Modification of Cysteine-Substituted Antibodies Using Enzymatic Oxidative Coupling Reactions

New protein bioconjugation chemistries are important for the generation of protein-based materials with novel biochemical properties and functions. These conjugates have a range of applications in basic and applied science. For example, biological probes are generated by installing fluorophores at specific sites on proteins without disturbing their cellular localizations and non-covalent interactions.^1,2,3 The immobilization of proteins on surfaces is routinely performed to study their biochemical properties and requires that proteins are properly oriented to preserve interactions with binding partners.^4,5 In the pharmaceutical industry, cytotoxic payloads are installed site-selectively on targeting agents in order to maintain recognition of malignant cell epitopes.^6,7,8 Regardless of the application, site selective modifications are required to preserve protein folding and function. Chemistries that modify proteins must therefore proceed under strict conditions, including the use of aqueous buffers from pH ~6.0 to 8.0, while having sufficiently rapid kinetics to achieve high conversion at temperatures ranging from 4 to 37 °C.

In the field of chemotherapeutics, site-selective modification chemistries are growing in importance for the development of targeted therapeutics. By targeting epitopes upregulated or unique to cancerous cells, these treatments have reduced the toxic side effects that have plagued their untargeted counterparts.⁹ In the last decade, antibody-drug conjugates (ADCs) have emerged as an expanding class of targeted chemotherapeutics. Their clinical potential is highlighted by their rate of FDA approval: 14 are currently on the market, three of which were approved last year alone, with over 100 more in clinical trials.¹⁰ Bioconjugation reactions are an important component in the development of ADCs, providing a vital means for the attachment of cytotoxic payloads to antibodies.

Over the last decades, various chemistries have been developed to attach payloads to antibodies covalently. For example, N-hydroxysuccinimide (NHS)-activated esters undergo rapid amide bond formation with the primary amine of lysine residues present on proteins. This chemistry was used to synthesize Kadcyla, an FDA approved ADC for the treatment of HER2(+) breast cancer tumors.¹¹ Despite the success of this strategy in the clinic, attachment of payload molecules using NHS esters is nonsite-specific and is complicated by the production of heterogenous mixtures of products with differential drug-to-antibody ratios (DAR) and pharmacokinetic properties.¹²

To produce more homogenous ADCs, others have explored the modification of cysteine residues present in antibodies, which can undergo Michael additions with maleimide-containing cargo.^9,13,14 Antibodies contain four to eight native disulfides that connect the light chains and the heavy chains. These can be reduced, producing four to eight reactive thiols for conjugation with maleimide-payload, as was done in the synthesis of Adcetris, an FDA approved ADC for the treatment of Hodgkin lymphoma.¹⁵ Other established chemistries developed for thiol modification include disulfide rebridging reagents, like bissulfones¹⁶, pyridazinediones¹⁷, and next generation maleimides¹⁸. All demonstrate higher homogeneity in ADC populations; however, common challenges still include intrachain misbridging and cysteine scrambling.¹⁹

Alternatively, engineered cysteine residues can be introduced onto the surface of the antibody, producing site-specific thiol handles for reaction (Figure 1a).^20,21 This method also allows for tuneability of the reactive cysteine, as well as the stability and physicochemical properties of the resulting conjugate. While thiol-maleimide chemistry can produce more homogenous ADC products, they suffer from poor stability under physiological conditions.²² The resulting thiosuccinamide linkages have been shown to be reversible, leading to migration of the cargo to other thiol-containing proteins.¹⁶ The result is off-target cytotoxicity that limits the potential applicability of this linking strategy. Others have turned to more stable alternatives to maleimides, including self-hydrolyzing maleimides²³, methylsulfonyl phenyloxadiazoles²⁴, and arylpropiolonitriles²⁵. While these strategies are promising, they often involve nontrivial, multi-step synthetic manipulations to achieve the desired linker payload. This can be especially challenging when varying payload attachments. As such, we sought to develop a thiolreactive chemistry that delivered stable conjugates, was synthetically convenient and versatile for the synthesis of ADCs.

Recently we have reported an enzymatic method for the attachment of phenolic cargo to free thiols present on proteins using the enzyme, tyrosinase.²⁶ In this strategy, tyrosinase uses molecular oxygen to oxidize phenols present in small molecules or proteins to highly reactive o-quinone intermediates, producing water as the sole byproduct. These o-quinones can then undergo nucleophilic attack with a thiol present on a protein, producing a new covalent linkage (Figure 1b). The resulting methodology is thus a two-step process in which the tyrosinase enzyme activates the phenolic moiety which is then serves as an electrophile for protein modification. We have demonstrated this approach for the attachment of highly hydrophobic molecules, tyrosine-containing peptides, and tyrosine-tagged proteins to thiol residues. Importantly, the bonds formed between o-quinones and thiols are stable under physiological conditions, suggesting this method would be applicable for the synthesis of ADCs.²⁶

Herein, we present an enzymatic oxidative coupling strategy for the synthesis of ADCs. With the tyrosinase-mediated thiol modification strategy, we install phenol-labeled cytotoxic payloads onto HER2-targeting antibodies bearing engineered cysteines for synthesis of ADCs with DARs 2-4 (Figure 1c). We demonstrate the cytotoxic potential of these conjugates in cell-killing assays. We then extend this strategy to synthesize antibody conjugates containing fluorophores, chelators, and oligonucleotides that have a range of applications in biomaterials and drug delivery.

To begin, we first validated the ability of tyrosinase to oxidize a phenolic drug payload. As a model compound, we chose monomethyl auristatin E (MMAE), an antimitotic drug that halts the cell cycle by blocking the polymerization of tubulin. We synthesized an N-terminal phenol-containing version of MMAE, which included a Val-Cit dipeptide followed by a p-aminobenzyl carbamate (PABC) moiety for intracellular release of the MMAE following cell uptake, compound 1. Previous work from our lab has shown that tyrosinase from Bacillus megatarium (megaTYR) is capable of oxidizing phenols to o-quinones that show excellent selectivity for free thiol additions.²⁷ We have demonstrated that megaTYR can be easily obtained through recombinant protein expression in E. coli and maintains high levels of activity under proper storage conditions at −80 °C.²⁸ Furthermore due to the size difference between the enzyme (~35 kDa) and the antibody, purification with high molecular weight cut off filters or size exclusion chromatography can be used for sample preparation. In addition, we have also demonstrated that resin-bound tyrosinases can be used for this chemistry, facilitating reaction work-up and allowing for enzyme recycling.²⁹ Initial tests demonstrated that 400 nM megaTYR could conjugate 100 μM 1 to a free thiol present at the C-terminus of an engineered GFP-binding nanobody (nbGFP-SG₄C), yielding quantitative conversion in 30 min (Figure 2a,b). Importantly, no off-target modification was observed, as only a single modification event was observed via LC-MS.

Reversion of payload on thiol-conjugated ADCs is a recognized problem in the development of these therapeutic molecules. In particular, this has been shown to occur in the presence of circulating thiols, such as glutathione and albumins.^30–32 As such, we sought to examine the stability of the oxidative coupling (OC) conjugates generated using the tyrosinase approach on the nbGFP-SG₄C substrate, which allows for the facile determination of conjugate purities using ESI-TOF MS. To do this, nbGFP-SG₄C was coupled to compound 1 with quantitative conversion using the above conditions. In parallel, we also modified nbGFP-SG₄C with maleimide analog 3. Next, we challenged the stability of these conjugates with 1 mM glutathione at room temperature or 37 °C for 7 d. At room temperature, both conjugates remained intact over 7 d without any alterations to the linkage (Supplementary Figure S1). However, over this time period at 37 °C, the OC conjugate remained largely intact, with only a 6% decrease in drug loading (Figure 2c, left). In addition, a small amount (<15%) of glutathione addition product was also observed, presumably due to the addition of the free thiol to an additional site on the catechol ring. However, this addition product did not alter the stability of the drug linkage. In contrast, the drug loading of the maleimide conjugates decreased by 30% over the same period, in addition to the hydrolysis of ~25% of the maleimide ring (Figure 2c, right). Taken together, this study indicated that the OC linking strategy led to both greater drug retention and higher purity over the 7-day period. Previous studies from our group corroborate these findings and have shown complete loss in detection of maleimide conjugates in serum compared to conjugates constructed using tyrosinase chemistry.²⁶

We next extended this approach towards the construction of ADCs. An important property of ADCs that determines their antitumor efficacy is their drug-to-antibody ratios (DAR). Previous studies have shown that higher DAR values can disrupt antibody folding, leading to faster clearance of ADCs, while lower DAR values may lead to decreased efficacy and cell killing.^33,34 Thus, it is important to develop chemistries that control the degree of drug payload installations on antibodies. In our studies, antibodies were expressed with four additional cysteine residues for bioconjugation, allowing for a maximum of DAR-4 ADCs to be generated. As candidate ADCs, we chose two antibody variants that were both based on the HER2 binding scaffold of Trastuzumab, an established anti-HER2 mAb already validated in clinical trials. Antibody A contained engineered cysteines at light chain (LC) E165C and heavy chain (HC) S375C, while antibody B contained engineered cysteines on HC positions E152C and S375C (EU numbering) (Figure 1d). These sites were chosen for mutagenesis due to their low solvent exposure, thus enhancing stability of the potential conjugate, while maintaining accessibility for modification³⁵. To ensure no advantageous small molecule disulfide bonds had capped the engineered thiols during expression, the antibodies were deblocked using a mild reductant and interchain disulfide bonds were reformed (see Supplementary Information).

Initial tests showed that 400 nM megaTYR could successfully conjugate 100 μM of 1 to each antibody in 30 min, affording DARs of ~3.5 for both antibodies A and B (Figure 3b, Supplementary Figure S2). As a comparison, we also conjugated 100 μM of a maleimide containing Val-Cit-PABC-MMAE (compound 3) to each antibody (Figure 3d, Supplementary Figure S2). This compound showed similar conversion for both antibodies as well, suggesting the tyrosinase-based approach is comparable with the conversion seen using conventional maleimides. However, unlike maleimides one can install a phenol into any compound using a variety of commercially available building blocks, offering a considerable gain in synthetic convenience.

After successfully conjugating 1 to antibodies A and B, we next set about optimizing the reaction conditions for the synthesis of ADCs using tyrosinase. As a test system, we chose the addition of compound 1 into antibody A as it can be easily reduced and screened for reactivity using LC-MS. Screening across pH ranges of 5.5 to 9.0 showed that megaTYR could successfully conjugate phenol-MMAE to both engineered thiols (Supplementary Figure S3). Even at the lowest pH of 5.5, appreciable conversion of both thiols was observed, suggesting that the addition of thiols into ortho-quinones is quite efficient even under acidic conditions (Supplementary Figure S3). We also screened the conversion of the ADCs using different concentrations of megaTYR, as it is desirable to minimize the amount of enzyme used for future purification steps (Supplementary Figure S4). We found >70% conversion of all four free thiols in antibody A at all concentrations tested (100 nM – 800 nM), with optimal modification occurring between 100-200 nM of megaTYR (corresponding to 0.25-0.5 mol%, based on reactive antibody thiols). Our lab has recently showed that megaTYR can be engineered to alter the affinity for phenols with different surrounding electronics, and thus alter rates of ortho-quinone generation. We screened across a panel of tyrosinases, including a commercially available tyrosinase from Agaricus bisporus (abTYR), as well as wild type megaTYR and a positively charged variant, D55R megaTYR (Supplementary Figure S4). Successful attachment of compound 1 was observed at all conditions tested, indicating that phenolic payloads could be attached to cysteine-containing antibodies using a variety of tyrosinases.

Using the optimized reaction conditions above, ADCs for A and B were generated by scaling up the tyrosinase-mediated oxidative couplings. In this scale up, we included phenolic compound 1 as well as a PEG-spaced variant (compound 2), and a maleimide comparison (compound 3) (Figure 3b–d). All reactions showed good conversion, with DARs averaging ~3.8. Reduced ESI-TOF MS analysis allowed us to analyze the relative conversion of the HC and LC cysteine mutants in antibody A. In all cases, full conversion was observed on the HC but not the LC, supporting the hypothesis that the location of the thiol handle dictates reactivity, and indicating that the thiol installed at LC E165C may have increased steric bulk which precludes full conversion.

With the panel of ADCs in hand, we next explored the antitumor efficacy on HER2(+) cells in a cell killing assay. For each ADC, the corresponding linker-MMAE was tested as well as the cysteine-antibody alone and MMAE alone at concentrations ranging from 10 nM to 1 pM (Figure 4b–g). HER2(+) SK-BR-3 cells were chosen as the experimental cells, while HER2(−) MDA-MB-231 cells were chosen as negative controls. The HER2(−) cells showed no response to any treatment except when treated with MMAE drug alone, supporting the notion that untargeted cytotoxic treatment can damage healthy cells and lead to toxic side effects. HER2(+) cells were more sensitive to free MMAE, with an EC₅₀ of 102 pM, which compared well to the literature-reported value.³⁶ No cell death was observed when exposed to the linker-MMAE constructs alone, compounds 1-3, possibly because the slightly larger, more polar linkers could not cross the cell membranes via diffusion and endocytic uptake was inefficient without the receptor binding step.

The A series of ADCs generated with the tyrosinase chemistry, phenol-alkyl-MMAE-ADC (ADC A1) and phenol-PEG-MMAE-ADC (ADC A2), showed similar EC₅₀ values of 174 pM and 172 pM, respectively. Compared to the maleimide-MMAE-ADC (ADC A3) with EC₅₀ 163 pM, the phenol ADCs (ADC A1 and A2) proved to be similarly cytotoxic. However, IgGs have been shown to remain in circulation for over 20 days³⁷, suggesting that long-term treatment using tyrosinase-synthesized ADCs may retain targeted cytotoxic functions for longer. Interestingly, despite having slightly lower DAR values (Supplementary Figure S2), the ADC B series demonstrated 2-3 fold better cytotoxicity then the corresponding ADC A series (Figure 4h, Supplementary Figure S5), with EC₅₀ values of 48 pM, 78 pM, and 45 pM for the phenol, phenol-PEG and maleimide ADC (ADC B1, B2, and B3), respectively. This result suggests that placement of the drug can influence various properties of ADCs including binding, stability, and delivery. In this scenario, placement of all MMAE molecules on the heavy chains induced a greater cytotoxic response.

In addition to small molecule drugs, other functional handles can be installed onto antibodies for therapeutics or biological studies. For example, fluorophores are useful tools for understanding their bioavailability and localization within cells.^1,2,3 Fluorophore-labeled antibodies can also be used for ligand binding assays to measure interactions between two molecules and imaging.^38–40 Another common substrate attached to antibodies are metal chelators for the purpose of targeted radioimmunotherapy^41,42 and imaging⁴³. As such, it is important to develop compatible bioconjugation chemistry with antibodies and their attachment substrates. To demonstrate the versatility of the tyrosinase-mediated oxidative coupling chemistry, we attached phenol-containing Alexa Fluor 488 (AF488) (Figure 5a,b) and tetraxetan (DOTA) (Figure 5c) to antibody A. In both reactions, the heavy chain modified quantitatively, and the light chain modified singly with AF488 and DOTA at 85% and 88% yields, respectively.

Another interesting, yet challenging substrate class for antibody conjugations are oligonucleotides. The conjugation of oligonucleotides to antibodies has growing applications in many fields, including the DNA-mediated barcoding of antibodies^44,45 as well as the delivery of oligonucleotide-derived immune agonists^46,47. Most notably, unmethylated cytosine-phosphate-guanine oligodeoxynucleotides (CpG ODNs), commonly found in bacterial and viral DNA, bind the toll-like receptor 9 (TLR9) pathway and induce an immune response for the treatment of cancer.^48,49

One of the most common methods for oligonucleotide conjugation to antibodies is via NHS ester coupling. This method, however, is typically associated with non-site-specific modification, generating heterogenous antibodies with various degrees of conjugation.¹² With the generation of several megaTYR variants, our lab has been able to install negatively charged oligonucleotide cargo onto proteins site-specifically.²⁷ The D55R megaTYR variant has the strongest preference for negatively charge substrates and has been used to attach DNA to proteins. We explored the conjugation of murine and human TLR9 agonists—ODN1826 and ODN2006, respectively—using D55R megaTYR (Supplementary Figure S6). Phosphorothioate oligonucleotides provide several advantages over the native phosphodiester bond for delivery in vivo, including improved permeation of the cell plasma membrane and resistance to degradation by nucleases. To test the compatibility of these substrates with tyrosinase chemistry, we used the phosphorothioate analog of ODN1826. With the amine DNA, we attached a phenol handle using an NHS ester-phenol. To confirm the formation of the phenol-DNAs, we conjugated both to GFP-S202C, then performed SDS PAGE gel and LC-MS analysis (Supplementary Figure S6). SDS PAGE showed a GFP band shift in the presence of tyrosinase and phenol-DNA. The analysis also identified a faint secondary band modification in addition to the expected product band for both ODN conjugates. However, MS showed quantitative conversion to the expected mass of tyrosinase-mediated addition of a single phenol-DNA for each ODN construct, but no over addition. Additionally, clean conversion to the expected conjugate with ODN1826 demonstrates the compatibility of the chemistry with phosphorothioates without side reactivity.

For antibody modification with DNA, optimal conditions included 5 eq of ODN per cysteine and 400 nM of D55R megaTYR, providing 82% HC and 61% LC single modification with ODN1826, and 73% HC and 67% LC single modification with ODN2006 by SDS PAGE gel densitometry (Figure 5d, Supplementary Figure S7). The varying degrees of modification on the LC and HC were similar to those observed with MMAE attachment. Furthermore, minor secondary bands, each corresponding to <20% modification, were observed in addition to the LC and HC single modification bands. LC-MS analysis successfully identified the HC-DNA conjugate, but the LC-DNA conjugate could not be observed using this method (Supplementary Figure S8). MS analysis did not identify a second addition of phenol-DNA using either of the ODN constructs.