hsCD38 4N/X) containing the following mutations on the four N-linked glycosylation sites (N100D, N164A, N209D and N219D), fused to a C-terminal 8-histidine peptide tag were synthesized and cloned into pcDNA3.1 derived vectors (Invitrogen). The expression vector also carries a CMV promoter, a bovine hormone poly-adenylation (poly(A)), the origin of plasmid replication of Epstein-Barr virus (oriP), and the murine VJ2C leader peptide for secretion of the encoded polypeptide chain.

Antigens were expressed in Expi293F^TM cells (ThermoFisher) according to manufacturer instructions. Five days post-transfection, cell-free culture supernatant containing the recombinant protein were prepared by centrifugation followed by filtration and used for further purification.

For antigen purification, the cleared supernatant containing the protein of interest was loaded on a 5 mL HisTrap^TM Excel column (Cytiva) equilibrated with PBS pH 7.4 at 10 mL/min using an Äkta^TM pure FPLC system (Cytiva). The column was then washed with 7 column volumes of PBS pH 7.4 at 5 mL min⁻¹ and the protein eluted with a 4-step gradient with 5%, 20%, 50%, and 100% of elution buffer composed of PBS pH 7.4 and 500 mm Imidazole, each step running over 7 column volumes. Peak fractions were pooled and concentrated using a Centricon^Ⓡ centrifugal filter device (Merck Millipore) equipped with a 10 kDa cutoff membrane. The concentrated sample was further purified by SEC using a HiLoad Superdex-200 26/600 (Cytiva) run at 2.8 mL/min and PBS pH 7.4.

Generation of anti-CD38 B6-D9 and anti-CD38 E2-RecA fabs

For expression of the anti-CD38 B6-D9 and anti-CD38 E2-RecA Fabs, cDNAs encoding the different antibody constant regions were gene synthetized by Geneart AG (Regensburg, Germany) and modified using standard molecular biology techniques. PCR products were digested with appropriate DNA restriction enzymes, purified, and ligated in modified pcDNA3.1 plasmids (Invitrogen). For reformatting scFv library clones into human IgG1 Fab fragments, each scFv clone in its phage library vector was used to amplify its individual variable heavy (VH) cDNAs by PCR, next the VH PCR product was cloned in the modified pcDNA 3.1 vector, described above, upstream of a cDNA encoding a human IgG1 heavy chain CH1 domain; whereas the fixed Vκ3–15/Jκ1 light chain was cloned in the modified pcDNA 3.1 vector described above upstream of a cDNA encoding a human kappa constant light chain domain.

For Fab expression, equal quantities of heavy chain and light chain vectors were co-transfected into Expi293F^TM cells (ThermoFisher). The expression was then performed as described previously for hsCD38 ECD. Five days post-transfection, cell-free culture supernatant containing the recombinant protein were prepared by centrifugation followed by filtration and used for further purification.

For purification, the cleared supernatant containing the protein of interest was loaded on a CaptureSelect™ CH1-XL (Thermo Scientific™) column equilibrated with PBS pH 7.4. Fab was then eluted with glycine 0.1 M pH 3.5. After neutralization with 1/10 volume of Tris-HCl pH 8.0, the eluate was further purified by SEC using a HiLoad Superdex-200 16/600 (Cytiva) equilibrated in PBS pH 7.4.

Crystallization and structure determination of the B6-D9/CD38 complex

For crystallization, the complex of hsCD38 with the anti-CD38 B6-D9 Fab was formed by mixing the receptor with a 1.3-fold excess of Fab followed by purification by SEC using a HiLoad Superdex-200 16/600 (Cytiva) equilibrated in 25 mm HEPES pH 7.4, 140 mm NaCl. The Fab/hsCD38 complex was concentrated to ~10 mg mL⁻¹. The crystals used for data collection were grown by the sitting drop vapor diffusion method with a reservoir solution containing 7% v/v T-mate pH 7.0, 0.1 M HEPES pH 7.2, 20% v/v PEG Smear Medium (BCS screen, Molecular Dimensions). Drops consisting of 1 μL protein solution +1 μL precipitant were set up at 20°C, and crystals appeared within 3–7 days. The resulting crystals were cryoprotected by soaking in well solution supplemented with 35% ethylene glycol, then flash cooled and stored in liquid nitrogen until data collection.

Diffraction data were collected at the Swiss Light Source (SLS, Villigen, Switzerland). The data were indexed in P4₃2₁2. Diffraction images were processed isotropically and anisotropically using the autoPROC program suite.⁵⁹ Anisotropically treated data improved the quality of the electron-density maps compared with the isotropically treated data. Therefore, the data treated with Staraniso in autoPROC were used to build the final model, extending the diffraction limit of the data set to 1.84 Å, and revealing the detailed binding interface. The overall resolution for isotropic scaling of data is 2.00 Å (I/sigI ≥1.2 in last shell from 2.03–2.00 Å). The phase information necessary to determine and analyze the structure was obtained by molecular replacement using Molrep from the CCP4 suite. A previously solved structure of Fab was used as a search model and one copy of the anti-CD38 B6-D9Fab/hsCD38 complex were found in the asymmetric unit. The model comprises residues Gln1 to Ser223 of the heavy chain, Glu1 to Cys214 of the light chain, and Gln48 to Cys296 of hsCD38. Subsequent model building and refinement was performed according to standard protocols with COOT and the software package CCP4, respectively. The initial model was refined using REFMAC5.⁶⁰ Between rounds of refinement, the model was built and adjusted using Coot.⁶¹ The water model was built with the “Find waters”-algorithm of COOT by putting water molecules in peaks of the Fo-Fc map contoured at 3.0 σ followed by refinement with REFMAC5 and checking all waters with the validation tool of COOT. The criteria for the list of suspicious waters were: B-factor greater 80 Ų, 2F_o-F_c map less than 1.2 σ, distance to the closest contact less than 2.3 Å or more than 3.5 Å. The suspicious water molecules and those in the ligand binding site (distance to ligand less than 10 Å) were checked manually. Final refinement statistics are summarized in Supplementary Table 4.

The structure was analyzed with the Protein Interfaces, Surfaces, and Assemblies (PISA) program. All figures were made with PyMol (DeLano Scientific).

Crystallization and structure determination of the E2-RecA/CD38 complex

For crystallization, the complex of hsCD38 with the anti-CD38 E2-RecA Fab was formed by mixing the receptor with a 1.3-fold excess of Fab followed by purification by SEC using a HiLoad Superdex-200 16/600 (Cytiva) equilibrated in 25 mm HEPES pH 7.4, 140 mm NaCl. The Fab/hsCD38 complex was concentrated to ~10 mg mL⁻¹. The crystals used for data collection were grown by the sitting drop vapor diffusion method with a reservoir solution containing 70% v/v MPD, 0.1 M HEPES, pH 7.5. Drops consisting of 1 μL protein solution +1 μL precipitant were set up at 20°C, and crystals appeared within 3–7 days. The resulting crystals were cryoprotected by soaking in well solution supplemented with 35% ethylene glycol, then flash cooled and stored in liquid nitrogen until data collection.

Diffraction data were collected at the SLS. The data were indexed in C2 and integrated and scaled using XDS and XSCALE⁶² to 2.7 Å resolution. The structure was determined by molecular replacement with Phaser. The Fab and CD38 from structures with PDB code 7DUN and 7DHA, respectively, were used as the initial search model and one copy of the anti-CD38 E2-RecA Fab/hsCD38 complex was found in the asymmetric unit. Rigid body refinement, simulated annealing and restrained refinement were carried out in Refmac5.⁶³ Between rounds of refinement, the model was built and adjusted using Coot.⁶¹ Final refinement statistics are summarized in Supplementary Table 4.

As previously, the structure was analyzed with the PISA program. All figures were made with PyMol2.5 (Schrödinger, LLC).

Electron microscopy and image processing

To investigate the formation of the ISB 1442/CD38 complex, negative stain EM experiments were conducted. A total of 4 μL of the sample was applied onto a Formvar carbon film 400 mesh copper grid, which had been pre-treated with glow discharge. After application, the grid was immediately blotted with filter paper and stained with 2% uranyl acetate. Excess stain was removed by blotting, and the grids were air-dried before imaging. Imaging was performed using a Tecnai T120 microscope operated at 120 keV, equipped with a 4K x 4K TemCam-F416 camera, with a pixel size of 1.68 Å. Data was acquired with SerialEM. The negative stain data was processed using Relion 3.0, which involved particle picking with the DoG picker, reference-free 2D classification, and 3D refinement with 17,000 particles. To create a composite model of the complex, UCSF Chimera to fit the structures from PDB 7DHA and 1GIG into the 3D volume.

Molecular dynamics simulations

All steps of protein preparation, system building, and MD simulations and analysis was done using Maestro Version 13.1.141 (Release 2022–1)⁶⁴ Bioluminate.^65–67

The proprietary X-ray structures of B6-D9 and E2-RecA Fabs in complex with CD38 were prepared using the Protein Preparation Workflow.⁶⁸ All non-relevant water molecules (not interacting with both the Fab and the target) and other molecules were removed. The system was then minimized before further steps. The system was prepared using Schrödinger System Builder (Desmond v6.7⁶⁹ – Force-Field OPLS4.^70–74 The molecules to simulate were enclosed in a cubic water box (using SPC as water model) with a 10 Å buffer (minimized). Counter-Ions were added to neutralize the overall charge of the system. Salt was added (concentration 0.15 M, positive ion Na⁺, negative ion Cl⁻) to reflect physiological conditions, OPLS4 was used as force-field.

MD simulations were run using Desmond v6.7⁶⁹ and OPLS4^70–74 as forcefield. Relaxation protocol and other parameters were kept with the standard values. For all systems, the target simulation time was set to 300 ns (generating 1000 frames). The E2-RecA/CD38 complex simulation was extended till 450 ns, to have an equal MD production time. To assess convergence of the simulations, the RMSD was calculated. In case of non-convergence, the simulations were extended. For analysis of the MD simulations the following Schrödinger scripts were used: analyze_simulation.py (RMSD), trajectory_bfactors.py (RMSF), analyze_trajectory_ppi.py (protein-protein interaction network), trj_cluster.py (selection of the 5 most representative structures from the MD). Lastly, Prime MM-GBSA⁴⁶ was used to calculate the individual forcefield energy contributions of each antibody residue toward binding. The values were then projected on the structure using PyMol.

ISB 1442 homology modeling

For all model building and refinement, Schrödinger release 2023–1 was used.⁷⁵ Water and other crystallization molecules were stripped from the B6-D9 and E2-RecA Fabs taken from the corresponding X-ray structures. After protein preparation and minimization, the structures were used for further modeling. Both B6-D9/CD38 and E2-RecA/CD38 protein complexes were prepared as described above and superimposed. The linker between the B6-D9 carboxyl and E2-RecA amino terminus was modeled de novo using the crosslink protein panel (Prime^44,45 from Schrödinger with the arbitrary number of the Gly-Gly-Gly-Ser monomers sufficient to bridge the two Fabs, followed by energy calculation in implicit solvent. The H2 binder was modeled using the antibody modeling panel from Maestro.^64–67 For the heavy and light chain frameworks PDB ID 5DR5 was used as template. The CDR loops were modeled using the following templates: CDR-L1 PDB ID: 2VXV,⁷⁶ CDR-L2 PDB ID: 5F9O,⁷⁷ CDR-L3: 3D9A,⁷⁸ CDR-H1 PDB ID: 3HI6,⁷⁹ CDR-H2 PDB ID: 7BQ5,⁸⁰ CDR-H3 PDB ID: 6O41.⁸¹

After the model was generated, it was submitted to protein preparation and minimization before further use. To model the CD38 transmembrane region, the AlphaFold model^48,49 (ID: A0A852MF64) was superimposed to the CD38 from the E2-RecA and B6-D9 crystal structures and used as template. To build the ISB 1442 full molecule model, a model of the A chain with the H2 binder and one of the B chain with the E2-RecA and B6-D9 with the full-length CD38 was created. As a template for the Fc and the hinge region, as well for orienting the A and B arms, PDB entry 1HZH⁸² was used as template. The model of the H2 binder was superimposed to the Fab in one of the 1HZH arms and then merged to the model, thus generating the A arm. The model for the E2-RecA was then superimposed on the other arm, generating a bispecific antibody with the H2 binder on the A arm and E2-RecA on the B arm. Following this step, the B6-D9 was manually added in the same entry, using the transmembrane region of CD38 to orient it in a way that would be compatible to have the two binders bind to two CD38 molecules on the same cellular surface. This was done setting the dihedral between the Cα carbon of residues 22 and 44 of one CD38 molecule and 44 and 22 on the second set to 0. The terminus of E2-RecA and B6-D9 that needed to be linked were positioned at a distance compatible with the (G₄S)₃ size (between 30 and 40 Å). The two units where then linked using the crosslink protein panel (Prime^44,45 from Schrödinger. The settings were set as default, with the linker set as exactly 3 repetitions of Gly-Gly-Gly-Gly-Ser motif, the look conformation prediction using “Loop lookup from curated PDB” option and energy calculation in implicit solvent model in Prime. After checking that no major structure distortions occurred after linker introduction, the system was re-prepared and minimized. To add the other ISB 1442 and CD38 molecules, the structures of E2-RecA and B6-D9 were superimposed (using CD38 as reference) to the previously generated model. In particular, the previous model was duplicated and the angle between the different Fabs was changed to allow the model to accommodate multiple instances of the antibody without steric clashes. Following the movement of the Fabs, the residues of the linkers were selected to perform a rapid minimization to reproduce a more natural conformation. A model of the membrane was created, using the residues 22–42 of all the CD38 units present in the model to orient the membrane. The membrane composition was set as POPC3 and the membrane was incorporated using the System builder,⁶⁹ selecting “None” as water model.

SPR binding at different CD38 surface densities

SPR analysis was used to measure the binding response, as well as the association and dissociation rate constants for the binding kinetics of ISB 1442, anti-CD47-BTA_anti-CD38 B6-D9-anti-CD38 B6-D9-BTB BEAT^Ⓡ (CD47_CD38–1 + 1), anti-CD47-BTA_anti-CD38 E2-RecA-anti-CD38 E2-RecA-BTB BEAT^Ⓡ (CD47_CD38–2 + 2), anti-CD47-BTA_anti-CD38 B6-D9-Dummy-BTB BEAT^Ⓡ (CD47_CD38-1_DU) and anti-CD47-BTA_Dummy-anti-CD38 E2-RecA-BTB BEAT^Ⓡ (CD47_CD38-2_DU) to recombinant human CD38 proteins immobilized at different surface densities. The binding kinetics were measured at 25°C on a Biacore 8K+ instrument (Cytiva). Biotinylated human CD38-Avi-His (Acrobiosystems, catalog NO: CD8-H82E7) was diluted at 3.5 nM, 14 nM, 56 nM and 224 nM and injected on flow path 2 of a Series S BiotinCAPture Chip (Cytiva, catalog NO: 28920234) for 30 s, resulting in immobilization responses of around 10 RU, 30 RU, 100 RU and 300 RU, respectively. Considering 1 RU as 1 pg mm⁻² bound to the surface,^83,84 surface densities roughly correspond to 2e5, 6e5, 2e6 and 6e6 CD38 molecules per μm², respectively. Antibodies were then injected in single cycle kinetics at different concentrations ranging from 0.41 to 100 nM in 1/3 dilution series. Experimental data were processed using the 1:1 binding kinetic fitting model on double reference subtracted sensorgrams (reference surface and buffer injection subtractions) to estimate affinities or avidity-induced apparent affinities for CD38, kinetic parameters of binding (ka, kd) and Rmax. Comparison of binding responses was performed by plotting relative response (RU) at end of association versus analyte concentration (nM). To compensate for the difference between the molecular weight of 2 + 1 molecules (around 192 kDa) and 1 + 1 molecules (around 145 kDa), binding responses of 1 + 1 molecules were multiplied by a factor of 1.32 before plotting. Theoretical Rmax was calculated using following formula: $\mathrm{Rmaxtheo}=\frac{\mathrm{M}{\mathrm{W}}^{\mathrm{analyte}}}{\mathrm{M}{\mathrm{W}}^{\mathrm{ligand}}}\mathit{\hspace{0.17em}}\cdot \mathrm{RL}\cdot {\mathrm{V}}^{\mathrm{ligand}}$, where MW = molecular weight (in kDa), RL = CD38 capture level (RU) and V = valency (in this case always set to 1 for comparison, independently of interaction mode) and percentage of theoretical Rmax (% Rmax_theo) was used to compare analyte binding capacity of CD38 surfaces. Stability of binding during the dissociation phase was assessed using following formula: $\mathrm{Stability}\mathrm{of}\mathrm{binding}=\frac{\mathrm{Stability}\mathrm{early}-\mathrm{Stability}\mathrm{late}}{\mathrm{Stability}\mathrm{early}}$, where Stability early = relative response (RU) at start of dissociation and Stability late = relative response (RU) at end of dissociation. Fold increase stability $(\mathrm{Fold}\mathrm{increase}\mathrm{stability}=\frac{\mathrm{Stabilit}{\mathrm{y}}^{\mathrm{XRU}}}{\mathrm{Stabilit}{\mathrm{y}}^{10\mathrm{RU}}}$) was reported for comparison of stability increase at different CD38 surface densities.

Cancer cell line staining with antibodies

Tumor cell lines were resuspended in the presence or absence of soluble CD38 and were incubated with increasing concentrations of antibodies or an isotype control at 4°C for 30 min. Cells were washed twice followed by an incubation with an APC-fluorescently labeled monoclonal anti-human IgG secondary antibody (Biolegend, Catalog NO: 366906) and incubated at 4°C for 30 min. Cells were washed twice and resuspended in FACS Buffer (PBS 2.5% FCS 2 mm EDTA and 0.05% NaN3) containing a viability dye (Dapi). Samples were acquired on a Cytoflex cytometer (Beckman Coulter). Data were analyzed using FlowJo software (BD). APC Geometric Mean of Fluorescence Intensities (geoMFI) of viable single cells for each sample were extracted. A normalization was then performed using the geoMFI of each isotype control antibody. The values of geoMFI from the control staining were subtracted from the geoMFI values of each molecule to generate the relative fluorescence intensity (RFI). Data were fitted with a non-linear regression curve (One site Specific Binding) using Graph Pad Prism software and Bmax was used for statistical comparison. For some curves, geoMFI values at the highest concentration were excluded due to the observation of a hook effect. For binding in the presence of sCD38, antibodies were used at the concentration corresponding to their respective EC50 previously measured on NCI-H929 in the absence of sCD38. Molar ratio CD38/antibody at IC50 was calculated by dividing the CD38 concentration at IC50 by the respective antibody concentration. Statistical analysis was conducted in Graph Pad Prism software using RM one-way ANOVA or one-way ANOVA followed by Šidák’s multiple comparisons tests.

Complement-dependent cytotoxicity assay

Tumor cells were labeled with 5 µM calcein AM and plated in 96-well plates with increasing concentration of test antibody and in the presence of 50% human serum for 4 h 30 min at 37°C, 5% CO2. Triton X-100 was used as a positive control for maximum tumor cell killing. After the completion of the assay, cells were centrifugated and fluorescence induced by calcein release was determined using a Synergy Plate reader at 485/515 nm. Specific killing was calculated according to the formula: %Killing = 100* (release of sample – spontaneous release)/(maximum release Triton X100 − spontaneous release).

Funding Statement

The author(s) reported there is no funding associated with the work featured in this article.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Authors contributions Statement

J.L. and T.M. contributed equally to the work. M.R.D. and C.D. contributed equally to the work. J.L., T.M., M.M., F.B., J.S., A.G., C. L. N., M.B., S.C., S.D., C.G. S.S, A.S., M.P., E.A.Z., M.R.D. and C.D. contributed to experimental design, data acquisition, interpretation, and analysis. J.L., T.M., S.B., M.L.M., S.S, A.S., M.P., E.A.Z., M.R.D. and C.D. contributed to project and resources administration and supervision. J.L., T.M., M.M., F.B., S.S, M.P., E.A.Z., M.R.D. and C.D wrote, reviewed, and edited the manuscript.

Data availability statement

Coordinates and structure factors are deposited in the Protein Data Bank (PDB code 9GOX for B6-D9 Fab/CD38 and 9GOY for E2-RecA Fab/CD38).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2025.2457471

Abbreviations

ADCC

Antibody-dependent cell-mediated cytotoxicity

ADCP

Antibody-dependent cellular phagocytosis

ADP

Adenosine diphosphate

AML

Acute myeloid leukemia

BCMA

B-cell maturation antigen

BEAT

Bispecific Engagement by Antibodies based on the T cell receptor

BLI

Bio-layer interferometry

BSA

Buried surface area

bsAbs

Bispecific antibodies

BsBpAb

Bispecific biparatopic antibody

CAR-T

Chimeric antigen receptor T cells

CD37

Cluster of differentiation 37

CD38

Cluster of differentiation 38

CD47

Cluster of differentiation 47

CDC

Complement-dependent cytotoxicity

cLC

Common light chain

CXCR4

C-X-C motif chemokine receptor 4

DLBCL

Diffuse large B cell lymphoma

DU

Dummy

ECD

Extracellular domain

EM

Cryo-electron microscopy

Fab

Antigen binding fragment

FcγR

Fcγ receptor

FR-alpha

Folate receptor alpha

GPRC5D

G protein – coupled receptor class C group 5 member D

hcAb

Heavy chain antibody

HCDR

Heavy chain complementarity-determining region

HER2

Human epidermal growth factor receptor 2

HIV

Human immunodeficiency virus

LCDR

Light chain complementarity-determining region

mAbs

Monoclonal antibodies

MD

Molecular dynamic

Met

Mesenchymal Epithelial Transition

MM

Multiple myeloma

MM-GBSA

Molecular mechanics with generalized Boltzman and surface area continuum solvation

NAD+

Nicotinamide adenine dinucleotide

NGS

Next-generation sequencing

PDB

Protein data bank

RBCs

Red blood cells

SARS-Cov2

Severe acute respiratory syndrome coronavirus 2

scFv

Single-chain variable fragments

SEC-MALS

Size-exclusion chromatography with multi-angle light scattering

SIRP

Signal-regulatory protein

SPR

Surface plasmon resonance

TAA

Tumor-associated antigen

TME

Tumor microenvironment

VHH

Single variable domain on a heavy chain