Decoupling stability and release in disulfide bonds with antibody-small molecule conjugates

A novel bioconjugation strategy utilizes a disulfide bond to site-specifically couple a small molecule drug to an antibody, enabling both high circulation stability and quick intracellular release.


Introduction
Utilizing proteins or other polymers to deliver small molecule payloads in vivo is a well-recognized and validated strategy. Many applications have emerged that rely on this approach and these conjugated systems are able to dramatically change the half-life, solubility, and therapeutic index of the small molecule. One of the critical components in these conjugates is the connection between the small molecule and large molecule delivery vehicle. The disulde bond is a bioactivatable connection that has been utilized for reversibly connecting protein toxins, 1 chemotherapeutic drugs, 2-8 and probes 9 to carrier molecules such as antibodies for around 40 years. 10,11 The majority of bioconjugates incorporating a disulde connection have introduced the disulde via a heterobifunctional crosslinker between the drug or probe and lysine (Lys) residues on the protein. Given the number of reactive Lys residues in most proteins, the result is a highly heterogeneous conjugate with a varying number of drugs at a large number of sites (Fig. 1a). 12 This heterogeneity presents challenges for characterization and analysis, but more importantly can reduce activity and increase toxicity. 13,14 The second even bigger challenge when attaching small molecules to proteins through heterobifunctional disulde linkers is that the stability of the conjugate is coupled to the ability to release the payload. In many applications, including antibody-drug conjugates (ADCs), high stability of the disulde in circulation and low stability of the disulde within the target cell are desired. Disuldes are reduced in the cytosol of cells where the concentration of reduced glutathione (GSH) is 1-10 mM whereas cysteine (Cys) is the most abundant reactive thiol in plasma with concentrations between 8-11 mM. 15 While this 1000 fold difference in reactive thiol concentration is the basis for the use of disuldes in drug delivery, the lack of exquisite selectivity for reduction by GSH versus Cys leads to a coupling of stability and release. Unhindered disuldes provide the most facile reduction in the cytosol but the lowest stability in circulation. Increasing the stability of the disulde bond with adjacent alkyl groups increases circulation half-life, but hinders release of the free drug in the target cell. Achieving the desired stability and release is therefore a balancing act. The disulde conjugates that have advanced into human clinic trials have typically had an intermediate level of alkyl substitution around the disulde (1-2 methyl groups on either side of the disulde). 16,17 For example, in maytansine disulde conjugates (Fig. 1a, R ¼ H, R 0 ¼ Me; SPDB-DM4), the drug becomes deconjugated from the antibody in circulation with a half-life of $9 days 18 while a major catabolite in tumors for the rst 4 days is the unreduced disulde. 19 While poorly reducible and even non-cleavable linkers for some payloads can afford efficacious ADCs, they are oen limited to targets with high and homogeneous antigen expression. 20 The approach of balancing stability and release has led to a sacrice in one or both potentially critical attributes.
Described herein is an approach to simultaneously achieve high stability in circulation and fast payload release in a target cell for disulde-linked antibody conjugates. To achieve high stability in circulation we connected a small molecule drug directly via a disulde to the two free thiols of a Cys-engineered antibody at a variety of mutant sites. Target cell internalization and degradation of the antibody by the lysosome then permitted facile cleavage of the disulde bond by cytosolic reductants (e.g. GSH) and release of the drug (Fig. 1b). Our approach has several advantages. Rather than extending the disulde away from the protein as is done with most disulde conjugates through the use of heterobifunctional linkers, the direct approach brings the disulde bond as close as possible to the antibody. By connecting the disulde directly to the antibody we can take advantage of a steric protection resulting from reduced solvent accessibility at some sites in a three-dimensional folded protein. While others have investigated attaching drugs through disulde bonds at different sites in smaller proteins, none of these approaches resulted in in vivo stability sufficient for antibody-mediated delivery and stability remained problematically coupled to release. [21][22][23] This approach also reduces the heterogeneity in the resulting conjugate since drugs are specically attached to the engineered Cys residues. Since only a single bond links the drug to the antibody, cleavage releases unmodied drug and antibody fragment without the need for a linker. 24

Results and discussion
To test the design concept we utilized thiol-containing derivatives of the natural product maytansine. 25 Derivatives DM1 (1) and DM3 (2), differing primarily in the addition of a methyl group next to the disulde, were rst treated with activating agent 3 followed by reaction with the engineered Cys of an anti-CD22 antibody (Scheme 1). We selected antibody site LC-V205C as a starting point as it offered the greatest stability for maleimide-linked conjugates in previous studies. 26 We evaluated the V205C-DM1 conjugate for in vivo stability (Fig. 2a). Interestingly, while we previously demonstrated that conjugates linked to V205C using maleimide chemistry were highly stable with minimal drug loss in vivo up to 21 days, we found that a DM1 conjugate (1a) linked at the same site through a disulde bond was quite unstable.
Within one day, about half of the unhindered drug (DM1, 1) was lost in circulation. Mass spectrometry analysis indicated that a disulde displacement had occurred, resulting in loss of drug and addition of Cys and GSH to the antibody (Fig. S1 †).
Based on these results we sought antibody sites that would generate more stable disulde conjugates. A screen of several mutation sites led ultimately to the identication of LC-K149C as a stable conjugation site for disuldes (manuscript in preparation). When we attached DM1 (1) through a disulde to K149C (1b), only about 10% of the drug was lost aer one day, and even aer seven days, more than 50% of the drug remained attached (versus 56% and 100% loss aer 1 and 7 days, respectively, for LC-V205C, 1a). The stability of unsubstituted disulde conjugate 1b is comparable to that of the Lys-linked disulde conjugate with two neighboring methyl groups (SPDB-DM4). 18 Furthermore, addition of just one methyl group next to the disulde (DM3, 2) resulted in conjugates possessing increased stability at both sites on the antibody with the LC-K149C conjugate (2b) losing only 10% of the drug aer seven days.
We next evaluated anti-tumor effects of anti-CD22 disulde conjugates in a human lymphoma tumor xenogra in mice Scheme 1 Synthesis of site-specific antibody-maytansine conjugates through disulfide activation. Drug-to-antibody ratios (DARs) were uniform, ranging between 1.8 and 2.0. ( Fig. 2b). Conjugates were administered at a single dose of 3 mg kg À1 and tumor volume was measured over time. Consistent with the in vivo stability data, the least stable conjugate, V205C-DM1 (1a), was also the least efficacious, resulting in modest tumor growth delay. Increasing stability either through site (K149C-DM1, 1b) or addition of a methyl group (V205C-DM3, 2a) resulted in complete tumor regression. Furthermore, this activity was target mediated as anti-Her2-K149C-DM3 (2c), the non-target control conjugate, showed no detectable activity and was equivalent to the vehicle control.
Having demonstrated that relatively high stability of the disulde in circulation can be achieved by appropriate choice of conjugation site, we evaluated the effectiveness of drug release in a target cell. It is well established that both Lys-and Cyslinked antibody-drug conjugates are completely catabolized in the lysosome to an amino acid-linker-drug species, wherein the amino acid portion is a remnant of the antibody attachment site. 19,20,28 For Lys linked antibody conjugates the catabolites observed are Lys-SPDB-DM1 (6), -DM3 (7), and -DM4 (8) (Fig. 3). 19,20 For our site-specic Cys disulde conjugates, the expected catabolites are Cys-DM1 (4) and Cys-DM3 (5). Based on previous data showing that endosomes/lysosomes are oxidizing, 9 we posit that disulde metabolites escape the lysosome and are reduced in the cytosol. Thus to compare release of maytansinoid payloads in a target cell we measured stability of the Lys-and Cys-linked catabolites in the presence of a 3.3 fold stoichiometric excess of reductants DTT or GSH ( Table 1).
The number of methyl groups on the carbon atom adjacent to the disulde was the primary determinant for the stability of the disulde bond, with an increasing number resulting in a sterically encumbered disulde more resistant to reduction. 29 Interestingly, within disuldes containing the same substitution, the Cys-catabolites were more readily reduced than Lys-catabolites. This trend is likely a result of the reduced pK a for the thiol in Cys, making it a better leaving group than the alkyl thiols present in the Lys-modied metabolites. These results, as well as those at higher glutathione concentrations (Table S1 †), suggest that drug release inside a target cell is more facile for Cys-linked conjugates than for Lys-linked conjugates.
With an understanding of the impact of disulde stability on efficacy and disulde substitution on chemical reduction, we set out to evaluate the inuence of disulde release on efficacy (Fig. 4). We selected the human lymphoma tumor xenogra BJAB, as this model discriminates between conjugates containing cleavable and non-cleavable linkers. 30 Anti-CD22 conjugates were administered at a single maytansine dose of 50 mg m À2 to allow comparison between Lys-and Cys-linked conjugates with differential drug loads. Conrming the importance of release, the non-cleavable maleimide control MPEO-DM1 (1d) was completely stable at this new site (Fig. S2 †) yet showed minimal activity. Signicantly, while the disulde conjugate K149C-DM1 (1b, Cys-linked) has similar circulation stability to SPDB-DM4 (9a, Lys-linked), it possessed superior efficacy, indicating that the ability to decouple stability from release results in improved activity in vivo. To conrm that this was not a result of antibody site (Cys vs. Lys) or drug loading, we made MBT-DM4 (9b), a Cys-linked conjugate with the same hindered disulde as SPDB-DM4 (9a), and demonstrated that the two conjugates were equally efficacious.
Lastly, we sought to develop Cys-linked disulde chemistry that would enable traceless release of non-thiol drugs (e.g., amines), while retaining high stability in circulation and rapid release of payload (Fig. 5). Traceless, self-immolative disulde linkers have been used to release phosphates, 31 alcohols, 32-34 hydrazides, 35 and amines 36,37 for a variety of imaging and therapeutic applications. 11 Immolation for some of these has been proposed to occur through either cyclization to a 3-membered ring (thiirane or episulde) or cyclization to a 5-membered cyclic thiocarbonate. [38][39][40] We sought to evaluate the immolative disulde linker for release of amines, a chemical functional group exemplied by many small molecule drugs and probes. We found that treatment of an immolative disulde linker-drug (100 mM) with a physiologically relevant concentration of a reductant (Cys, 2 mM) resulted in the production of free drug and thiirane (Fig. S3 †) as demonstrated by LC-MS.
Since immolating disuldes have not previously been connected to an antibody we wanted to establish whether stability was maintained in circulation. One point of concern is the fact that the electron withdrawing nature of the carbamate group lowers the pK a of the thiol and therefore makes it a better leaving group relative to the thiol in the maytansine disuldes described above. Nevertheless, we discovered that conjugates of the two immolating disuldes (9, 10, Fig. 5b) had an in vivo stability prole similar to that of the similarly substituted maytansines described above (Fig. S4 †). Furthermore, the stability of disuldes at site K149C is not antibody specic as ADCs with anti-Her2 have the same stability as those with anti-CD22 (Fig. S5 †).

Conclusions
In summary, we have demonstrated for the rst time, the decoupling of stability and release in a disulde antibody conjugate. This decoupling was accomplished by establishing conjugation chemistry in conjunction with antibody engineering that allowed antibody-mediated stabilization of the disulde while in circulation and rapid deconjugation upon complete antibody catabolism inside a target cell. Our site-specic conjugates are as or more stable than disulde conjugates employing heterobifunctional crosslinkers but generate catabolites that more readily release the payload. Lastly, we extended our work to immolating disulde linkers that combine the advantages of antibody protection, rapid cleavage inside a target cell and release of amine-containing drugs. In addition to providing stabilization to disuldes, we believe that the approach of utilizing an antibody to protect a labile chemical functional group will allow improvement in the targeting, half-life, and metabolic stability of small molecule drugs and probes for therapeutic, diagnostic and imaging applications. Fig. 5 Application of site-specific disulfide conjugates to aminecontaining drugs. (a) The disulfide conjugate is degraded by proteolysis in the lysosome and reduced by glutathione to generate a free thiol that cyclizes to generate thiirane, carbon dioxide, and the aminecontaining drug. (b) Disulfide MMAE antibody conjugates were generated with (10) and without (9) a neighboring methyl group.