Plant‐derived Durvalumab variants show efficient PD‐1/PD‐L1 blockade and therapeutically favourable FcR binding

Summary Immune checkpoint blocking therapy targeting the PD‐1/PD‐L1 inhibitory signalling pathway has produced encouraging results in the treatment of a variety of cancers. Durvalumab (Imfinzi®) targeting PD‐L1 is currently used for immunotherapy of several tumour malignancies. The Fc region of this IgG1 antibody has been engineered to reduce FcγR interactions with the aim of enhancing blockade of PD‐1/PD‐L1 interactions without the depletion of PD‐L1‐expressing immune cells. Here, we used Nicotiana benthamiana to produce four variants of Durvalumab (DL): wild‐type IgG1 and its ‘Fc‐effector‐silent’ variant (LALAPG) carrying further modifications to increase antibody half‐life (YTE); IgG4S228P and its variant (PVA) with Fc mutations to decrease binding to FcγRI. In addition, DL variants were produced with two distinct glycosylation profiles: afucosylated and decorated with α1,6‐core fucose. Plant‐derived DL variants were compared to the therapeutic antibody regarding their ability to (i) bind to PD‐L1, (ii) block PD‐1/PD‐L1 inhibitory signalling and (iii) engage with the neonatal Fc receptor (FcRn) and various Fcγ receptors. It was found that plant‐derived DL variants bind to recombinant PD‐L1 and to PD‐L1 expressed in gastrointestinal cancer cells and are able to effectively block its interaction with PD‐1 on T cells, thereby enhancing their activation. Furthermore, we show a positive impact of Fc amino acid mutations and core fucosylation on DL's therapeutic potential. Compared to Imfinzi®, DL‐IgG1 (LALAPG) and DL‐IgG4 (PVA)S228P show lower affinity to CD32B inhibitory receptor which can be therapeutically favourable. Importantly, DL‐IgG1 (LALAPG) also shows enhanced binding to FcRn, a key determinant of serum half‐life of IgGs.


Introduction
Immunological checkpoint inhibitors have emerged as novel therapy for cancer treatment (Li et al., 2018).Immune checkpoints are an essential part of the immune system and represent a group of membrane proteins expressed on effector cells, composed of multiple co-inhibitory and co-stimulatory pathways (Chen and Flies, 2013;Croft, 2003).Programmed cell death 1 (PD-1) is one of the most effective T-cell immune checkpoint molecules (Fife and Bluestone, 2008).PD-1 is a transmembrane protein receptor expressed on the surface of a variety of immune cells, including monocytes, T and B cells, acting as an 'off switch' to prevent excessive inflammation and maintaining immune tolerance to self-antigens under normal conditions (Keir et al., 2008).It does this when it attaches to its ligands, programmed cell death ligand PD-L1 and PD-L2, expressed on a variety of cells, including antigen-presenting cells (Keir et al., 2008).Engagement of PD-1 by PD-L1 transmits inhibitory signals that suppress T-cell function which can be reversed once signalling through PD-1 ceases.These inhibitory mechanisms should prevent aberrant immune responses and protect healthy tissues.However, by overexpressing PD-L1, tumour cells take advantage of this pathway to escape immune surveillance (Han et al., 2020;Zou and Chen, 2008).
PD-L1 is up-regulated in the tumour microenvironment and is found in a large variety of tumour cells.Therefore, immune checkpoint blockade can be a highly effective therapeutic strategy (Constantinidou et al., 2019;Iwai et al., 2017;Ribas and Wolchok, 2018).PD-L1 expression in tumours is currently used as a biomarker to inform therapeutic decisions.Monoclonal antibodies (mAbs) acting as PD-1 and PD-L1 inhibitors are a group of immunotherapy drugs that block the interaction between PD-1 and PD-L1 preventing the 'off' signal from being transmitted and, therefore, boosting T-cell immunity against a variety of human cancers (Gong et al., 2018;Waldman et al., 2020).Therapeutic antibodies used for the treatment of cancer patients normally belong to the IgG1 subclass and induce tumour destruction via the recruitment of immune effector cells by the Fc and hinge regions.However, antibodies have different requirements for Fcc receptor engagement to attain optimal anti-tumour activity (Yu et al., 2020).For blocking antibodies, such as the ones targeting PD-1 and PD-L1 which are not tumour-specific, the cytotoxicity brought about by antibody-dependent cellular cytotoxicity and phagocytosis (ADCC/ADCP) should be averted (Boulard et al., 2022).
A comparative study of FDA-approved antibodies targeting the PD-1/PD-L1 axis showed that PD-L1 antibodies seem to be superior to PD-1 antibodies in reverting PD-1 signalling (De Sousa Linhares et al., 2019).
All current FDA-approved PD-L1 antibodies are from the IgG1 isotype.Atezolizumab is engineered with a mutation of N297 to prevent N-glycosylation and thus exhibits decreased FccR-binding and effector functions (Herbst et al., 2014).However, the pronounced aggregation tendency of the non-glycosylated protein compromises the therapeutic efficacy of Atezolizumab.Durvalumab (DL) is engineered to contain amino acid substitutions that reduce Fc-mediated effector functions against cells expressing PD-L1 (Oganesyan et al., 2008;Stewart et al., 2015) while Avelumab retains intact Fc functions and can induce ADCC as a part of its mechanism of action (Hamilton and Rath, 2017).However, despite exerting enhanced cytotoxic effects towards PD-L1-expressing tumour cells, the effector function can diminish antitumor responses by attacking activated effector T cells and other immune cells, which also express high levels of PD-L1 (Knorr and Ravetch, 2019;Leitner et al., 2021).
PD-L1 is heavily glycosylated and glycosylation affects its binding to PD-1 and to diagnostic antibodies leading to inaccurate readouts (Lee et al., 2019).Removal of PD-L1 Nlinked glycosylation by enzymatic digestion eliminates structural hindrance and improves antibody-based detection and binding of Atezolizumab in lung cancer cells (Lee et al., 2019).In contrast, for breast cancer cells, binding of all three therapeutic PD-L1 antibodies favoured glycosylated PD-L1 over the non-glycosylated protein, and DL showed the highest affinity (Benicky et al., 2021).
Traditionally, therapeutic mAbs are manufactured in mammalian cell factories.Recently, plant-based expression platforms have emerged as cost-effective alternatives to produce high-quality proteins for research, diagnostic and therapeutic applications.Many plant-made pharmaceuticals have entered clinical testing, including antibodies for the treatment of cancer (Nessa et al., 2020;Shanmugaraj et al., 2020).In addition, advances in plant engineering have resulted in the ability to produce antibodies with tailor-made glycans (Castilho et al., 2015).mAbs produced in plants differ only in Fc N-glycosylation when compared to those produced in mammalian cells.When expressed in wild-type plants, mAbs are decorated with complex glycans carrying plant-specific glycosylation (core a1,3-fucose and a1,2-xylose).These sugar residues do not appear to affect the safety and efficacy of the mAbs but they are considered as cross-reactive carbohydrate determinants (CCDs) (Platts-Mills et al., 2021).

Generation of Durvalumab HC-variants
Durvalumab (Imfinzi â ) is a fully human IgG1 kappa antibody produced in Chinese hamster ovary (CHO) cells and engineered to contain three amino acid substitutions in the hinge region (L234F/L235E/P331S) to reduce FccR binding and ADCC or CDC cytotoxicity (Oganesyan et al., 2008;Stewart et al., 2015).The Fc domain is glycosylated with mainly biantennary complex-type Nglycans.
The completely Fc-silent DL-IgG4 S228P carries a combination of six mutations in the hinge region: S228P to prevent half-mAb formation and Fab-arm exchange (Silva et al., 2015); Y219C and G220C to increase the stability of the heavy-heavy chain interaction via incorporation of additional disulphide bonds (Handlogten et al., 2020); and E233P/F234V/P235A (PVA) to reduce the ability to bind to CD64 (Zhang et al., 2018).All DL HC variants carry the conserved glycosylation site (N297) in the CH2 domain known to impact their binding affinity to FccRs (Arnold et al., 2007).
DL variants were transiently expressed in N. benthamiana, a tobacco-related plant species widely used for recombinant protein expression, using a viral-based vector expression system developed by Icon Genetics (magnICON) (Castilho et al., 2015;Klimyuk et al., 2014).

Expression and glycoengineering of Durvalumab HC variants
Tobacco mosaic virus-based (TMVa) vectors carrying the DL HC variants were mixed with the potato virus X-based (PVXa)-DL-LC (1 : 1) and co-expressed in the N. benthamiana glycosylation mutant ΔXF (Strasser et al., 2008) via agroinfiltration.Total soluble proteins extracted after 4 days postinfiltration (4 dpi) were used to purify the antibodies via affinity chromatography with protein A. The quality and purity of plant-derived DL were confirmed by SDS-PAGE.Coomassie brilliant blue (CBB) staining of SDS-PAGE gels run under reducing conditions showed two bands representing the heavy and light chains which was confirmed by Western blotting using anti-human gamma and anti-human kappa antibodies (Figure 1b).Both chains seem to be stable in planta with no significant degradation observed.Size exclusion chromatography (SEC) showed similar profiles for all DL variants with a major dimer peak (>80%) indicating that the Plant-derived Durvalumab with favourable FcR binding 1225 majority of the protein is correctly assembled (Figure 1c).The overall yield after protein A purification (measured by UV absorbance) was similar for all DL variants (range: 0.6-1.0g/kg leaf wet weight).
DL produced in CHO cells is glycosylated with nearly all glycans core fucosylated (Figure 2).In contrast to mammalian cells that synthesize core fucosylation in a1,6-linkage, plant N-glycans are modified with a1,3-linked core fucose residues.Here, we transiently expressed DL variants in N. benthamiana ΔXF to produce afucosylated DL and over-express the human a1,6fucosyltransferase (FUT8, Castilho et al., 2015) to generate DL glyco-variants with human-like core fucosylation.Liquid chromatography-electrospray ionization-mass spectrometry (LC-ESI-MS) was used to compare the N-glycosylation profile of DL (Imfinzi â ) and plant-derived DL glyco-variants (Figure 2).DL HC variants expressed in N. benthamiana ΔXF are mainly decorated with N-glycans terminated by N-acetylglucosamine residues (GnGn) and lacking core fucose residues while co-expression of FUT8 resulted in the synthesis of GnGn structures decorated with human-like core a1,6-fucose (GnGnF6) and thus resembling DL (Imfinzi â ).High-mannose glycans (Man8 and Man9) were also detected in plant-derived DL variants (up to 25%).Compared to DL (Imfinzi â ) (0.5%), the percentage of unglycosylated protein is higher for plant-derived DL (up to 20%) (Figure 2).This underglycosylation is probably due to insufficient activity of the endogenous oligosaccharyltransferase (OST) complex responsible for the en bloc transfer of the N-glycan precursor Glc 3 Man 9 GlcNAc 2 to the consensus sequence for N-glycosylation of nascent polypeptides.Indeed, we have shown that expression of the single-subunit OST of Leishmania major (LmSTT3D) in plants substantially improves the N-glycosylation efficiency of several recombinant proteins including antibodies (Castilho et al., 2018).

Expression and glycoengineering of PD-L1
The human programmed cell death 1 ligand (PD-L1) consists of an ectodomain encompassing an immunoglobulin V-like domain (IgV), an immunoglobulin C-like domain (IgC), a transmembrane domain (TM) and a short intracellular tail (Dong et al., 2002).Here, we expressed the extracellular domain (ECD) of PD-L1 with a C-terminal His-tag (PD-L1 His ).Secreted proteins accumulating in the apoplastic fluid (AF) and analysed by SDS-PAGE showed that PD-L1 His is efficiently secreted.A broad protein band of ~40 kDa was detected in the AF of leaves infiltrated with agrobacteria encoding PD-L1 His (Figure S1A).Additional protein bands also visible in the secretome of leaves infiltrated with an empty vector result from agrobacteria infection.Recombinant PD-L1 His was efficiently purified from AF (Figure S1A) with yields of up to 60 lg/mL of AF (which corresponds to ~40 mg/kg leaf wet weight).The apparent molecular weight of plant-derived PD-L1 His is higher than its calculated molecular mass (~33 kDa), which is most probably due to protein glycosylation.Indeed, a shift in protein size is observed upon PNGase F digestion providing strong evidence that plant-derived PD-L1 His is glycosylated (Figure S1A).The soluble ECD of PD-L1 has been suggested to form weak homodimers (Chen et al., 2010).SDS-PAGE analysis under nonreducing conditions revealed that the plant-derived PD-L1 His is mainly monomeric (Figure S1B).
PD-L1 carries four potential N-glycosylation sites (N35, N192, N200 and N219) within the ECD.LC-ESI-MS analysis of PD-L1 His expressed in N. benthamiana ΔXF showed that the protein is mainly decorated with complex neutral glycans (GnGn).Mannosidic (Man5) and hybrid (Man5Gn) glycans were only found for glycosite N192.Overall, the extent of non-glycosylation is rather high (up to 40%) and glycosite N219 is mostly non-glycosylated (Figure S1C).

Binding of Durvalumab variants to PD-L1
The ability of plant-derived DL variants to bind to recombinant PD-L1 His was evaluated by immunoblotting and enzyme-linked immunosorbent assays (ELISA).Western blot results showed that all three DL variants specifically bind PD-L1 His (Figure S2A).Half maximal effective concentration (EC 50 ) values in ELISA binding assays confirmed that all plant-derived DL variants have similar PD-L1 His binding capacities as DL (Imfinzi â ) (Table 1 and  Figure S2B).Overall, these results indicate the integrity of the plant-produced DL proteins.
Alterations in glycosylation patterns, particularly increased sialylation, are a hallmark of cancer cells (Bellis et al., 2022;Pinho and Reis, 2015;Thomas et al., 2021).It has been shown that the steric hindrance of PD-L1 glycosylation significantly affects antibody recognition, with implications in patient therapeutic stratification (Lee et al., 2019).Therefore, we next evaluated the ability of plant-derived DL variants to bind to cell surface PD-L1 expressed by two gastrointestinal cancer cells upon in vitro interferon gamma (IFNc) stimulation: the metastatic gastric carcinoma cell line NCI-N87 and the colorectal carcinoma cell line SW48.Both cell lines have low basal expression of PD-L1, as demonstrated by flow cytometry (Figure 3a,b).Following in vitro IFNc stimulation, both cell lines start expressing similar PD-L1 levels at the cell surface in more than 80% of cells.We then assessed the ability of plant-derived DL variants to bind to PD-L1 expressing NCI-N87 and SW48 cells.Both cell lines showed significant binding to all DL variants and clinical-grade DL (Imfinzi â ) (Figure 3c,d).DL-IgG1 (GnGn), DL-IgG1 (GnGnF6) and DL-IgG1 (LALAPG) (GnGnF6) showed similar binding to clinicalgrade DL.The variants DL-IgG4 (PVA) S288P (GnGn), DL-IgG1 (LALAPG) (GnGn) and DL-IgG4 (PVA) S288P (GnGnF6) showed 20% decreased binding to both cell lines, when compared to clinicalgrade DL.Overall, the binding of plant-derived DL variants to cell surface PD-L1 was not affected by IgG isotype.Fc core fucosylation, however, influenced the binding of some plantderived DL variants.In particular, there was a significant increase in the binding of DL-IgG1 (LALAPG) (GnGnF6) to both cell lines as compared to DL-IgG1 (LALAPG) (GnGn) (Figure 3c,d).

Functional evaluation of Durvalumab variants
We used a previously described PD-1 reporter T-cell line (JE6-1-NF-kB::eGFP-PD1) to evaluate the antagonistic capacity of the plant-expressed DL variants in a functional assay (De Sousa Linhares et al., 2019).The PD-1 reporter cells were co-cultured with K562-based stimulator cells (K562S) expressing a membrane-bound anti-CD3 fragment, which engages the TCR-CD3 complex on the reporter T cells, thereby inducing NF-kB:: eGFP reporter gene expression.K562S-PD-L1 co-expresses PD-L1, which engages PD-1 on the reporter cells resulting in a significant inhibition of eGFP expression (Figure 4a).The presence of DL (Imfinzi â ) blocked PD-1 inhibition resulting in a dose-dependent increase in eGFP expression (Figure 4b).Plant-produced DL variants also efficiently reverted PD-1 mediated reporter inhibition similarly to the therapeutic antibody and independently of Fc glycosylation (Figure 4c and Figure S3A).Compared to other DL variants, EC 50 values for fucosylated DL-IgG4 (PVA) S288P were slightly higher.Similar results were obtained with a different batch of fucosylated DL-IgG4 (PVA) S288P excluding a batch effect.Interestingly, the EC 50 values obtained for the fucosylated DL-IgG4 S228P variant lacking the PVA mutation were also higher (EC50 41.4) (Figure S3B).

Binding of DL variants to human Fcc receptors
The choice of IgG heavy chain subclass and mutations of Fc 'hotspots', particularly at the hinge and upper CH2 domains, strongly affects the binding of antibodies to Fcc receptors and therefore has utmost pharmacological significance.In addition, Fc N-glycosylation has been shown to affect the affinity of IgG for all FccR classes (CD64, CD32 and CD16) and to C1q (Hayes et al., 2014).We used a flow cytometry-based assay to test the binding of our plant-derived DL variants to cells expressing high levels of different human FccRs-CD16A, the natural V176 highaffinity variant of CD16A (C16V), CD32A, CD32B and CD64 (Leitner et al., 2023).Cells not expressing human FccR were used as a control.Overall, DL-IgG1 showed the highest binding to all FccRs independently of glycosylation, although core fucosylation reduces the affinity (Figure 5a and Figure S4A).Core fucosylation also significantly decreased the affinity of DL-IgG1 (LALAPG) to all receptors except for CD32A (Figure 5a and Figure S4B) and the binding of DL-IgG4 (PVA) S288P particularly to CD16 and CD64 (Figure 5a and Figure S4C).
In order to evaluate the impact of the mutations introduced into DL-IgG4 (PVA) S288P on FccR binding, we compared it to DL-IgG4 S228P .The results showed that the E233P/F234V/P235A substitutions had no impact on binding to CD32A, while affinity to all other receptors was drastically decreased (up to 30-fold) (Figure S4D).

Binding to human FcRn
The circulating half-life of serum antibodies is roughly 10-21 days, depending on IgG isotype and attributes of the variable region.Fcsilent DL (Imfinzi â ) is within this range but a considerably lower halflife was reported for Avelumab, a wild-type IgG1 (De Sousa Linhares et al., 2019).The Fc contains the binding site for the neonatal Fc receptor (FcRn), a key determinant in extending the plasma half-life of antibodies by recycling them through transcytosis pathways (Pyzik et al., 2019).
Enhanced affinity to FcRn has been achieved through Fc mutagenesis approaches, and the M252Y/S254T/T256E (YTE) triple mutation (Dall'Acqua et al., 2002) has consistently shown a twofold to 11-fold improvement in FcRn binding at pH 6.0.
Surface plasmon resonance (SPR) was used to determine the binding kinetics of human FcRn to DL variants decorated with core a1,6-fucosylated glycans (GnGnF6).The results (Table 2 and Table S1 and Figure S5) showed that DL (Imfinzi â ) and DL-IgG4 (PVA) S288P have K D values higher than the other variants.To determine whether mutations introduced in DL-IgG4 (PVA) S288P were affecting its binding to FcRn, we also analysed DL-IgG4 S228P , which displayed a similar K D value as its PVA counterpart (Table 2 and Table S1 and Figure S5).DL-IgG1 has a slightly lower K D value and DL-IgG1(LALAPG) carrying the additional YTE triple mutation   2 and Table S1 and Figure S5).

Discussion
The PD-1/PD-L1 immune regulatory axis has a central role in the suppression of anti-tumour immunity (Baumeister et al., 2016).In recent years, the therapeutic potential of PD-1/PD-L1 blockade has been demonstrated in multiple human clinical trials and achieved remarkable clinical success in multiple solid cancers (Li et al., 2018).
In cancer therapy, antibodies of the IgG1 subclass favour tumour destruction through the recruitment of immune effectors.The role of the Fc-mediated effector functions in cancer therapy using PD-L1 blocking antibodies has been discussed controversially (Cohen Saban et al., 2023;Knorr and Ravetch, 2019;Leitner et al., 2021;Yu et al., 2020).While current human PD-1 therapeutics are based on IgG4 to minimize antibody-dependent cell lysis, with the exception of Avelumab, PD-L1 blockers are IgG1 mAbs with reduced Fc-effector function via amino acid substitutions on the FccR-binding domain or 'null' glycosylation.
The biological activity of mAbs is actively modulated by their glycosylation profile (Mastrangeli et al., 2019).Fc core fucosylation is implicated in FccRIIIa binding and negatively affects ADCC activity; terminal galactose was reported to be involved in FccRIIIa binding but not impacting ADCC activity, while the function of terminal sialylation is still being discussed (Mastrangeli et al., 2019;Park et al., 2020).
Improving efficacy and prolonging half-life is an important strategy to make checkpoint inhibitors more accessible and affordable.Mammalian cells are traditionally used for mAb production.However, in recent years, plants have carved their way as suitable alternative expression platforms.Plant expression systems have several advantages over conventional expression systems.The cultivation of plants under controlled conditions ensures the production of high-quality protein in accordance with GMP (Good Manufacturing Practice) standards at lower costs (Buyel and Fischer, 2012;Diamos et al., 2019;Ecker et al., 2015;Ma et al., 2017;Ridgley et al., 2023).Moreover, the rapid growth, low cost of nutrient components and the avoidance of risks from contaminations with viruses and prions of animal origin are other arguments in favour of plantderived therapeutics.Another important benefit of plant-based systems is that in comparison with mammalian-derived proteins which have a highly heterogeneous glycosylation, plant-derived therapeutics normally carry well-defined, homogeneous and consistent glycosylation patterns.Transient expression in N. benthamiana is particularly well suited for the generation of human antibodies at high yields and with tailored glycosylation profiles (Castilho et al., 2011(Castilho et al., , 2015;;Loos and Steinkellner, 2012;Montero-Morales and Steinkellner, 2018).
Nicotiana benthamiana plants lacking endogenous b1,2-xylose and core a1,3-fucose residues (DXF, (Strasser et al., 2008) have been used to produce a variety of antibodies with enhanced immune receptor binding and greater potency compared to commercial antibodies, including the well-described ZMAPP TM , an antibody cocktail developed for the treatment of Ebola virus infection (Zeitlin et al., 2011).In contrast, blocking antibodies with engineered Fcs designed to eliminate effector functions and reduce dose/administration frequency represent hold promise as replacements of current PD-1 and PD-L1 therapeutic antibodies.
DL variants expressed in N. benthamiana DXF show high integrity with no significant degradation or aggregation.Importantly, DL could be purified with yields up to 1 g/kg of fresh leaves, without the need for optimizing expression.Such recovery rates are significantly higher compared to previous reports on antibody accumulation levels (before purification) of 0.3 g/kg (Phakham et al., 2021) and 0.7 g/kg upon co-expression with a P19 silencing suppressor to optimize expression (Ridgley et al., 2023).
Apart from afucosylated DLs, we produced glyco-variants that truly mimic the glycosylation profile of therapeutic DL.Coexpression of human a1,6-fucosyltransferase (FUT8) was an efficient approach to produce homogenously fucosylated DL variants where the a1,3-linked core fucose typical of plant glycans was substituted by a non-immunogenic epitope.
All plant-derived DL variants were able to bind to soluble PD-L1, produced in N. benthamiana DXF plants at high yields (40 mg/kg of leaf material).PD-L1 is aberrantly overexpressed in malignancies of several origins (Dong et al., 2002) and possible alterations in the glycosylation pattern of PD-L1 might be related to the acquisition of molecular resistance to targeted therapeutic antibodies (Benicky et al., 2021).One of the most common glycan alterations in malignant cells is hypersialylation (Munkley, 2022).Recent studies showed that increased a2,6-sialylation of cell surface antigens leads to a marked resistance to therapeutic antibodies in gastrointestinal cancer (Duarte et al., 2021;Rodrigues et al., 2021;Yen et al., 2015).Glyco-profiling of PD-L1 expressed in MDA-MB-231 breast cancer cells showed that its complex glycans are fully sialylated (Benicky et al., 2021).
The pattern of expression and potential of PD-L1 as an immunohistochemical biomarker has been extensively studied in gastrointestinal neoplasms (Mastracci et al., 2022) but a comprehensive glycan analysis has not yet been reported.Here, we show that plant-derived DL variants are able to bind PD-L1 expressed in gastric and colorectal cancer cells.Interestingly, compared to DL (Imfinzi â ) and DL-IgG1, the binding of DL-IgG1 (LALAPG) and DL-IgG4 (PVA) S288P to PD-L1-positive gastrointestinal cancer cells was less effective.These differences were not observed in the binding to recombinant soluble PD-L1 and to PD-L1 expressed in the human erythroleukaemia K562 cell line.Recent studies have suggested that both variable and constant domains play a role in antibody-antigen recognition (Janda et al., 2016;Lua et al., 2018).Also, the hinge region allows a high flexibility between the Fab and Fc fragment in order to enhance the accessibility of the Fab fragment for antigen binding (Saphire et al., 2002).At the moment, the reasons underlying this unexpected observation are not clear.However, it is known that even minor structural variations (including Fc glycosylation) may affect antibody conformation, which may, in turn, impact antigen binding (Scallon et al., 2007;Strasser et al., 2009).We cannot exclude that the mutations introduced in IgG1 and IgG4 S288P impact their binding to PD-L1 on the cell surface.To note, DL (Imfinzi â ) also carries mutations in the hinge region (L234F/L235E/P331S).Also, the potential steric hindrance of extensive PD-L1 glycosylation in gastrointestinal cancer cells might affect its recognition by these two variants.
Of note, the cell-specific glycosylation profile of PD-L1 can also determine the binding affinity of PD-L1-targeting mAbs.Sitespecific glycoproteomic profiling of PD-L1 in cancer cells will certainly shed light on the molecular role played by receptor sitespecific glycosylation in the differential binding to DL variants and support the design of novel glycoengineered DL variants capable of circumventing glycan-mediated binding inhibition.
Overall, plant-derived DL variants showed similar ability to block the PD-L1/PD-1 interaction in a dose-dependent manner, comparable to therapeutic DL and independent from glycosylation and Fc mutations.These structural features, however, had a significant impact on the affinity to Fc receptors.Our results corroborated previous findings and demonstrated that apart from amino acid mutations in the Fc region, glycoengineering can also be used to finely tune the binding affinity to different Fcc receptors.We found that DL-IgG1 binds better to all FccRs, and although core fucosylation drastically reduces its binding affinities, it is not sufficient to generate a Fc-effector-silenced IgG.
IgG4 antibodies have naturally reduced affinities for most FccRs and C1q (Crescioli et al., 2016;Jiang et al., 2011) but still retain high affinity to FccRI and binding to FccRIIb, which can result in macrophage-mediated phagocytosis of PD-1-positive T cells (Arlauckas et al., 2017) and decreased anti-tumour activities.DL-IgG4 S228P binds to several Fcc receptors including CD64.For anti-PD-1 antibodies, it has been shown that cross-linking between PD-1 and CD64 could change the function of the antibody from blocking to activating (Zhang et al., 2018).Here, we show that mutations introduced in DL-IgG4 (PVA) S288P reduced the binding to CD16, CD64 and CD32B when compared to DL-IgG4 S228P.Interestingly, DL-IgG4(PVA) S288P carries only three out of the six amino acid substitutions previously reported (Zhang et al., 2018), showing that these mutations within the hinge region are sufficient to modulate FccR affinity.
From all DL variants, core-fucosylated DL-IgG4 (PVA) S288P and DL-IgG1 (LALAPG) showed consistently low affinity to all FccRs.Therapeutic DL (Imfinzi â ) contains a triple mutation (L234F/L235E/P331S) that causes a profound decrease in its binding to CD64, CD32A, CD16V and C1q (Oganesyan et al., 2008).The L234A/L235A/P331G mutations used in this investigation to produce DL-IgG1 (LALAPG) seem to be significantly effective to further reduce the affinity of DL to CD16V and to CD32B.This could be advantageous, since targeted blockade or genetic depletion of the inhibitory FccRIIB receptor have been used to overcome therapeutic resistance and boost activity of antibodies in cancer immunotherapy (Cohen Saban et al., 2023;Teige et al., 2019).Furthermore, introduction of the M252Y/S254T/T256E (YTE) mutations significantly enhanced binding of DL-IgG1 (LALAPG) to the FcRn at pH 6 without impacting the release at pH 7. Higher affinity for the FcRn is a common indicator of a potentially enhanced half-life of mAbs.
A recent study generated a production cost model showing that the expression and purification recovery levels using plant platforms were as competitive as mammalian cell-based platforms and therefore suitable to deliver accessible and more affordable diagnostic and therapeutic proteins (Ridgley et al., 2023).Our research consolidates the scientific, clinical and economic potential of plant-based expression platforms and dampens initial concerns relating to low yields and differences in glycosylation.Importantly, we show that plants can deliver biosimilars to therapeutic DL (Imfinzi â ) with (i) comparable biological effector functions, (ii) reduced interaction with FccR and (iii) enhanced binding to FcRn which could result in optimal FccR-mediated effector functions and a longer half-life in vivo.

Construction of Durvalumab HC variants
The cDNA sequences of wild-type human IgG1 and IgG4 S228P constant heavy chains (CH1-CH3 domains) lacking the variable regions (VH) and codon optimized for N. benthamiana were first cloned into the magnICON â tobacco mosaic virus-based (TMVa: pICHa26211, (Klimyuk et al., 2014)) vectors that include the signal peptide of barley a-amylase to target proteins to the secretory pathway and two BsaI restriction sites.
Similarly, the constant region of the human kappa light chain (LC) lacking the variable region was codon optimized for N. benthamiana and cloned into the magnICON â potato virus Xbased (PVX a: pICHa31150 vector, Klimyuk et al., 2014) carrying the signal peptide of barley a-amylase and two BsaI restriction sites.
The amino acid sequence for both the light and heavy chains of Durvalumab (DL) can be found in the international Codon-optimized sequences for the variable regions (VL: 1-107aa and VH: 1-121aa) were introduced upstream of HC and LC regions using the BsaI restriction sites.

Cloning of PD-L1
The DNA sequence coding for the extracellular domain (aa 18-283, Q9NZQ7) of human programmed cell death 1 ligand (PD-L1) with a C-terminal polyhistidine tag was codon optimized for N. benthamiana and cloned into viral-based magnICON â TMVa vectors (PD-L1 His ).

Transient expression and protein extraction
Recombinant proteins were expressed in N. benthamiana glycosylation mutant plants (DXF (Strasser et al., 2008)) via agroinfiltration.Infiltrated leaves harvested 4 days postinfiltration (dpi) were used for protein extraction.For detailed information see Appendix S1.

PD-L1 His purification
For purification of His-tagged PD-L1, a 0.7 9 5 cm Econocolumn (Bio-Rad) was packed with 1 mL of Ni-NTA His•Bind â resin (Sigma) and equilibrated with 10 column volumes of 20 mM Na2HPO4, 500 mM NaCl, pH 7.4.The apoplastic fluid was loaded to the column at 1 mL/min and the column was washed with 10 volumes of 20 mM Na2HPO4; 500 mM NaCl, 30 mM imidazole, pH7.4.Bound proteins were eluted with 1 mL of 20 mM Na 2 HPO 4 , 500 mM NaCl and 500 mM imidazole 7.4).Fractions containing PD-L1 His were pooled and dialysed overnight against PBS, pH 7.4 using SnakeSkin dialysis tubing (Thermo Fisher Scientific) with a 10 kDa molecular mass cut-off.

Antibody purification
DL variants were purified from total soluble protein extracts using an € AKTA pure protein purification system (Cytiva) and a 5 mL HiTrap Protein A HP (17040301 Cytiva) column.The column was equilibrated with 5 column volumes of 20 mM Tris/HCl, 150 mM NaCl, pH 7.4 (flow rate 4 mL/min) and the sample was loaded at a flow rate of 3 mL/min.After washing the column with 12 column volumes of 20 mM Tris/HCl, 150 mM NaCl, pH 7.4 (flow rate 4 mL/min), proteins were eluted with 5 column volumes of 0.1 M glycine/HCl, pH 3.5 (flow rate 2.5 mL/min), in 1 mL fractions.Pooled eluates were neutralized with 1 M Tris/HCl, pH 8.0 and dialysed against PBS, pH 7.4 as outlined above.

Protein analysis
Apoplastic and purified recombinant proteins were fractionated by 10% SDS-PAGE under reducing and non-reducing conditions and either stained with Coomassie brilliant blue (CBB, G-250) or analysed by immunoblotting.

Glycoengineering
DL HC variants were produced in N. benthamiana as two glycoforms.Expression of recombinant proteins in N. benthamiana DXF leads to the generation of afucosylated variants, while co-expression of the human-like core a1,6 fucosyltransferase (FUT8) enables the synthesis of core a1,6 fucosylated proteins (Castilho et al., 2015).
PD-L1 His was produced in N. benthamiana DXF as an afucosylated protein.

Glycan analysis
PD-L1 and DL were digested in solution with trypsin and subjected to LC-ESI-MS analysis (Grunwald-Gruber et al., 2017).Digested peptides were separated using a nanoEase C18 column.Detection was performed with a QTOF MS (maXis 4G, Bruker) equipped with the standard ESI source in positive ion, DDA mode.The possible glycopeptides were identified as sets of peaks consisting of the peptide moiety and the attached N-glycan varying in the number of HexNAc, hexose, deoxyhexose and pentose residues.Manual glycopeptide searches were performed using FreeStyle 1.8 (Thermo Scientific); deconvolution was done using the extract function.

Enzyme-linked immunosorbent assays (ELISA)
The binding properties of plant-derived DL variants (and controls) to recombinant PD-L1 His were determined by ELISA in three technical replicates (see Appendix S1).

Binding of plant-derived DL variants to gastrointestinal cancer cells
For functional assays, NCI-N87 gastric cancer and SW48 colorectal cancer cells were seeded in 6-well culture plates, in complete growth medium, and supplemented with IFNc (40 ng/mL) to induce the cell surface expression of PD-L1.Following 48 h of IFNc stimulation, cells were harvested and their reactivity to clinical-grade DL and plant-derived DL variants was evaluated by flow cytometry.The detailed cell line description and flow cytometry assay can be found in Appendix S1.

Cell reporter assays
Details of the culture of the reporter cell lines and flow cytometry can be found in Appendix S1.

Binding to FccRs
To evaluate the binding of DL variants to human FccRs, we used a set of BW5147 lines stably expressing high levels of human FccR -CD16A (FccRIIIA), the natural V176 high-affinity variant of CD16A (FccRIIIA 176 V), CD32A (FccRIIA), CD32B (FccRIIB) and CD64 (FccRI) (Leitner et al., 2023).BW5147 cells not expressing human FccR were used as a control.FccR-expressing and control cells were incubated with different concentrations of DL variants (0.3; 1; 3 and 10 lg/mL) for 30 min at 4 °C.Following a washing step bound, antibodies detected using Allophycocyaninconjugated AffiniPure F(ab)2 Fragment of Donkey Anti-Human IgG (H + L specific).Cells were subjected to another washing step and measured by flow cytometry.

Binding to FcRn
Binding of DL variants to hFcRn was determined by surface plasmon resonance (SPR) in three replicates, using the Biacore T200 system (GE Healthcare) at 25 °C.A Biacore CM5 Sensor Chip (Cytiva) was directly coated with 2.5 lg/mL of hFcRn (R&D Systems, 8639-FC-050, P55899) using an amine coupling kit (Cytiva, BR-1000-50) to approximately 80 response units (RU).PBS (pH 6) supplemented with 0.05% Tween-20 was used as running buffer.Plant-derived DL variants were injected at 6.87-440 nM for 60 s and allowed to dissociate for 60 s.DL (Imfinzi â ) was used as control.The chip was regenerated in PBS (pH 7.4).The binding kinetics, k on (1/Ms), k off (1/s) and K D (nM) were calculated from global fittings using a 1 : 1 binding model (BIAevaluation software).In the case of a very fast on-rate and a very fast off-rate, the K D value was calculated from steady state affinity measurements.

Statistics
EC 50 values were estimated by non-linear regression based on a four-parameter logistic curve (4PL) model with GraphPad Prism (version 9).
Statistical analysis was performed with multiple comparisons one-and two-way ANOVA.

Figure 1
Figure 1 Expression and purification of plant-derived Durvalumab variants.(a) Schematic representation of an IgG highlighting the differences in amino acid sequence within the hinge and CH2 domain of Durvalumab (Imfinzi â ) and plant-derived DL variants.(b) DL variants expressed in N. benthamiana DXF plants and purified by affinity chromatography were analysed in reducing conditions.SDS-PAGE gels were either stained with Coomassie brilliant blue (CBB) or used for immunoblotting with anti-gamma chain (heavy chain) and anti-kappa chain (light chain) antibodies conjugated to HRP.The apparent molecular mass of marker proteins is shown in kilo Dalton (kDa).(c) Size exclusion chromatography profiles and CBB of SDS-PAGE gels run under nonreducing conditions.

Figure 2
Figure2Glycosylation profiles of Durvalumab variants.Glycosylation profiles of Durvalumab (Imfinziâ) and DL variants expressed in N. benthamiana DXF plants without or with co-expression of core a1,6-fucosyltransferase (FUT8) are shown.MS spectra refer to the peptide carrying the N297 N-glycan: EEQFNSTYR for DL-IgG4(PVA) S288P and EEQYNSTYR for the other variants.The assigned N-glycan structures were labelled according to the ProGlycAn nomenclature.A cartoon illustration highlights the main glycan structures detected for each peptide.For details see http://www.functionalglycomics.org/.

Figure 4
Figure 4 Blocking PD-1/PD-L1 interaction by Durvalumab.(a) Simplified schematic representation of the use of a transcriptional PD-1 + NF-jB::eGFP reporter T-cell line to evaluate the impact of plant-derived DL on T-cell activation.Cross-linking of the TCR-CD3 complex with CD3-antibody fragments expressed on K562-based stimulator cells (K562S-PD-L1) results in a strong expression of the eGFP reporter gene.Negative co-stimulatory signals induced by engagement of PD-1 with K562S expressing PD-L1 lead to the inhibition of TCR/CD3 signalling and reduced eGFP expression.Antibodies targeting PD-L1 block PD-1 engagement and restore eGFP expression.(b) Example of flow cytometric measurement of eGFP expression in stimulated PD-1 reporter cells in the absence or presence of Durvalumab.(c) PD-1 expressing NF-jB::eGFP reporter cells were stimulated for 24 h with K562S-PD-L1 in presence of Durvalumab variants (DL-IgG1, DL-IgG1 (LALAPG) and DL-IgG4 (PVA) S288P) and glyco-variants (GnGn and GnGnF6) at indicated concentrations (1000 to 1 ng/mL).Untreated PD-1 reporter cells stimulated with K562S-PD-L1 used as controls.Data are derived from two independent experiments performed in triplicates (n = 6).Data were normalized to the eGFP expression of PD-1 reporter cells stimulated under conditions where PD-L1 was fully blocked (1 lg/mL of PD-L1 antibody).Inhibition curves (FigureS2) and half maximum effective concentrations (EC 50 ) were calculated from normalized data using a 4parameter logistic function.

Figure 5
Figure 5 Binding of Durvalumab variants to different FccRs.(a) Plant-derived Durvalumab variants (DL-IgG1, DL-IgG1 (LALAPG) and DL-IgG4 (PVA) S288P ) and glyco-variants (GnGn and GnGnF6) were compared to DL (Imfinziâ) regarding their ability to bind to cells expressing CD16, CD16V, CD32A, CD32B and CD64 Fcc receptors.Binding was determined at indicated antibody concentrations (0.3-10 lg/mL) and bound antibodies were detected using an Allophycocyanin-conjugated AffiniPure F(ab)2 Fragment of Donkey Anti-Human IgG (H + L specific).Cells not expressing human FccR served as control to normalize data.Two independent experiments were performed in duplicates.The table summarizes the reduction of binding to FccR upon core fucosylation.Statistical analysis was performed with multiple comparisons two-way ANOVA (n = 2).(b) Comparison of the binding of DL and plant-derived core-fucosylated (GnGnF6) DL variants (used at 10 lg/mL) to different FccRs.

ª
2023 The Authors.Plant Biotechnology Journal published by Society for Experimental Biology and The Association of Applied Biologists and John Wiley & Sons Ltd., 22, 1224-1237 Plant-derived Durvalumab with favourable FcR binding 1233 ImMunoGeneTics information database (IMGT/mAb-DB ID 528).

ª
2023 The Authors.Plant Biotechnology Journal published by Society for Experimental Biology and The Association of Applied Biologists and John Wiley & Sons Ltd., 22, 1224-1237 conditions where PD-L1 was fully blocked (1 lg/mL of PD-L1 antibody).

Figure S1
Figure S1 Binding of Durvalumab variants to PD-L1 His .Figure S2 Expression, purification and glycosylation of plantderived PD-L1 His .Figure S3 Blocking PD-1/PD-L1 interaction by Durvalumab. Figure S4 Binding of Durvalumab variants to Fcc receptors.Figure S5 SPR sensorgrams for the binding of Durvalumab variants to hFcRn.Table S1 Kinetic parameters of the binding of Durvalumab variants to hFcRn.Appendix S1 Transient expression-agroinfiltration.
Figure S1 Binding of Durvalumab variants to PD-L1 His .Figure S2 Expression, purification and glycosylation of plantderived PD-L1 His .Figure S3 Blocking PD-1/PD-L1 interaction by Durvalumab. Figure S4 Binding of Durvalumab variants to Fcc receptors.Figure S5 SPR sensorgrams for the binding of Durvalumab variants to hFcRn.Table S1 Kinetic parameters of the binding of Durvalumab variants to hFcRn.Appendix S1 Transient expression-agroinfiltration.
ª 2023 The Authors.Plant Biotechnology Journal published by Society for Experimental Biology and The Association of Applied Biologists and John Wiley & Sons Ltd., 22, 1224-1237

Table 2
Binding of Durvalumab to hFcRn ª 2023 The Authors.Plant Biotechnology Journal published by Society for Experimental Biology and The Association of Applied Biologists and John Wiley & Sons Ltd., 22, 1224-1237

Table S1
Kinetic parameters of the binding of Durvalumab variants to hFcRn.Appendix S1 Transient expression-agroinfiltration.ª 2023 The Authors.Plant Biotechnology Journal published by Society for Experimental Biology and The Association of Applied Biologists and John Wiley & Sons Ltd., 22, 1224-1237 Plant-derived Durvalumab with favourable FcR binding 1237