Mutation design
We selected IgG4 S228P as the development target (PDB: 5DK3). After analyzing the crystal structure of CH1-CL interface, we found that there were two sites where the interaction force was relatively weak. Sites 1 and 2 have hydrophilic amino acids that weaken the interaction between the two chains. As shown in Fig. 1A, these two sites are located within the interface, not completely exposed and could be modified without affecting the function. Since TCR α/β constant domains have two major hydrophobic areas (Figure S1), they were introduced to CH3, along with some site mutations, leading to heterodimerization of heavy chains23. We also grafted these two regions to the two weak interaction sites of CH1-CL, respectively, in order to obtain correct pairs of heavy and light chains, while leaving strong amino acid interaction in the CH1-CL interface unchanged.
It was found that the weak charge interaction of site 1 occurs among Asn26, Asn27, Ser85.1 of CL and His79 of CH1 (Fig. 1B). We used one of the two major hydrophobic regions of TCR α/β constant domain interface, i.e., Val86, Leu7 and Val22 in Cα and Val22 in Cβ23 to enhance the weak interaction. Thus, His79 was replaced by Val in CH1 and Asn27, Asn26 and Ser85.1 by Leu, Val and Val in CL, respectively (Fig. 1C). These mutations constructed a new hydrophobic region in the edge of CH1-CL. At the other end of which, the second largest hydrophobic region of TCR was introduced to include residues Trp88 and Tyr79 in Cα as well as Leu24 and Leu84 in TCR23. Similarly, Ser20 in CL was mutated to Trp, which has a large side chain, and Gln13 was mutated to Tyr. Within CH1, basic residue Lys26 and polar residue Gln84.2 were both mutated to non-polar amino acid Leu (Fig. 1C). With the addition of small non-polar residues nearby, another new large hydrophobic region was formed. After analyzing the entire modified CH1-CL structure, two large hydrophobic interaction areas of TCR Cα-Cβ were confirmed to present at both ends of the β sheet within CH1-CL interface. These modifications should in theory strengthen the acting force between CH1 and CL. Moreover, side chains of CH1 (Pro82 and Leu84.1) and CL (Ser81 and Glu79) will form respective polar bonds to stabilize the entire structure.
To prevent mismatch of heavy and light chains, the native disulfide bond in CH1-CL was also mutated (Fig. 1D), and the corresponding Cys was substituted by Val24. All amino acids of CH1-CL that were subjected to mutation are shown in Fig. 1E.
Heterodimer verification
In human IgG-like bispecific antibodies, heavy and light chain mismatch is generated because CH1-CL has two arms that form heterodimers with identical sequences when co-expressed in a single cell (Fig. 2A). To examine if the newly designed molecules could be assembled correctly, we studied all possible mismatches in the molecular format of one value M (Fig. 2B). The constructed plasmids bearing single-strand antibody coding were co-expressed and the results showed that mutated CL and CH1 could not be assembled with that of wild-type (Figure S2), while mutated CH1 and CL were able to assemble correctly (Fig. 2C) with an abundance close to 95%. The product after protein A purification displayed a relatively single peak and the retention time is consistent with that of the theoretical value (Fig. 2E). One previous study reported that native disulfide bonds replaced by a pair of non-native disulfide bonds can mitigated the problem of mismatch26. We repeated this design but found substitution of disulfide bonds did not significantly improve the expression and correct paring of our bispecific antibody (Figure S3).
Production and validation
Programed death-1 (PD-1) is an inhibitory receptor expressed on the surface of T cells. Inhibitory signal caused by PD-1 binding with its ligands PD-L1 and PD-L2 reduces T cell proliferation, cytokine secretion and cytotoxic activity25,26. In multiple syngeneic mouse tumor models, blockade of PD-1 or its ligands promoted the antitumor activity27–29, which could be further enhanced by antibodies against other negative regulators of T-cells, such as CTLA-4 and LAG330,31. To produce a bispecific antibody targeting both PD-1 and LAG-3, we selected IgG4 S228P32 to make the construct.
The PD-1×LAG3 bispecific antibody (PD-1×LAG3 TiMab) was produced by transient co-expression of 4 plasmids in Expi293 cells (Fig. 3A). The molar ratio of heavy and light chains was 1:2. Construction of plasmids encoding heavy and light chain genes is described in Materials and Methods. The PD-1×LAG3 TiMab yield was 30 mg/L (Fig. 3B) while that of PD-1 and LAG3 monospecific antibodies were 120 mg/L and 95 mg/L, respectively. Obviously, the expression level of the bispecific antibody is lower than its parental IgGs. PD-1×LAG3 TiMab was purified using standard protein A affinity chromatography. Sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and size exclusion chromatography were employed to analyze the purity and chain composition. Purified PD-1×LAG3 TiMab showed a single peak with very low levels of aggregates in SEC analysis. Fragments of free chains were also insignificant (Fig. 3C). The peak retention was similar to its parental molecules due to identical molecular mass of IgGs.
We subsequently carried out mass spectrometry to analyze whether the amino acids in the modified CH1-CL of PD-1×LAG3 TiMab were successfully mutated. Analysis of the different enzyme digested antibody demonstrated that the target residues in CH1 and CL were all mutated according to our design, e.g., Glu (Q) (Fig. 3D) in CH1 was changed to Leu (L) (Figure. 3E; other mutation sites were not shown).
Binding property
Antibody binding affinity to cell surface expressing antigens PD-1 and LAG3 was assessed by flow cytometry (FACS). PD-1×LAG3 TiMab exhibited similar binding affinity as the parental monospecific antibodies (Fig. 4A and 4B). It binds to CHO cells expressing PD-1 and LAG3 at 2.76 nM and 3.37 nM, respectively (Table. S1). The dual binding ability of PD-1×LAG3 TiMab was demonstrated in engineered cells that can bind to one arm of the antibody while the other arm is open for detection. When PD-1×LAG3 TiMab was saturated with PD-1 in the engineered cells, addition of LAG3 elicited a second binding signal. By calculating the stoichiometry of the binding events, we determined that PD-1×LAG3 TiMab is capable of binding both antigens simultaneously (Fig. 4C), suggesting correct Fab folding in the presence of mutated CH1-CL interface.
We then investigated whether mutations in CH1-CL interface affects Fc-mediated functional activity. Binding affinities of PD-1×LAG3 TiMab to various human Fcγ receptors, FcRn and C1q were determined by a steady-state equilibrium binding assay on Proteon. As shown in Table 1, the binding kinetics of the antibody to different Fcγ receptors and C1q are indistinguishable from the two parental antibodies and an isotype control of human IgG4 S228P. The interaction with human FcRn at pH 6 was also shown in Table S2. These data imply that Fc function is maintained following the mutations.
Thermostability
Thermostability of purified PD-1×LAG3 TiMab was measured at 1 mg/mL and 2 mL of which were incubated at 40°C for a specified period. Protein concentration and purity were checked at different time points. Compared to the control antibody, PD-1×LAG3 TiMab was not degraded significantly (Table S3). After incubation at 40°C for 21 days, it still displayed a similar binding ability as that of the parental (Figure S4A and S4B). The thermostability was also examined with differential scanning fluorimetry (DSF). As shown in Figure S4C, the melting temperature (Tm1) is 62.45°C, lower than the control IgG4 S228P whose Tm1 is 69°C, indicating that PD-1×LAG3 TiMab is relatively stable.
In vitro activity
The bioactivity of PD-1×LAG3 TiMab was compared to that of the monovalent forms of the parental antibodies. In a human T-cell response assay consisted of an allogeneic MLR, stimulation of human PBMC by super-antigenic DC and antigen-specific stimulation of T cells, blockade of PD-1 and LAG3 by PD-1×LAG3 TiMab resulted in a titratable enhancement of IFN-γ release (Fig. 5B). In some donor T-cell/DC pairs, enhanced T-cell proliferation was observed (Fig. 5A). It also enhanced IL-2 secretion in response to DC using PBMC compared to the isotype control (Fig. 5C). Taken together, these data demonstrated that PD-1×LAG3 TiMab can, at a very low concentration, enhance T-cell reactivity in the presence of a TCR stimulus. Specifically, there were significant releases of inflammatory cytokines, including IFN-γ and IL-2, from stimulated PBMC after co-incubation with the antibody.
Additionally, the ability of PD-1×LAG3 TiMab (0.003-50 mg/mL) to mediate ADCC activity in vitro was tested by using IL-2 activated PBMC as a source of natural killer (NK) cells as well as activated human CD4+ T cells expressing high levels of membrane PD-1 and LAG3 as target cells. Compared to a positive control of IgG1 antibody, PD-1×LAG3 was unable to mediate ADCC of T cells at high concentrations (Fig. 5D). It also failed to mediate complement-mediated cytotoxicity (CDC) of activated human CD4+ T cells in the presence of human complement (Fig. 5E).
Pharmacokinetic profile
We used a rat model to study the pharmacokinetic properties of PD-1×LAG3 TiMab by intravenous injection. Its circulating concentrations were determined by measuring that of PD-1 or LAG3 in the animal serum specifically captured by anti-Fc antibodies. In fact that the serum concentration of the bispecific antibody assessed by either antigen was very much alike, indicating that the molecule is intact in vivo and has the ability to bind both antigens. However, it is less stable than conventional IgG at the same intravenous dose (Fig. 6).
The drug clearance rate was faster, with a half-life shorter than 10 days.
This is expected because of the mutations in the CH1 and CL domains. The long half-life observed is consistent with previous observations on heterodimers formed by knob-into-hole method33.
Immunogenicity
We used free online IEDB software to predict the immunogenicity of PD-1×LAG3 TiMab in comparison with Trastuzumab and Pembrolizumab (Fig. 7A) which are non-immunogenic in the clinic. Immunogenicity can lead to the formation of anti-drug-antibody (ADA) immune complexes thereby affecting drug safety and pharmacokinetics. Therefore, ADA test was performed in PD-1×LAG3 TiMab coated plates and incubated with rat serum for 14 and 21 days. Mouse-anti-rat IgG was then added to reveal the results showing that our antibody did not produce significant ADA at both time points (Fig. 7C). Combined with the above prediction, introduction of site-directed mutation didn’t cause observable immunogenicity.