Proposed mechanism of off-target toxicity for antibody–drug conjugates driven by mannose receptor uptake

Abstract

Antibody–drug conjugates (ADCs) are developed with the goal of increasing compound therapeutic index by specific and targeted delivery of a toxic payload to the site of action while considerably reducing damage to normal tissues. Yet, off-target hepatic toxicities have been reported for several ADC. Locations of these off-target toxicities coincide with the reported locations of cell surface mannose receptor (MR). The relative proportion of agalactosylated glycans on the Fc domain (G0F vs. G1F and G2F components) in monoclonal antibody (mAb)–based biotherapeutics is closer to some disease state IgG rather than to a normal serum-derived immunoglobulin. The lack of the terminal galactose on a G0F glycan creates an opportunity for the mAb to interact with soluble and cell surface MRs. MR is a known multi-domain lectin that specifically binds and internalizes glycoproteins and immune complexes with relatively high G0F content and has been found on the surface of various cell types, including immune cells of myeloid lineage, endothelial cells, and hepatic and splenic sinusoids. In this review paper it is proposed that the mechanism of the off-target toxicities for ADC biotherapeutics is at least in part driven by the carbohydrates, specifically agalactosylated glycans, such as G0F, their interactions with MR and resulting glycan-derived cellular uptake of ADCs. Several case studies are reviewed presenting corroborating information.

Introduction

Monoclonal antibody (mAb)-based biotherapeutics are typically modified post-translationally by the intracellular machinery of a producing host cell in a manner similar to naturally produced immunoglobulins. The N-glycosylation site is located on the asparagine (Asn-297) amino acid within the CH2 domain and consists of biantennary core oligosaccharides with various degrees of terminal galactosylation. Glycans found at this site are generally categorized based on the number of terminal galactoses (G), that is, G0F, G1F and G2F. Although the nature of the oligosaccharides found within the Fc region is similar to that on serum-derived immunoglobulins, the G0F/G1F/G2F distribution on commercial mAbs is closer to that found on immunoglobulins from disease state sera where the relative fraction of the G0F component is elevated [1, 2]. The lack of the terminal galactose on a G0F glycan creates an opportunity for the mAb to interact with soluble and cell surface mannose receptor proteins, including mannose receptor (MR). The MR is an endocytic and phagocytic receptor and is a known multi-domain lectin that specifically binds and internalizes glycoproteins and immune complexes with relatively high G0F content. The intricate and multi-domain structure of MRs has been reviewed elsewhere [3].

MR binding affinity is high for mannose-containing glycans as well as for agalactosylated glycans (G0F) and N-Acetylglucosamine (GlcNac) as compared to galactosylated G1F and G2F glycans. A reverse correlation between terminal GlcNAc (tGlcNAc) content and half-life of a protein was demonstrated with MR viewed as the main cause of clearance of the protein [4]. Importantly, it was suggested that in proteins with a high G0F content, tGlcNAc residues in G0F structures may become accessible for mannose lectin binding [4].

MR has been found on the surface of various cell types [3, 5]. Upon internalization, MR presents bound protein ligand to the acidic endosomal and lysosomal environments. Antibody–drug conjugates (ADCs) are compounds designed for the targeted delivery of a toxic payload where the linker plays an important role and often is designed to deconjugate the toxin in a specific intracellular environment. Non-specific uptake of an ADC compound by a MR expressing cell can therefore result in non-specific delivery of the payload leading to various types of off-target toxicities. Hepatic toxicities consistent with the initial damage to sinusoidal endothelial cells (SECs) have been observed in toxicity studies with various ADCs [6–12]. Locations of these off-target toxicities coincide with the reported locations of cell surface MR. In this report, it is proposed that the off-target toxicities observed with ADCs can be explained in part by non-target-related uptake of ADCs through the antibody glycan interaction with the cell surface MR. It is proposed that such interaction occurs at least in part as a result of the elevated G0F oligosaccharide content on the IgG moiety of the ADC. Several case studies are reviewed presenting corroborating information.

Glycosylation of IgG molecules

Post-translational modifications, and in particular glycosylation, play a decisive role in ensuring proper protein function. Methodologies used when evaluating the nature, type and relative amounts of glycans found on glycoproteins, including immunoglobulins, have been extensively described and reviewed elsewhere [13–15].

IgGs are glycosylated predominantly in their heavy-chain (HC) within the Fc region. These Fc glycans are usually highly heterogeneous, containing a mixture of terminal sugar residues that include sialic acid, galactose, GlcNAc, core fucose and/or bisecting GlcNAc residues. Variation among individual IgGs include attachment of galactose and/or galactose-sialic acid at one or both of the terminal GlcNAc and/or attachment of a third GlcNAc arm. The glycans that occupy this site have been broadly characterized into three glycoforms, termed IgG–G2F, IgG–G1F and IgG–G0F, based on the number of terminal galactoses attached to the glycan [16, 17]. Increased levels of G0F are reported for certain disease conditions as described later in this review [17, 18].

Recombinant mAbs produced commercially are most commonly manufactured in Chinese hamster ovary (CHO) cells, murine myeloma NSO or Sp2/0 cell lines. The glycans on mAbs generated in these production cell lines are heterogeneous with a restricted glycoform profile relative to that observed in human IgG purified from normal human blood. The major glycan structures on recombinant mAbs have been identified as asialo biantennary complex-type glycans containing a-1,6 linked fucose with zero or one galactose [16]. Considerable differences were reported in the measured ratios of G0F, G1F and G2F levels for monoclonal humanized IgG preparations (52:38:10) and ratios observed for the neutral oligosaccharides of human polyclonal IgG (23:43:43) [15]. In IgG–G0F, the glycans do not have terminal galactose residues, thereby exposing GlcNAc residues.

Bevacizumab (Avastin) is a humanized form of a murine mAb. G0F accounts for approximately 80 % of the oligosaccharide structure [19]. CHO produced Rituxan, a chimeric mouse/human mAb, was reported to contain between 45 and 80 % of G0F. An analysis of four glycoforms found on rituximab revealed that all oligosaccharides had a Fuc residue in the innermost GlcNAc residue [20]. The most abundant (59.5 %) oligosaccharide lacked both galactose residues on the non-reducing ends of the biantennary chain (G0F). Two monogalactosylated positional isomers and double galactosylated isoforms were found at 28.3 and 8.5 % (G1F) and 3.8 % (G2F), respectively. The importance of this fact will be discussed later with respect to MR aided off-target toxicities.

In a study by Adamo et al. [1], the composition and relative % of monosaccharides found on a mAb drug candidate expressed in a B cell hybridoma and CHO cell transfectoma were evaluated. The hybridoma produced compound contained three major glycans: G0F, G1F and G2F and a more complex mix of glycans compared to the CHO cell derived product, presumably due to the expression of different types and amounts of glycosyltransferases. The CHO cell produced immunoglobulins were shown to contain a proportionally higher fraction of G0F component relative to G1F and even less of G2F (66–75, 22–30 and 3–4 %, respectively).

It was suggested that G0F oligosaccharides have increased mobility resulting from loss of binding between G0F oligosaccharides and the Fc protein surface [2]. Regions of the protein surface normally covered by the galactosylated oligosaccharide could be revealed as a result of reduced galactosylation. Consequently, abnormal lectin-like activity, elevated potential for interaction with mannose binding protein (MBP) or receptor could be speculated. Interactions between antennal terminal galactose and hydrophobic grooves of the CH2 protein backbone of the Fc have been shown to play an important role in maintaining IgG structure [2]. A relatively higher hydrophobicity of galactose versus glucose presents an opportunity for increased glycan binding with lectins, including MR. The exact degree of additional flexibility that enables MR interaction is unknown.

An investigation of the impact of Fc glycan structure on the pharmacokinetic (PK) clearance of therapeutic human IgGs in humans demonstrated that therapeutic IgGs containing Fc high-mannose glycans are cleared more rapidly in humans than other glycan forms [21]. Similar observations have been observed preclinically; however, they have not been formally published.

The proposed increased mobility of G0F component and its consequences could be contributing to the pathology of some disease states where glycosylation changes have been detected. Glycosylation patterns described for mAb compounds are similar to the glycoforms found on serum IgG from patients in certain disease states. Elevated levels of G0F are reported for sera from patients with various chronic inflammatory and infectious diseases [17, 18].

It is therefore possible that such abnormal glycosylation patterns can be viewed by the body as foreign. It could be speculated that abnormally glycosylated mAbs could be subjected to removal via specialized protective mechanisms, for example, by interaction with MBPs or receptors.

Mannose binding receptor (MR)

MR can specifically interact with carbohydrate moieties found on bacteria, fungi, parasites, and viruses and has an affinity to various endogenous molecules. MR plays an important role in microbial phagocytosis and clearance of a number of endogenous glycoproteins from the circulation and mediates the uptake of ligands for the purposes of both homeostasis and immunity [3].

The affinity of MR interaction with oligosaccharides is determined by the nature of the terminal residue found on the glycan; the highest affinity is for l-fucose followed by d-mannose and d-N-acetylglucosamine, and the lowest for d-galactose [3, 16].

It was demonstrated that uptake of chemically mannosylated and fucosylated bovine serum albumin (BSA) by cultured SECs is primarily driven through uptake by MR. Galactosylated BSA did not compete in binding with the receptor, demonstrating poor affinity for glycans terminating with galactose [22].

Upon interaction with its target, the MR-ligand complex is rapidly internalized into the cell in early endosomes where the ligand dissociates from MR. MR is then recycled back to the cell surface therefore maintaining high overall binding capacity. Bound ligand or MR-ligand complex can also be transported into the lysosomal compartment followed by proteolytic degradation [3, 22]. Subcellular fractionation of the liver indicated that the internalized ligand is transported to the lysosomes. Due to the ability of MR to quickly recycle, ligand uptake by cells is essentially continuous [19].

Another member of MBPs, a soluble mannose binding lectin (MBL), binds agalactosylated glycoforms of IgG (IgG–G0F), polymeric forms of IgA and certain glycoforms of IgM which have a high incidence of GlcNAc-terminating glycans. This interaction provides a route for clearance of immune complexes from the serum [18].

Selective clearance caused by tGlcNAc was demonstrated for a complex glycoprotein pharmaceutical that contains human IgG1-Fc domain and an extensively glycosylated extracellular domain of tumor necrosis factor receptor p55 [4]. Analysis of glycan distribution on the compound following re-purification from PK study samples demonstrated that glycans carrying tGlcNAc were preferentially cleared from circulation. Other glycans, including those with sialic acid and terminal galactose, showed only small changes. The data demonstrated a strong role of tGlcNAc and MR in the half-life of the glycosylated protein.

Galactose and sialic acid residues were removed from normal IgG–G2F via enzymatic procedures exposing tGlcNAc residues [23]. It was estimated that more than 75 % of galactose residues were removed by the treatment despite the fact that enzymatic removal of terminal galactose residues is challenging and possibly mAb dependent. Importantly, a correlation between the fraction of G0F found on IgG and binding affinity with soluble MBL was reported. Enzymatic removal of terminal galactose residues increased uptake of soluble IgG by MR on macrophages and dendritic cells. In contrast, enzymatically treated IgG was not taken up by B cells, consistent with the lack of MR exposure. The data suggested a substantial modulatory effect of the G0F containing Fc-glycoforms on the ability of MBL to bind IgG. This could be a result of increased degree in mobility of agalactosylated oligosaccharides as compared to G1F and G2F structures [2].

MR was first identified in the liver on SECs and as a specific uptake system in rat liver Kupffer cells for mannosylated/tGlcNAc and fucosylated glycoproteins in vivo. MR has since been found to be widely expressed among different tissues and has been found on the surface of various cell types [3, 5]. The tissue and subcellular distribution of MR suggests it is appropriately located to serve as a high-efficiency antigen uptake receptor of antigen presenting cells [5].

MR has been shown to play an important role in the mechanism of clearance for several biotherapeutic compounds. A mannosylated antibody–enzyme fusion protein, P. pastoris-derived mannosylated protein (MFECP1) evaluated for cancer treatment [24] was cleared by MR located on liver SECs. Clearance was inhibited in vivo by mannan, a MR inhibitor. Inhibitors of MR and the LDL receptor-related protein (LRP) considerably slowed clearance of tissue plasminogen activator (TPA) [25]. MR located on endothelial liver cells and LRP on parenchymal liver cells were reported to contribute to liver uptake. LRP and MR were viewed as major contributors to TPA clearance. Additionally, co-administration of MR inhibitor, a cluster mannosidase carrying six mannose groups, was used to extend exposure to TPA [4].

An analysis of the targeting of mannose-terminal glucocerebrosidase demonstrated predominant uptake by liver ECs [26]. A modified variant of glucocerebrosidase, alglucerase, used to treat Gaucher’s disease, an inherited lysosomal storage disorder, contains glycans with terminal and exposed mannose residues designed to promote uptake by target macrophages. The liver was the main tissue where uptake was observed (65.6 % uptake by liver ECs). Smaller fractions of 31 and 8.1 % of the enzyme were found in parenchymal and Kupffer cells, respectively. MR uptake of mannan substantially (~9-fold) increased the half-life of the enzyme in the plasma compartment.

New multi-valent, carbohydrate ligands that contain terminal N-acetylgalactosamine (GalNAc) were shown to have a high affinity for rat hepatocytes. Administration of intravenous infusion of Indium¹¹¹-tagged hexa-valent lactoside in rats and mice resulted in highly specific and nearly exclusive accumulation of radioactivity in the liver [27].

Overall, it has been concluded that MR-mediated clearance may significantly contribute to the mechanism of clearance of native and therapeutic glycoproteins.

Non-target toxicity of antibody–drug-conjugated compounds

In recent years, there has been an increase in applying biotherapeutic compounds to treat various neoplastic diseases. In a growing subset of biotherapeutics, a toxic payload (toxin) is delivered by an antibody that is specific to a tumor cell surface target, ideally, minimizing normal tissue exposure, reducing off-target toxicity and improving efficacy in the clinic. ADCs are composed of a cytotoxic or cytostatic agent (payload) that is attached to a target-specific antibody carrier via a linker. The traditional mechanism of action for ADCs includes several steps: The antibody binds to the cell surface target, the target-ADC complex is internalized and the payload is released within the targeted cell. The ideal “smart-linker” remains stable in circulation releasing the payload component once internalized by the cell. Recently developed linker types include disulfide, acid-labile hydrazone, peptide and thioether [28]. Payloads have included a range of radioisotopes, small molecular weight chemicals and bacterial toxins. Presently, radioisotope ADCs, anti-CD33-specific ADC conjugated to calicheamicin [29] and anti-CD30-specific ADC conjugated to monomethylauristatin E (Brentuximab, Vedotin, Adcetris^®) have been approved for the treatment of hematologic malignancies. Other ADCs currently in development have been reviewed elsewhere [28].

ADCs are developed with the goal of increasing compound therapeutic index by specific and targeted delivery of the toxic payload to the site of action while considerably reducing damage to normal tissues, that is, reducing off-target toxicity. Yet, off-target toxicity has been reported for many ADCs [6–12]. Off-target toxicity is defined as damage to the tissue not expressing the mAb target [6]. The exact mechanism of off-target toxicities is often not publically available, although internal investigations could be conducted. Damage to liver tissue is the most commonly described off-target toxicity [6].

One of the most typical examples of compounds with considerable off-target toxicities are ADCs with ricin toxin as the active payload. Ricin is a toxic protein containing two peptide domains: A and B. The A domain inhibits protein synthesis and is responsible for the toxic activity, while the B domain functions as a carrier delivering toxin via binding to cell surface glycan receptors.

Ricin molecules have been tested in conjugation with antibodies directed to anti-CD30 [7], human immune deficit virus antigens [30], colon cancer [31] and breast carcinoma cell targets [32]. The high potency of ricin toxin made it an attractive payload when developing ADCs while the off-target toxicity remained a substantial obstacle. In animal studies, liver tissue damage including hepatic necrosis has been reported [7]. Pharmacokinetics of ricin-based ADCs, including overall exposure values, have been effected as a result of quick uptake by hepatic tissue [33]. Antibody-ricin A-chain conjugates are rapidly removed from the circulation, primarily by liver cells [33].

The principle mechanism of hepatic damage is explained by hepatic uptake of ricin via MRs [34]. The B-chain of ricin contains high mannose type oligosaccharides, while the A-chain contains the complex unit (GlcNAc)₂-Fuc-Xyl-(Man)_4–6. Such glycans can be recognized by MRs located on various cell types, including hepatic cells. In a radiolabeled biodistribution study, a faster clearance of immunotoxin with a high ricin A-chain/antibody ratio as compared to unconjugated antibody was demonstrated. It was concluded that hepatic uptake is dependent on the Kupffer cell recognition of mannose-containing oligosaccharide structures on the ricin A moiety [33].

Mannose-containing blocking agents (ovalbumin, ovomucoid, mannosyl-lysine, d-mannose or l-fucose) given with ricin-antibody conjugate were shown to extend circulation half-life of the immunotoxin due to reduced liver uptake. Uptake was not inhibited by d-galactose saccharide [34].

Importantly, a carbohydrate-free form of ricin A-chain conjugate with a mAb was poorly taken up in vivo and in vitro by both hepatic parenchymal and non-parenchymal cells [33]. In vitro uptake of the glycan-free form of ricin A-antibody conjugate was unaffected by mannose [30]. Chemical deglycosylation was reported to reduce non-specific toxicity of ricin-antibody conjugates.

Gemtuzumab ozogamicin (Mylotarg^®) consists of a semisynthetic derivative of calicheamicin, a cytotoxic agent linked to a recombinant mAb directed against the CD33-antigen found on leukemic myeloblasts. Mylotarg^® was developed for treatment of relapsed acute myeloid leukemia (AML) in older patients [35] and was approved for marketing in 2000 by the United States Food and Drug Administration (FDA). It demonstrated moderate activity as a single agent in patients with CD33-positive refractory or relapsed AML [11]. The main type of observed toxicity was hematologic, which is expected for a CD33-directed cytotoxic therapy. Liver off-target toxicity was widely reported [11, 35]. Specifically, sinusoidal obstruction syndrome [SOS, referred to as venoocclusive disease (VOD)] was reported as a significant toxicity of Mylotarg^®-based combinations. SOS occurred when Mylotarg^® was used either as a single agent or when given with other cytotoxic agents [35].

Liver toxicity of Mylotarg^® revealed sinusoidal injury with extensive sinusoidal fibrosis [36]. Endothelial cell damage resulted in deposition of sinusoidal collagen, centrilobular congestion and hepatocyte necrosis. It was speculated that Mylotarg^® targets CD33 cells residing in hepatic sinusoids and triggers a series of cellular events that lead to sinusoidal obstruction rather than directly causing hepatocyte necrosis [36]. The authors admitted that this explanation required further confirmation and did not present any direct evidence. A targeted delivery of calicheamicin to hepatic sinusoids, via uptake by MR, can provide an alternate explanation to the observed liver toxicity phenomenon. Calicheamicins contain saccharides and could potentially interact with cell surface lectins, including MR, resulting in non-specific cellular uptake and off-target toxicity.

Importantly, due to a complete or partial release of the payload, off-target toxicity may be observed even when overall PK of the mAb component of the ADC or even the ADC is not significantly affected due to FcRn driven recycling mechanisms. The role of glycans in modulating clearance is not well understood. Studies of glycoprotein clearance have shown that alterations in the glycan structures found on IgG molecules can cause changes in the PK properties of immunoglobulins but the overall conclusions from these studies are conflicting. It has been demonstrated that therapeutic IgGs containing Fc associated high-mannose glycans are cleared more rapidly in humans than immunoglobulins with other glycan forms [37, 38]. Yet the overall impact on the drug exposure was shown to be minor and dependent upon the dosing regimen and relative level of the high-mannose glycan content found on the recombinant antibodies. The N-linked Fc glycans present on wild-type IgG molecules (containing low G0F component) are partially buried in the CH2 domain making them inaccessible to asialoglycoprotein and mannose receptors that could recognize the moieties. Even if immunoglobulin internalization occurs based on MR uptake, the FcRn driven mechanism of immunoglobulin recycling can effectively protect IgG molecules from intracellular degradation.

Uptake by MR will lead to presentation of the bound protein ligand, in this case ADC, to the endosomal and lysosomal compartments with an acidic environment. ADCs where the linker is designed to release the payload at low-pH conditions may therefore deconjugate under such conditions resulting in non-specific delivery of the toxic payload to normal tissue and in various types of off-target toxicities. Hepatic toxicities observed for Mylotarg^® are consistent with the initial damage to SECs. It is proposed that at least part of the off-target toxicity of Mylotarg^® can be explained by the G0F glycan–MR interaction.

Other antibody-toxin conjugates have been described where liver toxicity was identified as one of the major types of off-target damage [6–8]. Reported examples include effects of radioisotope labeled mAb [39]. Pretreating patients with unlabeled antibody was suggested as a way to reduce hepatic uptake of radioisotope-antibody conjugate, where unlabeled antibody was used to saturate non-target-related binding sites and therefore minimizing off-target binding of radiolabeled antibody to the liver [39]. MR is rapidly internalized and recirculated to the cell surface after engaging with the glycoprotein ligand. It would arguably be difficult to successfully saturate MR binding sites, if these are in fact the true culprits of off-target toxicity.

Drug biodistribution studies may facilitate in better understanding the true cause of off-target toxicity and have been reported for several ADCs [9]. Reported biodistribution studies support the proposal that liver tissue and some of the specific hepatic cell types may be internalizing or accumulating ADCs. The accumulation of antibody-saporin conjugate and its components by rat liver parenchymal and non-parenchymal cells was investigated [40]. It was suggested that the sensitivity of the liver cells is proportional to cellular uptake.

Immunoconjugate BMS-182248, a chimeric variant of anti-LewisY mAb conjugated with doxorubicin, was evaluated for the treatment of human lung adenocarcinoma [9]. Biodistribution of BMS-182248 was analyzed using athymic mice implanted with human tumor tissue after animals received several doses of the compound. The majority of the toxin found in plasma remained antibody conjugated, with relatively low levels of free toxin present. The conjugated form of the toxin was found in tumor, liver and heart tissue in order of decreasing concentration. While the tumor tissue retention of BMS-182248 was shown to be driven by antigen-specific binding, the exact mechanism of the non-target-related distribution was not discussed. Interestingly, another LewisY-directed mAb conjugated to calicheamicin was tested in a human biodistribution and PK study [41]. Biodistribution images showed a significant hepatic uptake for the ADC but not the parental antibody (not toxin conjugated).

Reported toxicities and information available about biodistribution of many of the ADCs and radiolabeled compounds suggest that MR uptake may play an important role in non-target tissue damage. MRs have evolved to quickly internalize and recycle back to the cell surface making it difficult to completely saturate binding capacity. MRs are also found on many normal cells and tissues. Cellular uptake of glycoproteins is directly linked to the expression of MRs on the cell type and the presence of certain types of glycans on glycoproteins, including the presence of agalactosylated components on immunoglobulins.

It is therefore proposed that the elevated G0F content on the mAb moiety of ADCs and similar biotherapeutics may lead to an important mechanism of off-target toxicity. Addressing the glycosylation pattern on the mAb component of the ADCs may therefore be an important consideration in the successful design of biotherapeutics.

Conclusion

Monoclonal antibodies developed as biotherapeutic compounds commonly contain glycans similar to those found on natural, serum-derived immunoglobulins. Within the Fc portion of the molecule, glycans consist of biantennary core oligosaccharides with various degrees of terminal galactosylation (G0F, G1F and G2F). Interestingly, the relative proportion of G0F/G1F/G2F in mAb-based biotherapeutics is closer to some disease state IgG rather than to a normal serum-derived immunoglobulin. Cell surface receptors, including MR, have evolved to effectively capture and remove proteins and immune complexes containing high-glycan structures that are agalactosylated. In this paper, it is proposed that G0F–MR interaction-based cellular uptake can be responsible for at least some of the off-target toxicity of ADCs where the linker is sensitive to conditions similar to those found in the endosomal and lysosomal compartments. MR is expressed on the surface of various cell types, including locations indicated as sites for reported off-target ADC toxicities. Off-target toxicity may be observed even in the absence of an altered PK profile. It is proposed that the mechanism of the off-target toxicities for ADCs is at least in part driven by the carbohydrates, specifically agalactosylated glycans, such as G0F, their interactions with cell surface lectins, for example MR and resulting glycan derived cellular uptake of ADCs.