Optimal Synthetic Glycosylation of a Therapeutic Antibody

Abstract Glycosylation patterns in antibodies critically determine biological and physical properties but their precise control is a significant challenge in biology and biotechnology. We describe herein the optimization of an endoglycosidase‐catalyzed glycosylation of the best‐selling biotherapeutic Herceptin, an anti‐HER2 antibody. Precise MS analysis of the intact four‐chain Ab heteromultimer reveals nonspecific, non‐enzymatic reactions (glycation), which are not detected under standard denaturing conditions. This competing reaction, which has hitherto been underestimated as a source of side products, can now be minimized. Optimization allowed access to the purest natural form of Herceptin to date (≥90 %). Moreover, through the use of a small library of sugars containing non‐natural functional groups, Ab variants containing defined numbers of selectively addressable chemical tags (reaction handles at Sia C1) in specific positions (for attachment of cargo molecules or “glycorandomization”) were readily generated.

Protein glycosylation is the most common and varied posttranslational modification, critically influencing protein function, [1] and yet is currently difficult to control. [2] Endoglycosidase (ENGase or "Endo")-catalyzed glycosylation is an attractive strategy to access homogeneous glycoproteins ( Figure 1a); [3] the initially limited scope of glycoproteins was expanded to antibodies (Abs) by the use of EndoS, [4] af amily 18 glycoside hydrolase (GH) from Streptococcus pyogenes capable of glycosylating immunoglobulins (Igs). [5] Monoclonal Abs (mAbs) and antibody-drug conjugates (ADCs) are arapidly growing class of therapeutics. [6] Glycans in Abs [7] modulate stability,t he rate of clearance,a nd the pharmacokinetic profile; [8] aggregation, folding, and immunogenicity; [9] complement activation; [10] binding to Fc receptors and Ab-dependent cell-mediated cytotoxicity; [11] and Abmediated inflammation. [12] They are therefore vital functional "switches" that cannot yet be controlled cleanly (see the Supporting Information for an extended discussion).
Antibodies are all N-glycosylated in the Fc region of each of two heavy chains.A ll therapeutic Abs are currently produced from cells as mixtures ( Figure 1b); more than 20 different glycoforms are typically identified. [13] By contrast, the chemoenzymatic ENGase method could potentially be used to access pure Abs.H owever,u ntil now it has been assumed that this method will necessarily give rise to homogeneous glycoforms by virtue of the direct reversal of selective enzymatic hydrolytic activity ( Figure 1). Herein we demonstrate that this assumption is incorrect:n ot only do nonspecific background chemical modifications compete,but we now reveal optimized methods that allow access to essentially homogenous (! 90 %p ure) glycoforms of ak ey therapeutic mAb.
Our preliminary studies [5a] had indicated that wild-type (WT) EndoS could be successfully used to trim glycans from mixtures of glycoforms of human IgG to reveal single GlcNAc moieties (Figure 1c,l eft). Subsequent treatment of the resulting IgG-GlcNAc with WT EndoS and an appropriately activated sugar oxazoline donor led to the formation of anew glycosidic linkage (Figure 1c,right). [5a] However, the inherent hydrolytic activity of EndoS prevented fully efficient reactions.T oo vercome this limitation, we explored the use of mutated variants of EndoS to access enzymes with enhanced transglycosylation:hydrolysis (T:H) activity ratios.S imilar strategies [14,15] have proven successful in other ENGase systems,b yp artial analogy with synthases described by Withers and co-workers. [16,17] Sequence alignment (see the Supporting Information) with other family 18 and 85 GHs [18] suggested residues D233, E235, Q303, and Y305, which enhance the role of the C2 amide in reactions involving oxazolinium intermediates (D233), act as ageneral acid/base (E235), or assist substrate binding (Q303, Y305). [ 10:100 (WT);n d= not determined;s ee the Supporting Information). Although, in our hands,none displayed completely abolished hydrolytic activity,i tw as substantially decreased in EndoS-D233Q as compared to EndoS-WT,thus giving rise to aT :H activity of 80:20. We therefore selected EndoS-D233Q.During the course of this study,W ang and co-workers also suggested that EndoS-D233Q and EndoS-D233A mutants possess useful "synthase" activity. [5b] The mutant EndoS-D233Q is sufficiently stable to be produced on scale.
We chose the therapeutic mAb Herceptin as ahighly representative substrate (see the Supporting Information). Our analysis of Herceptin (see Figure S4 in the Supporting Information) suggested at least seven major glycoforms with many other minor species,d ominated by complex biantennary structures,c onsistent with prior observations. [20] We estimate the most prevalent (asymmetric G0F/G1F) to account for less than 35 %; Herceptin is therefore highly heterogeneous.
We set out to create ap ure, single,s ymmetric glycoform of Herceptin bearing ar elevant complex biantennary glycan at each Fc Asn300 position. [21] A corresponding activated sugar oxazoline 2 was created on at ens-of-milligrams scale [22] to enable the creation of af ully sialylated G2F/G2F (S2G2F/ S2G2F) glycoform (S2G2F/ S2G2F-Herceptin). In principle, this glycan would convey designed anti-inflammatory properties, [12a] but at levels of incorporation not accessible in previous studies.I ts incorporation would, in turn, enable the ready creation, advantageously, of ADCs with reduced inflammatory properties in am anner not previously possible.
First, Herceptin was converted cleanly into Herceptin-GlcNAc (1)b yt he use of EndoS-WT.T hen, glycosylation at 30 8 8Ci np hosphate buffer (pH 6.5) with oxazoline donor 2 (2 70equivalents,s econd addition after 40 min) in the presence of EndoS-D233Q gave the desired glycosylated Ab 3,e ssentially as as ingle glycoform (Figure 2a  predominate. c) EndoS-WT cleaves the core of mixed N-glycans and can subsequentlyc atalyze glycosylationw ith oxazoline donors, [5a] to give, in principle, purer glycoformdistributions. d) Payload molecules may also, in principle,b eintroducedd irectlyt hrough glycosylation or indirectly by the incorporation of reactive handles in sugars and asubsequent selective reaction. Angewandte the reduction of inter-a nd intrachain disulfides to give two heavy and two light chains). This method of analysis has been the dominant and most successful approach to assess the chemistry of monomeric proteins by us and others. [23] However,t he mAb products of these reactions are heteromultimeric, and it occurred to us that the true details of these reactions (at, for example,e ach of the two Asn300 sites simultaneously) would only be revealed in this relevant protein complex. We elucidated conditions for evaluation by high-resolution native MS (nMS;s ee the Supporting Information). Strikingly,s uch nMS analysis without reduction revealed heterogeneity not detected in the monomeric state by rMS.A lthough the major reaction product was the expected diglycosylated Ab 3,w ea lso identified peaks corresponding to the attachment of only one and even three glycans (Figure 2c,d), along with small quantities of glycoforms containing only one core fucosyl moiety (see the Supporting Information). Thee stimated purity of 3 in this product mixture was below 75 %. [24] Thed etection of unexpected Ab products bearing three glycans had as triking implication:n onselective chemical reaction(s) occur(s) alongside the desired selective enzymatic reaction. Treatment of the glycosylation products with PNGaseF,w hich removes N-linked glycans from Herceptin, left these additional glycans attached (see Figures S25-S27). Tr yptic MS/MS mapping confirmed their non-Asn-linked nature (see the Supporting Information). Notably,W ang and co-workers have suggested that other endoglycosidases (i.e., EndoA) have broad acceptor-substrate specificity,t hus potentially glycosylating acceptors other than Asn-linked GlcNAc. [25] Finally,w et ested the direct reaction of 2 with Herceptin-GlcNAc (1;F igure 3; see also Figures S15 and S16);after 2hin the absence of an enzyme,chemical addition products ("glycation") were observed. At pH 7.4, around 20 %ofthemAb carried one glycan, 53 %carried two glycans, and 27 %c arried three glycans (Figure 3a). [5c, 26,27] When the pH value was lowered to 6.5, around 69 %ofthe mAb had not undergone any glycation after 2h:2 6% carried one glycan, and 5% carried two glycans.D irect high-resolution rMS of glycated samples indicated that most glycation occurred on the heavy chain, with small amounts on the light chain ( Figure 3b). LC-MS/MS following tryptic digestion revealed several possible sites (including heavy-chain site K30 and light-chain sites K188, H189, and K190;s ee Figure S41). Importantly,t he direct treatment of Herceptin with lactol 4 did not result in glycation, thus indicating that an excess of ar educing sugar (e.g., from the hydrolysis of 2 to 4)w as not the cause of the "glycation" process (see Figures S17 and S18).
Having discovered this previously unreported, but critical, aspect of the ENGase method, we set out to devise newly optimized methods to solve this apparent problem. Any solution would need to address three issues simultaneously: the optimization of transglycosylation (T) and minimization of two competing reactions,h ydrolysis (H) and "glycation" (G). Through experimentation, we logically optimized the T:H + Gr atio with EndoS-D233Q;w et hus used ah igher enzyme loading (10 %w/w rather than 2-5 %) and lower peak concentrations of oxazoline 2 ([Ab] = 13.3 mm ;[ 2] init < 0.2 mm,m ultiple additions (7 ) of fewer (15) equivalents at 15 min intervals). nMS revealed an efficient reaction with virtually no discernible "glycation" and at least 90 %desired bisglycosylation (fully sialylated mAb/mAb bearing four sialic acid residues,g iving > 75 % [28] 3,S 2G2F/S2G2F-Herceptin;F igure 4).  Herceptin is also known to internalize after binding to the HER2/neu receptor, [29] thereby valuably raising the potential of this mAb as ap latform for intracellular drug/toxin/ diagnostic targeting;t his potential has been recognized in the development of the ADC Kadcyla. We envisaged that judiciously selected sugar donors would therefore not only enable efficient access to Ab constructs with defined natural glycosylation but also modified, unnatural glycans bearing precise numbers of attachment sites (handles or tags) at set positions.T hese sites would enable the subsequent attachment of "cargo". Alternatively,s uch "cargo" could be attached to the glycan prior to enzymatic glycosylation (Figure 1d). Va rious appropriate representative (potentially selectively addressable) "handles" were incorporated into 4 to create activated sugar donors 6a-e (via 5a-e)a st est substrates for EndoS (Table 1). Our chemoselective synthetic strategy avoided the use of protecting groups and allowed direct conversion from free reducing sugars.Thus,the two C2 carboxylate groups found in the nonreducing terminal sialic acids of the decasaccharide lactol 4 [22c] were converted into amides with 4-(4,6-dimethoxy-1,3,5-triazin-2-yl)-4-methylmorpholinium chloride (DMTMM). [30] Through the use of an elevated temperature (37-50 8 8C) and an excess of the ammonium salt and coupling agent we were able to drive amidation to completion to give 5a-e.I ne ach case,t he product was readily purified by filtration and size exclusion chromatography.S ubsequent dehydration of lactols 5a-e to the oxazolines 6a-e was carried out with 2-chloro-1,3dimethyl-1H-benzimidazol-3-ium chloride (CDMBI) under mildly basic conditions. [22c] Given the ready access to 4, [22a,b] this route provides arapid method to attach reaction handles to biologically important glycans (and in turn to proteins,see below) with minimal operational manipulation and/or purification, thereby allowing its ready application and scale-up.It could also,i np rinciple,b ea pplied to any neuraminylcontaining sugar extracted from an atural source,e ven mixed samples.
Pleasingly,i ne very case EndoS-D233Q catalyzed the glycosylation of Herceptin-GlcNAc (1)w ith the functionalized, non-natural glycans 6a-e to give glycosylated mAbs 7ae (see Table S1) carrying two glycans each functionalized with two reaction handles as the major products,asjudged by rMS and reducing SDS-PAGE(see Figures S20-S24). However,as before,n MS analysis of these Abs under native conditions indicated varying levels of purity.S ignificant amounts of Ab carrying one or three glycans were observed (heterogeneity was dependent upon the sugar used;s ee Table S1). Larger and more "unnatural" tags led to ad ecrease in efficiency (from ! 90 %d esired bisglycosylation with "natural" 2 to 74 %b is-and 22 %m onoglycosylation with propargyl 6a,6 5% bisand 27 %m onoglycosylation with azide 6b,70%bisglycosylation with thiol 6c,a nd only 59 %b isglycosylation with 6e,w hich contains abulkier iodoaryl group;see Figures S20-S24). Thep resence of al abile disulfide in 6d apparently contributed to the high heterogeneity observed in its Ab product (< 40 %o f7d). We chose to optimize the transglycosylation reaction by using the non-natural oxazoline donor 6b bearing an azide group to create 7b ( Figure 5). Use of the previously optimized conditions (7 15equiv) led to only small amounts of glycation (ca. 2%). Glycation was almost entirely eliminated by further decreasing the rate of addition (20 5equiv,5 min intervals); under these conditions,90% bisglycosylation was observed (Figure 5b). Notably,circular dichroism analysis (Figure 5a)s howed that  [a] The disulfide linkage was reduced by the addition of dithiothreitol immediately followingoxazoline formation.
[b] The product was composed of around 70-80 %ofthe target 6e and 20-30 %unidentified material, which had undergone further dehydration. the gross secondary structural elements in Her were preserved during this optimized glycosylation to give azido-Herceptin Ab 7b.
Theazide "tag" sites in the glycans in azido-Herceptin Ab 7b were then used for the attachment of cargo by using ar hodamine alkyne and ac emadotin alkyne variant (generated by the use of standard solid-phase peptide synthesis techniques;s ee the Supporting Information). These compounds allowed us to demonstrate the functional integrity of as ynthetic Her with optimal glycosylation bearing cargo. Rhodamine-Herceptin generated in this way bound metastatic breast adenocarcinoma HER2(+ +)c ells (SK-BR-3) selectively (no binding to HER2(À)M CF-7 cells), as shown by fluorescence-activated cell sorting ( Figure 6). We were also able to generate Herceptin "A DC" variants with different loadings of the cemadotin toxin (ca. 2, 3, or 4; see the Supporting Information). Herceptin bearing approximately three cemadotin units showed enhanced killing of HER2(+ +) cells (SK-BR-3) over Her (EC 50 % 800 pm ;see the Supporting Information).
We have shown herein the first examples,t ot he best of our knowledge,oftheuse of nMS to direct the optimization of protein chemistry.T his approach has vitally enabled the precise and detailed remodeling of Abs,b yr evealing reactions that were not previously appreciated. Although rMS is an essential and useful tool for the characterization [23] and monitoring of chemical modification, [31] labile heteromultimers,s uch as the Ab Herceptin used in this study,m ay require such precision. Thei onization of native protein complexes is dependent on the surface composition, which is largely invariant between, for example,W T, cleaved, and remodeled IgG Herceptin. Therelative glycoform intensities detected and assigned by native MS can therefore be taken as proxy for the relative abundances,inline with the well-known minor effects of post-translational modification on intactprotein analysis. [32] Our increased control of the natural glycosylation pattern of Abs may allow the construction of new optimally efficacious Abs.Inthis study,highly pure sialylated Ab glycoforms have been created;Fcg lycan sialylation imparts anti-inflammatory properties to IgGs, [12a] and has led to the use of IVIg as an anti-inflammatory drug. However,o nly approximately 10 %o ft he total IgG content of IVIg carries sialylation. [33] Thep roduction of pure sialylated Abs (> 90 %s ialylated, as in this study) should improve anti-inflammatory properties and reduce dosage.I nterestingly,a berrant Ab glycation has also been linked to disease (see the Supporting Information). [34] Thei ncorporation of unnatural glycans into Abs also allowed dual simultaneous access to defined glycosylation and specifically positioned reaction handles.C urrent approaches to ADCs in the clinic use backbone Lysa nd Cys residues and give rise to heterogeneity.R ecently,o ther complementary approaches have been suggested that are based on the use of unnatural amino acids [35] or Fc N-glycosylation. [36] Some require either complete removal [36a] or significant truncation [5b, 36b] of the glycans,w hich may adversely  affect Ab stability and immunogenicity [8,9] as well as impacting on FcgRb inding. [37] Alternatively,r emodeling at the nonreducing termini [36c,d] ("tip sugars"), although in principle capable of partially reducing heterogeneity,i sl imited to certain residues (i.e.G al);i ta lso does not remove the variation arising from bisecting sugars [38] and/or hybrid/ triantennary/other branching glycans. [39] Theuse of unnatural AAs does not avoid the desire for or solve the problem of defined glycosylation. Thea pproach reported herein now enables the attachment of ad efined number of reaction handles at specified positions,w hile removing virtually all glycan heterogeneity.I nt his way,i ta llows access to as ingle key glycan motif with the potential for both the attachment of cargo and reduction of inflammation. Thus,A DCs can be created with improved anti-inflammatory properties by ag eneral optimized strategy.W en ote too that the in vivo attachment of cargo may have distinct advantages. [40] Other "pure and tagged" glycoproteins (created by the use of 6a-e, for example), besides Abs,a re anticipated to be of general utility as therapeutics,for diagnostic purposes,and as probes of organismal biology.