Cysteine-to-lysine transfer antibody fragment conjugation

Site-selective antibody fragment conjugation is achieved by using a disulfide bond to ‘hook’ and deliver thioester acylating agents to specific lysines.


Introduction
Over the past decades, antibody bioconjugation has emerged as a powerful tool, providing new avenues for the development of therapeutics and diagnostics. 1 Combining the exquisite targeting ability of antibodies with small molecules has enabled access to a broad range of constructs, including antibody-drug conjugates, 2 bispecics, 3 radioimmunoconjugates, 4 as well as antibody-nanoparticle conjugates 5 and targeted imaging agents. 6 Notably, antibody fragments have shown distinct advantages over full immunoglobulins, including enhanced tumour penetration, lower immunogenicity risk, accelerated renal clearance (tunable half-lives, e.g. by PEGylation) and production in cheaper prokaryotic expression systems. [7][8][9] For the next generation of antibody conjugates, it has been demonstrated that site-selective modication strategies that afford robustly stable constructs are vital to ensure superior in vivo outcomes. 1,10,11 The use of genetic engineering to incorporate cysteine mutants, 10 unnatural amino-acids 12 or enzymatically-recognised handles 13 has enabled antibody modication with an unprecedented degree of site-selectivity. However, with these approaches, further input of resources in the antibody development phase is required and variable protein expression yields, disulde scrambling or aggregation are limitations oen witnessed. 10,[14][15][16] Alternatively, we and others have recently described the development of reagents which are able to modify native disulde bonds by re-bridging the two cysteine residues, producing homogeneous antibody conjugates. [17][18][19][20][21][22][23][24] Importantly, the structural integrity of the antibody is maintained, contrary to targeting each cysteine residue independently, which has been shown to reduce the stability of the antibody in vivo. 25 Whilst disulde bridging is a promising strategy, the resultant conjugates are yet to be validated in the clinic.
Labeling via the primary amino groups on lysine residues has been heavily pursued, due to the advantages of using readily available acylating agents (e.g. NHS esters) to form robustly stable, clinically validated amide bonds. 26 However, due to the multitude of surface accessible lysine residues, heterogeneous mixtures are inevitably obtained with batch-to-batch variability and unpredictable pharmacokinetic properties. 27,28 An ideal approach to antibody modication would involve the siteselective labeling of lysines by acylation, as it would full both the criteria of homogeneity and robust stability. Reagents have been described which offer greater selectivity for certain lysines than conventional reagents by exploiting subtle differences in pK a 's of lysines or local environments. 29,30 We envisaged that an alternative approach to achieve selectivity would be to use proximal cysteine residues as ligating 'hooks', delivering acylating agents specically to certain lysine residues (Fig. 1). This cysteine-to-lysine transfer (CLT) methodology would offer new opportunities in accessing site-selective protein conjugates more widely, building on recent reports of using reversible bonds to deliver reactive functional groups to specic aminoacids, 31 including work by Bertozzi and coworkers on the use of nitrile reagents to modify specically designed cysteinecontaining peptides by S-to-N transfer to targeted lysine residues. 32

Results and discussion
We chose to develop the CLT strategy for antibody conjugation on the Fab fragment of Her2-targeting breast cancer drug trastuzumab. Specically, trastuzumab fragment conjugates are of widespread interest for drug delivery and imaging applications, [33][34][35] and the presence of 26 lysines and a single disulde bond would allow clear interpretation of the viability of the CLT strategy. Analysis of the region around the Fab disulde revealed three proximal lysine residues on the heavy-chain (K136, K221, K225) and none on the light chain (Fig. 2).
Thioesters present an ideal reactivity prole to achieve the desired initial cysteine acylation and subsequent S,N-acyl transfer onto proximal lysines. They are >100 times more reactive than the corresponding oxo-esters to nucleophiles such as amines, whilst being less reactive to hydrolysis. 36 This is exploited by nature, e.g. in acetoacetyl CoA in the Krebs cycle, 37 in the ubiquitination of proteins, 38 and in intein splicing. 39 Native chemical ligation (NCL) uses this reactivity to achieve selective amine-acylation on peptides and proteins via an S,Nacyl transfer involving a 5-membered ring intermediate. 40 This concept has also recently been applied to larger, macrocyclic intermediates, 41 such as the use of internal cysteines in peptides to accelerate ligation to the N-terminus. [42][43][44][45] In order for the CLT strategy to work, a thioester would need to react with the Fab fragment only upon reduction of the disulde bond. Guided by NCL, which employs aryl thioesters to achieve efficient thiol-thioester exchange, we began our studies by treating Fab and reduced Fab with thiophenyl thioester 2 along with a common acylating agent NHS ester 1. The number of acylations were identied by LC-MS (Fig. 3, see ESI † for raw LC-MS data). The NHS ester 1 generated a statistical distribution prole, consistent with non-selective lysine conjugation, upon reaction with either native or reduced Fab ( Fig. 3a and c, respectively). The aryl thioester 2 demonstrated high selectivity for transthioesterication with reduced Fab, but the presence of double acylated light chain (LC) along with unmodied LC conrmed that a small amount of background lysine conjugation was occurring (Fig. 3d). This was further observed in the control reaction with native Fab, which showed a single acylation (Fig. 3b); treatment of this species with TCEP Fig. 1 We describe the use of proximal cysteines to deliver acylation reagents to specific lysine residues in an antibody fragment, improving the homogeneity of these conjugates, whilst retaining the robustly stable amide linkages; CDRs ¼ complementarity-determining regions.  indicated that a reactive lysine(s) is present on the LC, which is consistent with literature reports (Fig. S6 †). 29,30 With the view to tuning down the reactivity of the thioester and hence avoid non-specic reactivity, we turned our attention to the use of alkyl thioesters. MESNa thioester 3 was found to be completely inert upon reaction with native Fab, even at 100 equiv. of reagent (Fig. 4b), while it was shown to undergo extremely selective transthioesterication with reduced Fab, affording solely the desired cysteinyl thioester conjugate 4 (Fig. 4c). 46 Further conrmation that the cysteines were the sites of selective reaction was obtained by the addition of thiols (cysteine, 100 equiv.) post conjugation which readily cleaved the thioesters regenerating native Fab (Fig. S10 †).
Having identied that alkyl thioesters can undergo selective transthioesterication, the next step was to attempt the S,N-acyl transfer to nearby lysine residues. While in NCL the initial transthioesterication is the rate determining step, in ligations proceeding via larger ring sizes, the S,N-acyl transfer becomes rate determining. 40 Given the macrocyclic intermediates involved in the CLT strategy, we anticipated that this reaction would not be rapid and optimization of conditions to limit competing hydrolysis would be required. We analysed reactions up to 72 h and used LC-MS to reveal the alkyne : antibody ratios (AARs) (see ESI Table S2 † for conditions examined). We observed that at physiological conditions (pH 7.4, RT), signicant transfer had taken place, but the reaction was incomplete. Enticingly however, the major product formed was an acylated conjugate derived from the reoxidised native Fab, conrming that the cysteines had been liberated and had spontaneously reoxidised (see also SDS-PAGE, Fig. 4f), which negated the requirement for an extra oxidation step.
By increasing the temperature to 37 C and pH to 8.0 the transfer reaction was pushed to completion in just 24 h. However, the AAR of 1.0 revealed that 50% of competing hydrolysis was taking place. To minimise this, we identied that lower temperature and higher pH gave the best yield of transfer. The nal optimised conditions of 12 C, pH 8.4, 72 h successfully afforded CLT conjugate 5 with an average AAR of 1.5 ( Fig. 4d and f, lane 5). Upon treatment with thiols (100 equiv. cysteine, pH 8.4, 37 C, 2 h), no change in the AAR was witnessed ( Fig. S14 †), conrming that a robustly stable acylated conjugate was obtained. Reduction of the conjugate with TCEP, followed by capping with N-methylmaleimide conrmed that the acylation had taken place exclusively on the heavy chain, which was consistent with a transfer mechanism occurring (Fig. 4e). The conjugate was then subjected to tryptic digestion, followed by LC-MS/MS analysis (see ESI †). 100% sequence coverage was obtained and the sites of modications were identied as K136, K221 and K225, being in accordance with predictions based on the Fab crystal structure. Size-exclusion analysis of CLT conjugate 5 conrmed that no aggregation had taken place and ELISA showed full retention of binding activity ( Fig. S26 and S27 †). Finally, CuAAC was employed to conjugate AlexaFluor488 azide to generate a functional antibody conjugate 6. The uorophore-to-antibody ratio (FAR) was determined to be 1.5 by UV absorbance (Fig. S16 †), which supported the loading obtained by LC-MS.
Next, we envisaged that the use of a bis-thioester would enable an alternative stoichiometry for this CLT strategy and postulated that the rigidity of a bridged system may further control the regioselectivity. We synthesised bis-thioester 7 using 2-methoxyethanethiol to infer water solubility, whilst avoiding purication issues associated with the highly polar bis-MESNa adducts. Treatment of reduced Fab with this reagent (100 equiv.) selectively afforded bridged conjugate 8 (Fig. 5b)   withdrawing effects of the b-carbonyl. Following the successful formation of conjugate 8, the S,N-acyl transfer was examined. Through optimization (see ESI †), we observed that the ideal conditions were 6 h at 37 C, with LC-MS analysis revealing that the dominant product was the AAR 1 CLT conjugate 9, in which the transfer was accompanied by hydrolysis of the second thioester. A further 2 h treatment with BME was identied as required to cleave off remaining traces of monothioesters which had formed due to competing hydrolysis, to afford a nal AAR 0.8 conjugate (Fig. 5c). Incubation with BME also served to conrm the robust stability of CLT construct 9, as no change in the AAR was witnessed. Following enzymatic digestion and LC-MS/MS analysis, we were pleased to nd that this reaction was site-selective for K136 ( Fig. 5d; 95% sequence coverage was obtained with 100% coverage of lysine residues). As only one lysine is reactive to acylation in this more constrained, bridged system, we infer that conformational effects must be playing a key role in driving this improved selectivity i.e. by holding a thioester in particular proximity of the K136 amino group. CLT conjugate 9 was also analysed by size exclusion chromatography, where no aggregation was observed to have taken place under the transfer conditions and ELISA analysis demonstrated full retention of binding activity ( Fig. S26 and S27 †). Subsequent AlexaFluor488 click conjugation generated uorescent conjugate 10 with a matching FAR (Fig. S23 †).

Conclusion
In summary, cysteine-to-lysine transfer (CLT) methodology allows the construction of highly homogenous antibody fragment conjugates, whilst incorporating robustly stable, clinically validated amide linkages. The readily available thioester reagents are shown to react selectively with the cysteines obtained from the reduced interchain disulde bond in a Fab, and then transfer at raised pH to specic proximal lysine residues, which are ideally placed distal from the binding site. By employing either mono-or bisthioesters we have shown it is possible to control the stoichiometry, to afford major products containing 2 or 1 acylations per disulde. Whilst hydrolysis of the thioesters represents an expected competing background reaction, the efficiencies of the macrocyclic S,N-acyl transfers are impressively still 75-80%, and it is likely that these can be further tuned by reagent design in future generations of CLT reagents. Current site-specic approaches, for example using cysteine mutants, 47 can achieve conjugation efficiencies over 90% and this should be the ultimate target. However, the use of an S,N-transfer reaction on a native protein to generate highly desirable amide linkages, with overall average loadings achieved of 1.5 and 0.8 per disulde, already place this methodology in a suitable position for use in antibody conjugations and protein modications more widely. It is also envisaged that CLT conjugation will allow the ready conversion of existing heterogeneous lysine reagents and conjugates to site-selective versions, building on the condence in the resultant amide linkages whilst offering the prospect of improved therapeutic indexes and production processes known for homogeneous conjugates.  the mass spectrometry experiments; N. F., V. C., K. K. and J. R. B. analysed the data; N. F., V. C. and J. R. B. co-wrote the paper.

Conflicts of interest
The authors have led a patent application on this technology with UCL Business.